Machine Learning May Help Identify Borderline Personality Disorder

Researchers developed a machine learning algorithm to screen for borderline personality disorder using a clinical expert's rating and de-identified electronic health records.

The following article is a part of conference coverage from Psych Congress 2021 , held October 29th through November 1, 2021, in San Antonio, Texas. The team at Psychiatry Advisor will be reporting on the latest news and research conducted by leading experts in psychiatry. Check back for more from the Psych Congress 2021.


A machine learning algorithm has demonstrated early efficacy in helping identify potential cases of borderline personality disorder (BoPD), according to study results recently presented at the 2021 Psych Congress, held from October 29 to November 1, 2021, in San Antonio, Texas.

Two cohorts of individuals, one with formal BoPD diagnoses (n=7112) and one without formal diagnoses but considered as “potential BoPD” (n=183,475), were extracted from the Cerner Health Facts database. The second cohort included individuals with either suicidal or intentional self-harm/bipolar disorder or other mental disorders across at least 3 diagnosis categories associated with BoPD. In the potential BoPD cohort, a clinical BoPD expert reviewed 456 records taken from a pair of independent random samples and assigned a likelihood of BoPD for each individual. The study researchers developed a machine learning algorithm to screen for BoPD using the clinical expert’s rating and de-identified electronic health records.

Across samples 1 and 2 within the potential BoPD cohort, the clinical expert identified 127 (28%) individuals deemed most likely to have BoPD; these comprised 29% of sample 1 (screening algorithm training set) and 27% of sample 2 (screening algorithm test set; P =.676). For the screening algorithm, an area under the receiver operating characteristics curve of 0.84 along with a positive predictive value of 0.72 were revealed by out-of-sample test results. The researchers indicated that 70% of those identified are highly likely to have a BoPD diagnosis. The accuracy of the machine learning algorithm was 0.82, its sensitivity was 0.54, and its specificity was 0.92.

The study authors concluded that “[initial] data support the utility of a machine learning algorithm to identify potential patients with BoPD.”

Disclosure: This clinical trial was supported by Boehringer Ingelheim International GmbH. Please see the original reference for a full list of authors’ disclosures.


Visit Psychiatry Advisor’s meetings section for complete coverage of Psych Congress 2021.



Goodman M, Yang L, Sharma VM, Shao N. Automatic screening of borderline personality disorder using electronic health record data. Presented at: Psych Congress 2021; October 29-November 1, 2021; San Antonio, Texas. Poster 22.