Truace
Subscribe

Improving early risk stratification for bronchopulmonary Dysplasia in preterm infants using machine learning: development and external validation

Background: Despite advances in perinatal care and improved survival of preterm infants, bronchopulmonary dysplasia (BPD) remains a major neonatal morbidity with long-term consequences. Accurate early risk prediction is therefore clinically important. Conventional models may be limited in capturing the complex and nonlinear interactions underlying BPD, whereas machine learning offers greater flexibility for high-dimensional clinical data. However, limited interpretability remains a barrier to clinical implementa…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1521· cite & verify
Improving early risk stratification for bronchopulmonary Dysplasia in preterm infants using machine learning: development and external validation
Recording noise, speech and music inside an incubator by DrTrumpet. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers built and compared seven machine learning models to predict bronchopulmonary dysplasia, a major complication of premature birth, using records from 1,083 infants born before 32 weeks at one hospital in China. A gradient boosting model performed best and was then tested on a separate cohort of 1,000 neonates.

If the approach holds up, clinicians could flag high-risk preterm infants earlier using data they already collect, and the interpretability tools show which factors drive risk. The model is meant only for risk stratification, not treatment decisions, and it has not yet been tested prospectively across multiple centers.

Main points

  1. Birth weight, gestational age, and 1-minute Apgar score were the leading predictors of BPD in the model.
  2. Neonatal anemia was strongly associated with BPD, with an adjusted odds ratio of 5.036 (95% CI 3.35-7.57).
  3. Risk rose sharply between 1,250 and 1,500 g birth weight, showing a nonlinear relationship.
  4. Internal sensitivity was 0.827, specificity 0.759, accuracy 0.793, and Brier score 0.141.

The gain

A gradient boosting machine model gave interpretable, individualized early BPD risk estimates for preterm infants using routinely available clinical variables, with internal AUC of 0.880.

The rundown

The retrospective cohort covered preterm infants under 32 weeks gestation and under 2,500 g birth weight admitted to the Affiliated Women's Hospital of Jiangnan University between January 2011 and December 2025. Data were split 7:3 into training and validation sets, and hyperparameters were tuned with 10-fold cross-validation using AUC.

Performance was judged with AUC-ROC, Brier score, and decision curve analysis, which indicated favorable net clinical benefit across relevant threshold probabilities. SHAP and partial dependence plots were used to address the interpretability barrier to clinical use.

Missing data were handled by complete case analysis. The best model was externally validated in an independent cohort of 1,000 neonates.

Sources

  1. Peer-reviewedAnnals of Medicine2026-09-25

The debate