Types of studies

We will include phase II and III randomised controlled studies, comparing ICIs alone or in combination with other forms of active control intervention(s) versus placebo or other active control interventions in any line of treatment for people with advanced pancreatic cancer, with or without blinding. There will be no language or publication status restrictions, and we will include meeting abstracts and unpublished online data.

We will not exclude studies based on the availability of relevant outcome data. We will exclude quasi‐randomised studies, defined as studies where the method of treatment allocation is not strictly random (e.g. alternate allocation).

Types of participants

We will include studies involving participants with locally advanced, unresectable, or metastatic pancreatic ductal adenocarcinoma.

Diagnosis of pancreatic adenocarcinoma will need to be established by either histological or cytological findings. There will be no restrictions for age or gender. We will incorporate data from studies that include a subset of eligible participants, provided the characteristics and outcomes of those participants can be extracted separately.

Types of interventions

The experimental intervention will consist of:

• ICIs, defined as drugs that are directed against immune checkpoint proteins (e.g. PD‐1 inhibitors, PD‐L1 inhibitors, CTLA‐4 inhibitors).

The control (comparator) intervention will consist of:

• chemotherapy, which encompasses all cytotoxic drug treatments;

• targeted therapy, defined as drugs directed against specific oncogenic signalling pathways (e.g. EGFR inhibitors, PARP inhibitors, NTRK inhibitors);

• inactive control intervention (e.g. no treatment or best supportive care).

We will apply no restrictions on drug dosage, treatment duration, or route of administration.

We will consider studies for inclusion if the interventions fall into the following comparisons.

• Single‐agent ICI versus inactive control intervention (e.g. no treatment or best supportive care).

• Single‐agent ICI versus other active control agents (e.g. chemotherapy or targeted therapy, or both).

• Doublet ICI (i.e. two separate ICI) versus inactive control intervention (e.g. no treatment or best supportive care).

• Doublet ICI (i.e. two separate ICI) versus other active control agents (e.g. chemotherapy or targeted therapy, or both).

• Chemotherapy or targeted therapy plus ICI versus the same chemotherapy or targeted therapy alone.

• Chemotherapy or targeted therapy plus ICI versus the same chemotherapy or targeted therapy plus other active control agents (e.g. chemotherapy or targeted therapy, or both).

Types of outcome measures

We will analyse the following outcomes in the review, but we will not use them as a basis for including or excluding studies.

Electronic searches

We will search the following electronic databases to identify all published and unpublished randomised controlled trials, with no language restrictions:

• Cochrane Central Register of Controlled Trials (CENTRAL);

• MEDLINE, accessed via PubMed (inception year to present; Appendix 1);

• Embase (inception year to present).

We will search all databases using both controlled vocabulary (namely medical subject headings (MeSH) in MEDLINE and EMTREE in Embase) and a wide range of free‐text terms. We will perform the MEDLINE search using the Cochrane Highly Sensitive Search Strategy, as described in the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2022).

Searching other resources

We will check the reference lists of all relevant studies and review articles to identify further studies that meet the inclusion criteria. We will contact study authors for additional information where necessary. Our Embase search will capture meeting abstracts from related conference proceedings.

We will also conduct searches in the following clinical trials registries to identify unpublished and ongoing trials:

• US National Institute of Health Ongoing Trials Register ClinicalTrials.gov (clinicaltrials.gov; Appendix 2);

• World Health Organization International Clinical Trials Registry Platform (ICTRP) (trialsearch.who.int/; Appendix 3).

Selection of studies

Two review authors (WYC and VC) will independently screen titles and abstracts for inclusion. All potential studies identified as a result of the search will be coded as either 'retrieve' (eligible, potentially eligible, or unclear) or 'do not retrieve'.

The same review authors will retrieve the full texts of potentially eligible studies and independently check the eligibility of each study against review eligibility criteria. We will resolve any disagreements through discussion or, if required, we will involve a third review author (HS). If a review author is also a contributor to a study that may be of interest to the review, that review author will not be involved in the selection, data extraction and management, assessment of risk of bias and assessment of the certainty of evidence processes.

