Latent class modeling of repetitive transcranial magnetic stimulation response trajectories in treatment-resistant depression.

Over 30% of patients with major depressive disorder (MDD) develop treatment-resistant depression (TRD). Repetitive transcranial magnetic stimulation (rTMS) is effective for TRD, yet there is substantial heterogeneity in treatment response. The ability to characterize distinct response trajectories could improve clinical decision-making and enhance early triaging of patients receiving acute rTMS treatment. We applied latent class mixture modeling (LCMM) to Patient Health Questionnaire-9 (PHQ-9) scores collected weekly during acute rTMS treatment in 308 patients with TRD treated at the UCSD Interventional Psychiatry Program. We also conducted a sensitivity analysis with baseline-adjusted PHQ-9 scores. LASSO-penalized multinomial logistic regression with nested cross-validation identified baseline clinical predictors of trajectory class membership. A three-class quadratic model best fit raw PHQ-9 trajectories: High Baseline, Minimal Improvement (HBMI); Moderate Baseline, Moderate Improvement (MBMI); and High Baseline, Large Improvement (HBLI). The HBLI group achieved the greatest symptom reduction (65.5%) and highest response rate (84%). Baseline-adjusted modeling yielded a 2-class solution (moderate vs. large improvement). Predictive modeling of raw PHQ-9 achieved strong multi-class discrimination (AUC = 0.82). History of suicidal ideation was associated with higher odds of HBMI class membership, while tobacco use and lack of previous suicide attempt was associated with HBLI membership. In conclusion, patients with high baseline depression severity and suicidal ideation are at greatest risk for nonresponse and may benefit from early consideration of alternative interventions.