Prediction of severe adverse neonatal outcomes at the second stage of labour using machine learning: a retrospective cohort study

Abstract

Objective To create a personalised machine learning model for prediction of severe adverse neonatal outcomes (SANO) during the second stage of labour. Design Retrospective Electronic‐Medical‐Record (EMR) ‐based study. Population A cohort of 73 868 singleton, term deliveries that reached the second stage of labour, including 1346 (1.8%) deliveries with SANO. Methods A gradient boosting model was created, analysing 21 million data points from antepartum features (e.g. gravidity and parity) gathered at admission to the delivery unit, and intrapartum data (e.g. cervical dilatation and effacement) gathered during the first stage of labour. Deliveries were allocated to high‐risk and low‐risk groups based on the Youden index to maximise sensitivity and specificity. Main outcome measures SANO was defined as either umbilical cord pH levels ≤7.1 or 1‐minute or 5‐minute Apgar score ≤7. Results The model for prediction of SANO yielded an area under the receiver operating curve (AUC) of 0.761 (95% CI 0.748–0.774). A third of the cohort (33.5%, n = 24 721) were allocated to a high‐risk group for SANO, which captured up to 72.1% of these cases (odds ratio 5.3, 95% CI 4.7–6.0; high‐risk versus low‐risk groups). Conclusions Data acquired throughout the first stage of labour can be used to predict SANO during the second stage of labour using a machine learning model. Stratifying parturients at the beginning of the second stage of labour in a ‘time out’ session, can direct a personalised approach to management of this challenging aspect of labour, as well as improve allocation of staff and resources. Tweetable abstract Personalised prediction score for severe adverse neonatal outcomes in labour using machine learning model.

Artificial intelligence in obstetrics

“…A third (33.5%; n=24,721) of cohorts were assigned to high-risk groups for SANOs, and SANOs in these high-risk groups were found to occur more significantly than in low-risk groups (odds ratio, 5.3; 95% CI. 4.7-6.0; high risk vs. low risk) [58].…”

“…The machine learning model for 1,346 cases (1.8%) with severe adverse neonatal outcomes (SANO) among 73,868 single-child pregnancies showed an AUC of 0.761 (95% CI, 0.748–0.774). A third (33.5%; n=24,721) of cohorts were assigned to high-risk groups for SANOs, and SANOs in these high-risk groups were found to occur more frequently than in low-risk groups (odds ratio, 5.3; 95% CI, 4.7–6.0; high risk vs. low risk) [ 58 ].…”

Evolving Frameworks for the Foundation and Practice of Electronic Fetal Monitoring

“…There is something of a tradeoff between the interpretability and accuracy of models 30 . Models are easier to understand when they are linear, additive, and less complex.…”

“…This encourages us to think about both clinical practice and model “re-specification.” The concept of a re-specification of a model involves opening it to inquiry about what should be in a model and what should be left out, how the relationships should be specified, and what the nature of each feature actually is. Guedalia et al 30 have summarized the impact of five features on a summary measure of adverse neonatal outcomes. Clinicians with experience in FHR monitoring might challenge the model to consider other aspects of FHR aside from simply keeping track of the number of non-reassuring FHR instances.…”

“…There is something of a tradeoff between the interpretability and accuracy of models. 30 Models are easier to understand when they are linear, additive, and less complex. However, especially with respect to complicated problems, non-linear models that are complex and have important interaction terms tend to be more accurate.…”