Integrating longitudinal clinical and microbiome data to predict growth faltering in preterm infants

Area of Science:

• Neonatal Medicine
• Microbiome Research
• Computational Biology

Background:

• Preterm birth impacts over 10% of global births, with infants highly susceptible to Growth Faltering (GF).
• Existing nutritional interventions have not fully resolved the challenge of predicting and preventing GF in preterm infants.
• A comprehensive clinical and microbiome dataset was collected from multiple international sites to address this gap.

Purpose of the Study:

• To develop and enhance machine learning models for early prediction of Growth Faltering risk in preterm infants.
• To investigate the combined utility of clinical and microbiome data for improved GF prediction.
• To identify cost-effective and accurate methods for GF risk assessment and intervention prediction.

Main Methods:

• Utilized and extended machine learning techniques for GF prediction using clinical data.
• Applied extended graphical models to integrate time-series clinical and microbiome data.
• Developed a hierarchical classifier employing a subset of taxa for specific infant populations.

Main Results:

• Models integrating both clinical and microbiome data demonstrated superior GF prediction accuracy compared to clinical data alone.
• A small subset of microbial taxa was sufficient to enhance prediction accuracy and identify potential interventions.
• The hierarchical classifier using a subset of taxa proved to be the most accurate and cost-effective prediction method.

Conclusions:

• Integrating clinical and microbiome data offers a powerful approach for early Growth Faltering prediction in preterm infants.
• Machine learning models, particularly hierarchical classifiers, can effectively identify high-risk infants and guide targeted interventions.
• This research provides a foundation for developing novel, data-driven strategies to improve preterm infant outcomes.