When Death Is a Choice: Statistical Implications of Medical Assistance in Dying for Oncology Clinical Trials

Background: Medical Assistance in Dying (MAID) is increasingly available in jurisdictions caring for patients with advanced cancer. Because MAID directly influences the timing of death, it introduces challenges for the interpretation and analysis of oncology time-to-event endpoints. Existing survival-analysis methods, competing-risk approaches, and estimand frameworks were largely developed without considering a patient-directed medical intervention that may alter the timing of mortality. Methods: We examined the potential implications of MAID across the oncology drug-development lifecycle, including phase I dose-finding studies, phase II efficacy studies, randomized phase III trials, real-world evidence analyses, and regulatory decision-making. Conceptual issues related to endpoint interpretation, estimands, transportability, and causal inference were reviewed. In addition, a Monte Carlo simulation study was conducted to evaluate several approaches for handling MAID-related observations in a two-arm randomized trial, including right-censoring, competing-risk analyses, inverse probability weighting, composite-event approaches, Cox regression adjustment, and interval-censored survival models. Simulations examined varying treatment effects, sample sizes, and differential MAID frequencies between treatment groups. Results: MAID has the potential to influence the interpretation of commonly used oncology endpoints, including overall survival, progression-free survival, eventfree survival, and disease-free survival, because observed treatment effects may reflect both biological and patient-mediated pathways affecting mortality. In simulation studies, conventional right-censoring, competing-risk analyses, and inverse probability weighted approaches exhibited substantial Type I error inflation when MAID frequencies differed between treatment groups despite identical underlying survival distributions. In contrast, interval-censored analyses consistently maintained nominal Type I error rates and achieved statistical power nearly identical to that obtained using the true underlying event times. Similar findings were observed regardless of which treatment group experienced the higher MAID frequency. Conclusions: MAID challenges traditional assumptions underlying oncology time-to-event analyses by introducing a patient-directed process that may influence mortality independently of disease progression. The conceptual and simulation findings suggest that MAID-related observations may often be more appropriately viewed as partially observed survival outcomes rather than as conventional censored observations or competing events. Interval-censored survival models provide a principled framework that preserves the overall-survival estimand while incorporating information regarding the timing of death. Additional methodological research is needed to evaluate the impact of MAID across the oncology drug-development lifecycle, including phase I dose-finding studies, phase II trial designs, randomized phase III studies, and real-world evidence analyses.