AI detects bipolar disorder in EHRs of patients with affective diagnoses with AUC 0.93

PreprintmedRxiv 2026 · 1 min read
AUC · bipolar disorder
0.93
500 EHRs of patients with affective diagnoses
Sensitivity
96.4%
95% CI 87.7–99.0
Specificity
84.4%
95% CI 71.2–92.3

medRxiv diagnostic accuracy preprint: EHR model reports AUC 0.93, sensitivity 96.4%, and specificity 84.4%.

Eugenio Ferro, Natalia Castaño-Villegas, Manuel F. Esteban Cárdenas, María Gómez Puentes, Carlos Torres-Delgado, Laura Ortiz Calderon, Katherine Monsalve, José Zea

This diagnostic accuracy study evaluated artificial intelligence for detecting bipolar disorder in electronic health records from patients with affective diagnoses. The reported dataset included n=500 EHRs and a validation subset of n=100, following STARD 2015 and TRIPOD-AI reporting standards.

Headline metrics include AUC 0.93, sensitivity 96.4% (95% CI: 87.7–99.0), specificity 84.4% (95% CI: 71.2–92.3), and domain-level concordance 91.3% (95% CI: 89.9–92.6). Among 387 patients without a prior bipolar diagnosis, 36.4% were newly classified as at risk.

Results at a glance

The headline numbers from the paper, in plain language.

How the model performed

Higher is better

Data: 500 EHRs of patients with affective diagnoses — medRxiv 2026

Scores reported on a 0–1 scale (AUC, kappa) are drawn ×100.

Show the numbers as a table
MetricValueWhat it means
AUC · bipolar disorder0.93500 EHRs of patients with affective diagnoses
Sensitivity96.4%95% CI 87.7–99.0
Specificity84.4%95% CI 71.2–92.3

How to cite

Eugenio Ferro, Natalia Castaño-Villegas, Manuel F. Esteban Cárdenas, María Gómez Puentes, Carlos Torres-Delgado, Laura Ortiz Calderon, Katherine Monsalve, José Zea. Artificial intelligence for detecting bipolar disorder in electronic health records of patients with affective diagnoses: a diagnostic accuracy study. medRxiv. 2026. https://doi.org/10.64898/2026.05.07.26352679

@article{ferro2026artificial,
  title     = {{Artificial intelligence for detecting bipolar disorder in electronic health records of patients with affective diagnoses: a diagnostic accuracy study}},
  author    = {Eugenio Ferro and Natalia Castaño-Villegas and Manuel F. Esteban Cárdenas and María Gómez Puentes and Carlos Torres-Delgado and Laura Ortiz Calderon and Katherine Monsalve and José Zea},
  journal   = {medRxiv},
  year      = {2026},
  doi       = {10.64898/2026.05.07.26352679},
  url       = {https://www.medrxiv.org/content/10.64898/2026.05.07.26352679v1},
  note      = {Preprint}
}

More research from Arkangel AI

All research →

For hospitals and insurers

Validate these results at your institution

We run the model on your own clinical records with your team, under your privacy and governance rules.

Start a research partnership
Read paperCite