Immunonutrition for patients undergoing surgery for head and neck cancer

Summary of findings for the main comparison. Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer

Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer
Patient or population: patients undergoing surgery for head and neck cancer Setting: hospitals (international) Intervention: immunonutrition Comparison: standard care
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with standard care	Risk with immunonutrition
Postoperative length of hospital stay (days) Follow‐up: 8 to 90 days post surgery or hospital discharge	The mean of reported length of hospital stay (mean values) across the standard care groups was 27.0 (17.4 to 36.1) days	The mean of reported length of hospital stay (mean values) across the immunonutrition groups was 23.2 (15.3 to 31.1) days	MD 2.5 lower (5.11 lower to 0.12 higher)	757 (10 RCTs)	⊕⊕⊝⊝ LOW¹	There may be a reduction in the length of hospital stay of 2.5 days with immunonutrition, but the estimate is imprecise (wide CI) and includes the null value.
Wound infection Follow‐up: 8 to 90 days post surgery or hospital discharge	Study population		RR 0.94 (0.70 to 1.26)	812 (12 RCTs)	⊕⊝⊝⊝ VERY LOW^1,2	Immunonutrition may have little or no effect on wound infection, but the evidence is very uncertain.
	145 per 1000	136 per 1000 (101 to 182)
Fistula formation Follow‐up: 8 to 90 days post surgery or hospital discharge	Study population		RR 0.48 (0.27 to 0.85)	747 (10 RCTs)	⊕⊕⊝⊝ LOW ¹	There may be an approximate halving of the risk of fistulae formation but the evidence is of low quality.
	113 per 1000	54 per 1000 (31 to 96)
Adverse events/tolerance of feeds Follow‐up: 10 to 90 days post surgery or hospital discharge	Study population		RR 1.33 (0.86 to 2.06)	719 (9 RCTs)	⊕⊝⊝⊝ VERY LOW^1,3	There may be little or no difference in adverse events such as diarrhoea between the treatment groups, but the evidence is very uncertain.
	91 per 1000	121 per 1000 (78 to 188)
All‐cause mortality Follow‐up: 30 days to greater than or equal to 16 months post surgery	Study population		RR 1.33 (0.48 to 3.66)	776 (14 RCTs)	⊕⊕⊝⊝ LOW¹	Immunonutrition may have little or no effect on mortality.
	18 per 1000	24 per 1000 (9 to 67)
*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; MD: mean difference; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: 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 certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹Downgraded by two levels for imprecision: most studies had small sample sizes and confidence intervals around the summary estimates were wide. ²Downgraded by one level for risk of bias: assessment of wound infection was poorly reported across studies. ³Downgraded by one level for risk of bias: assessment of adverse events was poorly reported across studies and not all studies measured the same adverse events.

Background

Description of the condition

The term 'head and neck cancer' encompasses several sites including oral and laryngeal cancers. In 2014, over 11,000 people in the UK were diagnosed with cancers at these sites (Cancer Research UK). Surgical treatment of head and neck cancer can be aggressive and highly complex, and people undergoing these surgeries may have a 30% to 60% incidence of postoperative complications including wound infections and other infections such as pneumonia (Kucur 2015; McMahon 2013; Perisanidis 2012; Yang 2014). This substantial morbidity has inevitable implications for both patients and healthcare systems. Furthermore, a recent systematic review and meta‐analysis showed that postoperative complications, especially infections, adversely affect long‐term survival (Pucher 2014).

Many patients with head and neck cancer are malnourished for a number of reasons including mechanical obstruction, tumour‐induced cachexia, poor dietary habits and excessive alcohol consumption. Poor nutrition is known to have an adverse impact on outcome in this patient group (van Bokhorst 2000). These patients have well‐documented immune defects including T‐lymphocytopenia and dysfunction, and reduced monocyte HLA‐DR expression (Hadden 1997). These defects, combined with the immune suppressive effects of surgery, may contribute to increased postoperative complications such as poor wound healing and sepsis.

Description of the intervention

Nutrition supports immune function by preventing or reversing immunosuppression related to malnutrition. Standard commercial nutritional supplements are described as polymeric, which means they contain whole protein, partially digested starch and triglycerides, along with electrolytes, trace elements and vitamins. More recently, specific nutritional components have been combined with standard polymeric enteral feeds with the aim of specifically improving immune function. Immunonutrition describes enteral feeding formulas usually supplemented with combinations of the amino acids arginine or glutamine, omega‐3 fatty acids and nucleic acids. Animal models and human studies have suggested that the individual components have beneficial (or potentially beneficial) effects on immune function. There is evidence that nutritional supplements with immunonutritional additives can favourably modulate the immune and inflammatory response both in vitro and in patients with trauma, burns or those undergoing gastrointestinal surgery (Di Carlo 1999; Wu 2001; Zhang 2012). Meta‐analysis suggests that immunonutrition reduces infectious complications in critically ill patients (Heyland 2001). They are usually given in liquid form and are designed to provide a patient's 'complete' nutritional requirements, provided they are given in an appropriate volume. Immunonutrition and standardised commercial nutrition supplements may be given either orally or via an enteral feeding tube.

How the intervention might work

The most studied nutrients in immunonutrition formulas are arginine, glutamine, omega‐3 fatty acids and nucleotides. Arginine is the most common immunonutrient given to patients with head and neck cancer. It is a non‐essential amino acid with a role in the synthesis of nucleotides, polyamines, nitric oxide and proline. Arginine may stimulate lymphocyte function and improve wound healing. Glutamine, also an amino acid, is a fuel for rapidly dividing cells in the body, in particular for enterocytes and colonocytes. The addition of omega‐3 fatty acids to enteral nutrition feeds reduces proinflammatory mediators in stressed patients and may reduce infections. The content of each immunonutrition formula varies between products. The biochemical and physiological properties of nutrients included in immunonutrition formulas have been discussed in detail (Worthington 2011).

Why it is important to do this review

Commercial enteral feed products containing specific nutritional components that may favourably affect immune function have been designed to improve the outcomes in surgical patients. Studies of head and neck cancer patients receiving immunonutrition in the perioperative period have not conclusively demonstrated benefit. We carried out a systematic review of randomised controlled trials, which was published in 2009, to determine whether perioperative immunonutrition has a role in the treatment of head and neck cancer (Stableforth 2009). In that review we examined 10 trials investigating the effects of immunonutrition in patients treated surgically for head and neck cancer. A reduction in the length of postoperative hospital stay was seen, but the reason for this reduction was not clear. Most trials were too small to provide precise estimates of intervention effects. There were insufficient data to exclude substantial effects of immunonutrition on clinical outcomes or biochemical and immunological parameters. Since the publication of that review in 2009 there have been further studies that merit evaluation and inclusion in an updated review (Azman 2015; Casas‐Rodera 2008; De Luis 2009; De Luis 2014; Falewee 2014; Felekis 2010; Ghosh 2012; Hanai 2018; Sorensen 2009; Turnock 2013).

Objectives

To assess the effects of immunonutrition treatment, compared to standard feeding, on postoperative recovery in adult patients undergoing elective (non‐emergency) surgery for head and neck cancer.

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs), including quasi‐randomised trials. We had planned to subject quasi‐randomised trials to a sensitivity analysis (see Sensitivity analysis). We included studies irrespective of language or publication status.

We excluded non‐randomised studies, such as cohort studies, because of the increased potential for bias. We also excluded cross‐over trials as this methodology is not suitable for evaluating an intervention that must be given at a specific time point.

Types of participants

We included all adult patients (18 years of age or older) undergoing an elective surgical procedure for head and neck cancer under a general anaesthetic.

Types of interventions

Intervention

The intervention was polymeric nutritional supplements with immunonutritional additives given by an oral or enteral route. In order to be included, studies needed to administer the immunonutrition either preoperatively or postoperatively or both pre‐ and postoperatively. Co‐intervention with other oral or parenteral substances was permitted as long as the dose of immunonutritional additives was quantified. The content of each immunonutrition formula can vary between products and we recorded the product used and its contents for each study.

Control

The control group received either standard care (intravenous fluids) and/or polymeric nutritional supplements.

The comparison was:

immunonutrition versus standard care (intravenous fluids) and/or polymeric nutritional supplements.

Types of outcome measures

We assessed the following outcomes in this review, but we did not use them as a sole basis for excluding studies.

Primary outcomes

Length of hospital stay: measured in days from the day of surgery to discharge from hospital.
Wound infections: as measured by the proportion of patients in whom any type or degree of wound infection was recorded, at any point postoperatively.
Fistula formation: as measured by the proportion of patients in whom a fistula was recorded at any point postoperatively.
Adverse events/tolerance of feeds, as defined by trial authors: as measured by the proportion of patients in whom adverse events relating to tolerance of feed was recorded, at any point postoperatively.

Secondary outcomes

We assessed the following secondary outcomes, measured postoperatively:

All‐cause mortality: as measured by the proportion of patients recorded as having died at any point postoperatively.
Postoperative complications, as defined by trial authors: as measured by the proportion of patients in whom any type or degree of complication (other than wound infection, fistula formation or relating to tolerance of feed) was recorded, at any point postoperatively.

Search methods for identification of studies

The Cochrane ENT Information Specialist conducted systematic searches for randomised controlled trials and controlled clinical trials. There were no language, publication year or publication status restrictions. The date of the search was 14 February 2018.

Electronic searches

The Information Specialist searched:

the Cochrane ENT Trials Register (searched 14 February 2018);
the Cochrane Central Register of Controlled Trials (CENTRAL) (searched via CRS Web 14 February 2018);
PubMed (1946 to 14 February 2018);
Ovid EMBASE (1974 to 14 February 2018);
Ovid CAB Abstracts (1910 to 14 February 2018);
EBSCO CINAHL (1982 to 14 February 2018);
LILACS, lilacs.bvsalud.org (searched 14 February 2018);
KoreaMed (searched via Google Scholar 14 February 2018);
IndMed, www.indmed.nic.in (searched 14 February 2018);
PakMediNet, www.pakmedinet.com (searched 14 February 2018);
Web of Knowledge, Web of Science (1945 to 14 February 2018);
ClinicalTrials.gov (searched via the Cochrane Register of Studies 14 February 2018);
World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP), www.who.int/ictrp (searched 14 February 2018);
ISRCTN, www.isrctn.com (searched 14 February 2018).

The Information Specialist modelled subject strategies for databases on the search strategy designed for CENTRAL. Where appropriate, they were combined with subject strategy adaptations of the highly sensitive search strategy designed by Cochrane for identifying randomised controlled trials and controlled clinical trials (as described in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0, Box 6.4.b. (Handbook 2011). Search strategies for major databases including CENTRAL are provided in Appendix 1.

Searching other resources

We scanned the reference lists of identified publications for additional trials and contacted trial authors where necessary. In addition, the Information Specialist searched PubMed to retrieve existing systematic reviews relevant to this systematic review, so that we could scan their reference lists for additional trials. The Information Specialist also ran non‐systematic searches of Google Scholar to retrieve grey literature and other sources of potential trials.

Data collection and analysis

Selection of studies

Two review authors independently examined the titles and abstracts of studies identified through the search strategy (either NH and ST or CA and SJL). Inconsistency between review authors regarding articles for full‐text reading was resolved by consultation with another review author. We obtained full‐text papers for all studies that could not be excluded on the basis of title and abstract. The same review authors then independently refined their selection by examining the selected articles and excluding those not relevant to this review. The review authors recorded agreement on study inclusion and resolved disagreement by consensus. We contacted original study authors where further clarity was needed in order to select a study for inclusion. We documented decisions on all studies and these are presented in the PRISMA flow chart (Figure 1).

Figure 1

Process for sifting search results and selecting studies for inclusion.

Data extraction and management

We minimally modified a data extraction form from the original provided by Cochrane. Three authors (NH, ST and SJL) tested this on several studies selected for inclusion, and revised it for ease of extraction and to include further useful data items. Two authors (NH and ST) independently extracted data from each study. The review authors were blinded to each other's data.

We extracted data regarding participant demographics, participant disease status, surgical procedures, control group postoperative care and the intervention (frequency and duration of supplementary feeding). SJL and CA combined the tabulated data and checked it for inconsistency.

