Risk prediction for type 2 diabetes mellitus (T2DM) often relies on clinical scoring systems, which may not fully capture underlying biological variation. A cohort analysis published in Diabetic Medicine evaluated whether integrating multi-omics data could improve risk stratification beyond established clinical and genetic approaches.

The study used UK Biobank data and included 21,312 participants, among whom 893 developed incident T2DM. Ridge Cox regression models were constructed using clinical risk factors based on the Finnish Diabetes Risk Score (CliS), plasma metabolomics (MetS), proteomics (ProS), and a polygenic risk score (PRS). Model performance was assessed using the C-index and net reclassification improvement (NRI). External validation of protein markers was conducted in the Liyang Cohort (n=10,056).

Among individual models, ProS showed the highest predictive performance with a C-index of 0.80. The combined multi-omics score (ComS) achieved the strongest discrimination with a C-index of 0.84, compared with 0.76 for CliS, and showed significant improvement in classification (NRI 0.328; p<0.001). The ComS reclassified 73 additional T2DM cases per 1000 individuals without increasing false positives. Kaplan-Meier analysis demonstrated improved risk separation. Both MetS and ProS identified high-risk individuals not captured by CliS. In validation, plasma fibroblast growth factor 23 levels were higher in T2DM cases.

Current screening approaches rely largely on clinical risk scores. These findings indicate that integrating multi-omics data may improve early identification of individuals at higher risk for T2DM and support more tailored screening strategies.