(460) Enhancing neonatal acidosis prediction: A Machine Learning approach using CTG features and clinical characteristics

Neonatal acidosis can lead to severe consequences such as hypoxic-ischemic encephalopathy, and mortality. Timely prediction during labor is essential for effective interventions and improved outcomes. We aimed to investigate the performance of machine learning techniques to predict neonatal acidemia at birth based on cardiotocography (CTG) features and maternal, pregnancy and labor characteristics.

Study Design:

A predictive modeling study was conducted on term singleton vaginal deliveries at a tertiary maternity hospital between January 2018 and July 2021. Patients with missing information or >30% signal loss were excluded. Neonates with cord pH< 7.10 were matched randomly with those with pH >7.10 to create a balanced dataset. After preprocessing of CTG tracings, features were extracted from each 10-minute interval of the 2h before delivery (baseline, short- and long-term variability, number of accelerations, decelerations, and contractions, deceleration area and reperfusion time). Maternal, pregnancy and labor characteristics were also extracted. A logistic regression model and a random forest model including all characteristics associated with neonatal acidosis with an area under the curve (AUC) >0.6 in the univariate analysis were developed. The dataset was split into training and test subsets, and a 10-fold cross-validation was performed to obtain the final AUC and accuracy of the models.

Results:

Among the 10561 included women, 395(3.7%) had a cord pH< 7.10 (mean pH 7.05±0.05) and were matched randomly with 395 women with a cord pH >7.10 (mean pH 7.26±0.07). Parity, oxytocin during labor and all CTG features but short time variation were associated with acidosis in univariate analysis. The logistic regression model on the test set achieved an AUC of 0.90 and an accuracy of 0.84. For the random forest model, the AUC and accuracy were 0.90 and 0.82, respectively.

Conclusion:

The findings from this study hold the potential to facilitate timely clinical interventions, enabling healthcare providers to make informed decisions and improve perinatal outcomes.