Assessment of risk of bias in included studies

Two review authors (ST and NH) independently assessed risk of bias. Disagreements were resolved by discussion and consensus, in which the other authors (CA, SJL) arbitrated. We developed our own risk of bias tool based on the criteria described in the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011), tailored to this review. We developed this tool as data extraction continued. We then discussed risk of bias for all studies to ensure uniformity and agreement. Where possible, we sought study protocols to aid assessment of selective outcome reporting bias.

We assessed each study according to the following domains:

random sequence generation;
allocation concealment;
blinding of participants and personnel;
blinding of outcome assessment;
incomplete outcome data;
selective reporting; and
any other potential threats to validity.

To assess risk of bias for these domains we looked for: evidence of, for example, use of randomisation tables or lists or randomisation by computer; allocation concealment via, for example, opaque, sealed envelopes or pharmacy assignment; explicit statements on blinding (or otherwise) and clear descriptions of who was blinded (we did not judge blinding of outcome assessment in relation to mortality, as it would not have been affected by the outcome assessor); specific statements regarding an intention‐to‐treat (ITT) analysis being conducted, statements about dropouts, or data presented in a way that allowed the number of participants included in analyses to be ascertained; all outcomes in protocols being reported in the manuscript; and factors such as poor recruitment rates, differences in baseline demographics, inadequate or poorly defined methods for assessing outcomes such as wound infections and length of hospital stay.

We classified risk of bias as 'high', 'low' or 'unclear' for each of these domains.

Measures of treatment effect

Categorical data are presented as a risk ratio (RR) with 95% confidence interval (CI). We present continuous data as a mean difference (MD) or standardised mean difference (SMD) with 95% CI, as appropriate.

Unit of analysis issues

The unit of analysis in all included studies was the individual participant. No studies used cluster‐randomisation.

Dealing with missing data

We contacted the nominated trial investigator for the included studies to obtain any missing data necessary for meta‐analysis (NH and SJL). We had planned to calculate missing standard deviations from the standard errors or confidence intervals, as described in the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011), but this was not required. Where standard deviations could not be calculated, we planned to impute these using the mean of the reported standard deviations from the other studies, but this was not needed.

Assessment of heterogeneity

We assessed statistical heterogeneity using visual inspection of the forest plot, the I² statistic (Handbook 2011), and the Chi² test. We considered an I² value of greater than 50% along with a P value of less than 0.10 in the Chi² test to be indicative of the need to further examine heterogeneity (Handbook 2011).

Assessment of reporting biases

We assessed publication bias and other small study effects in a qualitative manner using a funnel plot.

Data synthesis

We performed analyses in RevMan 5.3 (RevMan 2014). Analyses comprised only within‐study comparisons rather than individual‐level data. Comparisons were based on an intention‐to‐treat analysis. We used a random‐effects model for the meta‐analysis of results, as there was a high level of clinical heterogeneity among the included studies. Three authors (NH, ST and SJL) discussed the results for each outcome measure within each study, to determine the inclusion of data in the meta‐analyses.

Where complications were reported as percentage incidence, we converted this into the number of participants who experienced complications. In the case of Snyderman 1999 we estimated the number of wound infections, pneumonias and urinary tract infections per treatment group by visual inspection of 'Figure 2' within their manuscript.

All authors participated in double‐checking all of the continuous outcome data entered into RevMan for the included studies.

Subgroup analysis and investigation of heterogeneity

We planned to perform subgroup analyses as follows (and comparison of subgroups using an interaction term if appropriate).

Subgroup analysis of the participants, according to type of surgery:

anatomical site of surgery;
type of reconstruction ('primary closure' versus 'free flap').

Subgroup analysis of the intervention to assess clinical heterogeneity:

preoperative immunonutrition versus placebo drink;
postoperative immunonutrition versus postoperative polymeric feed.

Sensitivity analysis

Sensitivity analyses were based on the risk of bias of the studies (i.e. the removal of studies judged at high risk of bias for at least two of the factors assessed), or if they were quasi‐randomised trials. We also considered the appropriateness of comparing random‐effects and fixed‐effect estimates of each outcome variable. If publication bias was suspected we planned to perform a 'trim and fill' sensitivity analysis of the primary outcomes. To assess trial influence we planned to perform sensitivity analyses by sequentially excluding each study. If it was not possible to conduct an analysis in RevMan 5.3, we would have used Stata (Stata 11, StataCorp).

GRADE and 'Summary of findings' table

We used the GRADE approach to rate the overall quality of evidence (Ryan 2016). Two authors (CA and SJL) made the GRADE ratings and any differences were resolved by consensus of all authors. The quality of evidence reflects the extent to which we are confident that an estimate of effect is correct and we applied this in the interpretation of results. There are four possible ratings: high, moderate, low and very low. A rating of high quality of evidence implies that we are confident in our estimate of effect and that further research is very unlikely to change our confidence in the estimate of effect. A rating of very low quality implies that any estimate of effect obtained is very uncertain.

The GRADE approach rates evidence from RCTs that do not have serious limitations as high quality. However, several factors can lead to the downgrading of the evidence to moderate, low or very low. The degree of downgrading is determined by the seriousness of these factors:

study limitations (risk of bias);
inconsistency;
indirectness of evidence;
imprecision; and
publication bias.

We included a 'Summary of findings' table (summary of findings Table for the main comparison), constructed according to the recommendations described in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011).

We included the following outcomes in the 'Summary of findings' table: length of hospital stay, wound infection, fistula formation, adverse events/tolerance of feeds and postoperative mortality.

Results

Description of studies

See tables of Characteristics of included studies; Characteristics of excluded studies.

Results of the search

The electronic searches retrieved 1434 results. We identified three further records through scanning the reference lists of included studies. After screening titles and abstracts, we discarded 708 duplicates and 699 irrelevant records. We sought full texts for the remaining 29 records and retrieved trial information from ClinicalTrials.gov for one record (NCT03261180).

Upon screening we excluded a further nine records (see Characteristics of excluded studies). One of the records for which we sought a full text was the protocol associated with Palma‐Milla 2016; as such, only eight studies are shown in the Characteristics of excluded studies table). One relevant trial is not yet recruiting (NCT03261180).

Nineteen studies (with 20 publications) met the full inclusion criteria. We therefore included 19 unique studies comprising 1099 participants, as shown in Figure 1. The searches were completed in February 2018.

Included studies

We included 19 studies see Characteristics of included studies.

One study was only published as an abstract (Felekis 2005). We obtained additional unpublished data from 12 studies (Casas‐Rodera 2008; De Luis 2002; De Luis 2003; De Luis 2004; De Luis 2005; De Luis 2007; De Luis 2009; Falewee 2014; Ghosh 2012; Riso 2000; Sorensen 2009; Turnock 2013).

Design

All of the included studies were randomised trials of an active (immunonutrition) intervention versus control (see Table 1 for a description of the interventions used).

Table 1. Interventions

						Duration of supplements
Study	Groups	Control	Active	Isocaloric/isonitrogenous	Target energy intake	Pre‐operation	Post‐operation	Length of follow‐up
Snyderman 1999	1. Active pre‐ and postoperative 2. Active postoperative only 3. Control pre‐ and postoperative 4. Control postoperative only	Polymeric (Replete, Resource, Isosource, Jevity, Vivonex, Osmolite)	Polymeric + arginine (Impact)	Not stated	500 mL per day pre‐operation (500 kcal) 1000 mL per day post‐operation (1000 kcal)	≥ 5 days	≥ 7 days	1 month post‐operation
Riso 2000	1. Active postoperative enteral 2. Control postoperative enteral Both groups received parenteral nutrition for 3 days postoperatively to achieve nutritional goal	Polymeric (Nutrison protein plus)	Polymeric + arginine (Nutrison intensive)	Yes	31 kcal/kg per day by POD4	None	≥ 10 following total laryngectomy, ≥ 21 days following partial laryngectomy	To hospital discharge
Van Bokhorst 2000/2001	1. No pre‐operative nutritional support + postoperative standard formula 2. Pre‐operative + postoperative standard formula 3. Arginine supplemented pre‐ and postoperative	Polymeric	Polymeric + arginine ("41% of casein proteins were replaced by arginine")	Yes	150% of basal requirement	7 to 10 days	≥ 10 days	7 days post‐operation and greater than or equal to 16 months (survival)
De Luis 2002	1. Postoperative supplement with arginine + fibre 2. Postoperative polymeric control	Polymeric	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	Average duration 22 days (± 12 days) across groups	14 days postoperatively, and 3 months post hospital discharge (mortality)
De Luis 2003	1. Postoperative enteral supplement with arginine + fibre 2. Postoperative polymeric control	Polymeric	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	Average duration 20 days in group 1	5 days post‐operation
De Luis 2004	1. Postoperative polymeric + arginine + fibre 2. Postoperative polymeric control + fibre	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	≥ 10 days	Day 14
De Luis 2005	1. Postoperative polymeric + arginine 2. Postoperative polymeric control	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (12.5 g/day) + fibre (0.9 g/100 mL)	Yes	Not stated	None	Average duration 20 days in group 1	6 days post‐operation
Felekis 2005	1. Active pre‐ and postoperative 2. Regular oral diet and standard polymeric enteral feeding pre‐ and postoperatively	Polymeric	Undefined enteral immunonutrition	Yes	Not stated	6 days	8 days	Not stated
De Luis 2007	1. Postoperative polymeric + arginine 2. Postoperative polymeric control	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (0.85 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	≥ 10 days	12 days post‐operation
Casas‐Rodera 2008	1. Postoperative arginine enhanced formula 2. Polymeric control 3. Postoperative arginine enhanced formula RNA and omega‐3 fatty acids	Group 2 polymeric	Group 1 arginine‐enhanced formula (NB data from Group 1 not used in analyses) Group 3 arginine (1.3 g/100 mL), RNA (0.12 g/100 mL) and ratio of ω6:ω3 fats noted as "0.7 g"	No (control 122 kcal/100 mL, active 101 kcal/100 mL)	Requirements (used Harris Benedict formula with a stress factor correction of 1.4) by POD3	None	Average duration of 14.5 ± 8 days	14 days post‐surgery
De Luis 2009	1. Active postoperative (arginine) + fibre 2. Control postoperative (standard enteral nutrition)	Polymeric	Polymeric + arginine (0.85 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg; 1.7 g protein/kg on POD4	None	≥ 10 days	10 days post‐operation
Sorensen 2009	1. Active pre‐ and postoperative (arginine, glutamine, nucleotides and omega‐3 enriched) 2. Control pre‐ and postoperative (standard enteral nutrition)	Isosource	Polymeric + arginine and glutamine (Impact Recover (oral drink, containing 16.3 g arginine, 20 g glutamine, 1.6 g nucleotides, and 1.7 g ω3 fatty acids/L) or Impact glutamine (tube feed, containing 16.3 g arginine, 15 g glutamine, 1.6 g nucleotides and 2.7 g ω3 fatty acids/L))	No (Impact Recover = 960 kcal/L; Impact glutamine = 1300 kcal/L; Isosource = 1440 kcal/L)	˜1 L per day pre‐operatively, at least 1 L per day postoperatively (˜50% to 60% of basal requirements from intervention)	7 days	7 days	29 days post‐operation
Felekis 2010	1. Control pre‐ and postoperative 2. Active pre‐ and postoperative	Polymeric (Nutrison, Nutricia)	Polymeric + ω3 fatty acids, arginine, RNA (Impact, Novartis)	Yes	Requirement (based on Harris and Benedict equations) by POD4	5 days	8 days	8 days post‐operation
Ghosh 2012	1. Control pre‐ and postoperative 2. Active pre‐ and postoperative	Polymeric	Polymeric + ω3 fatty acids (1.7 g/L), arginine (1.25 g/L), RNA (1.2 g/L) (Impact)	Yes	Nutritional requirements based on Schofield calculation, 500 mL/day pre‐operatively, 1000 mL/day postoperatively (also allowed to eat and drink if able). Any calorific shortfall was made up using standard proprietary feeds.	5 days	7 days	30 days post‐operation
Turnock 2013	1. Control ‐ no supplements pre‐operation, postoperative standard supplements 2. Active ‐ pre‐ (oral) and postoperative (enteral then oral to day 5) supplements	Isosource standard (postoperative only)	Polymeric + ω3 fatty acids (3.3 and 1.7 g/L EPA and DHA for oral and enteral products, respectively), arginine (12.6 and 12.5 g/L for oral and enteral products), RNA (1.3 and 1.2 g/L for oral and enteral products) (Impact)	No (oral/enteral Impact 1000 kcal/L; Isosource standard 1200 kcal/L)	Requirements (25 to 30 kcal/kg)	5 days	≥ 5 days	Hospital discharge
De Luis 2014	1. Active postoperative 2. Control postoperative	Polymeric	Polymeric + arginine (8 g/L, 20 g/day) (NB control and active formulas contained the same ratio of ω6:ω3 fats and the same amount of fibre 13.8 g/L)	No (control 1.118 kcal/L, active 1.020 kcal/L)	Requirements (35 kcal/kg, 1.7 g protein/kg)	None	Minimal of 15 days	10 days post‐surgery
Falewee 2014	1. Pre‐ and postoperative control 2. Pre‐operative active, postoperative control 3. Pre‐ and postoperative active	Impact without immunonutrients	Polymeric + ω3 fatty acids (1.0 g per sachet (oral) and 1.65 g/500 mL (enteral)), arginine (3.8 g per sachet (oral) and 6.5 g/500 mL (enteral)), RNA (0.45 g per sachet (oral) and 0.65 g/500 mL (enteral)) (Oral Impact preoperatively and Enteral Impact postoperatively)	Yes	Requirements (calculated using SFNEP French recommendations)	7 days	7 to 15 days	90 days post‐surgery
Azman 2015	1. Postoperative active 2. Postoperative no intervention	None	Glutamine powder (30 g/day) (Glutamine Plus, Fresenius Kabi)	No control intervention	30 to 40 kcal/kg/day	None	4 weeks	4 weeks post‐operation
Hanai 2018	1. Pre‐ and postoperative active 2. Pre‐and postoperative no intervention	None	Prosure (an eicosapentaenoic acid (EPA) enriched oral nutritional supplement) at a dose of 2 packs per day (480 mL)	No control intervention	Prosure was administered in addition to a normal diet (or in lieu of part of a normal diet). Quote: "The dietary intake was not limited"	14 days	14 days	14 days post‐operation

DHA: docosahexaenoic acid
EPA: eicosapentaenoic acid
POD: postoperative day
RNA: ribonucleic acid
SFNEP: Société Francophone de Nutrition Entérale et Parentérale

Sample sizes

Total sample sizes ranged from 8 to 209 participants (Table 2). Twelve of the 19 studies had fewer than 25 patients in each treatment group.

Table 2. Baseline patient characteristics

	Number		Mean age		Sex M:F		Mean weight (kg)		BMI
Study	Control	Intervention	Control	Intervention	Control	Intervention	Control	Intervention	Control	Intervention
Snyderman 1999	47	82	61	63	32:15	63:19	67	71	Not reported	Not reported
Riso 2000	21	23	63	61	18:3	21:2	66	64	23.2	22.1
Van Bokhorst 2000/2001	15	17	60	59	7:8	12:5	55	62	Not reported	Not reported
De Luis 2002	24	23	59	63	3:21	2:21	68	68	24.1	26.2
De Luis 2003	18	18	59	63	1:27	1:17	69	69	24.1	26.2
De Luis 2004	45	45	61	60	3:42	3:42	69	70	25.1	25.2
De Luis 2005	15	14	63	61	3:12	2:12	Not reported	Not reported	24.1	24.6
Felekis 2005	17	20	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
De Luis 2007	37	35	62	62	3:34	4:31	69	68	25.1	24.0
Casas‐Rodera 2008*	15	14	54	50	15:0	14:0	66	68	Not reported	Not reported
De Luis 2009	34	38	61	63	27:7	30:8	71	73	26.4	26.5
Sorensen 2009	7	8	59	62	7:0	8:0	69	71	22.3	22.7
Felekis 2010	20	20	63	61	18:2	18:2	Not reported	Not reported	Not reported	Not reported
Ghosh 2012**	29	28	60	62	26:3	24:4	65	72	24	26
Turnock 2013	4	4	47	51	3:1	3:1	67	67	Not reported	Not reported
De Luis 2014	40	42	63.6	65.5	35:5	37:5	70	72	25.1	25.7
Falewee 2014***	104	105	59	59	86:18	87:18	69	70	23.7	23.6
Azman 2015****	22	22	Not reported	Not reported	15:7	9:13	Not reported	Not reported	Not reported	Not reported
Hanai 2018	14	13	66.1	61.5	8:6	8:5	Not reported	Not reported	Not reported	Not reported

* Two different values for weight reported in manuscript: data from Table II in manuscript are reported here.

** Data reported in manuscript (and here) as median values for age, weight and body mass index (BMI).

*** Additional data supplied by authors.

**** Median age across both treatment groups was 49 years (range 22 to 74 years). No data by treatment group available.

Setting

Studies were set in hospitals and conducted in eight countries. Seventeen studies were single‐site studies and one study was multicentre (Falewee 2014). We identified eight studies from Spain (Casas‐Rodera 2008; De Luis 2002; De Luis 2003; De Luis 2004; De Luis 2005; De Luis 2007; De Luis 2009; De Luis 2014). Two studies were from the USA (Snyderman 1999; Sorensen 2009), two studies from Greece (Felekis 2005; Felekis 2010), and one from each of the following countries: France (Falewee 2014), Italy (Riso 2000), Netherlands (Van Bokhorst 2000/2001), New Zealand (Turnock 2013), Malaysia (Azman 2015), Japan (Hanai 2018), and the UK (Ghosh 2012).

Participants

The 19 studies included in this review represent a total of 1099 participants undergoing head and neck cancer surgery of the upper aerodigestive tract (sites included mouth, pharynx and larynx) (see Characteristics of included studies).

Studies included adults only and the mean age of study participants across studies ranged from 47 to 66 years (Table 2). There were more males than females in most studies, and the mean body mass index (BMI) reported across studies ranged from 22.1 to 26.5 (Table 2). The stage of disease was reported in 18 studies with only one study not reporting this (Felekis 2005; published as an abstract).

Studies excluded people with a range of medical conditions including impaired renal or hepatic function (16 studies), ongoing infections (13 studies) and autoimmune disorders (13 studies), those on steroid treatment (10 studies) or nutritional oral supplementation in the previous six months and those who were malnourished/had severe cancer cachexia or sarcopenia (seven studies), those who were well nourished (two studies) or morbidly obese (one study), patients with contraindications to enteral nutrition/patients with inborn errors of metabolism relating to the composition of the formula (two studies), patients treated with chemotherapy and/or radiation therapy delivered to the head and neck area during the previous year (three studies) or chemoradiotherapy or other treatment protocols concurrent to the intervention (one study), patients testing positive for HIV (three studies), patients with diabetes (five studies), and pregnant or breast‐feeding women (four studies).

Interventions

Detailed descriptions of the interventions used in each of the included studies are shown in Table 1. Most studies (16/19) used immunonutrition formulas that contained arginine, one study used glutamine powder, one study used an eicosapentaenoic acid (EPA)‐enriched oral nutritional supplement and one used an unspecified product. When study feeds were given pre‐operatively, the length of intervention ranged from around 5 to 14 days (n = 9 studies). There was more variation when feeds were given postoperatively, with a range of around 5 days to an average duration of 22 days ± 12 days (n = 19 studies). Most studies (12/19) based the intake of study feeds on body weight or 'requirements', some (n = 5/19) used a set amount (e.g. 1000 mL per day, or 30 g powder per day) and others (n = 2/19) did not state the amount. Follow‐up time frames also varied considerably across studies, and ranged from five days post‐operation (De Luis 2003) to greater than or equal to 16 months (Van Bokhorst 2000/2001 for survival data).

Immunonutrition was given postoperatively in all studies, and nine studies gave immunonutrition pre‐operatively as well as postoperatively (Table 1). One study with three treatment groups gave pre‐operative immunonutrition alone in one group (Falewee 2014), but data from that group were not included in analyses. A commercial polymeric feed was used in the control group postoperatively in most studies (17/19), some of which contained additional fibre (Table 1). In six studies the control group received a standard polymeric feed preoperatively as well as postoperatively (Falewee 2014; Felekis 2005; Felekis 2010; Ghosh 2012; Sorensen 2009; Van Bokhorst 2000/2001). In one study, two groups (one of which had received the control feed both pre‐ and postoperatively and the other only postoperatively) were combined in their analyses (Snyderman 1999).

Outcomes

Of the outcomes considered, mortality was most commonly reported (14 studies; we obtained unpublished data on mortality for eight studies (Casas‐Rodera 2008; De Luis 2003; De Luis 2004; De Luis 2005; De Luis 2007; De Luis 2009; Falewee 2014; Sorensen 2009), with the remaining data being available in the paper or abstract), followed by wound infection (12 studies), adverse events/tolerance of feeds (11 studies), length of hospital stay (10 studies) and fistulae (10 studies).

Excluded studies

We excluded eight studies (see Characteristics of excluded studies).

We excluded one study (Buijs 2010), which was a follow‐up of patients from a study already included in the review (Van Bokhorst 2000/2001). Buijs 2010 reported on long‐term survival (≥ 10 years) and we felt that including mortality data from a much longer follow‐up time period than all other included studies (which measured mortality in a relatively short period of time post‐intervention) would make the results more difficult to interpret.

Three studies were not randomised (De Luis 2013; Linn 1988; Reis 2016), and four had no suitable control group (De Luis 2005a; De Luis 2010; De Luis 2015; Palma‐Milla 2016).

Risk of bias in included studies

The risk of bias for each study is described in detail in the Characteristics of included studies table. Details of risk of bias judgements for each study are presented in Figure 2, with an overall summary graph in Figure 3. Allocation concealment methods were most poorly reported, resulting in the greatest number of 'unclear' risk of bias assessments. Details of methodological quality are also presented in Table 3.

Figure 2

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

Figure 3

'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

See: Summary of findings for the main comparison Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer

Table 3. Methodological quality of trials

Study	Generation of allocation sequence	Allocation concealment	Power calculations	Patients blinded	Assessors blinded	Analysed as intention‐to‐treat (ITT)
Snyderman 1999	Tables	Not stated	Not stated	Unclear which treatment groups were blinded	Unclear which treatment groups were blinded	Yes ‐ stated ITT
Riso 2000	Computer‐generated*	Sealed envelopes* (not defined as opaque)	Not done*	Stated "double‐blindly performed" but no indication as to who was blinded*	Stated "double‐blindly performed" but no indication as to who was blinded*	Yes ‐ no attrition according to number of participants included in analyses
Van Bokhorst 2000/2001	Computer‐generated	Not stated	Yes	Yes**	Yes**	Yes ‐ no attrition according to number of participants included in analyses for relevant outcomes
De Luis 2002	Tables*	Sealed envelopes* (not defined as opaque)	Yes	Yes	Yes	Yes ‐ stated "Any drop‐outs were present in the study"
De Luis 2003	Tables*	Sealed envelopes (not defined as opaque)	Yes*	Yes	Yes	Yes ‐ stated ITT
De Luis 2004	Tables*	Sealed envelopes* (not defined as opaque)	Yes	States blinded, but no indication as to who was blinded*	States blinded, but no indication as to who was blinded*	Yes ‐ stated ITT
De Luis 2005	Tables*	Sealed envelopes (not defined as opaque)	Yes*	Stated "Study was blinded" but no indication as to who was blinded	Stated "Study was blinded" but no indication as to who was blinded	Yes ‐ stated ITT
Felekis 2005	Not stated	Not stated	Not stated	Not stated	Not stated	Not stated
De Luis 2007	Tables*	Envelopes* (not defined as opaque)	Yes	States blinded, but no indication as to who was blinded*	States blinded, but no indication as to who was blinded*	Yes ‐ stated ITT
Casas‐Rodera 2008	Stated "randomly allocated" but no information on how sequence was generated	Not stated	Not stated	Not stated	Not stated	Yes ‐ stated "no dropouts were present in the study"
De Luis 2009	Stated "randomly allocated" but no information on how sequence was generated	Not stated	Yes	Yes	Yes	Yes ‐ stated ITT
Sorensen 2009	Stated as a "randomized design" but no information on how sequence was generated	Envelopes (not defined as opaque)	Not done	Partial	Partial	Unclear ‐ no statement on dropouts
Felekis 2010	Stated "randomization generator"	Not clear	Not done*	Stated "double blinded" but no indication as to who was blinded	Stated "double blinded" but no indication as to who was blinded	Yes ‐ no loss to follow‐up described and stated "no dropouts occurred due to intolerance"
Ghosh 2012	"Randomisation lists" (pharmacy clinical trials unit)	Pharmacy (central telephone assignment)	Yes	Yes	Yes	Yes ‐ stated ITT
Turnock 2013	Computer‐generated	Opaque, sealed envelopes	Pilot study	No	No	Yes ‐ no attrition according to number of participants included in analyses
De Luis 2014	Not stated	Not stated	Not stated	Yes	Yes	Yes ‐ no attrition
Falewee 2014	Computer‐generated	Pharmacy clinical trials unit	Yes	Yes	Yes	Yes*
Azman 2015	Random ballot picking	Sealed envelopes (not defined as opaque)	Yes	No	No	Yes
Hanai 2018	Not stated	Not stated	Yes	Not stated, but assumed non‐blinded as intervention group received sachets and control group received no intervention	Not stated	Yes ‐ stated ITT (NB: 1 patient, quote "Excluded due to incomplete data")

* Additional information provided by authors.

