Development of a Predictive Model for the Latency Period in Preterm Premature Rupture of Membranes Before 32 Weeks' Gestation

Background: Only 13.7% of corticosteroid treatments occurred within the most effective treatment time window for preterm premature rupture of membranes (PPROM), mainly due to the lack of relevant predictive models. This study aims to analyze the risk factors associated with the duration of the latency period in PPROM to develop a predictive model for the latency period in cases of PPROM before 32 weeks' gestation, thereby increasing the effective use rate of prenatal corticosteroids. Methods: A retrospective analysis was conducted pregnant women with PPROM before 32 weeks' gestation between 2018 and 2022 at the Second Affiliated Hospital of Wenzhou Medical University. The group was divided based on the latency period: duration of <5d group (n=192) and duration of ≥5d group (n=162). Five machine learning (ML) algorithms were utilized to predict the probability of a latency duration of treatment of ≥5 days. Model performance was evaluated using the area under the curve (AUC), predictive accuracy, sensitivity, and specificity. Results: Significant differences (P<0.05) were observed between the two groups in terms of white blood cell count, gestational age at rupture, cervical dilation, uterine contractions, and color of vaginal discharge. The AUC values for the five predictive models ranged from 0.66 to 0.90. The Logistic Regression model was identified as the best predictive model for latency duration in PPROM patients. Conclusion: Elevated body temperature, cervical dilation, increased white blood cell count, and the color of vaginal discharge are risk factors for a longer latency period in PPROM before 32 weeks' gestation. The gestational age at the time of spontaneous membrane rupture is a protective factor for the latency period in PPROM. Keywords：Preterm Premature Rupture of Membranes (PPROM); the latency period; Predictive Factors; Machine Learning