Machine Learning Models Can Predict Gestational Diabetes
Effective discrimination achieved with 73-variable model, seven-variable logistic regression model in early pregnancy
THURSDAY, Jan. 7, 2021 (HealthDay News) -- Machine learning (ML) models are highly accurate for predicting gestational diabetes mellitus (GDM) in early pregnancy in a Chinese population, according to a study published online Dec. 22 in the Journal of Clinical Endocrinology & Metabolism.
Yan-Ting Wu, Ph.D., from the Shanghai Jiao Tong University, and colleagues extracted data from 73 variables during the first trimester from electronic medical record systems. Seventeen variables were selected for early GDM prediction based on an ML-driven feature selection method; to facilitate clinical application, seven variables were selected. Using the seven- and 73-variable datasets, advanced ML approaches were used to build models to predict early GDM. The training and testing sets included 16,819 and 14,992 cases, respectively.
The researchers found that the deep neural network model achieved high discriminative power using 73 variables, with an area under the curve (AUC) of 0.80. Effective discriminative power was also achieved with the seven-variable logistic regression model (AUC, 0.77). Compared with a body mass index (BMI) of 17 to 18 kg/m², low BMI (≤17 kg/m²) was related to an increased risk for GDM. For predicting GDM, total triiodothyronine and total tetraiodothyronine were superior to free triiodothyronine and free tetraiodothyronine. A promising predictive value was also seen for lipoprotein(a) (AUC, 0.66).
"These findings can help clinicians identify women at high risk of diabetes in early pregnancy and start interventions such as diet changes sooner," a coauthor said in a statement.