Prediction of Maternal Health Risk Using Clinical and Physiological Indicators in Rural Bangladesh: A Machine Learning Study

Deprecated: htmlspecialchars(): Passing null to parameter #1 ($string) of type string is deprecated in /var/www/html/lib/pkp/lib/vendor/smarty/smarty/libs/plugins/modifier.escape.php on line 36

Abstract

Identifying maternal health risk early is critical to minimizing preventable complications, especially in low-resource and rural areas. In this study, machine-learning methods have been used to assess the capability of clinical and physiological parameters to classify the maternal health risk. We performed a secondary analysis of the Maternal Health Risk dataset with maternal age, SBP, DBP, blood sugar, body temperature and heart rate as predictors. One physiologically impossible heart-rate observation was eliminated, and 451 records remained after the elimination of exact duplicate records. The following machine learning supervised models were developed: multinomial logistic regression, decision tree, random forest, support vector machine, and gradient boosting. Model performance was evaluated based on the accuracy, balanced accuracy, precision, recall, macro F1-score, Cohen's kappa, receiver operating characteristic, Area under the curve, and confusion matrices. Gradient Boosting performed the best overall (74.7% accuracy, 66.0% balanced accuracy, macro F1 score of 67.3% and Cohen's kappa score of 0.557). For low-risk cases, the model had a high recall, and for high-risk cases, a good discrimination, with mid-risk cases being relatively poor. The most significant predictor was blood sugar, followed by SBP. Routinely collected indicators can be used with machine-learning models to provide preliminary maternal risk screening. But more general predictors and external validation are needed before clinical use.