Axillary treatment for operable primary breast cancer

Summary of findings for the main comparison. No axillary surgery compared with full axillary surgery for operable primary breast cancer

No axillary surgery compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: no axillary surgery Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	No axillary surgery
All‐cause mortality	92% overall survival at 5 years^a	92% overall survival at 5 years (91% to 93%)	HR 1.06 (0.96 to 1.17)	3849 (10 studies)	⊕⊕⊕⊝ moderate^b
Locoregional recurrence	86% locoregional recurrence‐free survival at 5 years^c	71% locoregional recurrence‐free survival at 5 years (66% to 76%)	HR 2.35 (1.91 to 2.89)	20,863^d (5 studies)	⊕⊕⊕⊝ moderate^e
Lymphoedema Increase in arm circumference Follow‐up: 1 or more years	236 per 1000	87 per 1000 (66 to 117)	OR 0.31 (0.23 to 0.43)	1714 (4 studies)	⊕⊕⊝⊝ low^e,f
Arm or shoulder movement impairment Follow‐up: 1 or more years	91 per 1000	67 per 1000 (47 to 95)	OR 0.72 (0.49 to 1.05)	1495 (5 studies)	⊕⊝⊝⊝ very low^f,g
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk is taken from full axillary surgery arm of Institut Curie. ^bConfidence interval around the effect estimate includes both no effect and appreciable harm associated with no axillary surgery. ^cAssumed risk is taken from full axillary surgery arm of Institut Curie, local or axillary recurrence rates. ^dPerson‐years of follow‐up. ^eSubstantial heterogeneity (I² > 50%). ^fUnclear blinding of outcome assessment. ^gConsiderable heterogeneity (I² > 75%).

Summary of findings 2. Axillary sampling compared with full axillary surgery for operable primary breast cancer

Axillary sampling compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: axillary sampling Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Full axillary surgery	Axillary sampling
All‐cause mortality	82% overall survival at 5 years^a	83% overall survival at 5 years (79% to 87%)	HR 0.94 (0.73 to 1.21)	967 (3 studies)	⊕⊕⊝⊝ low^b,c
Local recurrence	85% local recurrence‐free survival at 5 years^d	80% local recurrence free survival at 5 years (71% to 86%)	HR 1.41 (0.94 to 2.12)	1404 (3 studies)	⊕⊕⊝⊝ low^e,f
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk is taken from full axillary surgery arm of E'dburgh Sample/Clear. ^bSubstantial heterogeneity. ^cConfidence interval for the effect includes both appreciable benefit and harm with axillary sampling. ^dAssumed risk taken from full axillary surgery arm of Cardiff. ^eNo blinding of outcome assessment or blinding not reported. ^fConfidence interval for effect includes both no difference and appreciable harm with axillary sampling. Low number of events.

Summary of findings 3. Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer

Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: sentinel node biopsy Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	Sentinel node biopsy
All‐cause mortality	96% overall survival at 5 years^a	96% overall survival at 5 years (95% to 96%)	HR 1.05 (0.89 to 1.25)	6352 (3 studies)	⊕⊕⊕⊝ moderate^b
Lymphoedema Patient‐reported lymphoedema of any severity Follow‐up: 12 months	132 per 1000	48 per 1000 (22 to 115)	OR 0.33 (0.15 to 0.86)	815 (3 studies)	⊕⊕⊝⊝ low^b,c
Subjective arm movement impairment Follow‐up: 12 months	100 per 1000	40 per 1000 (24 to 69)	OR 0.38 (0.22 to 0.67)	877 (2 studies)	⊕⊝⊝⊝ very low^b,d,e
Paraesthesia Follow‐up: 12 months	776 per 1000	343 per 1000 (238 to 444)	OR 0.15 (0.09 to 0.23)	495 (2 studies)	⊕⊕⊝⊝ low^d,e
Pain Follow‐up: 12 months	177 per 1000	86 per 1000 (61 to 126)	OR 0.44 (0.3 to 0.67)	877 (2 studies)	⊕⊕⊝⊝ low^d,e
Numbness Follow‐up: 12 months	346 per 1000	185 per 1000 (152 to 222)	OR 0.43 (0.34 to 0.54)	1799 (3 studies)	⊕⊕⊕⊝ moderate^f
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk taken from the full axillary surgery arm of Milan. ^bLow number of events. ^cIncomplete follow‐up for patient‐reported lymphoedema in ALMANAC. Event rates not reported in Addenbrookes 2. ^dModerate or substantial heterogeneity. ^eNo blinding or blinding not reported. ^fNo explanation provided.

See: Summary of findings for the main comparison No axillary surgery compared with full axillary surgery for operable primary breast cancer; Summary of findings 2 Axillary sampling compared with full axillary surgery for operable primary breast cancer; Summary of findings 3 Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer; Summary of findings 4 Radiotherapy alone compared with full axillary surgery for operable primary breast cancer

Summary of findings 4. Radiotherapy alone compared with full axillary surgery for operable primary breast cancer

Radiotherapy alone compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: radiotherapy alone Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	Radiotherapy alone
All‐cause mortality	81% overall survival at 5 years^a	79% overall survival at 5 years (77% to 81%)	HR 1.1 (1 to 1.21)	2469 (4 studies)	⊕⊕⊕⊕ high
Local recurrence	90% local recurrence‐free survival at 5 years^b	92% local recurrence‐free survival at 5 years^a (90% to 93%)	HR 0.8 (0.64 to 0.99)	22,256^c (4 studies)	⊕⊕⊕⊕ high
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk from full axillary surgery arm of NSABP B‐04 using mean 5‐year overall survival in combined N+ and N‐ groups. ^bAssumed risk from full axillary surgery arm of NSABP B‐04, using mean 5‐year risk for local or regional recurrence in combined lymph node‐positive and ‐negative groups. ^cPerson‐years of follow‐up.

Background

available in

Description of the condition

Invasive breast cancer occurs when uncontrolled, abnormal growth and division of cells in the lobules or ducts of the breast spreads to surrounding tissue. The Union Internationale Contre le Cancer staging system for breast cancer (UICC 1987) reflects how, when left untreated, cancer cells may spread locally to breast tissue and lymph glands in the axilla (stages I to III) and through the bloodstream and lymphatic system to other parts of the body (stage IV).

Description of the intervention

Removal of regional lymph nodes during attempts to achieve a curative excision for management of most cancers has a long history (Halsted 1895). Its aim consists of both local control of axillary disease and determination of stage to permit appropriate adjuvant therapy. Axillary surgery is a key component of breast cancer management, with UK clinical guidelines specifying that minimal surgery (preferably sentinel lymph node biopsy (SLNB)) should be performed to stage the axilla for patients with early invasive breast cancer and clinically negative axillary lymph nodes (NICE 2009).

Several alternative approaches to axillary surgery may be used.

Axillary clearance ‐ removal of all nodal tissue in the axilla by dissection up to the level of the axillary vein (Craig 1998) ‐ was previously the standard practice in many units. Full axillary clearance carries increased morbidity when compared with breast surgery alone, with 10% to 15% incidence of chronic arm lymphoedema (Kissin 1986), 9% incidence of late seroma, 2.2% infection rate, 12% breast oedema and 0.3% risk of damage to the long thoracic nerve (Senofski 1991). Other problems include shoulder stiffness ("frozen shoulder"), which can be severe (Kissin 1986). Immediate axillary node clearance is not considered appropriate in the absence of evidence of cancer spread determined by biopsy before surgery.
Axillary node sampling ‐ removal of four or five axillary nodes from the lower axilla (Craig 1998) ‐ involves removal of individual nodes, leaving axillary fat and most nodes and lymphatics intact. As a result, virtually none of the complications listed for axillary clearance are associated with this procedure. Women whose sampled axillary nodes contain cancer may need subsequent axillary clearance or radiotherapy. This previously popular approach was once considered appropriate.
Sentinel lymph node biopsy (Kelley 1998) ‐ a procedure in which the lymphatic pathway from the site of breast cancer is tracked with the use of a radio‐isotope or blue lymphatic dye ‐ allows biopsy of the first lymph node or nodes (sentinel node). Sentinel nodes are most likely to involve spread of cancer, and this approach allows accurate assessment of whether the cancer has spread along with removal of a small number of nodes (typically three or fewer).
In some patients who are not candidates for adjuvant therapies, surgeons may omit axillary surgery altogether to avoid additional morbidity (EBCTCG 1998, Walsh 1989). This has led some surgeons to spare some frail women with breast cancer from undergoing staging of the clinically uninvolved axilla by means of sentinel node biopsy or full clearance (Yancik 1989).

How the intervention might work

Removal of axillary nodes can improve local control of axillary disease while providing information on cancer stage that can be used to guide adjuvant therapy.

Why it is important to do this review

Arguments for and against each of these procedures are complicated and, as a result, practice is variable. Statistical synthesis of outcomes for these procedures will offer surgeons and patients a more reliable evidence base on which they can make difficult decisions concerning treatment.

Objectives

available in

Methods

available in

Criteria for considering studies for this review

Types of studies

Randomised controlled trials.

Types of participants

Women with clinically defined operable primary breast cancer, that is, primary tumour not fixed to underlying structures (includes tumour‐node‐metastasis (TNM) classifications T1‐3 and T4b with only minor skin involvement, N0‐1 and M0) nor to mobile lymph nodes (UICC 1987).

Types of interventions

Axillary lymph node dissection (ALND) versus no axillary surgery at the time of primary surgery
1. With the following subgroups for both arms:
  1. Radiotherapy
  2. No radiotherapy
ALND versus axillary sampling at the time of primary surgery
1. With the following subgroups for both arms:
  1. Radiotherapy
  2. No radiotherapy
2. And the following subgroups for the limited axillary staging arm:
  1. Further treatment for histologically node‐positive cases
  2. No further treatment for histologically node‐positive cases
ALND versus SLNB at the time of primary surgery
1. With the following subgroups for both arms:
  1. Radiotherapy
  2. No radiotherapy
2. And the following subgroups for the limited axillary staging arm:
  1. Further treatment for histologically node‐positive cases
  2. No further treatment for histologically node‐positive cases
Axillary sampling versus sentinel node biopsy at the time of primary surgery
1. With the following subgroups for both arms:
  1. Radiotherapy
  2. No radiotherapy
2. And the following subgroups for both arms:
  1. Further treatment for histologically node‐positive cases
  2. No further treatment for histologically node‐positive cases
Axillary sampling versus no axillary surgery at the time of primary surgery
1. With the following subgroups for both arms:
  1. Radiotherapy
  2. No radiotherapy
2. And the following subgroups for the limited axillary staging arm:
  1. Further treatment for histologically node‐positive cases
  2. No further treatment for histologically node‐positive cases
SLNB versus no axillary surgery at the time of primary surgery
1. With the following subgroups for both arms
  1. Radiotherapy
  2. No radiotherapy
2. And the following subgroups for the limited axillary staging arm:
  1. Further treatment for histologically node‐positive cases
  2. No further treatment for histologically node‐positive cases
ALND with no radiotherapy versus no axillary surgery with radiotherapy
1. With no subgroups

For all studies involving full axillary surgery or axillary sampling, the number of nodes removed and the method of node analysis used were recorded when available, to indicate whether an adequate sampling or clearance procedure was performed.

Types of outcome measures

Primary outcomes

Survival ‐ overall (interval between start of treatment or randomisation and death)
Disease control in the axilla (interval between start of treatment and the need for second‐line treatment or palliative treatment or regional recurrence in the axilla)
Breast cancer recurrence, either locally within the breast (local recurrence) or distantly as metastatic disease (distant recurrence), with time to recurrence and site of recurrence recorded
Adverse events (surgical complications) including acute local surgical complications, such as haematoma, infection, wound dehiscence or seroma, and acute systemic complications, such as chest infection, deep venous thrombosis, pulmonary embolism, cardiac failure, cardiac ischaemia and cerebrovascular accident
Long‐term complications including lymphoedema, shoulder stiffness, paraesthesia, pain, loss of functional capacity, winging of scapula and wound contracture or scarring

Secondary outcomes

Quality of life (measured on a validated scale)
Psychological and psychosocial variables (measured on validated scales)

Search methods for identification of studies

Electronic searches

The Trials Search Co‐ordinator for the Cochrane Breast Cancer Review Group searched the Specialised Register of the Group on 16 March 2015. Details of sources and search strategies used to populate this register are provided in the Group module in the Cochrane Library (http://onlinelibrary.wiley.com/o/cochrane/clabout/articles/BREASTCA/frame.html) . We have extracted for consideration studies coded as "AXILLARY NODE(S)", "EARLY BREAST CANCER", "LOCALLY ADVANCED BREAST CANCER", "PSYCHOSOCIAL" or "SURGERY" on the Specialised Register.

We searched the Cochrane Central Register of Controlled Trials (CENTRAL; Issue 2) in the Cochrane Library on 16 March 2015. See Appendix 1 for the search strategy used.

In addition, an information specialist searched the following databases while using the search terms and strategy identified in Appendix 2: MEDLINE via OvidSP (2007 to 12 March 2015), PreMEDLINE via OvidSP (12 March 2015) and Embase via OvidSP (2002 to 12 March 2015). We used a validated filter to identify reports of RCTs in our initial search of MEDLINE (Lefebvre 2001), and for updated searches, we used the revised filter (Lefebvre 2011). We used the Scottish Intercollegiate Guidelines Network RCT filter in our search of Embase (http://www.sign.ac.uk/methodology/filters.html).

We also searched on 16 March 2015 the World Health Organization International Clinical Trials Registry Portal (WHO ICTRP) (Appendix 3) and ClinicalTrials.gov (Appendix 4), for prospectively registered and ongoing trials.

Searching other resources

We searched (on 12 March 2015) conference proceedings from the American Society of Clinical Oncology (ASCO) 41st to 50th Annual Meetings (2005 to 2014) via Journal of Clinical Oncology (http://jco.ascopubs.org/site/meetings). We also searched (on 12 March 2015) conference proceedings from the San Antonio Breast Cancer (SABCS) 29th to 37th Annual Symposium Meetings (2006 to 2014) via the Cancer Research website (http://cancerres.aacrjournals.org/).

We contacted the authors of included and ongoing trials by email and asked them if they knew of any relevant studies. This yielded no additional studies. We also checked the reference lists of included studies and published reviews to look for relevant studies.

Data collection and analysis

Selection of studies

Two review authors (NB, MSH or MA) screened the titles and abstracts of references identified by electronic searches to identify publications of potentially eligible trials. We obtained a copy of the full‐text article for each reference reporting a potentially eligible trial, and we applied the review selection criteria to each trial. We reported all exclusions of potentially eligible trials in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) diagram (Figure 1) and, in some cases, in the Characteristics of excluded studies table. We used trial publications to assess each trial's eligibility, and for unpublished trials, we obtained information from the trial protocol or the next best available resource. When necessary and possible, we sought additional information from the principal investigator. Two review authors (NB, MSH or MA) independently assessed each potentially eligible trial for inclusion in the review and resolved discrepancies in eligibility judgements by discussion.

Figure 1

Study flow diagram.

Data extraction and management

We extracted data from published trial reports and entered them onto an electronic form (using Microsoft Word). Two review authors (NB, MSH or MA) independently extracted data from each trial and resolved disagreements regarding data extraction by discussion. The Early Breast Cancer Trialists' Collaborative Group (Clarke 2005) has published a meta‐analysis based on individual participant data for many of the included trials. We used this meta‐analysis as an additional source of outcome data for trials included in this review.

We contacted the authors of included and ongoing trials by email and asked them to share unpublished data from their trials and to clarify details about their trial that were unclear or missing from the published reports.

Assessment of risk of bias in included studies

We assessed the risk of bias of included studies by applying standard Cochrane methods for randomised trials as outlined in Higgins 2011. We assessed selection bias (random sequence generation, allocation concealment; two items) and reporting bias (selective reporting; one item) at study level, and detection bias (blinding of outcome assessment; one item) and attrition bias (incomplete outcome data; one item) at outcome level. We did not assess detection bias for the outcome of survival because this in an objective outcome, and we did not assess performance bias (one item) because blinding of healthcare personnel and participants is not possible for the interventions considered in this review.

Measures of treatment effect

For dichotomous data, we used odds ratio (OR) as the measure of treatment effect. For continuous data, we used the standardised mean difference (SMD). For time‐to‐event (survival) data, we used the hazard ratio (HR). For our meta‐analysis of time‐to‐event outcomes in Review Manager 5.3 (RevMan), we used 'O‐E' (observed minus expected) and 'V' (variance) statistics or hazard ratios for each trial. If these values were not reported for a given trial, we calculated them from available statistics, if possible, using the methods described in Tierney 2007.

Unit of analysis issues

Some trials performed serial measurements of arm volume and/or function over the first months and years after surgery. For our analysis, we used the measurement at one year post operation (or at the nearest time point after one year for trials not reporting data at the one‐year time point). One trial (NSABP B‐04) included three treatment comparison groups. This presented an issue only for analysis of less versus more axillary surgery (Analysis 5.1); to avoid double‐counting of the ALND group, we omitted the comparison of radiotherapy versus ALND in clinically node negative study participants.

Dealing with missing data

We analysed only data available in trial reports or obtained through contact with trial authors. We did not attempt data imputation.

Assessment of heterogeneity

We assessed statistical heterogeneity (variability in intervention effects) in meta‐analyses by using the I² statistic, which we interpreted alongside magnitude and direction of effects. We regarded an I² value of 30% to 60% as indicating potentially important heterogeneity and downgraded the overall quality of evidence for that outcome (owing to inconsistency) in the summary of findings tables. If heterogeneity was greater than 50%, we did not pool effect estimates but instead used the range of effects reported by individual studies.

Assessment of reporting biases

We checked reporting bias by using funnel plots and checked that outcomes measured in individual trials were reported in trial publications. If we suspected reporting bias for a given outcome, we downgraded the overall quality of the evidence in the summary of findings table owing to reporting/publication bias.

Data synthesis

We statistically synthesised time‐to‐event outcomes that were entered into RevMan as ‘O–E' and 'Variance’ outcomes by using a fixed‐effect model (the random‐effects model is not an option for this analysis in RevMan). We analysed dichotomous outcomes by using fixed‐effect (Mantel‐Haenszel method) and random‐effects (DerSimonian and Laird) models (Sensitivity analysis).

For summary of findings tables (summary of findings Table for the main comparison; summary of findings Table 2; summary of findings Table 3; summary of findings Table 4), we used the GRADE approach to assign an overall assessment of the quality of the evidence. In addition to the risk of bias assessment, the GRADE quality rating includes assessments of inconsistency, indirectness and imprecision of results, and of the likelihood of publication bias. We prioritised Primary outcomes for inclusion in summary of findings tables and organised them according to Types of interventions.

Subgroup analysis and investigation of heterogeneity

We planned the following subgroup analyses.

Radiotherapy versus no radiotherapy.
Further treatment versus no further treatment for histologically node‐positive participants.
Age groups (18 to 49 years; 50 to 69 years; 70 to 79 years; 80 years and older).

We were not able to analyse results by age group. When evidence suggested potentially important between‐study statistical heterogeneity (I² value of 30% to 60%), we compared fixed‐effect and random‐effects estimates to check whether the intervention effect was sensitive to the type of model used, although it should be noted that such comparisons were not possible for analyses of time‐to‐event outcomes, as already outlined in the Data synthesis section.

Sensitivity analysis

To examine the robustness of our results, we performed sensitivity analyses that included only studies with low risk of bias for allocation concealment. Moreover, we planned to undertake sensitivity analyses to examine short‐term and long‐term morbidity outcomes only for studies with low risk of bias for blinded assessment of these outcomes. However, we considered none of the studies to be at low risk of bias for these items, so we could not perform these analyses.

Results

Description of studies

Results of the search

In total, we screened 7436 references for inclusion in this review (Figure 1). We retrieved full‐text articles for 163 references to potentially relevant publications to check inclusion eligibility. Of these,13 full‐text articles reported on eight trials that appeared relevant but did not meet all of the inclusion criteria (AATRM‐048‐13‐2000; ACOSOG Z0011; Buenos Aires; Copenhagen; Edinburgh SES; IBCSG‐23‐01; IPO‐P; OTOASOR). See Excluded studies section.

We identified six articles reporting on eight possibly eligible ongoing trials (AMAROS; GF‐GS 01; KiSS; NCT01717131; NCT02167490; NCT02271828; SNAC2; SOUND). Two studies (ISRCTN88463711; Semiglazov 2003) await classification. We excluded 45 other full‐text articles for the following reasons: 23 used ineligible Types of interventions, four included ineligible Types of participants and 18 were the wrong Types of studies.

The remaining 97 articles were reports of 26 eligible RCTs included in this review. We contacted the authors of included studies by email to ask about other relevant trials for inclusion in the review, but this yielded no additional studies.

Included studies

This review includes 26 studies that performed 27 treatment comparisons.

Full axillary surgery versus no axillary surgery

Ten studies compared axillary lymph node dissection (ALND) versus no axillary surgery (N = 3849; Addenbrookes; Guy's; Hammersmith; IBCSG‐10‐93; Institut Curie; Institut Bergonie; Malmo; Milan 2; Milan 3; NSABP B‐04).

The Malmo trial compared ALND plus radiotherapy (RT) versus no ALND and no RT. In one trial (IBCSG‐10‐93), only those treated with conservative breast surgery received RT. In Addenbrookes; Guy's; Hammersmith; Institut Curie; Institut Bergonie; Milan 2; and Milan 3, all study participants received RT. NSABP B‐04 reported a three‐group comparison of ALND, no ANLD plus RT and no ALND for patients with clinically negative axillary nodes. Patients in the ALND arm received limited RT to the chest wall. We included the ALND and no ALND arms of NSABP B‐04 for this comparison.

Five studies excluded patients with clinically involved lymph nodes (Institut Bergonie; Institut Curie; Malmo; Milan 2; Milan 3), whereas the remaining five studies included these patients only when clinically involved nodes were mobile and were not fixed to underlying structures (Addenbrookes; Guy's; Hammersmith; IBCSG‐10‐93; NSABP B‐04).

Seven studies (Addenbrookes; Guy's; Hammersmith; IBCSG‐10‐93; Malmo; Milan 2; NSABP B‐04) did not provide extra treatment for participants with histologically positive axillary lymph nodes. In Institut Curie, Institut Bergonie and Milan 3, such individuals could receive chemotherapy or hormone therapy.

Full axillary surgery versus axillary sampling

Six trials compared ALND versus axillary sampling (N = 1559; Cape Town; Cardiff; E'dburgh Sample/Clear; Edinburgh 1; Ostersund; Xu 2003). Of these trials, only Cape Town did not provide RT as part of the randomised treatment.

In Cardiff, E'dburgh Sample/Clear, Edinburgh 1 and Ostersund, participants with histologically positive sampled axillary lymph nodes received additional RT. In Xu 2003, RT was provided only for participants with more than three positive axillary lymph nodes and for those with a primary tumour in the central quadrant. In Cape Town, participants with histologically positive sampled nodes did not receive additional treatment.

Four trials (Cape Town; Cardiff; E'dburgh Sample/Clear; Edinburgh 1) included patients with clinically involved axillary nodes, provided such nodes were mobile. In the Ostersund and Xu 2003 trials, inclusion criteria were unclear.

Full axillary surgery versus sentinel node biopsy

Seven trials compared ALND versus sentinel lymph node biopsy (SLNB) (N = 9426; Addenbrookes 2; ALMANAC; Genoa; GIVOM Sentinella; Milan; NSABP B‐32; SNAC).

In three studies (Genoa; GIVOM Sentinella;Milan), only participants treated with breast‐conserving surgery received RT, which meant that some of the participants in Genoa and GIVOM Sentinella did not receive RT. In the remaining trials (Addenbrookes 2; ALMANAC; NSABP B‐32; SNAC), participants received RT according to local treatment protocols, which meant that in practice, most participants received RT.

In all of these trials, participants with histologically positive sentinel lymph nodes received further treatment. Treatment for histologically positive lymph nodes consisted of ALND (Addenbrookes; Genoa; GIVOM Sentinella; NSABP B‐32; Milan; SNAC) or the choice of ALND or RT to the axilla (ALMANAC).

Addenbrookes 2; ALMANAC; Genoa; GIVOM Sentinella; NSABP B‐32 and SNAC excluded patients with clinically involved axillary nodes, but it was unclear whether the Milan trial excluded such individuals.

Axillary sampling versus SLNB

We identified no studies for this comparison.

Axillary sampling versus no axillary surgery

We identified no studies for this comparison.

SLNB versus no axillary surgery

We identified no studies for this comparison.

Full axillary surgery with no RT versus no axillary surgery with RT

Four trials compared ALND without RT versus RT alone (N = 2585; Manchester; NSABP B‐04; SE Scotland; WSSA Glasgow). One of these trials (NSABP B‐04) performed a three‐group comparison of ALND, no ANLD plus RT and no ALND with clinically negative axillary nodes. Participants in the ALND arm of this trial did receive limited RT to the chest wall. We included in this review the ALND and no ALND plus RT arms of NSABP B‐04. This trial randomised participants with clinically positive nodes to ALND or no ANLD plus RT; we analysed these results separately. All of these trials included patients with clinically involved axillary nodes provided such nodes were mobile. None of these trials specified that they provided extra treatments for participants with histologically positive axillary nodes.

Excluded studies

We excluded eight trials from this review (see Excluded studies table for full details). We excluded two otherwise relevant trials because treatment allocation was not randomised; instead, investigators decided treatment group on the basis of month of birth (Buenos Aires) or order of entry into the trial (Copenhagen). We excluded the Edinburgh South East Scotland trial (Edinburgh SES) because it did not involve axillary surgery or lymph node biopsy.

We excluded five trials comparing ALND versus no further axillary surgery because trial entry or inclusion depended on the results of SLNB (AATRM‐048‐13‐2000; ACOSOG Z0011; IBCSG‐23‐01; IPO‐P; OTOASOR). All of these trials excluded patients with clinically involved axillary nodes before their primary surgery. The IPO‐P trial included only those with negative SLNB. Remaining trials included only patients with a positive SLNB (AATRM‐048‐13‐2000; ACOSOG Z0011; IBCSG‐23‐01; OTOASOR). AATRM‐048‐13‐2000 included only patients with sentinel lymph node micrometastases.

Risk of bias in included studies

We summarised in Figure 2 the risk of bias of included studies.

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

In all, 17 trials clearly reported random sequence generation (Addenbrookes; Addenbrookes 2; ALMANAC; Cape Town; Cardiff; Edinburgh 1; Genoa; GIVOM Sentinella; Guy's; Hammersmith; IBCSG‐10‐93; Malmo; Milan; Milan 2; Milan 3; NSABP B‐32; SNAC), and the remaining nine trials provided unclear information on this (E'dburgh Sample/Clear; Institut Bergonie; Institut Curie; Manchester; NSABP B‐04; Ostersund;SE Scotland; WSSA Glasgow; Xu 2003).

Allocation concealment was adequate in 15 trials (Addenbrookes; ALMANAC; Cape Town; Cardiff; E'dburgh Sample/Clear; Edinburgh 1; Genoa; GIVOM Sentinella; IBCSG‐10‐93; Milan; Milan 2; Milan 3; NSABP B‐32; SE Scotland; SNAC) and unclear in the other 11 trials (Addenbrookes 2; Guy's; Hammersmith; Institut Bergonie; Institut Curie; Malmo; Manchester; NSABP B‐04; Ostersund; WSSA Glasgow; Xu 2003). In trials with unclear risk of selection bias, we did not observe obvious differences in the baseline characteristics of treatment groups, although Malmo, Ostersund and WSSA Glasgow poorly reported baseline characteristics.

Blinding

Two studies were at high risk of detection bias due to lack of blinding of outcome assessment or disease recurrence and adverse event outcomes (Addenbrookes 2; SNAC2). All other studies were at unclear risk of detection bias due to poor reporting.

Incomplete outcome data

Seventeen trials had low risk of incomplete overall survival data (ALMANAC; Cape Town; Cardiff; E'dburgh Sample/Clear; Edinburgh 1; Genoa; Guy's; IBCSG‐10‐93; Institut Bergonie; Malmo; Manchester; Milan; Milan 3; NSABP B‐32; SE Scotland; WSSA Glasgow; Xu 2003). The remaining trials were at unclear risk of bias due to incomplete outcome data because they did not report overall survival or the completeness of their reporting was uncertain. We observed a similar pattern for outcomes related to breast cancer recurrence and disease control in the axilla (Figure 2).

We judged five trials to be at low risk of bias because they provided incomplete data for short‐term adverse events (Addenbrookes 2; ALMANAC; NSABP B‐32; SNAC; Xu 2003); all of these trials involved SLNB. Three trials were at high risk (IBCSG‐10‐93; Ostersund; SE Scotland), and the remainder were at uncertain risk. We noted a similar pattern for long‐term adverse events, with three trials at low risk of bias (Addenbrookes 2; Hammersmith; Xu 2003), seven trials at high risk (E'dburgh Sample/Clear; Guy's; IBCSG‐10‐93; Milan; NSABP B‐32; Ostersund; SE Scotland) and the remainder at uncertain risk.

Selective reporting

Three trials were at low risk of bias due to selective reporting (Addenbrookes; ALMANAC; Edinburgh 1). Addenbrookes 2 and Milan 3 were at high risk of bias due to selective reporting of some outcomes on the basis of statistical significance. The remaining trials were at uncertain risk of bias due to selective reporting.

Other potential sources of bias

Trials typically reported intention‐to‐treat analyses, but in four trials it was unclear whether such analyses were performed (Cape Town; NSABP B‐04; Ostersund; WSSA Glasgow). We included two trials that performed per‐protocol analysis (Malmo; Milan) because study authors stated that per‐protocol results were similar to intention‐to‐treat results (Malmo), or because protocol violations were few (Milan).

Effects of interventions

We recorded in Table 1 time‐to‐event statistics extracted for each trial. We listed in Table 2 the definitions of adverse event outcomes used in each study, and we summarised in Table 3 adverse events at various time points after treatment.

