Dosing Strategy of Ramosetron to Prevent Postoperative Nausea and Vomiting and Development of Prediction Models Using...

The study aimed to compare the postoperative nausea and vomiting (PONV) preventive effect of repeated administration of ramosetron with the standard treatment group and compare models to predict the incidence of PONV using machine-learning techniques.

A total of 261 patients scheduled for breast surgery were analyzed to evaluate the effectiveness of repeated intravenous administration of ramosetron. All patients were administered 0.3 mg ramosetron just before the end of surgery. For the repeated dose of ramosetron group, an additional dose of 0.3 mg was administered at 4, 22, and 46 hours after the end of the surgery. Postoperative nausea, vomiting, and retching were evaluated using the Rhodes Index of Nausea, Vomiting, and Retching at 6, 24, and 48 hours postoperatively. Previously published randomized controlled data were combined with the data of this study to create a new dataset of 1390 patients, and machine-learning-based PONV prediction models (classification tree, random forest, extreme gradient boosting, and neural network) was constructed and compared with the Apfel model.

Fifty patients (38.5%) and 60 patients (45.8%) reported nausea, vomiting, or retching 48 hours postoperatively in the standard and repeated-dose groups, respectively (P = 0.317, χ² test). Median sensitivity, specificity, and accuracy of the Apfel model analyzed using the training set were 0.815, 0.344, and 0.495, respectively.

The repeated administration of ramosetron did not reduce the incidence of PONV. The Apfel model had high sensitivity, however, its specificity and accuracy were lower than that in machine-learning-based models.