A prognostic model for use at labour onset to estimate the risk of severe fetal compromise at birth: a retrospective, model development and validation study.

BACKGROUND: Absence of reliable risk assessment and monitoring of fetal wellbeing during labour remains a major clinical challenge worldwide. Cardiotocography (CTG) is commonly used to monitor the fetal heart rate, but its misinterpretation or delays in clinical escalation contribute to poor birth outcomes. We aimed to develop a novel prognostic model for the prediction of severe fetal compromise at birth, combining known clinical risk factors with the first 20 min of CTG monitoring.

METHODS: In this retrospective, model development and validation study, we used routinely collected, fully anonymised maternity data from three databases: the John Radcliffe Hospital (Oxford, UK; 1993-2024; model training and testing), the Rosie Hospital (Cambridge, UK; 2016-22; external validation), and the Stockholm-Gotland region (Sweden; 2018-20; independent external validation). Oxford and Cambridge datasets were accessed from maternity and neonatal unit electronic health records; Stockholm-Gotland data were accessed as part of the larger Stockholm-Gotland Perinatal Cohort. Singleton births at 36 weeks and 0 days of gestation or more undergoing intrapartum CTG monitoring, whether labour was spontaneous or induced, were included. Clinical risk factors used for model development were maternal age, gestational age, nulliparity, pre-eclampsia, diabetes in pregnancy, thick meconium, and maximal recorded maternal temperature in labour. Quantitative CTG features included in the model were fetal heart rate baseline, short-term variability, decelerative capacity, number of accelerations, and number of prolonged decelerations, and were extracted using the Oxford intrapartum computer-based algorithms. Birth outcomes were defined in three mutually exclusive groups: severe compromise, intermediate compromise, and healthy. Fit4Labour risk score (ie, estimated probability of severe compromise) was stratified into four risk groups: >95th centile of Fit4Labour risk score, >85th and ≤95th centile of Fit4Labour risk score, >25th and ≤85th centile of Fit4Labour risk score, and ≤25th centile of Fit4Labour risk score. Model accuracy was assessed by use of partial receiver operating characteristic curves and comparison with a clinical benchmark (sensitivity at 15% screen-positive rate).

FINDINGS: 147 350 births with at least 20 min of CTG data were analysed: 91 378 from Oxford 1993-2024 (73 443 for training and testing [Oxford 1993-2019] and 17 935 for internal validation [Oxford 2020-24]), 14 709 from Cambridge 2016-22 (external validation), and 41 263 from Stockholm-Gotland 2018-20 (independent external validation). The rate of severe compromise was higher in the highest-risk group (>95th centile) than in the lowest-risk group (≤25th centile) across all validation datasets: 0·9% (95% CI 0·28-1·51) in the highest risk group versus 0·1% (0·01-0·21) in the lowest-risk group in the Oxford 2020-24 dataset; 0·3% (0·00-0·65) versus 0·1% (0·00-0·22) in the Cambridge 2016-22 dataset; and 0·9% (0·47-1·27) versus 0·2% (0·08-0·23) in the Stockholm-Gotland 2018-20 dataset. At a fixed screen-positive rate of 15%, 20-min Fit4Labour had a sensitivity of 33·4% (95% CI 28·31-38·56), compared with the clinical benchmark sensitivity of 39·9% (95% CI 32·59-47·18, p=0·18), and showed consistent performance across all validation datasets.

INTERPRETATION: Fit4Labour is a robust prognostic model, readily applicable around labour onset and extensively validated with retrospective data from sites with different maternity demographics and clinical care practices. Fit4Labour relies on routinely available data and could produce interpretable output in real time, providing timely individualised risk stratification that is data-driven, objective, and reproducible.

FUNDING: National Institute for Health and Care Research.