基于机器学习的二尖瓣脱垂表型分组识别与心肌纤维化及心血管事件相关的特征
本研究在429例二尖瓣脱垂患者中利用超声心动图变量进行无监督聚类,识别出4个表型组,其中第3和第4组心肌纤维化更多且心血管事件率更高;决策树确定二尖瓣反流严重程度、左室收缩应变<21%和左房容积指数>42 mL/m2为分类关键变量,聚类分析较传统分析显著提高诊断准确性。
要点
- 二尖瓣脱垂患者可聚类为4个超声表型组,第3组为显著左室和左房重构伴重度二尖瓣反流,第4组为重构伴左室收缩应变下降。
- 第3和第4组心肌纤维化多于第1和第2组(P < 0.0001),且心血管事件率更高。
- 决策树识别出二尖瓣反流严重程度、左室收缩应变<21%和左房容积指数>42 mL/m2为分类最关键变量。
核心结论
- 聚类分析识别出4个具有不同超声左室和左房重构特征的组,这些组与心肌纤维化和临床结局相关;基于二尖瓣反流严重程度、左室收缩应变和左房容积指数3个关键变量的简单算法可能有助于二尖瓣脱垂患者的风险分层和决策。
研究设计与样本
- 双中心队列研究,纳入429例二尖瓣脱垂患者(年龄54 ± 15岁),使用超声心动图变量构建聚类,并与心脏磁共振评估的心肌纤维化和心血管结局进行关联分析。
诊断标准与关键数值
- 聚类分析较传统分析显著提高诊断准确性;决策树识别二尖瓣反流严重程度、左室收缩应变<21%和左房容积指数>42 mL/m2为正确分类的最相关3个变量。
临床应用与阅片要点
- 作者认为基于二尖瓣反流严重程度、左室收缩应变和左房容积指数3个关键变量的简单算法可能有助于二尖瓣脱垂患者的风险分层和决策。
局限性
- 原文未报告局限性
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
BACKGROUND: Structural changes and myocardial fibrosis quantification by cardiac imaging have become increasingly important to predict cardiovascular events in patients with mitral valve prolapse (MVP). In this setting, it is likely that an unsupervised approach using machine learning may improve their risk assessment.
OBJECTIVES: This study used machine learning to improve the risk assessment of patients with MVP by identifying echocardiographic phenotypes and their respective association with myocardial fibrosis and prognosis.
METHODS: Clusters were constructed using echocardiographic variables in a bicentric cohort of patients with MVP (n = 429, age 54 ± 15 years) and subsequently investigated for their association with myocardial fibrosis (assessed by cardiac magnetic resonance) and cardiovascular outcomes.
RESULTS: Mitral regurgitation (MR) was severe in 195 (45%) patients. Four clusters were identified: cluster 1 comprised no remodeling with mainly mild MR, cluster 2 was a transitional cluster, cluster 3 included significant left ventricular (LV) and left atrial (LA) remodeling with severe MR, and cluster 4 included remodeling with a drop in LV systolic strain. Clusters 3 and 4 featured more myocardial fibrosis than clusters 1 and 2 (P < 0.0001) and were associated with higher rates of cardiovascular events. Cluster analysis significantly improved diagnostic accuracy over conventional analysis. The decision tree identified the severity of MR along with LV systolic strain <21% and indexed LA volume >42 mL/m2 as the 3 most relevant variables to correctly classify participants into 1 of the echocardiographic profiles.
CONCLUSIONS: Clustering enabled the identification of 4 clusters with distinct echocardiographic LV and LA remodeling profiles associated with myocardial fibrosis and clinical outcomes. Our findings suggest that a simple algorithm based on only 3 key variables (severity of MR, LV systolic strain, and indexed LA volume) may help risk stratification and decision making in patients with MVP. (Genetic and Phenotypic Characteristics of Mitral Valve Prolapse, NCT03884426; Myocardial Characterization of Arrhythmogenic Mitral Valve Prolapse [MVP STAMP], NCT02879825).
引用
- Huttin O, Girerd N, Jobbe-Duval A, et al. Machine Learning-Based Phenogrouping in MVP Identifies Profiles Associated With Myocardial Fibrosis and Cardiovascular Events. JACC Cardiovasc Imaging. 2023. PMID: 37204382. doi:10.1016/j.jcmg.2023.03.009
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。