Impact of statistical models on the prediction of type 2 diabetes using non-targeted metabolomics profiling.

•Metabolites Risk Scores improve the prediction of type 2 diabetes on top of clinical and biological risk factors in both high and low-risk sub-populations.•Two predictive metabolites (1,5-anhydroglucitol and Dehydroisoandrosterone sulfate) were well conserved over 9 years.•Comparing two statistical approaches revealed that lipid metabolism distinguishes baseline risk from that of fast converters.