A Machine Learning Classifier Improves Mortality Prediction Compared With Pediatric Logistic Organ Dysfunction-2 Score: Model Development and Validation

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Pediatric critical illness (n=14237). Intervention: Random forest machine learning model vs. Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score. Primary outcome: PICU mortality prediction (AUC) (95% CI 0.863-0.895, p=0.003).

Remi D. Prince; Alireza Akhondi‐Asl; Nilesh M. Mehta; Alon Geva

doi:10.1097/cce.0000000000000426

May 1, 2021Critical Care ExplorationsOpen Access

A Machine Learning Classifier Improves Mortality Prediction Compared With Pediatric Logistic Organ Dysfunction-2 Score: Model Development and Validation

View Full Paper

Ask AI

Bookmark

Share

Key result

Random forest model outperforms PELOD-2 score in predicting PICU mortality with ~0.87 AUC.

95% CI 0.863-0.895
P=0.003
n=14,237

Figures

Why the study?

It was unknown whether machine learning algorithms could better predict PICU mortality than the Pediatric Logistic Organ Dysfunction-2 score.

Does a machine learning model improve prediction of PICU mortality compared to the Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score in critically ill pediatric patients?

Study Design

Type

Observational (n=14,237)

Multicenter

No

Structured PICO

Does a machine learning model improve prediction of PICU mortality compared to the Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score in critically ill pediatric patients?

P

Population

14,237 pediatric patients admitted to a quaternary care medical-surgical PICU from 2013 to 2019 (10,194 in training cohort, 4,043 in validation cohort), excluding patients whose primary reason for admission relates to congenital or acquired pediatric heart disease.

I

Intervention

Machine learning algorithms (specifically a random forest model, PELODRF) using the same variables as the Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score to predict PICU mortality.

C

Comparator

Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score and a locally retrained PELOD-2 logistic regression model.

O

Outcome

PICU mortality during the index PICU admission.hard clinical

Main Result

Absolute Event Rate: 0.867% vs 0.761%

p-value: p=0.003

A random forest machine learning model using standard clinical variables provides superior discrimination and calibration for predicting PICU mortality compared to the traditional logistic regression-based PELOD-2 score.

Limitations

Frequent missing values typical of ICU databases
Imputation method assumed missing values were normal
Limited to predictors included in the PELOD-2 model
Single-center study requiring external validation
Frequent missing values requiring imputation
Potential systematic biases due to differences in patient populations compared to previous PELOD-2 studies

Implication

May aid PICU risk stratification; leaves open prospective validation before clinical adoption.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Prince et al. (2021) conducted an observational in Pediatric critical illness (n=14,237). Random forest machine learning model vs. Pediatric Logistic Organ Dysfunction-2 (PELOD-2) score was evaluated on PICU mortality prediction (AUC) (95% CI 0.863-0.895, p=0.003). A random forest machine learning model significantly outperformed the Pediatric Logistic Organ Dysfunction-2 score in predicting PICU mortality, achieving an AUC of 0.867 compared to 0.761.

synapsesocial.com/papers/6a18cd86cac83a2bcecfbeb9 https://doi.org/10.1097/cce.0000000000000426

View Full Paper

Ask AI

Bookmark

Share

A Machine Learning Classifier Improves Mortality Prediction Compared With Pediatric Logistic Organ Dysfunction-2 Score: Model Development and Validation

Figures

Study Design

Structured PICO

Main Result

Limitations

Implication

Discussion

Cite This Study

A Machine Learning Classifier Improves Mortality Prediction Compared With Pediatric Logistic Organ Dysfunction-2 Score: Model Development and Validation

Figures

Study Design

Structured PICO

Main Result

Limitations

Implication

Discussion

Cite This Study

Keep reading

Keep reading