Modeling the Natural Course of Treatment-Resistant Depression Using the US STAR*D Data

Predictive equations were derived to characterize the natural course of treatment-resistant depression (TRD). Patients experiencing major depressive disorder symptoms after failing two antidepressants are considered TRD. Statistical analyses were conducted using data from the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study. During treatment induction, interrelated time-varying changes between patient’s severity and level of suicidal ideation were assessed. During treatment maintenance, relapses and remission were characterized. Changes in patient’s severity were measured by the 16-item Quick Inventory of Depressive Symptomatology Self-Report (QIDS-SR16) and modeled using repeated-measures mixed-effects with random intercept and slope. Changes in level of suicidal ideation were measured using Item-12 of QIDS-SR16 and modeled using multinomial logistic regression. Time-to-relapse was modeled using a Cox proportional hazard model with transition from response to remission as time-dependent covariate. Time-to-remission was modeled using a constant rate in an exponential distribution.

During induction, the equations replicated the observed mean QIDS-SR16, suicidal ideation levels distribution, and the total patients achieving both response and remission: 9.0% (observed) vs. 9.87% (equations). During maintenance, the equations closely replicated the number of relapses (198 [observed] vs. 195 [equations]) and the number of patients achieving remission (136 [observed] vs. 145 [equations]).

To our knowledge, these are the first predictive equations modeling disease progression in depression. The methods could be generalizable to other TRD studies using other depression symptoms scales such as MADRS. The equations reliably predict short- and long-term disease symptoms and endpoints of patients with TRD. The equations could inform health policy, clinical decisions and economic evaluations.