Metabolomic identification of a novel, externally validated predictive test for gestational diabetes mellitus.

Abstract

CONTEXT: Undiagnosed gestational diabetes (GDM) is a major preventable cause of stillbirth. In the UK, women are selected for diagnostic testing for GDM based on risk factors, including BMI >30kg/m 2. OBJECTIVE: To improve the prediction of GDM using metabolomics. METHODS: We performed metabolomics on maternal serum from the Pregnancy Outcome Prediction (POP) study at 12 and 20 weeks of gestational age (wkGA, 185 GDM cases and 314 non-cases). Predictive metabolites were internally validated using the 28wkGA POP study serum sample and externally validated using 24-28wkGA fasting plasma from the Born in Bradford (BiB) cohort (349 GDM cases and 2347 non-cases). The predictive ability of a model including the metabolites was compared with BMI >30kg/m 2 in the POP study. RESULTS: 47 predictive metabolites were identified using the 12 and 20wkGA samples. At 28wkGA, four of these (mannose, 4-hydroxyglutamate, 1,5-anhydroglucitol and lactosyl-N-palmitoyl-sphingosine (d18:1/16:0)) independently increased the bootstrapped area under the ROC curve (AUC) by >0.01. All four were externally validated in the BiB samples (P=2.6x10 -12, 2.2x10 -13, 6.9x10 -28 and 2.6x10 -17, respectively). In the POP study, BMI >30kg/m 2 had a sensitivity of 28.7% (95% CI 22.3 to 36.0%) and a specificity of 85.4% whereas at the same level of specificity, a predictive model using age, BMI and the four metabolites had a sensitivity of 60.2% (95% CI 52.6 to 67.4%) and an AUC of 0.82 (95% CI 0.78 to 0.86). CONCLUSIONS: We identified four strongly and independently predictive metabolites for GDM which could have clinical utility in screening for GDM.