Scholarlink

Background <p>Advances in wearable and machine learning technologies have opened new possibilities for predicting cardiovascular (CV) risk. Moreover, the association between obstructive sleep apnea (OSA) and CV risk has been well recognized. However, how to integrate such information effectively to enhance risk prediction remains unclear. This study seeks to explore effective strategies for incorporating OSA phenotypic information and overnight physiological information for precise prediction of CV adverse outcome, as well as to identify the most significant features among a broad spectrum of risk factors in the general and OSA populations.</p> Methods <p>A total of 1,874 participants, aged between 54 and 94 years and without a history of CVDs from the MESA dataset, were included for predicting CV adverse outcomes within five years, including myocardial infarction, angina, heart failure and all-cause mortality. Four OSA phenotypes were first identified via K-means clustering based on static polysomnographic (PSG) features. A phenotype-contrastive (<i>Contrast_pheno_DL</i>) deep learning (DL) model that integrates OSA phenotypic information and deep representations of overnight sleep-event feature sequences, was proposed and compared with several baseline models. Feature importance analysis was also conducted for the <i>Contrast_pheno_DL</i> model by calculating SHapley Additive exPlanations (SHAP) values of all features across the four phenotypes to provide model interpretability.</p> Results <p>The proposed <i>Contrast_pheno_DL</i> model performed the best, with an area under the receiver operating characteristic (AUROC) of 0.877 (95% CI, 0.816–0.908) and an area under the precision-recall curve (AUPRC) of 0.689 (95% CI, 0.576–0.742), respectively (<i>p</i> &lt; 0.05). Moreover, the PSG and FOOD FREQUENCY features were recognized as the most significant CV risk factors across all the phenotypes, with each phenotype emphasizing unique features.</p> Conclusion <p>Models that are aware of OSA phenotypes are preferred, and lifestyle factors should be an important focus for precise CV prevention and risk management in the general and OSA populations.</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

Integrating phenotypic information of obstructive sleep apnea and deep representation of sleep-event sequences for cardiovascular risk prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月16日

查看全文
  • Yali Zheng,
  • Zhengbi Song,
  • Bo Cheng,
  • Xiao Peng,
  • Yu Huang,
  • Qing Liu,
  • Min Min

摘要

Background

Advances in wearable and machine learning technologies have opened new possibilities for predicting cardiovascular (CV) risk. Moreover, the association between obstructive sleep apnea (OSA) and CV risk has been well recognized. However, how to integrate such information effectively to enhance risk prediction remains unclear. This study seeks to explore effective strategies for incorporating OSA phenotypic information and overnight physiological information for precise prediction of CV adverse outcome, as well as to identify the most significant features among a broad spectrum of risk factors in the general and OSA populations.

Methods

A total of 1,874 participants, aged between 54 and 94 years and without a history of CVDs from the MESA dataset, were included for predicting CV adverse outcomes within five years, including myocardial infarction, angina, heart failure and all-cause mortality. Four OSA phenotypes were first identified via K-means clustering based on static polysomnographic (PSG) features. A phenotype-contrastive (Contrast_pheno_DL) deep learning (DL) model that integrates OSA phenotypic information and deep representations of overnight sleep-event feature sequences, was proposed and compared with several baseline models. Feature importance analysis was also conducted for the Contrast_pheno_DL model by calculating SHapley Additive exPlanations (SHAP) values of all features across the four phenotypes to provide model interpretability.

Results

The proposed Contrast_pheno_DL model performed the best, with an area under the receiver operating characteristic (AUROC) of 0.877 (95% CI, 0.816–0.908) and an area under the precision-recall curve (AUPRC) of 0.689 (95% CI, 0.576–0.742), respectively (p < 0.05). Moreover, the PSG and FOOD FREQUENCY features were recognized as the most significant CV risk factors across all the phenotypes, with each phenotype emphasizing unique features.

Conclusion

Models that are aware of OSA phenotypes are preferred, and lifestyle factors should be an important focus for precise CV prevention and risk management in the general and OSA populations.