Toward precision psychiatry: integrating pharmacogenomics and multimodal clinical data for personalized antidepressant response prediction

Individual responses to antidepressant treatment are highly heterogeneous, resulting in prolonged trial-and-error processes that undermine clinical outcomes. Pharmacogenomics (PGx) holds substantial promise for advancing precision psychiatry by guiding personalized antidepressant selection. However, existing PGx-based prescribing tools remain insufficiently validated and underutilized in real-world clinical practice. This retrospective observational study analyzed real-world data from 4,216 inpatients diagnosed with major depressive disorder (MDD) who were admitted to Beijing Anding Hospital, Capital Medical University, between April 2022 and May 2025. Patients were 1:1 matched into two groups: a PGx group that received PGx-tested treatment, and a control group that received usual care without PGx testing. Multimodal data from the PGx group, including demographic characteristics, medication history, laboratory test results, psychometric assessments, and PGx reports, were used to develop a predictive model for personalized antidepressant response. PGx-tested treatment was associated with a statistically higher overall effective rate compared to usual care (88.1% vs. 85.6%, crude odds ratio [OR] = 1.24, 95% confidence interval [CI] = 1.03–1.48) but showed no significant differences in drug-specific response rate (73.2% vs. 70.3%) or average effective response time (12.4 days vs. 12.5 days). The developed predictive model achieved a mean AUC of 0.825 (95% CI = 0.785–0.865) for predicting treatment effectiveness across 10 commonly used antidepressants. In first-tier medication recommendations, the model demonstrated 35.4 percent higher recall and 54.2 percent higher precision than the commercial PGx tool. The Target Trial Emulation (TTE) validation showed that the model-assisted group exhibited superior outcomes: a HAMD score reduction rate of 77.5%, a drug-specific response rate of 93.0%, and an average effective response time of 10.3 days, all statistically better than the non-congruent and non-PGx groups. SHapley Additive exPlanations (SHAP) analysis confirmed known PGx associations consistent with existing literature and identified previously underexplored predictive factors. These findings highlight the potential to support precision psychiatry in real-world practice, reducing trial-and-error in antidepressant prescribing and optimizing clinical outcomes for patients with MDD.