Supervised machine learning on ECG features to classify sleep in non-critically ill children.

ABSTRACT

STUDY OBJECTIVES: Despite frequent sleep disruption in the paediatric intensive care unit (PICU), bedside sleep monitoring in real-time is currently not available. Supervised machine learning (ML) applied to electrocardiography (ECG) data may provide a solution, since cardiovascular dynamics are directly modulated by the autonomic nervous system (ANS) during sleep. METHODS: Retrospective study using hospital-based polysomnography (PSG) recordings obtained in non-critically ill children between 2017 and 2021. Six age categories were defined 6-12 months, 1-3 years, 3-5 years, 5-9 years, 9-13 years and 13-18 years. Features were derived in time, frequency and non-linear domain from pre-processed ECG data. Sleep classification models were developed for two, three, four and five state using logistic regression (LR), random forest (RF) and XGBoost (XGB) classifiers during five-fold nested cross-validation. Models were additionally validated across age categories. RESULTS: A total of 90 non-critically ill children were included with a median (Q1, Q3) recording length of 549.0 (494.8, 601.3) minutes. The three models obtained AUROC 0.72 - 0.78 with minimal variation across classifiers and age categories. Balanced accuracies were 0.70 - 0.72, 0.59 - 0.61, 0.50 - 0.51 and 0.41 - 0.42 for two, three, four and five state. Generally, the XGB model obtained the highest balanced accuracy (p < 0.05), except for five state where LR excelled (p = 0.67). CONCLUSIONS: ECG-based ML models are a promising and non-invasive method for automated sleep classification directly at the bedside of non-critically ill children aged 6 months to 18 years. Models obtained moderate-to-good performance for two and three state classification.