Nipple‐ and areola‐sparing mastectomy for the treatment of breast cancer

Summary of findings for the main comparison. Nipple‐sparing mastectomy for breast cancer treatment

Nipple‐sparing mastectomy for breast cancer treatment
Patient or population: women with breast cancer Setting: breast cancer therapy centres Intervention: nipple‐sparing mastectomy Comparison: modified radical mastectomy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
	Risk with other types of mastectomies	Risk with nipple‐sparing mastectomy
Overall survival	Study population		HR 0.72 (0.46 to 1.13)	1202 (2 observational studies)	⊕⊝⊝⊝ VERY LOW^1,2	We were unable to conclude non‐inferiority because of the high risk of confounding. The quality of evidence was downgraded to very low due to the risk of bias and imprecision.
	882 per 1000	785 per 1000 (626 to 910)
Local recurrence	Study population		HR 0.28 (0.12 to 0.68)	1311 (2 observational studies)	⊕⊝⊝⊝ VERY LOW ³	Conclusions could not be drawn because of the high risk of residual confounding. The quality of the evidence was graded as very low due to the risk of bias.
	17 per 1000	5 per 1000 (2 to 12)
Overall complications	Study population		RR 0.10 (0.01 to 0.82)	1067 (2 observational studies)	⊕⊕⊝⊝ LOW ⁴	We downgraded the quality of evidence from low to very low due to inconsistency (i.e. the magnitude of effects across the studies).
	184 per 1000	18 per 1000 (2 to 151)
Skin necrosis	Study population		RR 4.22 (0.59 to 30.03)	1948 (4 observational studies)	⊕⊝⊝⊝ VERY LOW ^5,6	We downgraded the quality of evidence from low to very low due to a wide confidence interval.
	19 per 1000	82 per 1000 (11 to 583)
Infection	Study population		RR 0.95 (0.44 to 2.09)	496 (2 observational studies)	⊕⊕⊝⊝ LOW
	48 per 1000	45 per 1000 (21 to 100)
*The risk in the intervention group (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; RR: Risk ratio; OR: Odds ratio; HR: Hazard ratio
GRADE Working Group grades of evidence High quality: we are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹There was no adjustment for confounding in Sakurai 2013 and we classified this as a serious risk of bias. We we were unable to conclude non‐inferiority because of the high risk of confounding ²The confidence interval of included studies both increased and decreased risk for nipple‐sparing mastectomy versus modified radical mastectomy ³Imprecision was downgraded because of the high risk of residual confounding due to lack of adjustment for tumour stage and radiotherapy. Adam 2014 matched the participants by tumour stage, Horiguchi 2001 seemed to have more initial tumours (Stages 0 and 1) in the NSM group (83 out of 123; 67.5%) than the MRM group (277 out of 910; 30.4%) and this fact may have influenced the results. Only 7.9% (103 out of 1303 participants) received radiotherapy, 7.2% (81 out of 1113 participants) in the modified radical mastectomy group and in 11.6% (22 out of 190 participants) in the nipple‐sparing mastectomy. No one received the treatment in Horiguchi 2001, probably because in that period the indication for radiotherapy treatment was more limited. Probably more patients would now receive post‐mastectomy radiotherapy, which would have an impact on the local recurrence rate ⁴There was evidence of considerable heterogeneity on statistical testing, I² = 85%, P = 0.01 ⁵There was evidence of considerable heterogeneity on statistical testing, I² = 79% , P = 0.008 ⁶Confidence intervals fail to exclude both clinically important and clinically unimportant harms.

Summary of findings 2. Nipple‐sparing mastectomy for breast cancer treatment

Nipple‐sparing mastectomy for breast cancer treatment
Patient or population: women with breast cancer Setting: breast cancer therapy centres Intervention: nipple‐sparing mastectomy Comparison: skin‐sparing mastectomy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
	Risk with other types of mastectomies	Risk with nipple‐sparing mastectomy
Overall survival	Study population		HR 0.70 (0.28 to 1.72)	781 (2 observational studies)	⊕⊝⊝⊝ VERY LOW ¹
	928 per 1000	841 per 1000 (521 to 989)
*The risk in the intervention group (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; RR: Risk ratio; OR: Odds ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹We downgraded the quality of evidence due to: (a) risk of bias of included studies: non‐randomised studies with low quality, and imprecision: wide confidence intervals that crossed null effect and comprised both directions of effect

Background

available in

Description of the condition

Breast cancer is the most frequent non‐skin cancer in women (23% of all cancers in women) with an estimated 1.67 million new cases and over 521,907 deaths reported worldwide in 2014 (Ferlay 2012).

Description of the intervention

The technique of subcutaneous mastectomy for the treatment of benign breast disease for women with a strong family history of breast cancer was first reported by Freeman 1962. In 1980, Gentil 1980 proposed nipple‐sparing mastectomy (NSM) for prophylactic contralateral mastectomy and to treat breast cancer. The traditional method of mastectomy consists of the removal of the skin that overlies the breast including the nipple‐areola complex. Today, NSM is commonly used for women who are considered to be high risk and who are undergoing surgery as primary prevention for breast cancer (Hartmann 2001; Josephson 2000; Lostumbo 2010; Pennisi 1989).

How the intervention might work

The surgical management of breast carcinoma has evolved during the last two decades and improvements in the techniques mean that more conservative surgery and better cosmetic results can be achieved without compromising oncological safety (Morrow 2002). Although the techniques for breast conserving surgery in the treatment of breast cancer are well‐established, many women prefer or require mastectomy to obtain local control of their disease. This is particularly true in cases of multifocal tumours and/or small volume breasts in relation to the tumour, extensive ductal carcinoma in situ, women with clinical contraindications for radiotherapy and treatment of local recurrences (Singletary 2003). Conventional surgical therapy in these situations is mastectomy with the removal of the nipple‐areola complex. This type of surgery has a cumulative incidence of local recurrence of about 2.3% after 20 years (Veronesi 1990).

Rising interest in improved cosmesis (i.e. cosmetic outcome) has led to the introduction of NSM or areola‐sparing mastectomy as an alternative to radical mastectomy (Chung 2008; Gerber 2009; Simmons 2002). Nipple‐areola complex preservation results in higher psychological satisfaction and the perception of less mutilation among women (Loewen 2008).

Occult nipple involvement in breast cancer ranges from 0% to 58% (Andersen 1979; Banerjee 2008; Lagios 1979; Laronga 1999; Loewen 2008; Luttges 1987; Menon 1989; Morimoto 1985; Parry 1977; Quinn 1981; Rusby 2008; Santini 1989; Schecter 2006; Smith 1976; Verma 1997; Vyas 1998; Wetheim 1980) and areola involvement is 0.9% in people with tumours less than 2 cm (Simmons 2002). This wide range may be explained by differences in the thoroughness of the pathological examinations and the study methods. The factors most commonly associated with the pathological involvement of the nipple are the size and region of the tumours, the distance of the tumours from the nipple, and axillary metastasis (Benediktsson 2008; Caruso 2006; Lagios 1979; Schecter 2006).

Despite the promising approach with nipple‐sparing and areola‐sparing mastectomy, the evidence from published studies seems preliminary. These studies include a small number of participants and a relatively short follow‐up period (Benediktsson 2008; Caruso 2006; Gerber 2009; Horiguchi 2001; Petit 2009a; Sacchini 2006). In these non‐randomised studies, the incidence of local recurrence ranges from 1.6% to 28% without radiotherapy (Benediktsson 2008; Caruso 2006), and 1.4% to 8.5% with radiation of the nipple‐areola complex (Benediktsson 2008; Petit 2009a; Petit 2009b). Surgical complications such as skin necrosis (i.e. death of localised tissue or cells) have been described in up to 11% of patients without radiotherapy (Sacchini 2006) and up to 15% with the addition of radiotherapy to the nipple‐areola complex (Petit 2009a).

Why it is important to do this review

NSM has been proposed for the treatment of breast cancer. This technique retains the entire natural envelope of the skin and areola complex, and aims to create an aesthetic result that is closer to the natural state than breast reconstruction techniques. The efficacy and effectiveness of NSM in the treatment of breast cancer is questionable, but it is estimated that local recurrence rates are very similar to those for breast‐conserving surgery followed by radiotherapy (Veronesi 1990; Veronesi 2002). No systematic reviews or articles have been published that have assessed the relevant studies with regards to their internal and external validity, and the risk of bias. Therefore, a systematic review on this topic is warranted.

