Objective Correlates of Activity Limitation in Korean Adults: A Combined Complex-Survey Regression and Explainable Machine-Learning Analysis of KNHANES 2019－2021

Abstract

Background/Objective: This study aimed to identify objective, potentially intervention-relevant correlates of activity limitation in Korean adults using complex-survey logistic regression and explainable machine learning.A secondary aim was to examine how inclusion of self-rated health (SRH), a conceptually overlapping global health measure, alters the interpretation of objective correlates.Design: Cross-sectional secondary analysis of nationally representative survey data.Methods: Data were obtained from 15,587 adults aged ≥19 years who participated in the Korea National Health and Nutrition Examination Survey 2019-2021.Activity limitation was defined as self-reported restriction in daily life or social activities due to health problems or physical or mental disability.Twenty-three predictors, including sociodemographic factors, musculoskeletal disorders, chronic diseases, health behaviors, perceived stress, and measured indices, were analyzed.SRH was excluded from the primary model and examined only in sensitivity analysis.Adjusted odds ratios were estimated using complex-survey logistic regression.Logistic regression, random forest, and XGBoost were compared using repeated stratified cross-validation, and variable contributions were interpreted using SHAP values.Results: The unweighted prevalence of activity limitation was 7.0%.Cross-validated AUC values ranged from 0.755 to 0.774.SHAP analysis identified employment, education, perceived stress, age, and arthritis as major contributors.Arthritis, high perceived stress, stroke, and osteoporosis were independently associated with activity limitation, whereas employment and higher education were protective.Adding SRH attenuated several objective disease-related associations.Conclusions: Arthritis and selected chronic conditions are objective correlates of activity limitation in Korean adults.SRH should be handled deliberately in explanatory models because it may obscure objective, potentially modifiable correlates.