Integrative Proteomic and Machine Learning Analysis Identifies Novel Predictors and Risk Model for Diabetic Macrovascular Complications

OBJECTIVE: Diabetic macrovascular complications continue to drive substantial morbidity, yet early detection tools and deeper mechanistic understanding remain scarce. This study aimed to pinpoint circulating protein biomarkers for diabetic macrovascular complications while elucidating their biological significance. METHODS: We combined proteome-wide Mendelian randomization (MR), Cox proportional hazards regression, and proteome-wide association study (PWAS) to rank priority proteins. From these prioritized proteins, we constructed a machine learning-derived protein risk score. Functional enrichment, multi-omics integration, and longitudinal trajectory modeling were conducted, supplemented by phenome-wide MR analyses. RESULTS: A total of 43 proteins were identified, clustering in pathways associated with immune-inflammatory cascades and extracellular matrix remodeling. The resulting 12-protein panel achieved robust discrimination (AUC = 0.793), maintained reliable performance over a 15-year period, and delivered clear risk stratification. Multi-omics integration revealed synchronized links to cardiometabolic dysregulation and cardiac structural alterations. Longitudinal trajectory analyses demonstrated that protein perturbations emerged as early as 10-12 years prior to clinical onset. Phenome-wide MR uncovered pleiotropic associations across various disease categories and provided causal support for several prioritized proteins. CONCLUSION: This work identifies a robust circulating protein panel suitable for early forecasting of diabetic macrovascular events and sheds light on core biological drivers of disease progression. The findings underscore the value of integrating proteomics, time-series evaluation, and causal inference for biomarker discovery and mechanistic insight.