Key result
Adding preoperative biomarkers to clinical data improves machine learning prediction of poor neonatal CPB outcomes.
Why the study?
Neonates undergoing surgery for congenital heart disease are at risk for poor postoperative outcomes after cardiopulmonary bypass, and current prognostic models rely predominantly on nonmodifiable clinical factors.
Does incorporating preoperative blood biomarkers into machine learning models improve the prediction of 30-day postoperative morbidity and mortality (ICU-30) compared with clinical features alone in neonates undergoing cardiopulmonary bypass for congenital heart disease?
Cohort (n=144)
No
Does incorporating preoperative blood biomarkers into machine learning models improve the prediction of 30-day postoperative morbidity and mortality (ICU-30) compared with clinical features alone in neonates undergoing cardiopulmonary bypass for congenital heart disease?
In neonates undergoing cardiopulmonary bypass for congenital heart disease, machine learning models utilizing preoperative serum biomarkers (particularly NGAL, TFF3, ADAMTS13, and Ang-2) improved the prediction of 30-day composite postoperative morbidity and mortality compared with clinical variables alone.
Answers are grounded in this paper and cite their sources.
May refine preoperative risk stratification after neonatal CPB; leaves open external validation before clinical adoption.
IMPORTANCE: Neonates undergoing surgery for congenital heart disease (CHD) are at risk for poor postoperative outcomes after cardiopulmonary bypass (CPB), including death, prolonged ICU stay, and ICU readmission. Current prognostic models rely predominantly on nonmodifiable clinical factors. OBJECTIVES: To utilize machine learning methods to compare the predictive utility of preoperative blood biomarkers vs. clinical features alone for ICU-30, a validated composite outcome after neonatal CPB that includes postoperative 30-day morbidity and mortality. DESIGN, SETTING, AND PARTICIPANTS: Prospective observational single-center cohort study of consecutively enrolled neonates younger than 30 days undergoing CPB for CHD. MAIN OUTCOMES AND MEASURES: Plasma and clinical data were collected from consecutively enrolled neonates younger than 30 days immediately before CPB. Thirty-seven biomarkers were measured by individual or multiplex enzyme-linked immunosorbent assay and tested for association with ICU-30, defined as 1) mortality within 30 days of CPB, 2) ICU stay longer than 30 days, or 3) ICU readmission within 30 days. Predictive performance of extreme gradient boosting (XGBoost), least absolute shrinkage and selection operator, and random forest (RF) models was compared using the area under the receiver operating characteristic curve (AUROC). For XGBoost, the top 20 features were identified based on model importance (gain). RESULTS: Biomarkers were available for 144 neonates. Most were male (51%), White (64%), and non-Hispanic (75%). Most operations were Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery (STAT) category 3 (32%) or STAT category 5 (29%). ICU-30 occurred in 29 neonates (20%). Using combined clinical and biomarker data, XGBoost demonstrated moderate discrimination for ICU-30 (AUROC, 0.74), outperformed RF (AUROC, 0.72), and LASSO (AUROC, 0.70). Biomarkers alone performed better (AUROC, 0.72) than clinical data alone (AUROC, 0.67) with XGBoost. The top two features in the combined XGBoost model were neutrophil gelatinase-associated lipocalin (NGAL) and trefoil factor 3 (TFF3). Using Shapley additive explanation to assess both predictive impact and directionality, the most influential features were NGAL, a disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13 (ADAMTS13), and angiopoietin-2 (Ang-2). CONCLUSIONS AND RELEVANCE: Among neonates with CHD requiring CPB, multibiomarker models improved prediction of poor postoperative outcome compared with clinical variables alone. The most informative model features included markers of renal injury/perfusion and gastrointestinal epithelial injury (NGAL and TFF3), as well as vascular injury and inflammation (ADAMTS13 and Ang-2). These findings support further investigation of biomarker-informed risk stratification and may provide insight into pathophysiologic mechanisms underlying adverse postoperative outcomes.
No takes yet. Share an insight, caveat, or question.
Gardner et al. (2026) conducted a cohort in Congenital heart disease requiring cardiopulmonary bypass (n=144). Preoperative blood biomarkers (37 biomarkers) vs. Clinical features alone was evaluated on ICU-30 (composite of mortality within 30 days of CPB, ICU stay >30 days, or ICU readmission within 30 days) (AUROC 0.72 vs 0.67 (0.74 combined)). Preoperative serum biomarkers improved XGBoost prediction of 30-day poor postoperative outcomes (ICU-30) after neonatal cardiopulmonary bypass vs clinical data alone.
Synapse has enriched 10 related papers. Consider them for comparative context: