# Loneliness severity and heart disease risk in older Chinese rural adults: a machine learning-based cross-sectional study

## TL;DR

Higher loneliness in older rural Chinese adults is linked to increased heart disease risk, with machine learning models showing strong predictive power.

## Contribution

This study is the first to explore loneliness as a predictor of heart disease in rural Chinese older adults using machine learning.

## Key findings

- Higher loneliness severity was associated with higher odds of heart disease, especially in rural areas.
- Logistic regression and gradient boosting decision tree models showed the best discrimination for predicting heart disease.
- Loneliness was identified as an independent predictor alongside dyslipidemia, age, sex, hypertension, and diabetes.

## Abstract

Loneliness is an important psychosocial determinant of cardiovascular health, yet its role in heart disease among older adults in rural China is underexplored.

Using 2020 China Health and Retirement Longitudinal Study (CHARLS) data, the analytic sample comprised 5767 rural residents aged ≥60 years. Loneliness was measured with a four-level 10-item Center for Epidemiologic Studies Depression Scale (CESD-10) item. Covariate-adjusted, residence-stratified logistic regressions assessed associations. Six machine learning classifiers (including logistic regression and gradient boosting decision tree) were trained with a stratified 70/30 split and evaluated by area under the receiver operating characteristic curve (ROC-AUC), area under the precision–recall curve (PR-AUC), accuracy, precision, recall, and F1; decision-curve analysis quantified net benefit. SHapley Additive exPlanations (SHAP) provided model interpretability.

A graded association was observed: higher loneliness severity corresponded to higher odds of heart disease, with stronger effects in rural than urban strata. Logistic regression and gradient boosting decision tree showed the best discrimination (ROC-AUC 0.753 and 0.750; PR-AUC 0.400 and 0.388). Decision-curve analysis indicated greater net benefit for these models across clinically relevant thresholds. SHAP ranked dyslipidemia, sex, age, hypertension, and diabetes as leading contributors, and identified loneliness as an independent, quantifiable predictor.

Loneliness is a salient correlate of self-reported heart disease in rural Chinese older adults. Incorporating brief loneliness screening into routine primary care may aid risk stratification in under-resourced settings. Interpretation should consider the cross-sectional design and self-reported outcomes; longitudinal studies with external validation and clinically adjudicated endpoints are warranted.