** Groups 2 and 3 included in analyses; authors stated that these groups were blinded.

Allocation

Random sequence generation

We classed 13 studies at low risk of bias due to acceptable randomisation sequence generation through the use of computer‐generated randomisation, randomisation lists, tables or a "randomization generator" (Azman 2015; De Luis 2002; De Luis 2003; De Luis 2004; De Luis 2005; De Luis 2007; Falewee 2014; Felekis 2010; Ghosh 2012; Riso 2000; Snyderman 1999; Turnock 2013; Van Bokhorst 2000/2001). In six studies the method of random sequence generation was not clear or not stated and therefore we classed this as an unclear risk of bias (Casas‐Rodera 2008; De Luis 2009; De Luis 2014; Felekis 2005; Hanai 2018; Sorensen 2009).

Allocation concealment

We considered 16 studies to be at unclear risk of bias due to inadequately reported methods of allocation concealment. Of these, eight reported the use of envelopes but did not state whether or not they were opaque (Azman 2015; De Luis 2002; De Luis 2003; De Luis 2004; De Luis 2005; De Luis 2007; Riso 2000; Sorensen 2009). We classed the other eight at unclear risk of bias because allocation concealment was either not stated (Casas‐Rodera 2008; De Luis 2009; De Luis 2014; Felekis 2005; Hanai 2018; Snyderman 1999; Van Bokhorst 2000/2001), or was not clear as documented (Felekis 2010). We classed three studies as low risk of bias due to the use of central telephone assignment (Falewee 2014; Ghosh 2012), or the use of opaque, sealed envelopes (Turnock 2013).

Blinding

Participants and personnel

Participants can be adequately blinded with this intervention, therefore we judged studies where patients were not blinded to be at high risk of bias. Two studies stated that they were not blinded and we classed them at a high risk of bias (Azman 2015; Turnock 2013), and one study used sachets in the intervention group and no treatment in the control group, so it was assumed to be unblinded and at high risk of bias (Hanai 2018). We classed nine studies at an unclear risk of bias because they either did not state whether or not participants and personnel were blinded (Casas‐Rodera 2008; Felekis 2005), or they stated that the study was blinded but did not state who was blinded (De Luis 2004; De Luis 2005; Felekis 2010; Riso 2000), or they described partial blinding (Snyderman 1999; Sorensen 2009). We classed all other studies (seven) as at low risk of bias.

Outcome assessment

Judgements for risk of bias in regards to the blinding of outcome assessment were as described above for blinding of participants and personnel with two exceptions. De Luis 2014 stated that "Blinding of patients and dietitians involved in patient treatment was maintained", but they did not indicate who the outcome assessor was; as such, we judged it at unclear (rather than low) risk of bias. Hanai 2018 did not indicate who the outcome assessor was or whether or not anyone was blinded; as such, we judged it at unclear (rather than high) risk of bias.

Incomplete outcome data

We considered most studies at low risk of bias for incomplete outcome data, with the exception of three studies that we classed at an unclear risk of bias because they did not mention whether or not an intention‐to‐treat (ITT) analysis had been conducted/whether any dropouts had occurred, or did not present data in such a way as to be able to determine the number of participants included in analyses (Azman 2015; Felekis 2005; Sorensen 2009). The remaining studies either had no attrition according to the numbers included in tables/figures, or stated that an ITT analysis had been conducted or that there were no dropouts/losses to follow‐up. (NB: we obtained additional data for one study to enable an ITT analysis to be conducted; Falewee 2014).

Selective reporting

We classed most studies (17) at an unclear risk of selective reporting bias because protocols were not available to judge whether or not selective reporting had occurred. We classed one study at a low risk of bias because the primary outcome stated in the protocol was reported in the paper (Falewee 2014) (no secondary outcomes were specified in the protocol so it was not possible to judge whether selective reporting of secondary outcomes had occurred). We classed one study at a high risk of bias because not all primary outcomes stated in the protocol were presented in the manuscript (Turnock 2013).

Other potential sources of bias

Five studies reported problems with recruitment (Falewee 2014; Ghosh 2012; Snyderman 1999; Turnock 2013; Van Bokhorst 2000/2001), three reported baseline differences between treatment groups (Ghosh 2012; Snyderman 1999; Van Bokhorst 2000/2001), and six had poorly defined methods for assessment of wound infection (Casas‐Rodera 2008; De Luis 2002; De Luis 2004; De Luis 2007; De Luis 2009; Felekis 2010). We classed these studies at high risk of bias for these other potential sources of bias. Of the 10 studies that reported length of hospital stay, six did not describe how this was determined (De Luis 2002; De Luis 2004; De Luis 2007; De Luis 2009; Riso 2000; Turnock 2013). The assessment of tolerance of feeds was not based on consistent descriptions across studies.

One study was available only as an abstract and we judged it at unclear risk of other bias (Felekis 2005). We classed the remaining studies at low risk of bias for other potential sources of bias as there was no evidence for this in the published data (Azman 2015; De Luis 2003; De Luis 2005; De Luis 2014; Sorensen 2009).

Most studies were small, with sample sizes that were unrealistically low for detecting clinical complications.

Effects of interventions

Immunonutrition versus standard care

Primary outcomes

Length of hospital stay

Length of hospital stay was reported in 10 studies. The mean length of stay ranged from 15.3 days to 31.1 days in immunonutrition groups and from 17.4 days to 36.1 days in control groups. We found no evidence of a difference between treatment groups in the length of hospital stay, but the confidence interval around the effect estimate was wide (mean difference ‐2.5 days, 95% confidence interval (CI) ‐5.11 to 0.12 (P = 0.06); 10 studies, 757 participants) (GRADE: low‐quality evidence). The results showed little evidence of heterogeneity between studies (Chi² = 12.89, P = 0.17, I² = 30%) (Analysis 1.1).

Wound infections

Wound infections were reported in 12 studies, of which 10 studies reported events. One study reported 'wound complications' as the number with Clavien‐Dindo grades above 3, above 2 or all grades and was not included in the meta‐analysis (Hanai 2018). Absolute risks ranged from 0% (0/45) to 61% (17/28) in the immunonutrition groups and from 0% (0/45) to 59% (17/29) in the control groups. Events were more common (in both treatment groups) in studies that had used pre‐ and postoperative intervention (a total of 95 events among 458 participants) than in studies that used only postoperative intervention (a total of 14 events among 354 participants). We found no evidence of a difference between treatment groups for this outcome. The combined risk ratio (RR) was 0.94 (95% CI 0.70 to 1.26, P = 0.66; 12 studies, 812 participants) (GRADE: very low‐quality evidence), with little evidence of heterogeneity between trials (Chi² = 4.12, P = 0.90, I² = 0%) (Analysis 1.2).

Fistula formation

Fistula formation was reported in 10 studies, all of which reported events. The absolute risk was 5.4% (range 0% (0/23) to 7% (7/105)) in the immunonutrition groups and 11.3% (range 2% (1/47) to 29% (2/7)) in the control groups. There was a reduction in fistula formation with immunonutrition compared to standard care; the combined RR was 0.48 (95% CI 0.27 to 0.85, P = 0.01; 10 studies, 747 participants) (GRADE: low‐quality evidence), with little evidence of heterogeneity between studies (Chi² = 6.99, P = 0.64, I² = 10%) (Analysis 1.3; Figure 4).

Figure 4

Forest plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.3 Fistula formation.

Adverse events/tolerance of feeds

Adverse events in relation to aspects of tolerance of feeds were reported in 11 studies, and included intolerance to feed (Snyderman 1999) and gastrointestinal intolerance (Falewee 2014), abdominal distension, abdominal cramps or emesis (Riso 2000), and diarrhoea (Azman 2015; Casas‐Rodera 2008; De Luis 2002; De Luis 2004; De Luis 2007; De Luis 2009; Hanai 2018). One study stated that "gastrointestinal tract tolerance of both formula diets was excellent in both groups, and no dropouts occurred because of intolerance" (Felekis 2010); for analysis we assumed that there were zero adverse events in both treatment groups. In two studies, a control feed was not used and adverse events were documented only in relation to withdrawals from the immunonutrition treatment groups (one per study); data from these two studies were not included in the meta‐analysis (Azman 2015; Hanai 2018). Among the nine studies that we included in the meta‐analysis, absolute risks ranged from 0% (0/20) to 40% (18/45) in the immunonutrition groups and from 0% (0/20) to 29% (6/21) in the control groups. There was no evidence of a difference between treatment groups for this outcome. The combined RR was 1.33 (95% CI 0.86 to 2.06, P = 0.20; 9 studies, 719 participants) (GRADE: very low‐quality evidence), with little evidence of heterogeneity between trials (Chi² = 7.11, P = 0.42, I² = 2%) (Analysis 1.4).

Secondary outcomes

All‐cause mortality

Mortality was reported in 14 studies (NB: additional unpublished information was obtained from eight authors) and ranged from 0% (0/105) to 14% (4/28) in the immunonutrition groups, and from 0% (0/45) to 8% (2/24) in the control groups. The follow‐up timeframes varied considerably across studies, and ranged from 30 days to greater than or equal to 16 months in those studies that were meta‐analysed (NB: one study did not state the follow‐up timeframe). There was no evidence of a difference between treatment groups for this outcome. The combined RR was 1.33 (95% CI 0.48 to 3.66, P = 0.59; 14 studies, 776 participants) (GRADE: low‐quality evidence), with little evidence of heterogeneity between studies (Chi² = 3.66, P = 0.45, I² = 0%) (Analysis 1.5).

Postoperative complications

Other clinical complications such as pneumonia and urinary tract infections were uncommonly reported (Table 4), but there was no evidence of a reduction (or an increase) with immunonutrition.

Table 4. Other complications

	Number		Other complications definition	Other complications ‐ total		Pneumonia		Urinary tract
Study	Control	Intervention	Other complications definition	Control	Intervention	Control	Intervention	Control	Intervention
Pre‐ and postoperative feeding
Snyderman 1999	47	82	"...infectious complications were judged using CDC criteria and were considered significant if antibiotic therapy was instituted" (NB: total for other complications provided here include "postoperative infection" butnot "wound healing problem" or "fistula")	19	19	‐	‐	‐	‐
Van Bokhorst 2000/2001	15	17	"Postoperative complications were categorized as absent, minor (including minor wound infections, redness and induration of the wound, pulmonary infections, and urinary tract infections), or major (including wound infections requiring surgical drainage, orocutaneous or pharyngocutaneous fistula, flap failure, radiologic signs of anastomotic leakage, respiratory insufficiency, cardiac failure, and septic shock)" (NB: data on individual complications not given in manuscript so the total for other complications provided here may include fistula and wound infection)	7 major	10 major	‐	‐	‐	‐
Felekis 2005	17	20	Reported as major and minor, but otherwise not defined	9 (6 major, 3 minor)	4 (2 major, 2 minor)	‐	‐	‐	‐
Sorensen 2009	7	8	"Serious wound complications were recorded as they occurred and included wound infection, wound dehiscence, and wound fistula. Wound assessments using the ASEPSIS scoring method were performed daily and photographic images were taken on several postoperative days." Authors also noted that "Patients may have had more than one complication" (NB: figures provided here include "wound dehiscence" (classed as major) and "other" (urinary tract infection and nosocomial pneumonia combined as one group) (classed as minor) but exclude "wound fistula" and "wound infection").	2 (0 major 2 minor)	5 (1 major ‐ wound dehiscence; 4 minor)	‐	‐	‐	‐
Felekis 2010	20	20	Minor described in results as "a slight increase of the temperature (< 38°), without an identifiable source of infection". Major described in results as including pneumonia, urinary tract infection, fistula and wound infection. NB: figures provided here do not include fistula and wound infection.	6 (3 minor, 3 major ‐ 2 pneumonia, 1 UTI)	2 (2 minor, 0 major)	2	0	1	0
Ghosh 2012	29	28	"The primary outcome event was defined as any patient with an infection of the lower respiratory tract, gastro‐intestinal tract, urinary tract or blood which required antibiotic treatment and occurred at any time, up to the 30th post‐operative day." "Secondary outcome measures included infections of primary surgical site, neck wound, PEG site, tracheostomy, free‐flap or split skin graft donor site." "...the surgical site/wound infections were defined according to CDC Definitions of Nosocomial Surgical Site infections, 1992 and the ASEPSIS wound score. The diagnosis of the non‐wound infection was as stipulated in the Trial Antibiotic Policy, which also governed how any infective complication diagnosed throughout the trial was to be treated."	8 with chest, urinary, gastrointestinal or blood infection 17 with neck, primary site, donor site, or PEG site infection	12 with chest, urinary, gastrointestinal or blood infection 17 with neck, primary site, donor site, or PEG site infection	4	9	1	2
Turnock 2013	4	4	"General infections (urinary tract infection, respiratory tract infection), flap anastomosis complications (venous or arterial), and wound complications (dehiscence, tissue necrosis, haematoma, chyle leak, salivary fistula or wound infection) were recorded. Infectious complications were judged using CDC criteria and were considered significant if antibiotic therapy was instituted." (NB: figures provided here do not include wound infection).	2 (infectious)	0	‐	‐	‐	‐
Falewee 2014*	104	105	Infectious complications: "systemic infection requiring antibiotic treatment (septicaemia, bacteraemia), surgical site infection (according to CDC Definitions of Nosocomial Surgical Site infections), documented nosocomial pneumopathy, up to the 30th post‐operative day." Surgical site infections (SSI): "primary surgical site, neck wound, free‐flap or split skin graft donor site, and tracheotomy"	‐	‐	‐	‐	‐	‐
Hanai 2018	14	13	Clinical complications presented as numbers of patients with "wound complications" classified according to Clavien‐Dindo system	7	4	‐	‐	‐	‐
Postoperative feeding only
Riso 2000	21	23	"Post‐operative complications were recorded as none, minor (urinary tract infection; respiratory tract infection: abnormal chest X‐ray), and major (fistula; wound infection; spontaneous or surgical purulent drainage and flap necrosis; anastomotic leakage)." (NB: figures provided here exclude reported fistula).	4 major (1 flap necrosis, 3 purulent drainage) 1 minor (respiratory tract)	2 (purulent drainage)	1 (respiratory tract)	0	0	0
De Luis 2002	24	23	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and local complications such as fistula and/or wound infection, assessing all complications using standard methods and the same investigator." (NB: figures provided here exclude reported fistula).	4	5	‐	‐	‐	‐
De Luis 2003	18	18	Not collected	‐	‐	‐	‐	‐	‐
De Luis 2004	45	45	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and local complications such as fistula and/or wound infection. All complications were assessed with standard methods by the same investigator." (NB: figures provided here exclude reported fistula).	4	2	‐	‐	‐	‐
De Luis 2005	15	14	Not collected	‐	‐	‐	‐	‐	‐
De Luis 2007	37	35	"Postoperative complications were recorded as none; general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and wound complications, such as fistula and/or wound infection, assessed all complications with standard methods by the same investigator surgeon." (NB: figures provided here exclude reported fistula).	2	2	‐	‐	‐	‐
Casas‐Rodera 2008	15	14	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen), and wound complications such as fistula and/or wound infection. All complications were assessed with standard methods by the same investigator." (NB: figures provided here exclude reported fistula and wound infection).	1	0	1	0	0	0
De Luis 2009	34	38	"Postoperative complications were registered as none, general infections (urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen and/or respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture) and local complications such as fistula and/or wound infection, assessed all complications with standard methods by the same investigator." (NB: figures provided here exclude reported fistula and wound infection).	8	9	‐	‐	‐	‐
De Luis 2014	40	42	No clinical complications recorded	‐	‐	‐	‐	‐	‐
Azman 2015	22	22	No clinical complications recorded	‐	‐	‐	‐	‐	‐

* Additional data supplied by authors for intention‐to‐treat (ITT) analysis to be conducted; as such, total n differs from that in the published manuscript.

ASEPSIS: Additional treatment, Serous discharge, Erythema, Purulent exudate, Separation of deep tissues, Isolation of bacteria, Stay duration as inpatient
CDC: Centers for Disease Control and Prevention
PEG: percutaneous endoscopic gastrostomy
UTI: urinary tract infection

Subgroup analysis

The direction of effect for length of stay, wound infection, fistula formation and mortality did not differ between studies that had given immunonutrition both pre‐ and postoperatively and studies that had given it only postoperatively (Analysis 1.1; Analysis 1.2; Analysis 1.3; Analysis 1.5). We observed some differences in effect sizes between the subgroups, but most subgroups included six or fewer studies (with the exception of eight studies in the postoperative feeding only subgroup for mortality) and P values for subgroup differences ranged from 0.08 for length of stay (Analysis 1.1) to 0.77 for mortality (Analysis 1.5) (corresponding I² values were 66.6% and 0%, respectively). Studies that had given immunonutrition only postoperatively showed a more beneficial effect on fistula formation than those that had given immunonutrition pre‐ and postoperatively (RR 0.31, 95% CI 0.14 to 0.71, P = 0.005, 6 studies, 354 participants, and RR 0.72, 95% CI 0.33 to 1.62, P = 0.43; 4 studies, 393 participants, respectively). There was a difference in the direction of effect for adverse events/tolerance of feeds between studies that had given immunonutrition both pre‐ and postoperatively and studies that had given it only postoperatively, but such a comparison is not warranted as only two studies were included in the pre‐ and postoperative feeding analysis.

For wound infections, more events were reported in the six studies that used pre‐ and postoperative immunonutrition (n = 48 events among 247 participants in the immunonutrition group and n = 47 events among 211 participants in the standard care group) than in the six studies that used only postoperative immunonutrition (n = 5 events among 178 participants in the immunonutrition group and n = 9 events among 176 participants in the standard care group).

Only one study gave pre‐operative immunonutrition alone so it was not possible to conduct a subgroup analysis of preoperative immunonutrition versus placebo drink (Falewee 2014).

We did not conduct subgroup analyses according to the type of surgery as data were not presented in such a way in the original publications as to allow this to be done.

Sensitivity analysis

We did not conduct sensitivity analyses in which we removed studies judged at high risk of bias for at least two of the factors assessed or where we removed quasi‐randomised trials because no studies met these criteria. We calculated random‐effects estimates for each outcome variable due to the extent of clinical heterogeneity. We did not judge fixed‐effect estimates to be appropriate.

Publication bias

We examined publication bias for all outcomes by visual inspection of funnel plots and we have presented three of these: length of hospital stay (Figure 5), wound infection (Figure 6) and fistula formation (Figure 7). Study numbers were relatively small for these outcomes, which made it difficult to definitively assess publication bias; however, there was a suggestion of publication bias for length of hospital stay (Figure 5). Given the absence of clear publication bias, however, we did not conduct a trim and fill analysis for the primary outcomes.

Figure 5

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.1 Postoperative length of hospital stay [days].

Figure 6

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.2 Wound infection.

Figure 7

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.3 Fistula formation.

Discussion

Summary of main results

Pooled estimates showed no evidence of a difference in the length of hospital stay between treatment groups. The mean difference was ‐2.5 days but the estimate was imprecise (95% confidence interval (CI) ‐5.11 to 0.12) and included the null value. There may be an approximate halving of the risk of fistulae formation (risk ratio (RR) 0.48) with immunonutrition, but the evidence was of low quality. Immunonutrition may have little or no effect on wound infection and mortality, and there was little or no difference in adverse events such as diarrhoea between the treatment groups. We did not formally meta‐analyse other complications because of their heterogeneity. The findings are summarised in summary of findings Table for the main comparison. Length of hospital stay was reduced in 8 of the 10 studies where it was recorded. No reduction in hospital stay was seen in the largest recent study (Falewee 2014). Reduced fistula formation was seen in patients receiving immunonutrition, but no other reductions in clinical complications such as wound infections were seen. No substantial differences in the findings were seen when looking at the timing of intervention (i.e. pre‐ and postoperative or just postoperative).

Where stated, all but two of the studies looking at in‐hospital postoperative immunonutrition used arginine as an immunonutrient; Azman 2015 used glutamine and Hanai 2018 used an eicosapentaenoic acid (EPA)‐enriched supplement. Ten studies evaluated postoperative nutrition alone and nine studies included pre‐ and postoperative nutrition. One study provided only preoperative immunonutrition in one of their treatment groups (Falewee 2014), but the numbers of patients and complications were too small to draw any conclusions. Studies that gave immunonutrition only postoperatively showed a larger beneficial effect on fistula formation than studies that gave it both pre‐ and postoperatively (RR 0.31 versus 0.72, respectively). One study that gave immunonutrition both pre‐ and postoperatively reported more fistulas with immunonutrition than standard care (Snyderman 1999). The reasons for this are not clear and are in contrast with all other studies (irrespective of timing of intervention), which reported fewer fistulas with immunonutrition than standard care. The cause of mortality was often poorly reported and it may be inappropriate to ascribe any difference in mortality to immunonutrition. However, no evidence of any effect of immunonutrition on mortality (which was low) was seen. We did not formally analyse potential clinical complications such as pneumonia, diarrhoea, cardiovascular effects and the relationship of nutritional status to outcomes due to either limited data or outcomes not having been recorded.

Overall, most studies were too small to provide precise estimates of intervention effects.

Overall completeness and applicability of evidence

Completeness

We attempted to identify and synthesise all existing research to provide a comprehensive estimate of the effect of immunonutrition on postoperative recovery following head and neck cancer surgery. We included 19 studies that recruited 1099 participants. We also conducted the largest systematic review prior to this one, which included 10 randomised controlled trials (RCTs) that had recruited 605 participants (Stableforth 2009). One of the trials (De Luis 2005a), which had been included in our previous review, was not included here as the immunonutrition intervention was administered at the point of discharge. Another systematic review from 2012 included 14 studies with 601 participants (Casas Rodera 2012); three of the studies in that review were not relevant to ours as two studies compared two doses/types of immunonutrients (De Luis 2005a; De Luis 2010), and one study, as noted above, was part of another publication (Buijs 2010). A third systematic review from 2014 included six studies and 397 participants, all of which were included in our review (Vidal‐Casariego 2014). Despite being the largest systematic review to date, it is possible that our search strategies may not have identified all of the existing literature.

We looked at similar outcomes to other systematic reviews, but we did not analyse further outcomes reported in studies such as pneumonia, cardiovascular effects and the relationship of nutritional status to outcomes due to limited data.

This review was systematic, using extensive searches of several databases and inclusive search terms. We did not include unpublished literature, but we felt it unlikely that there are large unpublished trials that demonstrate a substantial effect of immunonutritional interventions in head and neck cancer. We made attempts to contact senior or corresponding authors as published in the original papers. We received further data from Dr De Luis, Professor Jones (re: Ghosh 2012), Dr Riso and Dr Falewee regarding research methods and outcomes.

