The efficacy of machine learning in predicting fetal growth restriction: a systematic review and meta-analysis

Objectives Fetal growth restriction(FGR) is one of the most common causes of perinatal death and various short- and long-term complications. Accurate identification of fetal growth restriction is essential to reduce adverse perinatal outcomes. With the development of machine learning, many predictive models for fetal growth restriction have emerged. We assessed their performance by conducting systematic reviews and meta-analyses. Methods A systematic literature search was performed for relevant studies reported before May 31, 2022. The quality of the studies was assessed in strict accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (version 2020). Models with the reported area under the receiver operating characteristic (AUROC) indicator curve were meta-analyzed to determine the factors that contributed most to model performance. Results After the screening, 20 papers were eligible for synthesis, 19 were rated as high risk of bias, and 1 was rated as unclear risk of bias. From which 37 models were extracted, the c-statistic of the pooled random effects model (I2 = 96.0%) was 0.8123 (0.7824–0.8433, 95%CI). Among them, the c-statistic of the pooled risk model using the consensus developed by the Delphi procedure as the defined model was 0.81 (0.77–0.85, 95%CI), and the c-statistic of the pooled risk model using other defined models was 0.83 (0.78–0.87, 95%CI). In selecting predictors, most models were constructed by combining basic maternal characteristics, maternal pregnancy radionics, and serological indicators during pregnancy. Conclusion Machine learning has an excellent predictive value for FGR, which indicates that practical machine learning can be used as a possible means of FGR identification. However, improvements are needed in terms of quality and study design. We look forward to a unified definition and establishing an effective scoring tool to identify FGR accurately and intervene in time.