Applicability of risk scores for postoperative nausea and vomiting in a Taiwanese population undergoing general anaesthesia.

Five popular scoring systems for postoperative nausea and vomiting (PONV) were validated and compared with two new predictive models in a Taiwanese population. Nine hundred and ninety-two patients receiving general anaesthesia in a tertiary hospital were investigated in a prospective observational cohort study. Patient demographic data and the incidence of nausea or vomiting in the first 24 hours after surgery were recorded. The overall incidence of PONV was 42%. The area under the curve (AUC) of the five published PONV risk scoring systems was 0.62 to 0.67. Logistic regression analysis in this study cohort showed that female sex and a history of PONV/car sickness were the only statistically significant independent risk factors for PONV (likelihood ratio test P <0.001).The AUCs of our two-predictor and gender-only models were 0.668 and 0.643, respectively (Nagelkerke R² = 0.122 and 0.109). Goodness-of-fit showed that a two-predictor model predicted outcome that was in agreement with the observed outcome (P=0.973). Both the two-predictor model and the Apfel score had a similar AUC that was significantly different from the AUCs of the other models. The AUC for the gender-only model in our population was similar to that of the simplified Koivuranta and the Palazzo and Evans scores (AUC=0.659 and 0.632; P=0.137 and 0.513 respectively). All AUCs had only moderate discrimination power but our female gender-only model was much simpler. Using female gender as the only predictor of PONV had predictive power with 75% sensitivity and 54% specificity.