Applicability

Most studies applied exclusion criteria to individuals for study participation. These frequently included renal or hepatic impairment, existing infection and altered immune function. Three studies excluded participants who were well nourished (De Luis 2003; De Luis 2005; Van Bokhorst 2000/2001), one excluded those who were morbidly obese (Snyderman 1999), and seven excluded people who were malnourished (Azman 2015; De Luis 2002; De Luis 2004; De Luis 2007; De Luis 2009; De Luis 2014; Turnock 2013).

Studies included in this review were conducted in various countries, incorporating a range of cultures and healthcare systems. Eight were undertaken in Spain (seven of which were conducted by the same group), two in the United States, one in New Zealand, one in Malaysia, one in Japan and the remainder in western Europe. This may have had an effect on outcomes. For example, standard healthcare practice (e.g. discharge policies, implementation of Enhanced Recovery After Surgery (ERAS) protocols, etc.) is likely to vary across countries, thus making comparisons across studies less meaningful. In addition, the studies were reported over an approximate 18‐year period during which there are likely to have been changes in clinical practice. However, it was not possible from the information provided to assess the impact of this possibility. There were also differences between studies regarding patient populations and types of surgery. As such, it remains unknown who is most likely to benefit from immunonutrition (if indeed there is a true benefit) based on the included studies.

Quality of the evidence

Assessments of the quality of evidence for each outcome are presented in summary of findings Table for the main comparison.

Methodology

In general studies were poorly reported with a large proportion of unclear risk of bias assignments for several of the items assessed (Figure 2; Figure 3). We assigned blinding of outcome assessment an unclear risk of bias in 11 studies and 'other' bias a high risk of bias in 11 studies (examples of 'other' bias included difficulties in recruitment/not meeting target sample size, and evidence of some baseline differences between treatment groups).

The generation of the random allocation sequence was reported in 13 studies (Table 3). In nine studies allocation was concealed using sealed envelopes but it was not stated whether or not these were opaque in eight of the studies (one explicitly stated that opaque envelopes were used). Concealment of allocation was achieved by using a central telephone assignment in two studies. Where stated, all studies were double‐blind (as opposed to single‐blind), although there was an overall lack of description of how this was achieved. One study (published as an abstract; Felekis 2005) did not state whether an intention‐to‐treat (ITT) analysis was conducted, and for another study we obtained additional data from the authors to enable an ITT analysis to be conducted (Falewee 2014).

More than half of the studies reported use of sample size or power calculations (Table 3), but of those that did, several did not meet the target sample size. Many other studies included small sample sizes, reducing the power of the study to observe clinically important differences in outcomes.

The GRADE rating of the evidence varied for all of the outcomes assessed, but we classed none as high‐quality (summary of findings Table for the main comparison). The main reasons for downgrading the evidence were small sample sizes and wide confidence intervals around effect estimates, and poor descriptions of the methods used for assessing outcomes reported within studies.

Outcome assessment

Infection is the clinical outcome of interest in this research area. However, this was poorly defined in most studies. Wound infections were not classified by site or severity in most studies. Systemic infection was also poorly defined. Similarly, persistent postoperative fistula was not defined in most studies. A more precise measure of infection and fistula could improve the quality of the evidence base. Wound infections were considerably more common (in both treatment groups) in studies that used pre‐ and postoperative supplementation than in studies that used only postoperative supplementation, but the reasons for this are not clear. Four of the six studies that used pre‐ and postoperative supplementation and that reported wound infections were from the same research group, and it is possible that their definition of wound infection or their length of follow‐up meant that fewer wound infections were captured in their studies.

Blinding of participants with this intervention is possible and should be undertaken, as an awareness of treatment allocation may result in participants misreporting outcomes. Length of hospital stay is likely to be influenced by variation in discharge criteria, which may result in differences between studies. This lack of uniformity across centres may introduce variability for some outcomes.

Heterogeneity

We observed little evidence of heterogeneity for each outcome (Chi² ranged from 3.66 to 12.89, P = 0.17 to 0.91, and I² ranged from 0% to 30%). For length of hospital stay and fistula formation, if heterogeneity existed it would be more likely to indicate variation in size of effect as opposed to direction of effect given that most studies suggested a beneficial effect of immunonutrition on postoperative recovery outcomes. Visual inspection of the forest plots and associated data did not indicate that size of study substantially altered the effect size (although it must be noted that the majority of studies had sample sizes of fewer than 25 participants per treatment group).

Potential biases in the review process

Search strategy

Although we believe that our electronic (February 2018) and handsearching strategies have identified all relevant studies, it is possible that we may have missed some available literature or unpublished material. We stopped handsearching at the end of January 2017. We have read reviews and references of recent publications and in the time period until publication other studies may have been published or made available. These will be incorporated into future updates of this review.

Assumptions about the mechanism of effect

The various components of immunonutrition supplements have been shown in studies done in vitro and in vivo to produce what are considered beneficial changes in immunological function. However, there is little evidence from clinical trials that these mechanisms result in reduced postoperative complications. In particular there is little evidence of the superiority of any given immunonutrient over another, for any given dose regimen, or for the time periods for which the immunonutrients need to be taken to produce benefit. The small size of most studies and the variety of dosing regimens meant that it was not possible to comment on the relative merits of each type or dose of immunonutrient. Future studies are required to examine these issues.

Assumptions about the meta‐analyses and results

We think it unlikely that we have introduced bias through the methods used in the review process. The range of outcome metrics reported across studies was small and did not require conversion to common units for use in this review. We used a random‐effects model (due to a high level of clinical heterogeneity among included studies), which may have resulted in smaller studies being granted a larger weighting than necessary, potentially biasing the overall meta‐analysed results (Handbook 2011). We identified possible publication bias from visual inspection of the funnel plot for length of hospital stay (see Figure 5). Some of the studies published by De Luis et al have similar starting dates and trial designs (see Table 2), however the baseline data are different and in their 2004 paper they reference their 2002 paper as a different study. In their 2009 paper the patient characteristics (age, sex) and baseline data are different from their 2007 paper. However, we have been unable to obtain a response from the authors to clarify that there is no overlap in participants across studies.

We did not formally assess biochemical changes or immunological changes as secondary outcomes (as per the original protocol) because very few papers commented on such changes, and in each paper the markers chosen were different and assessed at differing time intervals. Meta‐analysis of the few papers was thus not possible. Furthermore, given the expected profound influence of the operative inflammatory response on levels of such markers, their interpretation is not straightforward.

Assumptions about study methodology

Studies were not excluded on the basis of methodological quality, but exclusion of poor‐quality studies would tend to move effect estimates towards the null. The major limitations of the review relate to the limitations of the literature, and we made a number of assumptions about the comparability of study methodology. In general, complications (especially wound infection) were poorly defined, and follow‐up timeframes differed considerably across studies (summary of findings Table for the main comparison). It may be difficult to detect any effects on postoperative outcomes in studies conducted at a late stage of disease. Different interventions may not have an equal effect, or even the same direction of effect, for different cancer sites and stages. We stratified our analysis according to whether or not immunonutrition was given both pre‐ and postoperatively or only postoperatively, but our analyses did not adjust for differences in the composition or volume of immunonutrition formulas provided, nor did they take into account the length of time for which participants were fed. Few studies reported on compliance with the intervention, but given that the feed was usually administered enterally (at least postoperatively) we assumed good compliance levels. Such diversity across studies may mean that results varied due to one or more of these factors, but we feel that this was unlikely to have greatly affected our findings. Of note, seven studies came from one centre, all of which had relatively small sample sizes (total sample sizes in these seven trials ranged from 29 to 90 participants). A number of factors that we were unable to control for also may have affected outcomes, such as the experience of the surgeon, the length of the operation and the success of the operation.

Agreements and disagreements with other studies or reviews

Other reviews have been published on this topic, with similarly positive results (Casas Rodera 2012; Stableforth 2009; Vidal‐Casariego 2014). The most recent systematic review and meta‐analysis included six head and neck cancer surgery studies (Vidal‐Casariego 2014). Compared to that review, we observed a lower reduction in length of hospital stay (2.5 days in our analysis versus 6.8 days in theirs). As done here, the authors of that review also conducted analyses based on the timing of administration of immunonutrition. In contrast with that review, however, we did not meta‐analyse infections other than wound infections due to the diverse range reported in studies. Our finding of little difference between treatments in wound infections is similar, as is the reduction in fistula formation with immunonutrition (Vidal‐Casariego 2014).

No prior review considered complications directly related to the immunonutrition intervention, and none were found in this review.

Some studies have reported on the tolerability of immunonutrition, but few in detail. In their systematic review, Vidal‐Casariego et al reported no increase in diarrhoea, although this was based on very few studies (Vidal‐Casariego 2014). Our review suggests that immunonutrition is generally as well tolerated as standard supplements in the head and neck cancer surgery patient group, but this finding is also based on very few studies.

Cost has not been reported in systematic reviews and nor were we able to assess costs due to a lack of published data. Snyderman 1999 considered costs in their study and suggested that a reduction in the infection rate between treatment groups (rates were reported as 23% and 45% in the immunonutrition and standard therapy groups, respectively) could reduce costs given the difference in length of hospital stay between those with and without infectious complications.

Overall, our findings are in agreement with other reviews in head and neck cancer surgery (Casas Rodera 2012; Stableforth 2009; Vidal‐Casariego 2014), as well as reviews in gastrointestinal surgery (Marimuthu 2012; Zhang 2012), and suggest a potential benefit of immunonutrition. However, in agreement with Vidal‐Casariego 2014, the findings for head and neck cancer are based on poor‐quality evidence.

There were insufficient data to exclude substantial effects of immunonutrition on other clinical outcomes or biochemical and immunological parameters.

Figure 1

Process for sifting search results and selecting studies for inclusion.

Figure 2

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

Figure 3

'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 4

Forest plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.3 Fistula formation.

Figure 5

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.1 Postoperative length of hospital stay [days].

Figure 6

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.2 Wound infection.

Figure 7

Funnel plot of comparison: 1 Immunonutrition versus standard care, outcome: 1.3 Fistula formation.

Analysis 1.1

Comparison 1 Immunonutrition versus standard care, Outcome 1 Postoperative length of hospital stay.

Analysis 1.2

Comparison 1 Immunonutrition versus standard care, Outcome 2 Wound infection.

Analysis 1.3

Comparison 1 Immunonutrition versus standard care, Outcome 3 Fistula formation.

Analysis 1.4

Comparison 1 Immunonutrition versus standard care, Outcome 4 Adverse events.

Analysis 1.5

Comparison 1 Immunonutrition versus standard care, Outcome 5 All‐cause mortality.

Summary of findings for the main comparison. Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer

Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer
Patient or population: patients undergoing surgery for head and neck cancer Setting: hospitals (international) Intervention: immunonutrition Comparison: standard care
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with standard care	Risk with immunonutrition
Postoperative length of hospital stay (days) Follow‐up: 8 to 90 days post surgery or hospital discharge	The mean of reported length of hospital stay (mean values) across the standard care groups was 27.0 (17.4 to 36.1) days	The mean of reported length of hospital stay (mean values) across the immunonutrition groups was 23.2 (15.3 to 31.1) days	MD 2.5 lower (5.11 lower to 0.12 higher)	757 (10 RCTs)	⊕⊕⊝⊝ LOW¹	There may be a reduction in the length of hospital stay of 2.5 days with immunonutrition, but the estimate is imprecise (wide CI) and includes the null value.
Wound infection Follow‐up: 8 to 90 days post surgery or hospital discharge	Study population		RR 0.94 (0.70 to 1.26)	812 (12 RCTs)	⊕⊝⊝⊝ VERY LOW^1,2	Immunonutrition may have little or no effect on wound infection, but the evidence is very uncertain.
	145 per 1000	136 per 1000 (101 to 182)
Fistula formation Follow‐up: 8 to 90 days post surgery or hospital discharge	Study population		RR 0.48 (0.27 to 0.85)	747 (10 RCTs)	⊕⊕⊝⊝ LOW ¹	There may be an approximate halving of the risk of fistulae formation but the evidence is of low quality.
	113 per 1000	54 per 1000 (31 to 96)
Adverse events/tolerance of feeds Follow‐up: 10 to 90 days post surgery or hospital discharge	Study population		RR 1.33 (0.86 to 2.06)	719 (9 RCTs)	⊕⊝⊝⊝ VERY LOW^1,3	There may be little or no difference in adverse events such as diarrhoea between the treatment groups, but the evidence is very uncertain.
	91 per 1000	121 per 1000 (78 to 188)
All‐cause mortality Follow‐up: 30 days to greater than or equal to 16 months post surgery	Study population		RR 1.33 (0.48 to 3.66)	776 (14 RCTs)	⊕⊕⊝⊝ LOW¹	Immunonutrition may have little or no effect on mortality.
	18 per 1000	24 per 1000 (9 to 67)
*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; MD: mean difference; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: 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 certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹Downgraded by two levels for imprecision: most studies had small sample sizes and confidence intervals around the summary estimates were wide. ²Downgraded by one level for risk of bias: assessment of wound infection was poorly reported across studies. ³Downgraded by one level for risk of bias: assessment of adverse events was poorly reported across studies and not all studies measured the same adverse events.

