Using Machine Learning to Achieve Accurate Estimates of Fetal Gestational Age and Personalized Predictions of Fetal Growth

Background: Preterm birth is a major global health challenge, and the leading cause of death in children under 5 years old 1. It is also a key measure of a population's general health and nutritional status 2. Current clinical methods of estimating fetal gestational age are often inaccurate; between 20 and 30 weeks of gestation, even the best ultrasound estimates have uncertainties of 9 - 18 days 3 (full widths of 18 - 36 days). Accurate estimates of fetal gestational age and personalized predictions of future growth can substantially improve the management of individual pregnancies and population-level health. Methods: Using ultrasound-derived, fetal biometric data, we present a novel machine-learning approach to accurately estimate the gestational age, and predict the future growth trajectory of each fetus. The accuracy of the method is determined by reference to exactly known facts pertaining to each fetus, rather than the start of the mother's last menstrual cycle. The data stem from a sample of healthy, well-nourished participants in a large, multicenter, population-based study, INTERGROWTH-21st 4. The generalizability of the algorithm is demonstrated with data from a different and more heterogeneous population (INTERBIO-21st). No new facilities are needed beyond those routinely available in clinical settings. Findings: We estimate the fetal gestational age to within 3 days, using measurements made in a 10-week window spanning the second and third trimesters. Fetal gestational age can thus be estimated into the third trimester with an accuracy of 3 days, which is 300% to 500% better than possible with any previous algorithm 5. This will enable improved management of individual pregnancies. Personalized forecasts of future fetal growth are also, for the first time, available. Six-week forecasts of the growth trajectory for a given fetus are accurate to within 7 days. This will help identify at-risk fetuses significantly more accurately than currently possible. At population level, the much higher accuracy will improve fetal growth charts and population health assessments. Upon publication of this paper, the new algorithm can be used free of charge via a web portal. Interpretation: Modern machine-learning can circumvent longstanding limitations in determining fetal gestational age and future growth trajectory without recourse to often inaccurately-known information, such as the date of the mother's last menstrual period. Our approach can be extended to other types of fetal-related data, such as measurements of cell-free RNA (cfRNA) transcripts in maternal blood 6. More generally, the approach has the potential to provide accurate forecasts of disease progression from spot measurements of the relevant biomarkers. Funding Statement: Bill & Melinda Gates Foundation; US Department of Energy, Office of Science, Basic Energy Sciences award DE-SC0002164 (underlying dynamical techniques); US National Science Foundation awards STC 1231306 (underlying data analytical techniques) and 1551489 (underlying analytical models); and National Institute for Health Research (NIHR) Oxford Biomedical Research Centre (BRC). Declaration of Interests: The authors stated: "None reported." Ethics Approval Statement: The INTERGROWTH-21st 255 Project was approved by the Oxfordshire Research Ethics Committee “C” (reference: 08/H0606/139), and the research ethics committees of the individual institutions and the regional health authorities where the project was implemented. Written informed consent was obtained from all participants.