JMIR Publications

ABSTRACT

Background:

Hyperbilirubinemia affects many newborns and if not appropriately treated can result in irreversible brain injury.

Objective:

We sought to develop predictive models of follow-up total serum bilirubin measurement and to compare accuracy with clinician predictions.

Methods:

Subjects were patients born between June 2015 and June 2019 at four Massachusetts hospitals. The prediction target was a follow-up total serum bilirubin measurement obtained < 72 hours after a prior measurement. Birth before versus after February 2019 was used to generate a training set (27,428 target measurements) and held-out test set (3,320 measurements) respectively. Multiple supervised learning models were trained. To assess model performance, predictions on the held-out test set were also compared with corresponding predictions from clinicians.

Results:

The best predictive accuracy on the held-out test set was obtained with the neural network (mean absolute error, MAE 1.05 mg/dL) and Xgboost (MAE 1.04 mg/dL) models. A limited number of predictors was sufficient for best performance and avoiding overfitting. Clinicians made a total of 210 prospective predictions. The neural network model accuracy on this subset of predictions had a MAE of 1.06 mg/dL, compared to clinician predictions with MAE 1.38 mg/dL (P < 0.0001).

Conclusions:

We have developed predictive models for newborn follow-up total serum bilirubin which outperform clinicians. This may be the first report of models that predict specific bilirubin values, are not limited to near-term patients without risk factors, and which take into account the effect of phototherapy.