Table 1. Summary time‐to‐event statistics

Study	Outcome reported	Observed	Expected	Variance	HR	95% CIs	P value	Follow‐up	Notes
Addenbrookes	Overall mortality	ALND: 107/112 No ALND: 108/121	o‐e = ‐3.1	46.5	0.94	(0.70 to 1.25)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition The number of patients reported by Clarke 2005 differs from that reported by Brinkley (1971).
Addenbrookes	Breast cancer mortality	ALND: 74/112 No ALND: 78/121	o‐e = ‐2.2	32.8	‐	‐	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Addenbrookes	Isolated local recurrence	ALND: 7 events/1148 women‐years No ALND: 15 events/1218 women‐years	o‐e = 3.3	5.4	1.8	(0.79 to 4.28)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition
ALMANAC	Overall mortality	ALDN: 7/476 SLNB: 7/478	NA	NA	NA	NA	NA	1 year	Cannot calculate o‐e. Not included in meta‐analysis
ALMANAC	Axillary recurrence	ALDN: 4/476 SLNB: 1/478	NA	NA	NA	NA	NA	1 year	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Overall mortality	ALND: 21/43 Simple: 30/52	o‐e = 4.74	12.35	1.47	(0.84 to 2.56)	0.1775	10 years	Tierney et al (2007) method 7 used log‐rank test results from figure 1. Cape Town
Cape Town	Overall mortality (node‐negative)	ALND: 14/21 Simple: 26/30	o‐e = 1.8	7.6	‐	‐	NA		Taken from Clarke 2005 (Appendix web figure 9a; Groote‐Schuur), then O‐E sign changed to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Cape Town	Overall mortality (node‐positive)	ALND: 19/22 Simple: 22/25	o‐e = ‐1.9	7.7	‐	‐	NA		Taken from Clarke 2005 (Appendix web figure 9b; Groote‐Schuur), then O‐E sign changed to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Cape Town	Isolated local recurrence (node‐negative)	ALND: 3/206 women‐years Simple: 8/232 women‐years	o‐e = 1.7	2.3	2.09	(0.58 to 7.63)	NA		Taken from Clarke 2005 (Appendix web figure 9a; Groote‐Schuur), then inverted to reflect that more surgery is our control and less surgery is our research condition
Cape Town	Isolated local recurrence (node‐positive)	ALND: 5/134 women‐years Simple: 9/173 women‐years	o‐e = 0.0	2.0	1.00	(0.25 to 4.00)	NA		Taken from Clarke 2005 (Appendix web figure 9b; Groote‐Schuur), then inverted to reflect that more surgery is our control and less surgery is our research condition
Cape Town	Axillary recurrence	ALND: 2/43 Simple: 8/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Any locoregional recurrence	ALND: 11/43 Simple: 19/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Distant metastases	ALND: 11/43 Simple: 13/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cardiff	Overall survival	ALND: N = 97 Sampling: N =103 Total events = 152 Fig 2 data: ALND: 23/97 Sampling: 13/103	o‐e: 7.4	38	1.21	(0.29 to 0.99)	0.23	20 years	HR calculated using log‐rank P value from Stewart et al (1994, page 42) by Tierney 2007 method 8, 9. Owing to non‐proportionality of hazard rates, HR cannot be included in meta‐analysis
Cardiff	Disease‐free survival	ALND: 97 Sampling: 103	5.87	7.75	2.13	(1.05 to 4.31)	0.035	20 years	Log‐rank P value Tierney 2007 method 8, 9 (page 43 & Fig 5 Stewart et al, 1994)
Cardiff	Locoregional recurrence (chest wall, axilla, supraclavicular/internal mammary nodes)	ALND: 19/94 Sampling: 31/99 Fig 4: ALND: 11/97 Sampling: 22/103	o‐e: 6.46	11.78	1.73	(0.87 to 3.42)	NA	20 years	Tierney et al (2007) method 4 used and data from Figure 4 & page 42 Stewart et al (1994)
Cardiff	Distant relapse	ALND: 43/94 Sampling: 59/99	o‐e: 8.4	24.87	1.4	(0.99 to 1.71)	0.092	20 years	Data from Table 2, Stewart et al (1994): excludes patients with radiotherapy violations. Per‐protocol analysis ‐ not included in meta‐analysis
Cardiff	Breast cancer recurrence (total) (locoregional and distant relapse)	ALND: 62/94 Sampling: 90/99	o‐e: 12.77	36.71	1.42	(1.18 to 1.61)	0.035	20 years	Calculated from Stewart et al (1994) (excludes RT violations) per‐protocol analysis Risk of overestimation not certain as these are first events or total events.‐ not included in meta‐analysis
Edinburgh 1	Overall survival	ALND: ?/232 Sampling: ?/234 Total events = 53 ALND: 207/232 Sampling: 190/234	o‐e: ‐4.66	13.25	0.7	(0.41 to 1.21)	0.20	5 years	HR calculated using log rank P ‐ figure 2, Chetty (2000)
Edinburgh 1	Axillary recurrence	ALND: /232 Sampling: /234	o‐e: ‐0.15	13.25	0.99	(0.58 to 1.69)	0.94	Up to 8 years	Log‐rank P value Tierney 2007 method 7, 8, 9 used Fig 3 Chetty (2000)
Edinburgh 1	Local recurrence in the breast	ALND: 14/232 Sampling: 15/234	o‐e: ‐0.10	7.24	0.99	(0.48 to 2.04)	0.97	Up to 8 years	Tierney 2007 method 7, 8, 9 used Table 2 & page 87 Chetty (2000)
Edinburgh 1	Distant recurrence	ALND: 29/232 Sampling: 29/234	Not available	Not available	Not available	Not available	NA	Up to 8 years	Table 2, Chetty (2000). Unable to estimate HR ‐ not included in analysis
E'dburgh Sample/Clear	Overall survival	ALND: 76/203 Sampling: 71/203	o‐e: ‐3.81	36.55	0.90	(0.62 to 1.25)	NA	13 years	Tierney 2007 method 3 used (using 1995 data – Clarke 2005 paper reports more deaths) Fig 1 and page 82 HR (CI) in Forrest et al (1995) ‐ inverted the HR
E'dburgh Sample/Clear	Distant metastases	ALND: 51/203 Sampling: 53/203	o‐e: 1.5	30.78	0.92	(0.67 to 1.35)	NA	13 years	Tierney 2007 method 3 used (using 1995 data), Fig 2 and HR (CI) page 82 in Forrest et al (1995), inverted the HR
E'dburgh Sample/Clear	Locoregional relapse (chest wall, axilla, supraclavicular)	ALND: 38/203 Sampling: 29/203	o‐e: ‐4.9	16.32	0.74	(0.46 to 1.20)	NA	13 years	Tierney 2007 method 3 used (using 1995 data) Method 3 Fig 3 from HR (CI), page 82 in Forrest et al (1995), inverted the HR
Genoa	Overall survival	ALND: 4/115 SLNB: 5/110	o‐e: 0.58	2.22	1.32	(0.35 to 4.92)	0.679	5 years	Log‐rank P value (Canavese 2009 ‐ fig 3) Tierney 2007 method 7 used Fig 3 KM curve gives P = 0.679. I assumed that was correct as it appears on the graph. The text value (page 20) may be a typo 0.697. HR are similar; CI differ
Genoa	Axillary recurrence	ALND: 1/115 SLNB: 0/110	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
Genoa	Breast cancer recurrence (local and contralateral recurrence, axillary and distant metastases)	ALND: 10/115 SLNB: 8/110	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
Genoa	5‐Year event‐free survival	ALND: 12/115 SLNB: 10/110	o‐e: ‐0.85	5.45	0.86	(0.37 to 1.98)	0.715	5 years	Log‐rank P value from Fig 2, Canavese (2009) method 7 Tierney 2007 used
GIVOM Sentinella	Overall survival	ALND: 14/352 SLNB: 21/345	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
GIVOM Sentinella	Disease‐free survival	ALND: 28/352 SLNB: 39/345	o‐e = 1.18	16.3	1.08		0.769	5 years	Method 7 Tierney 2007 used
GIVOM Sentinella	Axillary recurrence	ALND: 0/352 SLNB: 1/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
GIVOM Sentinella	Locoregional recurrence	ALND: 3/352 SLNB: 16/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
GIVOM Sentinella	Distant recurrence	ALND: 16/352 SLNB: 11/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
Guy's	Overall mortality (clinically node negative)	ALND: 178/241 No ALND (wide excision): 185/233	o‐e = 13.8	80.7	1.26	(0.98 to 1.63)	0.1	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Overall mortality (clinically node positive)	ALND: 82/85 No ALND (wide excision): 64/71	o‐e = 4.3	30.9	1.15	(0.81 to 1.64)	0.4	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Breast cancer mortality (clinically node negative)	ALND: 122/241 No ALND (wide excision): 142/233	o‐e = 13.8	58.8	‐	‐	0.07	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention Not included in meta‐analysis
Guy's	Breast cancer mortality (clinically node positive)	ALND: 53/85 No ALND (wide excision): 54/71	o‐e = 6.2	23.6	‐	‐	0.2	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention. Not included in meta‐analysis
Guy's	Isolated local recurrence (clinically node negative)	ALND: 35 events/3267 women‐years No ALND: 81 events/2383 women‐years	o‐e = 29.5	26.4	3.06	(2.09 to 4.48)	< .00001	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Isolated local recurrence (clinically node positive)	ALND: 17 events/873 women‐years No ALND: 31 events/519 women‐years	o‐e = 10.5	10.8	2.64	(1.46 to 4.80)	0.001	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Hammersmith	Overall survival	Radical: 35/76 Simple: 40/76	o‐e = 1.44	11.78	1.13	(0.64 to 2.00)	NA	8 years	Extracted from Fig 3, Burn et al (1968) Tierney 2007 method 10 on Simple is input as "research" and radical as "control". Min and max follow‐up input as 3‐96 months
Hammersmith	Local recurrence	Radical: 10/76 Simple: 11/76	NA	NA	NA	NA	NA	4‐9 years	Not included in meta‐analysis
Hammersmith	Mean time to recurrence	Radical: 15.7 months Simple: 25.9 months	NA	NA	NA	NA	NA	4‐9 years	Not included in meta‐analysis
IBCSG‐10‐93	Overall survival	ALND: 72/234 Surgery only: 71/239	o‐e = 1.76 (survival curves cross)	36.05	1.05	(0.76 to 1.46)	0.77	6‐7 years	HR reported on page 340 of IBCSG (2006), used Tierney 2007 method 3
IBCSG‐10‐93	Disease‐free survival	ALND: 92/234 Surgery only: 89/239	o‐e = 2.6	44.69	1.06	(0.79 to 1.42)	0.69	6‐7 years	HR reported on page 340 of IBCSG (2006), used Tierney 2007 method 3
IBCSG‐10‐93	Axilla recurrence (as first event)	ALND: 2/234 Surgery only: 6/239	NA	NA	NA	NA	NA	6‐7 years	Not included in meta‐analysis
Institut Bergonie	Overall survival (whole follow‐up period) ITT	no ALND: NR ALND: NR	o‐e = 6.42	7.04	2.49	90% CI (1.34 to 4.63)	NA	Whole follow‐up period (unclear how long that is)	HR reported on page 566 of Avril (2011), used Tierney 2007 method 3
Institut Bergonie	Event‐free survival (whole follow‐up period) ITT	no ALND: 44/297 ALND: 31/297	o‐e = 8.75	18.37	1.61	90% CI (1.1 to 2.37)	NA	Whole follow‐up period (unclear how long that is)	HR reported on page 566 of Avril (2011), used Tierney 2007 method 3
Institut Bergonie	Axillary event	Within 5 years: no ALND: 4/297 ALND: 0/310 After 5 years: no ALND: 2/297 ALND: 0/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Lymph node (excl axillary) event	Within 5 years: no ALND: 1/297 ALND: NA After 5 years: no ALND: 0/297 ALND: NA	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Breast/chest wall event	Within 5 years: no ALND: 5/297 ALND: 4/310 After 5 years: no ALND: 0/297 ALND: 8/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Metastatic event	Within 5 years: no ALND: 4/297 ALND: 1/310 After 5 years: no ALND: 2/297 ALND: 2/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Contralateral breast cancer	Within 5 years: no ALND: 2/297 ALND: 1/310 After 5 years: no ALND: 2/297 ALND: 1/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Other site cancer	Within 5 years: no ALND: 5/297 ALND: 5/310 After 5 years: no ALND: 5/297 ALND: 4/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Curie	Overall survival	RT: 43/331; ALND: 29/326	o‐e = 7	17.3	1.50	(0.94 to 2.40)	NA		Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
Institut Curie	Isolated local recurrence	RT: 39/2045 women‐years; ALND: 34/2126 women‐years	o‐e = 1.6	17.5	1.10	(0.69 to 1.75)	NA		Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
Institut Curie	Axilla recurrence	RT: 12/332; ALND: 5/326	o‐e = 3.86	3.53	3.93	‐	0.04		Table 2 in Louis‐Sylvestre (2004), method 7 in Tierney 2007
Institut Curie	Disease‐free survival	RT: 5 years : 82 (SD = 2.1)% 10 years : 72 (SD = 2.5)% 15 years : 65.5 (SD = 2.7)%	ALND: 5 years: 83.3 (SD 2)% 10 years: 72.6 (SD 2.5)% 15 years: 64.3 (SD 2.9)%.	NA	NA	NA	NA		o‐e cannot be extracted because P values not reported past NS in Table 2 in Louis‐Sylvestre (2004). Not included in meta‐analysis
Institut Curie	Metastases	RT: 5 years: 12.8 (SD 1.9)% 10 years: 21 (SD 2.3)% 15 years: 24.9 (SD 2.5)%	ALND: 5 years: 10.8 (SD 1.7)% 10 years: 18.3 (SD 2.2)% 15 years: 25.8 (SD 2.6)%	NA	NA	NA	NA		O‐e cannot be extracted because P values not reported past NS in Table 2 in Louis‐Sylvestre (2004). Not included in meta‐analysis
Malmo	Overall survival	ALND + RT: ?/97 Mastectomy alone: ?/98 (total event rate = 91)	o‐e = ‐4.19	22.75	0.83	(0.55 to 1.25)	0.38	15‐20 years	Using P = 0.38 reported on page 558 of Borgstrom (1994) and Tierney 2007 method 8. The o‐e is calculated on the basis of a total event rate of N = 91, and total N = 97 in the ALND + RT group and N = 98 in mastectomy alone group (i.e. intent‐to‐treat numbers), and using the only P value reported, which was for per‐protocol analysis that study authors stated did not differ from intention‐to‐treat analyses
Malmo	Chest wall recurrence	ALND + RT: 2/97 Mastectomy alone: 11/98	NA	NA	NA	NA	NA	15‐20 years	Cannot calculate o‐e. Not included in meta‐analysis
Manchester	Overall survival	Radical: 126/149 Simple + RT: 140/159	o‐e = 5.4	58.6	1.10	(0.85 to 1.42)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Manchester	Death from breast cancer	Radical: 100/149 Simple + RT: 112/159	o‐e = 2.8	46	1.06	(0.80 to 1.42)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Manchester	Local recurrence	Radical: 48 events/997 women‐years Simple + RT: 41 events/1113 women‐years	o‐e = ‐5.7	19.9	0.75	(0.48 to 1.17)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Milan	Death from any cause (OS)	ALND = 23/257 SLNB = 15/259	o‐e = ‐4.34	9.08	0.62	(0.32 to 1.19)	0.15	10 years	Log‐rank P (Tierney 2007 method 7); ALND is control
Milan	Breast cancer recurrence (local recurrence, regional lymph node metastases, distant metastases)	ALND = 26/257 SLNB = 23/259	o‐e = ‐2.25	12.02	0.83	(0.47 to 1.46)	0.52	10 years	Log‐rank P (Tierney 2007 method 7); ALND is control
Milan	Distant metastasis	ALND = 20/257 SLNB = 17/259	o‐e = ‐2.04	9.19	0.80	(0.42 to 1.53)	0.50	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan	Axillary metastasis	ALND = 0/257 SLNB = 2/259	o‐e = 0.97	0.50	6.96	(0.44 to 111.3)	0.17	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 8 and 9); ALND is control
Milan	Local recurrence	ALND = 4/257 SLNB = 4/259	o‐e = ‐0.12	2.00	0.94	(0.24 to 3.76)	0.93	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan	Supraclavicular metastasis	ALND = 2/257 SLNB = 0/259	o‐e = ‐1.02	0.50	0.13	(0.01 to 2.09)	0.15	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 8, 9); ALND is control
Milan	Contralateral breast cancer	ALND = 10/257 SLNB = 9/259	o‐e = ‐0.81	4.47	0.84	(0.34 to 2.07)	0.71	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan 2	Overall survival	ALND = 31/109 No ALND = 35/110	o‐e = ‐2.72	16.43	0.85	(0.52 to 1.37)		Median = 150 months	HR reported on page 922 of Martelli (2012). Using Tierney 2007 method 3 o Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status
Milan 2	Breast cancer deaths	ALND: 8/109 No ALND: 10/110	o‐e = 1.33	4.06	1.39	‐	‐	Median = 150 months	HR reported in Table 3 of Martelli (2012). Tierney 2007 method 3 o Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status. Not included in meta‐analysis
Milan 2	Axillary relapse	ALND: 0/109 No ALND: 4/110	NA	NA	NA	NA	NA	Median = 150 months	Table 2 of Martelli (2012), cannot calculate o‐e
Milan 2	Recurrence (ipsilateral breast tumour)	ALND: 4/109 No ALND: 7/110	NA	NA	NA	NA	NA	Median = 150 months	Table 2 of Martelli (2012), cannot calculate o‐e
Milan 2	Distant metastases	ALND: 9/109 No ALND: 9/110	o‐e = ‐2.68	5.93	0.64	(0.28 to 1.42)	NA	Median = 150 months	HR reported in Table 3 of Martelli (2012). Tierney 2007 method 3 Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status
Milan 3	Overall survival	10‐year ALND: 93.3% (95% CI 89.4‐95.8) no ALND: 91.5% (95% CI 87‐94.4)	o‐e = 1.76	12.33	1.15	(0.66 to 2.02)	P = .436	Median = 127.5 months	Agresti (2014) Figure 3A and Tierney 2007 method 11 Please note, the curves cross at the very end, also HR used for extraction of o‐e
Milan 3	Death from breast cancer	ALND: 17/272 no ALND: 15/245	NA	NA	NA	NA	P = 1.00	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Disease‐free survival	10‐year ALND: 92.4% (95% CI 88.5‐95.1) no ALND: 91.3% (95% CI 86.7‐94.3)	o‐e= ‐0.13	10.7	0.99	(0.54 to 1.8)	P = .97	Median = 127.5 months	Agresti (2014) Figure 3A andTierney 2007 method 11 Please note, the curves cross at the very end; also the HR used for extraction of o‐e
Milan 3	Distant metastases	ALND: 23/272 no ALND: 20/245	NA	NA	NA	NA	P = 1.00	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Axillary recurrence	ALND: 0/272; no ALND: 22/245	NA	NA	NA	NA	NA	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Local recurrence	ALND: 14/272 no ALND: 11/245	NA	NA	NA	NA	P = .839	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Contralateral breast cancer	ALND: 13/272 no ALND: 14/245	`NA	NA	NA	NA	P = .695	Median = 127.5 months	Not included in meta‐analysis
NSABP B‐04	Overall survival: node negative: ALND vs no ALND	ALND = 259/389 No ALND = 256/384	o‐e = ‐5	117.3	0.96	(0.80 to 1.15)	NA	15 years?	Taken from Clarke 2005 Lancet (Appendix web figure 9a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Overall survival: node negative: ALND vs no ALND + RT	ALND = 259/389 No ALND + RT = 271/386	o‐e = 8.6	122.2	1.07	(0.90 to 1.28)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Overall survival: node positive: ALND vs no ALND + RT	ALND = 244/301 No ALND + RT = 244/305	o‐e = 8.3	109.4	1.08	(0.89 to 1.30)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node negative: ALND vs no ALND	ALND = 35 events/3949 women‐years No ALND = 94 events/3335 women‐years	o‐e = 31.5	29.2	2.94	(2.05 to 4.23)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 9a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node negative: ALND vs no ALND + RT	ALND = 35 events/3949 women‐years No ALND + RT = 18 events/3896 women‐years	o‐e = ‐8.7	13	0.51	(0.30 to 0.88)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node positive: ALND vs no ALND + RT	ALND = 45 events/2268 women‐years No ALND + RT = 42 events/2025 women‐years	o‐e = ‐0.5	20.8	0.98	(0.64 to 1.50)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Disease‐free survival: node negative: ALND vs no ALND	ALND = 281/362 No ALND + RT = 287/365	o‐e = 9.36	138.3	1.07	(0.91 to 1.27)	0.39	25 years	FIsher (2008) page 568 (radical vs total mastectomy) Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Disease‐free survival: node negative: ALND vs no ALND + RT	ALND = 281/362 No ALND + RT = 292/352	o‐e = 8.3	142.39	1.06	(0.90 to 1.25)	0.49	25 years	FIsher (2008) page 568 (radical vs total mastectomy + RT) Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Disease‐free survival: node positive: ALND vs no ALND + RT	ALND = 254/292 No ALND + RT = 258/294	o‐e = 14.46	127.57	1.12	(0.94 to 1.33)	0.20	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Relapse‐free survival: node negative: ALND vs no ALND	ALND = 154/362 No ALND + RT = 182/365	o‐e = 10.17	77.61	1.14	(0.91 to 1.42)	0.27	25 years	FIsher (2008) page 568 Tierney 2007 method 3; calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Relapse‐free survival: node negative: ALND vs no ALND + RT	ALND = 154/362 No ALND + RT = 163/352	o‐e = ‐2.9	71.05	0.96	(0.76 to 1.21)	0.74	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Relapse‐free survival: node positive: ALND vs no ALND + RT	ALND = 178/292 No ALND + RT = 183/294	o‐e = 7.63	88.52	1.09	(0.89 to 1.35)	0.40	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Time to distant metastasis: node negative: ALND vs no ALND	ALND = 101/362 No ALND + RT = 107/365	o‐e = 8.44	88.52	1.1	(0.89 to 1.35)	0.39	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐04	Time to distant metastasis: node negative: ALND vs no ALND + RT	ALND = 101/362 No ALND + RT = 111/352	o‐e = 6.69	86.9	1.08	(0.88 to 1.34)	0.44	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐04	Time to distant metastasis: node positive: ALND vs no ALND + RT	ALND = 120/292 No ALND + RT = 127/294	o‐e = 5.98	88.41	1.07	(0.87 to 1.32)	0.51	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐32	Overall survival (all randomised participants, i.e. node+ and node‐)	ALND = 228 (deaths)/2807 SLN = 252 (deaths)/2804	10.32	119.7	1.09	(0.91 to 1.3)	0.35	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Disease‐free survival (all randomised participants, i.e. node+ and node‐)	ALND = 455/2807 SLN = 475/2804	4.6	232.39	1.02	(0.9 to 1.16)	0.72	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Local/regional recurrence (all randomised participants, i.e. node+ and node‐)	ALND = 121/2807 SLN = 112/2804	‐2.37	58.16	0.96	(0.74 to 1.24)	0.77	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Axillary recurrence (all randomised participants, i.e. node+ and node‐)	ALND = 6/2807 SLN = 14/2804	NA	NA	NA	NA	NA	10 years	o‐e cannot be calculated. Not included in meta‐analysis
NSABP B‐32	Overall survival (for SLN‐neg)	ALND = 219 (dead)/1975 SLN = 245 (dead)/2011	o‐e = 12.07	115.64	1.11	(0.93 to 1.33)	0.27	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Disease‐free survival (for SLN‐neg)	ALND = 456 (diseased)/1975 SLN = 465 (diseased)/2011	o‐e = 2.29	230.23	1.01	(0.89 to 1.15)	0.92	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Local regional recurrence	ALND = 85 (events)/1975 SLN = 80 (events)/2011	o‐e = ‐2.11	41.21	0.95	(0.7 to 1.29)	0.77	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Local recurrence in SLN‐negative participants	ALND = 54 (events)/1975 SLN = 49 (events)/2011	o‐e = ‐3.03	25.69	0.89	(0.6 to 1.31)	0.55	Mean = 95.6 months	From Krag (2010) page 930 using logrank P = 0.55 Tierney 2007 method 7
NSABP B‐32	Regional recurrence in SLN‐negative participants	ALND = 8 (events)/1975 SLN = 14 (events)/2011	o‐e = 2.77	5.09	1.72	(0.72 to 4.11)	0.22	Mean = 95.6 months	From Krag (2010) page 930 using log rank P = 0.22 Tierney 2007 method 7
NSABP B‐32	Distant recurrence in SLN‐negative patients	ALND = 55 (events)/1975 SLN = 64 (events)/2011	o‐e = 3.91	29.82	1.14	(0.8 to 1.64)		Mean = 95.6 months	From Krag (2010) Figure 4 Tierney 2007 method 3
Ostersund	Recurrence in the axilla	ALND: 0/57 Sampling: 1/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
Ostersund	Local recurrence	ALND: 4/57 Sampling: 1/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
Ostersund	Distant recurrence	ALND: 1/57 Sampling: 4/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
SE Scotland	Overall survival: node negative: ALND vs Simple + RT	ALND = 143/199 Simple + RT = 143/180	o‐e = 17.5	65.7	1.31	(1.02 to 1.66)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Overall survival: node positive: ALND vs Simple + RT	ALND = 72/89 Simple + RT = 77/93	o‐e = 6.3	34.1	1.20	(0.86 to 1.68)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Local isolated recurrence: node negative: ALND vs no ALND + RT	ALND = 26 events/2880 women‐years Simple + RT = 21 events/2204 women‐years	o‐e = ‐0.5	11.3	0.96	(0.53 to 1.71)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Local isolated recurrence: node positive: ALND vs no ALND + RT	ALND = 24 events/943 women‐years Simple + RT = 17 events/878 women‐years	o‐e = ‐2.9	9.8	0.74	(0.40 to 1.39)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Overall survival ‐ node negative	ALND: 56/101 Simple + RT to chest wall & axilla: 42/85	o‐e = ‐5.5	21.4	0.77	(0.51 to 1.18)	NA	15 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9a and 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Overall survival ‐ node positive	ALND: 13/17 Simple + RT to chest wall & axilla: 7/9	o‐e = ‐0.5	3.3	0.86	(0.29 to 2.53)	NA	15 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9b and 10b). then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Isolated local recurrence ‐ node negative	ALND: 15/510 py Simple + RT to chest wall & axilla: 13/483 py	o‐e = 0.0	6.7	1.00	(0.47 to 2.13)	NA	5 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9a and 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Isolated local recurrence ‐ node positive	ALND: 3/69 py Simple + RT to chest wall & axilla: 1/41 py	o‐e = ‐0.5	0.9	0.57	(0.07 to 4.53)	NA	5 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9b and 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Xu 2003	10‐year overall survival	Level I clearance: 75/93 ALND: 71/88	NA	NA	NA	NA	NA	10 years	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	10‐year disease‐free survival	Level I clearance: 72/93 ALND: 68/88	NA	NA	NA	NA	NA	10 years	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Breast cancer recurrence	Level I clearance: 19/93 ALND: 17/88	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Local recurrence	Level I clearance: 3.2% ALND: 2.3%	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Distant metastasis	Level I clearance: 19/93 ALND: 15/88	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis

Figures in bold were reported in the original publication; others were derived (see Notes column).

Table 2. Morbidity definitions

Study	Oedema	Shoulder function	Skin graft	Delayed healing	Activity	Attitude	Other	Notes
Guy's	Slight: 0‐2.5 cm Moderate: 2.5‐4.5 cm Severe > 4.5 cm Circumference of both arms measured 7.5 cm below the acromion, 18 cm above and 10 cm below the olecranon and at the wrist Presumably difference between arm circumference	Arm function: Good: uses arm freely Fair: cannot do usual tasks Poor: very unsatisfactory use of arm Appears to be assessed by patient questionnaire			Good: normal activity, back at work or resumed usual activities Fair: light work only because of operation; not resumed usual activities Poor: inactive. Assessed by patient questionnaire	Good: no complaints Fair: some complaints Poor: very unhappy about experience Assessed by patient questionnaire
ACOSOG Z0011	Lympheoedema (subjective) – according to patient self‐report or physician diagnosis Lympheoedema (objective): 2 cm or greater postop increase in ipsilateral arm circumference	Axillary paraesthesia – patient reported Brachial plexus injury – determined by physician on examining the patient
Addenbrookes	1. Mild oedema 2. Gross oedema (estimated by measuring the circumference of each arm with the arm extended at points 11 inches and 22 inches from the tips of the middle finger. An increase of 1 inch in the circumference of the arm on the side of the operation at either or both points was taken to indicate some degree of oedema)	Stiff shoulder	Need for skin graft	Sufficient to cause postponement of radiotherapy until at least 2 months after the operation. Although incidence of delayed healing varied between surgeons, each showed the same trend of higher incidence following a radical operation
Addenbrookes 2	Subjective lymphoedema: patient reported Objective lymphoedema: circumferential arm measurement at 4 cm intervals from the wrist (approximately 10 measurements) used to calculate arm volume. Volume corrected using measurements from contralateral arm	Range of movement measured by recording degrees of flexion, abduction and internal and external rotation using goniometer Sensory function tested using pinprick, light touch				Global Severity Index (GSI; low values better), Beck's Depression Inventory, Spielberger's State‐Trait anxiety, MAC, SF‐36 (measured psychological morbidity and quality of life)
ALMANAC	Change in ipsilateral arm volume at each follow‐up visit was expressed as a % increase from pretreatment value. Ratios of presurgery to postsurgery arm volumes were compared on a log‐transformed scale. The contralateral arm was used as a control for evaluations of arm volume Also patient rated as mild, moderate or severe	Assessed by goniometric measurement of arm movement (flexion, abduction, internal rotation and external rotation). Changes between visits calculated by subtraction The contralateral arm was used as a control for arm and shoulder function					QoL: Fact‐B+4 Anxiety: Spielberger STAI
Cardiff ‐ Local								No morbidity data
Cardiff ‐ St Mary's	Oedema of arm 72 cm	Restricted elevation 720 degrees				Measured but not reported	Axillary pain; numbness or paraesthesia on operated sides; aesthetic appearance of axillary scar
Edinburgh 1	Arm swelling measured by water displacement, circumference 15 cm above and below the olecranon process	Shoulder mobility assessed by measuring elevation through flexion, abduction, medial and lateral rotation					Shoulder muscle power assessed using graduated spring to measure flexion, extension, abduction and adduction of the shoulder joint
E'dburgh Sample/Clear	Arm welling (arm circumference 15 cm above and 10 cm below olecranon)	Objective assessment via adduction with internal rotation; abduction with external rotation, difference in height reached between treated and non‐treated arms by stretching above head, measurement of an abduction movement without shoulder rotation whilst lying on a flat, hard surface					Power (cm/kg) of pectoralis major by repeated lifting of a 3.5 kg weight as fast as possible over 45 seconds, comparing treated and untreated arm	Sample from study only, level B evidence
GIVOM Sentinella	Lympheodema was assessed by comparing the circumference of the operated vs the non‐operated arm at 15 cm above the epicondyle Unclear what difference in circumference constituted lymphoedema	Assessed by the surgeon by evaluating active and passive flexion, abduction, internal and external rotation, and classified on a scale 0 (normal mobility) to 3 (severe mobility) restriction Winged scapula reported as present/absent					Axillary and arm pain reported by patients on a scale from 0 (absent) to 3 (continuous/severe) Numbness assessed by the surgeon by comparing skin sensitivity in operated and non‐operated arms. Rated 0 (absent) to 3 (severe)
Guy's	Reports lymphoedema; categorised as none, slight, moderate and severe	Reports arm function as good, fair or poor			Reports activity as good, fair or poor	Reports attitude as good, fair or poor	Pts in no axillary surgery + RT arm reported fibrosis of breast and sometimes "marbling" of the overlying skin. Both occurred in <5% of cases
Hammersmith	Impairted function of the shoulder joint and swollen arm: no definitions given, but it is stated that the methodology included volumetric measurement of the upper limb and that an attempt was made to ally objective measurements with the patient’s subjective expression of discomfort or disability	Impairted function of the shoulder joint and swollen arm: no definitions given, but it is stated that the methodology included volumetric measurement of the upper limb and that an attempt was made to ally objective measurements with the patient’s subjective expression of discomfort or disability						In evaluating morbidity, attempts made to ally objective measurements with patient's subjective expression of disability or discomfort. Expectation that after RM, slight increase in volume of ipsilateral arm, or after RT, some discomfort and stiffness to shoulder, but these do not amount to morbidity
IBCSG‐10‐93	≥ 5% increase in arm circumference from baseline						QOL: A core questionnaire plus a surgical module specific to this trial. Four linear analogue scales on the core questionnaire were used: well‐being, mood, appetite and perceived adjustment/coping. After 1993, 6 additional scales were added: tiredness, hot flashes, nausea/vomiting, perceived social support, arm restriction and subjective health estimation. Surgical module measured swelling, numbness, weakness, pain, stiffness, performance of daily activities and global measure of arm/hand/shoulder/chest bother
IBCSG‐23‐01	No definitions for functional outcomes reported
Institut Bergonie	No definitions for functional outcomes reported
IPO‐P	An increase in arm volume was defined as an increase > 2 cm, comparing the circumference of the operated upper limb (at 3 points: the wrist, the midpoint of the forearm and the midpoint of the upper arm) with its non‐operated counterpart	Patients were asked to lift their operated arm (maximum possible abduction): abduction ≥ 90° was considered adequate; abduction < 90° was considered abnormal					Patients were asked: Is your arm painful in a resting position (yes/no)? Does the inside of your arm feel more numb (yes/no)?
Manchester
Milan	Arm swelling was assessed by comparing the circumference of treated and untreated arms 15 cm above the lateral epicondyle	Arm mobility was judged by asking the patient to rate restriction in movement on a scale 0 to 100 Numbness assessed by comparing skin sensitivity on inside and outside of the upper arm – classified as yes/no		Aesthetic appearance of scar judged by patient (rated good or bad)			Postoperative pain was evaluated as continuous (> 50% of the day), sporadic or absent
NSABP B‐04	Ipsilateral and contralateral measurement of arm circumference at 15 cm below the acromion process and 15 cm below the olecranon: An increase in arm circumference ≥ 2 cm in ipsilateral arm (below or above the elbow) indicated arm oedema
NSABP B‐32	Arm volume measured using volume of water displaced determined by the difference between treated and untreated arms (relative arm volume difference = [ipsilateral‐contralateral]/[contralateral] × 100%)	Arm mobility in degrees was determined by measuring the straight lateral abduction of both ipsilateral and contralateral arms using a standard orthopaedic goniometer to determine the angle between lateral chest wall and humerus (relative shoulder abduction deficit = [ipsilateral‐contralateral]/[contralateral] × 100%)					Numbness and tingling were assessed by self‐report by asking patients if they were currently experiencing any numbness or any tingling anywhere in ipsilateral and contralateral arms. OR of SLN compared with ALND Adverse events: no details reported
Ostersund	Arm volume measured using volume of water displaced. A cutoff of 10% increase in volume was used as the arbitrary cut point	Shoulder mobility (flexion, abduction and rotation) was determined with the help of a 360° scale placed on a wall with the centre at shoulder height
SNAC	Arm volume was estimated using 6 measures of arm circumference at 10 cm intervals starting 10 cm from the tip of the index finger. Upper limb swelling was expressed as percentage change in volume from baseline	Abduction and flexion measured using goniometer Arm morbidity measured using the 15‐item SSSS scale developed for the study, with each rated from 0 (no trouble at all) to 10 (worst I can imagine) and averaged to obtain overall score
SE Scotland	Increase in circumference of forearm by at least 3 cm	Failure to abduct the arm beyond a right angle
Xu 2003	Postoperative swelling: middle grade (diameter is 3–6 cm enlargement on the involved upper arm or forearm compared with the contralateral part)

Table 3. Morbidity data at each time point

Study	Outcome	Measurement	Follow‐up period 1	Follow‐up period 2	Notes
ACOSOG Z0011	Wound infection	Determined by treating physician	SLND: 11/371; SLND + ALND: 31/373
ACOSOG Z0011	Axillary seroma	Determined by treating physician	SLND: 21/371; SLND + ALND: 53/373
ACOSOG Z0011	Brachial plexus injury	Determined by treating physician	At 6 months: SLND: 3/415; SLND + ALND: 5/406 At 1 year: SLND: 0/415; SLND + ALND: 1/406
ACOSOG Z0011	Axillary paraesthesia	Patient reported	30 days: SLND: 43/371; SLND + ALND: 174/373	6 months: SLND: 35/288; SLND + ALND: 146/335
ACOSOG Z0011	Axillary paraesthesia	Patient reported	12 months: SLND: 24/268; SLND + ALND: 113/287
ACOSOG Z0011	Lymphoedema (objective)	Arm measurement	30 days: SLND: 17/272; SLND + ALND: 23/255	6 months: SLND: 21/271; SLND + ALND: 29/270
ACOSOG Z0011	Lymphoedema (objective)	Arm measurement	12 months: SLND: 14/226; SLND + ALND: 26/242
ACOSOG Z0011	Lymphoedema (subjective)	Patient reported/physician diagnosis	6 months: SLND: 19/339; SLND + ALND: 27/327	12 months: SLND: 12/268; SLND + ALND: 37/288
ACOSOG Z0011	Lymphoedema (subjective)	Patient reported/physician diagnosis	> 12 months: SLND: 14/253; SLND + ALND: 52/272
Addenbrookes	Mild oedema		Follow‐up was at least 12 months in most cases. ALND = 7/91; Simple = 5/113
Addenbrookes	Stiff shoulder		ALND = 6/91; Simple = 8/113
Addenbrookes	Skin graft	Need for skin graft	ALND = 4/91; Simple = 2/113
Addenbrookes	Delayed healing	Need to delay postoperative RT	ALND = 18/91; Simple = 7/113
Addenbrookes	Gross oedema	Arm measurement	ALND = 0/91; Simple = 0/113	ALND = 12/45; Simple = 6/53
Addenbrookes 2	Seroma		ALND: 33/155; SLNB: 20/143
Addenbrookes 2	Lymphoedema (objective)	Arm volume changes	12 months: ALND: mean (SE) = 56.4 (10.9); SLNB: mean (SE) = 18.6 (13.8), difference mean (SE) = 37.8 (17.6) Mean (1, 3, 6, 12 months): ALND: mean (SE) = 53.1 (8.1); SLNB: mean (SE) = 17.7 (9.2), difference mean (SE) = 35.4 (12.2)	Max: ALND: mean (SE) = 113.7 (9.7); SLNB: mean (SE) = 78.4 (12), difference mean (SE) = 35.3 (15.3)
Addenbrookes 2	Lymphoedema (subjective)	Patient reported	1 month: OR = 0.34 (95% CI 0.11 to 0.9); 3 months: OR = 0.4 (95% CI 0.16 to 0.94); 6 months: OR = 0.25 (95% CI 0.08 to 0.66)	12 months: OR = 0.36 (95% CI 0.15 to 0.86); mean: OR = 0.3 (95% CI 0.18 to 0.68)	Odds ratios: SLNB/ALND; i.e. lower favours SLNB
Addenbrookes 2	Paraesthesia		ALND: 130/155; SLNB: 92/140
Addenbrookes 2	Numbness		ALND: 115/155; SLNB: 68/143
Addenbrookes 2	Loss of pinprick		ALND: 118/155; SLNB: 77/140
Addenbrookes 2	Loss of light touch		ALND: 121/155; SLNB: 81/140
Addenbrookes 2	QOL (immediate postop)		Study authors note QOL scores were usually higher (better) in the SLND group and significantly so in the immediate postoperative period (P < 0.01). No significant effect of node positive/negative
Addenbrookes 2	MAC scale (12 months)		Study authors no significant difference in MAC scores during 1 year follow‐up. No significant effect of node positive/negative
Addenbrookes 2	BSI – somatisation (immediate postop)		SLND group scored lower (better) than ALND in the immediate postoperative period (P < 0.001)
Addenbrookes 2	Quality of life	GSI level	12 months: ALND: mean (SE, N) = 49.7 (1.1, 143); SLNB: mean (SE, N) = 48.4 (0.9, 134), difference mean (SE) = 1.3 (1.4)	OR for morbid GSI: study/control (95% CI) 0.55 (0.08 to 2.94)
Addenbrookes 2	Quality of life	SF‐36 (immediate postoperative)	Physical combined: ALND: mean (SD, N) = 38.6 (8.2, 143); SLNB: mean (SD, N) = 42.3 (10.4, 134), difference mean (95% CI) = 3.7 (1.2 to 6.1) Physical functioning: ALND: mean (SD, N) = 41.3 (9, 143); SLNB: mean (SD, N) = 44.5 (8.1, 134), difference mean (95% CI) = 3.2 (1.1 to5.4)	Vitality: ALND: mean (SD, N) = 48.2 (10.2, 143); SLNB: mean (SD, N) = 51.8 (9.8, 134), difference mean (95% CI) = 3.7 (1.1 to 6.2)
Addenbrookes 2	Shoulder movement (mean reduction)	Flexion, extension, abduction, internal rotation, external rotation	Flexion: ALND: mean (SD, N) = 13 (32.9, 141); SLNB: mean (SD, N) = 6.7 (15.6, 134), difference mean (95% CI) = 6.3 (0.1 to 12.6); Extension: ALND: mean (SD, N) = ‐1.5 (10.7, 139); SLNB: mean (SD, N) = ‐2.2 (8.1, 134), difference mean (95% CI) = 0.7 (‐1.5 to 3.3); Abduction: ALND: mean (SD, N) = 6.3 (11.5, 138); SLNB: mean (SD, N) = 3.1 (15.7, 132), difference mean (95% CI) = 3.2 (‐0.5 to 6.3)	Internal rotation: ALND: mean (SD, N) = 1.7 (12.7, 139); SLNB: mean (SD, N) = 0.3 (12, 134), difference mean (95% CI) = 1.4 (‐1.5 to 4.4); External rotation: ALND: mean (SD, N) = 2.9 (12.3, 139); SLNB: mean (SD, N) = 1.5 (11, 134), difference mean (95% CI) = 1.4 (‐1.5 to 4.4)
ALMANAC	Axillary drain usage		ALND: 359/453; SLNB: 75/449
ALMANAC	Infection rate of surgical wounds		ALND: 72/476; SLNB: 52/478
ALMANAC	Lymphoedema	Patient‐assessed; moderate/severe	1 month: ALND: 7/419; SLNB: 1/428 3 months: ALND: 12/395; SLNB: 4/417	6 months: ALND: 13/414; SLNB: 2/432 12 months: ALND: 10/403 SLNB: 4/412
ALMANAC	Lymphoedema	Mean (95% CI) change in arm vol compared with pretreatment	1 month: ALND = 1.022 (1.013‐1.032); SLNB = 1.003 (0.997‐1.01) 3 months: ALND = 1.044 (1.035‐1.053); SLNB = 1.019 (1.01‐1.028)	6 months: ALND = 1.058 (1.048‐1.069); SLNB = 1.022 (1.011‐1.032) 12 months: ALND = 1.061 (1.048‐1.074); SLNB = 1.028 (1.016‐1.039)
ALMANAC	Sensory loss	Median area of sensory loss (cm^2; range)	1 month: ALND = 40 (1‐489); SLNB = 32 (2‐254) 3 months: ALND = 47 (0‐1139); SLNB = 48 (0‐327)	6 months: ALND = 39 (0.4‐2827); SLNB = 32 (0‐201) 12 months: ALND = 35 (0.8‐1013); SLNB = 59 (0.2‐342)	Event rates for self‐assessed sensory loss also reported in Mansel 2006 for these follow‐up periods, but not extracted
ALMANAC	Intercostobrachial nerve damage	Clinician assessment; severe	1 month: ALND: 10/392; SLNB: 6/409 3 months: ALND: 10/373; SLNB: 4/397	6 months: ALND: 10/394; SLNB: 4/410 12 months: ALND: 5/384 SLNB: 5/400
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): flexion	1 month: ALND = 9.8; SLNB = 5.8 3 months: ALND = 3.7; SLNB = 2	6 months: ALND = 1.6; SLNB = 2 12 months: ALND = 0.1; SLNB = 2.7	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): abduction	1 month: ALND = 12.9; SLNB = 6.5 3 months: ALND = 4.2; SLNB = 1.9	6 months: ALND = 2.3; SLNB = 1.5 12 months: ALND = 1.9; SLNB = 2.5	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): external rotation	1 month: ALND = 1.2; SLNB = 0.7 3 months: ALND = 1.2; SLNB = 0.2	6 months: ALND = 1; SLNB = 0.6 12 months: ALND = 0.7; SLNB = 0.6	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): internal rotation	1 month: ALND = 0.9; SLNB = 0.4 3 months: ALND = 0.7; SLNB = 1	6 months: ALND = 0.8; SLNB = 0.2 12 months: ALND = 0.4; SLNB = 1.7	95% CI can also be extracted
ALMANAC	Quality of life	Measures: mean trial outcome index; trial outcome index reduced by ≥ 5 points from baseline (n/N); mean arm functioning subscale score; substantial arm swelling or tenderness (n/N); substantial numbness on ipsilateral side (n/N); mean FACT‐B+4 score			Means (95% CI) and event rates can be extracted for each time point (baseline, 1, 3, 6 and 12 months)
ALMANAC	State and trait anxiety				Mean and 95% CI can be extracted for each time point (baseline, 1, 3, 6 and 12 months)
Cardiff	Morbidity	Objective complaints: restricted elevation 720 degrees	Not stated: full axillary surgery, neg nodes = 25% (×2 = 7.47, P < 0.01); no axilary surgery, neg nodes = 0%; full axillary surgery + radical RT, positive nodes = 67%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Objective complaints: oedema of arm, 72 cm	Not stated: full axillary surgery, neg nodes = 46% (×2 = 6.02, P < 0.03); no axillary surgery, neg nodes = 15%; full axillary surgery + radical RT, positive nodes = 58%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Subjective complaints: limited arm movement	Not stated: full axillary surgery, neg nodes = 21%; no axillary surgery, neg nodes = 8%; full axillary surgery + radical RT, positive nodes = 8%; no axillary surgery + local RT = 21%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Subjective complaints: swollen arm	Not stated: full axillary surgery, neg nodes = 43%; no axillary surgery, neg nodes = 23%; full axillary surgery + radical RT, positive nodes = 58%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Edinburgh 1	Morbidity	Lateral shoulder rotation (mean (SE) difference (cm) from preoperative value (N))	6 months: Sampling + RT: 1.91 (SE = 0.56) (N = 64), sampling ‐ RT: 0.34 (SE = 0.59) (N = 59); ALND: 0.13 (SE = 0.39) (N = 132)	12 months: Sampling + RT: 1.75 (SE = 0.56) (N = 66), Sampling ‐ RT: 0.72 (SE = 0.62) (N = 55); ALND: 0.77 (0.4) (N = 128)	Figure 4, Chetty 2000 paper
Edinburgh 1	Morbidity	Lateral shoulder rotation (mean (SE) difference (cm) from preoperative value (N))	24 months: Sampling + RT: 1.57 (SE = 0.6) (N = 60), Sampling ‐ RT: ‐0.48 (SE = 0.65) (N = 52); ALND: 0.38 (SE = 0.43) (N = 117)	36 months: Sampling + RT: 2.19 (SE = 0.59) (N = 59), Sampling ‐ RT: 0.43 (SE = 0.64) (N = 50); ALND: 0.24 (SE = 0.43) (N = 110)	Figure 4, Chetty 2000 paper
Edinburgh 1	Morbidity	Arm volume (mean (SE) percentage of preoperative arm volume (N))	6 months: Sampling + RT: 100.69 (SE = 0.779) (N = 56), Sampling ‐ RT: 102.04 (SE = 0.766) (N = 58); ALND: 103.57 (SE = 0.519) (N = 126)	12 months: Sampling + RT: 100.95 (SE = 0.81) (N = 59), Sampling ‐ RT: 102.47 (SE = 0.85) (N = 54); ALND: 103.74 (SE = 0.57) (N = 119)	Figure 5, Chetty 2000 paper
Edinburgh 1	Morbidity	Arm volume (mean (SE) percentage of preoperative arm volume (N))	24 months: Sampling + RT: 100.84 (SE = 1.03) (N = 54), Sampling ‐ RT: 100.81 (SE = 1.06) (N = 51); ALND: 104.37 (SE = 0.73) (N = 108)	36 months: Sampling + RT: 100.01 (SE = 1.03) (N = 52), Sampling ‐ RT: 101.28 (SE = 1.07) (N = 48); ALND: 104.07 (SE = 0.73) (N = 103)	Figure 5, Chetty 2000 paper
E'dburgh Sample/Clear	Morbidity	Subjective arm	Not stated; full axillary surgery, positive node (Nil 8/12; intermittent 1/12; persistent 3/12); full axillary surgery, ‐negative node (nil 22/28; intermittent 1/28; persistent 5/28); Sample + RT, positive node (nil 17/28; intermittent 2/28; persistent 9/28); Sample, negative node (nil 23/26; intermittent 1/26; persistent 2/26)		Morbidity data to be included in discussion only; sample chosen from alphabetical pt list of patients free of local or systemic disease
E'dburgh Sample/Clear	Morbidity	Subjective mobility	Not stated; full axillary surgery, positive node (normal 12/12; reduced 0/12); full axillary surgery, negative node (normal 22/28; reduced 6/28); Sample + RT, negative node (normal 12/28; reduced 16/28); Sample, negative node (normal 24/26; reduced 2/26)		See comments in Aitken paper
E'dburgh Sample/Clear	Morbidity	Subjective interference with daily activities	Not stated; full axillary surgery, positive node (nil 12/12; occasional 0/12; severe 0/12); full axillary surgery, negative node (nil 24/28; occasional 4/28; severe 0/28); Sample + RT, positive node (nil 16/28; occasional 8/28; severe 4/28); Sample, negative node (nil 24/26; occasional 4/26; severe 0/26)		See comments in Aitken paper
E'dburgh Sample/Clear	Morbidity	Objective assessment ‐ shoulder joint mobility			See comments in Aitken paper
WSSA Glasgow	Psychological morbidity				Use in discussion only
GIVOM Sentinella	Lymphoedema	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.37 (0.2 to 0.7) 12 months: 0.48 (0.2 to 0.9)	18 months: 0.59 (0.3 to 1.2) 24 months: 0.52 (0.2 to 1.1)
GIVOM Sentinella	Shoulder movement restriction	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.47 (0.3 to 0.8) 12 months: 0.73 (0.4 to 1.4) 12 months: raw data extracted from graph (SLNB 17/336, ALND 23/341)	18 months: 0.62 (0.3 to 1.3) 24 months: 0.44 (0.2 to 1.0)
GIVOM Sentinella	Axillary/arm pain	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.52 (0.3 to 0.8) 12 months: 0.76 (0.5 to 1.3) 12 months: raw data extracted from graph (SLNB 30/336, ALND 39/341)	18 months: 0.84 (0.5 to 1.5) 24 months: 0.90 (0.5 to 1.6)
GIVOM Sentinella	Numbness	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.64 (0.4 to 0.9) 12 months: 0.53 (0.3 to 0.8) 12 months: raw data extracted from graph (SLNB 41/336, ALND 71/341)	18 months: 0.37 (0.2 to 0.6) 24 months: 0.54 (0.3 to 0.9)
GIVOM Sentinella	Winged scapula	Assessed by physician	Study authors report rate too low to analyse
GIVOM Sentinella	Health‐related quality of life: SF‐36 – physical component	Assessed by patients using validated questionnaires	No significant differences found between group means of SF‐36 physical component (Del Bianco, 2008)
GIVOM Sentinella	Health‐related quality of life: SF‐36 – mental component	Assessed by patients using validated questionnaires	No significant differences found between group means of SF‐36 mental component (Del Bianco, 2008)
GIVOM Sentinella	Health‐related quality of life: SF‐36 HRQOL domains	Assessed by patients using validated questionnaires	No significant differences found between groups on all HRQOL domains of SF‐36 (Zavagno, 2008)
GIVOM Sentinella	Health‐related quality of life: psychological general well‐being index	Assessed by patients using validated questionnaires	6, 12 months: significantly better PGWB general and anxiety domain scores in SLNB group than in ALND group (Del Bianco, 2008)	24 months: no significant differences between PGWB general and anxiety domain scores of both groups.(Del Bianco, 2008)
Guy's	Morbidity	Arm function	3 months: ALND: Good: 44/90, Fair: 41/90, Poor: 5/90; No ALND: Good: 59/77, Fair: 18/77, Poor: 0/77	15 months: ALND: Good: 83/100, Fair: 14/100, Poor: 3/100; No ALND: Good: 70/88, Fair: 17/88, Poor: 1/88	Sample only
Guy's	Morbidity	Lymphoedema	3 months: ALND: None: 18/93, Slight: 66/93, Moderate: 6/93, Severe: 3/93; No ALND: None: 36/81, Slight: 43/81, Moderate: 0/81, Severe: 2/81	15 months: ALND: None: 27/104, Slight: 71/104 Moderate: 6/104, Severe: 0/104; No ALND: None: 39/91, Slight: 52/91, Moderate: 0/91, Severe: 0/91	Sample only
Guy's	Morbidity	Activity	3 months: ALND: Good: 45/92, Fair: 46/92, Poor: 1/92; No ALND: Good: 62/80, Fair: 16/80, Poor: 2/80	15 months: ALND: Good: 85/101, Fair: 14/101, Poor: 2/101; No ALND: Good: 78/92, Fair: 13/92, Poor: 1/92	Sample only
Guy's	Morbidity	Attitude	3 months: ALND: Good: 81/92, Fair: 9/92, Poor: 2/92; No ALND: Good: 71/80, Fair: 7/80, Poor: 2/80	15 months: ALND: Good: 91/101, Fair: 8/101, Poor: 2/101; No ALND: Good: 87/92, Fair: 5/92, Poor: 0/92	Sample only
Hammersmith	Postoperative deaths		Radical: 0/95; Simple: 0/100
Hammersmith	Morbidity	Shoulder function	At 4‐year minimum follow‐up in survivors: Radical: 6/95; Simple = 18/100		Consequential morbidity, at time of publication Methodology not reported, all patients included
Hammersmith	Morbidity	Arm swelling (including volumetric measurement of upper limb)	At 4‐year minimum follow‐up in survivors: Radical: 7/95; Simple = 3/100		Consequential morbidity, at time of publication Methodology not reported, all patients included
IBCSG‐10‐93	Lymphoedema	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Arm circumference	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Performance of daily activities	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Arm pain	Physician reported	Baseline: ALND 5/175, surgery 8/194; 1st postoperative: ALND 38/164, surgery 12/168; 3 months: ALND 16/161, surgery 9/171; 6 months: ALND 17/174, surgery 11/177	9 months: ALND 21/160, surgery 8/164; 12 months: ALND 13/189, surgery 8/190; 18 months: ALND 14/173, surgery 7/183; 24 months: ALND 12/165, surgery 8/164
IBCSG‐10‐93	Restricted arm movement	Physician reported	Baseline: ALND 9/174, surgery 6/194; 1st postoperative: ALND 64/163, surgery 25/168; 3 months: ALND 23/161, surgery 10/170; 6 months: ALND 21/176, surgery 9/176	9 months: ALND 21/160, surgery 7/163; 12 months: ALND 19/188, surgery 6/187; 18 months: ALND 10/171, surgery 7/182; 24 months: ALND 12/165, surgery 7/164
IBCSG‐10‐93	QOL ‐ bothered scores	Patient reported	No significant differences at any time point (baseline, 1st postoperative, 3, 6, 9, 12, 18 and 24 months)
IBCSG‐10‐93	QOL ‐ arm movement scores	Patient reported	At 1st postoperative surgery alone, reported less restriction in use of their arm than ALND (P < .0001). Otherwise, no significant differences
IBCSG‐10‐93	QOL ‐ numbness scores	Patient reported	At 1st postoperative surgery alone, reported less severe postsurgery numbness than ALND (P < .0001). Otherwise, no significant differences
IBCSG‐10‐93	QOL ‐ coping scores	Patient reported	No significant differences at any time point (baseline, 1st postoperative, 3, 6, 9, 12, 18 and 24 months)
IBCSG‐23‐01	Postoperative infection	Physician assessed	Surgery alone: 0/467 ALND: 1/464
IBCSG‐23‐01	Sensory neuropathy	Physician assessed	Any: Surgery alone: 55/453 ALND: 82/447 Grade 3‐4: Surgery alone: 0/453 ALND: 1/447
IBCSG‐23‐01	Lymphoedema	Physician assessed	Defined as long term: Any: Surgery alone: 15/453 ALND: 59/447 Grade 3‐4: Surgery alone: 0/453 ALND: 3/447
IBCSG‐23‐01	Motor neuropathy	Physician assessed	Any: Surgery alone: 13/453 ALND: 37/447 Grade 3‐4: Surgery alone: 1/453 ALND: 3/447
Institut Bergonie	Arm fatigue	Unclear	Moderate/severe: no ALND: N = 4/258; ALND: N = 24/273
Institut Bergonie	Shoulder mobility	Unclear	Restricted somewhat or severely: no ALND: N = 5/257; ALND: N = 21/271
Institut Bergonie	Parasthesia	Unclear	Moderate/severe: no ALND: N = 6/258; ALND: N = 41/274
Institut Bergonie	Lymphoedema	Unclear	Minor/major difference: no ALND: N = 3/258; ALND: N = 29/275
Institut Bergonie	Other functional impairments	Unclear	Minor/major: no ALND: N = 12/263; ALND: N = 16/276
Institut Bergonie	Number of patients with functional impairments	Unclear	Minor: no ALND: N = 23/265; ALND: N = 78/278
IPO‐P	Upper limb circumference > 2 cm	Measured as per definition	6 months: Obs: 6/57; ALND: 10/49 12 months: Obs: 8/57; ALND: 15/49	24 months: Obs: 8/57; ALND: 14/49 48 months: Obs: 4/57; ALND: 19/49
IPO‐P	Pain at rest	Patient reported	6 months: Obs: 9/57; ALND: 9/49 12 months: Obs: 11/57; ALND: 14/49	24 months: Obs: 9/57; ALND: 10/49 48 months: Obs: 3/57; ALND: 7/49
IPO‐P	Parasthesias	Patient reported?	6 months: Obs: 10/57; ALND: 28/49 12 months: Obs: 6/57; ALND: 29/49	24 months: Obs: 5/57; ALND: 34/49 48 months: Obs: 6/57; ALND: 30/49
IPO‐P	Shoulder dysfunction	Measured as per definition	6 months: Obs: 5/57; ALND: 5/49 12 months: Obs: 4/57; ALND: 8/49	24 months: Obs: 0/57; ALND: 6/49 48 months: Obs: 2/57; ALND: 11/49
Milan	Morbidity	Axillary pain (sporadic/continuous)	6 months: ALND: 91/100; SNLB = 16/100	24 months: ALND: 39/100; SNLB = 8/100
Milan	Morbidity	Numbness/Parasthesia on operated side	6 months: ALND: 85/100; SNLB = 2/100	24 months: ALND: 68/100; SNLB = 1/100
Milan	Morbidity	Arm mobility, 80%‐100%	6 months: ALND: 73/100; SNLB = 100/100	24 months: ALND: 79/100; SNLB = 100/100
Milan	Morbidity	Arm mobility, 60%‐79%	6 months: ALND: 22/100; SNLB = 0/100	24 months: ALND: 18/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, 40%‐59%	6 months: ALND: 5/100; SNLB = 0/100	24 months: ALND: 2/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, 20%‐39%	6 months: ALND: 0/100; SNLB = 0/100	24 months: ALND: 1/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, < 20%	6 months: ALND: 0/100; SNLB = 0/100	24 months: ALND: 0/100; SNLB = 0/100
Milan	Morbidity	Aesthetic appearance of axillary scar: bad	6 months: ALND: 9/100; SNLB = 2/100	24 months: ALND: 15/100; SNLB = 0/100
Milan	Morbidity	Arm swelling < 1 cm difference in circumference	6 months: ALND: 44/100; SNLB = 11/100	24 months: ALND: 38/100; SNLB = 6/100
Milan	Morbidity	Arm swelling 1‐2 cm difference in circumference	6 months: ALND: 17/100; SNLB = 0/100	24 months: ALND: 25/100; SNLB = 1/100
Milan	Morbidity	Arm swelling >2 cm difference in circumference	6 months: ALND: 8/100; SNLB = 0/100	24 months: ALND: 12/100; SNLB = 0/100
Milan	Morbidity	Arm swelling, any	6 months: ALND: 69/100; SNLB = 11/100	24 months: ALND: 75/100; SNLB = 7/100
NSABP B‐04	Arm oedema	Arm swelling ≥ 2 cm difference in circumference	No. of patients with data: ALND: N = 577; no ALND + RT: N = 568 no ALND: N = 312 both node + and node‐ patients. Final measurement was 2 to 5 years after surgery Arm oedema recorded at least once: ALND: 58.1%; no ALND + RT: 38.2%; no ALND: 39.1% (P < 0.001) Oedema always: ALND: 3.6%; no ALND + RT: 0.9%; no ALND: 1% No measurement after first oedema: ALND: 9.2%; no ALND + RT: 5.8%; no ALND: 3.2% Oedema always after first oedema: ALND: 6.1%; no ALND + RT: 3.2%; no ALND: 2.6% Intermittent, final measurement oedema: ALND: 11.8%; no ALND + RT: 4.9%; no ALND: 8.6%; Total with oedema on final measurement: ALND: 30.7%; no ALND + RT: 14.8%; no ALND: 15.4% (P < 0.001)	Oedema once, then resolution: ALND: 15.9%; no ALND + RT: 15.3%; no ALND: 16.7% Intermittent, final measurement no oedema: ALND: 11.4%; no ALND + RT: 8.1%; no ALND: 7.1% Total with no oedema on final measurement (after at least 1 measurement of oedema): ALND: 27.3%; no ALND + RT: 23.4%; no ALND: 23.8 Arm oedema ≥ 4 cm difference in circumference recorded at least once: ALND: 21.5%; no ALND + RT: 11.4%; no ALND: 13.1%
NSABP B‐32	Adverse events (grade 3 or greater surgery related)	No details reported	ALND: 14/2788 SLN: 12/2800 Must include most of SLN positive and negative patients		Peri‐surgery
NSABP B‐32	Arm mobility/shoulder abduction deficit (objective)	Physician assessed	6 months: < 5%: ALND: 1299/1667; SLN: 1468/1744 5%‐10%: ALND: 218/1667; SLN: 176/1744 ≥ 10%: ALND: 150/1667; SLN: 99/1744
NSABP B‐32	Arm volume difference (objective)	Physician assessed	6 months: < 5%: ALND: 1187/1677; SLN: 1363/1759 5%‐10%: ALND: 277/1677; SLN: 236/1759 ≥ 10%: ALND: 211/1677; SLN: 158/1759	12 months: < 5%: ALND: 1170/1639; SLN: 1345/1705 5%‐10%: ALND: 252/1639; SLN: 215/1705 ≥ 10%: ALND: 216/1639; SLN: 147/1705	These data are also available for 18 and 30 months
NSABP B‐32	Arm volume difference (objective)	Physician assessed	24 months: < 5%: ALND: 1062/1517; SLN: 1184/1504 5%‐10%: ALND: 243/1517; SLN: 197/1504 ≥ 10%: ALND: 212/1517; SLN: 123/1504	36 months: < 5%: ALND: 990/1421; SLN: 1156/1459 5%‐10%: ALND: 227/1421; SLN: 194/1459 ≥ 10%: ALND: 203/1421; SLN: 109/1459	These data are also available for 18 and 30 months
NSABP B‐32	Tingling (subjective)	Self‐reported	6 months: ALND (N = 388/1693), SLN (N = 184/1766) 12 months: ALND (N = 305/1640), SLN (N = 158/1713) 18 months: ALND (N = 272/1566), SLN (N = 138/1638)	24 months: ALND (N = 236/1521), SLN (N = 137/1588) 30 months: ALND (N = 219/1448), SLN (N = 116/1502) 36 months: ALND (N = 193/1431), SLN (N = 110/1463)
NSABP B‐32	Numbness (subjective)	Self‐reported	6 months: ALND (N = 821/1693), SLN (N = 257/1769) 12 months: ALND (N = 679/1641), SLN (N = 216/1713) 18 months: ALND (N = 592/1567), SLN (N = 174/1638)	24 months: ALND (N = 554/1523), SLN (N = 157/1587) 30 months: ALND (N = 473/1450), SLN (N = 137/1504) 36 months: ALND (N = 445/1430), SLN (N = 119/1463)
NSABP B‐32	Shoulder abduction deficit ≥ 5% (in those with < 5% at baseline)	Physician assessed	6 months: ALND (N = 275/1449), SLN (N = 201/1519)
NSABP B‐32	Shoulder abduction deficit ≥ 5% (in those with < 5% at baseline)	Physician assessed	36 months: ALND (N = 314/1136), SLN (N = 192/1151)
NSABP B‐32	Numbness (in those with none at baseline)	Self‐reported	36 months: ALND (N = 407/1336), SLN (N = 103/1371)
NSABP B‐32	Tingling (in those with none at baseline)	Self‐reported	36 months: ALND (N = 175/1329), SLN (N = 90/1343)
Ostersund	Seroma	Patients with percutaneous aspiration in outpatient department	ALND: 17/50; sampling: 10/50		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Postoperative discharge (mL), median (range)		ALND: 250 (25‐1610); sampling: 130 (0‐1785)		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Duration of postop drainage (days) (median, range)		ALND: 4 (1‐11); sampling: 2.1 (1 ‐11)		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Arm volume increase	≥ 10%	ALND: 14/47; sampling: 0/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Subjective sensation of swelling in women without objective increase in arm volume	Any	ALND: 12/33; sampling: 9/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Shoulder mobility (mean decrease compared with baseline)		7.5° decrease for whole sample of 95 patients		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Axillary paraesthesia (impairment of sensibility in the axilla)		ALND: 17/48; sampling: 19/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Inner upper arm paraesthesia (impairment of sensibility in the inner upper arm)		ALND: 24/48; sampling: 4/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
SE Scotland	Delayed healing		ALND: 27/100; Simple + RT: 8/100
SE Scotland	Haematoma		ALND: 24/100; Simple + RT: 6/100
SE Scotland	Infection		ALND: 9/100; Simple + RT: 6/100
SE Scotland	DVT		ALND: 4/100; Simple + RT: 1/100
SE Scotland	Pulmonary embolism		ALND: 1/100; Simple + RT: 1/100
SE Scotland	Chest infection		ALND: 6/100; Simple + RT: 3/100
SE Scotland	Severe skin reaction		ALND: 0/100; Simple + RT: 5/100
SE Scotland	Nausea and vomiting		ALND: 0/100; Simple + RT: 2/100
SE Scotland	Tracheitis		ALND: 0/100; Simple + RT: 2/100
SE Scotland	Skin grafts		ALND: 10/100; Simple + RT: 0/100
SE Scotland	Arm oedema		ALND: 10/100; Simple + RT: 5/100
SE Scotland	Limitation of shoulder movement		ALND: 4/100; Simple + RT: 14/100
SNAC	Haematoma	Any	ALND: 30/539; SLNB: 38/544
SNAC	Seroma	Any	ALND: 195/539; SLNB: 93/544
SNAC	Infection	Any	ALND: 73/539; SLNB: 48/544
SNAC	Arm morbidity	Mean changes in arm morbidity (patient reported, overall summary average score of 15 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 7 (N = 457); SLNB: 4.4 (N = 456) 1 month: ALND: 2.2 (0.2); SLNB: 1.4 (0.15) 6 months: ALND: 1.1 (0.2); SLNB: 0.8 (0.15) 12 months: ALND: 1.05 (0.2); SLNB: 0.8 (0.15)	24 months: ALND: 1.05 (0.2); SLNB: 0.75 (0.15) 36 months: ALND: 1.05 (0.2); SLNB: 0.7 (0.15)
SNAC	Arm symptoms	Mean changes in arm symptoms (patient reported, average of 7 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 9.7 (N = 457); SLNB: 5.5 (N = 456) 1 month: ALND: 2.1 (0.2); SLNB: 1.2 (0.1) 6 months: ALND: 1.3 (0.15); SLNB: 0.8 (0.1) 12 months: ALND: 1.25 (0.15); SLNB: 0.7 (0.1)	24 months: ALND: 1.25 (0.15); SLNB: 0.7 (0.1) 36 months: ALND: 1.2 (0.2); SLNB: 0.65 (0.15)
SNAC	Arm swelling	Mean changes in arm swelling (patient reported, 1 item; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 7.3 (N = 457); SLNB: 3.4 (N = 456) 1 month: ALND: 1.25 (0.2); SLNB: 0.75 (0.15) 6 months: ALND: 0.9 (0.15); SLNB: 0.55 (0.1) 12 months: ALND: 0.95 (0.15); SLNB: 0.45 (0.1)	24 months: ALND: 1 (0.2); SLNB: 0.55 (0.15) 36 months: ALND: 1 (0.2); SLNB: 0.55 (0.15)
SNAC	Arm dysfunctions	Mean arm dysfunctions change (patient reported, average of 3 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 5.5 (N = 457); SLNB: 3.6 (N = 456) 1 month: ALND: 1.9 (0.15); SLNB: 1.35 (0.15) 6 months: ALND: 0.8 (0.1); SLNB: 0.65 (0.1) 12 months: ALND: 0.75 (0.1); SLNB: 0.6 (0.1)	24 months: ALND: 0.7 (0.1); SLNB: 0.55 (0.1) 36 months: ALND: 0.8 (0.1); SLNB: 0.5 (0.1)
SNAC	Arm disabilities	Mean arm disabilities (patient‐reported change, average of 4 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 3.4 (N = 457); SLNB: 2.9 (N = 456) 1 month: ALND: 2.2 (0.2); SLNB: 1.4 (0.15) 6 months: ALND: 0.75 (0.1); SLNB: 0.55 (0.1) 12 months: ALND: 0.65 (0.1); SLNB: 0.45 (0.1)	24 months: ALND: 0.6 (0.1); SLNB: 0.5 (0.1) 36 months: ALND: 0.7 (0.1); SLNB: 0.45 (0.1)
SNAC	Arm volume	Increase in arm volume (percentage change from clinician ratings from baseline; unclear if it is SEM or SD reported)	Average of measures taken at 6 and 12 months: ALND: 4.2% (N = 509); SLNB: 2.8% (N = 519) All patients: 1 month: ALND: 0.8% (0.4); SLNB: 0.9% (0.4), P = 0.67 6 months: ALND: 3.5% (0.8); SLNB: 2.4% (0.7), P = 0.02 12 months: ALND: 4.6% (0.8); SLNB: 3% (0.8), P = 0.001 Node‐negative patients: 1 month: ALND: 0.8% (0.4); SLNB: 0.3% (0.4), P = 0.16 6 months: ALND: 3.5% (0.8); SLNB: 1.9% (0.5), P = 0.004 12 months: ALND: 4.6% (0.8); SLNB: 2.2% (0.7), P = 0.001	All patients: 24 months: ALND: 5.8% (1); SLNB: 3.9% (0.7), P = 0.006 36 months: ALND: 5.8% (1); SLNB: 4.0% (1), P = 0.02 Node‐negative patients: 24 months: ALND: 5.8% (1); SLNB: 3% (0.7), P = 0.001 36 months: ALND: 5.8% (1); SLNB: 3.1% (1), P= 0.004
SNAC	Arm volume	Number with an increase in arm volume ≥ 15% (percentage change from clinician ratings from baseline)	All patients: 1 month: ALND: 5/544; SLNB: 3/544 6 months: ALND: 29/544; SLNB:21/544 12 months: ALND: 47/544; SLNB: 29/544 (P = 0.02) Node‐negative patients only: 1 month: ALND: 4/363; SLNB: 1/356 6 months: ALND: 16/363; SLNB: 9/356 12 months: ALND: 28/363; SLNB: 13/356 (P = 0.02)	All patients: 24 months: ALND: 81/544; SLNB: /544 (P = 0.001) 36 months: ALND: 82/544; SLNB: /544 (P = 0.01) Node‐negative patients only: 24 months: ALND: 47/363; SLNB: 25/356 (P = 0.01) 36 months: ALND: 49/363; SLNB: 25/356 (P = 0.006)
SNAC	Lateral abduction	Lateral abduction (change from clinician ratings from baseline; degrees; unclear if it is SEM or SD reported ‐ have assumed it is SEM for calculations)	Average of measures taken at 6 and 12 months (percentage change from baseline: ALND: 4.4% (N = 509); SLNB: 2.5% (N = 519) Node+ and node‐ patients (read off graph): Baseline: ALND: 158 (1); SLNB: 157 (1) 1 month: ALND: 131 (2); SLNB: 144 (2) 6 months: ALND: 150 (1); SLNB: 151 (1) 12 months: ALND: 150 (1); SLNB: 151 (1)	Node+ and node‐ patients (read off graph): 24 months: ALND: 151 (1); SLNB: 152 (1) 36 months: ALND: 150 (1); SLNB: 151 (1)
SNAC	Forward flexion	Forward flexion (degrees; unclear if it is SEM or SD reported ‐ have assumed it is SEM for calculations)	Node+ and node‐ patients (read off graph): Baseline: ALND: 157 (1); SLNB: 158 (1) 1 month: ALND: 137 (2); SLNB: 148 (1.5) 6 months: ALND: 150 (1); SLNB: 152 (1) 12 months: ALND: 151 (1); SLNB: 151 (1)	Node+ and node‐ patients (read off graph): 24 months: ALND: 152 (1); SLNB: 152 (1) 36 months: ALND: 152 (1); SLNB: 151 (1)
Xu 2003	Postoperative swelling (oedema)	Measurement of arm diameter	Level I clearance: 3/93 ALND: 7/88
Xu 2003	Involved upper limb disorder	Unclear	Level I clearance: 0/93 ALND: 0/88
Xu 2003	Cerebrovascular accident	Unclear	Level I clearance: 0/93 ALND: 2/88
Xu 2003	Cardiovascular events	Unclear	Level I clearance: 2/93 ALND: 1/88

