Predictive model of sleep disorders in pregnant women using machine learning and SHAP analysis

Sleep disorders are common among pregnant women and can adversely affect maternal and infant health. Traditional statistical methods have limitations in predicting these disorders, highlighting the need for advanced machine learning (ML) approaches. This study aimed to develop a reliable ML model for early prediction of pregnancy-related sleep disorders.

Data from 1,681 pregnant women in western China were analyzed. Logistic regression and LASSO regression identified key predictors, with 10 variables selected for Model training. Eight ML algorithms were evaluated using 5-fold cross-validation. SHAP analysis interpreted the model's decisions.

Ten predictors were identified: age, standardized gestational weight gain, gestational weeks, severity of morning sickness, pregnancy intention, pre-pregnancy health, underlying diseases, anxiety, depression, and the combined effect of anxiety and depression. LightGBM achieved the highest AUC (0.718) in the test set, with accuracy of 0.670 and specificity of 0.764. SHAP analysis revealed depression as the strongest predictor (mean |SHAP|=0.26), followed by gestational weeks and Std. GWG.

The interpretable and accurate LightGBM model, using clinically feasible variables, is a practical tool for early identification of pregnant women at high risk of sleep disorders. It enables targeted interventions to mitigate sleep - related adverse outcomes, thus improving maternal and infant health.