Machine Learning in the Differential Diagnosis of Bipolar and Borderline Personality Disorders: A Scoping Review
Aroldo Dargél, Tanya Tanya, Asma Alhadhrami, Sara Mahdiabadi, Risa Shorr, Caitlin Morris, Kathleen Pajer, Marion ImbaultBackground
Bipolar disorder (BD) and borderline personality disorder (BPD) are associated with substantial morbidity and can be difficult to distinguish when symptoms are assessed cross-sectionally. This scoping review examines machine learning (ML) and related predictive approaches designed to differentiate the two conditions.
Methods
We searched Ovid MEDLINE, APA PsycInfo, Embase, Web of Science, Scopus, IEEE Xplore, ACM Digital Library, and selected computer science conference proceedings through July 2, 2026. Full-text peer-reviewed journal articles and conference papers were eligible when they used a data-driven computational or statistical prediction approach to distinguish, classify, or characterize BD and BPD. Of 2,594 records identified, 2,523 unique records were screened, and 24 full texts were assessed. Screening followed the Population, Concept, Context framework, Joanna Briggs Institute guidance, and preferred reporting items for systematic reviews and meta-analyses extension for scoping reviews.
Results
Nine reports met the inclusion criteria. Seven participant-based reports contributed an aggregate of 986 report-level participants, although two reports used overlapping cohorts; one conference paper used synthetic data, and one study analyzed 149,798 Reddit posts. Studies including clinical samples reported BD versus BPD discrimination AUC (area under the receiver operating characteristic curve) values of 0.67 to 0.91 and accuracies ranging from 61.7% to 90%. These estimates were not directly comparable and were derived from small, often imbalanced samples using heterogeneous internal validation. No model was externally validated in an independent clinical cohort.
Conclusion
The evidence remains preliminary and does not support clinical deployment. Reported performance is constrained by small samples, class imbalance, uncertain ground-truth labels, spectrum bias, and reliance on pre-existing diagnoses. Prospective studies in consecutive at-risk cohorts, with transparent reporting and independent external validation, are required.