Blood pressure phenotypes and kidney impairment in adults with diabetes: a machine learning study

Background Diabetic kidney disease (DKD) is a leading cause of end-stage renal disease, with hypertension being a key modifiable risk factor. However, associations between specific blood pressure (BP) phenotypes and concurrent DKD risk remain poorly understood. Methods This cross-sectional study analyzed 1,240 adults with diabetes from community health records in Shanghai. Concurrent DKD was defined by UACR ≥30 mg/g or eGFR <60 mL/min/1.73 m 2 . BP phenotypes (140/90 mmHg threshold) were classified as normal (NBP), isolated systolic (ISH), isolated diastolic (IDH), and systolic-diastolic hypertension (SDH). Multivariable logistic regression (LR) estimated adjusted odds ratios (aORs). The cohort was split 80:20 into training (n = 992) and test (n = 248) sets. Five machine learning algorithms were evaluated using strict within-training-fold preprocessing. Results Concurrent DKD was present in 331/1,240 participants (26.69%). After adjustment, ISH and SDH were significantly associated with higher DKD odds versus NBP. On the independent test set, ROC-AUCs were 0.776 for XGBoost and 0.771 for LR, with no statistically significant discrimination difference. XGBoost showed a slightly lower Brier score, while LR offered superior calibration stability and nomogram interpretability. Conclusion ISH and SDH are associated with a higher prevalence of concurrent DKD in adults with diabetes. Routine clinical features and BP phenotypes provide moderate discrimination. These models function for opportunistic concurrent risk pattern recognition rather than predictive screening, supporting further external validation prior to clinical implementation.