Predicting Post-surgery Discharge Time in Pediatric Patients Using Machine Learning.

Area of Science:

• Pediatric Surgery
• Health Informatics
• Machine Learning in Healthcare

Background:

• Prolonged hospital stays after pediatric tonsillectomy and adenoidectomy increase costs and impact patient care.
• Identifying determinants of extended hospitalization is vital for optimizing postoperative management and resource allocation.

Purpose of the Study:

• To apply machine learning (ML) for predicting post-surgery discharge times in pediatric patients.
• To identify key variables influencing hospital stay duration after tonsillectomy and/or adenoidectomy.

Main Methods:

• Analysis of data from 423 pediatric patients undergoing tonsillectomy/adenoidectomy.
• Data preprocessing included handling missing values, outlier detection, and variable transformation.
• Four ML algorithms (Random Forest, Logistic Regression, RUSBoost, AdaBoost) were trained and evaluated using 10-fold cross-validation.

Main Results:

• Postoperative nausea and vomiting (PONV), dexmedetomidine infusion, fentanyl use, discharge pain, and extubation time were significant predictors of delayed discharge.
• The AdaBoost model achieved high accuracy and reliability in predicting discharge times.

Conclusions:

• ML models effectively predict pediatric discharge times and pinpoint critical factors for prolonged hospitalization.
• These predictive insights can refine postoperative care strategies and resource management in pediatric surgery.
• Integration of ML models into clinical practice can support real-time decision-making.