Systematic Review of Studies Investigating Infant Feeding Difficulties: Focus on Machine Learning Applications

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Not quantitatively reported; described only as 'generally achieved high classification accuracy' — no specific accuracy values provided

• Primary Outcome: Classification or prediction of infant feeding difficulties using ML algorithms applied to physiological signals (sucking, swallowing, breathing events)

• Nnt Or Sensitivity: Sensitivity, specificity, AUC-ROC, and other diagnostic performance metrics are not reported in the abstract. Individual study performance data are not synthesised.

• Confidence Interval: Not reported

Clinical Application

Currently limited. The review identifies that only 4 of 10 included studies directly applied ML models, and none appear to have been validated in prospective clinical trials or implemented in real-world clinical workflows. Sensor-based data acquisition adds complexity to bedside implementation. Standardisation of physiological signal measurement protocols and algorithm validation across diverse populations are prerequisites for clinical deployment. In Australia, infant feeding difficulties are managed across NICU, neonatal, and community paediatric settings under RACGP and Australian College of Neonatologists frameworks. Speech pathology-led instrumental swallowing assessments (videofluoroscopic swallow study, fibreoptic endoscopic evaluation of swallowing) remain the clinical standard. No ML-based feeding assessment tools are currently listed on the Australian Register of Therapeutic Goods (ARTG) or recommended in RACGP clinical guidelines. The TGA would require prospective clinical validation data before any such tool could be registered as a medical device software (SaMD) under the Software as a Medical Device framework. PBS listing is not applicable to diagnostic algorithms. Australian NICUs with existing high-density physiological monitoring infrastructure (e.g., major tertiary centres) would be the most feasible early implementation sites if validated tools emerge. Infants and newborns presenting with feeding difficulties in neonatal intensive care units (NICUs), neonatal nurseries, and paediatric outpatient settings. Most applicable to preterm infants and those with neurological or structural conditions affecting suck-swallow-breathe coordination.

Abstract

INTRODUCTION: This study aims to systematically review studies using machine learning for infant feeding difficulties and identify the clinical applicability and limitations of such approaches. METHODS: A literature search was conducted in accordance with the PRISMA guidelines. Studies on the use of machine learning for infants with feeding difficulties were analyzed. RESULTS: We screened 1,104 studies, and 10 were eligible for inclusion. These studies applied machine learning to measure physiological signals during feeding, such as the number of sucking, swallowing and breathing events, to classify or predict feeding difficulties. The most common algorithms in the analyzed studies included support vector machines (SVMs), k-nearest neighbors (KNNs), and decision trees (DTs), which generally achieved high classification accuracy. DISCUSSION: Although all included studies used sensor-based data, only four directly applied machine learning models. Future research should expand the direct applications of machine learning, standardize measurements, and validate algorithms to improve clinical utility.

References

1. 1.Ji, E. S., Lee, K. J., & Chung, Y. C. (2026). Systematic Review of Studies Investigating Infant Feeding Difficulties: Focus on Machine Learning Applications. Journal of Pediatric Health Care. Advance online publication. https://doi.org/10.1016/j.pedhc.2026.01.009