Developing a resiliency model for survival without major morbidity in preterm infants

Area of Science:

• Neonatal Medicine
• Machine Learning in Healthcare
• Biostatistics

Background:

• Preterm neonates (<32 weeks gestation) face significant risks of mortality and neonatal morbidity.
• Accurate prediction of outcomes is crucial for clinical decision-making and resource allocation.
• Existing predictive models may not fully capture the complexity of factors influencing preterm infant outcomes.

Purpose of the Study:

• To develop and validate a machine learning-based resiliency score.
• To predict survival and survival without severe neonatal morbidity in preterm neonates (<32 weeks gestation).
• To create a tool for adjusting variables in administrative datasets.

Main Methods:

• Utilized a population-based Californian administrative dataset.
• Developed predictive models using maternal, perinatal, and neonatal variables with the LASSO method.
• Assessed model discrimination using internal validation and an external dataset from a tertiary care center.

Main Results:

• Achieved excellent discrimination for survival (c-statistic 0.895) and survival without severe neonatal morbidity (c-statistic 0.867) in the internal validation dataset.
• Maintained high discrimination in the external validation dataset (c-statistics 0.817 and 0.804, respectively).
• The developed resiliency score effectively predicts outcomes in preterm neonates.

Conclusions:

• The validated resiliency score accurately predicts survival and survival without major morbidity in preterm infants (<32 weeks).
• This score offers a valuable tool for adjusting for multiple variables within administrative datasets.
• The findings support the use of machine learning for predicting complex neonatal outcomes.