To the Editor:—
Although we were happy to welcome a further addition to the literature on postoperative nausea and vomiting by Sinclair and colleagues, we were amazed by its omissions and a little more by its contents. In April 1998, an editorial posed the question “Can we predict who will vomit after surgery?”1and we were therefore surprised that this was not quoted by Sinclair and colleagues in their similarly entitled publication “Can postoperative nausea and vomiting be predicted?”, which was submitted about half a year later. 2We were even more surprised that the authors stated in the introduction that “the degree to which factors are predictors of PONV remains unknown.” This is plainly incorrect since a number of authors have attempted to quantify risk factors for postoperative nausea and vomiting (PONV) using logistic regression models. 3–7None of these studies were quoted in the introduction and some were briefly mentioned in the discussion only to be dismissed. The introduction gives the misleading impression that a completely new idea and concept has been developed. 2
Palazzo and Evans were the first to use logistic regression analysis to quantify the relative impact of fixed patient factors on the probability of PONV in 1993. 3Their study was criticized for being only applicable to one type of surgery. 2However, Dr Sinclair and colleagues failed to mention that this model was tested by Toner et al. in patients from a different hospital having a wide spectrum of operations with all sortsof anesthetists and anesthetics to test its robustness as a model. 8The study of Koivuranta et al . 6was criticized for no analysis of anesthesia-related factors. Again, this was misquoted as general and regional anesthesia were compared and described in the article. 6The model developed by Apfel et al . 7was criticized for lack of analysis of anesthesia-related factors in spite of the fact that induction agents, muscle relaxants, volatile anesthetics, etc., and their dosages were considered in the analysis. Anesthetic agents were not included in the final model because they were found to be statistically nonsignificant. 7What is more Apfel et al. were able to show, that their operation-independent score was able to predict postoperative vomiting in other types of surgery and that a score with consideration of the type of surgery did not increase the accuracy of their model. 9We wonder, why this study was also not quoted.
More importantly, we are concerned that flaws in data acquisition and analysis in Dr Sinclair’s study may have led to wrong conclusions.
Postoperative nausea and vomiting is usually defined as any episode of nausea or vomiting within 24 h. 10Sinclair et al. have defined in hospital PONV as “any volunteered report of nausea or observed active retching or vomiting requiring antiemetics.” It remains unclear whether “requiring antiemetic” is related to vomiting alone, active retching or vomiting, or to all three symptoms. Was the need for antiemetics standardized at all? As patients are usually not provided with rescue antiemetics for emetic sequelae after hospital discharge it seems that the definition of PONV outside hospital, which was based on a telephone interview, must have been different to that within hospital. It is possible that this difference in definitions could have been one reason for the low incidence of PONV of 9.1% (according to the text on page 114) or of 7.1% after general anesthesia (according to fig. 1). 2Most previous prospective studies, which have explicitly and separately asked for nausea and vomiting at repeated time intervals have reported average incidences between 20% to 30%. 3,5–9Sinclair et al. have suggested that the low incidence was most likely explained by an “under-reporting by PACU and ASU nurses.” Either way a score whether based on unclear definitions or possibly incomplete nurse assessment of PONV may lead to a systematic error with an underestimation of the real risk for PONV. Therefore, it remains unclear whether the reported three to sixfold incidence after orthopedic procedures is a reflection of the accuracy of documentation or the type of surgery. Thus, the impact of type of surgery on a score which has been based on incomplete data becomes questionable. This is particularly the case in this instance where the relatively low incidence of PONV after peripheral surgery is well known. 11
A further statistical difficulty with this study is that a broad spectrum of factors was fitted to a linear logistic regression model without testing for potential interactions or nonlinearities. For example Sinclair et al. should have considered an interaction between the duration and the type of anesthesia since previous work suggests that duration appears to be relevant for general 5–7but not for other techniques such as regional anesthesia. 12Failure to consider these interactions might lead to overestimates of the risk of PONV for long regional procedures. The importance of considering nonlinearity becomes more apparent when considering the relationship between age and PONV which is bimodal. In children the incidence increases with age, 11whereas in the adult population the tendency is for a decrease in PONV with increasing age. 5,7,13It is obvious that a score which includes age as a linear function would exaggerate the risk of very young children if based on a score developed mainly from adults.
Sinclair et al. have claimed that their study provides “the most comprehensive logistic regression model of patient-, anesthesia-, and surgery-related factors associated with PONV.”2However, their anesthesia related factors only considered whether general or regional anesthesia was given, with no reference to drugs given or extent of block. In fact, Carpenter et al. have shown that the choice of drugs for spinal anesthesia as well as the block height above Th4 appeared to be relevant. 12It would seem that the score proposed by Sinclair et al. is not so comprehensive as claimed.
In conclusion, aside from being disappointed that evidence of previous studies was either misrepresented or incomplete, there would appear to be some flaws in data acquisition and analysis which may have led to bias in the results thereby rendering the suggested model less applicable for other centers.
