Purpose: Decision trees can use clinical predictors to determine whether to continue the same antidepressant or switch to a different treatment in older patients with major depressive disorder (MDD). We examined whether pharmacogenetic and pharmacokinetic variables could improve their performance. Procedures: We analyzed 191 participants from the Incomplete Response in Late-Life Depression: Getting to Remission (IRL-Grey) trial who had not responded fully after 4 weeks of venlafaxine XR (150 mg/d) and for whom venlafaxine up to 300 mg/d was continued for 8 additional weeks. CYP2D6 genotypes were determined; venlafaxine, o-desmethylvenlafaxine (ODV), and active moiety (AM) exposures at week 4 were calculated using population pharmacokinetic modeling. Decision tree analysis was performed using 5 early clinical predictors of eventual nonresponse identified in previous research and 4 pharmacogenetic and pharmacokinetic potential predictors. One decision tree was designed to optimize specificity (k=0.3), and another to optimize sensitivity (k=0.7). Results: Longer episode duration and lack of partial response at week 4 were retained as clinical predictors, and lower AM and ODV exposures were identified as additional predictors. Negative predictive values (NPVs) of the high-specificity and high-sensitivity trees (77.7% and 73.0%, respectively) were similar to NPVs in trees based solely on clinical predictors. Implications: Our methods can guide future studies combining clinical and biomarker data to address applied pharmacological questions relevant to day-to-day practice.
Kim et al. (2026) studied this question.