We will identify, exclude duplicates, and collate multiple reports of the same study, so that each study rather than each report is the unit of interest in the review. We will record the selection process in sufficient detail to complete a PRISMA flow diagram and 'Characteristics of excluded studies' table.

Data extraction and management

Two review authors (WYC, VC) will independently extract outcome data from the included studies using a prepiloted data extraction form. To reach consensus, we will involve a third review author (HS) when necessary. When we encounter multiple publications of the same study, we will choose the first publication dealing with the primary endpoint in this review as a study identifier.

We will extract the following details from each included study and complete a 'Characteristics of included studies' table.

• Source: citation, study name if applicable, and contact details.

• Study details: study design, location, setting (type and stage of disease), number of participating centres, sample size, study start date, study completion date, study follow‐up.

• Participants: inclusion and exclusion criteria, number of participants, participant and tumour characteristics (age, sex, ethnicity, performance status, histology), stage (locally advanced pancreatic cancer, unresectable pancreatic cancer, metastatic pancreatic cancer).

• Intervention(s): type of intervention, dosage, route of administration, duration.

• Comparator(s): type of intervention, dosage, route of administration, duration.

• Outcomes: primary and secondary outcomes with definitions, and time points.

• Results: number of participants allocated to each group, and for each outcome of interest, sample size, summary data for each group, estimate of effect with confidence interval (CI) and P value, subgroup analyses, and whether analyses have been performed by intention‐to‐treat (ITT) or per‐protocol methods.

• Miscellaneous: funding source.

Assessment of risk of bias in included studies

Two review authors (WYC and VC) will independently assess the risk of bias for each included study, using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022) and RoB 2 tool (Sterne 2019). In the case that two review authors cannot reach consensus, they will consult with a third review author (HS). We will assess the risk of bias according to the following domains derived from the tool:

• bias arising from the randomisation process;

• bias due to deviations from intended interventions;

• bias due to missing outcome data;

• bias in measurement of the outcome;

• bias in selection of the reported result.

We will rate each domain of the RoB 2 tool as having 'high risk of bias', 'some concerns', or 'low risk of bias' and provide justification for our judgement of each domain with a brief description. We will provide a figure to summarise the risk of bias. Where information on risk of bias relates to unpublished data or correspondence with a study author, we will note this in the risk of bias table. When considering treatment effects, we will take into account the risk of bias for the studies that contribute to that outcome, as part of the GRADE methodology.

We will also present an overall risk of bias judgement for each study by evaluating the risk of bias across all five domains. We will consider the risk of bias for each trial as follows.

• Low risk of bias: the outcome is judged to be at low risk of bias for all domains.

• Some concerns: the outcome is judged to raise some concerns in at least one domain, but is not at high risk of bias for any domain.

• High risk of bias: the outcome is judged to be at high risk of bias in at least one domain, or there are some concerns for multiple domains such that our confidence in the result is substantially lowered.

Measures of treatment effect

For time‐to‐event outcomes (OS, PFS, duration of response, time to response) we will use hazard ratios (HRs) to measure treatment effects. We will report each HR along with the 95% CIs. An HR of one indicates that the hazard rate is equivalent between experimental and control groups, whilst an HR other than one indicates differences in hazard rates between the two groups. We will extract the HR from the included studies when it is available.

For dichotomous outcomes (AEs and ORR), we will use risk ratios (RRs) and 95% CIs if possible. For dichotomous outcomes related to OS and PFS at specific time points, we will use RRs to estimate survival rates and 95% CIs.

For continuous outcomes (HRQoL), we will use mean differences (MDs) between treatment arms when studies use a similar scale to measure outcomes, or standardised mean differences (SMDs) if studies use different scales to measure the same outcome. We will confirm that higher scores for continuous outcomes have the same meaning for the particular outcome, explain the direction, and report if directions were reversed. For HRQoL outcomes, we will present the SMD alongside the corresponding change in EORTC scale to improve interpretability.

We will undertake meta‐analyses only where this is meaningful (i.e. if the treatments, participants, and the underlying clinical question are similar enough for pooling to make sense). Where a trial reports multiple arms, we will include data from only the relevant arms but list all arms in the 'Characteristics of included studies' table.

