Machine learning to predict adverse events of sedatives in gastrointestinal endoscopy by using a random forest model

Aims Sedation has become standard practice for patients undergoing gastrointestinal (GI) endoscopy. However, considering the serious cardiopulmonary adverse events (AEs) of sedatives, it is important to identify patients at high risk. Nowadays, machine learning across a wide range of medical data can generate reasonable predictions for AEs in the clinical field. This study aims to perform a machine learning using a random forest model to identify predictors of AEs in sedative GI endoscopy.

Methods This observational prospective study enrolled the 462 patients who underwent sedative GI endoscopy in Korea university ansan hospital. The clinical data were used as predictor variables to construct random forest models to forecast the AEs of sedatives.

Results A total of 128 patients (27.7%) showed cardiopulmonary AEs in sedative endoscopy. Among them, 97 had hypoxia (pulse oximetry < 90%) and 31 had tachycardia (heart rate > 100 bpm). Patients group who developed AEs were older, had more male, smokers, heavy drinkers, higher BMI and neck circumference, had longer procedure time, and used propofol alone rather than with midazolam with/without propofol, and needed additional sedatives. The area under the receiver operating characteristic curve for the model based on the random forest for predicting AEs in sedative endoscopy was 0.82 (95% CI: 0.79-0.86).

Conclusions We constructed a random forest model to predict AEs during sedative GI endoscopy with acceptable performance.

Publikationsverlauf

Artikel online veröffentlicht:
14. April 2023