Predicting diabetic kidney disease with serum metabolomics and gut microbiota

), revealing 23 differential metabolites. Dysbiosis of the gut microbiota was evident in DKD patients, with lower relative abundances of g_Prevotella and g_Faecalibacterium compared to the DM and NC groups. Subgroup analysis indicated decreased relative abundances of g_Prevotella and g_Faecalibacterium in the DKD middle group, along with a decrease in g_Klebsiella compared to the DKD early group, which correlated positively with DKD patients' eGFR. There were 11 common metabolites among the three groups of differential metabolites. Among these, three serum metabolites-imidazolepropionic acid, adipoylcarnitine, and 1-methylhistidine-were identified as predictive serum metabolic markers. Disease prediction models (logistic regression models) were constructed based on these three metabolites combined with three genera of bacteria. These models demonstrated strong discriminatory power for diagnosing patients with DKD compared to patients with DM (area under the receiver operating characteristic curve (AUROC) = 0.939 and precision-recall curve (AUPR) = 0.940). The models also effectively discriminated between patients with DKD and NCs (0.976, 0.973). This study revealed distinctive serum metabolites and gut microbiota in patients with DKD. It demonstrated the potential utility of three specific serum metabolites and three genera of bacteria in diagnosing patients with DKD and assessing their renal dysfunction.