A recent study has highlighted the effectiveness of a deep learning model, specifically a multi-layer perceptron (MLP), in predicting the risk of gestational diabetes mellitus (GDM). The model, which surpasses traditional methods like logistic regression, offers promising potential for early intervention and improved patient outcomes.
Gestational diabetes affects 5% to 30% of pregnant women, with an increasing incidence among younger populations. The condition, which varies by region and ethnicity, poses significant risks for both mothers and infants, making it a key focus in public health.
Despite the proven advantages of deep learning (DL) in handling complex interactions and non-linear relationships between data points, there has been limited research comparing DL methods with traditional approaches in the context of GDM prediction. The study, published in Gynecological Endocrinology, sought to address this gap by evaluating the efficacy of DL models against logistic regression.
Researchers conducted a retrospective cohort study, analyzing prenatal health data from women who gave birth between 2008 and 2018. The cohort excluded women with pregestational diabetes, missing data, or abnormal health indicators.
The study followed a three-stage process. The first step identified risk factors for GDM, while the second dealt with data imbalances. The third stage involved the development of a classification model for risk prediction.
Initially, a multivariate logistic regression and nomogram model were used to create a baseline prediction. Subsequently, a neural network model was trained to predict GDM risk more accurately. This model included three layers: an initial linear layer, followed by a nonlinear rectified linear unit activation function, and a second linear layer with 64 hidden neurons. The output layer, consisting of two neurons, generated a risk score between 0 and 1.
Of the 42 variables assessed, 32 were included in the final model, with 7 showing significant differences between women with and without GDM. Key risk factors identified included history of hypertension, family history of diabetes, age, education level, body mass index (BMI), and folic acid supplementation.
Women diagnosed with GDM were more likely to be older, have a higher BMI, a lower education level, and took folic acid supplements. Differences in hematological, renal, and liver function indicators were also noted, such as higher hemoglobin (HGB), white blood cell (WBC) count, and platelet (PLT) levels in the GDM group.
In assessing the model’s performance, metrics such as the area under the receiver operating characteristic curve (auROC), average precision (auPR), and F1 score were used. The MLP model achieved impressive results with an auROC of 0.943, auPR of 0.855, and F1 score of 0.879, compared to the baseline model’s auROC of 0.774, auPR of 0.272, and F1 score of 0.377.
The researchers concluded that integrating genetic testing into future studies could further enhance predictive accuracy by identifying susceptibility genes and tailoring preventive strategies based on individual genetic profiles.
This study suggests that deep learning models could play a pivotal role in improving the early prediction and management of gestational diabetes, ultimately benefiting both maternal and infant health.
Related topics:
High Risk of Gestational Diabetes Among Pregnant Prisoners in England
Red Ginseng Extract Shows Potential in Managing Blood Sugar Levels, Study Finds
Leisure-Time Physical Activity Linked to Lower Risk of Type 2 Diabetes