Summary of findings for the main comparison. Immunonutrition compared to standard care for patients undergoing surgery for head and neck cancer

Table 1. Interventions

						Duration of supplements
Study	Groups	Control	Active	Isocaloric/isonitrogenous	Target energy intake	Pre‐operation	Post‐operation	Length of follow‐up
Snyderman 1999	1. Active pre‐ and postoperative 2. Active postoperative only 3. Control pre‐ and postoperative 4. Control postoperative only	Polymeric (Replete, Resource, Isosource, Jevity, Vivonex, Osmolite)	Polymeric + arginine (Impact)	Not stated	500 mL per day pre‐operation (500 kcal) 1000 mL per day post‐operation (1000 kcal)	≥ 5 days	≥ 7 days	1 month post‐operation
Riso 2000	1. Active postoperative enteral 2. Control postoperative enteral Both groups received parenteral nutrition for 3 days postoperatively to achieve nutritional goal	Polymeric (Nutrison protein plus)	Polymeric + arginine (Nutrison intensive)	Yes	31 kcal/kg per day by POD4	None	≥ 10 following total laryngectomy, ≥ 21 days following partial laryngectomy	To hospital discharge
Van Bokhorst 2000/2001	1. No pre‐operative nutritional support + postoperative standard formula 2. Pre‐operative + postoperative standard formula 3. Arginine supplemented pre‐ and postoperative	Polymeric	Polymeric + arginine ("41% of casein proteins were replaced by arginine")	Yes	150% of basal requirement	7 to 10 days	≥ 10 days	7 days post‐operation and greater than or equal to 16 months (survival)
De Luis 2002	1. Postoperative supplement with arginine + fibre 2. Postoperative polymeric control	Polymeric	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	Average duration 22 days (± 12 days) across groups	14 days postoperatively, and 3 months post hospital discharge (mortality)
De Luis 2003	1. Postoperative enteral supplement with arginine + fibre 2. Postoperative polymeric control	Polymeric	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	Average duration 20 days in group 1	5 days post‐operation
De Luis 2004	1. Postoperative polymeric + arginine + fibre 2. Postoperative polymeric control + fibre	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (0.625 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	≥ 10 days	Day 14
De Luis 2005	1. Postoperative polymeric + arginine 2. Postoperative polymeric control	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (12.5 g/day) + fibre (0.9 g/100 mL)	Yes	Not stated	None	Average duration 20 days in group 1	6 days post‐operation
Felekis 2005	1. Active pre‐ and postoperative 2. Regular oral diet and standard polymeric enteral feeding pre‐ and postoperatively	Polymeric	Undefined enteral immunonutrition	Yes	Not stated	6 days	8 days	Not stated
De Luis 2007	1. Postoperative polymeric + arginine 2. Postoperative polymeric control	Polymeric + fibre (0.9 g/100 mL)	Polymeric + arginine (0.85 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg per day by POD4	None	≥ 10 days	12 days post‐operation
Casas‐Rodera 2008	1. Postoperative arginine enhanced formula 2. Polymeric control 3. Postoperative arginine enhanced formula RNA and omega‐3 fatty acids	Group 2 polymeric	Group 1 arginine‐enhanced formula (NB data from Group 1 not used in analyses) Group 3 arginine (1.3 g/100 mL), RNA (0.12 g/100 mL) and ratio of ω6:ω3 fats noted as "0.7 g"	No (control 122 kcal/100 mL, active 101 kcal/100 mL)	Requirements (used Harris Benedict formula with a stress factor correction of 1.4) by POD3	None	Average duration of 14.5 ± 8 days	14 days post‐surgery
De Luis 2009	1. Active postoperative (arginine) + fibre 2. Control postoperative (standard enteral nutrition)	Polymeric	Polymeric + arginine (0.85 g/100 mL) + fibre (0.9 g/100 mL) (NB control and active formulas contained the same ratio of ω6:ω3 fats)	Yes	32 kcal/kg; 1.7 g protein/kg on POD4	None	≥ 10 days	10 days post‐operation
Sorensen 2009	1. Active pre‐ and postoperative (arginine, glutamine, nucleotides and omega‐3 enriched) 2. Control pre‐ and postoperative (standard enteral nutrition)	Isosource	Polymeric + arginine and glutamine (Impact Recover (oral drink, containing 16.3 g arginine, 20 g glutamine, 1.6 g nucleotides, and 1.7 g ω3 fatty acids/L) or Impact glutamine (tube feed, containing 16.3 g arginine, 15 g glutamine, 1.6 g nucleotides and 2.7 g ω3 fatty acids/L))	No (Impact Recover = 960 kcal/L; Impact glutamine = 1300 kcal/L; Isosource = 1440 kcal/L)	˜1 L per day pre‐operatively, at least 1 L per day postoperatively (˜50% to 60% of basal requirements from intervention)	7 days	7 days	29 days post‐operation
Felekis 2010	1. Control pre‐ and postoperative 2. Active pre‐ and postoperative	Polymeric (Nutrison, Nutricia)	Polymeric + ω3 fatty acids, arginine, RNA (Impact, Novartis)	Yes	Requirement (based on Harris and Benedict equations) by POD4	5 days	8 days	8 days post‐operation
Ghosh 2012	1. Control pre‐ and postoperative 2. Active pre‐ and postoperative	Polymeric	Polymeric + ω3 fatty acids (1.7 g/L), arginine (1.25 g/L), RNA (1.2 g/L) (Impact)	Yes	Nutritional requirements based on Schofield calculation, 500 mL/day pre‐operatively, 1000 mL/day postoperatively (also allowed to eat and drink if able). Any calorific shortfall was made up using standard proprietary feeds.	5 days	7 days	30 days post‐operation
Turnock 2013	1. Control ‐ no supplements pre‐operation, postoperative standard supplements 2. Active ‐ pre‐ (oral) and postoperative (enteral then oral to day 5) supplements	Isosource standard (postoperative only)	Polymeric + ω3 fatty acids (3.3 and 1.7 g/L EPA and DHA for oral and enteral products, respectively), arginine (12.6 and 12.5 g/L for oral and enteral products), RNA (1.3 and 1.2 g/L for oral and enteral products) (Impact)	No (oral/enteral Impact 1000 kcal/L; Isosource standard 1200 kcal/L)	Requirements (25 to 30 kcal/kg)	5 days	≥ 5 days	Hospital discharge
De Luis 2014	1. Active postoperative 2. Control postoperative	Polymeric	Polymeric + arginine (8 g/L, 20 g/day) (NB control and active formulas contained the same ratio of ω6:ω3 fats and the same amount of fibre 13.8 g/L)	No (control 1.118 kcal/L, active 1.020 kcal/L)	Requirements (35 kcal/kg, 1.7 g protein/kg)	None	Minimal of 15 days	10 days post‐surgery
Falewee 2014	1. Pre‐ and postoperative control 2. Pre‐operative active, postoperative control 3. Pre‐ and postoperative active	Impact without immunonutrients	Polymeric + ω3 fatty acids (1.0 g per sachet (oral) and 1.65 g/500 mL (enteral)), arginine (3.8 g per sachet (oral) and 6.5 g/500 mL (enteral)), RNA (0.45 g per sachet (oral) and 0.65 g/500 mL (enteral)) (Oral Impact preoperatively and Enteral Impact postoperatively)	Yes	Requirements (calculated using SFNEP French recommendations)	7 days	7 to 15 days	90 days post‐surgery
Azman 2015	1. Postoperative active 2. Postoperative no intervention	None	Glutamine powder (30 g/day) (Glutamine Plus, Fresenius Kabi)	No control intervention	30 to 40 kcal/kg/day	None	4 weeks	4 weeks post‐operation
Hanai 2018	1. Pre‐ and postoperative active 2. Pre‐and postoperative no intervention	None	Prosure (an eicosapentaenoic acid (EPA) enriched oral nutritional supplement) at a dose of 2 packs per day (480 mL)	No control intervention	Prosure was administered in addition to a normal diet (or in lieu of part of a normal diet). Quote: "The dietary intake was not limited"	14 days	14 days	14 days post‐operation
DHA: docosahexaenoic acid EPA: eicosapentaenoic acid POD: postoperative day RNA: ribonucleic acid SFNEP: Société Francophone de Nutrition Entérale et Parentérale

Table 1. Interventions

Table 2. Baseline patient characteristics

	Number		Mean age		Sex M:F		Mean weight (kg)		BMI
Study	Control	Intervention	Control	Intervention	Control	Intervention	Control	Intervention	Control	Intervention
Snyderman 1999	47	82	61	63	32:15	63:19	67	71	Not reported	Not reported
Riso 2000	21	23	63	61	18:3	21:2	66	64	23.2	22.1
Van Bokhorst 2000/2001	15	17	60	59	7:8	12:5	55	62	Not reported	Not reported
De Luis 2002	24	23	59	63	3:21	2:21	68	68	24.1	26.2
De Luis 2003	18	18	59	63	1:27	1:17	69	69	24.1	26.2
De Luis 2004	45	45	61	60	3:42	3:42	69	70	25.1	25.2
De Luis 2005	15	14	63	61	3:12	2:12	Not reported	Not reported	24.1	24.6
Felekis 2005	17	20	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
De Luis 2007	37	35	62	62	3:34	4:31	69	68	25.1	24.0
Casas‐Rodera 2008*	15	14	54	50	15:0	14:0	66	68	Not reported	Not reported
De Luis 2009	34	38	61	63	27:7	30:8	71	73	26.4	26.5
Sorensen 2009	7	8	59	62	7:0	8:0	69	71	22.3	22.7
Felekis 2010	20	20	63	61	18:2	18:2	Not reported	Not reported	Not reported	Not reported
Ghosh 2012**	29	28	60	62	26:3	24:4	65	72	24	26
Turnock 2013	4	4	47	51	3:1	3:1	67	67	Not reported	Not reported
De Luis 2014	40	42	63.6	65.5	35:5	37:5	70	72	25.1	25.7
Falewee 2014***	104	105	59	59	86:18	87:18	69	70	23.7	23.6
Azman 2015****	22	22	Not reported	Not reported	15:7	9:13	Not reported	Not reported	Not reported	Not reported
Hanai 2018	14	13	66.1	61.5	8:6	8:5	Not reported	Not reported	Not reported	Not reported
* Two different values for weight reported in manuscript: data from Table II in manuscript are reported here. Data reported in manuscript (and here) as median values for age, weight and body mass index (BMI). * Additional data supplied by authors. **** Median age across both treatment groups was 49 years (range 22 to 74 years). No data by treatment group available.

