Abstract
BACKGROUND: Metabolic syndrome (MetS) has been widely recognized as a risk factor for lung function. However, the evidence regarding the causal effect of MetS on lung function is limited, and it differs according to multidimensional individual characteristics. This study sought to investigate the causal effects and heterogeneity in the association between MetS and lung function through the development and validation of causal models. METHODS: This cohort study included adults from the China Health and Retirement Longitudinal Study (CHARLS) aged ≥ 45 years. We applied propensity score overlap weighting to balance baseline characteristics. The CausalForestDML model was used to estimate the causal and heterogeneous treatment effects, and SHapley Additive exPlanations analysis was implemented to explain the importance of features. Model evaluation was conducted using total operating characteristic (TOC) curves and QINI curves, and a heterogeneous analysis placebo test was conducted to verify the robustness of the model. RESULTS: Over the two years, 6,468 participants were included in our analysis, of which 4,498 (69.5%) had MetS. After applying overlap weighting, MetS exposure demonstrated a significant adverse causal effect on the peak expiratory flow as a percentage of predicted value (PEF%pred) (average treatment effect = -4.20, p < 0.001), and all participants exposed to MetS demonstrated individual treatment effects below zero. The body mass index (BMI), baseline PEF%pred, high-density lipoprotein cholesterol (HDL-C), and triglyceride glucose (TyG) index were the most influential factors. Subgroup analysis found that subgroup 4 (HDL-C ≤ 53.544 mg/dl, male, baseline PEF%pred ≥ 86.499) demonstrated the worst conditional average treatment effect (CATE) (-5.10). The CausalForestDML model demonstrates strong performance in distinguishing subgroups with varying treatment effects, and the placebo test also supports the causal interpretation (p < 0.05). CONCLUSIONS: Despite heterogeneity across individuals, the adverse causal effects of MetS exposure on lung function are universal. Preventing and managing MetS is essential for safeguarding lung function. Such causal machine learning models could evolve into clinically useful tools for personalized treatment decisions in MetS. TRIAL REGISTRATION: Not applicable.