Types of studies

All randomised controlled trials (RCTs) and cluster‐RCTs looking at pre‐ or intra‐operative interventions to reduce the need of opioids after breast reconstruction. The interventions may include utilization of specific anesthetics, fast‐track protocols, or enhanced recovery pathway protocols. In the absence of RCTs, we will include well‐designed cohort or case‐control studies.

Types of participants

Women with breast cancer, aged ≥ 18 years, who are undergoing unilateral or bilateral therapeutic mastectomy and immediate or delayed breast reconstruction (of any technique). We will also include women at high‐risk of breast cancer (e.g. BRCA1 or BRCA2 mutations, personal history of ductal or lobular carcinoma in situ, history of chest radiation, Li‐Fraumeni syndrome, Cowden/PTEN syndromes) undergoing risk‐reducing mastectomy and immediate or delayed breast reconstruction (of any technique).

We will exclude studies that report on other participants and do not report results separately on those participants relevant to this review.

Types of interventions

The peri‐operative phase is defined as immediately pre‐operative and a few days postoperative.

Types of outcome measures

Electronic searches

We will search the following databases.

• The Cochrane Breast Cancer Specialised Register. Details of the search strategies used by the Cochrane Breast Cancer Group for the identification of studies and the procedure used to code references are outlined in the Group's module (www.mrw.interscience.wiley.com/cochrane/clabout/articles/BREASTCA/frame.html). Trials with the key words "opioids", "oxycodone", "morphine", "breast reconstruction", "mammoplasty" will be extracted and considered for inclusion in the review.

• CENTRAL (the Cochrane Library, latest issue) (see Appendix 1).

• MEDLINE (via OvidSP) from 1946 to present (see Appendix 2).

• EMBASE (via OvidSP) from 1947 to present (see Appendix 3).

• World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) search portal (http://apps.who.int/trialsearch/Default.aspx) for all prospectively registered and ongoing trials (see Appendix 4).

• ClinicalTrials.gov (http://clinicaltrials.gov/) (see Appendix 5).

Searching other resources

• Bibliographic searching.

We will try to identify further studies from reference lists of identified relevant trials or reviews. We will obtain a copy of the full article for each reference reporting a potentially eligible trial. Where this is not possible, we will try to contact study authors to obtain additional information.

Selection of studies

Following removal of duplicate references, two review authors (CS, MAC) will independently screen all titles and abstracts and discuss the results of the first step of screening. A third review author (GDR) will resolve any disagreements regarding eligibility. Two review authors (CS, MAC) will screen the selected references based on the full‐text publication, and we will follow a similar process to resolve any disagreements. The result of the selection process will be recorded in the PRISMA flow diagram. In addition, we will record the reasons for excluding key full‐text articles in the ‘Characteristics of excluded studies’ table. We will apply no language restrictions and, if necessary, we will translate non‐English papers.

Data extraction and management

Two review authors (CS, MAC) will develop a standard data extraction form and independently extract data from the included studies. A third review author (GDR) will resolve any disagreements regarding data collection and extraction. We will collect information on study design, participants, setting, interventions, outcomes, follow‐up, methods employed to control for confounders and biases, sources of funding, and notable conflicts of interest of trial authors, amongst other items. For observational studies where we extract adjusted estimates, we will select the most fully adjusted model. Should we identify multiple papers that report results from the same study during the data extraction process, we will select a single paper as the primary reference based on the completeness of reporting and sample size if overlapping population exists.

Assessment of risk of bias in included studies

For RCTs, two review authors (CS, MAC) will assess risk of bias using Cochrane's ‘Risk of bias' tool described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This will include an evaluation of the following items that are reported in a ‘Risk of bias’ table:

• Random sequence generation (to assess possible selection bias).

• Allocation concealment (to assess possible selection bias).

• Blinding of participants and personnel (to assess possible performance bias).

• Blinding of outcome assessment (to assess possible detection bias).

• Incomplete outcome data (to assess possible attrition bias).

• Selective outcome reporting (to assess possible reporting bias).

• Other bias.

For non‐randomised studies, we will assess risk of bias using the Risk of Bias in Non‐randomized studies of Interventions (ROBINS‐I) tool (Sterne 2016). This includes an evaluation of multiple items at the pre‐intervention period, intervention period, and postintervention period. In particular, it will cover participant baseline demographics and clinical characteristics (to assess for possible selection bias), heterogeneity of surgical interventions, adherence to interventional protocols, missing data, reported results, and adjustment for potential confounding variables (to assess for bias due to confounding).

Examples of possible confounding factors include the following.

• Demographic characteristics, such as younger age (age < 50 years, Katz 2005), higher body mass index (Spivey 2018), unmarried status (Katz 2005).

• Presence of pre‐operative pain (Wang 2016) and pre‐operative use of opioids.

• Comorbidities that alter the metabolism of pain medications, such as liver or kidney disease.

• Pre‐existing psychiatric conditions, such as anxiety (Katz 2005) or depression (Spivey 2018).

• Types of surgical interventions, such as sentinel lymph node biopsy versus axillary dissection (Spivey 2018) where intraoperative nerve division may be related with reduced pain (Bruce 2012), type of breast reconstruction performed, type of implant used (Siotos 2019b), and volume of implant or tissue expander inserted.

