Using natural language processing to assess MDD psychotherapy patterns among Veterans Affairs patients with suicide...

Major depressive disorder (MDD) is a leading risk factor for suicide. Within the US Department of Veterans Affairs (VA), psychotherapy is widely used to treat MDD and prevent suicide. Little is known about how classified suicide risk impacts this treatment. Addressing this gap, this study assesses patterns in MDD-diagnosed VA patients' unstructured electronic health record (EHR) notes using Latent Dirichlet Allocation, a methodology that evaluates textual patterns. We developed a suicide-risk-stratified sample (high-, moderate-, and low-) of MDD-diagnosed VA patients who died by suicide in 2017-2018 (cases) and risk-matched patients with similar demographics, diagnoses, and care who did not die by suicide (controls). After collecting all psychotherapy EHR notes within 1-year of case death, we derived topics, monitored topic differences across patient groups, and evaluated clinical relevance. We identified five topics: Disclosure, Suicide Risk, Treatment Goals, Symptoms, and Medication. Risk-matched cases and controls' topic patterns were consistently significantly different except for the Disclosure and Medication topics for high-risk patients and the Suicide Risk and Symptoms topics for low-risk patients. Our analyses highlighted how classified suicide risk-tier may impact psychotherapy delivery and utilization for suicide-risk-matched cases and controls. Most topics displayed decreased prominence for higher-risk versus lower-risk patients, with cases tending to receive more illness-focused care and controls tending to receive more supportive care. Identified differences offer to enhance risk modeling and improve suicide prevention psychotherapy.