# Factors affecting feeding ability in children with neonatal intensive care unit stay: a cluster analysis using machine learning methods

## TL;DR

This study uses machine learning to identify factors affecting oral feeding in critically ill neonates in the NICU, finding that sucking power and rooting reflex are key predictors.

## Contribution

The novel use of XGBoost and SHAP values to analyze oral feeding readiness in NICU neonates, identifying distinct profiles and key predictors.

## Key findings

- Three distinct profiles of oral feeding readiness were identified using cluster analysis.
- XGBoost outperformed other models with 85.2% accuracy in predicting oral feeding outcomes.
- Sucking power positively and rooting reflex negatively influenced feeding predictions.

## Abstract

Oral feeding introduction is challenging in Neonates Intensive Care Unit (NICU) daily care with limited measuring methods. Our study aimed to depict the oral feeding related features in neonates with critical conditions who were administered to NICU and its major predictors.

A total of 1,419 neonates with critical conditions who were administered to NICU were enrolled. The related features were acquired by using the Preterm Infant Oral Feeding readiness assessment scale (PIOFRA). The Shapley Additive Explanation (SHAP) values were used in XGBoost models established based on selected features. In addition, the ANOVA analysis was adopted to depict the group differences.

Three profiles with distinct PIOFRA features were identified in cluster analysis (p > 0.05). Compared to other prediction models (e.g., Logistic Regression, Random Forest), the XGBoost model achieved the highest accuracy (85.2%). Sucking power and rooting reflex were identified as the features with largest impact in oral feeding predations that exhibited positive and negative influence respectively.

Oral feeding difficulty can be commonly observed in neonates in NICU, and more detailed assessments are needed to illustrate the difference in gestational features (e.g., born weight, gestational age) between difference profiles. PIOFRA features can be strong predictors in predicting whether neonates had achieved full oral feedings or not. However, more studies are needed to verify the detailed mechanism to illustrate how sucking and rooting reflex functions ensure the safe and efficient content transportation in neonates administered to NICU.