We reported relative effects of treatments on time‐to‐event outcomes and noted that HRs less than 1.0 favour the 'less axillary surgery' arm, and HRs greater than 1.0 favour the 'more axillary surgery' arm. Similarly, for adverse event rates, ORs less than 1.0 favour the 'less axillary surgery' arm, and ORs greater than 1.0 favour the 'more axillary surgery' arm.

No axillary surgery versus full axillary surgery

Overall survival

All 10 trials comparing ALND versus no axillary surgery reported overall survival. The HR for death from any cause was 1.06 (95% confidence interval (CI) 0.96 to 1.17; 3849 participants; 10 studies; Analysis 1.1) with no statistically significant heterogeneity (I² = 26%; P = 0.20). We downgraded evidence for this outcome from high to moderate quality owing to imprecision: The confidence interval of the effect estimate includes both no difference between treatment groups and appreciable harm associated with no axillary surgery (summary of findings Table for the main comparison). For the single trial that did not use RT (NSABP B‐04), the HR was 0.96 (95% CI 0.80 to 1.15; 773 participants; one study; Analysis 1.1). For trials that used RT, the HR was 1.11 (95% CI 0.98 to 1.25; 3076 participants; nine studies; Analysis 1.1) with no statistically significant heterogeneity (I² = 24%; P = 0.23).

For the subgroup of studies that provided additional treatment to participants with histologically positive axillary nodes (Institut Bergonie; Institut Curie; Milan 3), no axillary surgery was associated with increased risk of overall mortality (HR 1.51, 95% CI 1.09 to 2.09; 1174 participants; three studies; Analysis 1.2.1) with no statistically significant heterogeneity (I² = 25%; P = 0.27).

For the subgroup of studies that did not provide additional treatment to participants with histologically positive axillary nodes (Addenbrookes; Guy's; Hammersmith; IBCSG‐10‐93; Malmo; Milan 2; NSABP B‐04), the HR for overall mortality was 1.02 (95% CI 0.92 to 1.13; 2675 participants; seven studies; Analysis 1.2.2) with no statistically significant heterogeneity (I² = 0%; P = 0.59).

For the subgroup of studies with adequate allocation concealment (Addenbrookes; IBCSG‐10‐93; Milan 2; Milan 3), the HR for death from any cause was 0.98 (95% CI 0.81 to 1.18; 1442 participants; four studies; Analysis 1.13.1) with no statistically significant heterogeneity (I² = 0%; P = 0.81).

Disease control in the axilla

Trials comparing full axillary surgery with no axillary surgery did not report disease control in the axilla.

Breast cancer recurrence

Local recurrence

Included studies not separately report time to local recurrence.

Locoregional recurrence

We were able to extract locoregional recurrence time‐to‐event data for four of the nine included trials. No axillary surgery was associated with increased risk of locoregional recurrence (with HR ranging from 1.10 to 3.06; 20,863 person‐years of follow‐up; four studies; Analysis 1.3) but heterogeneity was substantial (I² = 71%; P = 0.007); for this reason, we downgraded evidence for this outcome to moderate quality (summary of findings Table for the main comparison).

For the single trial that provided additional treatment to participants with histologically positive axillary nodes (Institut Curie), the HR for locoregional recurrence was 1.10 (95% CI 0.69 to 1.75; 4171 person‐years of follow‐up; one study; Analysis 1.4.1). For the remaining trials (Addenbrookes; Guy's; NSABP B‐04), which provided no specific additional treatment to participants with histologically positive axillary nodes, no axillary surgery was associated with increased risk of locoregional recurrence (HR 2.83, 95% CI 2.25 to 3.57; 16,692 person‐years of follow‐up; three studies) with no statistically significant heterogeneity (I² = 0%; P = 0.74).

In subgroup analyses of trials according to use of RT (Analysis 1.3), no axillary surgery was associated with increased risk of locoregional recurrence (HR ranging from 1.10 to 3.06; 13,579 person‐years of follow‐up; three studies; Analysis 1.3.2) but heterogeneity was substantial (I² = 75%; P = 0.008). For the single trial that did not use RT (NSABP B‐04), no axillary surgery was associated with increased risk of locoregional recurrence (HR 2.94, 95% CI 2.05 to 4.23; 7284 person‐years of follow‐up; one study; Analysis 1.3.1).

We judged allocation concealment as adequate in only one of the trials reporting locoregional recurrence (Addenbrookes). We were uncertain about whether no axillary surgery was associated with increased risk of locoregional recurrence in this trial (HR 1.84, 95% CI 0.79 to 4.28).

Distant metastasis

We were able to extract distant metastasis time‐to‐event data for two trials (Milan 2; NSABP B‐04). The HR for distant metastasis was 1.06 (95% CI 0.87 to 1.30; 946 participants; two studies; Analysis 1.5) with moderate heterogeneity (I² = 40%; P = 0.20).

One of the trials (Milan 2) had adequate allocation concealment, but its results indicate uncertainty about the relative rates of distant metastasis with the two treatment options (HR 0.64, 95% CI 0.28 to 1.42; 219 participants; one study).

Institut Curie reported the rate of metastases but provided insufficient detail for extraction of time‐to‐event outcomes. In this trial, at 15 years of follow‐up, the rate of metastasis was 24.9% for no axillary surgery versus 25.8% for axillary lymph node dissection (P reported as not significant).

Long‐term adverse events

Lymphoedema

Four of the included trials reported the rate of lymphoedema, defined as an increase in arm circumference, at 12 or more months after surgery (Addenbrookes; Guy's; Institut Bergonie; NSABP B‐04). The Addenbrookes, Guy's and Institut Bergonie trials used RT. NSABP B‐04 was a three‐arm trial, but we included the two "no radiotherapy" arms for this comparison. No axillary surgery was associated with decreased risk of lymphoedema at 12 or more months post surgery (OR 0.31, 95% CI 0.23 to 0.43; fixed‐effect model; 1714 participants; four studies; Analysis 1.6). We downgraded evidence for this outcome to low quality owing to substantial heterogeneity (I² = 69%; P = 0.02) and unclear blinding of the outcome assessment (summary of findings Table for the main comparison). A random‐effects model yielded a similar result (OR 0.22, 95% CI 0.08 to 0.57; random‐effects model; 1714 participants; four studies; I² = 69%; P = 0.02; Analysis 1.7).

Subgroup analysis of trials that did not provide additional treatment to participants with histologically positive axillary lymph nodes (Addenbrookes; Guy's, NSABP B‐04) revealed that no axillary surgery was associated with decreased risk of lymphoedema (OR 0.40, 95% CI 0.28 to 0.55; 1182 participants; three studies) and showed no important heterogeneity (I² = 0%; P = 0.54).

We judged allocation concealment as adequate in only one of the trials reporting lymphoedema (Addenbrookes). Its results were consistent with results of the pooled analysis (HR 0.35; 95% CI 0.12 to 1.03; 98 participants).

Arm or shoulder movement impairment

Five trials (Addenbrookes; Guy's; Hammersmith; IBCSG‐10‐93; Institut Bergonie), involving 1495 participants, reported impairment of arm or shoulder function at 12 or more months after surgery (Analysis 1.8). Results show considerable heterogeneity (I² = 78%; P = 0.001), with the OR for any impairment of function ranging from 0.24 to 3.26. We downgraded evidence for this outcome to very low quality owing to heterogeneity and unclear blinding of outcome assessment (summary of findings Table for the main comparison).

Differences between trials in the definitions of arm and shoulder impairment are a possible source of this heterogeneity. All trials provided RT, but in both Guy's and Hammersmith trials, the no axillary surgery group received more extensive RT than the ALND group.

Analysis restricted to trials with adequate allocation concealment (Addenbrookes; IBCSG‐10‐93) suggests fewer participants with arm or shoulder movement impairment in the no axillary surgery than in the ALND group (HR 0.46, 95% CI 0.23 to 0.93) but with potentially important heterogeneity (I² = 59%; P = 0.12).

Arm pain

One study reported arm pain. In IBCSG‐10‐93, the OR for arm pain at 12 or more months was 0.60 (95% CI 0.24 to 1.47; 379 participants; Analysis 1.9).

Paraesthesia

One study reported on paraesthesia. In Institut Bergonie, paraesthesia at 12 or more months after surgery was less likely in the no axillary surgery group (OR 0.14, 95% CI 0.06 to 0.32; 532 participants; Analysis 1.10).

Short‐term adverse events

One trial (Addenbrookes) reported acute adverse events (surgical complications).

Delayed healing

Delayed healing was less likely in the no axillary surgery group (OR 0.27, 95% CI 0.11 to 0.67; 204 participants; one study; Analysis 1.11).

Skin grafts

Skin grafts were less likely in the no axillary surgery group (OR 0.39, 95% CI 0.07 to 2.19; 204 participants; one study; Analysis 1.12).

Quality of life

IBCSG‐10‐93 was the only trial that measured quality of life outcomes; investigators reported no statistically significant differences in quality of life, bother and coping scores between treatment groups during the two years of postoperative follow‐up.

Psychological and psychosocial outcomes

The included studies did not report on these outcomes.

Axillary sampling versus full axillary surgery

Overall survival

Five trials (Cape Town; Cardiff; E'dburgh Sample/Clear; Edinburgh 1; Xu 2003) reported time to death from any cause, but we excluded Cardiff data from the meta‐analysis owing to non‐proportionality of hazard rates (i.e. survival curves cross at 12 years' follow‐up) and the published report provided insufficient detail to include Xu 2003. In the remaining three trials (Cape Town; E'dburgh Sample/Clear; Edinburgh 1), heterogeneity in the HR for overall mortality was substantial (HR 0.94, 95% CI 0.73 to 1.21; 967 participants; three studies; I² = 45%; P = 0.16; Analysis 2.1). We downgraded this evidence to low quality owing to substantial heterogeneity and serious imprecision (summary of findings Table 2).

Subgroup analysis of the two trials that provided RT (E'dburgh Sample/Clear; Edinburgh 1) yielded an HR of 0.84 (95% CI 0.64 to 1.11; 872 participants; two studies; Analysis 2.1) with no significant heterogeneity (I² = 0%; P = 0.44), and for the trial that did not use RT (Cape Town), an HR of 1.47 (95% CI 0.84 to 2.56; 85 participants).

We conducted no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

Disease control in the axilla

Included studies did not report disease control in the axilla, but two trials reported axillary recurrence (see below).

Breast cancer recurrence

Local recurrence

Five trials that performed six treatment comparisons reported local recurrence (Cape Town (1) and (2); Cardiff; Edinburgh 1; Ostersund; Xu 2003), but we could not extract time‐to‐event data from Ostersund and Xu 2003. The HR for local recurrence was 1.41 (95% CI 0.94 to 2.12; 1404 participants; three studies; Analysis 2.2) with no heterogeneity (I² = 0%; P = 0.91). In the Ostersund trial, one out of 54 participants in the axillary sampling arm experienced local recurrence compared with four of 57 participants in the ALND arm. In Xu 2003, local recurrence rates were 3.2% and 2.3% in the axillary sampling and ALND arms, respectively (181 participants; P value reported as greater than 0.05). We downgraded evidence for local recurrence to low quality on the basis of few events and serious imprecision (summary of findings Table 2). We performed no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

Axillary recurrence

Two trials reported axillary recurrence rates (Cape Town; Edinburgh 1), but we were able to extract time‐to‐event data only from Edinburgh 1, yielding an HR for axillary recurrence of 0.99 (95% CI 0.58 to 1.69; 466 participants; Analysis 2.3) with axillary lymph node sampling versus dissection. In Cape Town, rates of axillary recurrence were 8/52 for axillary lymph node sampling and 2/43 for ALND.

Locoregional recurrence

Two trials (Cape Town; E'dburgh Sample/Clear) reported locoregional recurrence, but we could extract time‐to‐event data only from E'dburgh Sample/Clear, yielding an HR for locoregional recurrence of 0.74 (95% CI 0.46 to 1.20; 406 participants; one study; Analysis 2.4). In the Cape Town trial, 19 of 52 participants in the axillary sampling group experienced locoregional recurrence compared with 11 of 43 in the ALND group.

Distant metastasis

Four trials reported distant metastasis (Cape Town; Cardiff; E'dburgh Sample/Clear; Xu 2003). We were able to extract time‐to‐event data only extracted from the Cardiff and E'dburgh Sample/Clear trials, but we did not include data from Cardiff in the meta‐analysis owing to the non‐proportionality of HRs. In E'dburgh Sample/Clear, the HR for distant metastasis was 1.05 (95% CI 0.74 to 1.49; 406 participants; Analysis 2.5). In the Cape Town trial, distant metastasis occurred at a rate of 13 of 52 participants in the axillary sampling group compared with 11 of 43 participants in the ALND group. In Xu 2003, distant metastasis rates were 19/93 and 15/88 in the axillary sampling and ALND arms, respectively (181 participants; P value reported as greater than 0.05).

Long‐term adverse events

Lymphoedema

Two trials reported on lymphoedema. In the Cardiff trial, lymphoedema at 12 or more months after surgery (defined as an increase in arm circumference) was less likely in the axillary sampling group than in the ALND group (OR 0.32, 95% CI 0.13 to 0.81; 85 participants; one study; Analysis 2.6). In Xu 2003, postoperative lymphoedema occurred in 3/93 participants in the axillary sampling group compared with 7/88 in the ALND group, but it was unclear at what time this measurement was taken.

Arm or shoulder movement impairment

One trial (Edinburgh 1) reported shoulder lateral rotation at 12‐months follow‐up, noting a relatively small decrease in range of movement when compared with baseline in both the axillary sampling and ALND groups (mean difference (MD) ‐0.05 cm, 95% CI ‐1.50 to 1.40; 191 participants; one study; Analysis 2.7).

Short‐term adverse events

Seroma

One trial collected data on seroma formation. In the Ostersund trial, seroma occurred at a rate of 10 of 50 participants in the axillary sampling group compared with 17 of 50 participants in the ALND group (OR 0.49, 95% CI 0.20 to 1.20; 100 participants; one study; Analysis 2.8).

Quality of life

The included studies did not report this outcome.

Psychological and psychosocial outcomes

The included studies did not report these outcomes.

Sentinel node biopsy versus full axillary surgery

Overall survival

Five trials reported overall mortality (ALMANAC; Genoa; GIVOM Sentinella; Milan; NSABP B‐32), but we were able to extract time‐to‐event data from only three studies (Genoa; Milan; NSABP B‐32). The HR for overall mortality was 1.05 (95% CI 0.89 to 1.25; 6352 participants; three studies; Analysis 3.1) with minimal heterogeneity (I² = 28%; P = 0.25). We rated evidence for overall mortality as moderate quality owing to imprecision. The confidence interval of the effect estimate included both no differences between treatment groups and appreciable harm associated with SLNB (summary of findings Table 3). In the ALMANAC trial, the overall mortality rate for the year after surgery was seven out of 478 women (1.5%) in the sentinel node group versus seven out of 476 women (1.5%) in the full axillary surgery group. In the GIVOM Sentinella trial, the overall mortality rate over the five years after surgery was 21 out of 345 women (6.1%) in the sentinel node group versus 14 out of 352 women (4.0%) in the full axillary surgery group.

We conducted no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

Disease control in the axilla

The included studies did not report disease control in the axilla, although five trials reported axillary recurrence (see below).

Breast cancer recurrence

Local recurrence

Data reveal uncertainty about the relative effectiveness of SLNB and ALND in terms of local recurrence (HR 0.94, 95% CI 0.24 to 3.77; 516 participants; one study; Milan; Analysis 3.2).

Axillary recurrence

Five trials, involving 7487 participants, reported axillary recurrence (ALMANAC; GIVOM Sentinella; Genoa; NSABP B‐32; Milan), but event rates were low, and we were able to extract time‐to‐event data only from Milan. Results derived from Milan suggest uncertainty about whether axillary recurrence is more likely with SLNB than with ALND (HR 6.96, 95% CI 0.44 to 111.25; 516 participants; one study; Analysis 3.3). In ALMANAC, the rate of axillary local recurrence during the first year after surgery was 1/478 (0.2%) in the SLNB group versus 4/476 (0.8%) in the ALND group. In GIVOM Sentinella, axillary recurrence rates over the five years after surgery were 1/345 (0.3%) in the SLNB group versus 0/352 (0%) in the ALND group. In Genoa, axillary recurrence rates were 0/110 (0%) in the SLNB group versus 1/115 (0.8%) in the ALND group. In NSABP B‐32, axillary recurrence rates were 14/2804 (0.5%) in the SLNB group versus 6/2807 (0.2%) in the ALND group.

We conducted no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

Locoregional recurrence

Two trials reported locoregional recurrence (GIVOM Sentinella; NSABP B‐32), but we were able to extract time‐to‐event data only from NSABP B‐32. Data reveal uncertainty about whether SLNB or ALND was more effective in terms of locoregional recurrence (HR 0.96, 95% CI 0.74 to 1.24; 5611 participants; one study; Analysis 3.4). In GIVOM Sentinella, locoregional recurrence rates were 16/345 (4.6%) in the SLNB group versus 3/352 (0.9%) in the ALND group.

Distant metastasis

Two studies reported distant metastases (GIVOM Sentinella; Milan), but we were able to extract time‐to‐event data only from Milan. The relative effectiveness of SLNB and ALND in terms of distant metastasis was uncertain (HR 0.80, 95% CI 0.42 to 1.53; 516 participants; one study; Analysis 3.5). In GIVOM Sentinella, distant metastasis rates were 11/3345 (3.2%) in the SLNB group versus 16/352 (4.5%) in the ALND group.

Long‐term adverse events

Lymphoedema

Four studies reported objectively measured lymphoedema at 12 or more months after surgery (ALMANAC; GIVOM Sentinella; Milan; SNAC). Investigators measured lymphoedema by using arm circumference (GIVOM Sentinella; Milan) or arm volume (ALMANAC; SNAC). Increased arm circumference at 12 months after surgery was less likely with SLNB than with ALND (OR 0.48, 95% CI 0.26 to 0.92; 677 participants ‐ Analysis 3.6 OR 0.04, 95% CI 0.00 to 0.60; 200 participants ‐ Analysis 3.6 and OR 0.60, 95% CI 0.37 to 0.96, 1088 participants ‐ Analysis 3.6) for the GIVOM Sentinella, Milan and SNAC trials, respectively. We did not pool results owing to heterogeneity (I² = 51%; P = 0.13), and we conducted no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

The ALMANAC trial reported the mean ratio in arm volume at baseline compared with 12 months after surgery. In the sentinel lymph node group, this was 1.03 (95% CI 1.02 to 1.04) compared with 1.06 (95% CI 1.05 to 1.07) in the ALND group (P = 0.096; two sided t‐test).

In ALMANAC, Addenbrookes 2 and SNAC, patient‐reported lymphoedema (of any severity) was less likely in the SLNB group than in the ALND group (OR 0.33, 95% CI 0.23 to 0.47; fixed‐effect model; 1903 participants; three studies; Analysis 3.7) with no heterogeneity (I² = 0%; P = 0.96). The random‐effects model produced the same result. We downgraded evidence on patient‐reported lymphoedema to moderate quality owing to incomplete follow‐up (summary of findings Table 3). Restricting this analysis to trials with adequate allocation concealment (ALMANAC and SNAC) yielded a similar result (OR 0.33, 95% CI 0.22 to 0.48; fixed‐effect model).

Shoulder or arm movement impairment

The Addenbrookes 2, ALMANAC and SNAC trials measured change in the range of shoulder movement from baseline to 12 months after surgery. Results showed no statistically significant differences between SLNB and ALND groups when change in the range of movement was compared from baseline to 12 months post surgery, for flexion (MD 1.55°, 95% CI ‐0.19° to 3.29°; 2257 participants; three studies; Analysis 3.8), abduction (MD ‐1.02°, 95% CI ‐2.79° to 0.75°; 2252 participants; three studies; Analysis 3.9), internal rotation (MD 0.50°; 95% CI ‐1.10° to 2.09°; 1227 participants; two studies; Analysis 3.10) or external rotation (MD ‐0.56°; 95% CI ‐2.21° to 1.09°; 1227 participants; two studies; Analysis 3.11). Except for external rotation, heterogeneity was substantial or considerable for all shoulder movement comparisons.

In two trials (GIVOM Sentinella and Milan), subjective arm movement impairment was less likely with SLNB than with ALND. This difference was statistically significant in the Milan trial (OR 0.02, 95% CI < 0.00 to 0.31; 200 participants; Analysis 3.12) but not in the GIVOM Sentinella trial (OR 0.74, 95% CI 0.39 to 1.41; 677 participants; Analysis 3.12), and heterogeneity in the pooled estimate was considerable (I² = 88%; P = 0.004). We downgraded evidence on subjective arm movement impairment to low quality owing to heterogeneity and lack of blinding (summary of findings Table 3). We conducted no sensitivity analysis for this outcome because all trials were at low risk of bias owing to allocation concealment.

The SNAC trial reported subjective arm disability rated on a scale from 0 (no trouble at al) to 10 (the worst I can imagine). At one year postoperatively, mean arm disability ratings were low in both groups: 0.65 (standard error (SE) 0.1) in the ALND group compared with 0.45 (SE 0.1) in the SLNB group.

Pain

Two trials reported pain at 12 or more months after surgery (GIVOM Sentinella; Milan). Pain was less likely to be reported in the sentinel lymph node group than in the axillary dissection group. This difference was statistically significant in the Milan trial (OR 0.14, 95% CI 0.06 to 0.31; 200 participants; Analysis 3.13) but not in the GIVOM Sentinella trial (OR 0.76, 95% CI 0.46 to 1.25; 677 participants; Analysis 3.13), and heterogeneity was considerable in the pooled estimate (I² = 92%; P = 0.0005). We downgraded evidence on pain to low quality owing to heterogeneity and lack of blinding (summary of findings Table 3).

