Identifying early-life, environmental, and social stressors as key predictors of U.S. respiratory failure mortality: a machine learning study.

Background: Traditional risk factors like smoking and air pollution remain insufficient to explain the persistent geographical variation in respiratory failure mortality across the United States, with significantly higher mortality observed in specific regions such as the Mississippi River Basin. Methods: We conducted a state-level ecological study using Centers for Disease Control and Prevention Wide-ranging Online Data for Epidemiologic Research (CDC WONDER) mortality data and 68 socio-environmental risk factors from Global Burden of Disease (GBD). Predictive modeling leveraged 12 machine learning algorithms (from linear to tree-based ensembles) with Recursive Feature Elimination (RFE) and SHapley Additive exPlanations (SHAP) interpretation. Results: The optimal model Gradient Boosting Machine (GBM) achieved high accuracy (Root Mean Square Error (RMSE) = 0.7447, R-squared (R Conclusion: These findings suggest an "early-life and cumulative exposure" framework as a hypothesis-generating model. This framework identifies early-life vulnerability, environmental factors, and social stress as key state-level predictors that warrant further investigation. Our findings suggest a potential shift in research focus, moving from a reactive focus on adult-life risk factors toward proactive interventions targeting early-life vulnerabilities and cumulative environmental and social exposures.