Using multistate models with clinical trial data for a deeper understanding of complex disease processes

Music City Center 

Clinical trials are costly and require significant commitment. Maximizing the use of collected data is crucial so we can learn as much as possible. A multistate model describes longitudinal events, allowing multiple clinical endpoints to be treated as outcomes and covariates to be estimated simultaneously. Proportional hazards models are fitted for each transition, enabling calculations of absolute risks, probability of being in a state, expected number of visits to a state, and time spent in a state. For this presentation, we use a publicly available clinical trial dataset, myeloid, from the survival package in R to demonstrate the application of multistate hazards models. In the myeloid dataset, treatment B results in a longer duration of complete response (CR) compared to treatment A. Mutation status does not affect the rate of transition to CR but significantly influences the duration of CR. We also found that more patients in treatment A received transplants without achieving complete response (CR); in contrast, more patients in treatment B received transplants after achieving CR. Additionally, the mutation status significantly influences the transition rate from CR to transplant, whereas the treatment has no impact on this particular transition. Our observations on these datasets were made possible by multistate models. Clinical trial data offers much more than a simple yes/no answer if we, the statisticians, are willing to explore it.

multistate model

disease process