Study of postoperative nausea and vomiting: recommending risk models for group comparisons

Having read with great interest the recent paper (Thomas et al. Anaesthesia 2002; 57: 1119–28), as it questions the validity of predictive models [1–4] that we endorse, we would like to make the following observations. The authors compared the incidence of postoperative nausea and vomiting (PONV) predicted by several risk models, based on the demographic characteristics of patients from their previous study [5]. The predicted risks did not correlate very well. However, to conclude that the use of risk scores for group comparisons is limited, ignores the fact that some scores may be considerably better than others [6,7]. Furthermore, if the authors wish to determine whether one of the risk models provides an acceptable prediction, the method of Bland and Altman is certainly not the appropriate choice [8]. This relatively simple but elegant method was originally developed ‘for assessing agreement between two methods of clinical measurement’ in the absence of a gold standard. However, in risk modelling the gold standard is clearly the actual outcome, i.e. the estimated incidence should be compared with the number of patients that actually suffered PONV or not. Unfortunately, such a comparison was not performed. Therefore, neither the presented material nor the performed statistical analysis in the discussed paper [4] allows one to conclude that risk scores in general do not provide an appropriate estimate for patient risk (and therefore are not useful for group comparisons). In contrast, the simplified risk scores [3] did provide a realistic estimate of the patients risk for PONV in adults as recently shown in several other centres [6,7,9]. For this and other reasons, we would strongly recommend the use of a simplified risk score for group comparisons in anti-emetic trials in adults [10]. The only exception is paediatric anaesthesia due to the lack of a reliable predictive model for children. This is also the reason why a risk score for group comparison was not used in a recent study, since 593 out of 1180 patients were children [11]. If the authors are still not convinced, we would be pleased to support them by performing a validation study of risk scores in their own centre. We thank Dr Apfel and his colleagues for their interest in our paper and would like to respond to the comments they have made. Our study demonstrated that different scoring systems produce very different estimates of the incidence of postoperative nausea and vomiting (PONV). We found this both perplexing and surprising bearing in mind the similarity of the demographic details that make up the systems that we analysed. While we accept Apfel's statement that some scoring systems may be considerably better than others, without a placebo group, no researcher is going to be able to know which system will be the best or most accurate. We still maintain that the inclusion of a placebo group within an anti-emetic trial involving ambulatory surgery for gynaecological patients is unethical. For the reason stated above, by necessity, we were working without a known ‘gold standard’. Our decision to use the method of Bland and Altman was therefore quite correct. Furthermore, it would have been impossible to perform a comparison in the absence of such a gold standard. This limitation of our study was covered fully in our original discussion. We quite understand the attractiveness of a simplified risk score to be able to either predict anti-emetic requirements or to use within the structure of a randomised control trial. What we believe we have demonstrated is that all the current instruments are too blunt, since moderate accuracy is probably the best one can hope for when trying to construct risk scores by current techniques of modelling. We thank Professor Apfel for the offer of support in performing a validation of Apfel's risk score in our own centre and would welcome such co-operation. Indeed we are still confused as to how we should be scoring occasional smokers and patients with no previous anaesthetic history (do they really have the same risk of PONV as patients who have had a general anaesthetic and did not suffer from PONV?). Even gender has become contentious and the boundaries blurred. Is the risk of PONV genotypically, phenotypically or hormonally derived? Moreover, we wish to draw on professor Apfel's teams expertise in testing the utility of his scoring system by challenging the null hypothesis: ‘there is no difference in PONV in a population prescribed the currently available anti-emetics based on derived risk (using Apfel's scoring system) and a population where prescribing is left to chance'. R. ThomasN. Jones P. Strike Salisbury District Hospital, Salisbury, UK E-mail: [email protected]