Performance and Fairness of a Peripheral Artery Disease Detection Model Across Patient Phenotypes and Health Systems
- 1University of California San Diego Department of Surgery
Abstract
BackgroundPeripheral artery disease (PAD) is a major cause of cardiovascular events but remains underdiagnosed. Electronic health record (EHR)-based machine learning models show promise for earlier detection, but developing generalizable and fair models across diverse populations remains challenging.
MethodsUsing the University of California Health Data Warehouse, containing EHR data from five health systems, we identified patients with and without PAD. We used unsupervised clustering to define PAD phenotypes and trained a LightGBM classifier using 14,023 features spanning demographics, comorbidities, medications, laboratory values, healthcare utilization, and diagnosis, procedure, and medication codes. We evaluated performance overall and across demographic groups and phenotypes, and assessed fairness using selection rates and subgroup differences in true- and false-positive rates.
ResultsThe study included 33,739 cases and 33,739 matched controls. Clustering identified four phenotypes: patients with limited healthcare documentation (cluster 1), younger patients with severe metabolic disease (cluster 2), patients with a traditional atherosclerotic risk profile (cluster 3), and frail elderly patients with multimorbidity (cluster 4). Overall, the model demonstrated consistent performance across institutions (AUROC 0.76-0.79; AUC-PR 0.76-0.79) with well- calibrated probabilities. Performance was similar across genders, with modest variation by race and age, and was stronger in clusters 2-4. Cluster 2 demonstrated the highest sensitivity (TPR 0.87, 95% CI 0.87-0.88), while cluster 1 showed the lowest performance (TPR 0.40, 95% CI 0.39-0.41).
ConclusionsThe EHR-based PAD detection model demonstrated consistent performance across five health systems. Phenotypic clustering revealed clinically meaningful differences in model performance adding an additional consideration in ML fairness and performance evaluations.
Related
- Behavioral-metabolic phenotypes and differential risks of diabetic kidney disease and peripheral neuropathy in type 2 diabetes: a latent class analysis
- Post-exercise Ankle brachial Index Increases the Detection of Peripheral Artery Disease in Patients with Exertional Limb Symptoms
- Evaluating the impact of social determinants of health on model performance and fairness in artificial intelligence for glaucoma progression prediction
- Assessment of Fairness and Bias of an Image-based Surgical Site Infection Detection AI Model
- Diagnostic Performance of Deep Learning Using Retinal Images for Detection of Chronic Kidney Disease: A Systematic Review and Meta-Analysis
BibTeX
@misc{kallis2026performance,
title = {{Performance and Fairness of a Peripheral Artery Disease Detection Model Across Patient Phenotypes and Health Systems}},
author = {K. Kallis and C. R. Quitevis and M. Ramsis and N.-K. Kabutey and M. S. Conte and V. L. Rowe and M. D. Humphries and T. Hernandez-Boussard and M. B. Malas and E. G. Ross},
howpublished = {medRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.19.26360861},
url = {https://www.medrxiv.org/content/10.64898/2026.08.19.26360861v2},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Performance and Fairness of a Peripheral Artery Disease Detection Model Across Patient Phenotypes and Health Systems
AU - K. Kallis
AU - C. R. Quitevis
AU - M. Ramsis
AU - N.-K. Kabutey
AU - M. S. Conte
AU - V. L. Rowe
AU - M. D. Humphries
AU - T. Hernandez-Boussard
AU - M. B. Malas
AU - E. G. Ross
PY - 2026
DA - 2026/08/19
PB - medRxiv
DO - 10.64898/2026.08.19.26360861
UR - https://www.medrxiv.org/content/10.64898/2026.08.19.26360861v2
AB - BackgroundPeripheral artery disease (PAD) is a major cause of cardiovascular events but remains underdiagnosed. Electronic health record (EHR)-based machine learning models show promise for earlier detection, but developing generalizable and fair models across diverse populations remains challenging. MethodsUsing the University of California Health Data Warehouse, containing EHR data from five health systems, we identified patients with and without PAD. We used unsupervised clustering to define PAD phenotypes and trained a LightGBM classifier using 14,023 features spanning demographics, comorbidities, medications, laboratory values, healthcare utilization, and diagnosis, procedure, and medication codes. We evaluated performance overall and across demographic groups and phenotypes, and assessed fairness using selection rates and subgroup differences in true- and false-positive rates. ResultsThe study included 33,739 cases and 33,739 matched controls. Clustering identified four phenotypes: patients with limited healthcare documentation (cluster 1), younger patients with severe metabolic disease (cluster 2), patients with a traditional atherosclerotic risk profile (cluster 3), and frail elderly patients with multimorbidity (cluster 4). Overall, the model demonstrated consistent performance across institutions (AUROC 0.76-0.79; AUC-PR 0.76-0.79) with well- calibrated probabilities. Performance was similar across genders, with modest variation by race and age, and was stronger in clusters 2-4. Cluster 2 demonstrated the highest sensitivity (TPR 0.87, 95% CI 0.87-0.88), while cluster 1 showed the lowest performance (TPR 0.40, 95% CI 0.39-0.41). ConclusionsThe EHR-based PAD detection model demonstrated consistent performance across five health systems. Phenotypic clustering revealed clinically meaningful differences in model performance adding an additional consideration in ML fairness and performance evaluations.
ER -