Table 2. Baseline patient characteristics

Table 3. Methodological quality of trials

Study	Generation of allocation sequence	Allocation concealment	Power calculations	Patients blinded	Assessors blinded	Analysed as intention‐to‐treat (ITT)
Snyderman 1999	Tables	Not stated	Not stated	Unclear which treatment groups were blinded	Unclear which treatment groups were blinded	Yes ‐ stated ITT
Riso 2000	Computer‐generated*	Sealed envelopes* (not defined as opaque)	Not done*	Stated "double‐blindly performed" but no indication as to who was blinded*	Stated "double‐blindly performed" but no indication as to who was blinded*	Yes ‐ no attrition according to number of participants included in analyses
Van Bokhorst 2000/2001	Computer‐generated	Not stated	Yes	Yes**	Yes**	Yes ‐ no attrition according to number of participants included in analyses for relevant outcomes
De Luis 2002	Tables*	Sealed envelopes* (not defined as opaque)	Yes	Yes	Yes	Yes ‐ stated "Any drop‐outs were present in the study"
De Luis 2003	Tables*	Sealed envelopes (not defined as opaque)	Yes*	Yes	Yes	Yes ‐ stated ITT
De Luis 2004	Tables*	Sealed envelopes* (not defined as opaque)	Yes	States blinded, but no indication as to who was blinded*	States blinded, but no indication as to who was blinded*	Yes ‐ stated ITT
De Luis 2005	Tables*	Sealed envelopes (not defined as opaque)	Yes*	Stated "Study was blinded" but no indication as to who was blinded	Stated "Study was blinded" but no indication as to who was blinded	Yes ‐ stated ITT
Felekis 2005	Not stated	Not stated	Not stated	Not stated	Not stated	Not stated
De Luis 2007	Tables*	Envelopes* (not defined as opaque)	Yes	States blinded, but no indication as to who was blinded*	States blinded, but no indication as to who was blinded*	Yes ‐ stated ITT
Casas‐Rodera 2008	Stated "randomly allocated" but no information on how sequence was generated	Not stated	Not stated	Not stated	Not stated	Yes ‐ stated "no dropouts were present in the study"
De Luis 2009	Stated "randomly allocated" but no information on how sequence was generated	Not stated	Yes	Yes	Yes	Yes ‐ stated ITT
Sorensen 2009	Stated as a "randomized design" but no information on how sequence was generated	Envelopes (not defined as opaque)	Not done	Partial	Partial	Unclear ‐ no statement on dropouts
Felekis 2010	Stated "randomization generator"	Not clear	Not done*	Stated "double blinded" but no indication as to who was blinded	Stated "double blinded" but no indication as to who was blinded	Yes ‐ no loss to follow‐up described and stated "no dropouts occurred due to intolerance"
Ghosh 2012	"Randomisation lists" (pharmacy clinical trials unit)	Pharmacy (central telephone assignment)	Yes	Yes	Yes	Yes ‐ stated ITT
Turnock 2013	Computer‐generated	Opaque, sealed envelopes	Pilot study	No	No	Yes ‐ no attrition according to number of participants included in analyses
De Luis 2014	Not stated	Not stated	Not stated	Yes	Yes	Yes ‐ no attrition
Falewee 2014	Computer‐generated	Pharmacy clinical trials unit	Yes	Yes	Yes	Yes*
Azman 2015	Random ballot picking	Sealed envelopes (not defined as opaque)	Yes	No	No	Yes
Hanai 2018	Not stated	Not stated	Yes	Not stated, but assumed non‐blinded as intervention group received sachets and control group received no intervention	Not stated	Yes ‐ stated ITT (NB: 1 patient, quote "Excluded due to incomplete data")
* Additional information provided by authors. ** Groups 2 and 3 included in analyses; authors stated that these groups were blinded.

Table 3. Methodological quality of trials

Table 4. Other complications

	Number		Other complications definition	Other complications ‐ total		Pneumonia		Urinary tract
Study	Control	Intervention	Other complications definition	Control	Intervention	Control	Intervention	Control	Intervention
Pre‐ and postoperative feeding
Snyderman 1999	47	82	"...infectious complications were judged using CDC criteria and were considered significant if antibiotic therapy was instituted" (NB: total for other complications provided here include "postoperative infection" butnot "wound healing problem" or "fistula")	19	19	‐	‐	‐	‐
Van Bokhorst 2000/2001	15	17	"Postoperative complications were categorized as absent, minor (including minor wound infections, redness and induration of the wound, pulmonary infections, and urinary tract infections), or major (including wound infections requiring surgical drainage, orocutaneous or pharyngocutaneous fistula, flap failure, radiologic signs of anastomotic leakage, respiratory insufficiency, cardiac failure, and septic shock)" (NB: data on individual complications not given in manuscript so the total for other complications provided here may include fistula and wound infection)	7 major	10 major	‐	‐	‐	‐
Felekis 2005	17	20	Reported as major and minor, but otherwise not defined	9 (6 major, 3 minor)	4 (2 major, 2 minor)	‐	‐	‐	‐
Sorensen 2009	7	8	"Serious wound complications were recorded as they occurred and included wound infection, wound dehiscence, and wound fistula. Wound assessments using the ASEPSIS scoring method were performed daily and photographic images were taken on several postoperative days." Authors also noted that "Patients may have had more than one complication" (NB: figures provided here include "wound dehiscence" (classed as major) and "other" (urinary tract infection and nosocomial pneumonia combined as one group) (classed as minor) but exclude "wound fistula" and "wound infection").	2 (0 major 2 minor)	5 (1 major ‐ wound dehiscence; 4 minor)	‐	‐	‐	‐
Felekis 2010	20	20	Minor described in results as "a slight increase of the temperature (< 38°), without an identifiable source of infection". Major described in results as including pneumonia, urinary tract infection, fistula and wound infection. NB: figures provided here do not include fistula and wound infection.	6 (3 minor, 3 major ‐ 2 pneumonia, 1 UTI)	2 (2 minor, 0 major)	2	0	1	0
Ghosh 2012	29	28	"The primary outcome event was defined as any patient with an infection of the lower respiratory tract, gastro‐intestinal tract, urinary tract or blood which required antibiotic treatment and occurred at any time, up to the 30th post‐operative day." "Secondary outcome measures included infections of primary surgical site, neck wound, PEG site, tracheostomy, free‐flap or split skin graft donor site." "...the surgical site/wound infections were defined according to CDC Definitions of Nosocomial Surgical Site infections, 1992 and the ASEPSIS wound score. The diagnosis of the non‐wound infection was as stipulated in the Trial Antibiotic Policy, which also governed how any infective complication diagnosed throughout the trial was to be treated."	8 with chest, urinary, gastrointestinal or blood infection 17 with neck, primary site, donor site, or PEG site infection	12 with chest, urinary, gastrointestinal or blood infection 17 with neck, primary site, donor site, or PEG site infection	4	9	1	2
Turnock 2013	4	4	"General infections (urinary tract infection, respiratory tract infection), flap anastomosis complications (venous or arterial), and wound complications (dehiscence, tissue necrosis, haematoma, chyle leak, salivary fistula or wound infection) were recorded. Infectious complications were judged using CDC criteria and were considered significant if antibiotic therapy was instituted." (NB: figures provided here do not include wound infection).	2 (infectious)	0	‐	‐	‐	‐
Falewee 2014*	104	105	Infectious complications: "systemic infection requiring antibiotic treatment (septicaemia, bacteraemia), surgical site infection (according to CDC Definitions of Nosocomial Surgical Site infections), documented nosocomial pneumopathy, up to the 30th post‐operative day." Surgical site infections (SSI): "primary surgical site, neck wound, free‐flap or split skin graft donor site, and tracheotomy"	‐	‐	‐	‐	‐	‐
Hanai 2018	14	13	Clinical complications presented as numbers of patients with "wound complications" classified according to Clavien‐Dindo system	7	4	‐	‐	‐	‐
Postoperative feeding only
Riso 2000	21	23	"Post‐operative complications were recorded as none, minor (urinary tract infection; respiratory tract infection: abnormal chest X‐ray), and major (fistula; wound infection; spontaneous or surgical purulent drainage and flap necrosis; anastomotic leakage)." (NB: figures provided here exclude reported fistula).	4 major (1 flap necrosis, 3 purulent drainage) 1 minor (respiratory tract)	2 (purulent drainage)	1 (respiratory tract)	0	0	0
De Luis 2002	24	23	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and local complications such as fistula and/or wound infection, assessing all complications using standard methods and the same investigator." (NB: figures provided here exclude reported fistula).	4	5	‐	‐	‐	‐
De Luis 2003	18	18	Not collected	‐	‐	‐	‐	‐	‐
De Luis 2004	45	45	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and local complications such as fistula and/or wound infection. All complications were assessed with standard methods by the same investigator." (NB: figures provided here exclude reported fistula).	4	2	‐	‐	‐	‐
De Luis 2005	15	14	Not collected	‐	‐	‐	‐	‐	‐
De Luis 2007	37	35	"Postoperative complications were recorded as none; general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen) and wound complications, such as fistula and/or wound infection, assessed all complications with standard methods by the same investigator surgeon." (NB: figures provided here exclude reported fistula).	2	2	‐	‐	‐	‐
Casas‐Rodera 2008	15	14	"Postoperative complications were recorded as none, general infections (respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture and/or urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen), and wound complications such as fistula and/or wound infection. All complications were assessed with standard methods by the same investigator." (NB: figures provided here exclude reported fistula and wound infection).	1	0	1	0	0	0
De Luis 2009	34	38	"Postoperative complications were registered as none, general infections (urinary tract infection was diagnosed if the urine culture showed at least 10⁵ colonies of a pathogen and/or respiratory tract infection was diagnosed when the chest radiographic examination showed new or progressive unfiltration, temperature above 38.5°C and isolation of pathogens from the sputum or blood culture) and local complications such as fistula and/or wound infection, assessed all complications with standard methods by the same investigator." (NB: figures provided here exclude reported fistula and wound infection).	8	9	‐	‐	‐	‐
De Luis 2014	40	42	No clinical complications recorded	‐	‐	‐	‐	‐	‐
Azman 2015	22	22	No clinical complications recorded	‐	‐	‐	‐	‐	‐
* Additional data supplied by authors for intention‐to‐treat (ITT) analysis to be conducted; as such, total n differs from that in the published manuscript. ASEPSIS: Additional treatment, Serous discharge, Erythema, Purulent exudate, Separation of deep tissues, Isolation of bacteria, Stay duration as inpatient CDC: Centers for Disease Control and Prevention PEG: percutaneous endoscopic gastrostomy UTI: urinary tract infection

Table 4. Other complications

Comparison 1. Immunonutrition versus standard care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Postoperative length of hospital stay Show forest plot	10	757	Mean Difference (IV, Random, 95% CI)	‐2.50 [‐5.11, 0.12]

1.1 Pre‐operative and postoperative feeding	4	403	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐2.82, 2.21]
1.2 Postoperative feeding only	6	354	Mean Difference (IV, Random, 95% CI)	‐4.47 [‐8.46, ‐0.48]
2 Wound infection Show forest plot	12	812	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.70, 1.26]

2.1 Pre‐operative and postoperative feeding	6	458	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.72, 1.33]
2.2 Postoperative feeding only	6	354	Risk Ratio (M‐H, Random, 95% CI)	0.55 [0.19, 1.59]
3 Fistula formation Show forest plot	10	747	Risk Ratio (M‐H, Random, 95% CI)	0.48 [0.27, 0.85]

3.1 Pre‐operative and postoperative feeding	4	393	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.33, 1.62]
3.2 Postoperative feeding only	6	354	Risk Ratio (M‐H, Random, 95% CI)	0.31 [0.14, 0.71]
4 Adverse events Show forest plot	9	719	Risk Ratio (M‐H, Random, 95% CI)	1.33 [0.86, 2.06]

4.1 Pre‐operative and postoperative feeding	2	325	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.30, 2.50]
4.2 Postoperative feeding only	7	394	Risk Ratio (M‐H, Random, 95% CI)	1.42 [0.82, 2.46]
5 All‐cause mortality Show forest plot	14	776	Risk Ratio (M‐H, Random, 95% CI)	1.33 [0.48, 3.66]

5.1 Pre‐operative and postoperative feeding	6	357	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.28, 4.60]
5.2 Postoperative feeding only	8	419	Risk Ratio (M‐H, Random, 95% CI)	1.57 [0.29, 8.53]

Comparison 1. Immunonutrition versus standard care