2026-10-14T14:57:18Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33307305 2026-09-29T12:52:02Z category_7 category_21 category_45 category_734 category_811 category_64 category_106 portal_87 item_type_6 month_year_09_2026

Estimand and Causal Framework for Cardiovascular Trials with Non-Susceptible Fractions Ayon Mukherjee (24610314) Atanu Bhattacharjee (2897717) Yongming Qu (3563060) Medicine Biotechnology Sociology Biological Sciences not elsewhere classified Mathematical Sciences not elsewhere classified Cancer Science Policy Causal inference Directed acyclic graphs (DAGs) Estimands Mixture cure models Non-susceptibility Time-to-event analysis <p>Cardiovascular outcome trials assume all randomized patients remain susceptible to future events, yet contemporary therapies may render a non-negligible fraction effectively non-susceptible. When this occurs, conventional estimands—hazard ratios and restricted mean survival times—conflate distinct treatment mechanisms, producing bias and limiting causal interpretability. We develop an estimand and causal inference framework for survival data with a latent non-susceptible fraction, aligned with ICH E9(R1) guidelines. Using mixture cure models and potential outcomes theory, treatment effects are decomposed into two components: the effect on long-horizon non-susceptibility (incidence) and the effect on event timing among susceptible individuals (latency), accommodating intercurrent events via strategy-specific counterfactuals. Simulation studies show that Cox proportional hazards estimands exhibit substantial bias in the presence of non-susceptible fractions, even under correct model specification. We additionally evaluate robustness to non-proportional hazards among susceptibles and quantify bias in both incidence and latency estimands across identifiability conditions. Non-susceptibility-aware estimands recover distinct causal mechanisms with superior interpretability. Application to two large cardiovascular outcome trials suggests that, to the extent an effect is present, it is concentrated in the incidence component—an increase in the probability of long-term event-free survival—with little evidence of an effect on event timing among susceptible individuals; however, both components are estimated imprecisely under the observed follow-up, so this decomposition is indicative rather than definitive. This distinction is entirely obscured by conventional hazard ratios. The proposed framework—with explicit formal identification conditions, explicitly layered by causal role and testability, clarified estimand targets, and an open-source interactive dashboard and Statistical Analysis Plan template—equips statisticians and regulators with practical tools to define and interpret treatment effects when sustained event-free survival is biologically plausible.</p> 2026-09-29T12:52:02Z Text Journal contribution 10.6084/m9.figshare.33307305.v2 https://figshare.com/articles/journal_contribution/Estimand_and_Causal_Framework_for_Cardiovascular_Trials_with_Non-Susceptible_Fractions/33307305 CC BY 4.0