保留射血分数的心力衰竭合并阻塞性睡眠呼吸暂停患者急性冠脉综合征长期风险预测的多变量机器学习模型

文献信息

中文摘要

背景： 保留射血分数的心力衰竭是一种异质性综合征，合并阻塞性睡眠呼吸暂停会进一步增加急性冠脉综合征的风险。然而，针对该人群进行长期急性冠脉综合征风险分层的有效工具仍然有限。本研究旨在开发并外部验证一种基于机器学习的预后模型，用于预测保留射血分数的心力衰竭合并阻塞性睡眠呼吸暂停患者在多个时间点的急性冠脉综合征风险。
方法： 我们回顾性纳入了来自2家三级医院的2272例保留射血分数的心力衰竭合并阻塞性睡眠呼吸暂停患者。采用最小绝对收缩和选择算子回归及Boruta算法，从临床特征、睡眠监测、实验室检查和超声心动图中筛选候选预测因子。共开发了33种生存模型，并根据一致性指数选择最优模型并进行外部验证。
结果： 共识别出10个核心预测因子，包括代谢、睡眠相关和心脏参数。加速斜向随机生存森林模型在所有评估模型中取得了最佳区分度，在外部验证队列中一致性指数为0.731。该模型表现出良好的时间依赖性区分能力，在20、30和40个月时的曲线下面积值分别为0.742、0.734和0.755。校准分析和决策曲线分析支持其可靠性和临床实用性。
结论： 我们开发并外部验证了一种基于机器学习的模型，用于保留射血分数的心力衰竭合并阻塞性睡眠呼吸暂停患者的多时间点急性冠脉综合征风险预测。该加速斜向随机生存森林模型以网络计算器形式实现，可能有助于个体化风险评估和临床决策。
注册： 网址：https://www.chictr.org.cn/；唯一标识符：ChiCTR2300075727。

英文摘要

BACKGROUND: Heart failure with preserved ejection fraction is a heterogeneous syndrome, and comorbid obstructive sleep apnea further increases the risk of acute coronary syndrome. However, effective tools for long-term acute coronary syndrome risk stratification in this population remain limited. This study aimed to develop and externally validate a machine learning-based prognostic model for predicting acute coronary syndrome risk at multiple time points in patients with heart failure with preserved ejection fraction and obstructive sleep apnea.
METHODS: We retrospectively enrolled 2272 patients with heart failure with preserved ejection fraction and obstructive sleep apnea from 2 tertiary hospitals. Candidate predictors from clinical characteristics, sleep monitoring, laboratory tests, and echocardiography were selected using least absolute shrinkage and selection operator regression and the Boruta algorithm. Thirty-three survival models were developed, and the optimal model was selected based on the concordance index and externally validated.
RESULTS: Ten core predictors were identified, including metabolic, sleep-related, and cardiac parameters. The accelerated oblique random survival forest model achieved the best discrimination among all evaluated models, with a concordance index of 0.731 in the external validation cohort. The model demonstrated good time-dependent discrimination, with area under the curve values of 0.742, 0.734, and 0.755 at 20, 30, and 40 months, respectively. Calibration analyses and decision curve analysis supported its reliability and clinical utility.
CONCLUSIONS: We developed and externally validated a machine learning-based model for multitime-point acute coronary syndrome risk prediction in patients with heart failure with preserved ejection fraction and obstructive sleep apnea. The accelerated oblique random survival forest model, implemented as a web-based calculator, may facilitate individualized risk assessment and clinical decision-making.
REGISTRATION: URL: https://www.chictr.org.cn/; Unique Identifier: ChiCTR2300075727.