• Variations in opioid allowance and non‐opioid medications use in Enhanced Recovery Protocols (Stone 2019).

• Receipt of neoadjuvant or adjuvant therapy, such as chemotherapy (Bruce 2014), or radiotherapy (Poleshuck 2006).

We will classify items as either ‘low', ‘moderate', ‘serious', or ‘critical' risk of bias or ‘no information', and will record the reasons for each classification. A third author (GDR) will be involved in the process to resolve any disagreements.

Measures of treatment effect

For dichotomous outcomes (e.g. opioid use, presence of pain, postoperative nausea and vomiting, anti‐emetic use, number of complications), we will measure the effect using risk ratios (RRs) and 95% confidence intervals (CIs). We will report the ratios of treatment effects for response so that RRs less than 1.0 favor the intervention.

For continuous outcomes collected using different scales (e.g. opioid use, intensity of pain, severity of postoperative nausea and vomiting), we will measure the effect using standardized mean differences (SMD) with 95% CIs. For continuous outcomes collected using the same scale (e.g. operation time, length of hospital stay, cost of care), we will measure the effect using mean differences (MD) with 95% CIs. We will report the differences of treatment effects for response so that mean difference values of less than 0.0 favor the intervention.

Unit of analysis issues

We will conduct the analysis per participant. If we identify both cluster and individually‐based RCTs, we will adjust the cluster's sample size using appropriate methods, as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will also consider whether the matching technique used in the included studies may impact the analysis. In non‐randomised studies, matching refers to selecting study participants in a way that the distribution of specific confounders are more similar among the comparative groups (Costanza 1995).

In those studies contributing more than one intervention group, we will combine the intervention groups to create a single pair‐wise comparison in order to overcome the unit‐of‐analysis error.

Dealing with missing data

Where missing data occurs (e.g. effect estimates are reported without their corresponding standard error), one review author will contact the corresponding author or co‐authors of the study and request further information. If we do not receive a response within four weeks after initial email contact, we will send a follow‐up email.

We will perform the analysis on an intention‐to‐treat basis (i.e. all participants are analysed according to the groups to which they were randomised). We will perform a sensitivity analysis to assess the impact of missing data on our results. We will stratify the included studies into the three categories of low missing data (≤ 10% of the original sample size), medium missing data (> 10% and < 50%), and high missing data (≥ 50%), and will compare the results in each category.

Assessment of heterogeneity

We will assess clinical and methodological heterogeneity among included trials by reviewing patient characteristics, types of interventions, and types of breast reconstruction.

Statistical heterogeneity will be assessed using the Chi² test, the Tau² and the I² statistic (Higgins 2003). We will consider P values < 0.05 (generated by the Chi² test) to indicate statistically significant heterogeneity, and I² values as follows.

• < 40%: heterogeneity might not be important.

• 30 to 60%: moderate heterogeneity.

• 50 to 75%: substantial heterogeneity.

• > 75%: considerable heterogeneity.

We will perform subgroup analysis to investigate substantial heterogeneity if appropriate. However, if there is considerable heterogeneity, meta‐analyses may not be appropriate.

Assessment of reporting biases

If we have 10 or more studies in our meta‐analysis of the primary outcome, we will assess reporting bias by generating funnel plots (Higgins 2011) and evaluating the results of the Egger's test (Egger 1997).

Data synthesis

We will perform statistical synthesis of data using Review Manager 5 (RevMan 5) software (Review Manager 2014). Dichotomous outcomes will be synthesized using the Dersimonian and Laird model, and continuous outcomes will be synthesized using the inverse‐variance method. If data are available to perform meta‐regression, we will undertake this in Stata (Stata 2015).

We will perform all analyses using the random‐effects model. If different estimates of effect are available (e.g. odds ratios and RRs), we will perform the appropriate conversions before pooling the data, according to the guidelines provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

Subgroup analysis and investigation of heterogeneity

If there are 10 or more studies included in our meta‐analysis and we observe high heterogeneity, we will fit a meta‐regression model using the study characteristics provided below to investigate heterogeneity.

• Type of intervention (i.e. technical, pharmacological, and postoperative management).

• Study design if trial studies are available (i.e. experimental versus observational studies).

• Type of breast reconstruction (i.e. implant‐based versus autologous‐based versus other methods of reconstruction).

• Timing of breast reconstruction (i.e. immediate versus staged versus delayed).

• Type of mastectomy (i.e. risk‐reducing mastectomy versus therapeutic mastectomy).

We will compare the heterogeneity (i.e. I² and Tau²) from each of these meta‐regression models to a model without any covariates. This will give an indication of the degree of variation between the studies explained by the covariate.

Sensitivity analysis

If an adequate number of studies are available, we plan to perform a sensitivity analysis by excluding studies with high risk of bias. Also, we will perform a sensitivity analysis based on the presence of missing data in the included studies. We will stratify the included studies into three categories of low missing data (≤ 10% of the original sample size), medium missing data (> 10% and < 50%), and high missing data (≥ 50%), and will compare the results in each category.

For each sensitivity analysis, we will report the effect estimate and associated CI, and heterogeneity for each outcome, and compare the results to those from the meta‐analyses prior to each sensitivity analysis.