A Machine Learning–Derived Risk Score Based on Dietary Nutrient Intake for Early Detection and Prognostic Prediction of Preserved Ratio Impaired Spirometry

Background: Preserved Ratio Impaired Spirometry (PRISm) is a subclinical pulmonary phenotype associated with increased risk of chronic obstructive pulmonary disease (COPD), cardiovascular disease, and all-cause mortality. Early identification and stratified prevention of PRISm remain a clinical challenge. Methods: Using data from the US National Health and Nutrition Examination Survey (NHANES) 2007– 2012, we developed and validated a stacked machine learning (ML) model integrating dietary intake and demographic features to generate a continuous PRISm risk score. The dataset was split into training, validation, and test sets. Model performance was evaluated using ROC curves and calibration. The associations between the risk score and adverse health outcomes were assessed using logistic regression and Kaplan–Meier analysis. Subgroup analysis was performed to assess the impact of lifestyle across risk strata. Results: The stacked ML model demonstrated strong predictive ability, achieving an AUC of 0.818 in the test set. The risk score was significantly associated with multiple chronic conditions, including hypertension, diabetes, cardiovascular disease, and COPD. High-risk individuals had substantially increased mortality rates compared to the low-risk group. In the low-risk group, adherence to a healthy lifestyle was associated with significantly lower odds of adverse outcomes, while no such association was observed in the high-risk group. Conclusion: This study presents a non-invasive, data-driven model for PRISm risk prediction and health outcome stratification based on dietary and demographic features. The PRISm risk score may aid early screening and inform personalized prevention strategies. Keywords: preserved ratio impaired spirometry, machine learning, dietary intake, prognosis, stratification