What is the clinical evidence from this study?

Study design: Observational. Population: Pediatric patients undergoing noncardiac surgery (n=1023639). Intervention: XGBoost machine learning model vs. Pediatric Risk Assessment (PRAm) score. Primary outcome: 30-day mortality.

What does this research mean for the field?

An XGBoost machine learning model accurately predicts 30-day mortality in pediatric noncardiac surgery but performs similarly to the existing regression-based PRAm score, offering no substantial added clinical benefit. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

August 28, 2026Anesthesia & Analgesia

Machine Learning Modeling for Predicting Mortality in Pediatric Patients Undergoing Elective Noncardiac Surgery: Comparison to a Regression Model

Q: What are the key findings of this study?

An XGBoost machine learning model predicted 30-day mortality with an AUC-ROC of 0.956, performing similarly to the regression-based PRAm score (AUC-ROC 0.958) without substantial added net benefit.

View Full Paper

Ask AI

Bookmark

Key result

XGBoost model shows no benefit over PRAm score for predicting 30-day mortality.

n=1,023,639

Why the study?

Perioperative mortality in children is rare, but accurate preoperative risk stratification is critical to mitigate risk; this study aimed to develop machine learning predictive models for 30-day mortality in pediatric noncardiac surgery and compare them to the regression-based PRAm score.

Do machine learning models improve the prediction of 30-day mortality compared to the regression-based PRAm score in pediatric patients undergoing noncardiac surgery?

Population

1,023,639 pediatric patients under 18 years undergoing noncardiac surgery

Comparison

Machine learning models vs regression-based PRAm score

Design

Retrospective database study

Follow-up

30 days

Authors

SSSteven J. StaffaBoston Children's Hospital EVEleonore ValenciaBoston Children's Hospital VTVirginia TangelUniversity Medical Center Groningen

Discussion

Member takes

Implication

ML models may enhance pediatric perioperative risk stratification; leaves open prospective validation versus PRAm.

Key Points

To develop and internally validate machine learning models for predicting 30-day mortality in pediatric patients undergoing noncardiac surgery and compare their performance against the regression-based Pediatric Risk Assessment (PRAm) score.
Retrospective analysis of the NSQIP-Pediatric database (2012–2022, excluding 2020) including N=1,023,639 pediatric patients (<18 years) undergoing noncardiac surgery.
Random forest and XGBoost models were trained on a 70% training split (n=716,662) and evaluated on a 30% validation split (n=306,977) against the PRAm score.
Thirty-day mortality occurred in 3,522 of 1,023,639 encounters (0.34%).
The XGBoost model achieved an AUC-ROC of 0.956, AUC-PR of 0.179, accuracy of 99.4%, precision of 0.247, and Brier score of 0.003 in the validation set.
The regression-based PRAm score achieved an AUC-ROC of 0.958, showing no substantial difference in net clinical benefit compared to the XGBoost model.

Study Design

Type

Observational (n=1,023,639)

Multicenter

Yes

Structured PICO

Do machine learning models improve the prediction of 30-day mortality compared to the regression-based PRAm score in pediatric patients undergoing noncardiac surgery?

Population

1,023,639 pediatric patients <18 years undergoing noncardiac surgery, evaluated for 30-day mortality risk.

Exposure

Machine learning models (random forest and XGBoost) for predicting 30-day mortality

Comparator

Regression-based Pediatric Risk Assessment (PRAm) score

Outcome

30-day mortalityhard clinical

Machine learning models performed similarly to the regression-based PRAm score for predicting 30-day mortality in pediatric noncardiac surgery, suggesting no meaningful clinical benefit from the added complexity.

Limitations

The ML-based models performed similarly to the regression-based PRAm score, questioning whether their added complexity yields meaningful clinical benefit.

Cite This Study

Staffa et al. (2026) conducted an observational in Pediatric patients undergoing noncardiac surgery (n=1,023,639). XGBoost machine learning model vs. Pediatric Risk Assessment (PRAm) score was evaluated on 30-day mortality. An XGBoost machine learning model predicted 30-day mortality with an AUC-ROC of 0.956, performing similarly to the regression-based PRAm score (AUC-ROC 0.958) without substantial added net benefit.

synapsesocial.com/papers/6a91467bd15324a1df3aa224 https://doi.org/10.1213/ane.0000000000008266

View Full Paper

Ask AI

Bookmark

August 28, 2026Anesthesia & Analgesia

Machine Learning Modeling for Predicting Mortality in Pediatric Patients Undergoing Elective Noncardiac Surgery: Comparison to a Regression Model

View Full Paper

Ask AI

Bookmark

Key result

XGBoost model shows no benefit over PRAm score for predicting 30-day mortality.

n=1,023,639

Why the study?

Do machine learning models improve the prediction of 30-day mortality compared to the regression-based PRAm score in pediatric patients undergoing noncardiac surgery?

Population

1,023,639 pediatric patients under 18 years undergoing noncardiac surgery

Comparison

Machine learning models vs regression-based PRAm score

Design

Retrospective database study

Follow-up

30 days

Authors

SSSteven J. StaffaBoston Children's Hospital EVEleonore ValenciaBoston Children's Hospital VTVirginia TangelUniversity Medical Center Groningen

Discussion

Member takes

Implication

ML models may enhance pediatric perioperative risk stratification; leaves open prospective validation versus PRAm.

Key Points

To develop and internally validate machine learning models for predicting 30-day mortality in pediatric patients undergoing noncardiac surgery and compare their performance against the regression-based Pediatric Risk Assessment (PRAm) score.
Retrospective analysis of the NSQIP-Pediatric database (2012–2022, excluding 2020) including N=1,023,639 pediatric patients (<18 years) undergoing noncardiac surgery.
Random forest and XGBoost models were trained on a 70% training split (n=716,662) and evaluated on a 30% validation split (n=306,977) against the PRAm score.
Thirty-day mortality occurred in 3,522 of 1,023,639 encounters (0.34%).
The XGBoost model achieved an AUC-ROC of 0.956, AUC-PR of 0.179, accuracy of 99.4%, precision of 0.247, and Brier score of 0.003 in the validation set.
The regression-based PRAm score achieved an AUC-ROC of 0.958, showing no substantial difference in net clinical benefit compared to the XGBoost model.

Study Design

Type

Observational (n=1,023,639)

Multicenter

Yes

Structured PICO

Do machine learning models improve the prediction of 30-day mortality compared to the regression-based PRAm score in pediatric patients undergoing noncardiac surgery?

Population

1,023,639 pediatric patients <18 years undergoing noncardiac surgery, evaluated for 30-day mortality risk.

Exposure

Machine learning models (random forest and XGBoost) for predicting 30-day mortality

Comparator

Regression-based Pediatric Risk Assessment (PRAm) score

Outcome

30-day mortalityhard clinical

Limitations

The ML-based models performed similarly to the regression-based PRAm score, questioning whether their added complexity yields meaningful clinical benefit.

Cite This Study

synapsesocial.com/papers/6a91467bd15324a1df3aa224 https://doi.org/10.1213/ane.0000000000008266

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark