Development of prediction model of problematic social media use in pregnant women using machine learning

Background: Problematic social media use (PSMU) has emerged as a significant addictive behavior, influenced by a complex interplay of sociodemographic, behavioral, and psychosocial factors. Few explored PSMU in pregnant women, though women could be more vulnerable for PSMU. This study aimed to identify key factors and develop a machine learning-based prediction model for classifying pregnant women with PSMU.

Methods: A total of 945 pregnant women (median age 30 years) were recruited in a tertiary hospital in Shanghai, China. PSMU was measured using the Bergen Social Media Addiction Scale, with scores ?19 indicating PSMU. Potential factors were selected based on the literature, including sociodemographic and pregnancy- related characteristics, psychological status (prenatal depression, prenatal anxiety, resilience, Big Five personality traits), screen time, partner phubbing behavior, family communication level, family well-being, and positive and adverse childhood experiences.The dataset was randomly split into a training set (80%, n = 756) and a testing set (20%, n = 189), with no significant differences. Feature selection was conducted using Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation. The factors selected by LASSO were subsequently used to construct multiple machine learning models, including logistic regression, support vector machine (SVM), decision tree, AdaBoost, and extreme gradient boosting (XGBoost). Model performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, predictive values, and F1 score. SHAP (SHapley Additive exPlanations) analysis was used to quantify feature contributions and interpret the model.

Results: 194 (20.5%) pregnant women were identified with PSMU. Significant differences were observed in age, employment status, monthly household income, gestational age, screen time, conscientiousness level, neuroticism level, adverse childhood experience number, prenatal depression, prenatal anxiety, partner phubbing behavior, family well-being between those with or without PSMU. LASSO indicated that 12 factors (employment status, screen time, openness, agreeableness, conscientiousness, age, prenatal depression, partner phubbing, prenatal anxiety, resilience, family communication level, and gestational age) are associated with PSMU. The XGBoost model was selected as the final classification model, achieving an AUC (95% CI) of 0.91 (0.84–0.98), accuracy of 0.91, sensitivity of 0.82, specificity of 0.92, positive predictive value of 0.70, negative predictive value of 0.96, and an F1 score of 0.76. SHAP analysis indicated prenatal depression as the strongest predictor, followed by gestational week, prenatal anxiety, and partner phubbing behavior, while age and resilience had lower contributions.

Conclusions: We have developed a prediction model of PSMU in pregnant women in China, providing a useful tool for early identification. Targeted intervention and prevention programs for PSMU can be delivered to those with prenatal depression, larger gestational week, prenatal anxiety, and higher partner phubbing behavior.