Improving the Efficiency of Individual Treatment Rule Estimation Using External Data

Oregon Convention Center 

Individualized treatment rule (ITR), which tailors treatments to individual patient's characteristics,
can be estimated using data from a randomized clinical trial (RCT). However, RCT is often conducted
under restricted inclusion criteria, so they are not representative of a target population. As the result, the ITR benefit
shown in the trial may not accurately reflect what will be attained in the target population. In this work, using a
large feature data from the target population, we first evaluate the benefit of ITRs in the target population, and then
we propose a new estimate for ITR based on two derived features which essentially correspond to a predictive score
and a residual score. The new ITR estimator leads to the smallest variance of the benefit in the target population.
Finally, we demonstrate the performance of the proposed method through simulation studies and an application to
treating type 2 diabetes.