NHANES-Derived Machine Learning Model for Early Identification of Frailty Risk in CKM: Optimizing Resource Allocation...

Area of Science:

• Gerontology
• Cardiovascular Health
• Metabolic Health
• Nephrology

Background:

• Cardiorenal metabolic (CKM) dysfunction presents a complex health challenge.
• Frailty is a significant concern in older adults, linked to adverse health outcomes.
• Existing frailty risk prediction tools are lacking for middle-aged and older CKM patients (Stages 0-3).

Purpose of the Study:

• Develop a machine learning model to predict frailty risk in middle-aged and older adults with CKM Stages 0-3.
• Identify key predictive factors for frailty in this specific patient population.

Main Methods:

• Utilized NHANES data (2007-2018) from 6749 participants aged 40+.
• Employed LASSO regression for feature selection and compared five machine learning models (LR, LightGBM, XGBoost, RF, DT).
• Evaluated model performance using 10-fold cross-validation and SHAP values for feature importance.

Main Results:

• The Random Forest (RF) model demonstrated superior performance with an AUC of 0.90.
• Key predictors identified include poverty-income ratio (PIR), antihypertensive medication use, hemoglobin, red blood cell count, albumin, and diabetes.
• The model showed strong predictive accuracy (0.81) and recall (0.84).

Conclusions:

• The developed machine learning model accurately predicts frailty risk in middle-aged and older adults with CKM Stages 0-3.
• This tool facilitates early identification of high-risk individuals for improved clinical management.
• The findings support optimized healthcare resource allocation and targeted interventions for CKM patients.