Non-elective caesarean section risk assessment using Machine Learning techniques

Background

The sustained increase and the disparities in the rate of caesarean deliveries (CS) pose a public health challenge. Optimising the use of CS is of global concern and a challenge for public health. Machine Learning (ML) techniques can assist clinicians in decision-making, improving treatment modalities and patient outcomes.

Methods

Original data correspond to the 41,037 deliveries that took place at La Paz University Maternity Hospital (Madrid, Spain) between 2010 and 2018. Machine Learning (ML) model algorithm Random Forest (RF) was performed to determine the risk of CS. The first analysis was Mean Decrease Accuracy carried out on 50 permutations. The second analysis was ceteris-paribus and partial-dependence profiles.

Results

The RF models obtained identify a higher risk of CS delivery in multiple gestations, macrosomic foetuses and in those with prolonged gestation associated with other maternal–foetal characteristics. Results deny the consideration that older nulliparous mothers represent a specific obstetrtic risk goup.

Conclusions

ML techniques can be very useful in identifying risk factors to be addressed to optimise the number of CS. Macrosomia prevention programmes, reduction in the rate of multiple pregnancies or finishing pregnancy before the onset of risks associated with prolonged pregnancy may be ways to optimise the number of CS.