Abstract
The neonatal oxycardiorespirogram (OCRG) captures synchronized, multi-channel recordings of respiratory patterns, heart rate variability, transcutaneous oxygen tension, and relative skin perfusion in neonates. As a non-invasive, point-of-care modality, OCRG is routinely used to assess cardiorespiratory stability in high-risk infants, particularly preterm neonates at risk for apnea, bradycardia, and desaturation. These events can persist beyond hospital discharge, elevating morbidity and mortality, yet no standardized tool reliably predicts which infants will experience clinically significant post-discharge episodes. Although OCRG is established in clinical practice, its rich time-series data remains largely underutilized for predictive modeling. In contrast, machine learning methods have achieved strong performance in related neonatal monitoring tasks - such as apnea detection, sepsis prediction, sleep staging, and extubation readiness - by integrating multimodal biosignals and temporal modeling strategies. These advances highlight the opportunity to apply machine learning analytics and explainability methods to OCRG data, enabling the discovery of physiological patterns, refining risk stratification, and informing individualized interventions such as the timing of caffeine withdrawal, initiation of home monitoring, or discharge planning. Given the multimodal and sequential structure of OCRGs, time-series-based machine learning, including both shallow and deep learning approaches, represents particularly promising analytic strategies for future applications. This mini-review synthesizes current gaps in OCRG-based analytics, examines transferable lessons from existing machine learning applications in neonatal biosignals, and outlines a translational roadmap for evolving OCRG from a descriptive monitoring tool into a predictive platform for precision neonatal care.