Identifying the comorbidity status of diabetes and hypertension among Chinese elderly population using both traditional statistical method and machine learning: a cross-sectional study

To explore the probability factors for diabetes comorbidity with hypertension among the elderly population using machine learning (ML) methods and to compare the performance of ML models with traditional statistical methods. This cross-sectional study included 4,704 older adults who underwent health examinations between 2019 and 2021 at a tertiary hospital in Chengdu, Sichuan Province, China. Sociodemographic data, dietary habits, lifestyle factors, chronic disease diagnoses, and family history were assessed. Participants were divided into three groups: no diabetes (No DM), diabetes without hypertension (DM no HTN), and diabetes with hypertension (DM with HTN). We used four ML methods (random forest, LASSO, neural network, ridge regression) to build three classifier models: task 1 (DM no HTN vs. No DM), task 2 (DM with HTN vs. No DM), and task 3 (DM with HTN vs. DM no HTN). Traditional logistic regression models were also performed for comparison. Machine learning (ML) models achieved numerically higher AUC values compared to traditional logistic regression in classification tasks, though these differences did not reach statistical significance. Neural networks achieved the best performance, with an AUC of 0.790 (95% CI: 0.707–0.873), in distinguishing individuals with no diabetes mellitus (DM) from those with DM but no hypertension (HTN). LASSO regression demonstrated the best performance, with an AUC of 0.820 (95% CI: 0.757–0.882), in distinguishing individuals with no DM from those with DM and HTN. Random forests achieved the best performance, with an AUC of 0.783 (95% CI: 0.700-0.867), in distinguishing individuals with DM but no HTN from those with DM and HTN. Coronary heart disease, excessive intake of meat and fat, and age over 75 years were important features in task (1) Insufficient intake of vegetables and fruits, coronary heart disease, and sedentariness were important features in task (2) Excessive intake of meat and fat, family history of hypertension, and insufficient consumption of whole grains were important features in task 3. This study is the first to predict different statuses of hypertension comorbidity in Chinese elderly with diabetes using ML methods. Our findings provide valuable insights for preventing comorbidity in elderly population with diabetes.