Tailoring Treatment Information Using Personal Characteristics and Health Outcome Preferences
*Laura Hatfield, Harvard Medical School
Keywords: Bayesian models, risk-benefit analysis, health utilities, shared decision-making
Patients choosing among treatments face a complex decision process. For example, patients with colon cancer must weigh toxicity of chemotherapy against potential life extension. In elderly patients with limited life expectancy, the risk-benefits tradeoffs are particularly acute. This calculation is further complicated when individuals respond differently to treatments. Methods are needed to synthesize multiple sources of information into clear, decision-relevant, personalized recommendations. We analyze clinical trials of colon cancer treatments that measure multiple health outcomes using hierarchical Bayesian models for expected outcomes of each treatment, allowing variation across observed patient characteristics. Then, we add a second stage to convert expected outcomes using health-related utilities specific to colon cancer and its treatments. Specifically, we translate into quality-adjusted life expectancy (QALE). Informed by these data, we use simulation to compare the performance of assigning individuals to optimal treatment using combinations of 1) population- vs. individual-level outcomes, 2) single- vs. multi-outcome models, and 3) fixed vs. person-level utilities.