A1575
Title: Estimating treatment decision rules for ordinal outcomes with applications to an antidepressant treatment trial
Authors: Adam Ciarleglio - George Washington University (United States) [presenting]
Abstract: In most studies of antidepressant treatment effects, response data are collected at multiple time points, yielding a treatment response trajectory for each subject. In many instances, these trajectories can reasonably be assumed to arise from latent classes that are characterized by both the timing and quality of the response, which can be considered as an ordinal outcome (e.g., non-responders, delayed responders, gradual responders, or early responders). Using data from a randomized controlled trial comparing different antidepressant treatment combinations, we sought to develop and evaluate a treatment decision rule for selecting an optimal combination treatment for falling in a more desirable trajectory class using patient characteristics measured at baseline. A two-stage approach is employed to estimate the decision rule. The first stage fits predictive models in each treatment arm using parametric and flexible non-parametric methods. The second stage uses predictions from the first stage to estimate a generalized odds ratio, which is then modeled as a function of the baseline characteristics, yielding the estimated treatment decision rule. This approach is assessed via a simulation study and applied and evaluated using the antidepressant trial data.
