基于加速度计活动行为表型与心血管疾病风险的关联研究:衰弱状态的影响及代谢组学特征OA

Objective This study aims to investigate the associations among accelerometer-derived activity phenotypes identified via unsupervised machine learning,frailty,and cardiovascular disease risk.Methods A total of 87 824 participants were enrolled from a large-scale European cohort study.Accelerometer-derived physical activity indicators were extracted,including moderate-to-vigorous physical activity(MVPA),light physical activity(LPA),sedentary time and overall activity intensity.Gaussian mixture modeling was adopted to identify activity behavior phenotypes.Cox proportional hazards regression models were constructed to analyze the associations between the identified phenotypes and cardiovascular disease risk,and LASSO regression was utilized to screen key metabolites associated with the activity phenotypes.Results Three distinct physical activity behavior phenotypes were identified in this study.Over the follow-up period,4 832 cardiovascular disease events were documented.Phenotype A was characterized by relatively low MVPA,prolonged sedentary time,and lower overall activity inten-sity;Phenotype B exhibited an intermediate activity level;and Phenotype C was characterized by relatively high MVPA and LPA,shorter sedentary time,and higher overall activity intensity.Compared with Phenotype A,Pheno-types B and C were associated with a significantly lower risk of cardiovascular disease,with hazard ratios of 0.80 and 0.74 respectively,and both comparisons reached statistical significance(P<0.001).This association remained consistent among non-frail and pre-frail participants,and no significant interaction was detected between physical activity behavior phenotype and frailty status.LASSO regression identified 26 core metabolites,which revealed that Phenotype A was mainly characterized by inflammation-and energy metabolism-related abnormalities,while Pheno-type C presented relatively elevated levels of fatty acids and HDL-related lipid components.Conclusions Accelerometer-derived physical activity behavioral phenotypes identified via unsupervised machine learning are independently associated with cardiovascular disease risk and are characterized by distinct metabolic profiles.These data-driven phenotypes can provide additional information for cardiovascular disease risk stratification beyond the scope of traditional frailty assessments.