Predictors of Hospitalization in Patients With Newly Diagnosed Major Depressive Disorder: A Real-World Evidence Study

OBJECTIVES: Hospitalization for major depressive disorder (MDD) places a heavy burden on patients and the healthcare system. This study examines predictors of hospitalization in patients with newly-diagnosed MDD.

METHODS: A retrospective cohort study of electronic health record (EHR)-derived de-identified data from the NeuroBlu Database, a longitudinal behavioral health real-world database comprising structured and unstructured patient-level clinical data, was conducted in adults (≥18 years) with ≥2 MDD diagnoses (second diagnosis=index date) between 09/2000-06/2020. Insurance plan enrolment was required for ≥60 days pre-index and ≥30 days post-index. Patients with bipolar disorder/schizophrenia/schizoaffective disorder diagnosed before day 30 were excluded. Hospitalizations <30 days of index, and missing sex/race/marital status/clinical global impression scale (CGIS) were excluded. Patients were randomly split into train and test sets; C-index was estimated by 10-fold cross-validation (CV) using elastic-net regularized Cox models. Bipolar disorder/schizophrenia/schizoaffective disorder diagnoses were entered as time-varying covariates in the models. IRB approval was obtained prior to study conduct and included a waiver of HIPAA authorization.

RESULTS: 7,286 of 104,631 patients with MDD met study criteria. 43.5% of train (N=5,116) and 41.9% of test (N=2,170) patients were hospitalized; median time to hospitalization was 12.6 months and 12.9 months, respectively. Train patients were mostly single (39.9%), white (81.2%), female (68.2%) with non-severe MDD (85%), median age 42 years, and median CGIS 4. Test patients had similar characteristics. Age (HR=0.99), marital status (widowed HR=1.47, divorced/separated HR=1.23, single HR=1.32, reference=married), CGIS (HR=1.20), substance use (HR=1.70), family (HR=1.35), financial (HR=1.15), legal (HR=1.25), and occupational stressors (HR=1.16) were important predictors of hospitalization (all p<0.05). C-index was 0.687 (95% CI 0.668-0.688) in train set with CV and 0.672 (95% CI 0.653-0.691) in test set.

CONCLUSIONS: The 8-predictor model performed well in the test set. Stressors and CGIS are not routinely collected in EHR or claims data but are important in determining clinical outcomes.