Types of studies

We included randomised controlled trials (RCTs) reported as full‐text, those published as abstract only, and those with unpublished data.

Types of participants

We included adults, or children, or both, undergoing major upper gastrointestinal, liver and pancreatic surgery such as transhiatal oesophagectomy, gastrectomy (irrespective of whether a total or subtotal distal gastrectomy is performed), liver resection (irrespective of the number of segments resected and the aetiology), pancreatic resection (irrespective of whether a pancreaticoduodenectomy or a distal pancreatectomy is performed, and irrespective of whether the pylorus is preserved), pancreatic drainage procedures (for chronic pancreatitis) and open pancreaticojejunostomy or pancreaticogastrostomy for pseudocyst.

We excluded upper gastrointestinal, liver and pancreatic surgeries that are included in the British Association of Day Surgery Directory of Procedures (BADS 2012), including laparoscopic cholecystectomy or laparoscopic fundoplication, as patients are discharged on the same day when they have mobilised adequately.

Types of interventions

We included trials comparing an enhanced recovery protocol with usual care, provided that the only difference between randomly assigned groups is the use of an enhanced recovery protocol. We will accept the definition proposed by Kehlet 1997, which requires inclusion of one or more of the following elements.

1. Preoperative information and teaching.

2. Decreased stress related to surgery.

3. Pain relief.

4. Exercise (early mobilisation).

5. Enteral nutrition.

6. Growth factors.

We excluded trials comparing different enhanced recovery protocols. We also excluded trials comparing laparoscopic and open surgeries, as the issues surrounding laparoscopic or open surgery are different for different procedures.

Types of outcome measures

Electronic searches

We conducted a literature search to identify all published and unpublished RCTs until 26th March 2015. This literature search identified potential studies published in all languages. We translated the non‐English language papers and fully assessed them for potential inclusion in the review as necessary.

We searched the following electronic databases to identify potential studies.

1. Cochrane Central Register of Controlled Trials (CENTRAL; Cochrane Library; 2015, Issue 3) (Appendix 1).

2. MEDLINE (OvidSP) (1966 to March 2015) (Appendix 2).

3. EMBASE (OvidSP) (1988 to March 2015) (Appendix 3).

4. Science Citation Index (Web of Knowledge) (1982 to March 2015) (Appendix 4).

We also conducted a search of ClinicalTrials.gov (Appendix 5) and the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (Appendix 6) on 26th March 2015.

Searching other resources

We checked reference lists of all primary studies and review articles for additional references. We contacted authors of identified trials and asked them to identify other published and unpublished studies.

We searched for errata or retractions from eligible trials on http://www.ncbi.nlm.nih.gov/pubmed on 26th March 2015, but did not find any errata.

Selection of studies

Two review authors (GBS and AB) independently screened titles and abstracts for inclusion of all potential studies identified as a result of the search and coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve.'

We retrieved full‐text study reports, and the two review authors (GBS and AB) independently screened them, identified studies for inclusion and identified and record reasons for exclusion of ineligible studies.

We resolved disagreements through discussion and, when required, consulted a third person (KG). We identified and excluded duplicates and collated multiple reports of the same study, so that each study rather than each report is the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) flow diagram (Moher 2009; Figure 1), and a Characteristics of excluded studies table.

Figure 1

---

---

Study flow diagram.

Data extraction and management

We used a standard data collection form for study characteristics and outcome data that was piloted on two studies in the review. The two review authors (GBS and AB) extracted the following study characteristics from included studies.

1. Methods: study design, total duration of study and run‐in, number of study centres and locations, study settings, withdrawals, dates of study.

2. Participants: number (N), mean age, age range, gender, inclusion criteria, exclusion criteria.

3. Interventions: interventions, comparisons, concomitant interventions.

4. Outcomes: primary and secondary outcomes specified and collected, time points reported.

5. Notes: funding for trial, notable conflicts of interest of trial authors.

Two review authors (GBS and AB) independently extracted outcome data from the included studies. If outcomes were reported multiple times for the same time point, for example, if short‐term health‐related quality of life was reported at six weeks and at three months, we planned to choose the later time point (i.e. three months) for data extraction. For time‐to‐event outcomes, we planned to extract data to calculate the natural logarithm of the hazard ratio and its standard error using the methods suggested by Parmar 1998.

We planned to include all randomly assigned participants for the medium‐ and long‐term outcomes (e.g. mortality, quality of life), and this was not conditional upon short‐term outcomes (e.g. being alive at three months, having a low or high quality of life index at three months).

We noted in the Characteristics of included studies table whether outcome data were reported in an unusable way. We resolved disagreements by reaching consensus or by involving a third person (KG). One review author (KG) copied data from the data collection form into the Review Manager file (RevMan 2014). We double‐checked that data were entered correctly by comparing data in the study reports with data presented in the systematic review.

Assessment of risk of bias in included studies

Two review authors (GBS and AB) independently assessed risk of bias for each study, using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Disagreements were resolved by discussion or through involvement of a third assessor (KG). We assessed risk of bias according to the following domains.

1. Random sequence generation.

2. Allocation concealment.

3. Blinding of participants and personnel.

4. Blinding of outcome assessment.

5. Incomplete outcome data.

6. Selective outcome reporting.

7. Other bias.

We graded each potential source of bias as high, low or unclear risk and provided a quote from the study report together with a justification for our judgement in the 'Risk of bias' table. We summarised risk of bias judgements across different studies for each of the domains listed. We considered blinding separately for different key outcomes when necessary (e.g. for unblinded outcome assessment, risk of bias for all‐cause mortality may be very different than for a patient‐reported pain scale). When information on risk of bias was related to unpublished data or correspondence with a trialist, we noted this in the 'Risk of bias' table.