Objectives

available in

To assess the efficacy and safety of nipple‐sparing mastectomy and areola‐sparing mastectomy for the treatment of ductal carcinoma in situ and invasive breast cancer in women.

Methods

available in

Criteria for considering studies for this review

Types of studies

We searched for randomised controlled trials (RCTs) as they provide the highest level of evidence. As no RCTs were found, we expanded our criteria to include non‐randomised comparative studies (cohort and case‐control studies).

Types of participants

Women with a diagnosis of ductal carcinoma in situ or invasive breast cancer, regardless of age, time of onset, or disease stage. The diagnosis must have been in accordance with the histopathological criteria of the World Health Organization (WHO) (Lakhani 2012).

We included women who had undergone breast reconstructive surgery.

Types of interventions

Nipple‐sparing mastectomy (NSM) (that is, removal of all glandular breast tissue and preservation of the entire skin, nipple and areola) and areola‐sparing mastectomy (that is, removal of all glandular breast tissue and nipple, and preservation of the entire skin and areola) compared with conventional mastectomy (removal of the skin that overlies the breast including the nipple and areola, also known as modified radical mastectomy (MRM)) for the treatment of ductal carcinoma in situ or invasive breast cancer, regardless of any adjuvant therapy.

We included studies where women underwent breast reconstructive surgery.

Types of outcome measures

Primary outcomes

Overall survival, considered separately for ductal carcinoma in situ and early breast cancer if possible.

Secondary outcomes

Local recurrence incidence rate and time‐to‐recurrence during the follow‐up period, considered separately for ductal carcinoma in situ and early breast cancer if possible.

Adverse events: local surgical complications and overall complications (systemic surgical complications, e.g. thromboembolic events)

- Local Complications

- - Explantation of implant/expander
  - Hematoma
  - Seroma
  - Rehospitalization
  - Skin necrosis (nipple, areola or flap necrosis)
  - Skin necrosis with revision surgery
  - Infection

Cosmetic results: participants' and professionals' opinions
Quality of life including satisfaction with the decision to have NSM, satisfaction with the cosmetic outcome, satisfaction with the medical process, psychological well‐being, impact on body image, and impact on primary relationships and sexuality.

Search methods for identification of studies

See: Breast Cancer Group for search methods used in reviews.

There were no language restrictions on included studies. We undertook full translations of all non‐English language papers using local resources.

Electronic searches

We searched the following databases.

The Cochrane Breast Cancer Group's Specialised Register. The Cochrane Breast Cancer Group searched their Specialised Register on 30 September 2014. Details of the search strategies used by the Group for the identification of studies and the procedure used to code references are outlined in the Group’s module (onlinelibrary.wiley.com/o/cochrane/clabout/articles/BREASTCA/frame.html). We extracted trials with the key words ’surgery’, ’nipple sparing mastectomy’, ’areola sparing mastectomy’ and ’breast conserving surgery’
The Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2014, issue 9) (Appendix 1) on 30 September 2014.
MEDLINE (via Pubmed) (Appendix 2) on 30 September 2014.
Embase (via OVID) (Appendix 3) on 30 September 2014.
LILACS (via Biblioteca Virtual em Saúde (BVS)) (Appendix 4) on 30 September 2014.
The WHO International Clinical Trials Registry Platform (ICTRP) search portal (www.who.int/ictrp/en/) for all prospectively registered and ongoing trials (Appendix 5) on 30 September 2014.
ClinicalTrials.gov (www.clinicaltrials.gov/) (Appendix 6) on 30 September 2014.

Searching other resources

Bibliography Searching

We searched bibliographies of all included studies and review papers in order to identify other potentially suitable studies. We obtained a copy of the full article for each reference reporting a potentially eligible trial, where possible. We contacted the study authors to provide additional information when it was not available.

Unpublished Literature

We contacted experts in this field and sent letters to all authors of included studies requesting information on unpublished data or ongoing studies.

Data collection and analysis

Selection of studies

Two review authors (BS, RR) independently examined the titles and abstracts of articles identified in the search as potentially relevant trials. From this initial assessment, we obtained full versions of all potentially relevant articles. We consulted a third review author (JLBB) to help to resolve any disagreements.

Data extraction and management

We extracted and recorded the data onto data extraction forms which we had developed for this review. Two review authors (BS and RR) independently undertook full data extraction and consulted a third review author (JLBB) to help resolve disagreements. We sought unpublished data concerning outcomes of interest from study authors by letter as stated above.

We included the following information from individual studies on data extraction forms:

publication details;
study design, study setting, inclusion/exclusion criteria;
patient population (e.g. age, type of surgical procedure, histological classification);
details of intervention;
outcome measures; and
withdrawals, length and method of follow‐up and the number of participants followed up.

For non‐randomised studies we also recorded the following information:

methods used to control for confounders;
adjusted and unadjusted outcome measures.

Assessment of risk of bias in included studies

We applied the Cochrane tool for assessing risk of bias as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a), the Cochrane EPOC Group’s risk of bias criteria (Cochrane EPOC Group 2013), and recommendations by Norris 2013. Two review authors (BS, RR )independently assessed the methodological quality of each study and risk of bias for the following domains: selection bias, performance/detection bias, attrition bias, reporting bias and other bias. For each risk of bias domain and its associated specific questions outlined below, we assigned either ’High risk’, ’Low risk’, or ’Unclear risk’.

Selection Bias

Was the allocation sequence adequately generated?

Scored “Low risk” if a random component in the sequence generation process was described (e.g. referring to a random number table).
Scored “High risk” when a non‐random method was used (e.g. performed by date of admission). Non‐randomised studies should be scored “High risk”.
Scored “Unclear risk” if not specified in the paper.

Was the allocation adequately concealed?

Scored “Low risk” if participants or investigators enrolling participants could not foresee assignment (e.g. because a centralised randomisation scheme, an on‐site computer system or sealed opaque envelopes were used).
Scored “High risk” if participants or investigators enrolling participants could possibly foresee assignments. Non‐randomised studies were scored “High risk”.
Scored “Unclear risk” if not specified in the paper.

For non‐randomised studies, we also considered the following questions.

Were baseline characteristics similar?

Scored “Low risk” if baseline characteristics of the study and control group were reported and similar. Important baseline characteristics were adjuvant radiotherapy, age, surgical techniques, stage of disease, ductal carcinoma in situ or invasive breast cancer, and chemotherapy. We considered intervention and control groups similar for categorical variables if category membership agreed within no more than a 2 percentage point difference between groups. We considered mean ages within two years and tumour sizes within 1 cm similar. When we used statistical tests to compare baseline characteristics between groups, we took statistical significance into account, but as the study may have been underpowered due to small sample sizes, we also considered the magnitude of the difference.
Scored “Unclear risk” if it was not clear in the paper (e.g. characteristics were mentioned in the text, but no data were presented).
Scored “High risk” if there was no report of characteristics in text or tables or if there were differences between control and intervention groups.

Was there adequate adjustment for confounding?

Scored “Low risk” if appropriate methods were used to adjust for potential confounding (e.g. adjuvant radiotherapy, age, surgical techniques, stage of disease, ductal carcinoma in situ or invasive breast cancer, and chemotherapy).
Scored "Unclear risk" if the methods used to adjust for confounding were not reported in the paper.
Scored “High risk” if no appropriate methods were used to adjust for potential confounding.

Performance/detection bias

Was knowledge of the allocated interventions adequately prevented during the study?

Scored “Low risk” if the study authors stated explicitly that the primary outcome variables were assessed blindly, or the outcomes were objective, e.g. overall survival, hospitalisation time
Scored “High risk” if the outcomes were not assessed blindly
Scored “Unclear risk” if not specified in the paper

Attrition bias

Were incomplete outcome data adequately addressed?

Scored “Low risk” if missing outcome measures were unlikely to bias the results (e.g. reasons for missingness unlikely to be related to true outcome, missing outcome data balanced and small numbers across study groups with similar reasons for missing data or missing data were imputed using appropriate methods).
Scored “High risk” if missing outcome data were likely to bias results.
Scored “Unclear risk” if not specified in the paper.

Reporting bias

Were reports of the study free from selective outcome reporting?

Scored “Low risk” if there was no evidence that outcomes were selectively reported (e.g. the study had a protocol pre‐specifying the outcomes, or all relevant outcomes described in the methods section were reported in the results section).
Scored “High risk” if some pre‐specified outcomes were subsequently omitted from the results.
Scored “Unclear risk” if not specified in the paper.

