Machine Learning vs. Logistic Regression for Predicting Outcomes in Preterm Infants
Author Information
Author(s): Afifi Jehier, Ahmad Tahani, Guida Alessandro, Vincer Michael John, Stewart Samuel Alan
Primary Institution: Dalhousie University
Hypothesis
In very preterm infants (<31 weeks’ gestation), EN and ML methods (RF and XBG) will outperform the traditionally used LR in predicting NDI at 36 months corrected age (CA).
Conclusion
Random Forest was superior to conventional Logistic Regression in predicting neurodevelopmental impairment in very preterm infants.
Supporting Evidence
- Of 811 eligible infants, 663 were included in the analysis.
- 195 infants (29%) developed neurodevelopmental impairment (NDI).
- Random Forest had a higher AUC of 0.79 compared to other methods.
- All models provided comparable discrimination but ML methods were superior to LR.
Takeaway
This study looked at how well different methods can predict if very preterm babies will have development problems later. It found that a new method called Random Forest works better than the old method.
Methodology
A retrospective cohort study comparing four prediction methods (Logistic Regression, Elastic Net, Random Forest, and Gradient Boosting) using clinical predictors from a population-based sample of very preterm infants.
Potential Biases
The population in Nova Scotia is predominantly Caucasian, which may limit the generalizability of the findings to other ethnic groups.
Limitations
The study is retrospective and may have excluded potential predictors affecting the predictive performance; some known predictors were not included due to missing data.
Participant Demographics
Mean gestational age was 28 weeks, mean birth weight was 1137 g, and 52% of participants were male.
Statistical Information
P-Value
0.05
Confidence Interval
95% CI (0.70–0.88)
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website