Unit of analysis issues

For studies that compare more than one treatment arm with a control arm in the same meta‐analysis, we will divide the number of participants in the control group by the number of treatment arms.

For cluster‐randomised trials, we will use analyses that have properly accounted for the cluster design. Alternatively, we will estimate the treatment effect using the intraclass correlation coefficient (ICC), as described in Section 23.1 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022). If the ICC is unavailable, we will use external estimates obtained from similar studies.

For crossover trials, we will only include data from the first period since carryover effects are typically a concern in cancer drug trials. In addition, crossover is often unidirectional from the control arm to the experimental arm at the time of progression.

Dealing with missing data

When we identify missing or unclear data, we will contact the study author directly. If we are unable to obtain the information from the investigators or study sponsors, we will follow Cochrane recommendations when dealing with such issues related to data, as described in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021), and will analyse only the available data. We will address the impact of including such studies indicated in a sensitivity analysis.

Assessment of heterogeneity

We will assess clinical and methodological heterogeneity by inspecting the characteristics of the included studies, including the types of studies, participants, and interventions. We will assess statistical heterogeneity by using I² statistic to measure heterogeneity amongst the trials in each analysis (Higgins 2003). If we identify substantial heterogeneity as per the Cochrane Handbook for Systematic Reviews of Interventions (greater than 50%), we will explore it using prespecified subgroup analysis (Higgins 2022). We will also assess heterogeneity by evaluating whether there is good overlap of CIs.

Assessment of reporting biases

We will investigate publication bias using funnel plots where there are at least 10 studies in each analysis. We will consider P < 0.05 to be a statistically significant reporting bias.

Data synthesis

We will perform the meta‐analysis using Review Manager Web (RevMan Web 2022). We will use random‐effects model by default. For studies that cannot be synthesised using meta‐analysis, we will provide a narrative description of the study characteristics and findings according to the Synthesis Without Meta‐analysis (SWiM) reporting guideline.

Subgroup analysis and investigation of heterogeneity

If sufficient data are available, we will conduct the following subgroup analyses.

• Stage of disease (locally advanced or unresectable versus metastatic)

• PD‐L1 status (positive versus negative)

• Type of ICI (anti‐PD‐1 versus anti‐PD‐L1 versus anti‐CTLA‐4)

• Type of active comparator (chemotherapy versus targeted therapy)

Sensitivity analysis

We will perform sensitivity analysis by excluding studies at high risk of bias from the meta‐analysis. For testing the robustness of our findings regardless of which method was chosen, we will conduct a sensitivity analysis for the primary outcome using a fixed‐effect model. In case of divergence between the two models, we will present both results; otherwise, we will present only results from the random‐effects model.

Summary of findings and assessment of the certainty of the evidence

We will create a summary of findings table for each main comparison to summarise data on our outcomes. We have selected the following main comparisons because we believe these to be the most important for decision makers.

• Single‐agent ICI versus inactive control intervention (e.g. no treatment or best supportive care).

• Single‐agent ICI versus other active control agents (e.g. chemotherapy or targeted therapy, or both).

• Doublet ICI (i.e. two separate ICI) versus inactive control intervention (e.g. no treatment or best supportive care).

• Doublet ICI (i.e. two separate ICI) versus other active control agents (e.g. chemotherapy or targeted therapy, or both).

We will report the following outcomes in each summary of findings table.

• OS determined at the end of study follow‐up.

• PFS determined at the end of study follow‐up.

• Overall ORR based on the best response assessment at the end of study follow‐up – we will report RECIST v1.1 as the default.

• HRQoL: we will report the EORTC QLQ‐C30 questionnaire at the 12‐week time point as the default.

• AEs determined at the end of study follow‐up.

Two review authors (WYC and VC) will independently assess the certainty of the evidence. We will resolve any disagreement by discussion, or by involving a third review author (HS). We will use the GRADE assessment process to determine the certainty of the body of evidence for each outcome, as described in Chapter 14.2 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022). The GRADE system uses five considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to judge the certainty of evidence as high, moderate, low, or very low. We will justify all decisions to downgrade the certainty of the evidence using footnotes, and we will make comments to aid the reader's understanding of the review where necessary. We will incorporate the GRADE judgements about the certainty of the evidence in our reporting of the results.