Types of studies

We only included randomised controlled trials (RCTs).

Types of participants

We included all adults (18 years or older) with histologically‐confirmed early‐stage NSCLC (stages I to III) after surgical resection (with or without chemotherapy), and those with unresectable locally advanced stage III NSCLC who had received radiotherapy, or radiotherapy combined with chemotherapy with curative intent.

Types of interventions

• Surgical treatment + immunotherapy agents versus surgical treatment with placebo, best supportive care, or no intervention.

• Radical radiotherapy (with or without chemotherapy) + immunotherapy agents versus radical radiotherapy (with or without chemotherapy) with placebo, best supportive care, or no intervention.

Types of outcome measures

Electronic searches

We conducted a literature search to identify all published or unpublished RCTs. The literature search identified potential trials in all languages.

We searched the following electronic databases for potential trials.

• Cochrane Central Register of Controlled Trials (CENTRAL; 2021, Issue 5) in the Cochrane Library, searched 19 May 2021 (Appendix 1).

• MEDLINE PubMed (1966 to 19 May 2021; Appendix 2).

• Embase (1988 to 19 May 2021; Appendix 3).

• CINAHL (1982 to 19 May 2021).

The search string for MEDLINE was developed according to the Cochrane Highly Sensitive Search Strategy, sensitivity‐maximising version, as referenced in Chapter 6.4.11.1 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021). We adapted the terms and the search strategies for CENTRAL, MEDLINE, and Embase to search CINAHL. 

Searching other resources

We searched Clinical Trial Registers (www.clinicaltrials.gov; www.controlled-trials.com), databases of the Food and Drug Administration (FDA), the European Medicines Agency (EMA), and the World Health Organization (WHO) International Clinical Trials Registry Platform (who.int/clinical-trials-registry), to identify information about ongoing trials. We searched all databases from their inception to 19 May 2021. 

We checked reference lists of all included trials and related reviews for additional references. We asked experts in the field and manufacturers of relevant drugs to provide details of outstanding clinical trials and any relevant unpublished material. We also contacted authors of identified trials and asked them to identify other published and unpublished trials.

We manually checked for potential trials in abstracts or reports from the following relevant conference proceedings (from 1990 to present): American Society of Clinical Oncology (ASCO), European Society of Medical Oncology (ESMO), European Cancer Conference Organisation (ECCO), and International Association for the Study of Lung Cancer (IASLC) World Lung Cancer Conference.

We searched for errata or retractions from eligible trials on www.ncbi.nlm.nih.gov/pubmed on 19 May 2021. 

Selection of studies

Two review authors (YY and XW for this updated review, or OT and ET for the original review) independently screened titles and abstracts of all trials that we identified as a result of the search, and labelled them as 'retrieve' (eligible, potentially eligible, or unclear) or 'do not retrieve'. For the ones coded as 'retrieve', we then referred to their full‐text study reports or publication. Two review authors (HS, RL) independently screened the full‐text reports. The procedure of study identification for inclusion and exclusion was well documented by using standard screening forms. We resolved any disagreement through discussion, or consulted a third review author (CS).

We identified and excluded duplicates and collated multiple reports of the same study, so that each study, rather than each report, was the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (Liberati 2009), and 'Characteristics of excluded studies' table.

Data extraction and management

We used a data collection form for study characteristics and outcome data, which we tested on one study in the review. One review author (RL) extracted study characteristics from included trials. We extracted the following study characteristics.

• Methods: study design (for example, parallel or cross‐over design), number of study centres and location (country), total duration (for example, date of study, follow‐up period, early stopping of trial), method of randomisation (including imbalanced randomisation ratio), methods of allocation concealment, blinding

• Participants: N, age, gender, Eastern Co‐operative Oncology Group (ECOG) performance status, medical history, the severity of condition (stage), diagnostic criteria, inclusion criteria, exclusion criteria

• Interventions: intervention, comparison, concomitant medications, excluded medications

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

• Notes: funding for trial, or any notable conflicts of interest of trial authors

Two review authors (YY and XW for this updated review, HS and RL for the original review) independently extracted outcome data from the included trials. We noted in the 'Characteristics of included studies' table if outcome data were reported in an unusable way. We resolved disagreements by consensus, or by involving a third review author (CS). One review author (RL) copied the data from the data collection form into the Review Manager file (Review Manager 2020). We double‐checked that the data were entered correctly by comparing the study reports with the data in the systematic review. A second review author (JZ) spot‐checked study characteristics for accuracy against the trial report.

