Events that occur after participants are randomly assigned to treatment in a clinical trial are known as intercurrent events. Examples include participants stopping or switching treatment, or dying before the study is over. These events can alter the course of the randomized clinical trials and make it more difficult to compare treatment effectiveness and use the study findings for decision making in regulatory science.
The standard intention-to-treat approach compares participants according to the treatment they were originally assigned, regardless of these later, intercurrent events. This preserves the trial validity of randomization but ignores differences in how participants respond to treatment and how intercurrent events influence their outcomes. The 2018 ICH E9(R1) addendum describes principal stratification as an alternative approach to handle intercurrent events. However significant gaps exist between the theory and practice of principal stratification in regulatory science. In particular, researchers lack clear and accessible analytical methods, practical guidance, and software for applying this approach in the context of regulatory science.
Project Description and Goals
This project aimed to develop clear and accessible statistical analysis tools, software and educational material to help researchers use principal stratification to analyze intercurrent events in clinical trials. We focused on the most prevalent type of intercurrent events: nonadherence to assigned treatment, including treatment switching and stopping. We developed estimand, computational, visualization, and sensitivity analysis tools, with a special emphasis on time-to-event outcomes. We also developed two companion R packages and a tutorial illustrating these methods with data from a real oncology clinical trial. In addition, we examined situations in which an event, such as death, makes it challenging to apply the principal strata strategy in these cases. This work provides researchers, regulators and industry sponsors with (1) new methodological tools for addressing a pressing and prevalent complication in clinical trials, and (2) freely available open-source software and educational material for addressing a common challenge in clinical trials. We shared the findings through scientific publications and presentations, as well as FDA research forums.
Research Results
The project delivered three main products: (1) two analytical methods known as Bayesian mixture modeling and multiply-robust weighting for conducting principal stratification analysis of non-adherence with survival outcomes, which measure the time until an event such as disease progression or death occurs; (2) an implementation-focused tutorial illustrating the principal stratum strategy with data from a real oncology trial, and (3) PStrata and mrPStrata, the first and to our knowledge, only software packages for implementing this strategy. Both packages are publicly available through CRAN, the repository for R statistical software. Equally important, the project provided a clear assessment of when the strategy is practical. When implemented properly, principal stratification can provide a useful picture of treatment effect heterogeneity. However, the strategy is technically and conceptually demanding, relies on strong assumptions that cannot be tested using the observed data, and often becomes impractical when a clinical trial involves more than one intercurrent event, particularly in oncology.
Publications (to date)
Cheng, C., Liu, B., Wruck, L., & Li, F. (2026). Multiply robust estimation for causal survival analysis with treatment noncompliance. The Annals of Applied Statistics, 20(1), 1–24. https://doi.org/10.1214/25-AOAS2117
Manuscript Preprints
Liu, B., & Li, F. (2023). PStrata: An R Package for Principal Stratification (arXiv:2304.02740). arXiv. https://doi.org/10.48550/arXiv.2304.02740
Zhou, X., Chen, J., Mishra-Kalyani, P., Li, X., Shen, Y. L., Wang, S., Halabi, S., & Li, F. (2026). Implementing the principal stratum strategy for intercurrent events with survival outcomes: A tutorial (arXiv:2605.27655; Version 1). arXiv. https://doi.org/10.48550/arXiv.2605.27655
Presentations
Li, F. (2024, April 4). Principal Stratification Analysis of Noncompliance with Time-to-Event Outcomes [Oral Presentation]. Duke Industry Statistical Symposium.
Zhou, X. (2025, May 2). Implementing the Principal Stratum Strategy for Intercurrent Events in the CALGB 90206 Trial: A Case Study in Oncology Randomized Trial [Seminar]. Yale-Mayo Clinic CERSI Scholar Seminar. https://youtu.be/4L8lo6bor6Y?si=9Tk26ZoFq0_WD1Ug
Halabi, S. (2026, January 11). Implementing the Principal Stratum Strategy for Intercurrent Events with Survival Outcomes: A Tutorial [Poster]. 2026 UCSF-Stanford CERSI Innovations in Regulatory Science Summit. https://pharm.ucsf.edu/cersi/2026-ucsf-stanford-cersi-innovations-in-regulatory-science-summit
Zhou, X. (2026, April 6). Implementing the Principal Stratum Strategy for Intercurrent Events with Survival Outcomes: A Tutorial [Poster]. UAB School of Public Health Annual Public Health Research Poster Session.
Li, F. (2026, March 6). Implementing the principal stratum strategy for intercurrent events with survival outcomes: A Tutorial [Oral Presentation]. FDA Oncology Center of Excellence Research Rounds.
Zhou, X. (2025, May 2). Implementing the Principal Stratum Strategy for Intercurrent Events in the CALGB 90206 Trial: A Case Study in Oncology Randomized Trial [Seminar]. Yale-Mayo Clinic CERSI Scholar Seminar. https://youtu.be/4L8lo6bor6Y?si=9Tk26ZoFq0_WD1Ug
Zhou, X., & Li, F. (2026a, May 11). Implementing the principal stratum strategy for intercurrent events with survival outcomes: A tutorial [Oral Presentation]. Society for Causal Inference. https://sci-info.org/abstract-search/?abst_id=20643
Zhou, X., & Li, F. (2026b, August 3). Implementing the principal stratum strategy for intercurrent events with survival outcomes: A tutorial [Oral Presentation]. Joint Statistical Meetings. https://ww2.amstat.org/meetings/jsm/2026/onlineprogram/abstract.cfm?sid=5528&tid=5532
Software
Liu, B., & Li, F. (2026). PStrata: Principal Stratification Analysis in R (Version 1.0.1) [Computer software]. https://cran.r-project.org/web/packages/PStrata/index.html
Cheng, C., Liu, B., Li, F., & Li, F. (2026). mrPStrata: Multiply Robust Estimation in Causal Survival Analysis with Treatment Noncompliance (Version 0.1.0) [Computer software]. https://cran.r-project.org/web/packages/mrPStrata/index.html