Paraesthesia

Two trials reported paraesthesia at 12 or more months after surgery (Addenbrookes 2; Milan). Both trials found that paraesthesia was less likely in the sentinel lymph node group than in the axillary dissection group. For the Milan trial (OR < 0.00, 95% CI <0.00 to 0.04; 200 participants; Analysis 3.14) and the Addenbrookes 2 trial (OR 0.37, 95% CI 0.21 to 0.64; 295 participants; Analysis 3.14), heterogeneity was considerable in the pooled estimate (I² = 95%; P < 0.00001). We downgraded evidence on paraesthesia to low quality owing to heterogeneity and lack of blinding (summary of findings Table 3).

Numbness

Three trials reported numbness or sensory deficit at 12 or more months after surgery (Addenbrookes 2;ALMANAC; GIVOM Sentinella). All found that numbness was less likely in the SLNB group than in the ALND group (OR 0.43, 95% CI 0.34 to 0.54; 1799 participants; Analysis 3.15) with limited heterogeneity (I² = 20%; P = 0.29). Restricting this analysis to trials with adequate allocation concealment (ALMANAC; GIVOM Sentinella) yielded a similar result (OR 0.47, 95% CI 0.36 to 0.61).

Short‐term adverse events

Seroma

The Addenbrookes 2 and SNAC trials reported that seroma was less likely with SLNB than with ALND (OR 0.60, 95% CI 0.33 to 1.11; 298 participants; Analysis 3.16; OR 0.36; 95% CI 0.27 to 0.48; 1083 participants; Analysis 3.16 respectively) but with considerable heterogeneity (I² = 53%; P = 0.14).

Wound infection

The ALMANAC and SNAC trials reported that wound infection was less likely with SLNB than with ALND (OR 0.65, 95% CI 0.50 to 0.85; 2074 participants; Analysis 3.17).

Brachial plexus injury

The ALMANAC trial reported the rate of brachial plexus injury at six months postoperatively (OR 0.38, 95% CI 0.12 to 1.22; 804 participants).

Quality of life

We did not conduct statistical meta‐analysis because of differences in the scales used, but results from three trials (Addenbrookes 2; ALMANAC; GIVOM Sentinella) suggested that SLNB was associated with better quality of life, at least in the immediate postoperative period.

Addenbrookes 2 reported that quality of life scores were usually higher (better) in the SLND group than in the ALND group, and significantly so in the immediate postoperative period (P < 0.01).

ALMANAC measured a trial outcome index (TOI, derived from the sum of scores on physical and well‐being subscales and on breast cancer concerns subscales of the FACT‐B+4 (Functional Assessment of Cancer Therapy, Breast, for patients with lymphoedema questionnaire) before surgery and repeatedly in the following 18 months. Participants in the SLND group recovered more quickly to their baseline TOI value than those in the ALND group. This occurred at 12 months for the SLND group compared with 18 months for the ALND group (P < 0.01). Global quality of life (measured with the total FACT‐B+4 score) was significantly better in the SLND group than in the ALND group at most time points following surgery (at one month, P < 0.001; at three months, P = 0.04; at six months, P = 0.059; at 12 months, P = 0.024; at 18 months, P = 0.019). .

GIVOM Sentinella reported no significant differences between SLNB and ALND groups on the physical and health‐related quality of life components of the Short Form (SF)‐36 measure.

Psychological and psychosocial outcomes

Although three trials reported psychological outcomes, we did not pool their results owing to insufficient detail in reporting and differences in measurement scales used.

The Addenbrookes 2 trial reported no significant differences between SLND and ALND groups in Mental Adjustment to Cancer scores, depressive symptoms (measured on the Beck Depression Inventory) or state anxiety (measured by the Spielberger State/Trait Anxiety Inventory) during the first year after surgery.

ALMANAC reported that Spielberger State/Trait Anxiety Inventory scores were slightly lower (better) in the SLNB group than in the ALND group during the first year after surgery, but this difference was not statistically significant.

GIVOM Sentinella reported no significant differences between SLNB and ALND groups on the mental health‐related quality of life components of the SF‐36. Participants in the SLNB group scored significantly better than those in the ALND group in general and anxiety domains of the psychological well‐being measure within the first 12 months after surgery, but this difference was no longer statistically significant at two years after surgery.

Full axillary surgery with no radiotherapy versus no axillary surgery with radiotherapy

Overall survival

Four studies involving seven treatment comparisons reported that overall survival was reduced among participants treated with RT compared with those treated with ALND (HR 1.10, 95% CI 1.00 to 1.21; 2469 participants; Analysis 4.1) with no heterogeneity (I² = 0%; P = 0.63). We graded this evidence as high quality (summary of findings Table 4). Only one of the trials (SE Scotland) was at low risk of bias owing to allocation concealment; this trial was consistent with the pooled analysis showing reduced overall survival among patients treated with RT compared with those treated with ALND (HR 1.27, 95% CI 1.04 to 1.54).

Disease control in the axilla

Trials included in this comparison did not report disease control in the axilla.

Breast cancer recurrence

Local recurrence

Four studies involving seven treatment comparisons reported that local recurrence was less likely among participants treated with RT compared in those treated with ALND (HR 0.80, 95% CI 0.64 to 0.99; 22256 person‐years of follow‐up; four studies;Analysis 4.2) with no heterogeneity (I² = 0%; P = 0.63). We graded this evidence as high quality (summary of findings Table 4). Only one trial (SE Scotland) was at low risk of bias owing to allocation concealment; results showed uncertainty about whether local recurrence was less likely in patients treated with RT compared with those treated with ALND (HR 0.85, 95% CI 0.56 to 1.30).

Locoregional recurrence

The trials included for this comparison did not report locoregional recurrence.

Distant metastasis

One trial (NSABP B‐04) that performed two treatment comparisons reported that the HR for distant metastasis for RT alone versus ALND alone was 1.07 (95% CI 0.93 to 1.25; 1313 participants; Analysis 4.3).

Long‐term adverse events

Lymphoedema

One trial (SE Scotland) reported lymphoedema at 12 or more months after treatment and used a definition of 2 cm or greater increase in arm circumference. In the RT group, 5 out of 100 participants had lymphoedema compared with 10 out of 100 in the axillary surgery group (OR 0.47, 95% CI 0.16 to 1.44; 200 participants; Analysis 4.4).

Short‐term adverse events

Delayed healing, wound infection and skin graft

One trial (SE Scotland) involving 200 participants reported that acute adverse events ‐ delayed healing (OR 0.24, 95% CI 0.10 to 0.55; Analysis 4.5), wound infection (OR 0.65, 95% CI 0.22 to 1.89; Analysis 4.6), skin graft (OR 0.04, 95% CI 0.00 to 0.74; Analysis 4.7) and haematoma (OR 0.20, 95% CI 0.08 to 0.52; Analysis 4.8) ‐ were less likely with radiotherapy than with axillary surgery.

Quality of life

The trials included for this comparison did not report quality of life.

Psychological and psychosocial outcomes

The trials included for this comparison did not report psychological and psychosocial outcomes.

Less axillary surgery versus axillary lymph node dissection

Overall survival

When all trials were combined, the HR for overall mortality was 1.08 (95% CI 1.01 to 1.16, when HR > 1 favours ALND; 12,864 participants; 19 studies; Analysis 5.1) with no significant heterogeneity (I² = 12%; P = 0.30).

Trials comparing no axillary surgery (with or without RT) versus ALND reported increased mortality with less axillary surgery (HR 1.10, 95% CI 1.02 to 1.19; 5545 participants; 13 studies; I² = 6%; obtained by combining analyses 5.1.1 and 5.1.4 (not shown)), but trials comparing axillary sampling or SLNB versus ALND did not report increased mortality (HR 1.01, 95% CI 0.88 to 1.17; 7319 participants; six studies; obtained by combining analyses 5.1.2 and 5.1.3 (not shown)).

We performed subgroup analysis that was based on use of radiotherapy. Trials using RT in both treatment groups reported no difference in overall survival between less axillary surgery and more axillary surgery groups (HR 1.06, 95% CI 0.96 to 1.16; 10,075 participants; 13 studies; Analysis 5.2.1) with no important heterogeneity (I² = 28%; P = 0.15). Similarly, results showed no differences between groups for trials that did not use RT in either group (HR 1.00, 95% CI 0.85 to 1.19; 1093 participants; three trials; Analysis 5.2.3) with no important heterogeneity (I² = 8%; P = 0.34). Trials that used RT only in the less axillary surgery arm reported reduced overall survival for the less axillary surgery arm compared with the ALND arm (HR 1.10, 95% CI 1.00 to 1.21; 2469 participants; four trials; Analysis 5.2.2) with no heterogeneity (I² = 0%; P = 0.52).

We conducted subgroup analysis according to whether additional treatment was given to participants with histologically positive nodes and excluded trials in which one of the treatment arms received no axillary staging. Trials that provided additional treatment to participants with histologically positive axillary nodes (E'dburgh Sample/Clear; Edinburgh 1; Genoa; Milan) reported uncertainty whether less axillary surgery was the more effective treatment in terms of overall survival (HR 0.82, 95% CI 0.64 to 1.05; 1613 participants; four trials; Analysis 5.3) with no heterogeneity (I² = 0%; P=0.61). They also described uncertainty about relative effectiveness in the only trial (Cape Town) that did not provide additional treatment to those with histologically positive nodes (HR 1.47, 95% CI 0.84 to 2.56; 95 participants; Analysis 5.3).

Breast cancer recurrence

Local recurrence

Study results show uncertainty about whether local recurrence was reduced with less axillary surgery when compared with ALND (HR 0.90, 95% CI 0.75 to 1.09, when HR > 1 favours ALND; 24,176 participants; eight studies; Analysis 5.4).

Locoregional recurrence

Locoregional recurrence was more likely with less surgery than with ALND (HR 1.53, 95% CI 1.31 to 1.78, when HR > 1 favours ALND; 26,880 participant years of follow‐up; seven studies; Analysis 5.5).

Distant metastasis

Results reveal uncertainty about whether distant metastasis was more likely in patients treated with less axillary surgery than in those receiving ALND (HR 1.07, 95% CI 0.95 to 1.20, when HR >1 favours ALND; 2665 participants; five studies; Analysis 5.6).

Long‐term adverse effects

Lymphoedema (defined as an increase in arm circumference at 12 or more months postoperatively) was less likely with less axillary surgery than with ALND (OR 0.37, 95% CI 0.29 to 0.46; fixed‐effect model; 3964 participants; nine studies; I² = 52%; Analysis 5.7). The random‐effects model produced a similar result (OR 0.35, 95% CI 0.23 to 0.53; random‐effects model; 3964 participants; nine studies; I² = 52%).

Paraesthesia

Three trials reported paraesthesia at 12 or more months after surgery (Institut Bergonie; Addenbrookes 2; Milan). All trials found paraesthesia less likely in the less axillary surgery group than in the more axillary surgery group. For Institut Bergonie (OR 0.14, 95% CI 0.06 to 0.32; 532 participants; Analysis 5.8), for Milan (OR < 0.00, 95% CI <0.00 to 0.04; 200 participants; Analysis 5.8) and for Addenbrookes 2 (OR 0.37, 95% CI 0.21 to 0.64; 295 participants; Analysis 5.8); heterogeneity was considerable in the pooled estimate (I² = 91%; P < 0.0001).

Pain

Three trials reported pain at 12 or more months after surgery (IBCSG‐10‐93; GIVOM Sentinella; Milan). Pain was less likely to be reported in the less surgery group than in the more surgery group. This difference was statistically significant in the Milan trial (OR 0.14, 95% CI 0.06 to 0.31; 200 participants; Analysis 5.9) but not in the GIVOM Sentinella trial (OR 0.76, 95% CI 0.46 to 1.25; 677 participants; Analysis 5.9) or the IBCSG‐10‐93 trial (OR 0.60 95% CI 0.24 to 1.47; 379 participants; Analysis 5.9), and heterogeneity was considerable in the pooled estimate (I² = 84%; P < 0.0001).

Short‐term side effects

Delayed healing

The Addenbrookes and SE Scotland trials reported delayed wound healing was less likely with less surgery than with more surgery (OR 0.25, 95% CI 0.13 to 0.46; 404 participants; fixed‐effect model; two studies; I² = 0%; Analysis 5.10). The random‐effects model produced a similar result (OR 0.25, 95% CI 0.13 to 0.47; 404 participants; random‐effects model; two studies; I² = 0%).

Seroma

Seroma was less likely with less axillary surgery than with ALND (OR 0.40, 95% CI 0.32 to 0.52; 1481 participants; fixed‐effect model; three studies; I² = 14%; Analysis 5.11). The random‐effects model produced a similar result (OR 0.42, 95% CI 0.31 to 0.56; 1481 participants; random‐effects model; three studies; I² = 14%).

Wound infection

Wound infection was less likely with less axillary surgery than with ALND (OR 0.65, 95% CI 0.50 to 0.84; fixed‐effect model; 2274 participants; three studies; I² = 0%; Analysis 5.12). The random‐effects model yielded the same result.

Skin graft

Data reveal uncertainty about whether skin graft was less likely with less axillary surgery than with ALND (OR 0.15, 95% CI 0.04 to 0.57; fixed‐effect model; 404 participants; two studies; I² = 49%; Analysis 5.13). The random‐effects model suggested that skin graft was less likely with less axillary surgery than with ALND (OR 0.17, 95% CI 0.02 to 1.64; random‐effects model; 404 participants; two studies; I² = 49%).

Haematoma

The SNAC and SE Scotland trials reported haematoma. In the SNAC trial there were similar rates of haematoma in the less surgery group than more surgery group (OR 1.27, 95% CI 0.78 to 2.09; 1083 participants; Analysis 5.14). In the SE Scotland trial haematoma was less likely in the less surgery group than the more surgery group (OR 0.20, 95% CI 0.08 to 0.52; 200 participants; Analysis 5.14. There was considerable heterogeneity in the pooled estimate (I² = 91%; P = 0.0007).

Quality of life, psychological and psychosocial outcomes

Only trials comparing SLND versus ALND reported these outcomes, so we could perform no additional analyses.

Discussion

available in

Summary of main results

Risk of overall mortality was not increased when participants were treated with axillary sampling or sentinel lymph node biopsy (SLNB) versus axillary lymph node dissection (ALND). Treatment omitting all axillary surgery was associated with increased risk of overall mortality compared with ALND, but this was noted only in trials comparing radiotherapy (RT) alone versus ALND.

Axillary lymph node dissection was associated with increased risk of lymphoedema and surgical adverse events compared with less axillary surgery.

Overall completeness and applicability of evidence

We found no trials that performed the following comparisons: sentinel node biopsy versus axillary sampling, no axillary surgery versus axillary sampling and no axillary surgery versus sentinel node biopsy.

Adverse event data were limited, particularly for older trials comparing no surgery, RT or axillary sampling versus ALND. Quality of life data were limited to three trials. Sentinel lymph node trials provided limited data on long‐term overall survival and breast cancer recurrence; these trials were often designed to compare quality of life and adverse effects. Substantial heterogeneity in adverse event trial results was often due to differences among adverse event definitions between trials.

Some trials reported data in a way that precluded inclusion in the time‐to‐event meta‐analysis, and although we contacted study authors, we obtained no additional data.

Applicability of some of the comparisons in this review to current breast cancer practice is questionable, particularly for comparisons involving no axillary surgery. Use of adjuvant therapies differs between current practice and many of the included trials ‐ more effective adjuvant systemic therapies are available today. Similarly, RT regimens used in the older trials are most likely less effective and are associated with more side effects:

Patients with breast cancer today are likely to differ from those who participated in older trials, and breast cancer is more likely to be detected at an earlier stage.

Quality of the evidence

The included studies were at low or unclear risk of selection bias. Selection bias was typically unclear because trial publications did not fully report methods of random sequence generation or allocation concealment used and study authors did not reply when we contacted them to request additional information about conduct of the trial. We performed sensitivity analyses for trials with adequate allocation concealment and found that these results generally were consistent with findings of the main analyses.

Risk of attrition bias tended to be lower for survival and for breast cancer recurrence than for adverse events. This sometimes occurred because adverse event assessments were done for a subset of the trial population. This subgroup of participants assessed for adverse events could be systematically different from the trial population as a whole, especially in the case of assessment for long‐term adverse events when patients may have died or may have been too sick to participate.

The included trials did not include blinding (and it was probably infeasible), but this was considered a source of bias only for outcomes with potential subjectivity in measurement (i.e. breast cancer recurrence and adverse events). Detection bias could lead to overestimation of adverse events in patients with more extensive axillary surgery. Similarly, patients receiving less extensive axillary surgery could be checked more carefully for breast cancer recurrence.

For these reasons, we downgraded the quality of the evidence for adverse effects (summary of findings Table for the main comparison; summary of findings Table 2; summary of findings Table 3; summary of findings Table 4).

Potential biases in the review process

The meta‐analyses of time‐to‐event outcomes conducted for this review used the fixed‐effect model because only fixed‐effect meta‐analytical methods are available in RevMan for ‘O–E' and 'Variance’ outcomes. This could affect interpretation of results by yielding narrower confidence intervals for the pooled hazard ratio in the presence of heterogeneity than would be obtained with a random‐effects model. This is particularly the case for Analysis 5.1 (which compares overall survival with more surgery vs less surgery), in which the underlying assumption of the fixed‐effect model is unlikely to be true, given the different types of interventions and patient populations included.

Agreements and disagreements with other studies or reviews

Kell 2010 reported a meta‐analysis of seven trials of SLNB versus axillary clearance (Addenbrookes 2; ALMANAC; GIVOM Sentinella; Milan; SNAC; ACOSOG Z0011; NSABP B‐32). Compared with axillary clearance, SLNB was associated with reduced risk of postoperative wound infection (odds ratio (OR) 0.58, 95% confidence interval (CI) 0.42 to 0.80), of postoperative seroma (OR 0.40, 95% CI 0.31 to 0.51) and of arm swelling at six months postoperatively (OR 0.30, 95% CI 0.14 to 0.66). These results are consistent with findings of the current review.

Wang 2011 also analysed trials examining the sentinel lymph node versus axillary clearance (Addenbrookes 2; ALMANAC; GIVOM Sentinella; Genoa; Milan; SNAC; ACOSOG Z0011; NSABP B‐32). Comparison of SLNB with ALND revealed no statistically significant difference in overall survival (hazard ratio (HR) 1.07, 95% CI 0.90 to 1.27) or regional lymph node recurrence (OR 1.65, 95% CI 0.77 to 3.56). Postoperative complications were less likely with SLNB than with ALND, including lymphoedema (OR 0.24, 95% CI 0.11 to 0.53), numbness (OR 0.19, 95% CI 0.11 to 0.33), infection (OR 0.50, 95% CI 0.36 to 0.70) and seroma (OR 0.39, 95% CI 0.31 to 0.49). These results are consistent with findings of the current review.

In the Early Breast Cancer Trialists Group, meta‐analysis of individual participant data (Clarke 2005) revealed that axillary clearance versus effective axillary RT involved little absolute difference (< 10%) in five‐year risk of local recurrence, as well as little difference in breast cancer mortality (when combined with other local treatment comparisons). The current review observed an increase in overall mortality with RT with no axillary surgery compared with axillary clearance, but the absolute difference at five years was on the order of a few percent (summary of findings Table 4), and had a random‐effects model been possible, greater uncertainty would surround this estimate.

Figure 1

Study flow diagram.

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Analysis 1.1

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 1 All‐cause mortality (radiotherapy subgroups).

Analysis 1.2

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 2 All‐cause mortality (extra treatment for positive node subgroups).

Analysis 1.3

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 3 Locoregional recurrence (radiotherapy subgroups).

Analysis 1.4

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 4 Locoregional recurrence (extra treatment for positive‐node subgroups).

Analysis 1.5

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 5 Distant metastasis.

Analysis 1.6

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 6 Lymphoedema (≥ 12 months postop) ‐ fixed‐effect model.

Analysis 1.7

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 7 Lymphoedema (≥ 12 months postop) ‐ random‐effects model.

Analysis 1.8

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 8 Arm or shoulder movement impairment (≥ 12 months postop).

Analysis 1.9

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 9 Pain (≥ 12 months postop).

Analysis 1.10

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 10 Paraesthesia (≥ 12 months postop).

Analysis 1.11

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 11 Delayed healing.

Analysis 1.12

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 12 Skin graft.

Analysis 1.13

Comparison 1 No axillary surgery versus full axillary surgery, Outcome 13 All‐cause mortality (allocation concealment subgroups).

Analysis 2.1

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 1 All‐cause mortality.

Analysis 2.2

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 2 Local recurrence.

Analysis 2.3

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 3 Axillary recurrence.

Analysis 2.4

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 4 Locoregional recurrence.

Analysis 2.5

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 5 Distant metastasis.

Analysis 2.6

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 6 Lymphoedema. Increase in arm circumference (≥ 12 months postop).

Analysis 2.7

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 7 Shoulder lateral rotation (12 months postop).

Analysis 2.8

Comparison 2 Axillary sampling versus full axillary surgery, Outcome 8 Seroma.

Analysis 3.1

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 1 All‐cause mortality.

Analysis 3.2

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 2 Local recurrence.

Analysis 3.3

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 3 Axillary recurrence.

Analysis 3.4

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 4 Locoregional recurrence.

Analysis 3.5

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 5 Distant metastasis.

Analysis 3.6

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 6 Lymphoedema. Increase in arm circumference (≥ 12 months postop).

Analysis 3.7

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 7 Lymphoedema. Patient reported (at 12 or more months postop).

Analysis 3.8

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 8 Shoulder flexion (12 months postop).

Analysis 3.9

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 9 Shoulder abduction (12 months postop).

Analysis 3.10

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 10 Shoulder internal rotation (12 months postop).

Analysis 3.11

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 11 Shoulder external rotation (12 months postop).

Analysis 3.12

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 12 Subjective arm movement impairment (≥ 12 months postop).

Analysis 3.13

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 13 Pain (≥ 12 months postop).

Analysis 3.14

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 14 Paraesthesia (≥ 12 months postop).

Analysis 3.15

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 15 Numbness (≥ 12 months postop).

Analysis 3.16

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 16 Seroma.

Analysis 3.17

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 17 Wound infection.

Analysis 3.18

Comparison 3 Sentinel node biopsy versus full axillary surgery, Outcome 18 Brachial plexus injury at 6 months postop.

Analysis 4.1

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 1 All‐cause mortality.

Analysis 4.2

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 2 Local recurrence.

Analysis 4.3

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 3 Distant metastasis.

Analysis 4.4

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 4 Lymphoedema. Increase in arm circumference (≥ 12 months postop).

Analysis 4.5

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 5 Delayed healing.

Analysis 4.6

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 6 Wound infection.

Analysis 4.7

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 7 Skin graft.

Analysis 4.8

Comparison 4 Radiotherapy versus full axillary surgery, Outcome 8 Haematoma.

Analysis 5.1

Comparison 5 Less surgery versus ALND, Outcome 1 All‐cause mortality.

Analysis 5.2

Comparison 5 Less surgery versus ALND, Outcome 2 All‐cause mortality (radiotherapy subgroups).

Analysis 5.3

Comparison 5 Less surgery versus ALND, Outcome 3 All‐cause mortality (additional treatment for histologically positive nodes).

Analysis 5.4

Comparison 5 Less surgery versus ALND, Outcome 4 Local recurrence.

Analysis 5.5

Comparison 5 Less surgery versus ALND, Outcome 5 Locoregional recurrence.

Analysis 5.6

Comparison 5 Less surgery versus ALND, Outcome 6 Distant metastasis.

Analysis 5.7

Comparison 5 Less surgery versus ALND, Outcome 7 Lymphoedema. Increase in arm volume at 12 months postop.

Analysis 5.8

Comparison 5 Less surgery versus ALND, Outcome 8 Paraesthesia (≥ 12 months postop).

Analysis 5.9

Comparison 5 Less surgery versus ALND, Outcome 9 Pain (≥ 12 months postop).

Analysis 5.10

Comparison 5 Less surgery versus ALND, Outcome 10 Delayed healing.

Analysis 5.11

Comparison 5 Less surgery versus ALND, Outcome 11 Seroma.

Analysis 5.12

Comparison 5 Less surgery versus ALND, Outcome 12 Wound infection.

Analysis 5.13

Comparison 5 Less surgery versus ALND, Outcome 13 Skin graft.

Analysis 5.14

Comparison 5 Less surgery versus ALND, Outcome 14 Haematoma.

Summary of findings for the main comparison. No axillary surgery compared with full axillary surgery for operable primary breast cancer

No axillary surgery compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: no axillary surgery Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	No axillary surgery
All‐cause mortality	92% overall survival at 5 years^a	92% overall survival at 5 years (91% to 93%)	HR 1.06 (0.96 to 1.17)	3849 (10 studies)	⊕⊕⊕⊝ moderate^b
Locoregional recurrence	86% locoregional recurrence‐free survival at 5 years^c	71% locoregional recurrence‐free survival at 5 years (66% to 76%)	HR 2.35 (1.91 to 2.89)	20,863^d (5 studies)	⊕⊕⊕⊝ moderate^e
Lymphoedema Increase in arm circumference Follow‐up: 1 or more years	236 per 1000	87 per 1000 (66 to 117)	OR 0.31 (0.23 to 0.43)	1714 (4 studies)	⊕⊕⊝⊝ low^e,f
Arm or shoulder movement impairment Follow‐up: 1 or more years	91 per 1000	67 per 1000 (47 to 95)	OR 0.72 (0.49 to 1.05)	1495 (5 studies)	⊕⊝⊝⊝ very low^f,g
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk is taken from full axillary surgery arm of Institut Curie. ^bConfidence interval around the effect estimate includes both no effect and appreciable harm associated with no axillary surgery. ^cAssumed risk is taken from full axillary surgery arm of Institut Curie, local or axillary recurrence rates. ^dPerson‐years of follow‐up. ^eSubstantial heterogeneity (I² > 50%). ^fUnclear blinding of outcome assessment. ^gConsiderable heterogeneity (I² > 75%).

Summary of findings for the main comparison. No axillary surgery compared with full axillary surgery for operable primary breast cancer

Summary of findings 2. Axillary sampling compared with full axillary surgery for operable primary breast cancer

Axillary sampling compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: axillary sampling Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Full axillary surgery	Axillary sampling
All‐cause mortality	82% overall survival at 5 years^a	83% overall survival at 5 years (79% to 87%)	HR 0.94 (0.73 to 1.21)	967 (3 studies)	⊕⊕⊝⊝ low^b,c
Local recurrence	85% local recurrence‐free survival at 5 years^d	80% local recurrence free survival at 5 years (71% to 86%)	HR 1.41 (0.94 to 2.12)	1404 (3 studies)	⊕⊕⊝⊝ low^e,f
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk is taken from full axillary surgery arm of E'dburgh Sample/Clear. ^bSubstantial heterogeneity. ^cConfidence interval for the effect includes both appreciable benefit and harm with axillary sampling. ^dAssumed risk taken from full axillary surgery arm of Cardiff. ^eNo blinding of outcome assessment or blinding not reported. ^fConfidence interval for effect includes both no difference and appreciable harm with axillary sampling. Low number of events.

Summary of findings 2. Axillary sampling compared with full axillary surgery for operable primary breast cancer

Summary of findings 3. Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer

Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: sentinel node biopsy Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	Sentinel node biopsy
All‐cause mortality	96% overall survival at 5 years^a	96% overall survival at 5 years (95% to 96%)	HR 1.05 (0.89 to 1.25)	6352 (3 studies)	⊕⊕⊕⊝ moderate^b
Lymphoedema Patient‐reported lymphoedema of any severity Follow‐up: 12 months	132 per 1000	48 per 1000 (22 to 115)	OR 0.33 (0.15 to 0.86)	815 (3 studies)	⊕⊕⊝⊝ low^b,c
Subjective arm movement impairment Follow‐up: 12 months	100 per 1000	40 per 1000 (24 to 69)	OR 0.38 (0.22 to 0.67)	877 (2 studies)	⊕⊝⊝⊝ very low^b,d,e
Paraesthesia Follow‐up: 12 months	776 per 1000	343 per 1000 (238 to 444)	OR 0.15 (0.09 to 0.23)	495 (2 studies)	⊕⊕⊝⊝ low^d,e
Pain Follow‐up: 12 months	177 per 1000	86 per 1000 (61 to 126)	OR 0.44 (0.3 to 0.67)	877 (2 studies)	⊕⊕⊝⊝ low^d,e
Numbness Follow‐up: 12 months	346 per 1000	185 per 1000 (152 to 222)	OR 0.43 (0.34 to 0.54)	1799 (3 studies)	⊕⊕⊕⊝ moderate^f
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio; OR: odds ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk taken from the full axillary surgery arm of Milan. ^bLow number of events. ^cIncomplete follow‐up for patient‐reported lymphoedema in ALMANAC. Event rates not reported in Addenbrookes 2. ^dModerate or substantial heterogeneity. ^eNo blinding or blinding not reported. ^fNo explanation provided.

Summary of findings 3. Sentinel node biopsy compared with full axillary surgery for operable primary breast cancer

Summary of findings 4. Radiotherapy alone compared with full axillary surgery for operable primary breast cancer

Radiotherapy alone compared with full axillary surgery for operable primary breast cancer
Patient or population: women with operable primary breast cancer Settings: hospital Intervention: radiotherapy alone Comparison: full axillary surgery
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)
	Assumed risk	Corresponding risk
	Full axillary surgery	Radiotherapy alone
All‐cause mortality	81% overall survival at 5 years^a	79% overall survival at 5 years (77% to 81%)	HR 1.1 (1 to 1.21)	2469 (4 studies)	⊕⊕⊕⊕ high
Local recurrence	90% local recurrence‐free survival at 5 years^b	92% local recurrence‐free survival at 5 years^a (90% to 93%)	HR 0.8 (0.64 to 0.99)	22,256^c (4 studies)	⊕⊕⊕⊕ high
The basis for the assumed risk* (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HR: hazard ratio.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
^aAssumed risk from full axillary surgery arm of NSABP B‐04 using mean 5‐year overall survival in combined N+ and N‐ groups. ^bAssumed risk from full axillary surgery arm of NSABP B‐04, using mean 5‐year risk for local or regional recurrence in combined lymph node‐positive and ‐negative groups. ^cPerson‐years of follow‐up.

Summary of findings 4. Radiotherapy alone compared with full axillary surgery for operable primary breast cancer

Table 1. Summary time‐to‐event statistics

Study	Outcome reported	Observed	Expected	Variance	HR	95% CIs	P value	Follow‐up	Notes
Addenbrookes	Overall mortality	ALND: 107/112 No ALND: 108/121	o‐e = ‐3.1	46.5	0.94	(0.70 to 1.25)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition The number of patients reported by Clarke 2005 differs from that reported by Brinkley (1971).
Addenbrookes	Breast cancer mortality	ALND: 74/112 No ALND: 78/121	o‐e = ‐2.2	32.8	‐	‐	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Addenbrookes	Isolated local recurrence	ALND: 7 events/1148 women‐years No ALND: 15 events/1218 women‐years	o‐e = 3.3	5.4	1.8	(0.79 to 4.28)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 9b), then inverted to reflect that more surgery is our control and less surgery is our research condition
ALMANAC	Overall mortality	ALDN: 7/476 SLNB: 7/478	NA	NA	NA	NA	NA	1 year	Cannot calculate o‐e. Not included in meta‐analysis
ALMANAC	Axillary recurrence	ALDN: 4/476 SLNB: 1/478	NA	NA	NA	NA	NA	1 year	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Overall mortality	ALND: 21/43 Simple: 30/52	o‐e = 4.74	12.35	1.47	(0.84 to 2.56)	0.1775	10 years	Tierney et al (2007) method 7 used log‐rank test results from figure 1. Cape Town
Cape Town	Overall mortality (node‐negative)	ALND: 14/21 Simple: 26/30	o‐e = 1.8	7.6	‐	‐	NA		Taken from Clarke 2005 (Appendix web figure 9a; Groote‐Schuur), then O‐E sign changed to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Cape Town	Overall mortality (node‐positive)	ALND: 19/22 Simple: 22/25	o‐e = ‐1.9	7.7	‐	‐	NA		Taken from Clarke 2005 (Appendix web figure 9b; Groote‐Schuur), then O‐E sign changed to reflect that more surgery is our control and less surgery is our research condition. Not included in meta‐analysis
Cape Town	Isolated local recurrence (node‐negative)	ALND: 3/206 women‐years Simple: 8/232 women‐years	o‐e = 1.7	2.3	2.09	(0.58 to 7.63)	NA		Taken from Clarke 2005 (Appendix web figure 9a; Groote‐Schuur), then inverted to reflect that more surgery is our control and less surgery is our research condition
Cape Town	Isolated local recurrence (node‐positive)	ALND: 5/134 women‐years Simple: 9/173 women‐years	o‐e = 0.0	2.0	1.00	(0.25 to 4.00)	NA		Taken from Clarke 2005 (Appendix web figure 9b; Groote‐Schuur), then inverted to reflect that more surgery is our control and less surgery is our research condition
Cape Town	Axillary recurrence	ALND: 2/43 Simple: 8/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Any locoregional recurrence	ALND: 11/43 Simple: 19/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cape Town	Distant metastases	ALND: 11/43 Simple: 13/52	NA	NA	NA	NA	NA	10 years	Cannot calculate o‐e. Not included in meta‐analysis
Cardiff	Overall survival	ALND: N = 97 Sampling: N =103 Total events = 152 Fig 2 data: ALND: 23/97 Sampling: 13/103	o‐e: 7.4	38	1.21	(0.29 to 0.99)	0.23	20 years	HR calculated using log‐rank P value from Stewart et al (1994, page 42) by Tierney 2007 method 8, 9. Owing to non‐proportionality of hazard rates, HR cannot be included in meta‐analysis
Cardiff	Disease‐free survival	ALND: 97 Sampling: 103	5.87	7.75	2.13	(1.05 to 4.31)	0.035	20 years	Log‐rank P value Tierney 2007 method 8, 9 (page 43 & Fig 5 Stewart et al, 1994)
Cardiff	Locoregional recurrence (chest wall, axilla, supraclavicular/internal mammary nodes)	ALND: 19/94 Sampling: 31/99 Fig 4: ALND: 11/97 Sampling: 22/103	o‐e: 6.46	11.78	1.73	(0.87 to 3.42)	NA	20 years	Tierney et al (2007) method 4 used and data from Figure 4 & page 42 Stewart et al (1994)
Cardiff	Distant relapse	ALND: 43/94 Sampling: 59/99	o‐e: 8.4	24.87	1.4	(0.99 to 1.71)	0.092	20 years	Data from Table 2, Stewart et al (1994): excludes patients with radiotherapy violations. Per‐protocol analysis ‐ not included in meta‐analysis
Cardiff	Breast cancer recurrence (total) (locoregional and distant relapse)	ALND: 62/94 Sampling: 90/99	o‐e: 12.77	36.71	1.42	(1.18 to 1.61)	0.035	20 years	Calculated from Stewart et al (1994) (excludes RT violations) per‐protocol analysis Risk of overestimation not certain as these are first events or total events.‐ not included in meta‐analysis
Edinburgh 1	Overall survival	ALND: ?/232 Sampling: ?/234 Total events = 53 ALND: 207/232 Sampling: 190/234	o‐e: ‐4.66	13.25	0.7	(0.41 to 1.21)	0.20	5 years	HR calculated using log rank P ‐ figure 2, Chetty (2000)
Edinburgh 1	Axillary recurrence	ALND: /232 Sampling: /234	o‐e: ‐0.15	13.25	0.99	(0.58 to 1.69)	0.94	Up to 8 years	Log‐rank P value Tierney 2007 method 7, 8, 9 used Fig 3 Chetty (2000)
Edinburgh 1	Local recurrence in the breast	ALND: 14/232 Sampling: 15/234	o‐e: ‐0.10	7.24	0.99	(0.48 to 2.04)	0.97	Up to 8 years	Tierney 2007 method 7, 8, 9 used Table 2 & page 87 Chetty (2000)
Edinburgh 1	Distant recurrence	ALND: 29/232 Sampling: 29/234	Not available	Not available	Not available	Not available	NA	Up to 8 years	Table 2, Chetty (2000). Unable to estimate HR ‐ not included in analysis
E'dburgh Sample/Clear	Overall survival	ALND: 76/203 Sampling: 71/203	o‐e: ‐3.81	36.55	0.90	(0.62 to 1.25)	NA	13 years	Tierney 2007 method 3 used (using 1995 data – Clarke 2005 paper reports more deaths) Fig 1 and page 82 HR (CI) in Forrest et al (1995) ‐ inverted the HR
E'dburgh Sample/Clear	Distant metastases	ALND: 51/203 Sampling: 53/203	o‐e: 1.5	30.78	0.92	(0.67 to 1.35)	NA	13 years	Tierney 2007 method 3 used (using 1995 data), Fig 2 and HR (CI) page 82 in Forrest et al (1995), inverted the HR
E'dburgh Sample/Clear	Locoregional relapse (chest wall, axilla, supraclavicular)	ALND: 38/203 Sampling: 29/203	o‐e: ‐4.9	16.32	0.74	(0.46 to 1.20)	NA	13 years	Tierney 2007 method 3 used (using 1995 data) Method 3 Fig 3 from HR (CI), page 82 in Forrest et al (1995), inverted the HR
Genoa	Overall survival	ALND: 4/115 SLNB: 5/110	o‐e: 0.58	2.22	1.32	(0.35 to 4.92)	0.679	5 years	Log‐rank P value (Canavese 2009 ‐ fig 3) Tierney 2007 method 7 used Fig 3 KM curve gives P = 0.679. I assumed that was correct as it appears on the graph. The text value (page 20) may be a typo 0.697. HR are similar; CI differ
Genoa	Axillary recurrence	ALND: 1/115 SLNB: 0/110	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
Genoa	Breast cancer recurrence (local and contralateral recurrence, axillary and distant metastases)	ALND: 10/115 SLNB: 8/110	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
Genoa	5‐Year event‐free survival	ALND: 12/115 SLNB: 10/110	o‐e: ‐0.85	5.45	0.86	(0.37 to 1.98)	0.715	5 years	Log‐rank P value from Fig 2, Canavese (2009) method 7 Tierney 2007 used
GIVOM Sentinella	Overall survival	ALND: 14/352 SLNB: 21/345	NA	NA	NA	NA	NA	5 years	Not included in meta‐analysis
GIVOM Sentinella	Disease‐free survival	ALND: 28/352 SLNB: 39/345	o‐e = 1.18	16.3	1.08		0.769	5 years	Method 7 Tierney 2007 used
GIVOM Sentinella	Axillary recurrence	ALND: 0/352 SLNB: 1/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
GIVOM Sentinella	Locoregional recurrence	ALND: 3/352 SLNB: 16/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
GIVOM Sentinella	Distant recurrence	ALND: 16/352 SLNB: 11/345	NA	NA	NA	NA	NA	5 years	Cannot calculate o‐e. Not included in meta‐analysis
Guy's	Overall mortality (clinically node negative)	ALND: 178/241 No ALND (wide excision): 185/233	o‐e = 13.8	80.7	1.26	(0.98 to 1.63)	0.1	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Overall mortality (clinically node positive)	ALND: 82/85 No ALND (wide excision): 64/71	o‐e = 4.3	30.9	1.15	(0.81 to 1.64)	0.4	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Breast cancer mortality (clinically node negative)	ALND: 122/241 No ALND (wide excision): 142/233	o‐e = 13.8	58.8	‐	‐	0.07	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention Not included in meta‐analysis
Guy's	Breast cancer mortality (clinically node positive)	ALND: 53/85 No ALND (wide excision): 54/71	o‐e = 6.2	23.6	‐	‐	0.2	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention. Not included in meta‐analysis
Guy's	Isolated local recurrence (clinically node negative)	ALND: 35 events/3267 women‐years No ALND: 81 events/2383 women‐years	o‐e = 29.5	26.4	3.06	(2.09 to 4.48)	< .00001	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Guy's	Isolated local recurrence (clinically node positive)	ALND: 17 events/873 women‐years No ALND: 31 events/519 women‐years	o‐e = 10.5	10.8	2.64	(1.46 to 4.80)	0.001	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research intervention
Hammersmith	Overall survival	Radical: 35/76 Simple: 40/76	o‐e = 1.44	11.78	1.13	(0.64 to 2.00)	NA	8 years	Extracted from Fig 3, Burn et al (1968) Tierney 2007 method 10 on Simple is input as "research" and radical as "control". Min and max follow‐up input as 3‐96 months
Hammersmith	Local recurrence	Radical: 10/76 Simple: 11/76	NA	NA	NA	NA	NA	4‐9 years	Not included in meta‐analysis
Hammersmith	Mean time to recurrence	Radical: 15.7 months Simple: 25.9 months	NA	NA	NA	NA	NA	4‐9 years	Not included in meta‐analysis
IBCSG‐10‐93	Overall survival	ALND: 72/234 Surgery only: 71/239	o‐e = 1.76 (survival curves cross)	36.05	1.05	(0.76 to 1.46)	0.77	6‐7 years	HR reported on page 340 of IBCSG (2006), used Tierney 2007 method 3
IBCSG‐10‐93	Disease‐free survival	ALND: 92/234 Surgery only: 89/239	o‐e = 2.6	44.69	1.06	(0.79 to 1.42)	0.69	6‐7 years	HR reported on page 340 of IBCSG (2006), used Tierney 2007 method 3
IBCSG‐10‐93	Axilla recurrence (as first event)	ALND: 2/234 Surgery only: 6/239	NA	NA	NA	NA	NA	6‐7 years	Not included in meta‐analysis
Institut Bergonie	Overall survival (whole follow‐up period) ITT	no ALND: NR ALND: NR	o‐e = 6.42	7.04	2.49	90% CI (1.34 to 4.63)	NA	Whole follow‐up period (unclear how long that is)	HR reported on page 566 of Avril (2011), used Tierney 2007 method 3
Institut Bergonie	Event‐free survival (whole follow‐up period) ITT	no ALND: 44/297 ALND: 31/297	o‐e = 8.75	18.37	1.61	90% CI (1.1 to 2.37)	NA	Whole follow‐up period (unclear how long that is)	HR reported on page 566 of Avril (2011), used Tierney 2007 method 3
Institut Bergonie	Axillary event	Within 5 years: no ALND: 4/297 ALND: 0/310 After 5 years: no ALND: 2/297 ALND: 0/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Lymph node (excl axillary) event	Within 5 years: no ALND: 1/297 ALND: NA After 5 years: no ALND: 0/297 ALND: NA	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Breast/chest wall event	Within 5 years: no ALND: 5/297 ALND: 4/310 After 5 years: no ALND: 0/297 ALND: 8/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Metastatic event	Within 5 years: no ALND: 4/297 ALND: 1/310 After 5 years: no ALND: 2/297 ALND: 2/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Contralateral breast cancer	Within 5 years: no ALND: 2/297 ALND: 1/310 After 5 years: no ALND: 2/297 ALND: 1/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Bergonie	Other site cancer	Within 5 years: no ALND: 5/297 ALND: 5/310 After 5 years: no ALND: 5/297 ALND: 4/310	NA	NA	NA	NA	NA		Not included in meta‐analysis
Institut Curie	Overall survival	RT: 43/331; ALND: 29/326	o‐e = 7	17.3	1.50	(0.94 to 2.40)	NA		Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
Institut Curie	Isolated local recurrence	RT: 39/2045 women‐years; ALND: 34/2126 women‐years	o‐e = 1.6	17.5	1.10	(0.69 to 1.75)	NA		Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
Institut Curie	Axilla recurrence	RT: 12/332; ALND: 5/326	o‐e = 3.86	3.53	3.93	‐	0.04		Table 2 in Louis‐Sylvestre (2004), method 7 in Tierney 2007
Institut Curie	Disease‐free survival	RT: 5 years : 82 (SD = 2.1)% 10 years : 72 (SD = 2.5)% 15 years : 65.5 (SD = 2.7)%	ALND: 5 years: 83.3 (SD 2)% 10 years: 72.6 (SD 2.5)% 15 years: 64.3 (SD 2.9)%.	NA	NA	NA	NA		o‐e cannot be extracted because P values not reported past NS in Table 2 in Louis‐Sylvestre (2004). Not included in meta‐analysis
Institut Curie	Metastases	RT: 5 years: 12.8 (SD 1.9)% 10 years: 21 (SD 2.3)% 15 years: 24.9 (SD 2.5)%	ALND: 5 years: 10.8 (SD 1.7)% 10 years: 18.3 (SD 2.2)% 15 years: 25.8 (SD 2.6)%	NA	NA	NA	NA		O‐e cannot be extracted because P values not reported past NS in Table 2 in Louis‐Sylvestre (2004). Not included in meta‐analysis
Malmo	Overall survival	ALND + RT: ?/97 Mastectomy alone: ?/98 (total event rate = 91)	o‐e = ‐4.19	22.75	0.83	(0.55 to 1.25)	0.38	15‐20 years	Using P = 0.38 reported on page 558 of Borgstrom (1994) and Tierney 2007 method 8. The o‐e is calculated on the basis of a total event rate of N = 91, and total N = 97 in the ALND + RT group and N = 98 in mastectomy alone group (i.e. intent‐to‐treat numbers), and using the only P value reported, which was for per‐protocol analysis that study authors stated did not differ from intention‐to‐treat analyses
Malmo	Chest wall recurrence	ALND + RT: 2/97 Mastectomy alone: 11/98	NA	NA	NA	NA	NA	15‐20 years	Cannot calculate o‐e. Not included in meta‐analysis
Manchester	Overall survival	Radical: 126/149 Simple + RT: 140/159	o‐e = 5.4	58.6	1.10	(0.85 to 1.42)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Manchester	Death from breast cancer	Radical: 100/149 Simple + RT: 112/159	o‐e = 2.8	46	1.06	(0.80 to 1.42)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Manchester	Local recurrence	Radical: 48 events/997 women‐years Simple + RT: 41 events/1113 women‐years	o‐e = ‐5.7	19.9	0.75	(0.48 to 1.17)	NA	15 years	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Milan	Death from any cause (OS)	ALND = 23/257 SLNB = 15/259	o‐e = ‐4.34	9.08	0.62	(0.32 to 1.19)	0.15	10 years	Log‐rank P (Tierney 2007 method 7); ALND is control
Milan	Breast cancer recurrence (local recurrence, regional lymph node metastases, distant metastases)	ALND = 26/257 SLNB = 23/259	o‐e = ‐2.25	12.02	0.83	(0.47 to 1.46)	0.52	10 years	Log‐rank P (Tierney 2007 method 7); ALND is control
Milan	Distant metastasis	ALND = 20/257 SLNB = 17/259	o‐e = ‐2.04	9.19	0.80	(0.42 to 1.53)	0.50	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan	Axillary metastasis	ALND = 0/257 SLNB = 2/259	o‐e = 0.97	0.50	6.96	(0.44 to 111.3)	0.17	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 8 and 9); ALND is control
Milan	Local recurrence	ALND = 4/257 SLNB = 4/259	o‐e = ‐0.12	2.00	0.94	(0.24 to 3.76)	0.93	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan	Supraclavicular metastasis	ALND = 2/257 SLNB = 0/259	o‐e = ‐1.02	0.50	0.13	(0.01 to 2.09)	0.15	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 8, 9); ALND is control
Milan	Contralateral breast cancer	ALND = 10/257 SLNB = 9/259	o‐e = ‐0.81	4.47	0.84	(0.34 to 2.07)	0.71	10 years	Log‐rank P from table 4 Veronesi (2010) (Tierney 2007 method 7); ALND is control
Milan 2	Overall survival	ALND = 31/109 No ALND = 35/110	o‐e = ‐2.72	16.43	0.85	(0.52 to 1.37)		Median = 150 months	HR reported on page 922 of Martelli (2012). Using Tierney 2007 method 3 o Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status
Milan 2	Breast cancer deaths	ALND: 8/109 No ALND: 10/110	o‐e = 1.33	4.06	1.39	‐	‐	Median = 150 months	HR reported in Table 3 of Martelli (2012). Tierney 2007 method 3 o Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status. Not included in meta‐analysis
Milan 2	Axillary relapse	ALND: 0/109 No ALND: 4/110	NA	NA	NA	NA	NA	Median = 150 months	Table 2 of Martelli (2012), cannot calculate o‐e
Milan 2	Recurrence (ipsilateral breast tumour)	ALND: 4/109 No ALND: 7/110	NA	NA	NA	NA	NA	Median = 150 months	Table 2 of Martelli (2012), cannot calculate o‐e
Milan 2	Distant metastases	ALND: 9/109 No ALND: 9/110	o‐e = ‐2.68	5.93	0.64	(0.28 to 1.42)	NA	Median = 150 months	HR reported in Table 3 of Martelli (2012). Tierney 2007 method 3 Please note, the curves cross; also the HR used for extraction of o‐e and its variance is adjusted for tumour grade and oestrogen‐receptor status
Milan 3	Overall survival	10‐year ALND: 93.3% (95% CI 89.4‐95.8) no ALND: 91.5% (95% CI 87‐94.4)	o‐e = 1.76	12.33	1.15	(0.66 to 2.02)	P = .436	Median = 127.5 months	Agresti (2014) Figure 3A and Tierney 2007 method 11 Please note, the curves cross at the very end, also HR used for extraction of o‐e
Milan 3	Death from breast cancer	ALND: 17/272 no ALND: 15/245	NA	NA	NA	NA	P = 1.00	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Disease‐free survival	10‐year ALND: 92.4% (95% CI 88.5‐95.1) no ALND: 91.3% (95% CI 86.7‐94.3)	o‐e= ‐0.13	10.7	0.99	(0.54 to 1.8)	P = .97	Median = 127.5 months	Agresti (2014) Figure 3A andTierney 2007 method 11 Please note, the curves cross at the very end; also the HR used for extraction of o‐e
Milan 3	Distant metastases	ALND: 23/272 no ALND: 20/245	NA	NA	NA	NA	P = 1.00	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Axillary recurrence	ALND: 0/272; no ALND: 22/245	NA	NA	NA	NA	NA	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Local recurrence	ALND: 14/272 no ALND: 11/245	NA	NA	NA	NA	P = .839	Median = 127.5 months	Not included in meta‐analysis
Milan 3	Contralateral breast cancer	ALND: 13/272 no ALND: 14/245	`NA	NA	NA	NA	P = .695	Median = 127.5 months	Not included in meta‐analysis
NSABP B‐04	Overall survival: node negative: ALND vs no ALND	ALND = 259/389 No ALND = 256/384	o‐e = ‐5	117.3	0.96	(0.80 to 1.15)	NA	15 years?	Taken from Clarke 2005 Lancet (Appendix web figure 9a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Overall survival: node negative: ALND vs no ALND + RT	ALND = 259/389 No ALND + RT = 271/386	o‐e = 8.6	122.2	1.07	(0.90 to 1.28)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Overall survival: node positive: ALND vs no ALND + RT	ALND = 244/301 No ALND + RT = 244/305	o‐e = 8.3	109.4	1.08	(0.89 to 1.30)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node negative: ALND vs no ALND	ALND = 35 events/3949 women‐years No ALND = 94 events/3335 women‐years	o‐e = 31.5	29.2	2.94	(2.05 to 4.23)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 9a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node negative: ALND vs no ALND + RT	ALND = 35 events/3949 women‐years No ALND + RT = 18 events/3896 women‐years	o‐e = ‐8.7	13	0.51	(0.30 to 0.88)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Local isolated recurrence: node positive: ALND vs no ALND + RT	ALND = 45 events/2268 women‐years No ALND + RT = 42 events/2025 women‐years	o‐e = ‐0.5	20.8	0.98	(0.64 to 1.50)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
NSABP B‐04	Disease‐free survival: node negative: ALND vs no ALND	ALND = 281/362 No ALND + RT = 287/365	o‐e = 9.36	138.3	1.07	(0.91 to 1.27)	0.39	25 years	FIsher (2008) page 568 (radical vs total mastectomy) Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Disease‐free survival: node negative: ALND vs no ALND + RT	ALND = 281/362 No ALND + RT = 292/352	o‐e = 8.3	142.39	1.06	(0.90 to 1.25)	0.49	25 years	FIsher (2008) page 568 (radical vs total mastectomy + RT) Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Disease‐free survival: node positive: ALND vs no ALND + RT	ALND = 254/292 No ALND + RT = 258/294	o‐e = 14.46	127.57	1.12	(0.94 to 1.33)	0.20	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of disease‐free survival were the first local, regional or distant recurrence of tumour; contralateral breast cancer or a second primary tumour other than a tumour in the breast; and death with no evidence of cancer
NSABP B‐04	Relapse‐free survival: node negative: ALND vs no ALND	ALND = 154/362 No ALND + RT = 182/365	o‐e = 10.17	77.61	1.14	(0.91 to 1.42)	0.27	25 years	FIsher (2008) page 568 Tierney 2007 method 3; calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Relapse‐free survival: node negative: ALND vs no ALND + RT	ALND = 154/362 No ALND + RT = 163/352	o‐e = ‐2.9	71.05	0.96	(0.76 to 1.21)	0.74	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Relapse‐free survival: node positive: ALND vs no ALND + RT	ALND = 178/292 No ALND + RT = 183/294	o‐e = 7.63	88.52	1.09	(0.89 to 1.35)	0.40	25 years	FIsher (2008) page 568, Tierney 2007 method 3, calculated from the date of mastectomy, events considered in determination of relapse‐free survival were the first local, regional or distant recurrence; or an event in the contralateral breast
NSABP B‐04	Time to distant metastasis: node negative: ALND vs no ALND	ALND = 101/362 No ALND + RT = 107/365	o‐e = 8.44	88.52	1.1	(0.89 to 1.35)	0.39	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐04	Time to distant metastasis: node negative: ALND vs no ALND + RT	ALND = 101/362 No ALND + RT = 111/352	o‐e = 6.69	86.9	1.08	(0.88 to 1.34)	0.44	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐04	Time to distant metastasis: node positive: ALND vs no ALND + RT	ALND = 120/292 No ALND + RT = 127/294	o‐e = 5.98	88.41	1.07	(0.87 to 1.32)	0.51	25 years	FIsher (2008) page 569, Tierney 2007 method 3
NSABP B‐32	Overall survival (all randomised participants, i.e. node+ and node‐)	ALND = 228 (deaths)/2807 SLN = 252 (deaths)/2804	10.32	119.7	1.09	(0.91 to 1.3)	0.35	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Disease‐free survival (all randomised participants, i.e. node+ and node‐)	ALND = 455/2807 SLN = 475/2804	4.6	232.39	1.02	(0.9 to 1.16)	0.72	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Local/regional recurrence (all randomised participants, i.e. node+ and node‐)	ALND = 121/2807 SLN = 112/2804	‐2.37	58.16	0.96	(0.74 to 1.24)	0.77	10 years	From Julian (2013) using Tierney 2007 method 4. Contacted author (Krag) to confirm direction of effect
NSABP B‐32	Axillary recurrence (all randomised participants, i.e. node+ and node‐)	ALND = 6/2807 SLN = 14/2804	NA	NA	NA	NA	NA	10 years	o‐e cannot be calculated. Not included in meta‐analysis
NSABP B‐32	Overall survival (for SLN‐neg)	ALND = 219 (dead)/1975 SLN = 245 (dead)/2011	o‐e = 12.07	115.64	1.11	(0.93 to 1.33)	0.27	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Disease‐free survival (for SLN‐neg)	ALND = 456 (diseased)/1975 SLN = 465 (diseased)/2011	o‐e = 2.29	230.23	1.01	(0.89 to 1.15)	0.92	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Local regional recurrence	ALND = 85 (events)/1975 SLN = 80 (events)/2011	o‐e = ‐2.11	41.21	0.95	(0.7 to 1.29)	0.77	10 years	From Julian (2013) using Tierney 2007 method 4
NSABP B‐32	Local recurrence in SLN‐negative participants	ALND = 54 (events)/1975 SLN = 49 (events)/2011	o‐e = ‐3.03	25.69	0.89	(0.6 to 1.31)	0.55	Mean = 95.6 months	From Krag (2010) page 930 using logrank P = 0.55 Tierney 2007 method 7
NSABP B‐32	Regional recurrence in SLN‐negative participants	ALND = 8 (events)/1975 SLN = 14 (events)/2011	o‐e = 2.77	5.09	1.72	(0.72 to 4.11)	0.22	Mean = 95.6 months	From Krag (2010) page 930 using log rank P = 0.22 Tierney 2007 method 7
NSABP B‐32	Distant recurrence in SLN‐negative patients	ALND = 55 (events)/1975 SLN = 64 (events)/2011	o‐e = 3.91	29.82	1.14	(0.8 to 1.64)		Mean = 95.6 months	From Krag (2010) Figure 4 Tierney 2007 method 3
Ostersund	Recurrence in the axilla	ALND: 0/57 Sampling: 1/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
Ostersund	Local recurrence	ALND: 4/57 Sampling: 1/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
Ostersund	Distant recurrence	ALND: 1/57 Sampling: 4/54	NA	NA	NA	NA	NA	Median: 30 (range, 5‐76) months	From Borup‐Chistesen (1993) table IV. Recurrence is reported only out of N = 111 (57 + 54) participants who did not have metastases in axillary lymph nodes after dissection or biopsy. Cannot calculate o‐e on the basis of available data
SE Scotland	Overall survival: node negative: ALND vs Simple + RT	ALND = 143/199 Simple + RT = 143/180	o‐e = 17.5	65.7	1.31	(1.02 to 1.66)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Overall survival: node positive: ALND vs Simple + RT	ALND = 72/89 Simple + RT = 77/93	o‐e = 6.3	34.1	1.20	(0.86 to 1.68)	NA	15 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Local isolated recurrence: node negative: ALND vs no ALND + RT	ALND = 26 events/2880 women‐years Simple + RT = 21 events/2204 women‐years	o‐e = ‐0.5	11.3	0.96	(0.53 to 1.71)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
SE Scotland	Local isolated recurrence: node positive: ALND vs no ALND + RT	ALND = 24 events/943 women‐years Simple + RT = 17 events/878 women‐years	o‐e = ‐2.9	9.8	0.74	(0.40 to 1.39)	NA	5 years?	Taken from Clarke 2005 (Appendix web figure 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Overall survival ‐ node negative	ALND: 56/101 Simple + RT to chest wall & axilla: 42/85	o‐e = ‐5.5	21.4	0.77	(0.51 to 1.18)	NA	15 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9a and 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Overall survival ‐ node positive	ALND: 13/17 Simple + RT to chest wall & axilla: 7/9	o‐e = ‐0.5	3.3	0.86	(0.29 to 2.53)	NA	15 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9b and 10b). then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Isolated local recurrence ‐ node negative	ALND: 15/510 py Simple + RT to chest wall & axilla: 13/483 py	o‐e = 0.0	6.7	1.00	(0.47 to 2.13)	NA	5 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9a and 10a), then inverted to reflect that more surgery is our control and less surgery is our research condition
WSSA Glasgow	Isolated local recurrence ‐ node positive	ALND: 3/69 py Simple + RT to chest wall & axilla: 1/41 py	o‐e = ‐0.5	0.9	0.57	(0.07 to 4.53)	NA	5 years?	CAUTION: same control group used twice for these data Taken from Clarke 2005 (Appendix web figures 9b and 10b), then inverted to reflect that more surgery is our control and less surgery is our research condition
Xu 2003	10‐year overall survival	Level I clearance: 75/93 ALND: 71/88	NA	NA	NA	NA	NA	10 years	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	10‐year disease‐free survival	Level I clearance: 72/93 ALND: 68/88	NA	NA	NA	NA	NA	10 years	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Breast cancer recurrence	Level I clearance: 19/93 ALND: 17/88	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Local recurrence	Level I clearance: 3.2% ALND: 2.3%	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Xu 2003	Distant metastasis	Level I clearance: 19/93 ALND: 15/88	NA	NA	NA	NA	NA	10 years?	o‐e could not be calculated as no P values reported. Not included in meta‐analysis
Figures in bold were reported in the original publication; others were derived (see Notes column).

