Comparing interpretable machine learning models for fall risk in middle-aged and older adults with and without pain DOI Creative Commons

Shangmin Chen,

Yongshan Gao,

Lin Du

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 16, 2025

Pain is common in middle-aged and older adults, has also been identified as a fall risk factor, whereas the mechanism of falls pain unclear. This study included 13,074 adults from China health retirement longitudinal (wave 2011-2015) to separately develop four-year prediction models for with without pain, using five machine learning algorithms 145 input variables candidate features. Shapley Additive exPlanations (SHAP) was used model explanations. Adjusted logistic regression (LR) showed that (OR 1.40 [1.29, 1.53]) associated higher risk. Among characteristics, lower limb had highest 1.71 [1.22, 2.18]), followed by severe 1.53 [1.36, 1.73]) multisite 1.43 [1.28, 1.55]). non-pain, LR performed best AUC-ROC values 0.732 0.692, respectively. Common important features history height. Unique were functional limitation, SPPB, WBC, chronic disease score, life satisfaction, platelets, cooking fuel, quantity, while marital status, age, depressive symptoms, cognitive function, hearing, rainy days, tidiness, sleep duration exclusive non-pain model. characteristics are among adults. Prediction can help identify people at high pain. Important differ between populations, prevention strategies should target specific populations prediction.

Language: Английский

Comparing interpretable machine learning models for fall risk in middle-aged and older adults with and without pain DOI Creative Commons

Shangmin Chen,

Yongshan Gao,

Lin Du

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 16, 2025

Pain is common in middle-aged and older adults, has also been identified as a fall risk factor, whereas the mechanism of falls pain unclear. This study included 13,074 adults from China health retirement longitudinal (wave 2011-2015) to separately develop four-year prediction models for with without pain, using five machine learning algorithms 145 input variables candidate features. Shapley Additive exPlanations (SHAP) was used model explanations. Adjusted logistic regression (LR) showed that (OR 1.40 [1.29, 1.53]) associated higher risk. Among characteristics, lower limb had highest 1.71 [1.22, 2.18]), followed by severe 1.53 [1.36, 1.73]) multisite 1.43 [1.28, 1.55]). non-pain, LR performed best AUC-ROC values 0.732 0.692, respectively. Common important features history height. Unique were functional limitation, SPPB, WBC, chronic disease score, life satisfaction, platelets, cooking fuel, quantity, while marital status, age, depressive symptoms, cognitive function, hearing, rainy days, tidiness, sleep duration exclusive non-pain model. characteristics are among adults. Prediction can help identify people at high pain. Important differ between populations, prevention strategies should target specific populations prediction.

Language: Английский

Citations

0