Predicting mental and psychomotor delay in very pre-term infants using machine learning.
Gözde M Demirci1, Phyllis M Kittler2,3, Ha T T Phan2,3
1Computer Science Department, The Graduate Center of the City University of NY, New York, NY, USA.
Pediatric Research
|July 27, 2023
Summary
Machine learning models can predict neurodevelopmental delays in very preterm infants early. Combining perinatal and longitudinal data improves prediction accuracy, enabling timely interventions for better outcomes.
Area of Science:
- Neonatal Medicine
- Developmental Pediatrics
- Computational Biology
Background:
- Very preterm infants face a higher risk of neurodevelopmental delays.
- Early prediction of these delays is crucial for timely intervention and improved long-term outcomes.
- Machine learning (ML) offers a promising approach for predicting developmental trajectories.
Purpose of the Study:
- To develop and validate ML models for predicting mental and psychomotor delay at 25 months in very preterm infants.
- To identify key predictors of neurodevelopmental delay using both perinatal and longitudinal data.
- To assess the efficacy of combining different data types for enhanced predictive accuracy.
Main Methods:
- A RandomForest classifier was applied to data from 1109 very preterm infants.
- Key predictors were selected from 52 perinatal and 16 longitudinal variables (1-22 months).
- SHapley Additive exPlanations (SHAP) were used for model interpretability.
Main Results:
- Models using only perinatal variables achieved 62% (mental) and 61% (psychomotor) balanced accuracy.
- Top predictors included birth year, hospital stay duration, antenatal magnesium sulfate, intubation duration, birth weight, cranial ultrasound findings, gestational age, maternal age and education, and intrauterine growth restriction.
- Combining 19-month follow-up scores with perinatal variables yielded the highest balanced accuracy at 72% (mental) and 73% (psychomotor).
Conclusions:
- ML models integrating perinatal and longitudinal data can predict 24-month mental/psychomotor delay in very preterm infants approximately six months in advance.
- This early prediction facilitates earlier intervention, potentially improving developmental outcomes.
- While perinatal features alone were insufficient for clinical application, their combination with longitudinal data significantly enhanced predictive power. Recent birth year correlated with a reduced likelihood of delay, possibly due to medical advancements.
