Artificial neural network for risk assessment in preterm neonates

Area of Science:

• Neonatal Medicine
• Artificial Intelligence in Healthcare
• Predictive Analytics

Background:

• Preterm infants (< 32 weeks gestational age and/or < 1500 g birthweight) face significant mortality risks.
• Accurate prediction of individual mortality risk is crucial for clinical management.
• Existing predictive models may have limitations in precision.

Purpose of the Study:

• To develop and evaluate an artificial neural network (ANN) for predicting individual neonatal mortality risk in preterm infants.
• To compare the performance of the ANN against traditional logistic regression models.

Main Methods:

• A retrospective study involving 890 preterm neonates.
• An ANN was trained on admission data from infants born between 1990-1993.
• The ANN's predictive accuracy was validated on infants born in subsequent years (1994-1996).

Main Results:

• The ANN demonstrated superior predictive performance compared to logistic regression (AUC 0.95 vs 0.92).
• A predicted mortality risk > 0.50 was associated with high morbidity.
• Mortality risks for two non-survivors with birthweights > 2000 g and severe congenital disease were underestimated.

Conclusions:

• An ANN trained on admission data can accurately predict mortality risk for the majority of preterm infants.
• Despite high overall accuracy, prediction failures indicate the model is currently unsuitable for individual treatment decisions.
• Further refinement is needed to improve accuracy for complex cases and guide clinical interventions.