Implementing the principal stratum strategy for intercurrent events with survival outcomes: a tutorial

Xiaoxiao Zhou, Joyce Chen, Pallavi Mishra-Kalyani, Xiaoxue Li, Yuan Li Shen, Shu Wang, Susan Halabi, Fan Li, [1em] Department of Biostatistics, University of Alabama at Birmingham, Alabama, USA, [0.5em] Corresponding author: Fan Li, Department of Statistical Science, Duke University [email protected] · 2026

Read Implementing the principal stratum strategy for intercurrent events with survival outcomes: a tutorial by Xiaoxiao Zhou, Joyce Chen, Pallavi Mishra-Kalyani, Xiaoxue Li, Yuan Li Shen, Shu Wang, Susan Halabi, Fan Li, [1em] Department of Biostatistics, University of Alabama at Birmingham, Alabama, USA, [0.5em] Corresponding author: Fan Li, Department of Statistical Science, Duke University [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-27655.txt.

Introduction

Xiaoxiao Zhou and colleagues tutor trial staff in the principal stratum strategy for intercurrent events with survival outcomes. The tutorial defines causal effects for a binary treatment with a single binary stop-treatment event, reviewing mixture-model and weighting methods with assumptions, models, sensitivity checks, and example code. Simulated studies mimicking a real cancer trial test how the methods operate.

Catalog record

CREATORXiaoxiao Zhou, Joyce Chen, Pallavi Mishra-Kalyani, Xiaoxue Li, Yuan Li Shen, Shu Wang, Susan Halabi, Fan Li, [1em] Department of Biostatistics, University of Alabama at Birmingham, Alabama, USA, [0.5em] Corresponding author: Fan Li, Department of Statistical Science, Duke University [email protected]
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE81837
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.27655