When considering treatment effects, we took into account the risk of bias for studies that contribute to those outcomes.

Measures of treatment effect

We analysed dichotomous data as risk ratios (RRs) and continuous data as mean differences (MDs) when the outcome was reported or converted to the same units in all trials (e.g. hospital stay) or as standardised mean differences (SMDs) when different scales were used in measuring the outcome (e.g. quality of life). We ensured that higher scores for continuous outcomes had the same meaning for the particular outcome, explained the direction to the reader and reported when the directions were reversed.

We calculated rate ratios for outcomes such as adverse events and serious adverse events when it was possible for the same person to experience more than one adverse event (or serious adverse event). If the study authors had calculated the rate ratio of adverse events (or serious adverse events) for intervention versus control on the basis of Poisson regression, we planned to obtain the rate ratio by using the Poisson regression method in preference to the rate ratio calculated using the number of adverse events (or serious adverse events) reported during a specified period. We planned to calculate the hazard ratio for time‐to‐event outcomes such as long‐term mortality and long‐term recurrence.

We undertook meta‐analyses since all the surgeries were major hepato pancreato biliary surgeries. Trialists commonly indicate when they have skewed data by reporting medians and interquartile ranges. It was not possible to determine whether the data were skewed. We attempted to contact the trial authors to provide this information, but we were unable to obtain this information. So, if the median and interquartile range were reported, we imputed the mean and standard deviation (as mentioned in the Dealing with missing data), but we performed a sensitivity analysis, excluding the data from these trials (as mentioned in the Sensitivity analysis).

When multiple trial arms were reported in a single trial, we planned to include only the relevant arms. When two comparisons (e.g. enhanced recovery protocol A versus standard care and enhanced recovery protocol B versus standard care) had to be entered into the same meta‐analysis, we planned to pool the results of enhanced recovery protocol A and enhanced recovery protocol B and compare them with standard care. The alternative way of including such trials is to half the control group and compare it with enhanced recovery protocol A and with enhanced recovery protocol B to avoid double counting. We planned to perform a sensitivity analysis to determine whether results obtained using the two methods of dealing with multi‐arm trials led to different conclusions.

Unit of analysis issues

The unit of analysis was individual study participants undergoing major upper gastrointestinal, liver and pancreatic surgery. If cluster‐randomised trials were identified, we planned to obtain the effect estimate adjusted for the clustering effect. If this was not available, we planned to perform a sensitivity analysis by excluding the trial from the meta‐analysis, as the variance of the effect estimate unadjusted for cluster effect is less than the actual variance, which is adjusted for cluster effect, inappropriately giving more weight to the cluster‐RCT in the meta‐analysis.

Dealing with missing data

We attempted to contact the investigators or study sponsors to verify key study characteristics and to obtain missing numerical outcome data whenever possible (e.g. when a study is identified as an abstract only). We received additional data from two study authors (Jones 2013; Kim 2012). If we were unable to obtain information from investigators or study sponsors, we imputed mean from median (i.e. consider median as the mean) and standard deviation from standard error, interquartile range or P values according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), but we assessed the impact of including such studies as indicated in a sensitivity analysis.

If we were unable to calculate the standard deviation from the standard error, interquartile range or P values, we imputed the standard deviation as the highest standard deviation in the remaining trials included in the outcome, while remaining fully aware that this method of imputation will decrease the weight of the studies in the meta‐analysis of MDs and will shift the effect towards no effect for SMDs.

Assessment of heterogeneity

We used the I² statistic to measure heterogeneity among the trials in each analysis. If we identified substantial heterogeneity as per the Cochrane Handbook for Systematic Reviews of Interventions (greater than 50% to 60%), we planned to explore this by performing prespecified subgroup analysis (Higgins 2011). 

Assessment of reporting biases

We attempted to contact study authors to ask them to provide missing outcome data. When this was not possible, and the missing data were thought to introduce serious bias, we explored the impact of including such studies in the overall assessment of results by a sensitivity analysis. 

Since there were fewer than 10 trials for all the outcomes, we did not create and examine a funnel plot to explore possible publication biases. We planned to use Egger's test to determine the statistical significance of the reporting bias (Egger 1997). We planned to use a P value less than 0.05 to show statistically significant reporting bias.

Data synthesis

We performed analyses using Review Manager 5.3 (RevMan 2014). We used the Mantel Haenszel method for dichotomous data, inverse variance method for continuous data and generic inverse variance for count data. We planned to use generic inverse variance for time‐to‐event data. We used both the fixed‐effect model (DerSimonian 1986), and random‐effects model for the analysis (Demets 1987). In case of discrepancy between the two models, we reported both results; otherwise we reported only the results from the fixed‐effect model.

Subgroup analysis and investigation of heterogeneity

We planned to carry out the following subgroup analyses.

1. Different surgeries (e.g. total or subtotal gastrectomy, distal gastrectomy, pancreaticoduodenectomy, distal pancreatectomy).

2. Different enhanced recovery protocols (according to the element that the enhanced recovery protocol is meant to address, for example, pain relief, nutrition).

3. Adults versus children.

We planned to use the primary outcomes in the subgroup analysis. We used the formal Chi² test for subgroup differences to test for subgroup interactions.

Sensitivity analysis

We performed sensitivity analysis, as defined a priori, to assess the robustness of our conclusions. This involved:

1. excluding trials at unclear or high risk of bias (one of more of the risk of bias domains is classified as unclear or high);

2. excluding trials for which mean or standard deviation or both were imputed; and

3. excluding cluster‐RCTs for which adjusted effect estimates are not reported.

4. different methods of dealing with multi‐arm trials (please see Measures of treatment effect).