Skip to content


Joint Modeling of Longitudinal Endpoints in Clinical Trial

Chairs: Sudhakar Rao (Janssen), Xiang Li (Janssen) and Qing Xu (Taiho Oncology)

Speaker: Liangcai Zhang, PhD (Janssen)
Title: Multivariate Joint Modeling & Statistical Learning of Longitudinal Endpoints and Its Application to Trial Monitoring and Planning 
Abstract: In clinical development, multiple endpoints are measurable events or outcomes that can help determine whether a medical intervention or investigational drug is benefiting patients in both safety and efficacy. Modeling of multiple endpoints jointly is a way to analyze trial data taking “totality of information” into account quantitively, and has the potential to increase precision of estimates and therefore increase study power. Different joint modeling methods have been discussed in the literature. In one class of models, an endpoint is modeled conditioned on another endpoint, the so called “shared term” strategy of jointly analyzing a marker process with a survival outcome (Tsiatis et al 1995, Pawitan and Self 1993). Another class of models specifies a marginal process for each endpoint and these marginal models are then joined in one way or another via latent variables (Catalano and Ryan 1992).

In this session proposal, we present an alternative approach to model the association between different endpoints using random effects, whereby the average evolution of a specific endpoint is described with some parametric function of time, and random effects are used to describe subject-specific deviations from the average evolution. A joint multivariate distribution is imposed on each endpoint’s mean process on random effects to induce association between different endpoints. 

The implementation of the model and its added value will be discussed through a real-world example and a simulation study:

(a) Application to trial analysis: Dose response curves are typically characterized for the primary endpoint marginally (without joint modeling). Often multivariate longitudinal data is available. For example, in immunology trials, many endpoints such as IgG, ClinESSDAI, ESSPRI, ESSDAI, and STAR-Composite Score, etc, are collected over time, in addition to the primary endpoint. By making reasonable assumptions and leveraging such longitudinal data across different endpoints, we show that efficiency can be gained in estimating treatment effects and as a result better understand relationships among outcomes.

(b) Application to trial planning: A second case study uses simulation to illustrate power gain through modeling jointly secondary and auxiliary endpoints. Potential savings in terms of sample size and/or trial duration will be compared against conventional approaches, and we will discuss implications of type-I error inflation in this context

Speaker: Fang Chen, PhD (SAS)
Title: Bayesian Joint Modeling using SAS
Abstract: In clinical trials and observational studies, repeated measurements of longitudinal data and time-to-event data are often collected for patients. The two type of data may be correlated and associated with each other. Joint models offer an effective approach that uses a single model to account for the dependency among the data and to provide better inference and assessment on any potential treatment effects. This talk demonstrates how to use SAS software, specifically the MCMC procedure, to simultaneously fit Bayesian joint models for the longitudinal and survival components of the data. We will discuss how to fit various
survival models (parametric and Cox models), repeated measurement models, and random-effects models using the software. Comparison to the frequentist approach of fitting joint models (e.g. using PROC NLMIXED) will also be discussed, to demonstrate the pros and cons of different approaches.

Speakers: Sofia Pozsonyiova (GSK) 
Title: Joint Modelling to Assess the Correlation Between Weight Change over Time and Hypertension in HIV Clinical Trials
Abstract: Hypertension is one of the most common comorbidities and a leading risk factor for mortality worldwide. It is important to identify risk factors that are associated with the hypertension. Weight change over time could be one of the key risk factors for hypertension. We model the longitudinal post-baseline weight change over time and time to occurrence of hypertension jointly using the linear mixed effect models and proportional hazards model to assess the relationship between the hypertension, weight change, and some other factors including treatment in clinical trials. Modelling, parameter estimations, and simulation results will be presented. The results from pooled clinical trials data will be discussed.

Speaker: Hsien Ming J Hung, PhD (FDA)
Title: Some Regulatory Experiences in Joint Modeling of Longitudinal Endpoints in CNS Trials
Abstract: For decades, there are several prominent ideas for joint modeling for longitudinal data analysis in clinical trials. In CNS trials for some neurological diseases such as ALS, joint rank analysis has been used for evaluating treatment effects in regulatory applications. However, this analysis has recently received some challenges, when there are only a few deaths in a trial. Some alternative analyses have been proposed to generate more interpretable trial results. Some of them rely on parametric modeling with time-to-event endpoints considered together. Additionally, the endpoint may not have to be at a specific timepoint. This talk is to share some of the experiences with different approaches in terms of limitations and utilities for regulatory applications.

Disclaimer: This abstract reflects the views of the author and should not be construed to represent FDA’s views or policies.