Longitudinal trajectories of urinary albumin-to-creatinine ratio and risk of proteinuria among Chinese patients with type 2 diabetes: a single−center retrospective cohort study

Background Urinary albumin-to-creatinine ratio (UACR) is a key marker for monitoring proteinuria progression in type 2 diabetes mellitus (T2DM). However, UACR trajectory patterns and their association with proteinuria risk remain underexplored. Methods This retrospective cohort study included 3,101 T2DM patients (baseline UACR <30 mg/g) with regular follow-up from March 2018 to October 2024 at the Ningbo Metabolic Management Center (MMC) subcenter. Clinical data were obtained from electronic medical records. Group-based trajectory modeling (GBTM) identified UACR trajectory patterns. Partial Least Squares Discriminant Analysis (PLS-DA) with Boruta algorithm selected key variables associated with trajectories. Additionally, we used the Light Gradient Boosting Machine (LightGBM) algorithm for multi-class classification modeling and Shapley Additive exPlanations (SHAP) values to quantify individual feature contributions to distinct trajectories. Multivariable Cox regression evaluated proteinuria risk by trajectory. Results GBTM analysis identified three distinct UACR trajectories: low-normal, mid-range normal, and rising with fluctuation groups. PLS-DA with Boruta algorithm selected 11 significant features, including baseline UACR, sex, height, Hb, HCT, BMI, waist circumference, RBC, FCP, SCR, and FINS. Meanwhile, we explained the prediction results of a multi-class LightGBM model by assigning SHAP values to 11 features. Multivariable Cox regression showed significantly increased proteinuria risk in both mid-range normal (HR = 48.40, 95% CI: 14.67-159.67) and rising with fluctuation groups (HR = 509.56, 95% CI: 157.01–1653.70) compared to low-normal group. Conclusions Identifying distinct UACR trajectories in Chinese T2DM patients, particularly the strong association between rising with fluctuation pattern and proteinuria risk, indicates that trajectory-informed patient classification could enhance early-stage diabetic kidney disease screening and preventive strategies.