Abstract:
|
To identify patients in the placebo group who are likely to suffer liver function abnormality over the entire study duration, we propose to extend the Mixture-of-Experts model to analyze longitudinal records of Alanine transaminase (ALT). We separate patients into multiple groups by ALT trajectories and identify the important characteristics associated with liver function abnormality. There are several features of the proposed model. First, we employ a nonparametric functional form to capture the underlying change patterns and possible "spikes" of ALT curves. Second, we investigate covariates to simultaneously explain between-cluster (i.e., serving as predictors of cluster membership) and within-cluster (i.e., accounting for within-cluster variability) heterogeneity with the proposed model. Third, the proposed model allows for individually different measurement times. The proposed model is evaluated using a simulation study. Our simulation study demonstrates that the proposed model is capable of separating trajectories into latent classes and generally estimating the parameters unbiasedly and precisely with appropriate empirical coverage for a nominal 95% confidence interval.
|