Prediction of gestational diabetes mellitus in the first 19 weeks of pregnancy using machine learning techniques.

AIM: Our objective was to develop a first 19 weeks risk prediction model with several potential gestational diabetes mellitus (GDM) predictors including hepatic and renal and coagulation function measures.
METHODS: A total of 490 pregnant women, 215 with GDM and 275 controls, participated in this case-control study. Forty-three blood examination indexes including blood routine, hepatic and renal function, and coagulation function were obtained. Support vector machine (SVM) and light gradient boosting machine (lightGBM) were applied to estimate possible associations with GDM and build the predict model. Cutoff points were estimated using receiver operating characteristic curve analysis.
RESULTS: It was observed that a cutoff of Prothrombin time (PAT-PT) and Activated partial thromboplastin time (PAT-APTT) could reliably predict GDM with sensitivity of 88.3% and specificity of 99.47% (AUC of 94.2%). If we only use hepatic and renal function examination, a cutoff of DBIL and FPG with sensitivity of 82.6% and specificity of 90.0% (AUC of 91.0%) was obvious and a negative correlation with PAT-PT (r=-0.430549) and patient activated partial thromboplastin time (PAT-APTT) (r=-0.725638). A negative correlation with direct bilirubin (DBIL) (r=-0.379882) and positive correlation with fasting plasma glucose (FPG) (r = 0.458332) neglect coagulation function examination.
CONCLUSION: The results of this study point out the possible roles of PAT-PT and PAT-APTT as potential novel biomarkers for the prediction and earlier diagnosis of GDM. A first 19 weeks risk prediction model, which incorporates novel biomarkers, accurately identifies women at high risk of GDM, and relevant measures can be applied early to achieve the prevention and control effects.