Related Experiment Video
Updated: Aug 15, 2025
A Novel Method for Involving Women of Color at High Risk for Preterm Birth in Research Priority Setting
Published on: January 12, 2018
Predicting the risk of 7-day readmission in late preterm infants in California: A population-based cohort study
Ribka Amsalu1, Scott P Oltman1,2, Melissa M Medvedev3,4
1California Preterm Birth Initiative University of California San Francisco San Francisco California USA.
Insights
This study developed a model to predict 7-day hospital readmission for late preterm infants. The model identified key risk factors, helping to stratify infants and target interventions for better health outcomes.
Area of Science:
- Neonatalogy
- Pediatrics
- Public Health
Background:
- Late preterm infants (34-36 weeks gestation) face higher risks of morbidity and readmission than term infants.
- Existing prediction models for specific morbidities are limited for early readmission risk in this population.
Purpose of the Study:
- To derive and validate a predictive model for 7-day hospital readmission in late preterm infants.
- To identify key maternal and infant variables associated with early readmission.
Main Methods:
- Population-based retrospective cohort study using California birth and hospital discharge data (2007-2011).
- Utilized random forest and logistic regression to develop and validate a predictive model.
- Evaluated model performance using discrimination (c-statistic) and calibration (Brier score).
Main Results:
- A model with 24 variables demonstrated better predictive ability (c-statistic 0.644) than an 8-variable logistic model.
- Identified eight key predictors: length of stay, delivery method, parity, gestational age, birthweight, race/ethnicity, phototherapy, and diabetes.
- Risk stratification identified approximately 19% of infants at highest risk for readmission.
Conclusions:
- The developed 7-day readmission predictive model shows moderate performance in identifying at-risk late preterm infants.
- Further research should incorporate additional variables and focus on hospital practices to mitigate readmission risks.
Background And Aims:
The American Academy of Pediatrics describes late preterm infants, born at 34 to 36 completed weeks' gestation, as at-risk for rehospitalization and severe morbidity as compared to term infants. While there are prediction models that focus on specific morbidities, there is limited research on risk prediction for early readmission in late preterm infants. The aim of this study is to derive and validate a model to predict 7-day readmission.
Methods:
This is a population-based retrospective cohort study of liveborn infants in California between January 2007 to December 2011. Birth certificates, maintained by California Vital Statistics, were linked to a hospital discharge, emergency department, and ambulatory surgery records maintained by the California Office of Statewide Health Planning and Development. Random forest and logistic regression were used to identify maternal and infant variables of importance, test for association, and develop and validate a predictive model. The predictive model was evaluated for discrimination and calibration.
Results:
We restricted the sample to healthy late preterm infants (n = 122,014), of which 4.1% were readmitted to hospital within 7-day after birth discharge. The random forest model with 24 variables had better predictive ability than the 8 variable logistic model with c-statistic of 0.644 (95% confidence interval 0.629, 0.659) in the validation data set and Brier score of 0.0408. The eight predictors of importance length of stay, delivery method, parity, gestational age, birthweight, race/ethnicity, phototherapy at birth hospitalization, and pre-existing or gestational diabetes were used to drive individual risk scores. The risk stratification had the ability to identify an estimated 19% of infants at greatest risk of readmission.
Conclusions:
Our 7-day readmission predictive model had moderate performance in differentiating at risk late preterm infants. Future studies might benefit from inclusion of more variables and focus on hospital practices that minimize risk.