Were reports of the study free from selective analysis reporting?

Scored “Low risk” for each outcome if there was no evidence that analyses were selectively reported (e.g. analyses were defined in the methods section of the protocol or paper).
Scored “High risk” if there was evidence of selective analysis reporting (e.g. multiple adjusted analyses were carried out and only one reported, or unusual cut‐points were used for categorizing an outcome).
Scored “Unclear” risk if unclear from the paper.

Classification of study designs

We included various study designs and defined them as follows:

Prospective cohort study: a group of exposed and non‐exposed individuals who were followed over time to compare incidence (or rate of death from disease) between the groups (Gordis 1996). In prospective cohort studies, the recruitment, exposure/intervention, and outcomes must all have occurred after setting up the study.
Retrospective cohort study: a group of exposed and non‐exposed individuals who were followed over time to compare incidence (or rate of death from disease) between the groups (Gordis 1996). In retrospective cohort studies, outcomes could have occurred prior to setting up the study or collected afterwards, or both.
Case‐control study: a study that compared people with a specific outcome of interest (cases) with people from the same source population but without the outcomes (controls), to examine the association between the outcome and exposure.

Measures of treatment effect

We reported time‐to‐event outcomes (e.g. overall survival and local recurrence) as hazard ratios (HRs) with 95% confidence intervals (CIs). Where necessary we estimated HRs using the methods of Parmar 1998.

We reported dichotomous outcomes (e.g. distant disease, explantation of implant/expander, hematoma, seroma, rehospitalization, skin necrosis (nipple, areola or flap necrosis), skin necrosis (nipple, areola or flap necrosis) with revision surgery, infection and cosmetic results) as risk ratios (RRs) with 95% CIs. Participants reported as lost‐to‐follow‐up were excluded from the analyses.

We reported continuous outcomes (e.g. quality of life) as mean differences (MDs) or standardised mean differences (SMD) with 95% CIs.

Considering the current approach for ductal carcinoma in situ or invasive breast cancer, it seemed reasonable to consider the treatment effect for the primary outcome as a non‐inferiority question. Thus, the non‐inferiority bound for the HR was 1.13 (based on a 10‐year death rate of 24% for conventional mastectomy with a cut‐off value of 27% for nipple/areola‐sparing mastectomy, 1.13 = 0.27/0.24). For local recurrence the non‐inferiority bound for the HR was 2.67 (based on 3% 10‐year local recurrence for conventional mastectomy with a cut‐off value of 8% for nipple/areola‐sparing mastectomy, 2.67 = 0.08/0.03). Non‐inferiority could be claimed if the upper limit of the two‐sided 95% confidence interval was less than the non‐inferiority bound.

Outcomes relating to adverse events were assessed for superiority.

Unit of analysis issues

The unit of analysis was the individual participant.

Two studies were three‐arm studies. The three surgical interventions in the Gerber 2009 and Kim 2010 studies contributed to the NSM versus SSM comparison and also the NSM versus MRM comparison.

Dealing with missing data

Where data were missing or unsuitable for analysis (e.g. intention‐to‐treat analysis was not used), we contacted the study authors to request further information as indicated in the Characteristics of included studies table. Where data were missing to the extent that the study could not be included in the meta‐analysis, and attempts to retrieve data were exhausted, we presented the results in the review and discussed them in the context of the findings.

Assessment of heterogeneity

We assessed statistical heterogeneity using the Chi² statistic (P value less than 0.1). We also assessed heterogeneity between studies using the I² statistic to examine the percentage of total variation across studies due to heterogeneity rather than chance (Higgins 2003). An I² value of 30% to 60% may represent moderate heterogeneity, while values greater than 50% may be considered substantial heterogeneity (Deeks 2011).

We investigated the following factors as potential causes of heterogeneity in the included studies using the framework below.

Clinical diversity: included study location and setting, full characteristics of participants, co‐morbidity and treatments that participants were probably receiving on trial entry. We considered how outcomes were measured, the definition of outcomes, and how they were recorded. Depending upon the extent of the clinical diversity, we either analysed studies separately or presented the results using a narrative approach.
Methodological diversity: included assessment of the randomisation process, study quality, and analytical method.

Assessment of reporting biases

We searched for protocols of included trials using PubMed (National Library of Medicine) and through the UK and other trial registries, where possible. We contacted study authors to attempt to establish a full data set or reasons for the non‐reporting of certain outcomes as outlined in the Characteristics of included studies table.

Data synthesis

We synthesised data using Cochrane’s statistical software, Review Manager (RevMan) (RevMan 2014).

We based the choice of using a fixed‐effect or random‐effects model for data synthesis on the extent of the heterogeneity. Where substantial clinical or methodological heterogeneity existed, we used a random‐effects model. Otherwise we used a fixed‐effect model (Deeks 2011).

We combined data using the inverse variance method and the log‐HR for time‐to‐event outcomes, the log‐RR for dichotomous outcomes and the MD for continuous outcomes. For random‐effects meta‐analysis we used the DerSimonian and Laird method (Deeks 2011).

Where the data were too diverse for combining effect sizes in a meaningful or valid manner, we presented the results of individual studies in table and graphical format and used a narrative approach to summarise the data.

We used the criteria of the GRADE working group to evaluate the evidence. The quality of evidence for each outcome was classified as high, moderate, low or very low quality. The classification criteria considered the study design, the risk of bias, the inconsistency of data, subjectivity (indirectness, or indirect evidence), imprecision and publication bias (Guyatt 2008).

Subgroup analysis and investigation of heterogeneity

As we expected a small number of published studies, we anticipated that it would not be possible to perform subgroup analyses.

However, we will consider the following subgroups in future updates of this review.

Participants who received adjuvant radiotherapy versus participants who did not have radiotherapy.
Younger (< 50 years ) versus older (≥ 50 years) women.
Surgical techniques.
Cancer stage based on the TNM (the size and/or extent of the primary tumour (T), the amount of spread to nearby lymph nodes (N), and the presence of metastases (M) or secondary tumours) classification system (NCI 2013).
Systemic therapy (chemotherapy or endocrine therapy) versus no systemic therapy.

Sensitivity analysis

We planned to conduct sensitivity analyses by excluding either studies of low methodological quality or, if RCT evidence was available, quasi‐randomised studies. This was not possible due to a lack of good quality studies, but will be considered for future updates of this review.

Results

Description of studies

Results of the search

See: Figure 1.

Figure 1

Study flow diagram

Based on our search strategy, we identified and screened 5130 references, with an additional six references identified from other sources. After removing duplicates, we screened the title and abstracts of 3662 references. Of these, we discarded 3610 records and assessed 52 full‐text articles. We excluded 41 articles due to being either case series or duplicate data. Eleven cohort studies met the inclusion criteria for this review. There were no RCTs or quasi‐RCTs.

Included studies

We included 11 studies in the review (Adam 2014; Boneti 2011; Burdge 2013; Gerber 2009; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014). See: Characteristics of included studies table.

Study design

All studies were cohort studies; 10 were retrospective (Adam 2014; Boneti 2011; Burdge 2013; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014) and one was prospective (Gerber 2009).

Characteristics of participants

In total, the studies included 6502 participants involving 7018 procedures, as 516 participants underwent bilateral surgery. Out of these, 2529 participants underwent nipple‐sparing mastectomy (NSM), while there were no participants who had areola‐sparing mastectomy, 818 participants underwent skin‐sparing mastectomy (SSM) and 3671 underwent a modified radical mastectomy (MRM).

The mastectomy indications were 99.2% for invasive breast carcinoma or ductal carcinoma in situ and 0.8% for prophylaxis, that mostly involved the contralateral breast after invasive breast cancer. It was possible to identify the indications for NSM in 10 studies.

The criteria for mastectomy varied across the included studies. In six studies, the criteria comprised any tumour without skin or areola involvement (Boneti 2011; Burdge 2013; Kim 2010; Poruk 2015; Sakurai 2013; Shi 2012). Four studies considered tumours located at least 2 cm from the nipple‐areola complex (Adam 2014; Gerber 2009; Oura 1994; Stanec 2014).

Eleven studies classified the stage of the tumour. Eight studies used the AJCC classification (Gerber 2009; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014). In total, 387 participants had Stage 0, 1992 participants had stage 1; 2776 participants had stage 2; 655 participants had stage 3 and 11 participants had stage 4 tumours (refer to Table 1). In the three remaining studies, one study (Burdge 2013) enrolled participants with, at least, stage 2 (NSM: mean tumour size was 3.4 cm (± 2.2 cm) and SSM: mean tumour size was 4.6 cm (± 2.9 cm)). The Boneti 2011 study described the mean tumour size and there was no difference between groups (the mean tumour size was 1.9 cm (± 1.6 cm) in the NSM group and 2.1 cm (± 1.7 cm) in the SSM group, P = 0.42 in SSM group) and Adam 2014 described the tumour size according to TNM staging (NSM: T1 ‐ 37 participants (53.6%), T2 ‐ 14 participants (20.3%), T3 ‐ three participants (4.3%), T4 ‐ one participant (1.4%); MRM T1 ‐ 111 participants (53.9%), T2 ‐ 46 participants (22.3%), T3 ‐ nine participants (4.4%), T4 ‐ one participant (0.5), P = 0.86). In two studies there were significant differences in staging between groups with more advanced disease in the SSM group (Poruk 2015) and MRM group (Horiguchi 2001) when compared to NSM. In six of the remaining studies there were differences between intervention and control groups of more than 2 percentage points for at least one tumour stage (Gerber 2009; Kim 2010; Oura 1994; Sakurai 2013; Shi 2012; Stanec 2014) (Table 2). Only two studies used appropriate methods to adjust for tumour stage (Kim 2010; Poruk 2015), while Adam 2014 matched participants in the intervention and control groups according to their tumour stage.

Table 1. Tumour stage by study

Study	Stage 0	Stage 1	Stage 2	Stage 3	Stage 4
Adam 2014	Study did not use the AJCC classification
Boneti 2011	Study did not use the AJCC classification
Burdge 2013	Study did not use the AJCC classification
Gerber 2009	0	22	168	13	0
Horiguchi 2001	71	289	581	102	0
Kim 2010	173	752	1274	311	0
Oura 1994	0	368	125	0	0
Poruk 2015	34	70	77	34	6
Sakurai 2013	26	361	399	140	5
Shi 2012	10	0	81	14	0
Stanec 2014	73	130	71	41	0
Total	387	1992	2776	655	11

AJCC: American Joint Committee on Cancer

Table 2. Number of participants by tumour stage and intervention

Study	Stage 0	Stage 1	Stage 2	Stage 3	Stage 4
	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM
Adam 2014	This study did not use the AJCC classification
Boneti 2011	This study did not use the AJCC classification
Burdge 2013	This study did not use the AJCC classification
Gerber 2009¹		11(18.3);11(22.9);35 (26.9)	44(73.3);36(75.0);88(67.7)	5(8.3);1(2.1);7(5.4)	0;0;0
Horiguchi 2001²	19(14.3); 0;52(5.7)	64(48.1); 0;225(24.7)	48(36.1); 0; 533(58.6)	2(1.5); 0; 100(11.0)	0;0;0
Kim 2010³	19(12.5);65(17.7); 89(4.5)	70(46.1); 150 (40.8);532(26.7)	55(36.2);121(32.9);1098(55.2)	8(5.3);32(8.7)271(13.6)	0;0;0
Oura 1994	0;0;0	244(81.6); 0; 124(63.9)	55(18.4); 0; 70(36.1)	0;0;0	0;0;0
Poruk 2015⁴	24(23.3); 10 (8.5); 0	36(35.0); 34(28.8); 0	34(33.0); 43(36.4); 0	7(6.8); 27(22.9);0	2(1.9);4(3.4); 0
Sakurai 2013	21(2.7); 0; 5(3.5)	304(38.6); 0; 57(39.6)	364(46.3); 0; 35(24.3)	95(12.1); 0; 45(31.3)	3(0.4); 0; 2(1.4)
Shi 2012	2 (5.7); 0; 8 (8.0)	10 (28.6); 0; 20(20)	18 (51.4); 0; 63 (63.0)	5(14.3); 0; 9(9.0)	0;0;0
Stanec 2014	51(19.8); 22(22.9); 0	97(37.7);33(34.4);0	88(34.2); 21(21.9); 0	21(8.2); 20 (20.8); 0	0;0;0
Total	136(7.4); 97(15.4);154(4.4)	836(45.8);228(36.2);993(28.6)	706(38.7);221(35.1);1887(54.4)	143(7.8);80(12.7); 432(12.5)	5(0.3); 4(0.6); 2(0.1)

Numbers are given by absolute values with percentages in brackets.
AJCC: American Joint Committee on Cancer
MRM: modified radical mastectomy
NSM: nipple‐sparing mastectomy
SSM: skin‐sparing mastectomy

¹Stage 0 and 1 have been combined; P = 0.47
²P < 0.01
³P = 0.21
⁴P = 0.001

The mean age was described in 10 studies while one study (Horiguchi 2001) described the age groups as younger or older than 50 years (see Table 3). Two studies reported similar ages between intervention and control groups (Adam 2014; Boneti 2011). Three studies reported significant differences between groups (Gerber 2009; Horiguchi 2001; Poruk 2015) with the NSM group being younger on average. The remaining five studies reported more than two years' difference in mean age between intervention and control groups (Burdge 2013; Oura 1994; Sakurai 2013; Shi 2012; Stanec 2014). Only three studies used appropriate methods to adjust for tumour stage disease (Horiguchi 2001; Kim 2010; Poruk 2015). One study matched participants in the intervention and control groups according to three age categories (Adam 2014).

Table 3. Mean age of participants by intervention

Age	NSM	SSM	MRM	Comments
Adam 2014	49 (24‐74)	‐	48.5 (21‐87)	P = 0.38
Boneti 2011	51.2 (10.9)	53,1 (11.5)	‐	P = 0.24
Burdge 2013	48.1 (10.4); (29‐75)	53.9 (10.4); (29‐73)	‐
Gerber 2009	46 (10)	48 (10)	58 (6)	P = 0.001
Horiguchi 2001	<= 50 y: 57 >50 y: 76	‐	<= 50 y: 504 50 y >: 406	P < 0.01
Kim 2010	41.5 (7.4)	42.8 (6.6)	data not available	P = 0.06
Oura 1994	Stage 1: 49 Stage 2: 48	‐	Stage 1: 56 Stage 2: 54
Poruk 2015	45 (12)	55 (14.5)	‐	P < 0.001
Sakurai 2013	51 (25‐89)	‐	58 (31‐88)
Shi 2012	35.6	‐	50.8
Stanec 2014	50.9	53.2	‐

MRM: modified radical mastectomy
NSM: nipple‐sparing mastectomy
SSM: skin‐sparing mastectomy

Interventions

Five studies compared NSM and MRM (Adam 2014;Horiguchi 2001; Oura 1994; Sakurai 2013; Shi 2012), four studies compared NSM and SSM (Boneti 2011; Burdge 2013; Poruk 2015; Stanec 2014) and two studies contained three arms involving NSM, SSM and MRM (Gerber 2009; Kim 2010).

Regarding adjuvant therapy, only two studies described chemotherapy (Gerber 2009; Poruk 2015). Poruk 2015 reported that use of both neoadjuvant and adjuvant chemotherapy was higher in the SSM group (neoadjuvant: 17%: 22/131 participants and adjuvant: 57%: 74/131 participants) compared to the group receiving NSM (neoadjuvant: 8.5%: 11/130 participants and adjuvant: 44.5%: 57/130; P = 0.04 and 0.05, respectively). There was no difference between NSM and SSM groups for neo‐ or adjuvant chemotherapy in Gerber 2009: 90% (43/48 participants) in the NSM group versus 88% (53/60) in the SSM group, but it was lower in the MRM group: 83% (109/130).

Radiotherapy was a co‐intervention in seven studies. In five studies, radiotherapy was performed according to tumour stage (Adam 2014; Boneti 2011; Gerber 2009; Poruk 2015; Stanec 2014). In the remaining two studies, all participants received radiotherapy (Burdge 2013; Shi 2012). Only two studies compared radiotherapy between intervention groups. Poruk 2015 reported 28% of radiotherapy treatment (36 out of 130 participants) for the NSM group versus 50.4% (65 out of 131 participants) for the SSM group (P < 0.001). Gerber 2009 did not find a difference in radiotherapy treatment between the NSM and SSM groups: 29% (14 out of 48 participants) and 27% (16 out of 60 participants) respectively, but it was lower for the MRM group: 24% (31 out of 130 participants).

In three studies, participants did not receive radiotherapy (Horiguchi 2001; Oura 1994; Sakurai 2013).

Follow‐up

Six studies had a mean follow‐up of 60 months or greater (Gerber 2009; Horiguchi 2001; Kim 2010; Oura 1994; Sakurai 2013; Shi 2012). The shortest follow‐up ranged from 3 to 102 months (mean: 25.3 months; Boneti 2011) while the longest ranged from 0 to 231 months (mean: 87 months; Sakurai 2013). Boneti 2011 used methods to adjust for differential follow‐up between intervention and control groups; this study had a mean follow‐up of 25.3 months for the NSM group and 38.2 months for the SSM group (P < 0.001). (See Table 4.)

See: Summary of findings for the main comparison Nipple‐sparing mastectomy for breast cancer treatment; Summary of findings 2 Nipple‐sparing mastectomy for breast cancer treatment

Table 4. Mean follow‐up (in months) by intervention

Follow‐up (months)	NSM	SSM	MRM	Comments
Adam 2014	35	‐	36
Boneti 2011	25.3 (18.8) 3‐102	38.2 (26.3) 4‐144	‐	P < 0.001
Burdge 2013	25 (18.8)	38 (26.3)	‐
Gerber 2009	101 (32‐126)	101 (32‐126)	101 (32‐126)
Horiguchi 2001	66	‐	81
Kim 2010	60	67	not mentioned
Oura 1994	Stage 1: 49 Stage 2: 38	‐	Stage 1: 85 Stage 2: 73
Poruk 2015	25.8 (18)	29.9 (15.7)	‐	P = 0.86
Sakurai 2013	87 (10‐252)	‐	87 (0‐231)
Shi 2012	68 (10‐104)	‐	68 (10‐104)
Stanec 2014	43	58	‐

MRM: modified radical mastectomy
NSM: nipple‐sparing mastectomy
SSM: skin‐sparing mastectomy

Excluded studies

Of the 41 excluded studies, all were case series with two being duplicate reports. See Excluded studies.

Risk of bias in included studies

The risk of bias for each of the selected studies is shown in Figure 2. Overall, we identified 11 relevant studies (11 cohort ) published between 2001 and 2014. The major implication for risk of bias assessment was selection bias due to the lack of random assignment to intervention or control. None of the studies performed adequate adjustment for confounding, and there were differences in baseline characteristics between intervention and comparison groups for some studies.

Figure 2

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

Allocation

The most common source of potential bias was selection bias. All studies were cohort studies; 10 were retrospective (Adam 2014; Boneti 2011; Burdge 2013; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014) while one was prospective (Gerber 2009). Therefore none had an allocation sequence adequately generated and concealed, and so all were judged as having a high risk of bias.

Differences in baseline characteristics between mastectomy groups were observed for nine studies (Burdge 2013; Gerber 2009; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014). For two studies (Adam 2014; Boneti 2011) the stage of disease and age at baseline of participants were similar, but differences in chemotherapy and radiotherapy between groups were not reported.

None of the studies used appropriate techniques to deal with all the potential confounders that were pre‐specified for this review. Some studies partially adjusted for confounding. Adam 2014 matched participants for age, tumour stage and follow up, but made no adjustment for adjuvant radiotherapy, surgical techniques or chemotherapy. Horiguchi 2001 adjusted for age, lymph node status and oestrogen receptor status for local recurrence, but did not adjust for tumour stage, surgical techniques or chemotherapy. Kim 2010 adjusted overall survival for age, tumour stage and oestrogen receptor status, but did not adjust for radiotherapy use, surgical techniques or chemotherapy use. Poruk 2015 controlled for age, stage, and surgery laterality, but did not adjust for radiotherapy or chemotherapy use. The remaining studies made no adjustment for confounding.

Blinding

Blinding of participants and surgeons was not possible because of the nature of the procedure. The lack of blinding may have influenced the results for some outcomes. The time to event outcomes were judged as low risk of bias. Adverse events, cosmesis and quality of life were judged as high risk of bias due to the lack of blinding.

Incomplete outcome data

Three studies reported loss‐to‐follow‐up. Stanec 2014 reported almost 12%, which may have been high enough to influence results, so the risk of bias was considered to be unclear, Boneti 2011 reported 5% and Gerber 2009 reported 3%. The low rates of attrition in Boneti 2011 and Gerber 2009 were unlikely to have affected their results and they were classified at low risk of bias. It was impossible to determine if the remaining studies had any missing outcome measures (Adam 2014; Burdge 2013; Horiguchi 2001; Kim 2010; Oura 1994; Poruk 2015; Sakurai 2013; Shi 2012). These studies and Stanec 2014 were considered to have unclear risk of bias.

Selective reporting

Ten of the eleven studies (Adam 2014; Boneti 2011; Burdge 2013; Gerber 2009; Horiguchi 2001; Kim 2010; Poruk 2015; Sakurai 2013; Shi 2012; Stanec 2014) appeared to be free from selective outcome and analysis reporting, because the defined statistical methods and outcomes in the methods section of the study were employed and reported in the results section.

Other potential sources of bias

We did not identify any other potential sources of bias.

Effects of interventions

The 11 cohort studies, involving 13 treatment comparisons, enrolled 6502 participants. Four studies included participants who underwent either nipple‐sparing mastectomy (NSM) or skin‐sparing mastectomy (SSM) (Boneti 2011; Burdge 2013; Poruk 2015; Stanec 2014), five studies included participants who underwent either NSM or modified radical mastectomy (MRM) (Adam 2014; Horiguchi 2001; Oura 1994; Sakurai 2013; Shi 2012) and two studies involved three treatment comparisons, that is, NSM, SSM and MRM, and therefore these latter two studies contributed data for the analysis related to NSM versus SSM, and NSM versus MRM (Gerber 2009; Kim 2010).

We tried to analyse the outcomes of overall survival and local recurrence in two different comparisons: NSM versus SSM, and NSM versus MRM. When this was not possible (e.g. for adverse effects), we drew the comparison between NSM versus other types of mastectomy (i.e. SSM or MRM, or both).

For most studies, we could not calculate the HRs because outcomes were not reported in sufficient detail. Many studies only reported outcomes at the end of follow‐up, rather than at specific time points.

It was not possible to stratify the analyses for participants with DCIS and those with invasive breast cancer, as had originally been planned, because insufficient data were available.

Overall survival

Overall survival was evaluated in four of the 11 studies (Adam 2014; Kim 2010; Poruk 2015; Sakurai 2013).

Nipple‐sparing mastectomy versus skin‐sparing mastectomy

Two studies involving 781 participants evaluated overall survival for NSM versus SSM (Kim 2010; Poruk 2015), both of which had adjusted for some confounders including age and tumour stage. The HR was 0.70 (95% CI 0.28 to 1.72; 2 studies; 781 participants; Analysis 1.1; Figure 3) with no heterogeneity across studies (I² = 0%; P =0.65). The confidence interval is too wide to draw any conclusions, and indeed crossed the specified non‐inferiority bound of 1.13. We downgraded the evidence for this outcome from low to very low quality due to imprecision, as further research is very likely to have an important impact on our confidence in the estimate of effect and may change the estimate.

Figure 3

Forest plot of comparison: 1 Overall survival, outcome: 1.1 NSM vs SSM ‐ overall survival.

Nipple‐sparing mastectomy versus modified radical mastectomy

Two studies involving 1202 participants evaluated overall survival for NSM versus MRM (Adam 2014; Sakurai 2013). Adam 2014 matched participants for age and tumour stage, but the larger study, Sakurai 2013 did not adjust for confounding. The HR was 0.72 (95% CI 0.46 to 1.13; 2 studies, 1202 participants; Analysis 1.2; Figure 4) with no heterogeneity (I² = 0%, P = 0.41). Although the confidence interval did not cross the specified non‐inferiority bound of 1.13, we were unable to conclude non‐inferiority because of the high risk of confounding. We downgraded the quality of evidence from low to very low due to the risk of bias and imprecision.

Figure 4

Forest plot of comparison: 1 Overall survival, outcome: 1.2 NSM vs MRM ‐ overall survival.

Local recurrence

Eleven studies with 13 treatment comparisons described local recurrence (Adam 2014; Boneti 2011; Burdge 2013; Gerber 2009; Horiguchi 2001; Poruk 2015; Kim 2010; Oura 1994; Sakurai 2013; Shi 2012; Stanec 2014).

In total, there were 144 events in 2105 women who had undergone NSM. The recurrent events occurred in 38 out of 118 (32.2%) women in the nipple or areola, or both, 80 out of 118 (67.8%) women in the mastectomy flap or regional lymph nodes while the location was not described in 26 participants.

Two studies involving 1303 participants evaluated the HR for local recurrence for NSM versus MRM (Adam 2014; Horiguchi 2001). One study matched participants for age and tumour stage (Adam 2014), and the other adjusted for age, lymph node status and oestrogen receptor status. The HR was 0.28 (95% CI 0.12 to 0.68, P = 0.005; 2 studies, 1303 participants; Analysis 2.1; Figure 5) with no heterogeneity (I² = 0%, P= 0.97). However we were unable to draw firm conclusions because of the high risk of residual confounding. We graded the quality of the evidence as very low due to the risk of bias.

Figure 5

Forest plot of comparison: 2 Local recurrence, outcome: 2.1 NSM vs MRM ‐ local recurrence.

Adverse events

We evaluated overall complications for two studies involving 1067 participants (Sakurai 2013; Shi 2012). Neither study made any adjustment for confounding. The RR was 0.10 (95% CI 0.01 to 0.82, P = 0.03; 2 studies, 1067 participants; Analysis 3.1) with significant heterogeneity (I² = 85%; P = 0.01). We downgraded the quality of evidence from low to very low due to inconsistency (i.e. the magnitude of effects across the studies).

Local surgical complications

Explantation of implant/expander

Two studies reported on explantation of the implant/expander (Boneti 2011; Burdge 2013). Burdge 2013 described the necessity of implant removal in 5% of the participants (3 out of 60 participants) and Boneti 2011 described the necessity of implant removal for 1% of participants (2 out of 281 in NSM group and 3 out of 227 in the SSM group).

In terms of capsular contracture, three studies described an incidence of 3.1% (9 out of 293 participants; Boneti 2011),10% (6 out of 60 participants; Burdge 2013) and 15% (22 out of 145 participants; Stanec 2014).

Hematoma

Two studies evaluated hematomas (Shi 2012; Stanec 2014). Shi 2012 reported that 2 out of 35 participants had a hematoma or infection in the NSM group while 5 out of 100 participants had a hematoma (two participants) or infection (three participants) in the MRM group. Stanec 2014 reported that 9 out of 252 participants had a hematoma in the NSM group while 3 out of 109 participants had a hematoma in the SSM group.

Seroma

Two studies evaluated seromas (Shi 2012; Stanec 2014). Shi 2012 described an incidence of 2.9% (one participant) in the NSM group versus 14% (14 participants) in the MRM group (P = 0.13). Stanec 2014 described seroma of the donor site and axilla as a frequent postoperative complication with 25.6% of participants (108/361) being treated with aspirative punction in the NSM and SSM groups.

Rehospitalization/re‐exploration

Boneti 2011 described that 0.7% (2 out of 281 breasts) in the NSM group versus 0.4% (1 out of 227 breasts) in the SSM group required further exploration for postoperative bleeding. Stanec 2014 described that 1.1% of participants (4 out of 361) needed further exploration for the drainage of breast abscesses.

Skin necrosis (nipple, areola or flap necrosis)

Four studies described skin necrosis (Kim 2010; Sakurai 2013; Shi 2012; Stanec 2014). The RR was 4.22 (95% CI 0.59 to 30.03, 4 studies, P = 0.15; 1948 participants; Analysis 3.2; Figure 6) with significant heterogeneity (I² = 79%; P = 0.008). We downgraded the quality of evidence from low to very low due to a wide confidence interval.

Figure 6

Forest plot of comparison: 3 Adverse events, outcome: 3.2 Skin necrosis.

Skin necrosis with revision surgery

Only two studies described this outcome (Boneti 2011; Stanec 2014). Boneti 2011 showed 0.9% (4 out of 404) of participants had nipple necrosis with revision surgery and Stanec 2014 described 3.6% (13 out of 361) of participants needing further exploration.

Infection

Two studies involving 496 participants evaluated infections following NSM versus other types of mastectomy (Shi 2012; Stanec 2014). Neither study made any adjustment for confounding. The RR was 0.95 (95% CI 0.44 to 2.09, P = 0.91, 2 studies; 496 participants; Analysis 3.3) with no significant heterogeneity (I² = 0%; P = 0.39).

Due to a lack of data, it was not possible to perform subgroup analysis or analyse thromboembolic events as proposed in the protocol.

Cosmetic results

Five studies performed an evaluation of the aesthetic outcome after NSM (Boneti 2011; Burdge 2013; Gerber 2009; Shi 2012; Stanec 2014). Boneti 2011 and Burdge 2013 asked the participants to rate their cosmetic result on a scale from 0 to 10 (with 10 being the most favourable outcome). In Boneti 2011, the NSM group gave a mean score of 9.2 versus 8.3 in the SSM group (P = 0.04). In Burdge 2013, the average rating was 8 by the participants and 9 by the doctors for both groups in total.

In three studies, cosmesis was assessed through personal opinion and the evaluation was either excellent, good, average or poor (Gerber 2009; Shi 2012; Stanec 2014).

Stanec 2014 described the aesthetic outcome by the type of primary reconstruction (that is, latissimus dorsal, implant and deep inferior epigastric perforators (DIEP)). The best results were with autologous reconstruction with almost 90% of participants having excellent or good evaluations compared to almost 70% of participants with implants. Unsatisfactory reconstruction was higher for participants who had implants (18.6%) compared to participants who had autologous reconstruction (7%).

Gerber 2009 described excellent and/or good results for almost 98% of the SSM group and 100% for the NSM group (P = 0.004). After 59 months of follow‐up, satisfaction (excellent and/or good) decreased to 88% in the SSM group and 96% in the NSM group (P = 0.025).

Shi 2012 asked the participants if they were satisfied with the local aesthetic (yes or no). The participants in the NSM group who received breast reconstruction felt more satisfied with the resulting aesthetic appearance (34 out of 35 participants) than those in the MRM group (20 out of 100 participants, P < 0.001).

Quality of life

Only one study reported this outcome. Shi 2012 used a psychological questionnaire and the NSM group had higher levels of self‐confidence (33 out of 35 versus 30 out of 100 participants, P < 0.001), social activity (35 out of 35 versus 60 out of 100 participants, P < 0.001) and sexual activity (31 out of 35 versus 82 out of 100 participants, P = 0.52) compared to the MRM group.

Discussion

available in

Summary of main results

In general the results were inconclusive because none of the included studies had used appropriate methods to deal with confounding. In particular, the modified radical mastectomy (MRM) and skin‐sparing mastectomy (SSM) groups tended to have participants with more advanced disease than the nipple‐sparing mastectomy (NSM) group, and the use of radiotherapy tended to be different in the different treatment groups. Currently, the indications for postmastectomy radiation have been extended and this will probably have an impact on the results in future studies.

Overall survival

The results of this review were inconclusive as to whether there was a difference in overall survival between NSM and other types of mastectomy (SSM and MRM) due to a lack of studies reporting survival outcomes in sufficient detail to extract hazard ratios, and the high risk of confounding.

Evidence from two studies involving 781 people contributed data to the overall survival outcome for NSM compared to SSM. The results were inconclusive because the confidence interval was wide. Only one of these studies had more than five years of follow‐up (Kim 2010), the other had a mean follow‐up of about 25 months (Poruk 2015).

Evidence from two studies involving 1202 people contributed data to the overall survival outcome for NSM compared to MRM. The confidence interval did not cross the non‐inferiority bound, but we were unable to draw any conclusions because of the high risk of residual confounding.

Local recurrence

For local recurrence, we could extract hazard ratios for only two studies comparing NSM to MRM, involving 1303 participants. The confidence interval indicated a lower risk of local recurrence in the NSM group, but we were unable to draw any conclusions due to the high likelihood of residual confounding due to lack of adjustment for tumour stage and radiotherapy.

Adam 2014 matched the participants assigned to each group taking into account the tumour stage. However, Horiguchi 2001 seemed to have more initial tumours (Stages 0 and 1) in the NSM group (83 out of 123 participants; 67.5%) than in the MRM group (277 out of 910 participants; 30.4%) and this imbalance may have influenced the results.

Only 7.9% (103 out of 1303 participants) received radiotherapy: 7.2% (81 out of 1113 participants) in the MRM group and 11.6% (22 out of 190 participants) in the NSM group. No one received this treatment in Horiguchi 2001, probably because during that period the indication for radiotherapy treatment was more limited. Probably more patients would now receive post‐mastectomy radiotherapy, which would have an impact on the local recurrence rate.

Adverse events

Again, we could not draw any firm conclusions because of the high risk of confounding. There seemed to be evidence that the global rate of complications was lower in the NSM group. The frequency of this outcome was 0.85% versus 18.4% for NSM and other types of mastectomy respectively. This was lower than in the published meta‐analysis of Endara 2013, which evaluated 6615 women undergoing NSM for therapeutic or prophylactic purposes and observed an incidence of 22% for overall complications. There was no evidence that infection differed between treatment groups in this review. One possible reason for the lower frequency in this Cochrane review is due to the presence of retrospective cohort studies, where the reporting of medical data may be flawed.

Strengths of this systematic review

Strengths of this systematic review include the following.

A sensitive search strategy was carried out for all electronic databases (so as to avoid missing relevant studies). In addition, a manual search was made of reference lists of relevant studies and we screened clinical trials registries
We applied a rigorous method recommended by Cochrane (Higgins 2011b) when conducting the review.
The methodological quality of observational studies included was evaluated and considered in the presentation of findings. We also evaluated the quality of evidence through an appropriate tool (GRADE).

Main limitations

The main limitations of this systematic review are a result of the limited strength of the evidence due to the methodological deficiencies of the existing studies.

The evidence in this review came from observational studies (mostly retrospective and low methodological quality), subject to important biases which increased the uncertainty of the results and limited the quality of existing evidence.
It was not possible to calculate the hazard ratio for the assessment of survival data for all studies because many studies did not report time‐to‐event analyses in sufficient detail.
Inability to perform the subgroup analyses initially proposed in the published protocol. The proposed analyses included: presence or absence of adjuvant treatment, participant age, surgical technique performed, and tumour staging according to the TNM system.
This was a systematic review that used aggregated data (in which the subject of analysis was the study), and not a meta‐analysis of individual data (in which the subject of analysis is the person or the participant). Therefore, only data published or provided subsequently by the authors of the studies were available.

Overall completeness and applicability of evidence

In this systematic review, the evidence was incomplete due to a lack of good‐quality studies in this area which have used appropriate methods to adjust for confounding. Additional research is therefore likely to have an important impact on the estimated effect. Decisions regarding choice of surgical method should be made jointly by the surgeon and woman after extensive information on the risks and benefits is provided.

Quality of the evidence

See the 'Summary of findings' tables: summary of findings Table for the main comparison; summary of findings Table 2.

Potential biases in the review process

This systematic review has several strengths. We asked a specific clinical question and the search strategy was comprehensive. We included publications of all relevant studies irrespective of language. Finally, we rigorously applied the GRADE criteria for each of the relevant outcomes (Guyatt 2008)

There were several potential biases in the review process. We made efforts to limit the bias in several ways: two review authors assessed the eligibility for inclusion and independently assessed the risks of bias. Although the review authors’ views varied, we decided to accept the final conclusions after extensive discussion and reaching a consensus. Carrying out reviews, however, may require a number of subjective judgements, and it is possible that a different review team may have reached different decisions regarding the assessments of eligibility and risks of bias. Feedback from readers will serve to improve the next review update.

Agreements and disagreements with other studies or reviews

We found a meta‐analysis by de La Cruz 2015 addressing the same clinical question as our Cochrane review. The de La Cruz 2015 meta‐analysis included eight studies and the authors found no difference between NSM and MRM/SSM: 3.4 % risk difference in overall survival (P = 0.073), 9.6 % risk difference in disease‐free survival (P = 0.056), and a 0.4 % risk difference in local recurrence (P = 0.567). These results are quite similar to our results and it was not possible to conclude that NSM was not inferior to other surgeries, because of the limited methodological quality and risk of bias of the included studies.

The most important methodological differences between the meta‐analysis by de La Cruz 2015 and our review are:

de La Cruz 2015 limited their search to Pubmed, Scopus and Google Scholar, included only studies in English, and used outcome (survival) in the search strategy. We included additional searches in Embase and CENTRAL databases, we did not use a language limit and did not specify any outcomes in the search strategy. Limiting the search string by outcome enhances specificity but reduces sensitivity which is not recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b).

de La Cruz 2015 used the risk difference to compare time‐to‐event outcomes between intervention groups. We performed these analyses using hazard ratios (which was more appropriate). Moreover, we split the analysis into two subgroups NSM versus SSM and NSM versus MRM because we considered SSM and MRM to be different control groups, which could influence the results.

de La Cruz 2015 did not assess the methodological quality and risk of bias of the included studies nor the overall quality of the evidence for each outcome. We did both assessments in accordance with the Cochrane Handbook for Systematic Reviews of Interventions using the 'Risk of bias table' and GRADE approach, respectively (Higgins 2011b).

Figure 1

Study flow diagram

Figure 2

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

Figure 3

Forest plot of comparison: 1 Overall survival, outcome: 1.1 NSM vs SSM ‐ overall survival.

Figure 4

Forest plot of comparison: 1 Overall survival, outcome: 1.2 NSM vs MRM ‐ overall survival.

Figure 5

Forest plot of comparison: 2 Local recurrence, outcome: 2.1 NSM vs MRM ‐ local recurrence.

Figure 6

Forest plot of comparison: 3 Adverse events, outcome: 3.2 Skin necrosis.

Analysis 1.1

Comparison 1 Overall survival, Outcome 1 NSM vs SSM ‐ overall survival.

Analysis 1.2

Comparison 1 Overall survival, Outcome 2 NSM vs MRM ‐ overall survival.

Analysis 2.1

Comparison 2 Local recurrence, Outcome 1 NSM vs MRM ‐ local recurrence.

Analysis 3.1

Comparison 3 Adverse events, Outcome 1 Overall complication.

Analysis 3.2

Comparison 3 Adverse events, Outcome 2 Skin necrosis.

Analysis 3.3

Comparison 3 Adverse events, Outcome 3 Infection.

Summary of findings for the main comparison. Nipple‐sparing mastectomy for breast cancer treatment

Nipple‐sparing mastectomy for breast cancer treatment
Patient or population: women with breast cancer Setting: breast cancer therapy centres Intervention: nipple‐sparing mastectomy Comparison: modified radical mastectomy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
	Risk with other types of mastectomies	Risk with nipple‐sparing mastectomy
Overall survival	Study population		HR 0.72 (0.46 to 1.13)	1202 (2 observational studies)	⊕⊝⊝⊝ VERY LOW^1,2	We were unable to conclude non‐inferiority because of the high risk of confounding. The quality of evidence was downgraded to very low due to the risk of bias and imprecision.
	882 per 1000	785 per 1000 (626 to 910)
Local recurrence	Study population		HR 0.28 (0.12 to 0.68)	1311 (2 observational studies)	⊕⊝⊝⊝ VERY LOW ³	Conclusions could not be drawn because of the high risk of residual confounding. The quality of the evidence was graded as very low due to the risk of bias.
	17 per 1000	5 per 1000 (2 to 12)
Overall complications	Study population		RR 0.10 (0.01 to 0.82)	1067 (2 observational studies)	⊕⊕⊝⊝ LOW ⁴	We downgraded the quality of evidence from low to very low due to inconsistency (i.e. the magnitude of effects across the studies).
	184 per 1000	18 per 1000 (2 to 151)
Skin necrosis	Study population		RR 4.22 (0.59 to 30.03)	1948 (4 observational studies)	⊕⊝⊝⊝ VERY LOW ^5,6	We downgraded the quality of evidence from low to very low due to a wide confidence interval.
	19 per 1000	82 per 1000 (11 to 583)
Infection	Study population		RR 0.95 (0.44 to 2.09)	496 (2 observational studies)	⊕⊕⊝⊝ LOW
	48 per 1000	45 per 1000 (21 to 100)
*The risk in the intervention group (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; RR: Risk ratio; OR: Odds ratio; HR: Hazard ratio
GRADE Working Group grades of evidence High quality: we are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹There was no adjustment for confounding in Sakurai 2013 and we classified this as a serious risk of bias. We we were unable to conclude non‐inferiority because of the high risk of confounding ²The confidence interval of included studies both increased and decreased risk for nipple‐sparing mastectomy versus modified radical mastectomy ³Imprecision was downgraded because of the high risk of residual confounding due to lack of adjustment for tumour stage and radiotherapy. Adam 2014 matched the participants by tumour stage, Horiguchi 2001 seemed to have more initial tumours (Stages 0 and 1) in the NSM group (83 out of 123; 67.5%) than the MRM group (277 out of 910; 30.4%) and this fact may have influenced the results. Only 7.9% (103 out of 1303 participants) received radiotherapy, 7.2% (81 out of 1113 participants) in the modified radical mastectomy group and in 11.6% (22 out of 190 participants) in the nipple‐sparing mastectomy. No one received the treatment in Horiguchi 2001, probably because in that period the indication for radiotherapy treatment was more limited. Probably more patients would now receive post‐mastectomy radiotherapy, which would have an impact on the local recurrence rate ⁴There was evidence of considerable heterogeneity on statistical testing, I² = 85%, P = 0.01 ⁵There was evidence of considerable heterogeneity on statistical testing, I² = 79% , P = 0.008 ⁶Confidence intervals fail to exclude both clinically important and clinically unimportant harms.

