INVERSE PREFERENCE ELICITATION FOR DYNAMIC TREATMENT REGIMES

Purpose: To provide decision support that allows room for patient and clinician preferences, we assess each potential treatment choice in a dynamic treatment regime, and determine "What health outcome preferences are consistent with this choice of treatment?"

Method: Dynamic treatment regimes can be constructed from data to identify tailored treatment choices based on both baseline characteristics and on accumulating patient information (response to previous treatments, etc.) in order to optimize the mean value of a specified health outcome. A difficulty with this approach is that there may not be a single outcome that is appropriate to optimize for all patients. For example, minimizing symptom severity and minimizing side-effect burden are both desirable objectives. However, depending on available treatments, the use of one or the other as the health outcome to be optimized will result in different dynamic treatment regimes. For example, a regime that minimizes expected symptom severity will tend to choose more aggressive drugs that are very effective but have a more severe side-effect burden. On the other hand, a regime that minimizes expected side-effect burden will choose drugs that are less effective but have a milder side-effect profile. A more flexible approach is to consider weighted combinations of conflicting health outcomes according to patient and clinician preference. We present a method that allows us to determine, for each possible treatment choice in a dynamic treatment regime, the range of preferences (say for symptoms versus side-effect reduction) that are consistent with that choice. In effect, our method computes the optimal treatment choice for all preferences and all tailorings simultaneously.

Result: Using data from the CATIE (Clinical Antipsychotic Trials of Intervention Effectiveness) trial, we illustrate how different treatment choices are consistent with different preferences for having fewer symptoms versus less weight gain. 

Conclusion: Our methodology allows us to use dynamic treatment regimes to help inform clinical decision making while respecting patient and clinician preferences for different health outcomes.