Purpose: This study aimed to develop and validate a prediction model for rescue antiemetic treatment in the post-anesthetic care unit (PACU) using machine learning techniques.
Methods: Data from two academic hospitals (one for the derivation cohort a...
Purpose: This study aimed to develop and validate a prediction model for rescue antiemetic treatment in the post-anesthetic care unit (PACU) using machine learning techniques.
Methods: Data from two academic hospitals (one for the derivation cohort and the other for the external validation cohort) were collected. A gradient boosting machine (GBM) model was developed to predict rescue antiemetic treatment in the PACU, with input variables selected by the Boruta algorithm from automatically collected electronic health records. The model’s performance, measured by area under the receiver operating characteristic curve (AUROC), was compared with the Apfel and the Koivuranta scores in both internal and external validation.
Results: Four variables were selected by the Boruta algorithm: sex, age, anesthetic duration, and volatile anesthetics. In the internal validation, the GBM model (AUROC, 0.804) showed significantly higher predictive performance than the Apfel (AUROC, 0.637) and Koivuranta (0.698) scores (P<0.001 for both). In the external validation, the GBM model (AUROC, 0.731) also showed significantly higher predictive performance than the Apfel (AUROC, 0.638) and Koivuranta (0.675) scores (P<0.001 for both).
Conclusions: The GBM model with automatically collectible variables showed better performance than traditional risk scores in predicting rescue antiemetics in the PACU. This model can help clinicians identify high-risk patients, adjust anesthetic methods, or prepare timely interventions in the PACU.