Abstract

We created a cohort of new users of antidepressant drugs between 2008 and 2018 using observational data from Kaiser Permanente Washington. The data contain information on patients' demographics, anthropometric measurements, medication use, and patient health questionnaires (PHQ). The PHQ is a depression severity score ranging from 0 to 27, with 27 the most severe. Our goal is to find tailoring variables for the choice of an antidepressant drug based on patient characteristics to minimize patients' PHQ score in the first year of follow-up. Since PHQ measurements are not always available at cohort entry and they are measured sporadically during follow-up, we propose a sequential approach to multiple imputations to recover the baseline and monthly PHQ values. Dynamic weighted survival modeling is further used to develop an adaptive treatment strategy for the choice of antidepressant drug. The results are discussed and compared with those from a previous study using United Kingdom's data.