Assessment of risk of bias in included studies

Two review authors (YY and XW for this updated review, HS and RL for the original review) independently assessed the risk of bias for each study, using RoB 1 (Higgins 2011). Any disagreements were resolved by discussion, or by involving a third assessor (JZ). We assessed the risk of bias according to the following domains.

• Random sequence generation

• Allocation concealment

• Blinding of participants and personnel

• Blinding of outcome assessment

• Incomplete outcome data

• Selective outcome reporting

• Other potential bias

We graded each potential source of bias as high, low, or unclear, and provided a quote from the study report together with a justification for our judgement in the risk of bias table. We summarised the risk of bias judgements across different trials for each of the domains listed. For overall risk of bias, we considered trials that had adequate random sequence generation, adequate allocation concealment, adequate blinding, adequate handling of incomplete outcome data, no selective outcome reporting, and were without other bias risks, as being at overall low risk of bias. We considered trials that were assessed as being at high or unclear risk of bias in the majority of domains as being at overall high risk of bias; and the remaining trials to be at low risk of bias. We considered blinding separately for different key outcomes where necessary e.g. for unblinded outcome assessment, the risk of bias for all‐cause mortality may be very different than for a participant‐reported pain scale. Where information on the risk of bias 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 the trials that contributed to that outcome.

Measures of treatment effect

We analysed the primary outcomes based on intention‐to‐treat (ITT) analyses, where available. We measured effect estimates by hazard ratios (HRs) for time‐to‐event variables, and risk ratios (RRs) for dichotomous variables. For continuous variables, we calculated mean differences (MDs) with their corresponding 95% confidence intervals (CIs) if trials used the same measurement, and standardised mean differences (SMDs) and 95% CIs when trials use different scales. We contacted the corresponding authors for missing information about standard deviations or standard errors. For reports without available data for pooling, we tried to measure rates from figures, and calculated effect estimates from P values, t statistics, ANOVA tables, or other statistics as appropriate.

Unit of analysis issues

We did not find any trials with non‐standard design (such as cluster‐randomised trials for potential 'unit of analysis error') for this updated review. But if such eligible trials emerge in future literature searches, we will carefully assess these trials (in terms of recruitment bias, baseline imbalance, loss of clusters, and comparability with individually‐randomised trials). Furthermore, we will apply proper statistical methods (such as multilevel models and generalised estimating equations) for analysis, according to the Handbook for Systematic Review of Interventions (Higgins 2021). The individual participant was the unit of analysis for this updated review.

If there had been data from a cross‐over trial with eligible intervention performances, we would have used the data from the first phase only, i.e. from randomisation to the point of cross‐over.

Had multiple trial arms been reported in a single trial, we would only have included the relevant arms. In future, if we enter two comparisons (e.g. drug A versus placebo and drug B versus placebo) from the same trial into the same meta‐analysis, we will halve the control group to avoid double counting.
 

Dealing with missing data

We contacted investigators or study sponsors to verify key study characteristics and obtain missing outcome data where applicable (e.g. when a study was identified as an abstract only). For full‐text reports with missing details relevant to our analysis, we also contacted the authors of the trials by email. In the case of non‐response after repeated attempts, we dropped these incomplete data from the analysis, stating this clearly in the Results section, and discussed it further under the Potential biases in the review process section of the Discussion.

