Fetal growth and gestational age prediction by machine learning

Abstract

Current strategies to identify fetuses with pathological fetal growth restriction are inadequate. Conventionally, fetal growth restriction is defined as an estimate of fetal size at a specific gestational age that is below a predefined threshold, usually the bottom 10th percentile, based on a specific growth standard or population reference.1Ananth CV Brandt JS Vintzileos AM Standard vs population reference curves in obstetrics: which one should we use?.Am J Obstet Gynecol. 2019; 220: 293-296Summary Full Text Full Text PDF PubMed Scopus (12) Google Scholar By this definition, most fetuses diagnosed with fetal growth restriction are constitutionally small but not at increased risk for adverse outcomes due to hypoxic-ischaemic stress, and instead are at high risk for interventions such as iatrogenic preterm birth due to false-positive antenatal testing. Recent efforts to improve the identification of at-risk fetuses have focused on a new international consensus definition for fetal growth restriction2Gordijn SJ Beune IM Thilaganathan B et al.Consensus definition of fetal growth restriction: a Delphi procedure.Ultrasound Obstet Gynecol. 2016; 48: 333-339Crossref PubMed Scopus (664) Google Scholar and the development of growth standards, which are nomograms of fetal growth based on healthy populations.3Papageorghiou AT Ohuma EO Altman DG et al.International standards for fetal growth based on serial ultrasound measurements: the Fetal Growth Longitudinal Study of the INTERGROWTH-21st Project.Lancet. 2014; 384: 869-879Summary Full Text Full Text PDF PubMed Scopus (525) Google Scholar, 4Buck Louis GM Grewal J Albert PS et al.Racial/ethnic standards for fetal growth: the NICHD Fetal Growth Studies.Am J Obstet Gynecol. 2015; 213: 449.e1-449.e41Summary Full Text Full Text PDF PubMed Scopus (291) Google Scholar, 5Kiserud T Piaggio G Carroli G et al.The World Health Organization fetal growth charts: a multinational longitudinal study of ultrasound biometric measurements and estimated fetal weight.PLoS Med. 2017; 14e1002220Crossref PubMed Scopus (358) Google Scholar In their recent guidelines, the Society for Maternal Fetal Medicine recommends the use of population references (nomograms based on general populations including healthy and abnormal pregnancies) in clinical practice.6Martins JG Biggio JR Abuhamad A Society for Maternal-Fetal Medicine (SMFM) consult series #52: diagnosis and management of fetal growth restriction.Am J Obstet Gynecol. 2020; (published online May 12.)DOI:10.1016/j.ajog.2020.05.010Summary Full Text PDF PubMed Scopus (140) Google Scholar Population references, even those based on ultrasound parameters, have inherent limitations since menstrual dates directly or indirectly form the basis upon which predictions are made. In The Lancet Digital Health, Russell Fung and colleagues7Fung R Villar J Dashti A et al.Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study.Lancet Digital Health. 2020; 2: e368-e375Summary Full Text Full Text PDF Scopus (22) Google Scholar applied a novel machine learning algorithm to improve gestational dating and create personalised projections of growth. The authors used data from the multicentre, International Fetal and Newborn Growth Consortium for the 21st Century (INTERGROWTH-21st) Project to test the algorithm, which they also validated using the INTERBIO-21st Fetal Study, a large, heterogeneous, multinational study enriched with patients at risk for fetal growth restriction. For patients who had a first scan between 20 and 30 weeks gestation and a second scan in the ensuing 10 weeks, the algorithm estimated gestational age with a 95% CI half-width to within 3 days. The personalised forecasts of fetal growth were accurate to within 7 days across a 6-week intervisit interval. The study has several strengths. The application of geometric machine learning can circumvent the inadequacies of current methods to assign gestational dating. Fung and colleagues7Fung R Villar J Dashti A et al.Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study.Lancet Digital Health. 2020; 2: e368-e375Summary Full Text Full Text PDF Scopus (22) Google Scholar used three methods to assess fetal size, assign gestational age, and predict fetal growth. The resultant algorithm does not require conversion of fetal biometric data to gestational age based on assumptions of reliable ascertainment of last menstrual period nor is it biased by the intentional disregard of expected variability in fetal growth. Rather, the algorithm uses observable ultrasound measures to assign gestational age and predict the individualised forecast of fetal growth, although the basis for this forecast is the observed gestational age and fetal biometry. Personalised forecasts of fetal growth provide insight into individualised growth potential with easier applicability and better accuracy than current iterations of customised growth curves. The algorithm uses only observable parameters on ultrasound, yet it can generate accurate forecasts of fetal growth to within 7 days. Deviation of predicted growth should identify fetuses at risk for complications due to impaired or accelerated growth, prompting additional surveillance or diagnostic testing. The study raises several important issues. First, there is uncertainty regarding the application of machine learning in the prediction of fetal biometry or measures associated with them.8Naimi AI Platt RW Larkin JC Machine learning for fetal growth prediction.Epidemiology. 2018; 29: 290-298Crossref PubMed Scopus (24) Google Scholar, 9Kuhle S Maguire B Zhang H et al.Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study.BMC Pregnancy Childbirth. 2018; 18: 333Crossref PubMed Scopus (54) Google Scholar, 10Wosiak A, Zamecznik A, Niewiadomska-Jarosik K. Supervised and unsupervised machine learning for improved identification of intrauterine growth restriction types. Federated Conference on Computer Science and Information Systems (FedCSIS); Gdansk; Sept 11–14, 2016: 323–29.Google Scholar The optimal cut-points associated with intrauterine fetal demise and related complications remain unknown. Second, the focus of this work was on errors around the predictions. Quantification of uncertainty intervals around these predictions remain essential. Third, application of resampling methods (eg, k-fold cross-validation) could have provided more realistic predictions. Additionally, the espoused methods underscore the need for prospective validation in external cohorts with wide geographical diversity and differing patient populations. Fourth, how does the algorithm predict pathological fetal growth restriction and what are the potential costs associated with overdiagnosis of fetal growth restriction? Fifth, and most importantly, the algorithm by Fung and colleagues7Fung R Villar J Dashti A et al.Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study.Lancet Digital Health. 2020; 2: e368-e375Summary Full Text Full Text PDF Scopus (22) Google Scholar is only applicable to patients having a first ultrasound at 20–30 weeks gestation. In developed countries where access to first trimester scanning is common, the effect and clinical utility of this algorithm could be less relevant. Fung and colleagues7Fung R Villar J Dashti A et al.Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study.Lancet Digital Health. 2020; 2: e368-e375Summary Full Text Full Text PDF Scopus (22) Google Scholar have developed a machine learning algorithm to assign gestational age that overcomes many of the limitations that hamper the accuracy of conventional dating methods. The algorithm also forecasts personalised fetal growth based exclusively on observable ultrasound parameters, which could mitigate the effects of population averaging to assess healthy versus abnormal growth. We believe that machine learning for gestational dating and assessment of fetal growth is here to stay, but several issues should be addressed before enthusiasm for widespread application increases. Nevertheless, as the results of this study reach the scientific community, we hope that the discussion will steer away from standards and population references. The applications of machine learning might just be the future of research for fetal growth restriction, but we are not there yet. We declare no competing interests. Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning studyMachine learning can circumvent long-standing 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. Using this algorithm in clinical practice could facilitate the management of individual pregnancies and improve population-level health. Upon publication of this study, the algorithm for gestational age estimates will be provided for research purposes free of charge via a web portal. Full-Text PDF Open Access

Topics & Concepts