Scholarlink

Purpose <p>Obstructive sleep apnoea syndrome (OSA) is a common yet underdiagnosed condition associated with significant health risks. Although polysomnography is the diagnostic gold standard, it is resource-intensive and unsuitable for widespread screening. Heart rate variability (HRV) derived from electrocardiogram (ECG) recordings has emerged as a promising, accessible alternative for OSA detection. Recent developments in machine learning have enabled automated HRV analysis, potentially offering a scalable screening tool for OSA. This study aimed to evaluate the diagnostic accuracy of machine learning-based models trained on HRV for detecting OSA in adults.</p> Methods <p>We searched PubMed, Embase, Scopus, Web of Science, and IEEE Xplore (up to 20 October 2024) for eligible studies that assessed the accuracy of OSA diagnosis using AI models trained on HRV, compared to the apnoea-hypopnea index (AHI). Bayesian bivariate random-effects meta-analysis estimated pooled sensitivity and specificity. Risk of bias was assessed using QUADAS-2, and GRADE was used to rate evidence certainty.</p> Results <p>Nine studies with 2,019 participants met inclusion criteria. Pooled sensitivity was 79.0% (95% CrI: 74.9%–82.7%) and specificity was 75.0% (95% CrI: 67.9%–82.3%). The diagnostic odds ratio was 11.3 (95% CrI: 7.21–19.0%). Meta-regression showed specificity varied with demographic factors, while model architecture and validation methods had no significant impact. No publication bias was detected.</p> Conclusions <p>Machine learning models trained on HRV show good diagnostic accuracy for OSA, with higher specificity than STOP-BANG and comparable performance to home sleep tests. Their scalability and potential integration into wearable devices offer a practical, cost-effective screening option. Further real-world validation is warranted.</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

Electrocardiogram heart rate variability for machine learning diagnosis of obstructive sleep apnoea: A Bayesian meta-analysis

  • 期刊论文

  • 发表日期：

    2025年9月30日

查看全文
  • Yunrui Hao,
  • Nicole Kye Wen Tan,
  • Esther Yanxin Gao,
  • Joy Xin Yi Au,
  • Novelle En Xian Toh,
  • Cai Ling Yong,
  • Yao Hao Teo,
  • Adele Chin Wei Ng,
  • Zhou Hao Leong,
  • Chu Qin Phua,
  • Thun How Ong,
  • Leong Chai Leow,
  • Guang-Bin Huang,
  • Benjamin Kye Jyn Tan,
  • Song Tar Toh

摘要

Purpose

Obstructive sleep apnoea syndrome (OSA) is a common yet underdiagnosed condition associated with significant health risks. Although polysomnography is the diagnostic gold standard, it is resource-intensive and unsuitable for widespread screening. Heart rate variability (HRV) derived from electrocardiogram (ECG) recordings has emerged as a promising, accessible alternative for OSA detection. Recent developments in machine learning have enabled automated HRV analysis, potentially offering a scalable screening tool for OSA. This study aimed to evaluate the diagnostic accuracy of machine learning-based models trained on HRV for detecting OSA in adults.

Methods

We searched PubMed, Embase, Scopus, Web of Science, and IEEE Xplore (up to 20 October 2024) for eligible studies that assessed the accuracy of OSA diagnosis using AI models trained on HRV, compared to the apnoea-hypopnea index (AHI). Bayesian bivariate random-effects meta-analysis estimated pooled sensitivity and specificity. Risk of bias was assessed using QUADAS-2, and GRADE was used to rate evidence certainty.

Results

Nine studies with 2,019 participants met inclusion criteria. Pooled sensitivity was 79.0% (95% CrI: 74.9%–82.7%) and specificity was 75.0% (95% CrI: 67.9%–82.3%). The diagnostic odds ratio was 11.3 (95% CrI: 7.21–19.0%). Meta-regression showed specificity varied with demographic factors, while model architecture and validation methods had no significant impact. No publication bias was detected.

Conclusions

Machine learning models trained on HRV show good diagnostic accuracy for OSA, with higher specificity than STOP-BANG and comparable performance to home sleep tests. Their scalability and potential integration into wearable devices offer a practical, cost-effective screening option. Further real-world validation is warranted.