Table 1. Summary time‐to‐event statistics

Table 2. Morbidity definitions

Study	Oedema	Shoulder function	Skin graft	Delayed healing	Activity	Attitude	Other	Notes
Guy's	Slight: 0‐2.5 cm Moderate: 2.5‐4.5 cm Severe > 4.5 cm Circumference of both arms measured 7.5 cm below the acromion, 18 cm above and 10 cm below the olecranon and at the wrist Presumably difference between arm circumference	Arm function: Good: uses arm freely Fair: cannot do usual tasks Poor: very unsatisfactory use of arm Appears to be assessed by patient questionnaire			Good: normal activity, back at work or resumed usual activities Fair: light work only because of operation; not resumed usual activities Poor: inactive. Assessed by patient questionnaire	Good: no complaints Fair: some complaints Poor: very unhappy about experience Assessed by patient questionnaire
ACOSOG Z0011	Lympheoedema (subjective) – according to patient self‐report or physician diagnosis Lympheoedema (objective): 2 cm or greater postop increase in ipsilateral arm circumference	Axillary paraesthesia – patient reported Brachial plexus injury – determined by physician on examining the patient
Addenbrookes	1. Mild oedema 2. Gross oedema (estimated by measuring the circumference of each arm with the arm extended at points 11 inches and 22 inches from the tips of the middle finger. An increase of 1 inch in the circumference of the arm on the side of the operation at either or both points was taken to indicate some degree of oedema)	Stiff shoulder	Need for skin graft	Sufficient to cause postponement of radiotherapy until at least 2 months after the operation. Although incidence of delayed healing varied between surgeons, each showed the same trend of higher incidence following a radical operation
Addenbrookes 2	Subjective lymphoedema: patient reported Objective lymphoedema: circumferential arm measurement at 4 cm intervals from the wrist (approximately 10 measurements) used to calculate arm volume. Volume corrected using measurements from contralateral arm	Range of movement measured by recording degrees of flexion, abduction and internal and external rotation using goniometer Sensory function tested using pinprick, light touch				Global Severity Index (GSI; low values better), Beck's Depression Inventory, Spielberger's State‐Trait anxiety, MAC, SF‐36 (measured psychological morbidity and quality of life)
ALMANAC	Change in ipsilateral arm volume at each follow‐up visit was expressed as a % increase from pretreatment value. Ratios of presurgery to postsurgery arm volumes were compared on a log‐transformed scale. The contralateral arm was used as a control for evaluations of arm volume Also patient rated as mild, moderate or severe	Assessed by goniometric measurement of arm movement (flexion, abduction, internal rotation and external rotation). Changes between visits calculated by subtraction The contralateral arm was used as a control for arm and shoulder function					QoL: Fact‐B+4 Anxiety: Spielberger STAI
Cardiff ‐ Local								No morbidity data
Cardiff ‐ St Mary's	Oedema of arm 72 cm	Restricted elevation 720 degrees				Measured but not reported	Axillary pain; numbness or paraesthesia on operated sides; aesthetic appearance of axillary scar
Edinburgh 1	Arm swelling measured by water displacement, circumference 15 cm above and below the olecranon process	Shoulder mobility assessed by measuring elevation through flexion, abduction, medial and lateral rotation					Shoulder muscle power assessed using graduated spring to measure flexion, extension, abduction and adduction of the shoulder joint
E'dburgh Sample/Clear	Arm welling (arm circumference 15 cm above and 10 cm below olecranon)	Objective assessment via adduction with internal rotation; abduction with external rotation, difference in height reached between treated and non‐treated arms by stretching above head, measurement of an abduction movement without shoulder rotation whilst lying on a flat, hard surface					Power (cm/kg) of pectoralis major by repeated lifting of a 3.5 kg weight as fast as possible over 45 seconds, comparing treated and untreated arm	Sample from study only, level B evidence
GIVOM Sentinella	Lympheodema was assessed by comparing the circumference of the operated vs the non‐operated arm at 15 cm above the epicondyle Unclear what difference in circumference constituted lymphoedema	Assessed by the surgeon by evaluating active and passive flexion, abduction, internal and external rotation, and classified on a scale 0 (normal mobility) to 3 (severe mobility) restriction Winged scapula reported as present/absent					Axillary and arm pain reported by patients on a scale from 0 (absent) to 3 (continuous/severe) Numbness assessed by the surgeon by comparing skin sensitivity in operated and non‐operated arms. Rated 0 (absent) to 3 (severe)
Guy's	Reports lymphoedema; categorised as none, slight, moderate and severe	Reports arm function as good, fair or poor			Reports activity as good, fair or poor	Reports attitude as good, fair or poor	Pts in no axillary surgery + RT arm reported fibrosis of breast and sometimes "marbling" of the overlying skin. Both occurred in <5% of cases
Hammersmith	Impairted function of the shoulder joint and swollen arm: no definitions given, but it is stated that the methodology included volumetric measurement of the upper limb and that an attempt was made to ally objective measurements with the patient’s subjective expression of discomfort or disability	Impairted function of the shoulder joint and swollen arm: no definitions given, but it is stated that the methodology included volumetric measurement of the upper limb and that an attempt was made to ally objective measurements with the patient’s subjective expression of discomfort or disability						In evaluating morbidity, attempts made to ally objective measurements with patient's subjective expression of disability or discomfort. Expectation that after RM, slight increase in volume of ipsilateral arm, or after RT, some discomfort and stiffness to shoulder, but these do not amount to morbidity
IBCSG‐10‐93	≥ 5% increase in arm circumference from baseline						QOL: A core questionnaire plus a surgical module specific to this trial. Four linear analogue scales on the core questionnaire were used: well‐being, mood, appetite and perceived adjustment/coping. After 1993, 6 additional scales were added: tiredness, hot flashes, nausea/vomiting, perceived social support, arm restriction and subjective health estimation. Surgical module measured swelling, numbness, weakness, pain, stiffness, performance of daily activities and global measure of arm/hand/shoulder/chest bother
IBCSG‐23‐01	No definitions for functional outcomes reported
Institut Bergonie	No definitions for functional outcomes reported
IPO‐P	An increase in arm volume was defined as an increase > 2 cm, comparing the circumference of the operated upper limb (at 3 points: the wrist, the midpoint of the forearm and the midpoint of the upper arm) with its non‐operated counterpart	Patients were asked to lift their operated arm (maximum possible abduction): abduction ≥ 90° was considered adequate; abduction < 90° was considered abnormal					Patients were asked: Is your arm painful in a resting position (yes/no)? Does the inside of your arm feel more numb (yes/no)?
Manchester
Milan	Arm swelling was assessed by comparing the circumference of treated and untreated arms 15 cm above the lateral epicondyle	Arm mobility was judged by asking the patient to rate restriction in movement on a scale 0 to 100 Numbness assessed by comparing skin sensitivity on inside and outside of the upper arm – classified as yes/no		Aesthetic appearance of scar judged by patient (rated good or bad)			Postoperative pain was evaluated as continuous (> 50% of the day), sporadic or absent
NSABP B‐04	Ipsilateral and contralateral measurement of arm circumference at 15 cm below the acromion process and 15 cm below the olecranon: An increase in arm circumference ≥ 2 cm in ipsilateral arm (below or above the elbow) indicated arm oedema
NSABP B‐32	Arm volume measured using volume of water displaced determined by the difference between treated and untreated arms (relative arm volume difference = [ipsilateral‐contralateral]/[contralateral] × 100%)	Arm mobility in degrees was determined by measuring the straight lateral abduction of both ipsilateral and contralateral arms using a standard orthopaedic goniometer to determine the angle between lateral chest wall and humerus (relative shoulder abduction deficit = [ipsilateral‐contralateral]/[contralateral] × 100%)					Numbness and tingling were assessed by self‐report by asking patients if they were currently experiencing any numbness or any tingling anywhere in ipsilateral and contralateral arms. OR of SLN compared with ALND Adverse events: no details reported
Ostersund	Arm volume measured using volume of water displaced. A cutoff of 10% increase in volume was used as the arbitrary cut point	Shoulder mobility (flexion, abduction and rotation) was determined with the help of a 360° scale placed on a wall with the centre at shoulder height
SNAC	Arm volume was estimated using 6 measures of arm circumference at 10 cm intervals starting 10 cm from the tip of the index finger. Upper limb swelling was expressed as percentage change in volume from baseline	Abduction and flexion measured using goniometer Arm morbidity measured using the 15‐item SSSS scale developed for the study, with each rated from 0 (no trouble at all) to 10 (worst I can imagine) and averaged to obtain overall score
SE Scotland	Increase in circumference of forearm by at least 3 cm	Failure to abduct the arm beyond a right angle
Xu 2003	Postoperative swelling: middle grade (diameter is 3–6 cm enlargement on the involved upper arm or forearm compared with the contralateral part)

Table 2. Morbidity definitions

Table 3. Morbidity data at each time point

Study	Outcome	Measurement	Follow‐up period 1	Follow‐up period 2	Notes
ACOSOG Z0011	Wound infection	Determined by treating physician	SLND: 11/371; SLND + ALND: 31/373
ACOSOG Z0011	Axillary seroma	Determined by treating physician	SLND: 21/371; SLND + ALND: 53/373
ACOSOG Z0011	Brachial plexus injury	Determined by treating physician	At 6 months: SLND: 3/415; SLND + ALND: 5/406 At 1 year: SLND: 0/415; SLND + ALND: 1/406
ACOSOG Z0011	Axillary paraesthesia	Patient reported	30 days: SLND: 43/371; SLND + ALND: 174/373	6 months: SLND: 35/288; SLND + ALND: 146/335
ACOSOG Z0011	Axillary paraesthesia	Patient reported	12 months: SLND: 24/268; SLND + ALND: 113/287
ACOSOG Z0011	Lymphoedema (objective)	Arm measurement	30 days: SLND: 17/272; SLND + ALND: 23/255	6 months: SLND: 21/271; SLND + ALND: 29/270
ACOSOG Z0011	Lymphoedema (objective)	Arm measurement	12 months: SLND: 14/226; SLND + ALND: 26/242
ACOSOG Z0011	Lymphoedema (subjective)	Patient reported/physician diagnosis	6 months: SLND: 19/339; SLND + ALND: 27/327	12 months: SLND: 12/268; SLND + ALND: 37/288
ACOSOG Z0011	Lymphoedema (subjective)	Patient reported/physician diagnosis	> 12 months: SLND: 14/253; SLND + ALND: 52/272
Addenbrookes	Mild oedema		Follow‐up was at least 12 months in most cases. ALND = 7/91; Simple = 5/113
Addenbrookes	Stiff shoulder		ALND = 6/91; Simple = 8/113
Addenbrookes	Skin graft	Need for skin graft	ALND = 4/91; Simple = 2/113
Addenbrookes	Delayed healing	Need to delay postoperative RT	ALND = 18/91; Simple = 7/113
Addenbrookes	Gross oedema	Arm measurement	ALND = 0/91; Simple = 0/113	ALND = 12/45; Simple = 6/53
Addenbrookes 2	Seroma		ALND: 33/155; SLNB: 20/143
Addenbrookes 2	Lymphoedema (objective)	Arm volume changes	12 months: ALND: mean (SE) = 56.4 (10.9); SLNB: mean (SE) = 18.6 (13.8), difference mean (SE) = 37.8 (17.6) Mean (1, 3, 6, 12 months): ALND: mean (SE) = 53.1 (8.1); SLNB: mean (SE) = 17.7 (9.2), difference mean (SE) = 35.4 (12.2)	Max: ALND: mean (SE) = 113.7 (9.7); SLNB: mean (SE) = 78.4 (12), difference mean (SE) = 35.3 (15.3)
Addenbrookes 2	Lymphoedema (subjective)	Patient reported	1 month: OR = 0.34 (95% CI 0.11 to 0.9); 3 months: OR = 0.4 (95% CI 0.16 to 0.94); 6 months: OR = 0.25 (95% CI 0.08 to 0.66)	12 months: OR = 0.36 (95% CI 0.15 to 0.86); mean: OR = 0.3 (95% CI 0.18 to 0.68)	Odds ratios: SLNB/ALND; i.e. lower favours SLNB
Addenbrookes 2	Paraesthesia		ALND: 130/155; SLNB: 92/140
Addenbrookes 2	Numbness		ALND: 115/155; SLNB: 68/143
Addenbrookes 2	Loss of pinprick		ALND: 118/155; SLNB: 77/140
Addenbrookes 2	Loss of light touch		ALND: 121/155; SLNB: 81/140
Addenbrookes 2	QOL (immediate postop)		Study authors note QOL scores were usually higher (better) in the SLND group and significantly so in the immediate postoperative period (P < 0.01). No significant effect of node positive/negative
Addenbrookes 2	MAC scale (12 months)		Study authors no significant difference in MAC scores during 1 year follow‐up. No significant effect of node positive/negative
Addenbrookes 2	BSI – somatisation (immediate postop)		SLND group scored lower (better) than ALND in the immediate postoperative period (P < 0.001)
Addenbrookes 2	Quality of life	GSI level	12 months: ALND: mean (SE, N) = 49.7 (1.1, 143); SLNB: mean (SE, N) = 48.4 (0.9, 134), difference mean (SE) = 1.3 (1.4)	OR for morbid GSI: study/control (95% CI) 0.55 (0.08 to 2.94)
Addenbrookes 2	Quality of life	SF‐36 (immediate postoperative)	Physical combined: ALND: mean (SD, N) = 38.6 (8.2, 143); SLNB: mean (SD, N) = 42.3 (10.4, 134), difference mean (95% CI) = 3.7 (1.2 to 6.1) Physical functioning: ALND: mean (SD, N) = 41.3 (9, 143); SLNB: mean (SD, N) = 44.5 (8.1, 134), difference mean (95% CI) = 3.2 (1.1 to5.4)	Vitality: ALND: mean (SD, N) = 48.2 (10.2, 143); SLNB: mean (SD, N) = 51.8 (9.8, 134), difference mean (95% CI) = 3.7 (1.1 to 6.2)
Addenbrookes 2	Shoulder movement (mean reduction)	Flexion, extension, abduction, internal rotation, external rotation	Flexion: ALND: mean (SD, N) = 13 (32.9, 141); SLNB: mean (SD, N) = 6.7 (15.6, 134), difference mean (95% CI) = 6.3 (0.1 to 12.6); Extension: ALND: mean (SD, N) = ‐1.5 (10.7, 139); SLNB: mean (SD, N) = ‐2.2 (8.1, 134), difference mean (95% CI) = 0.7 (‐1.5 to 3.3); Abduction: ALND: mean (SD, N) = 6.3 (11.5, 138); SLNB: mean (SD, N) = 3.1 (15.7, 132), difference mean (95% CI) = 3.2 (‐0.5 to 6.3)	Internal rotation: ALND: mean (SD, N) = 1.7 (12.7, 139); SLNB: mean (SD, N) = 0.3 (12, 134), difference mean (95% CI) = 1.4 (‐1.5 to 4.4); External rotation: ALND: mean (SD, N) = 2.9 (12.3, 139); SLNB: mean (SD, N) = 1.5 (11, 134), difference mean (95% CI) = 1.4 (‐1.5 to 4.4)
ALMANAC	Axillary drain usage		ALND: 359/453; SLNB: 75/449
ALMANAC	Infection rate of surgical wounds		ALND: 72/476; SLNB: 52/478
ALMANAC	Lymphoedema	Patient‐assessed; moderate/severe	1 month: ALND: 7/419; SLNB: 1/428 3 months: ALND: 12/395; SLNB: 4/417	6 months: ALND: 13/414; SLNB: 2/432 12 months: ALND: 10/403 SLNB: 4/412
ALMANAC	Lymphoedema	Mean (95% CI) change in arm vol compared with pretreatment	1 month: ALND = 1.022 (1.013‐1.032); SLNB = 1.003 (0.997‐1.01) 3 months: ALND = 1.044 (1.035‐1.053); SLNB = 1.019 (1.01‐1.028)	6 months: ALND = 1.058 (1.048‐1.069); SLNB = 1.022 (1.011‐1.032) 12 months: ALND = 1.061 (1.048‐1.074); SLNB = 1.028 (1.016‐1.039)
ALMANAC	Sensory loss	Median area of sensory loss (cm^2; range)	1 month: ALND = 40 (1‐489); SLNB = 32 (2‐254) 3 months: ALND = 47 (0‐1139); SLNB = 48 (0‐327)	6 months: ALND = 39 (0.4‐2827); SLNB = 32 (0‐201) 12 months: ALND = 35 (0.8‐1013); SLNB = 59 (0.2‐342)	Event rates for self‐assessed sensory loss also reported in Mansel 2006 for these follow‐up periods, but not extracted
ALMANAC	Intercostobrachial nerve damage	Clinician assessment; severe	1 month: ALND: 10/392; SLNB: 6/409 3 months: ALND: 10/373; SLNB: 4/397	6 months: ALND: 10/394; SLNB: 4/410 12 months: ALND: 5/384 SLNB: 5/400
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): flexion	1 month: ALND = 9.8; SLNB = 5.8 3 months: ALND = 3.7; SLNB = 2	6 months: ALND = 1.6; SLNB = 2 12 months: ALND = 0.1; SLNB = 2.7	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): abduction	1 month: ALND = 12.9; SLNB = 6.5 3 months: ALND = 4.2; SLNB = 1.9	6 months: ALND = 2.3; SLNB = 1.5 12 months: ALND = 1.9; SLNB = 2.5	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): external rotation	1 month: ALND = 1.2; SLNB = 0.7 3 months: ALND = 1.2; SLNB = 0.2	6 months: ALND = 1; SLNB = 0.6 12 months: ALND = 0.7; SLNB = 0.6	95% CI can also be extracted
ALMANAC	Shoulder function	Mean change in shoulder function (degrees): internal rotation	1 month: ALND = 0.9; SLNB = 0.4 3 months: ALND = 0.7; SLNB = 1	6 months: ALND = 0.8; SLNB = 0.2 12 months: ALND = 0.4; SLNB = 1.7	95% CI can also be extracted
ALMANAC	Quality of life	Measures: mean trial outcome index; trial outcome index reduced by ≥ 5 points from baseline (n/N); mean arm functioning subscale score; substantial arm swelling or tenderness (n/N); substantial numbness on ipsilateral side (n/N); mean FACT‐B+4 score			Means (95% CI) and event rates can be extracted for each time point (baseline, 1, 3, 6 and 12 months)
ALMANAC	State and trait anxiety				Mean and 95% CI can be extracted for each time point (baseline, 1, 3, 6 and 12 months)
Cardiff	Morbidity	Objective complaints: restricted elevation 720 degrees	Not stated: full axillary surgery, neg nodes = 25% (×2 = 7.47, P < 0.01); no axilary surgery, neg nodes = 0%; full axillary surgery + radical RT, positive nodes = 67%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Objective complaints: oedema of arm, 72 cm	Not stated: full axillary surgery, neg nodes = 46% (×2 = 6.02, P < 0.03); no axillary surgery, neg nodes = 15%; full axillary surgery + radical RT, positive nodes = 58%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Subjective complaints: limited arm movement	Not stated: full axillary surgery, neg nodes = 21%; no axillary surgery, neg nodes = 8%; full axillary surgery + radical RT, positive nodes = 8%; no axillary surgery + local RT = 21%		Sample of 85 patients only from Cardiff site
Cardiff	Morbidity	Subjective complaints: swollen arm	Not stated: full axillary surgery, neg nodes = 43%; no axillary surgery, neg nodes = 23%; full axillary surgery + radical RT, positive nodes = 58%; no axillary surgery + local RT = 37%		Sample of 85 patients only from Cardiff site
Edinburgh 1	Morbidity	Lateral shoulder rotation (mean (SE) difference (cm) from preoperative value (N))	6 months: Sampling + RT: 1.91 (SE = 0.56) (N = 64), sampling ‐ RT: 0.34 (SE = 0.59) (N = 59); ALND: 0.13 (SE = 0.39) (N = 132)	12 months: Sampling + RT: 1.75 (SE = 0.56) (N = 66), Sampling ‐ RT: 0.72 (SE = 0.62) (N = 55); ALND: 0.77 (0.4) (N = 128)	Figure 4, Chetty 2000 paper
Edinburgh 1	Morbidity	Lateral shoulder rotation (mean (SE) difference (cm) from preoperative value (N))	24 months: Sampling + RT: 1.57 (SE = 0.6) (N = 60), Sampling ‐ RT: ‐0.48 (SE = 0.65) (N = 52); ALND: 0.38 (SE = 0.43) (N = 117)	36 months: Sampling + RT: 2.19 (SE = 0.59) (N = 59), Sampling ‐ RT: 0.43 (SE = 0.64) (N = 50); ALND: 0.24 (SE = 0.43) (N = 110)	Figure 4, Chetty 2000 paper
Edinburgh 1	Morbidity	Arm volume (mean (SE) percentage of preoperative arm volume (N))	6 months: Sampling + RT: 100.69 (SE = 0.779) (N = 56), Sampling ‐ RT: 102.04 (SE = 0.766) (N = 58); ALND: 103.57 (SE = 0.519) (N = 126)	12 months: Sampling + RT: 100.95 (SE = 0.81) (N = 59), Sampling ‐ RT: 102.47 (SE = 0.85) (N = 54); ALND: 103.74 (SE = 0.57) (N = 119)	Figure 5, Chetty 2000 paper
Edinburgh 1	Morbidity	Arm volume (mean (SE) percentage of preoperative arm volume (N))	24 months: Sampling + RT: 100.84 (SE = 1.03) (N = 54), Sampling ‐ RT: 100.81 (SE = 1.06) (N = 51); ALND: 104.37 (SE = 0.73) (N = 108)	36 months: Sampling + RT: 100.01 (SE = 1.03) (N = 52), Sampling ‐ RT: 101.28 (SE = 1.07) (N = 48); ALND: 104.07 (SE = 0.73) (N = 103)	Figure 5, Chetty 2000 paper
E'dburgh Sample/Clear	Morbidity	Subjective arm	Not stated; full axillary surgery, positive node (Nil 8/12; intermittent 1/12; persistent 3/12); full axillary surgery, ‐negative node (nil 22/28; intermittent 1/28; persistent 5/28); Sample + RT, positive node (nil 17/28; intermittent 2/28; persistent 9/28); Sample, negative node (nil 23/26; intermittent 1/26; persistent 2/26)		Morbidity data to be included in discussion only; sample chosen from alphabetical pt list of patients free of local or systemic disease
E'dburgh Sample/Clear	Morbidity	Subjective mobility	Not stated; full axillary surgery, positive node (normal 12/12; reduced 0/12); full axillary surgery, negative node (normal 22/28; reduced 6/28); Sample + RT, negative node (normal 12/28; reduced 16/28); Sample, negative node (normal 24/26; reduced 2/26)		See comments in Aitken paper
E'dburgh Sample/Clear	Morbidity	Subjective interference with daily activities	Not stated; full axillary surgery, positive node (nil 12/12; occasional 0/12; severe 0/12); full axillary surgery, negative node (nil 24/28; occasional 4/28; severe 0/28); Sample + RT, positive node (nil 16/28; occasional 8/28; severe 4/28); Sample, negative node (nil 24/26; occasional 4/26; severe 0/26)		See comments in Aitken paper
E'dburgh Sample/Clear	Morbidity	Objective assessment ‐ shoulder joint mobility			See comments in Aitken paper
WSSA Glasgow	Psychological morbidity				Use in discussion only
GIVOM Sentinella	Lymphoedema	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.37 (0.2 to 0.7) 12 months: 0.48 (0.2 to 0.9)	18 months: 0.59 (0.3 to 1.2) 24 months: 0.52 (0.2 to 1.1)
GIVOM Sentinella	Shoulder movement restriction	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.47 (0.3 to 0.8) 12 months: 0.73 (0.4 to 1.4) 12 months: raw data extracted from graph (SLNB 17/336, ALND 23/341)	18 months: 0.62 (0.3 to 1.3) 24 months: 0.44 (0.2 to 1.0)
GIVOM Sentinella	Axillary/arm pain	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.52 (0.3 to 0.8) 12 months: 0.76 (0.5 to 1.3) 12 months: raw data extracted from graph (SLNB 30/336, ALND 39/341)	18 months: 0.84 (0.5 to 1.5) 24 months: 0.90 (0.5 to 1.6)
GIVOM Sentinella	Numbness	Assessed by physician, reported as odds ratio (95% CI): SLNB/ALND	6 months: 0.64 (0.4 to 0.9) 12 months: 0.53 (0.3 to 0.8) 12 months: raw data extracted from graph (SLNB 41/336, ALND 71/341)	18 months: 0.37 (0.2 to 0.6) 24 months: 0.54 (0.3 to 0.9)
GIVOM Sentinella	Winged scapula	Assessed by physician	Study authors report rate too low to analyse
GIVOM Sentinella	Health‐related quality of life: SF‐36 – physical component	Assessed by patients using validated questionnaires	No significant differences found between group means of SF‐36 physical component (Del Bianco, 2008)
GIVOM Sentinella	Health‐related quality of life: SF‐36 – mental component	Assessed by patients using validated questionnaires	No significant differences found between group means of SF‐36 mental component (Del Bianco, 2008)
GIVOM Sentinella	Health‐related quality of life: SF‐36 HRQOL domains	Assessed by patients using validated questionnaires	No significant differences found between groups on all HRQOL domains of SF‐36 (Zavagno, 2008)
GIVOM Sentinella	Health‐related quality of life: psychological general well‐being index	Assessed by patients using validated questionnaires	6, 12 months: significantly better PGWB general and anxiety domain scores in SLNB group than in ALND group (Del Bianco, 2008)	24 months: no significant differences between PGWB general and anxiety domain scores of both groups.(Del Bianco, 2008)
Guy's	Morbidity	Arm function	3 months: ALND: Good: 44/90, Fair: 41/90, Poor: 5/90; No ALND: Good: 59/77, Fair: 18/77, Poor: 0/77	15 months: ALND: Good: 83/100, Fair: 14/100, Poor: 3/100; No ALND: Good: 70/88, Fair: 17/88, Poor: 1/88	Sample only
Guy's	Morbidity	Lymphoedema	3 months: ALND: None: 18/93, Slight: 66/93, Moderate: 6/93, Severe: 3/93; No ALND: None: 36/81, Slight: 43/81, Moderate: 0/81, Severe: 2/81	15 months: ALND: None: 27/104, Slight: 71/104 Moderate: 6/104, Severe: 0/104; No ALND: None: 39/91, Slight: 52/91, Moderate: 0/91, Severe: 0/91	Sample only
Guy's	Morbidity	Activity	3 months: ALND: Good: 45/92, Fair: 46/92, Poor: 1/92; No ALND: Good: 62/80, Fair: 16/80, Poor: 2/80	15 months: ALND: Good: 85/101, Fair: 14/101, Poor: 2/101; No ALND: Good: 78/92, Fair: 13/92, Poor: 1/92	Sample only
Guy's	Morbidity	Attitude	3 months: ALND: Good: 81/92, Fair: 9/92, Poor: 2/92; No ALND: Good: 71/80, Fair: 7/80, Poor: 2/80	15 months: ALND: Good: 91/101, Fair: 8/101, Poor: 2/101; No ALND: Good: 87/92, Fair: 5/92, Poor: 0/92	Sample only
Hammersmith	Postoperative deaths		Radical: 0/95; Simple: 0/100
Hammersmith	Morbidity	Shoulder function	At 4‐year minimum follow‐up in survivors: Radical: 6/95; Simple = 18/100		Consequential morbidity, at time of publication Methodology not reported, all patients included
Hammersmith	Morbidity	Arm swelling (including volumetric measurement of upper limb)	At 4‐year minimum follow‐up in survivors: Radical: 7/95; Simple = 3/100		Consequential morbidity, at time of publication Methodology not reported, all patients included
IBCSG‐10‐93	Lymphoedema	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Arm circumference	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Performance of daily activities	Physician reported	Not significantly different between treatments
IBCSG‐10‐93	Arm pain	Physician reported	Baseline: ALND 5/175, surgery 8/194; 1st postoperative: ALND 38/164, surgery 12/168; 3 months: ALND 16/161, surgery 9/171; 6 months: ALND 17/174, surgery 11/177	9 months: ALND 21/160, surgery 8/164; 12 months: ALND 13/189, surgery 8/190; 18 months: ALND 14/173, surgery 7/183; 24 months: ALND 12/165, surgery 8/164
IBCSG‐10‐93	Restricted arm movement	Physician reported	Baseline: ALND 9/174, surgery 6/194; 1st postoperative: ALND 64/163, surgery 25/168; 3 months: ALND 23/161, surgery 10/170; 6 months: ALND 21/176, surgery 9/176	9 months: ALND 21/160, surgery 7/163; 12 months: ALND 19/188, surgery 6/187; 18 months: ALND 10/171, surgery 7/182; 24 months: ALND 12/165, surgery 7/164
IBCSG‐10‐93	QOL ‐ bothered scores	Patient reported	No significant differences at any time point (baseline, 1st postoperative, 3, 6, 9, 12, 18 and 24 months)
IBCSG‐10‐93	QOL ‐ arm movement scores	Patient reported	At 1st postoperative surgery alone, reported less restriction in use of their arm than ALND (P < .0001). Otherwise, no significant differences
IBCSG‐10‐93	QOL ‐ numbness scores	Patient reported	At 1st postoperative surgery alone, reported less severe postsurgery numbness than ALND (P < .0001). Otherwise, no significant differences
IBCSG‐10‐93	QOL ‐ coping scores	Patient reported	No significant differences at any time point (baseline, 1st postoperative, 3, 6, 9, 12, 18 and 24 months)
IBCSG‐23‐01	Postoperative infection	Physician assessed	Surgery alone: 0/467 ALND: 1/464
IBCSG‐23‐01	Sensory neuropathy	Physician assessed	Any: Surgery alone: 55/453 ALND: 82/447 Grade 3‐4: Surgery alone: 0/453 ALND: 1/447
IBCSG‐23‐01	Lymphoedema	Physician assessed	Defined as long term: Any: Surgery alone: 15/453 ALND: 59/447 Grade 3‐4: Surgery alone: 0/453 ALND: 3/447
IBCSG‐23‐01	Motor neuropathy	Physician assessed	Any: Surgery alone: 13/453 ALND: 37/447 Grade 3‐4: Surgery alone: 1/453 ALND: 3/447
Institut Bergonie	Arm fatigue	Unclear	Moderate/severe: no ALND: N = 4/258; ALND: N = 24/273
Institut Bergonie	Shoulder mobility	Unclear	Restricted somewhat or severely: no ALND: N = 5/257; ALND: N = 21/271
Institut Bergonie	Parasthesia	Unclear	Moderate/severe: no ALND: N = 6/258; ALND: N = 41/274
Institut Bergonie	Lymphoedema	Unclear	Minor/major difference: no ALND: N = 3/258; ALND: N = 29/275
Institut Bergonie	Other functional impairments	Unclear	Minor/major: no ALND: N = 12/263; ALND: N = 16/276
Institut Bergonie	Number of patients with functional impairments	Unclear	Minor: no ALND: N = 23/265; ALND: N = 78/278
IPO‐P	Upper limb circumference > 2 cm	Measured as per definition	6 months: Obs: 6/57; ALND: 10/49 12 months: Obs: 8/57; ALND: 15/49	24 months: Obs: 8/57; ALND: 14/49 48 months: Obs: 4/57; ALND: 19/49
IPO‐P	Pain at rest	Patient reported	6 months: Obs: 9/57; ALND: 9/49 12 months: Obs: 11/57; ALND: 14/49	24 months: Obs: 9/57; ALND: 10/49 48 months: Obs: 3/57; ALND: 7/49
IPO‐P	Parasthesias	Patient reported?	6 months: Obs: 10/57; ALND: 28/49 12 months: Obs: 6/57; ALND: 29/49	24 months: Obs: 5/57; ALND: 34/49 48 months: Obs: 6/57; ALND: 30/49
IPO‐P	Shoulder dysfunction	Measured as per definition	6 months: Obs: 5/57; ALND: 5/49 12 months: Obs: 4/57; ALND: 8/49	24 months: Obs: 0/57; ALND: 6/49 48 months: Obs: 2/57; ALND: 11/49
Milan	Morbidity	Axillary pain (sporadic/continuous)	6 months: ALND: 91/100; SNLB = 16/100	24 months: ALND: 39/100; SNLB = 8/100
Milan	Morbidity	Numbness/Parasthesia on operated side	6 months: ALND: 85/100; SNLB = 2/100	24 months: ALND: 68/100; SNLB = 1/100
Milan	Morbidity	Arm mobility, 80%‐100%	6 months: ALND: 73/100; SNLB = 100/100	24 months: ALND: 79/100; SNLB = 100/100
Milan	Morbidity	Arm mobility, 60%‐79%	6 months: ALND: 22/100; SNLB = 0/100	24 months: ALND: 18/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, 40%‐59%	6 months: ALND: 5/100; SNLB = 0/100	24 months: ALND: 2/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, 20%‐39%	6 months: ALND: 0/100; SNLB = 0/100	24 months: ALND: 1/100; SNLB = 0/100
Milan	Morbidity	Arm mobility, < 20%	6 months: ALND: 0/100; SNLB = 0/100	24 months: ALND: 0/100; SNLB = 0/100
Milan	Morbidity	Aesthetic appearance of axillary scar: bad	6 months: ALND: 9/100; SNLB = 2/100	24 months: ALND: 15/100; SNLB = 0/100
Milan	Morbidity	Arm swelling < 1 cm difference in circumference	6 months: ALND: 44/100; SNLB = 11/100	24 months: ALND: 38/100; SNLB = 6/100
Milan	Morbidity	Arm swelling 1‐2 cm difference in circumference	6 months: ALND: 17/100; SNLB = 0/100	24 months: ALND: 25/100; SNLB = 1/100
Milan	Morbidity	Arm swelling >2 cm difference in circumference	6 months: ALND: 8/100; SNLB = 0/100	24 months: ALND: 12/100; SNLB = 0/100
Milan	Morbidity	Arm swelling, any	6 months: ALND: 69/100; SNLB = 11/100	24 months: ALND: 75/100; SNLB = 7/100
NSABP B‐04	Arm oedema	Arm swelling ≥ 2 cm difference in circumference	No. of patients with data: ALND: N = 577; no ALND + RT: N = 568 no ALND: N = 312 both node + and node‐ patients. Final measurement was 2 to 5 years after surgery Arm oedema recorded at least once: ALND: 58.1%; no ALND + RT: 38.2%; no ALND: 39.1% (P < 0.001) Oedema always: ALND: 3.6%; no ALND + RT: 0.9%; no ALND: 1% No measurement after first oedema: ALND: 9.2%; no ALND + RT: 5.8%; no ALND: 3.2% Oedema always after first oedema: ALND: 6.1%; no ALND + RT: 3.2%; no ALND: 2.6% Intermittent, final measurement oedema: ALND: 11.8%; no ALND + RT: 4.9%; no ALND: 8.6%; Total with oedema on final measurement: ALND: 30.7%; no ALND + RT: 14.8%; no ALND: 15.4% (P < 0.001)	Oedema once, then resolution: ALND: 15.9%; no ALND + RT: 15.3%; no ALND: 16.7% Intermittent, final measurement no oedema: ALND: 11.4%; no ALND + RT: 8.1%; no ALND: 7.1% Total with no oedema on final measurement (after at least 1 measurement of oedema): ALND: 27.3%; no ALND + RT: 23.4%; no ALND: 23.8 Arm oedema ≥ 4 cm difference in circumference recorded at least once: ALND: 21.5%; no ALND + RT: 11.4%; no ALND: 13.1%
NSABP B‐32	Adverse events (grade 3 or greater surgery related)	No details reported	ALND: 14/2788 SLN: 12/2800 Must include most of SLN positive and negative patients		Peri‐surgery
NSABP B‐32	Arm mobility/shoulder abduction deficit (objective)	Physician assessed	6 months: < 5%: ALND: 1299/1667; SLN: 1468/1744 5%‐10%: ALND: 218/1667; SLN: 176/1744 ≥ 10%: ALND: 150/1667; SLN: 99/1744
NSABP B‐32	Arm volume difference (objective)	Physician assessed	6 months: < 5%: ALND: 1187/1677; SLN: 1363/1759 5%‐10%: ALND: 277/1677; SLN: 236/1759 ≥ 10%: ALND: 211/1677; SLN: 158/1759	12 months: < 5%: ALND: 1170/1639; SLN: 1345/1705 5%‐10%: ALND: 252/1639; SLN: 215/1705 ≥ 10%: ALND: 216/1639; SLN: 147/1705	These data are also available for 18 and 30 months
NSABP B‐32	Arm volume difference (objective)	Physician assessed	24 months: < 5%: ALND: 1062/1517; SLN: 1184/1504 5%‐10%: ALND: 243/1517; SLN: 197/1504 ≥ 10%: ALND: 212/1517; SLN: 123/1504	36 months: < 5%: ALND: 990/1421; SLN: 1156/1459 5%‐10%: ALND: 227/1421; SLN: 194/1459 ≥ 10%: ALND: 203/1421; SLN: 109/1459	These data are also available for 18 and 30 months
NSABP B‐32	Tingling (subjective)	Self‐reported	6 months: ALND (N = 388/1693), SLN (N = 184/1766) 12 months: ALND (N = 305/1640), SLN (N = 158/1713) 18 months: ALND (N = 272/1566), SLN (N = 138/1638)	24 months: ALND (N = 236/1521), SLN (N = 137/1588) 30 months: ALND (N = 219/1448), SLN (N = 116/1502) 36 months: ALND (N = 193/1431), SLN (N = 110/1463)
NSABP B‐32	Numbness (subjective)	Self‐reported	6 months: ALND (N = 821/1693), SLN (N = 257/1769) 12 months: ALND (N = 679/1641), SLN (N = 216/1713) 18 months: ALND (N = 592/1567), SLN (N = 174/1638)	24 months: ALND (N = 554/1523), SLN (N = 157/1587) 30 months: ALND (N = 473/1450), SLN (N = 137/1504) 36 months: ALND (N = 445/1430), SLN (N = 119/1463)
NSABP B‐32	Shoulder abduction deficit ≥ 5% (in those with < 5% at baseline)	Physician assessed	6 months: ALND (N = 275/1449), SLN (N = 201/1519)
NSABP B‐32	Shoulder abduction deficit ≥ 5% (in those with < 5% at baseline)	Physician assessed	36 months: ALND (N = 314/1136), SLN (N = 192/1151)
NSABP B‐32	Numbness (in those with none at baseline)	Self‐reported	36 months: ALND (N = 407/1336), SLN (N = 103/1371)
NSABP B‐32	Tingling (in those with none at baseline)	Self‐reported	36 months: ALND (N = 175/1329), SLN (N = 90/1343)
Ostersund	Seroma	Patients with percutaneous aspiration in outpatient department	ALND: 17/50; sampling: 10/50		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Postoperative discharge (mL), median (range)		ALND: 250 (25‐1610); sampling: 130 (0‐1785)		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Duration of postop drainage (days) (median, range)		ALND: 4 (1‐11); sampling: 2.1 (1 ‐11)		Adverse events reported only for the 1987‐89 sample; i.e. for N = 100/200
Ostersund	Arm volume increase	≥ 10%	ALND: 14/47; sampling: 0/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Subjective sensation of swelling in women without objective increase in arm volume	Any	ALND: 12/33; sampling: 9/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Shoulder mobility (mean decrease compared with baseline)		7.5° decrease for whole sample of 95 patients		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Axillary paraesthesia (impairment of sensibility in the axilla)		ALND: 17/48; sampling: 19/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
Ostersund	Inner upper arm paraesthesia (impairment of sensibility in the inner upper arm)		ALND: 24/48; sampling: 4/48		Adverse events reported only for the 1987‐89 sample; i.e. for ca N = 100/200
SE Scotland	Delayed healing		ALND: 27/100; Simple + RT: 8/100
SE Scotland	Haematoma		ALND: 24/100; Simple + RT: 6/100
SE Scotland	Infection		ALND: 9/100; Simple + RT: 6/100
SE Scotland	DVT		ALND: 4/100; Simple + RT: 1/100
SE Scotland	Pulmonary embolism		ALND: 1/100; Simple + RT: 1/100
SE Scotland	Chest infection		ALND: 6/100; Simple + RT: 3/100
SE Scotland	Severe skin reaction		ALND: 0/100; Simple + RT: 5/100
SE Scotland	Nausea and vomiting		ALND: 0/100; Simple + RT: 2/100
SE Scotland	Tracheitis		ALND: 0/100; Simple + RT: 2/100
SE Scotland	Skin grafts		ALND: 10/100; Simple + RT: 0/100
SE Scotland	Arm oedema		ALND: 10/100; Simple + RT: 5/100
SE Scotland	Limitation of shoulder movement		ALND: 4/100; Simple + RT: 14/100
SNAC	Haematoma	Any	ALND: 30/539; SLNB: 38/544
SNAC	Seroma	Any	ALND: 195/539; SLNB: 93/544
SNAC	Infection	Any	ALND: 73/539; SLNB: 48/544
SNAC	Arm morbidity	Mean changes in arm morbidity (patient reported, overall summary average score of 15 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 7 (N = 457); SLNB: 4.4 (N = 456) 1 month: ALND: 2.2 (0.2); SLNB: 1.4 (0.15) 6 months: ALND: 1.1 (0.2); SLNB: 0.8 (0.15) 12 months: ALND: 1.05 (0.2); SLNB: 0.8 (0.15)	24 months: ALND: 1.05 (0.2); SLNB: 0.75 (0.15) 36 months: ALND: 1.05 (0.2); SLNB: 0.7 (0.15)
SNAC	Arm symptoms	Mean changes in arm symptoms (patient reported, average of 7 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 9.7 (N = 457); SLNB: 5.5 (N = 456) 1 month: ALND: 2.1 (0.2); SLNB: 1.2 (0.1) 6 months: ALND: 1.3 (0.15); SLNB: 0.8 (0.1) 12 months: ALND: 1.25 (0.15); SLNB: 0.7 (0.1)	24 months: ALND: 1.25 (0.15); SLNB: 0.7 (0.1) 36 months: ALND: 1.2 (0.2); SLNB: 0.65 (0.15)
SNAC	Arm swelling	Mean changes in arm swelling (patient reported, 1 item; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 7.3 (N = 457); SLNB: 3.4 (N = 456) 1 month: ALND: 1.25 (0.2); SLNB: 0.75 (0.15) 6 months: ALND: 0.9 (0.15); SLNB: 0.55 (0.1) 12 months: ALND: 0.95 (0.15); SLNB: 0.45 (0.1)	24 months: ALND: 1 (0.2); SLNB: 0.55 (0.15) 36 months: ALND: 1 (0.2); SLNB: 0.55 (0.15)
SNAC	Arm dysfunctions	Mean arm dysfunctions change (patient reported, average of 3 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 5.5 (N = 457); SLNB: 3.6 (N = 456) 1 month: ALND: 1.9 (0.15); SLNB: 1.35 (0.15) 6 months: ALND: 0.8 (0.1); SLNB: 0.65 (0.1) 12 months: ALND: 0.75 (0.1); SLNB: 0.6 (0.1)	24 months: ALND: 0.7 (0.1); SLNB: 0.55 (0.1) 36 months: ALND: 0.8 (0.1); SLNB: 0.5 (0.1)
SNAC	Arm disabilities	Mean arm disabilities (patient‐reported change, average of 4 items; unclear if it is SEM or SD reported) from baseline	Node+ and node‐ patients: average of measures taken at 6 and 12 months: ALND: 3.4 (N = 457); SLNB: 2.9 (N = 456) 1 month: ALND: 2.2 (0.2); SLNB: 1.4 (0.15) 6 months: ALND: 0.75 (0.1); SLNB: 0.55 (0.1) 12 months: ALND: 0.65 (0.1); SLNB: 0.45 (0.1)	24 months: ALND: 0.6 (0.1); SLNB: 0.5 (0.1) 36 months: ALND: 0.7 (0.1); SLNB: 0.45 (0.1)
SNAC	Arm volume	Increase in arm volume (percentage change from clinician ratings from baseline; unclear if it is SEM or SD reported)	Average of measures taken at 6 and 12 months: ALND: 4.2% (N = 509); SLNB: 2.8% (N = 519) All patients: 1 month: ALND: 0.8% (0.4); SLNB: 0.9% (0.4), P = 0.67 6 months: ALND: 3.5% (0.8); SLNB: 2.4% (0.7), P = 0.02 12 months: ALND: 4.6% (0.8); SLNB: 3% (0.8), P = 0.001 Node‐negative patients: 1 month: ALND: 0.8% (0.4); SLNB: 0.3% (0.4), P = 0.16 6 months: ALND: 3.5% (0.8); SLNB: 1.9% (0.5), P = 0.004 12 months: ALND: 4.6% (0.8); SLNB: 2.2% (0.7), P = 0.001	All patients: 24 months: ALND: 5.8% (1); SLNB: 3.9% (0.7), P = 0.006 36 months: ALND: 5.8% (1); SLNB: 4.0% (1), P = 0.02 Node‐negative patients: 24 months: ALND: 5.8% (1); SLNB: 3% (0.7), P = 0.001 36 months: ALND: 5.8% (1); SLNB: 3.1% (1), P= 0.004
SNAC	Arm volume	Number with an increase in arm volume ≥ 15% (percentage change from clinician ratings from baseline)	All patients: 1 month: ALND: 5/544; SLNB: 3/544 6 months: ALND: 29/544; SLNB:21/544 12 months: ALND: 47/544; SLNB: 29/544 (P = 0.02) Node‐negative patients only: 1 month: ALND: 4/363; SLNB: 1/356 6 months: ALND: 16/363; SLNB: 9/356 12 months: ALND: 28/363; SLNB: 13/356 (P = 0.02)	All patients: 24 months: ALND: 81/544; SLNB: /544 (P = 0.001) 36 months: ALND: 82/544; SLNB: /544 (P = 0.01) Node‐negative patients only: 24 months: ALND: 47/363; SLNB: 25/356 (P = 0.01) 36 months: ALND: 49/363; SLNB: 25/356 (P = 0.006)
SNAC	Lateral abduction	Lateral abduction (change from clinician ratings from baseline; degrees; unclear if it is SEM or SD reported ‐ have assumed it is SEM for calculations)	Average of measures taken at 6 and 12 months (percentage change from baseline: ALND: 4.4% (N = 509); SLNB: 2.5% (N = 519) Node+ and node‐ patients (read off graph): Baseline: ALND: 158 (1); SLNB: 157 (1) 1 month: ALND: 131 (2); SLNB: 144 (2) 6 months: ALND: 150 (1); SLNB: 151 (1) 12 months: ALND: 150 (1); SLNB: 151 (1)	Node+ and node‐ patients (read off graph): 24 months: ALND: 151 (1); SLNB: 152 (1) 36 months: ALND: 150 (1); SLNB: 151 (1)
SNAC	Forward flexion	Forward flexion (degrees; unclear if it is SEM or SD reported ‐ have assumed it is SEM for calculations)	Node+ and node‐ patients (read off graph): Baseline: ALND: 157 (1); SLNB: 158 (1) 1 month: ALND: 137 (2); SLNB: 148 (1.5) 6 months: ALND: 150 (1); SLNB: 152 (1) 12 months: ALND: 151 (1); SLNB: 151 (1)	Node+ and node‐ patients (read off graph): 24 months: ALND: 152 (1); SLNB: 152 (1) 36 months: ALND: 152 (1); SLNB: 151 (1)
Xu 2003	Postoperative swelling (oedema)	Measurement of arm diameter	Level I clearance: 3/93 ALND: 7/88
Xu 2003	Involved upper limb disorder	Unclear	Level I clearance: 0/93 ALND: 0/88
Xu 2003	Cerebrovascular accident	Unclear	Level I clearance: 0/93 ALND: 2/88
Xu 2003	Cardiovascular events	Unclear	Level I clearance: 2/93 ALND: 1/88

