Prediction Model for Failed Vacuum Assisted Delivery: A Retrospective Cohort Study

Background/Objectives: We aimed to determine risk factors and to design a clinically based predictive model for a failed vacuum assisted delivery (VAD). Methods: We conducted a retrospective cohort study in a single tertiary university-affiliated medical center between 2011 and 2023. The study population consisted of singleton pregnancies with a VAD trial. The study group comprised cases of a failed VAD, defined as the occurrence of any of the following: (1) more than two vacuum cup detachments; (2) extraction duration exceeding 20 min; or (3) abandonment of the vacuum attempt by the operating physician, with conversion to urgent cesarean delivery (CD). The control group comprised cases of successful VADs. Factors associated with failed VAD were examined by univariate and multivariate analyses. A prediction score was developed to predict failed VAD. A receiver-operating characteristic curve (ROC) was utilized for the model. Internal validation was performed by means of a 70/30 train–test split, with model performance evaluated on the validation set using ROC analysis. Results: A total of 131,019 women delivered in our center during the study period. VAD was attempted in 8885 (6.8%) cases, of which 172 (1.9%) failed trials that led to urgent CDs. Several risk factors for a failed VAD were identified, including induction of labor, fetal head station below +2 cm relative to the ischial spines, duration of the second stage of delivery >3.5 h, preeclampsia, birthweight >3750 g, and male gender. The prediction score demonstrated good discriminatory performance, with an AUC of 0.723 (95% CI 0.637–0.810). Internal validation using a 30% holdout cohort revealed that the model maintained good performance, with an AUC of 0.764 (95% CI 0.619–0.909; p < 0.001). Conclusions: Our model has the potential to assist obstetricians with VAD decision-making and parturient counseling, as well as identifying parturients at high risk for complicated deliveries.