PS28: A World of Patient-Centricity: Modern Statistical Methods for Information-Rich and Clinically Relevant RCTs

Session overview: Randomized clinical trials have long been designed and analyzed with a focus on a single primary clinical outcome, often considered the most important measure of treatment efficacy. This approach has proven effective in providing clear, decisive results, but it also comes with limitations. Patients, in practice, often care about multiple outcomes, each varying in significance. Attempting to capture the full scope of patient experiences through a single endpoint may therefore fall short of clinical relevance. In recent years, several patient-centric statistical methodologies have emerged to address this gap, particularly in late-phase trials. These include Generalized Pairwise Comparisons (GPC), Desirability of Outcome Ranking (DOOR), and Ordinal Longitudinal Models for State Occupancies. For conciseness, a description of all three methods is deferred to the end of this session proposal. What unites these methods is their ambition to integrate outcomes of different types into a single statistical analysis, guided by clinical considerations. These outcomes may reflect multiple aspects of treatment effects, but also intercurrent events, which are essential elements of the real-world patient experience. Combining measures of effectiveness with intercurrent event directly aligns with the 'composite strategy' of estimands. Unlike traditional methods that analyze endpoints separately, potentially overlooking clinically-important interactions between outcomes, these methodologies offer a more comprehensive view of treatment effects and patient experiences. In that sense, they serve as a bridge between the statistician's toolbox and patient-centric concerns. All three methods aim to ensure that patients with varied trial experiences (from a clinical perspective) -while having similar baseline characteristics-are represented differently in statistical analyses. Achieving this requires some effort compared to simple (but often overly simplistic) statistical approaches. A key challenge, therefore, is to show that information-rich analyses can sustain and, in many cases, enhance clinical interpretability. In other words, clinical trials can benefit from these methods not only in depth of information they provide, but also in terms of interpretability and clinical pertinence of the results. Conference sessions focused on these methods, especially those involving a diverse mix of academic, industry, and regulatory professionals, are crucial for addressing this challenge. Session Objective: Despite their shared objective, the methods of interest vary considerably in their underlying philosophy and approach. The goal of this session is therefore threefold: 1. To raise awareness of patient-centric statistical methods in clinical trials and illustrate their potential for more clinically-pertinent statistical analyses, 2. To explore the philosophical differences between these approaches, 3. To compare them in terms of clinical areas of application. The session will consist of three talks, each focused on one of the methods. Each presenter will focus on one method, presenting its core philosophy and developing one or two examples to illustrate its most compelling use. These contrasting examples should serve as the foundation for a lively debate, encouraging collaboration among advocates of these innovative methods. Ultimately, the session aims to enhance the clinical relevance of clinical trial designs by fostering a better understanding of these patient-centric approaches. Brief description of the methods i. Generalized Pairwise Comparisons Generalized Pairwise Comparisons (GPC) builds upon the Wilcoxon test, specifically its U-statistic version. Specifically, it is based on comparing pairs of patients across different treatment groups, determining whether one patient has a more desirable outcome than the other in the pair. The key advancement of GPC over the Wilcoxon test is its broader definition of "more desirable," which is not limited to univariate outcomes. It can incorporate more complex aspects, such as recognizing that a 1-day difference in survival between two patients, while numerically significant, may not be clinically meaningful. GPC can therefore account for the full range of outcomes relevant to the definition of "more desirable." Over the years, it has gained popularity in certain disease areas, particularly cardiology (where it is known as the "Win Ratio," one of the GPC-based statistics). ii. Desirability of Outcome Ranking Desirability of Outcome Ranking (DOOR) involves first creating a list of potential patient experiences during a trial, with each experience being multivariate in nature (e.g., surviving to the end of the trial without more than two severe adverse events). The next step in DOOR is to rank these experiences by clinical desirability, creating a new outcome for each patient based on the rank of their experience relative to all possible experiences. This "DOOR outcome" can then be analyzed using conventional methods for univariate ordinal outcomes. iii. Ordinal Longitudinal Models of State Occupancies Ordinal Longitudinal Models of State Occupancies (OLM-SO), while not always introduced explicitly as such, can be seen as a longitudinal version of DOOR. It acknowledges that defining and ranking overall experiences can be complex and subjective-for example, deciding whether a mild effect occurring early in a trial is more or less significant than a more severe effect occurring later. Instead, OLM-SO focuses on defining and ranking experiences within smaller time intervals, for which consensus can be easily achieved (e.g., what events occur to a patient on a specific day). The clinical trial data for each patient is then transformed into a series of rank values over time, reflecting the patient's status during each time interval. This data can then be analyzed using conventional methods for longitudinal analysis of ordinal outcomes.