Predicting Diagnostic Conversion From Major Depressive Disorder to Bipolar Disorder: An EHR Based Study From Colombia

Area of Science:

• Psychiatry and Mental Health
• Health Informatics
• Clinical Epidemiology

Background:

• Most patients with bipolar disorder (BD) initially present with depressive symptoms, leading to delayed diagnosis and poorer outcomes.
• Early identification of individuals with major depressive disorder (MDD) who are at risk of converting to BD is critical for timely intervention.
• Electronic Health Records (EHR) offer a valuable resource for identifying predictive features of BD conversion.

Purpose of the Study:

• To identify clinical and demographic features that predict the conversion from MDD to BD using EHR data.
• To leverage a multivariable Cox regression model for identifying significant predictors of BD conversion.
• To explore the utility of natural language processing (NLP) for extracting novel risk factors from clinical notes.

Main Methods:

• Analysis of 15 years of EHR data from 13,607 patients diagnosed with MDD.
• Application of a multivariable Cox regression model to identify predictors of MDD to BD conversion.
• Utilized NLP to extract suicidality as a predictive feature from clinical notes.

Main Results:

• 11.8% of MDD patients (1610 out of 13,607) transitioned to BD within the study period.
• Key predictors of BD conversion included initial MDD episode severity, psychosis, hospitalization, family history of BD, and female gender.
• NLP-derived suicidality and specific medication classes (mood stabilizers, antipsychotics) were also significant predictors.

Conclusions:

• EHR data analysis confirms known risk factors (e.g., psychotic depression, female gender) and identifies novel ones (e.g., suicidality from clinical notes).
• The findings validate the use of EHR data for predicting BD conversion and discovering new risk factors.
• Improved early diagnosis of BD is achievable through the identification of these predictive features.