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Implementing the principal stratum strategy for intercurrent events with survival outcomes: a tutorial Xiaoxiao Zhou, Joyce Chen, Pallavi Mishra-Kalyani, Xiaoxue Li, Yuan Li Shen, Shu Wang, Susan Halabi, Fan Li, [1em] Department of Biostatistics, University of Alabama at Birmingham, Alabama, USA, [0.5em] Corresponding author: Fan Li, Department of Statistical Science, Duke University [email protected] Abstract The International Council for Harmonization (ICH) E9 (R1) addendum provides the estimand framework to formulate treatment effects in a clinical trial. One of the attributes of an estimand the framework describes is intercurrent events. Among the five strategies to intercurrent events the guidance lists, the principal stratum strategy is the most conceptually and technically challenging because it defines treatment effects on unobserved strata. Its application to survival outcomes is particularly inaccessible to practitioners. This tutorial reviews the methodology and implementation of the estimand framework with the principal stratum strategy to address intercurrent events with survival outcomes. We illustrate using a clinical trial in oncology and focus on a simple case with binary treatment and a single binary intercurrent event of discontinuation of the assigned treatment. We define the causal effects and review two main methods for estimating the effects: the mixture model method and the weighting method. For each method, we elaborate the associated assumptions, models, sensitivity analysis, software and provide example R code. We conduct simulation studies that mimic the real study to study the operation characteristics of these methods. Keywords: estimand; intercurrent events; oncology; principal stratum; survival outcomes. 1 Introduction Intercurrent events (ICEs) are common in clinical trials; these are post-randomization events that can affect either the interpretation or the existence of measurements relevant to the clinical question of interest. There are many types of ICEs in clinical trials and we review two main types. The first type is non-adherence to the assigned treatment, including switching or discontinuation of the treatment. For example, in a trial to compare the efficacy of two prostate cancer treatments, the primary endpoint is the overall survival (OS) time, but a substantial portion of patients discontinued the assigned treatment due to toxicity, which is potentially caused by the treatment and affects the OS time. The second type – common in oncology trials – is the intermediate disease or complication status between the initial treatment and the final clinical endpoint. We give three examples. A first example concerns the progression-free survival time (PFS) in a slow-growing tumor trial, where the ICE is the symptomatic progression of the cancer, e.g., tumor growth, new lesions, or worsening disease. A second example concerns the survival time for patients who had achieved complete response/remission (CR) or partial response (PR) after the initial treatment in a malignant hematology trial, where the ICE is the CR/PR status. A third example concerns the complication-free survival time after allogeneic stem cell or bone marrow transplantation, where the ICE is the post-treatment complication of Graft-versus-Host Disease (GVHD). Often both types of ICEs are present in a clinical trial. The presence of ICEs in a clinical trial complicates the definition of causal estimands and statistical analysis. The ICH E9 (R1) addendum extends the original ICH E9 guidelines to include statistical principles related to estimands and sensitivity analysis in clinical trials. The addendum provides the estimand framework – a structured approach to discuss the treatment effect(s) of interest that a clinical trial should address and align design, conduct, analysis and interpretation with the clinical trial objectives and questions of interest. The estimand consists of five key attributes: population, treatment, endpoint, population-level summary, and strategies to address ICEs. The addendum lists five strategies to address ICEs: treatment policy, hypothetical, principal stratum, while on treatment, and composite strategies. These can be applied separately or in combination, depending on the nature of the ICEs and the study objectives. Recent research has extensively discussed the general principles of the estimand framework, focusing on selecting and formulating appropriate estimands. 26 , 22 , 38 , 16 , 31 , 2 However, practical guidance on the implementation of each ICE strategy remains limited. In particular, the principal stratum strategy has been relatively less adopted, partially due to its conceptual and computational complexity. This strategy is based on principal stratification, 14 which is a causal inference framework for defining estimands in the presence of post-treatment confounding factors, e.g., an ICE. Principal stratification is a classification of subjects based on their joint counterfactual ICE statuses under both treatment and control, i.e., the principal strata, and the target estimand is the treatment effects within a principal stratum, the mathematical formulation of which is given in Section 2 . The choice of target estimands depends on the specific clinical context. For example, when the ICE is treatment discontinuation, a target estimand is the treatment effect in overall survival time for the principal stratum of patients who can tolerate and continue the assigned treatment regardless of the initial assignment. When the ICE is cancer progression, a target estimand is the treatment effect in survival time for the principal stratum of patients whose cancer would not progress regardless of the initial assignment. Because each patient is observed under only one treatment, patients within a principal stratum cannot be identified directly, neither before the trial nor using data from a parallel-group randomized trial. Therefore, principal strata are latent subpopulations, differing from the subpopulations in the per-protocol analyses, where the membership is determined solely by the observed ICE status under the assigned treatment. Additional structural and modeling assumptions are necessary to identify these latent strata and associated causal effects. 42 , 19 , 23 In the literature, there are two main estimation approaches: (i) the mixture model approach based on the exclusion restriction assumption; 18 , 42 , 28 (ii) the principal score weighting approach based on the principal ignorability assumption. 20 , 12 , 19 , 5 Implementing either approach requires substantial effort in statistical computation and programming. Survival or time-to-event endpoints, prevalent in clinical research, bring additional challenges to implementing the principal stratum strategy. To this end, Liu and Li 24 developed the mixture-model-based estimation method with the associated PStrata R package, and Cheng et al. 7 developed the principal-score-weighting estimation method with the associated mrPStrata R package. 6 Despite these advances, the gap between theory and practice remains wide: the principal stratum strategy continues to be poorly understood in regulatory studies. Noirjean et al. 32 provided a recent case study that compares the hypothetical strategy with the principal stratum strategy in a vaccine trial, but did not supply detailed information on implementation. This gap motivates us to provide a tutorial for implementing the estimand framework with a principal stratum strategy with survival outcomes. We acknowledge that the principal strata strategy is conceptually controversial in the causal inference literature. Some authors caution against causal effects defined in the latent principal strata because they lack transparency and require implausible assumptions to identify and instead advocate mediation analysis, 38 , 41 while other authors 40 , 13 adopt a more balanced view, recognizing that principal causal effects can provide refined clinical insights beyond the conventional intention-to-treat effects. Instead of taking a firm position in the ongoing conceptual debate, our goal is to provide potential users with detailed information on the principal stratum strategy so that

Read the complete text