Summary of findings for the main comparison. Nipple‐sparing mastectomy for breast cancer treatment

Summary of findings 2. Nipple‐sparing mastectomy for breast cancer treatment

Nipple‐sparing mastectomy for breast cancer treatment
Patient or population: women with breast cancer Setting: breast cancer therapy centres Intervention: nipple‐sparing mastectomy Comparison: skin‐sparing mastectomy
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
	Risk with other types of mastectomies	Risk with nipple‐sparing mastectomy
Overall survival	Study population		HR 0.70 (0.28 to 1.72)	781 (2 observational studies)	⊕⊝⊝⊝ VERY LOW ¹
	928 per 1000	841 per 1000 (521 to 989)
*The risk in the intervention group (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; RR: Risk ratio; OR: Odds ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹We downgraded the quality of evidence due to: (a) risk of bias of included studies: non‐randomised studies with low quality, and imprecision: wide confidence intervals that crossed null effect and comprised both directions of effect

Summary of findings 2. Nipple‐sparing mastectomy for breast cancer treatment

Table 1. Tumour stage by study

Study	Stage 0	Stage 1	Stage 2	Stage 3	Stage 4
Adam 2014	Study did not use the AJCC classification
Boneti 2011	Study did not use the AJCC classification
Burdge 2013	Study did not use the AJCC classification
Gerber 2009	0	22	168	13	0
Horiguchi 2001	71	289	581	102	0
Kim 2010	173	752	1274	311	0
Oura 1994	0	368	125	0	0
Poruk 2015	34	70	77	34	6
Sakurai 2013	26	361	399	140	5
Shi 2012	10	0	81	14	0
Stanec 2014	73	130	71	41	0
Total	387	1992	2776	655	11
AJCC: American Joint Committee on Cancer

Table 1. Tumour stage by study

Table 2. Number of participants by tumour stage and intervention

Study	Stage 0	Stage 1	Stage 2	Stage 3	Stage 4
	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM	NSM; SSM; MRM
Adam 2014	This study did not use the AJCC classification
Boneti 2011	This study did not use the AJCC classification
Burdge 2013	This study did not use the AJCC classification
Gerber 2009¹		11(18.3);11(22.9);35 (26.9)	44(73.3);36(75.0);88(67.7)	5(8.3);1(2.1);7(5.4)	0;0;0
Horiguchi 2001²	19(14.3); 0;52(5.7)	64(48.1); 0;225(24.7)	48(36.1); 0; 533(58.6)	2(1.5); 0; 100(11.0)	0;0;0
Kim 2010³	19(12.5);65(17.7); 89(4.5)	70(46.1); 150 (40.8);532(26.7)	55(36.2);121(32.9);1098(55.2)	8(5.3);32(8.7)271(13.6)	0;0;0
Oura 1994	0;0;0	244(81.6); 0; 124(63.9)	55(18.4); 0; 70(36.1)	0;0;0	0;0;0
Poruk 2015⁴	24(23.3); 10 (8.5); 0	36(35.0); 34(28.8); 0	34(33.0); 43(36.4); 0	7(6.8); 27(22.9);0	2(1.9);4(3.4); 0
Sakurai 2013	21(2.7); 0; 5(3.5)	304(38.6); 0; 57(39.6)	364(46.3); 0; 35(24.3)	95(12.1); 0; 45(31.3)	3(0.4); 0; 2(1.4)
Shi 2012	2 (5.7); 0; 8 (8.0)	10 (28.6); 0; 20(20)	18 (51.4); 0; 63 (63.0)	5(14.3); 0; 9(9.0)	0;0;0
Stanec 2014	51(19.8); 22(22.9); 0	97(37.7);33(34.4);0	88(34.2); 21(21.9); 0	21(8.2); 20 (20.8); 0	0;0;0
Total	136(7.4); 97(15.4);154(4.4)	836(45.8);228(36.2);993(28.6)	706(38.7);221(35.1);1887(54.4)	143(7.8);80(12.7); 432(12.5)	5(0.3); 4(0.6); 2(0.1)
Numbers are given by absolute values with percentages in brackets. AJCC: American Joint Committee on Cancer MRM: modified radical mastectomy NSM: nipple‐sparing mastectomy SSM: skin‐sparing mastectomy ¹Stage 0 and 1 have been combined; P = 0.47 ²P < 0.01 ³P = 0.21 ⁴P = 0.001

Table 2. Number of participants by tumour stage and intervention

Table 3. Mean age of participants by intervention

Age	NSM	SSM	MRM	Comments
Adam 2014	49 (24‐74)	‐	48.5 (21‐87)	P = 0.38
Boneti 2011	51.2 (10.9)	53,1 (11.5)	‐	P = 0.24
Burdge 2013	48.1 (10.4); (29‐75)	53.9 (10.4); (29‐73)	‐
Gerber 2009	46 (10)	48 (10)	58 (6)	P = 0.001
Horiguchi 2001	<= 50 y: 57 >50 y: 76	‐	<= 50 y: 504 50 y >: 406	P < 0.01
Kim 2010	41.5 (7.4)	42.8 (6.6)	data not available	P = 0.06
Oura 1994	Stage 1: 49 Stage 2: 48	‐	Stage 1: 56 Stage 2: 54
Poruk 2015	45 (12)	55 (14.5)	‐	P < 0.001
Sakurai 2013	51 (25‐89)	‐	58 (31‐88)
Shi 2012	35.6	‐	50.8
Stanec 2014	50.9	53.2	‐
MRM: modified radical mastectomy NSM: nipple‐sparing mastectomy SSM: skin‐sparing mastectomy

Table 3. Mean age of participants by intervention

Table 4. Mean follow‐up (in months) by intervention

Follow‐up (months)	NSM	SSM	MRM	Comments
Adam 2014	35	‐	36
Boneti 2011	25.3 (18.8) 3‐102	38.2 (26.3) 4‐144	‐	P < 0.001
Burdge 2013	25 (18.8)	38 (26.3)	‐
Gerber 2009	101 (32‐126)	101 (32‐126)	101 (32‐126)
Horiguchi 2001	66	‐	81
Kim 2010	60	67	not mentioned
Oura 1994	Stage 1: 49 Stage 2: 38	‐	Stage 1: 85 Stage 2: 73
Poruk 2015	25.8 (18)	29.9 (15.7)	‐	P = 0.86
Sakurai 2013	87 (10‐252)	‐	87 (0‐231)
Shi 2012	68 (10‐104)	‐	68 (10‐104)
Stanec 2014	43	58	‐
MRM: modified radical mastectomy NSM: nipple‐sparing mastectomy SSM: skin‐sparing mastectomy

Table 4. Mean follow‐up (in months) by intervention

Comparison 1. Overall survival

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 NSM vs SSM ‐ overall survival Show forest plot	2	781	Hazard Ratio (Fixed, 95% CI)	0.70 [0.28, 1.72]

2 NSM vs MRM ‐ overall survival Show forest plot	2	1202	Hazard Ratio (Fixed, 95% CI)	0.72 [0.46, 1.13]

Comparison 1. Overall survival

Comparison 2. Local recurrence

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 NSM vs MRM ‐ local recurrence Show forest plot	2	1303	Hazard Ratio (Fixed, 95% CI)	0.28 [0.12, 0.68]

Comparison 2. Local recurrence

Comparison 3. Adverse events

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Overall complication Show forest plot	2	1067	Risk Ratio (M‐H, Random, 95% CI)	0.10 [0.01, 0.82]

2 Skin necrosis Show forest plot	4	1948	Risk Ratio (IV, Random, 95% CI)	4.22 [0.59, 30.03]

3 Infection Show forest plot	2	496	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.44, 2.09]

Comparison 3. Adverse events