Hepatotoxicity (liver damage) is a common problem in drug treatment trials but is observed only indirectly through biomarkers measured in the blood. This creates the need to infer an individual's unobserved liver function dynamically using blood tests and other patient baseline characteristics. Major statistical challenges include high dimensionality, irregular time observation points over patients, presence of missing observations, and noise involved in measurement and biological processes. This article introduces a class of multivariate Bayesian dynamic stochastic models for detecting and forecasting changes in an individual's liver function in two situations: without and with drug. These models separate measurement error from variation inherent in a biological process, and describe the underlying process of liver detoxification, whereby, drug affects liver function which in turn induces changes in observed analytes. We apply the Bayesian methodology to make an inference. A clinical toxicity study is examined, together with simulated data. The results suggest that changes in observed analytes can be captured by the proposed models.
All Science Journal Classification (ASJC) codes
- Statistics and Probability