Abstract:
|
Patients engaging with decisions about their medical care require relevant, interpretable, personalized information to help synthesize risks and benefits of treatment. Ideally, these methods should incorporate individual characteristics that modify responses to treatment and their preferences for possible health states. Thus we construct hierarchical Bayesian models that jointly predict multiple outcomes of treatments for individuals, accounting for outcome correlations and effect modification. The key benefits include the ability to make joint inference on multiple outcomes and to incorporate individual preferences. We demonstrate our methods in the setting of interpreting randomized clinical trial data on cancer treatments. Here, patients must weigh potential survival benefits against increased risks of adverse effects from treatment, but these outcomes are usually analyzed separately. Informed by real clinical trial data, we also use simulation to compare allocation methods that assign patients to "best" treatment using various combinations of 1) population- vs individual-level expected outcomes and 2) single vs multiple health-related outcomes.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.