A Performance Evaluation of Computerised Antepartum Fetal Heart Rate Monitoring: The Dawes-Redman Algorithm at Term

Abstract

Objectives: This study aims to rigorously evaluate the Dawes-Redman computerised cardiotocography algorithm's effectiveness in assessing antepartum fetal wellbeing. It focuses on analysing the algorithm's performance using extensive clinical data, examining accuracy, sensitivity, specificity, and predictive values in various scenarios. The objectives include assessing the algorithm's reliability in identifying fetal wellbeing across different risk prevalences, its efficacy in the context of temporal proximity to delivery, and its performance across ten specific adverse pregnancy outcomes. This comprehensive evaluation seeks to clarify the algorithm's utility and limitations in contemporary obstetric practice, particularly in high-risk pregnancy scenarios. Methods: Antepartum fetal heart rate recordings from term singleton pregnancies between 37 and 42 gestational weeks were extracted from the Oxford University Hospitals database, spanning 1991 to 2021. Traces with significant data gaps or incomplete Dawes-Redman analyses were excluded. For the ten adverse outcomes, only traces performed within 48 hours prior to delivery were considered, aligning with clinical decision-making practices. A healthy cohort was established using rigorous inclusion and exclusion criteria based on clinical indicators. Propensity score matching, controlling for gestational age and fetal sex, ensured balanced comparisons between healthy and adverse outcome cohorts. The Dawes-Redman algorithm's categorisation of FHR traces as either 'criteria met' (an indicator of wellbeing) or 'criteria not met' (indicating a need for further evaluation) informed the evaluation of predictive performance metrics. Performance was assessed using accuracy, sensitivity, specificity, and predictive values (PPV, NPV), adjusted for various risk prevalences. Results: 4,196 term antepartum FHR traces were identified, matched by fetal sex and gestational age. The Dawes-Redman algorithm showed a high sensitivity of 91.7% for detecting fetal wellbeing. However, specificity for adverse outcomes was low at 15.6%. The PPV varied with population prevalence, high in very low-risk settings (99.1%) and declined with increased risk. Temporal proximity to delivery indicated robust sensitivity (>91.0%). Specificity notably decreased over time, impacting the algorithm's discriminative power for identifying adverse outcomes. Across different adverse conditions, the algorithm's performance remained consistent, with high sensitivity but varying NPVs, confirming its utility in detecting fetal wellbeing rather than adverse outcomes. Conclusion: These findings reveal the Dawes-Redman algorithm is effective for detecting fetal wellbeing in term pregnancies, evidenced by its high sensitivity and PPV. However, its low specificity suggests limitations in its ability to identify fetuses at risk of adverse outcomes. The predictive accuracy of the algorithm is significantly affected by the prevalence of healthy pregnancies within the population. Clinical interpretation of FHR traces that do not satisfy the Dawes-Redman criteria should be approached with caution, as they do not necessarily correlate with heightened risk. While the algorithm proves reliable for its primary objective in low-risk contexts, the development of algorithms optimised for high-risk pregnancy scenarios remains an area for future enhancement.