Artificial intelligence for detecting bipolar disorder in electronic health records of patients with affective diagnoses: a diagnostic accuracy study

BackgroundBipolar disorder (BD) is frequently underdiagnosed, particularly in patients presenting with depressive disorders, leading to delays in appropriate treatment. Artificial intelligence (AI) applied to electronic health records (EHRs) may improve early detection by identifying clinically relevant symptom patterns. ObjectiveTo evaluate the diagnostic performance of a natural language processing (NLP)-based AI model for detecting BD-related features in EHRs of patients with affective diagnoses. MethodsA retrospective diagnostic accuracy study was conducted using 500 EHRs from a psychiatric referral hospital in Bogota, Colombia (2020-2024). The model extracted 18 predefined clinical domains from unstructured text and classified patients into four risk categories. Diagnostic performance was assessed in a validation subset of 100 records using independent psychiatric evaluation as the reference standard. Sensitivity, specificity, positive and negative predictive values, F1-score, and area under the receiver operating characteristic curve (AUC-ROC) were calculated. ResultsThe model achieved high agreement in symptom extraction (mean 91.1%). Sensitivity was 96.4% (95% CI: 87.7%-99.0%) and specificity was 84.4% (95% CI: 71.2%-92.3%), with an F1-score of 0.92 and an AUC-ROC of 0.932 (95% CI: 0.881-0.975). A substantial proportion of patients with depressive diagnoses were identified as having confirmed BD or clinically relevant risk. The model analyzed complete EHRs 120 times faster than human reviewers. ConclusionsNLP-based analysis of EHRs can achieve clinically meaningful performance in identifying BD-related patterns while substantially reducing review time. The model may be useful as a clinical decision support tool for earlier identification of bipolar disorder.