Correspondence  |   May 2000
Bias in a Further Model for Predicting PONV May Not Advance Current Knowledge
Author Notes
  • Pittsburgh, Pennsylvania 15213
  • University of Pittsburgh
  • Pittsburgh, Pennsylvania 15213
  • Toronto Western Hospital
  • University of Toronto
  • Toronto, Ontario
Article Information
Correspondence   |   May 2000
Bias in a Further Model for Predicting PONV May Not Advance Current Knowledge
Anesthesiology 5 2000, Vol.92, 1489. doi:
Anesthesiology 5 2000, Vol.92, 1489. doi:
In Reply:—
We are grateful for the opportunity to respond to the remarks of Apfel et al.  Although we are concerned about the degree to which we have been misunderstood and misquoted, we regret that Apfel et al.  have been misled by our introductory statement. By simply stating that we have developed and validated a mathematical model to calculate the risk of postoperative nausea and vomiting (PONV) in a large population of ambulatory surgical patients, we did not intend to suggest that this was a new concept. Our work is different from previous studies, as our study focus is ambulatory surgical patients whereas other work focused more on inpatients.
This would not be supported by the studies which we quoted in the discussion. In our view, presenting the limitations of previous mathematical models does not “dismiss” them. Rather, it outlines areas in which improvements can be made. Although we were not aware of the editorial by Kortilla, due to its recent publication in Acta Anaesthesiol Scand at the time of our submission to Anesthesiology, we welcome his contribution, which highlights the persistent problem of predicting PONV among our patients.
We applaud Palazzo and Evans for the introduction of logistic regression analysis as a tool for quantifying the impact of patient factors on the probability of PONV. Their model was developed for the prediction of PONV among patients undergoing minor orthopedic surgery. When tested by Toner et al.  in a larger, heterogenous patient group, the overall correct prediction rate of the model was 71%. Despite our omission of this finding in our study, we agree with Toner et al.  that this is in fact not substantially greater than chance alone would allow. Although the model is most effective in predicting the risk of PONV in groups of patients, it is less capable of estimating PONV for individual patients. In clinical practice, we believe that most anesthesiologists would be interested in predicting the risk of PONV in their individual patient, based upon patient, anesthesia, and surgery related factors.
Our study does not claim that the studies of Koivuranta et al.  and Apfel et al.  were criticized for “no analysis” and “lack of analysis,” respectively, of anesthesia-related factors. In our view, mathematical models, which do not incorporate a variable for anesthesia related factors, are limited. Since many patient and surgery-related factors can not be changed in the perioperative period, there must be, within the equation of the model, provision to allow for input of anesthesia-related variables. The mathematical models proposed by Koivuranta et al.  and Apfel et al.  do not permit users to determine the impact of a modification in anesthesia technique. Our comprehensive mathematical model includes patient, anesthesia, and surgery related factors associated with PONV.
Nurses in the PACU and ASU collected our data. Postoperative nausea and vomiting was defined in the check-off forms, which were used by the nurses in the PACU and ASU. Although the definition was printed on the check-off forms for clarification regarding patients in the PACU and ASU, patients who were contacted by phone were nonetheless asked about episodes of nausea or vomiting at home. The main difference is that patients in the PACU and ASU who experienced nausea or vomiting were treated, since antiemetics are standardized for those patients.
Apfel et al.  have tried to explain a method of “under-reporting by PACU and ASU nurses,” demonstrating their lack of understanding regarding the structure of our PACU and ASU staffing. Nursing assignment in the PACU and ASU is not according to surgical subspecialty. Rather, all nurses provide postoperative care for any surgical subspecialty patient. Therefore, the potential for an “incomplete assessment,” as proposed by Apfel et al.  , does not exist. The claim that we have suggested that the low incidence of PONV was “most likely explained by” under-reporting by PACU and ASU nurses is incorrect. Although under-reporting because of high work load may be a limitation of the study, the high number of D+C procedures and ophthalmology surgery performed at our hospital, both with low incidences of PONV, may be the most likely explanation for the low incidence of PONV.
Apfel et al.  allude to a discrepancy between two percentages quoted in the text and figure 1. The two percentages, 9.1% and 7.2%, refer to different incidences. The 9.1% incidence of PONV refers to all subjects at 24 h (those who responded to the 24 h interview), while the 7.2% is the incidence of PONV among general anesthetic patients during their postoperative stay.
Apfel et al.  recommended the inclusion of an interaction term between duration and type of anesthesia into the model, suggesting that duration is important for general but not regional anesthesia. During the data analysis, this interaction term was not found to be statistically significant (P  = 0.45, although we had sufficient sample size to detect an association). Therefore, it was not included in the final model. We did not mention it in our manuscript because of space limitations.
Apfel et al.  discussed the bimodal, nonlinear association between age and PONV, stating that there is an increase PONV incidence with age among young children, and decrease of PONV incidence among adults. Because our patient population did not include pediatric patients (out of 17,638 patients, only 14 patients were younger than 14 yr of age, and only an additional 32 patients were 14 yr of age), the association between age and PONV incidence was fairly linear in our data set. Our model is only applicable in the age range of the study patient population (11–98 yr). We caution against extrapolating our model to an age group, which is outside of the age range of our study population (e.g.  , to very young patients, where, in fact, the predicted risk can be exaggerated using our model). However, the scope of our current investigation did not include pediatric cases. The increase in the incidence of PONV among pediatric patients and the decrease in the incidence of PONV with increasing adult age means that the association is not linear if we combine pediatric and adult patients. It does not mean that a bimodal distribution exists between PONV and age, in which there should be two peaks in the distribution. There is one peak (i.e.  , one mode) in late childhood, with a lower incidence of PONV in early childhood and adulthood.
We have developed and validated a mathematical model to calculate the risk of PONV among ambulatory surgical patients. We believe that this model will predict patients’ risk of PONV and promote efforts to reduce the incidence of PONV.