Identification of age-specific risk factors for hyperuricemia: a machine learning-driven stratified analysis in...

Hyperuricemia (HUA) as a global public health challenge, although its overall epidemiological characteristics have been widely reported, its age-specific risk pattern remains controversial. This study aims to reveal the risk factors of HUA in healthy physical examination populations of different age groups and construct a machine learning-driven risk prediction model to achieve precise intervention.

A cross-sectional study design was adopted. A total of 2821 physical examinees from a tertiary hospital from January 2022 to December 2024 were included and divided into 5 groups according to age (Group I: 18-30 years old, n = 185;) Group II: 31-40 years old, n = 532; Group III: 41-50 years old, n = 753; Group IV: 51-60 years old, n = 714; Group V > 60 years old, n = 637. Sociodemographic and health behavior data were collected through electronic questionnaires. Univariate analysis and binary Logistic regression were used in SPSS 27.0 to screen for independent risk factors (P < 0.05). Then, logistic regression (LR), random forest (RF) and eXtreme Gradient Boosting (XGBoost) models were constructed using Python 3.8.2, and rank the feature importance of the optimal model.

The overall detection rate of HUA was 22.8%, and the level of serum uric acid increased significantly with age. There were significant differences in risk factors among different age groups. The risk factors were significantly different among different age groups: In group I, fasting blood glucose was an independent risk factor for HUA; In Group II, six factors such as BMI and alanine aminotransferase were included in the regression model. The logistic regression model (AUC = 0.840) showed that creatinine, triglycerides and serum total protein were the key predictors. In group III, HUA was significantly correlated with six factors such as gender and alcohol consumption. The ranking of the importance of characteristics by logistic regression (AUC = 0.766) was creatinine, and gender, alcohol consumption. In Group IV, four factors such as gender and alcohol consumption were included in the regression model. Logistic regression (AUC = 0.749) showed that creatinine, serum total protein, and alcohol consumptionwere the main predictive indicators. In Group V, the influencing factors such as educational level and dietary taste on HUA were prominent. The top three characteristics in terms of importance in logistic regression (AUC = 0.756) were creatinine, gender, and alcohol consumption.

The high HUA detection rate in the health examination population and the significant differences in clinical characteristics across different age groups were confirmed. Machine learning models helped to deeply explore risk factors, confirming the association between health behaviors and examination status, providing a reference for age-stratified HUA intervention.

Not applicable.