We are pleased that our new score to predict a patient’s risk of postdischarge nausea and vomiting (PDNV score) that helps clinicians to decide on an antiemetic regimen has been positively received.1
Generally, the advantage of performing a multiple logistic regression analysis is to determine the effect of one variable while all other variables are held constant, hence inherently eliminating any source of confounding. This works fairly well unless there is a very high correlation among two variables, which was not the case in our dataset, i.e., the use of opioids and the development of nausea in the postanesthesia care unit (PACU) for the prediction of PDNV did not correlate strongly enough to affect prediction. In other words, the coefficients for opioids and nausea in the PACU would remain very similar even if one of these parameters were taken out of the equation. Of note, in some datasets, gender and history of postoperative nausea and vomiting do not only correlate, but also interact in the way they affect outcome. However, we have previously established that considering such an interaction does little to improve the predictive accuracy as measured by the area under the receiver operating characteristic curve.2 These previous results corroborate the notion to use a simplified model.
That said, nausea and vomiting in the PACU were highly correlated as shown in our Venn diagram (fig. 1).1 Furthermore, in simple bivariate analyses, both nausea and vomiting in the PACU were predictive for PDNV. However, because vomiting in the PACU was only 3.9% and nausea was 19.9%, nausea in the PACU is more informative for predicting PDNV and vomiting in the PACU no longer provides statistically significant additional information for predicting PDNV.
Also, certain factors that are significant for nausea in the PACU are not significant for PDNV; yet, nausea in the PACU itself is significant for PDNV. As there is no statistically significant direct connection between these certain factors and PDNV, suggesting that they actually are mediated through nausea in the PACU, they give us no additional clinically useful information compared to simply considering nausea in the PACU.
We have developed this risk score for PDNV as a key parameter to assess the need for prophylactic regimens. However, we do not think it should be the only factor. For example, we might be more liberal with prophylactic antiemetics in a patient who underwent retinal surgery where vomiting after surgery might lead to retinal bleeding and irreversible blindness. Therefore, we did not explicitly link the patient’s risk for PDNV with certain antiemetics regimen. The number needed to treat has inherent limitations when it is used to compare the efficacy of different antiemetics, especially because it is so dependent on the baseline risk.3 Assuming a relative risk reduction of 26% for the average prophylactic antiemetic,4 the numbers needed to treat vary from 35 (0 risk score) to 5 (risk score of 5) for single-drug prophylaxis and from 20 (risk score of 0) to 3 (risk score of 5) for two-drug prophylaxis (table 1). In light of these numbers needed to treat, we suggest that single-drug antiemetic prophylaxis be considered in patients with a score of at least 1 and two-drug or more antiemetic prophylaxis be considered in patients with a score of 3 or higher.
Using this risk score improves predictive value over simply looking at laparoscopic versus nonlaparoscopic procedures. It is a valid point that laparoscopic surgery is a significant risk factor for postoperative nausea and vomiting in the PACU, but adding it as an additional risk factor for PDNV does not improve the area under the receiver operating characteristic curve of our simplified risk score.
The perpetual question with creating risk scores is how they will be implemented in clinical practice and how they affect patient outcomes. Based on specific patient characteristics, we can accurately establish a patient’s risk for experiencing PDNV. We also have validated data that certain interventions are shown to improve PDNV outcome. Thus, studies that seek to determine if risk scores really improve outcome are actually testing how effectively physicians/hospitals are able to implement the use of risk scores in making clinical decisions. Kalkman et al. are doing fascinating research and their work may provide further insight into factors that influence clinical adaptation of risk scores (stay tuned!).
In conclusion, we appreciate the thoughtful comments and hope that our response has helped to understand the clinical validity of our risk score and why we would like to leave it to the practitioner how to utilize patient’s risk as objective information in the approach to prevent nausea and vomiting.