Development and validation of a machine learning model predicting post-tonsillectomy hemorrhage
Anker Stubberud1,2,3, Sverre Morten Zahl4, Tor Åge Myklebust5,6
1Department of Otolaryngology, Helse Møre Og Romsdal Hospital Trust, Ålesund, Norway. anker.stubberud@ntnu.no.
Summary
Machine learning models can predict post-tonsillectomy hemorrhage with moderate accuracy. Key predictors include age, sex, and surgical hemostasis techniques, offering potential clinical decision-support.
Area of Science:
- Otorhinolaryngology
- Medical Informatics
- Machine Learning
Background:
- Post-tonsillectomy hemorrhage is a significant complication.
- Predicting hemorrhage risk is crucial for patient management.
Purpose of the Study:
- To develop and validate machine learning models for predicting post-tonsillectomy hemorrhage.
- To identify key predictors of post-tonsillectomy bleeding.
Main Methods:
- Analysis of a large cohort from the Norwegian tonsil registry (32,037 patients).
- Development of supervised machine learning models using perioperative data.
- Evaluation of model performance using Area Under the Curve (AUC) and SHAP plots.
Main Results:
- The Adaboost classifier achieved an AUC of 0.71 for hemorrhage prediction.
- Significant predictors included older age, male sex, and bipolar diathermy for hemostasis.
- The model demonstrated superior performance compared to alternative strategies.
Conclusions:
- Machine learning models can predict post-tonsillectomy hemorrhage with moderate accuracy.
- Further research is needed to establish clinical utility as a decision-support tool.