Comparison of Multivariable Logistic Regression and Machine Learning Models for Predicting Bronchopulmonary Dysplasia or Death in Very Preterm Infants.

Faiza Khurshid,Helen Coo,Joseph Y Ting,Amal Khalil,Jonathan Wong,Jonathan Messiha,Prakesh S Shah

doi:10.3389/fped.2021.759776

Abstract

Bronchopulmonary dysplasia (BPD) is the most prevalent and clinically significant complication of prematurity. Accurate identification of at-risk infants would enable ongoing intervention to improve outcomes. Although postnatal exposures are known to affect an infant's likelihood of developing BPD, most existing BPD prediction models do not allow risk to be evaluated at different time points, and/or are not suitable for use in ethno-diverse populations. A comprehensive approach to developing clinical prediction models avoids assumptions as to which method will yield the optimal results by testing multiple algorithms/models. We compared the performance of machine learning and logistic regression models in predicting BPD/death. Our main cohort included infants <33 weeks' gestational age (GA) admitted to a Canadian Neonatal Network site from 2016 to 2018 (n = 9,006) with all analyses repeated for the <29 weeks' GA subcohort (n = 4,246). Models were developed to predict, on days 1, 7, and 14 of admission to neonatal intensive care, the composite outcome of BPD/death prior to discharge. Ten-fold cross-validation and a 20% hold-out sample were used to measure area under the curve (AUC). Calibration intercepts and slopes were estimated by regressing the outcome on the log-odds of the predicted probabilities. The model AUCs ranged from 0.811 to 0.886. Model discrimination was lower in the <29 weeks' GA subcohort (AUCs 0.699–0.790). Several machine learning models had a suboptimal calibration intercept and/or slope (k-nearest neighbor, random forest, artificial neural network, stacking neural network ensemble). The top-performing algorithms will be used to develop multinomial models and an online risk estimator for predicting BPD severity and death that does not require information on ethnicity.

Highlights

Bronchopulmonary dysplasia (BPD), a form of chronic lung disease, is the most common morbidity among very preterm neonates [1]
Assessing model calibration is important with machine learning approaches: while logistic regression models can be recalibrated by reestimating the intercept [22] and updating the regression coefficients [51], methods for improving the calibration of machine learning models are “unclear” [59]
It is possible that the relationship between the predictors we examined and the outcome of BPD/death is adequately captured by fitting standard logistic regression models, even without the inclusion of interaction terms

Summary

Introduction

Bronchopulmonary dysplasia (BPD), a form of chronic lung disease, is the most common morbidity among very preterm neonates (i.e., those born before 33 weeks of gestation) [1]. The Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) hosts an online BPD risk estimator that allows users to predict the probability of mild, moderate and severe BPD, and death, on postnatal days 1, 3, 7, 14, 21, and 28 [12]. This risk estimator has not been validated in the Canadian population and was developed using data from 2000 to 2004 [13]. Most other BPD prediction models only allow an infant’s risk to be estimated at one point in time [reviewed in [16, 17]] and have limited clinical utility, given that postnatal exposures are known to affect the likelihood of developing BPD [1]

Objectives

Methods

Results

Discussion

Conclusion