Predicting Reintubation in Postoperative Pediatric Cardiac Surgery: A Machine Learning Approach

Area of Science:

• Pediatric Cardiac Surgery
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Accurate prediction of reintubation is vital for pediatric cardiac surgery patients.
• This study identifies key predictors and develops a predictive model.

Purpose of the Study:

• To identify significant predictors of reintubation in pediatric patients post-cardiac surgery.
• To train a multilayer perceptron (MLP) neural network for reintubation prediction.

Main Methods:

• Retrospective analysis of 294 pediatric patients (1-24 months) undergoing cardiac surgery and mechanical ventilation.
• Pearson Chi-square (PC²) test and binomial logistic regression analysis (BLRA) for predictor identification.
• MLP neural network trained on clinical covariates for prediction.

Main Results:

• Key predictors identified: low BMI, emergency surgery, previous infection, pre-reintubation arterial blood gas (ABG) levels, and procedure type (aortoplasty).
• Duration of ventilation and RACHS2 score were also significant predictors.
• MLP model achieved high accuracy: 93.7% sensitivity, 90.5% specificity, 0.94 F1-score, and 0.94 AUC.

Conclusions:

• The MLP neural network demonstrates excellent predictive accuracy for reintubation risk factors.
• This model can enhance postoperative care for pediatric cardiac surgery patients.