Use of artificial intelligence to predict mean time to delivery following cervical ripening with dinoprostone vaginal insert

Objective: To validate a mathematical model to predict the mean time to delivery (TTD) following cervical ripening with dinoprostone vaginal insert (DVI), and assess its impact on the risk of nocturnal deliveries.

Methods: We performed a case-control retro-prospective study at Angers University Hospital. In the control group, we retrospectively included 405 patients who underwent cervical ripening with DVI between 01/2015 and 09/2016. Based on the delivery outcomes, we developed a mathematical model that integrates all the factors influencing TTD following cervical ripening with DVI. In the study group, we prospectively included 223 patients who underwent cervical ripening with DVI between 11/2017 and 11/2018. The timing of insertion was calculated using the mathematical model developed in the control group, in order to prevent the occurrence of nocturnal deliveries.

Results: The calculated mean TTD was significantly shorter than the real mean TTD (21h46 min ± 3h28 min versus 25h38 min ± 12h10 min, p < 0.001), and for 44% of patients, there was at least 10 h difference between the two. The real TTD (25h38 min ± 12H10 min versus 20h39 min ± 10h49, p < 0.001), and the rate of nocturnal deliveries (30.5% versus 21.2%, p = 0.01) were significantly higher in the study group compared to the control group.

Conclusion: The mathematical model did not help predicting TTD following cervical ripening with DVI, and or reducing the number of nocturnal deliveries.