Scholarlink

Purpose <p>Postural orthostatic tachycardia syndrome (POTS) is a common autonomic disorder characterized by orthostatic intolerance and excessive tachycardia upon standing. Despite its prevalence, POTS is often underdiagnosed or diagnosed late, largely due to limited access to autonomic specialists and testing. This study aimed to evaluate the performance of machine learning (ML) models in diagnosing POTS using validated symptom surveys and physiological measurements.</p> Methods <p>We retrospectively analyzed data from patients evaluated at the Autonomic Laboratory at Brigham and Women’s Faulkner Hospital (2017–2025), with POTS diagnoses confirmed by autonomic testing. ML models based on a multilayer perceptron were trained using patient-reported surveys (Survey of Autonomic Symptoms [SAS], COMPASS-31) and autonomic testing data. Importantly, no orthostatic heart rate criteria for the POTS diagnosis were provided to the models.</p> Results <p>A total of 3210 patients were included, of whom 810&#xa0;had confirmed POTS. All patients completed SAS; 1337 also completed COMPASS-31 (334 with POTS). Models incorporating heart rate data achieved the highest diagnostic accuracy (PyTorch/LightGBM: AUC 0.98/0.99; precision 0.94/0.93; F1 score 0.88/0.93; sensitivity 83%/92%; specificity 98%/98%). In contrast, models trained solely on SAS (AUC 0.68/0.63) or COMPASS-31 (AUC 0.66/0.62) performed poorly.</p> Conclusion <p>ML models accurately diagnosed POTS when incorporating heart rate data alongside survey responses, with heart rate data being the strongest predictor. These findings suggest ML could assist in POTS diagnosis. The study also highlights the importance of heart rate measures in POTS diagnosis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Utilization of machine learning in diagnosis of postural tachycardia syndrome (POTS)

  • 期刊论文

  • 发表日期：

    2026年3月2日

查看全文
  • Seiji Akera,
  • Peter Novak

摘要

Purpose

Postural orthostatic tachycardia syndrome (POTS) is a common autonomic disorder characterized by orthostatic intolerance and excessive tachycardia upon standing. Despite its prevalence, POTS is often underdiagnosed or diagnosed late, largely due to limited access to autonomic specialists and testing. This study aimed to evaluate the performance of machine learning (ML) models in diagnosing POTS using validated symptom surveys and physiological measurements.

Methods

We retrospectively analyzed data from patients evaluated at the Autonomic Laboratory at Brigham and Women’s Faulkner Hospital (2017–2025), with POTS diagnoses confirmed by autonomic testing. ML models based on a multilayer perceptron were trained using patient-reported surveys (Survey of Autonomic Symptoms [SAS], COMPASS-31) and autonomic testing data. Importantly, no orthostatic heart rate criteria for the POTS diagnosis were provided to the models.

Results

A total of 3210 patients were included, of whom 810 had confirmed POTS. All patients completed SAS; 1337 also completed COMPASS-31 (334 with POTS). Models incorporating heart rate data achieved the highest diagnostic accuracy (PyTorch/LightGBM: AUC 0.98/0.99; precision 0.94/0.93; F1 score 0.88/0.93; sensitivity 83%/92%; specificity 98%/98%). In contrast, models trained solely on SAS (AUC 0.68/0.63) or COMPASS-31 (AUC 0.66/0.62) performed poorly.

Conclusion

ML models accurately diagnosed POTS when incorporating heart rate data alongside survey responses, with heart rate data being the strongest predictor. These findings suggest ML could assist in POTS diagnosis. The study also highlights the importance of heart rate measures in POTS diagnosis.