OBJECTIVES: Identify potential biases in indirect treatment comparisons of treatments when different starting points are considered to measure outcomes, and evaluate assumptions required to assess causality.

METHODS: We consider trials in resectable NSCLC with pre-operative and adjuvant immunotherapies, where outcomes are measured starting from different time points. We assess the practical feasibility of the treatment decision problem, the conditions for the transitivity assumption and additional assumptions needed to perform network meta-analyses (NMA).

RESULTS: The decision problem is only relevant to patients for whom all treatment options are equally applicable and it may be limited in clinical practice. The target population of the trials could differ, as patients may be excluded from trials with one treatment option, if the alternative treatment is more suitable. Therefore it is possible that specific patient characteristics, not necessarily known or collected, are effect modifiers which can bias the comparisons. The different treatment timing relative to surgery results in patient attrition causing selection bias, which is difficult to be accounted for with analytical methods, if individual patient data is not available for all studies. The outcome definition and assessment is also different, if the starting point is different.

CONCLUSIONS: Indirect treatment comparisons in resectable NSCLC comparing pre-operative to adjuvant immunotherapies are methodologically challenging for several reasons: it is unclear if the clinical trial populations can be considered exchangeable, the outcome definition and assessment differs, and the patient attrition is also different, if the starting point is different. Hence, it is unlikely that transitivity holds to employ traditional NMA methods for continuous endpoints assuming proportional hazards. Alternative methods would need to account for effect modifiers and the factors causing patient attrition which otherwise induces selection bias.