Modeling laboratory data from clinical trials

Abnormal laboratory data from clinical trials are considered precursors of potential organ dysfunction. Routine analysis of these data focusses primarily on the incidence of extremely high or low values relative to normal ranges, or the change of these incidences compared to baseline. Alternatively, the worst values from each individual patient are analyzed. Where feasible, graphical representations of the time course of lab measurements are displayed, often for each subject individually. However, only rarely is an attempt made to model the dynamics of the measurements over time. Stochastic processes can be used to model laboratory measurements and specifically Ornstein-Uhlenbeck processes have some properties that are suitable for this purpose. Models based on these processes can account for potential systematic effects relevant for all or a group of subjects as well as the random variation between subjects. The idea is to estimate parameters of these processes in order to quantify effects of drugs or other medical treatments. Results based on simulations and real data will be presented.