Assessment of heterogeneity

We carried out tests for heterogeneity using the Chi² test, to assesses whether observed differences in results were compatible with chance alone. We used the I² statistic to quantify inconsistency across trials. The presence of heterogeneity was defined by P < 0.05 from the Chi² test, and I² > 50% (Higgins 2021). If we detected moderate or higher heterogeneity (50% to 100%), we applied a thorough exploration of possible sources of heterogeneity by means of subgroup and sensitivity analyses (as stated below). Given the limitations of the methods, the P value from the Chi² test and the value of I² were only referred to as a guide, and we exercised caution when interpreting the results.

Assessment of reporting biases

We attempted to contact study authors, asking 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 trials in the overall assessment of results by conducting a sensitivity analysis.

We did not include sufficient trials for each outcome in the current updated review to create a funnel plot. But if we are able to pool more than 10 trials in future updates, we will do so to explore possible publication biases (intervention effect estimate versus standard error of intervention effect estimate). If we find funnel plot asymmetry, we will further investigate clinical diversity of trials as a possible explanation. If there are sufficient trials (> 10), we also will use the 'contour‐enhanced' funnel plot to differentiate asymmetry due to other factors (Peters 2008). If the supposed missing trials are in areas of higher statistical significance, the cause of the asymmetry is highly suggestive of being due to factors other than publication bias.

Data synthesis

We used Review Manager 5 for pooling data and for statistical analysis (Review Manager 2020). We used random‐effects models for primary analyses since the agents of interest had different mechanisms of action. In the future (since currently no subgroup analysis was successfully conducted), provided that the trials in some subgroups are found to be homogeneous (in terms of age, diagnostic subtype, intervention type, intervention duration), we will use both fixed‐effect and random‐effects models, and compare the results. In the absence of heterogeneity and significant reporting bias, these two models should yield the same results. In this case, we will report the results from the fixed‐effect model only. If the results are different, indicating significant heterogeneity, we will report the results from the random‐effects model only.

Subgroup analysis and investigation of heterogeneity

We had planned to perform the following exploratory subgroup analyses on the primary outcomes, using Review Manager 5 (Review Manager 2020).

• Participants receiving immunotherapy who present with a different stage of NSCLC (stages I, II, or III).

• Participants receiving different types of immunotherapy.

• Participants with a specific biomarker: e.g. participants with a gene‐signature profile (MAGE‐A3‐positive).

In the current updated review, we did not include sufficient trials for each population (subgroup) of interest to conduct subgroup analyses. In future updates, we will perform a subgroup analysis for primary outcomes when we have at least three trials for a subgroup.

Considering the differences between subgroups, we will first examine them by visual inspection of their CIs; non‐overlapping CIs indicate a statistically significant difference in treatment effect between subgroups. Also, we will use the approach of Borenstein 2008 to formally investigate differences between two or more subgroups.

Sensitivity analysis

We performed sensitivity analyses, defined a priori, to assess the robustness of our conclusions. This was achieved by repeating the analyses to explore the influence of the following factors on effect size.

• Exclusion of unpublished trials (in the current updated review, we did not include unpublished data, but will conduct this analysis if unpublished data are included in future updates).

• Exclusion of lower quality trials (those at high or unclear risk of bias).

Summary of findings and assessment of the quality of the evidence

We created a summary of findings table presenting all our primary and secondary outcomes, except overall response and health‐related quality of life, using GRADEpro software (GRADEpro GDT). We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of the body of evidence as it related to the trials that contributed data to the meta‐analyses for the prespecified outcomes. We justified decisions to downgrade or upgrade the quality of the evidence using footnotes, and made comments to aid the reader's understanding of the review where necessary (Higgins 2021).

Summary of findings and assessment of the certainty of the evidence

We created a 'Summary of findings' table presenting all our primary and secondary outcomes except overall response and health‐related quality of life, using the GRADEpro software, version 3.2 (GRADEpro GDT). We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of the body of evidence as it related to the studies that contributed data to the meta‐analyses for the pre‐specified outcomes. We justified decisions to downgrade or upgrade the quality of the evidence using footnotes, and made comments to aid the reader's understanding of the review where necessary.