Can maternal characteristics improve due date prediction without ultrasound? Development and internal validation of a machine-learning prediction model in 600 000 pregnancies in Sweden

Abstract: To assess whether estimated due dates based on Naegele’s rule can be improved by incorporating routinely collected maternal characteristics into a non-parametric machine learning model, particularly in settings where ultrasound is unavailable.

Design

Population-based retrospective cohort study; development and internal validation of a machine-learning prediction model.

Setting

Nationwide register data from Sweden.

Participants

603 861 singleton pregnancies recorded in the Swedish Medical Birth Register between 2008 and 2017. We restricted the sample to spontaneous singleton births with complete information on last menstrual period, ultrasound dating and maternal characteristics and further limited the primary analysis to pregnancies in which both Naegele’s rule and ultrasound-based gestational age at delivery fell between 37 and 42 completed weeks. Primary and secondary outcome measures: The primary outcome was prediction accuracy of the date of delivery. Performance was evaluated using exact-day accuracy, accuracy within ±1 and ±7 days, mean squared error and root mean squared error. Machine-learning predictions were compared with those obtained using Naegele’s rule and ultrasound-based dating under identical inclusion criteria.