When patients’ voices aren’t heard: estimands and statistical methods for handling missing patient-reported outcomes in oncology studies

Introduction: Patient-reported outcomes (PROs) are integral to oncology clinical trials, yet missing data - especially due to intercurrent events (ICEs) including disease progression and death - poses challenges for robust and interpretable analysis. While regulatory and best practice guidelines now emphasize the explicit definition of estimands, including strategies for handling ICEs, and supplementary analyses to examine their robustness, practical recommendations for their implementation in PRO analyses remain limited. Methods We present a methodological framework for defining estimands and statistical analysis for a longitudinal change in a PRO confirmatory endpoint, including strategies for handling the main ICE of disease progression in addition to handling treatment discontinuation and death, using a simulated clinical trial. We propose for the ICE of disease progression using either a hypothetical or treatment policy strategy for the main and supplementary analysis, and present implementation of two methods targeting a hypothetical approach and one method for treatment policy approach (implicit multiple imputation in a longitudinal model, a joint modelling of longitudinal PROs and time-to-progression, and multiple imputation using control-based imputation post progression). Results We present the occurrence of ICEs and missing data and provide a tutorial for conducting analysis in the presence of disease progression using hypothetical and treatment policy strategies respectively. Despite the occurrence of disease progression events and other missing data, conducting supplementary analysis provided confidence in our overall interpretation for the simulated trial. Conclusions Our recommendations provide practical guidance for specifying estimands, selecting statistical analysis methods, and interpreting PRO analyses in oncology trials with missing data. Accurately estimating the treatment effect on quality-of-life, in a way which is interpretable, is crucial to aid patients and other stakeholders when making treatment decisions.