Defining the treatment effects in clinical trials that collect multivariate outcome data longitudinally is a difficult and open problem. The problem is further complicated by the heterogeneity of data, outcome scales, missing data, and correlation within and between outcomes of the same subject. To address this problem, this project proposes to develop the multidimensional latent trait linear mixed model (MLTLMM), define the treatment effect, and build the necessary complexity of the model to incorporate the major components of the data that could lead to strong biases in treatment effect estimation. The overall objectives of this proposal are to: 1) develop a modeling framework for analyzing multivariate longitudinal data and build an increasingly more sophisticated class of models that account for known, and currently ignored, problems in the data; 2) provide fast inferential and statistically principled approaches to inference; 3) develop a class of sensitivity analysis approaches to modeling choices; 4) develop tools for personalized dynamic predictions to facilitate targeted treatments; 5) apply these methods to data from current clinical trials; and 6) develop and standardize the newly proposed approaches via professional software development and web deployment. Our methods of defining and estimating the overall treatment effects in multivariate longitudinal data address the critical need across trials of many medical conditions (e.g., Alzheimer's disease, Huntington's disease) with a similar data structure.
This project provides statistical analysis methods for large clinical trial datasets where multiple health outcomes are measured at multiple visits. Methods are applied to two double-blind, placebo-controlled multi-site randomized phase III clinical trials of Parkinson's disease to provide a clear and simple clinical interpretation of the overall treatment effects.
Showing the most recent 10 out of 32 publications
