Clinical and Surgical Department, Faculty of Medicine and Pharmacy, "Dunărea de Jos" University, 47 Domnească Street, 800008 Galati, Romania.
Department of Mother and Child Care, "Grigore T. Popa" University of Medicine and Pharmacy Iasi, 700115 Iasi, Romania.
Medicina (Kaunas). 2024 Sep 30;60(10):1604. doi: 10.3390/medicina60101604.
: Intra/postpartum hemorrhage stands as a significant obstetric emergency, ranking among the top five leading causes of maternal mortality. The aim of this study was to assess the predictive performance of four machine learning algorithms for the prediction of postpartum and intrapartum hemorrhage. : A prospective multicenter study was conducted, involving 203 patients with or without intra/postpartum hemorrhage within the initial 24 h postpartum. The participants were categorized into two groups: those with intra/postpartum hemorrhage (PPH) and those without PPH (control group). The PPH group was further stratified into four classes following the Advanced Trauma Life Support guidelines. Clinical data collected from these patients was included in four machine learning-based algorithms whose predictive performance was assessed. : The Naïve Bayes (NB) algorithm exhibited the highest accuracy in predicting PPH, boasting a sensitivity of 96.3% and an accuracy of 98.6%, with a false negative rate of 3.7%. Following closely were the Decision Tree (DT) and Random Forest (RF) algorithms, each achieving sensitivities exceeding 94% with a false negative rate of 5.9%. Regarding severity classification I, the NB and Support Vector Machine (SVM) algorithms demonstrated superior predictive capabilities, achieving a sensitivity of 96.4%, an accuracy of 92.1%, and a false negative rate of 3.6%. The most severe manifestations of HPP were most accurately predicted by the NB algorithm, with a sensitivity of 89.3%, an accuracy of 82.4%, and a false negative rate of 10.7%. : The NB algorithm demonstrated the highest accuracy in predicting PPH. A notable discrepancy in algorithm performance was observed between mild and severe forms, with the NB and SVM algorithms displaying superior sensitivity and lower rates of false negatives, particularly for mild forms.
产后出血是一种严重的产科急症，是导致产妇死亡的五大原因之一。本研究旨在评估四种机器学习算法对预测产后和产时出血的预测性能。
本研究采用前瞻性多中心研究，共纳入 203 例产后 24 小时内发生或未发生产后出血的患者。根据产后出血的严重程度分为产后出血（PPH）组和无产后出血（对照组）。PPH 组根据高级创伤生命支持指南进一步分为四组。将这些患者的临床数据纳入四个基于机器学习的算法中，评估其预测性能。
朴素贝叶斯（NB）算法在预测 PPH 方面表现出最高的准确性，其敏感性为 96.3%，准确性为 98.6%，假阴性率为 3.7%。紧随其后的是决策树（DT）和随机森林（RF）算法，它们的敏感性均超过 94%，假阴性率为 5.9%。在严重程度分类 I 方面，NB 和支持向量机（SVM）算法表现出更好的预测能力，敏感性为 96.4%，准确性为 92.1%，假阴性率为 3.6%。最严重的 HPP 表现形式由 NB 算法预测的准确性最高，敏感性为 89.3%，准确性为 82.4%，假阴性率为 10.7%。
NB 算法在预测 PPH 方面表现出最高的准确性。在轻度和重度表现形式之间，算法性能存在显著差异，NB 和 SVM 算法表现出更高的敏感性和更低的假阴性率，特别是对于轻度表现形式。
