To develop and validate a risk assessment model for kidney damage in community-dwelling patients with essential hypertension.

A total of 2 338 patients with essential hypertension from four community health service centers affiliated with Beijing University of Chinese Medicine Shenzhen Hospital (Longgang) were retrospectively enrolled between January and December 2025. They were randomly divided into a training set (n=1 637) and a validation set (n=701) in a 7:3 ratio. Relevant clinical data were collected. After variable screening using univariate analysis, LASSO regression, and multivariable regression analysis, risk assessment models were constructed using seven machine learning algorithms. Model performance was evaluated on the validation set using metrics including accuracy and the area under the receiver operating characteristic curve(AUC). The importance of individual variables was assessed using SHapley additive exPlanations (SHAP)values.

Univariate analysis revealed that 28 variables, including age and sex, were associated with kidney damage (P<0.05). Lasso regression screened and retained seven feature variables. Multivariable logistic regression confirmed that male sex(OR=0.168, 95% CI: 0.113-0.250), age (OR=1.155, 95% CI: 1.133-1.178), history of alcohol consumption (OR=1.604, 95%CI: 1.415-2.879), urine occult blood (OR=1.923, 95% CI: 1.230-3.010), red blood cell count (OR=0.702, 95% CI: 0.508-0.968), and uric acid (OR=1.01, 95% CI: 1.008-1.012) were independent risk factors for kidney damage (P<0.05). The AUC values for all models exceeded 0.80. The support vector machine (SVM) demonstrated relatively better performance (AUC=0.85), with balanced performance across various metrics. SHAP value analysis indicated that age, sex, and uric acid level were key features influencing model predictions, and nonlinear interaction effects were observed among the variables.

Male sex, age, history of alcohol consumption, urine occult blood, red blood cell count, and uric acid are independent risk factors for kidney damage in community-dwelling patients with essential hypertension. The SVM model constructed using these variables exhibits good predictive performance and holds potential clinical application value.