JMIR Publications

ABSTRACT

Background:

To date, a reliable, validated pain-level assessment approach to evaluate consciousness status has yet to be established. With appropriate algorithms, computers can learn to detect patterns and associations in large datasets.

Objective:

This study aimed to apply machine learning to electrocardiography (ECG) waveforms to create an algorithm to predict and monitor regular uterine contraction cycle-induced labor pain.

Methods:

Pregnant women undergoing natural spontaneous delivery (NSD) with regular uterine contraction pain were recruited from National Taiwan University Hospital for prospective data collection and cross-sectional analysis. Machine learning was used to fine-tune a nociception-related algorithm (NoP) based initially on the complex analysis of features in high-fidelity ECG waveform recordings in 4 pregnant women. The algorithm was then validated in another 12 pregnant women. Data on uterine contractions were collected as raw data from tocometry and ECG waveforms by re-cording simultaneously using the same computer. Every uterine contraction cycle with VAS pain score was also confirmed and recorded by an investigating nurse for further analysis.

Results:

The machine learning model that most accurately predicted pain was the XGBoost model, which exhibited the highest ROC (0.98), followed by the random forest and GBDT models with an ROC of 0.89. NoP threshold values and NoP indexes were the highest and most important features in the XGBoost models.

Conclusions:

The newly developed algorithm supports obstetricians in clinical practice by providing early detection of alterations in ECG waveforms during uterine contraction cycles. The integration of ECG systems should be considered in artificial intelligence (AI) models for future parturient care.