Biostatistics Weekly Seminar

When studying health-economic outcomes, particularly for new treatments, it is common for subject-level cost and effectiveness data to be unavailable. Researchers have therefore developed techniques to synthesize data from published literature. One useful method is microsimulations, where a model framework consisting of health states is developed, and simulated patients move through the model, transitioning from state to state probabilistically. Motivated by a study of metastatic prostate cancer, we develop a general microsimulation framework to estimate the cost-effectiveness of therapy order (e.g. A→B vs. B→A). We discuss the advantages of microsimulation models over Markov cohort models for this type of analysis, and develop strategies to infer time-dependent state-transition probabilities based on survival curves available in published literature. We also develop a novel calibration method which both ensures that model-based results match overall survival data from target trials, and accounts for dependence of outcomes within patients across health states. We demonstrate how the Nelder-Mead method can be used to select parameters which minimize the difference between the target trial results and the model-based results. We demonstrate our method in a study of therapy sequence for prostate cancer.

Zoom (Link to Follow)
08 December 2021
1:30pm