Table 3. Morbidity data at each time point

Comparison 1. No axillary surgery versus full axillary surgery

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 All‐cause mortality (radiotherapy subgroups) Show forest plot	10	3849	Hazard Ratio (95% CI)	1.06 [0.96, 1.17]

1.1 no radiotherapy	1	773	Hazard Ratio (95% CI)	0.96 [0.80, 1.15]
1.2 radiotherapy	9	3076	Hazard Ratio (95% CI)	1.11 [0.98, 1.25]
2 All‐cause mortality (extra treatment for positive node subgroups) Show forest plot	10	3849	Hazard Ratio (95% CI)	1.06 [0.96, 1.17]

2.1 additional treatment for node‐positive patients	3	1174	Hazard Ratio (95% CI)	1.51 [1.09, 2.09]
2.2 no specific additional treatment for node‐positive patients	7	2675	Hazard Ratio (95% CI)	1.02 [0.92, 1.13]
3 Locoregional recurrence (radiotherapy subgroups) Show forest plot	4	20863	Hazard Ratio (95% CI)	2.35 [1.91, 2.89]

3.1 no radiotherapy	1	7284	Hazard Ratio (95% CI)	2.94 [2.05, 4.23]
3.2 radiotherapy	3	13579	Hazard Ratio (95% CI)	2.11 [1.64, 2.72]
4 Locoregional recurrence (extra treatment for positive‐node subgroups) Show forest plot	4	20863	Hazard Ratio (95% CI)	2.35 [1.91, 2.89]

4.1 additional treatment for node‐positive patients	1	4171	Hazard Ratio (95% CI)	1.10 [0.69, 1.75]
4.2 no specific additional treatment for node‐positive patients	3	16692	Hazard Ratio (95% CI)	2.83 [2.25, 3.57]
5 Distant metastasis Show forest plot	2	946	Hazard Ratio (95% CI)	1.06 [0.87, 1.30]

5.1 no radiotherapy	1	727	Hazard Ratio (95% CI)	1.10 [0.89, 1.35]
5.2 radiotherapy	1	219	Hazard Ratio (95% CI)	0.64 [0.28, 1.42]
6 Lymphoedema (≥ 12 months postop) ‐ fixed‐effect model Show forest plot	4	1714	Odds Ratio (M‐H, Fixed, 95% CI)	0.31 [0.23, 0.43]

6.1 additional treatment for node‐positive patients	1	532	Odds Ratio (M‐H, Fixed, 95% CI)	0.07 [0.02, 0.22]
6.2 no additional treatment for node‐positive patients	3	1182	Odds Ratio (M‐H, Fixed, 95% CI)	0.39 [0.28, 0.54]
7 Lymphoedema (≥ 12 months postop) ‐ random‐effects model Show forest plot	4	1714	Odds Ratio (M‐H, Random, 95% CI)	0.22 [0.08, 0.57]

7.1 additional treatment for node‐positive patients	1	532	Odds Ratio (M‐H, Random, 95% CI)	0.07 [0.02, 0.22]
7.2 no additional treatment for node‐positive patients	3	1182	Odds Ratio (M‐H, Random, 95% CI)	0.40 [0.28, 0.55]
8 Arm or shoulder movement impairment (≥ 12 months postop) Show forest plot	5	1495	Odds Ratio (M‐H, Fixed, 95% CI)	0.72 [0.49, 1.05]

8.1 radiotherapy	5	1495	Odds Ratio (M‐H, Fixed, 95% CI)	0.72 [0.49, 1.05]
9 Pain (≥ 12 months postop) Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

9.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
10 Paraesthesia (≥ 12 months postop) Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

10.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
11 Delayed healing Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

11.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
12 Skin graft Show forest plot	1		Odds Ratio (M‐H, Random, 95% CI)	Totals not selected

12.1 radiotherapy	1		Odds Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]
13 All‐cause mortality (allocation concealment subgroups) Show forest plot	10	3849	Hazard Ratio (95% CI)	1.06 [0.96, 1.17]

13.1 adequate allocation concealment	4	1442	Hazard Ratio (95% CI)	0.98 [0.81, 1.18]
13.2 unclear or inadequate allocation concealment	6	2407	Hazard Ratio (95% CI)	1.09 [0.97, 1.23]

Comparison 1. No axillary surgery versus full axillary surgery

Comparison 2. Axillary sampling versus full axillary surgery

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 All‐cause mortality Show forest plot	3	967	Hazard Ratio (95% CI)	0.94 [0.73, 1.21]

1.1 radiotherapy	2	872	Hazard Ratio (95% CI)	0.84 [0.64, 1.11]
1.2 no radiotherapy	1	95	Hazard Ratio (95% CI)	1.47 [0.84, 2.56]
2 Local recurrence Show forest plot	3	1404	Hazard Ratio (95% CI)	1.41 [0.94, 2.12]

2.1 radiotherapy	2	659	Hazard Ratio (95% CI)	1.40 [0.89, 2.19]
2.2 no radiotherapy	1	745	Hazard Ratio (95% CI)	1.48 [0.58, 3.82]
3 Axillary recurrence Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

4 Locoregional recurrence Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

4.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
4.2 no radiotherapy	0		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
5 Distant metastasis Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

5.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
5.2 no radiotherapy	0		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
6 Lymphoedema. Increase in arm circumference (≥ 12 months postop) Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7 Shoulder lateral rotation (12 months postop) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

7.1 radiotherapy	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
8 Seroma Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

8.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]

Comparison 2. Axillary sampling versus full axillary surgery

Comparison 3. Sentinel node biopsy versus full axillary surgery

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 All‐cause mortality Show forest plot	3	6352	Hazard Ratio (95% CI)	1.05 [0.89, 1.25]

1.1 radiotherapy	2	6127	Hazard Ratio (95% CI)	1.05 [0.88, 1.25]
1.2 no radiotherapy	1	225	Hazard Ratio (95% CI)	1.30 [0.35, 4.84]
2 Local recurrence Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

2.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
2.2 no radiotherapy	0		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
3 Axillary recurrence Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

3.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
4 Locoregional recurrence Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

4.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
4.2 no radiotherapy	0		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
5 Distant metastasis Show forest plot	1		Hazard Ratio (95% CI)	Totals not selected

5.1 radiotherapy	1		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
5.2 no radiotherapy	0		Hazard Ratio (95% CI)	0.0 [0.0, 0.0]
6 Lymphoedema. Increase in arm circumference (≥ 12 months postop) Show forest plot	3		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6.1 radiotherapy	3		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7 Lymphoedema. Patient reported (at 12 or more months postop) Show forest plot	3		Odds Ratio (Fixed, 95% CI)	0.33 [0.23, 0.47]

7.1 adequate allocation concealment	2		Odds Ratio (Fixed, 95% CI)	0.33 [0.22, 0.48]
7.2 unclear allocation concealment	1		Odds Ratio (Fixed, 95% CI)	0.36 [0.15, 0.86]
8 Shoulder flexion (12 months postop) Show forest plot	3	2257	Mean Difference (IV, Fixed, 95% CI)	1.55 [‐0.19, 3.29]

8.1 radiotherapy	3	2257	Mean Difference (IV, Fixed, 95% CI)	1.55 [‐0.19, 3.29]
9 Shoulder abduction (12 months postop) Show forest plot	3	2252	Mean Difference (IV, Fixed, 95% CI)	‐1.02 [‐2.79, 0.75]

9.1 radiotherapy	3	2252	Mean Difference (IV, Fixed, 95% CI)	‐1.02 [‐2.79, 0.75]
10 Shoulder internal rotation (12 months postop) Show forest plot	2	1227	Mean Difference (IV, Fixed, 95% CI)	0.50 [‐1.10, 2.09]

10.1 radiotherapy	2	1227	Mean Difference (IV, Fixed, 95% CI)	0.50 [‐1.10, 2.09]
11 Shoulder external rotation (12 months postop) Show forest plot	2	1227	Mean Difference (IV, Fixed, 95% CI)	‐0.56 [‐2.21, 1.09]

11.1 radiotherapy	2	1227	Mean Difference (IV, Fixed, 95% CI)	‐0.56 [‐2.21, 1.09]
12 Subjective arm movement impairment (≥ 12 months postop) Show forest plot	2	877	Odds Ratio (M‐H, Fixed, 95% CI)	0.38 [0.22, 0.67]

12.1 radiotherapy	2	877	Odds Ratio (M‐H, Fixed, 95% CI)	0.38 [0.22, 0.67]
13 Pain (≥ 12 months postop) Show forest plot	2	877	Odds Ratio (M‐H, Fixed, 95% CI)	0.44 [0.30, 0.67]

13.1 radiotherapy	2	877	Odds Ratio (M‐H, Fixed, 95% CI)	0.44 [0.30, 0.67]
14 Paraesthesia (≥ 12 months postop) Show forest plot	2	495	Odds Ratio (M‐H, Fixed, 95% CI)	0.15 [0.09, 0.23]

14.1 radiotherapy	2	495	Odds Ratio (M‐H, Fixed, 95% CI)	0.15 [0.09, 0.23]
15 Numbness (≥ 12 months postop) Show forest plot	3	1799	Odds Ratio (M‐H, Fixed, 95% CI)	0.43 [0.34, 0.54]

15.1 radiotherapy	3	1799	Odds Ratio (M‐H, Fixed, 95% CI)	0.43 [0.34, 0.54]
16 Seroma Show forest plot	2	1381	Odds Ratio (M‐H, Fixed, 95% CI)	0.40 [0.31, 0.51]

16.1 radiotherapy	2	1381	Odds Ratio (M‐H, Fixed, 95% CI)	0.40 [0.31, 0.51]
17 Wound infection Show forest plot	2	2074	Odds Ratio (M‐H, Fixed, 95% CI)	0.65 [0.50, 0.85]

17.1 radiotherapy	2	2074	Odds Ratio (M‐H, Fixed, 95% CI)	0.65 [0.50, 0.85]
18 Brachial plexus injury at 6 months postop Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

18.1 radiotherapy	1		Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]

Comparison 3. Sentinel node biopsy versus full axillary surgery

Comparison 4. Radiotherapy versus full axillary surgery

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 All‐cause mortality Show forest plot	4	2469	Hazard Ratio (95% CI)	1.10 [1.00, 1.21]

2 Local recurrence Show forest plot	4	22256	Hazard Ratio (95% CI)	0.80 [0.64, 0.99]

3 Distant metastasis Show forest plot	1	1313	Hazard Ratio (95% CI)	1.07 [0.93, 1.25]

4 Lymphoedema. Increase in arm circumference (≥ 12 months postop) Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

5 Delayed healing Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6 Wound infection Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

7 Skin graft Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

8 Haematoma Show forest plot	1		Odds Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 4. Radiotherapy versus full axillary surgery

Comparison 5. Less surgery versus ALND

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 All‐cause mortality Show forest plot	19	12864	Hazard Ratio (95% CI)	1.08 [1.01, 1.16]

1.1 no axillary surgery vs ALND	9	3076	Hazard Ratio (95% CI)	1.11 [0.98, 1.25]
1.2 axillary sampling vs ALND	3	967	Hazard Ratio (95% CI)	0.94 [0.73, 1.21]
1.3 SLNB vs ALND	3	6352	Hazard Ratio (95% CI)	1.05 [0.89, 1.25]
1.4 radiotherapy vs ALND	4	2469	Hazard Ratio (95% CI)	1.10 [1.00, 1.21]
2 All‐cause mortality (radiotherapy subgroups) Show forest plot	19	13637	Hazard Ratio (95% CI)	1.07 [1.00, 1.14]

2.1 radiotherapy (same in both groups)	13	10075	Hazard Ratio (95% CI)	1.06 [0.96, 1.16]
2.2 radiotherapy (in less surgery group only)	4	2469	Hazard Ratio (95% CI)	1.10 [1.00, 1.21]
2.3 no radiotherapy	3	1093	Hazard Ratio (95% CI)	1.00 [0.85, 1.19]
3 All‐cause mortality (additional treatment for histologically positive nodes) Show forest plot	5	1708	Hazard Ratio (95% CI)	0.90 [0.72, 1.14]

3.1 additional treatment for histologically positive nodes	4	1613	Hazard Ratio (95% CI)	0.82 [0.64, 1.05]
3.2 no additional treatment for histologically positive nodes	1	95	Hazard Ratio (95% CI)	1.47 [0.84, 2.56]
4 Local recurrence Show forest plot	8	24176	Hazard Ratio (95% CI)	0.90 [0.75, 1.09]

4.1 axillary sampling vs ALND	3	1404	Hazard Ratio (95% CI)	1.41 [0.94, 2.12]
4.2 SLNB vs ALND	1	516	Hazard Ratio (95% CI)	0.94 [0.24, 3.77]
4.3 radiotherapy vs ALND	4	22256	Hazard Ratio (95% CI)	0.80 [0.64, 0.99]
5 Locoregional recurrence Show forest plot	6	26880	Hazard Ratio (95% CI)	1.53 [1.31, 1.78]

5.1 no axillary surgery vs ALND	4	20863	Hazard Ratio (95% CI)	2.35 [1.91, 2.89]
5.2 axillary sampling vs ALND	1	406	Hazard Ratio (95% CI)	0.74 [0.46, 1.20]
5.3 SLNB vs ALND	1	5611	Hazard Ratio (95% CI)	0.96 [0.74, 1.24]
6 Distant metastasis Show forest plot	3	2665	Hazard Ratio (95% CI)	1.07 [0.95, 1.20]

6.1 no axillary surgery vs ALND	2	946	Hazard Ratio (95% CI)	1.06 [0.87, 1.30]
6.2 axillary sampling vs ALND	1	406	Hazard Ratio (95% CI)	1.05 [0.74, 1.49]
6.3 radiotherapy vs ALND	1	1313	Hazard Ratio (95% CI)	1.07 [0.93, 1.25]
7 Lymphoedema. Increase in arm volume at 12 months postop Show forest plot	9	3964	Odds Ratio (M‐H, Fixed, 95% CI)	0.37 [0.29, 0.46]

7.1 no axillary surgery vs ALND	4	1714	Odds Ratio (M‐H, Fixed, 95% CI)	0.31 [0.23, 0.43]
7.2 axillary sampling vs ALND	1	85	Odds Ratio (M‐H, Fixed, 95% CI)	0.32 [0.13, 0.81]
7.3 SLNB vs ALND	3	1965	Odds Ratio (M‐H, Fixed, 95% CI)	0.48 [0.33, 0.69]
7.4 radiotherapy vs ALND	1	200	Odds Ratio (M‐H, Fixed, 95% CI)	0.47 [0.16, 1.44]
8 Paraesthesia (≥ 12 months postop) Show forest plot	3	1027	Odds Ratio (M‐H, Fixed, 95% CI)	0.14 [0.10, 0.21]

8.1 no axillary surgery vs ALND	1	532	Odds Ratio (M‐H, Fixed, 95% CI)	0.14 [0.06, 0.32]
8.2 SLNB vs ALND	2	495	Odds Ratio (M‐H, Fixed, 95% CI)	0.15 [0.09, 0.23]
9 Pain (≥ 12 months postop) Show forest plot	3	1256	Odds Ratio (M‐H, Fixed, 95% CI)	0.47 [0.32, 0.68]

9.1 no axillary surgery vs ALND	1	379	Odds Ratio (M‐H, Fixed, 95% CI)	0.60 [0.24, 1.47]
9.2 SLNB vs ALND	2	877	Odds Ratio (M‐H, Fixed, 95% CI)	0.44 [0.30, 0.67]
10 Delayed healing Show forest plot	2	404	Odds Ratio (M‐H, Fixed, 95% CI)	0.25 [0.13, 0.46]

10.1 no axillary surgery vs ALND	1	204	Odds Ratio (M‐H, Fixed, 95% CI)	0.27 [0.11, 0.67]
10.2 radiotherapy vs ALND	1	200	Odds Ratio (M‐H, Fixed, 95% CI)	0.24 [0.10, 0.55]
11 Seroma Show forest plot	3	1481	Odds Ratio (M‐H, Fixed, 95% CI)	0.40 [0.32, 0.52]

11.1 SLNB vs ALND	2	1381	Odds Ratio (M‐H, Fixed, 95% CI)	0.40 [0.31, 0.51]
11.2 axillary sampling vs ALND	1	100	Odds Ratio (M‐H, Fixed, 95% CI)	0.49 [0.20, 1.20]
12 Wound infection Show forest plot	3	2274	Odds Ratio (M‐H, Fixed, 95% CI)	0.65 [0.50, 0.84]

12.1 SLNB vs ALND	2	2074	Odds Ratio (M‐H, Fixed, 95% CI)	0.65 [0.50, 0.85]
12.2 radiotherapy vs ALND	1	200	Odds Ratio (M‐H, Fixed, 95% CI)	0.65 [0.22, 1.89]
13 Skin graft Show forest plot	2	404	Odds Ratio (M‐H, Fixed, 95% CI)	0.15 [0.04, 0.57]

13.1 no axillary surgery vs ALND	1	204	Odds Ratio (M‐H, Fixed, 95% CI)	0.39 [0.07, 2.19]
13.2 radiotherapy vs ALND	1	200	Odds Ratio (M‐H, Fixed, 95% CI)	0.04 [0.00, 0.74]
14 Haematoma Show forest plot	2	1283	Odds Ratio (M‐H, Fixed, 95% CI)	0.80 [0.53, 1.20]

14.1 SLNB vs ALND	1	1083	Odds Ratio (M‐H, Fixed, 95% CI)	1.27 [0.78, 2.09]
14.2 radiotherapy vs ALND	1	200	Odds Ratio (M‐H, Fixed, 95% CI)	0.20 [0.08, 0.52]

Comparison 5. Less surgery versus ALND