Whether nitrous oxide induces nausea and vomiting is actually well known--when it is administered as the sole anesthetic drug, volunteers become nauseated and vomit . The same is true for ether, cyclopropane, and other anesthetics. Therefore, the question of interest is whether addition of N2 O increases the incidence of postoperative vomiting compared with anesthetic regimens differing only by its omission.
This month, Hartung  reexamines this issue, not by performing an additional clinical trial, but by tabulating results from published clinical trials. Such reanalysis has become popular in recent years, and stems from the recognition that certain end points (e.g., the effect of different anesthetic techniques on the incidence of death after cardiac surgery) are difficult to examine without large clinical trials that are logistically difficult for individual investigators to perform. If a trend appears repeatedly in small clinical trials--even if no individual trial achieves statistical significance--it is tempting to pool the results of these trials to obtain a larger sample size and thereby attain additional statistical power. The typical approach to this problem is termed meta-analysis . Examples are uncommon in the anesthesia literature, but include Sorenson and Pace's examination of outcome after regional versus general anesthesia in high-risk patients , Pace's determination of whether succinylcholine induces myalgias , and Dexter and Tinker's evaluation of recovery times after three different anesthetic regimens .
The potential advantages of meta-analysis can be fragile. First, the meta-analysts rarely have access to the original data and must depend on published summaries; if studies are not standardized in their methods and reporting techniques, it may be difficult to pool data from different sources. For example, investigators may fail to distinguish adequately between nausea, retching, and vomiting. Second, meta-analysts typically exclude unpublished trials. As a member of the Editorial Board of Anesthesiology, I can verify that studies in "hot" areas are sometimes not published, not because of scientific limitations, but rather because many similar trials already have been published. In addition, although trials with negative outcomes (i.e., the inability to demonstrate a difference between therapies) often are rejected because of inadequate sample size, similarly sized trials with positive outcomes might be published . Thus, the database of published trials may not include all studies and may not even represent a cross section of the available data. Finally, meta-analysts must either accept all studies as having equal scientific quality or they must judge the relative quality of each, resulting unavoidably in studies of varying quality being included in the analysis.
Cote et al.  attempted to overcome these limitations using an approach they called a "combined analysis" to examine the age at which premature neonates no longer have appreciable risk for developing postoperative apnea. Investigators from several studies collaborated to reexamine the original data from each of their studies, bypassing problems associated with using only published information; however, this approach still does not avoid the problem of ignoring the results of unpublished trials. They noted that inconsistencies in study design (predominantly the intensity of monitoring) markedly influenced the results--centers that monitored more intensely were more likely to detect apnea, suggesting a serious limitation of meta-analyses that pool results of nonstandardized studies.
In the present issue, Hartung uses yet another approach to examine the role of N2 O in inducing postoperative vomiting. Rather than reexamine the results of each study, Hartung determined whether each of 27 clinical trials concluded that N2 O increased, decreased, or had no influence on the incidence of vomiting. His statistical approach was similar to that of flipping coins: If a coin is flipped 27 times, it is likely to land in its "heads" position roughly 13 or 14 times.1 However, it would not be surprising if "heads" occurred 12 or 15 times and not astounding if "heads" occurred 11 or 16 times. But if "heads" occurred 24 of 27 times and tails only three, we would be most surprised, i.e., we would not be likely to accept this as a chance occurrence. If gambling, we might expect that the coin was fixed. In science, we would reject the null hypothesis (that N2 O did not influence the occurrence of vomiting) and conclude that the incidence of vomiting is greater when N2 O is used than when it is not.
1 With 27 coin flips, percentages associated with various numbers of "heads" are as follows: 3 (0.0022%), 11 or 16 (9.7% each), 12 or 15 (13.0% each), 13 or 14 (14.9%).
Hartung reports a P value <0.00005 that N2 O increases the incidence of vomiting. In light of this remarkable degree of statistical significance, should clinicians immediately ban N2 O from their practice? Probably not. First, the effect of N2 O on postoperative vomiting must be considered in the context of the other drugs in the anesthetic regimen. If cyclopropane consistently induced postoperative vomiting, then addition of N2 O to a cyclopropane anesthetic probably would not worsen the incidence of vomiting. Rather, reducing the dose of cyclopropane might decrease (or leave unchanged) the incidence of vomiting. At the other extreme, consider the contribution of N2 O to postoperative vomiting during propofol anesthesia (during which the incidence of vomiting is low). Although N2 O might permit a decrease in propofol requirements, this, per se, is not likely to reduce the incidence of propofol-induced vomiting and might increase its incidence. Unfortunately, no clinical trial has examined whether N2 O influences the incidence of vomiting after propofol anesthesia. Although several studies report the effect of N2 O on vomiting with other modern anesthetics [e.g., desflurane ], these trials are too small to determine statistical significance. Second, statistical significance is not identical to clinical importance. For example, removing N2 O from one's anesthetic practice might decrease the incidence of vomiting by only 1%-2%, a change too small for an individual anesthesiologist to detect. Finally, N2 O may have other redeeming qualities (e.g., its use may hasten wake up by permitting a reduction in the dose of other anesthetics) that overcome its potential disadvantages.
Hartung's analysis may settle the issue of whether N2 O increases the incidence of vomiting associated with older anesthetic regimens, but further analyses may be needed to define the role of N2 O in inducing vomiting with new anesthetic regimens.
1. Hornbein TF, Eger EI II, Winter PM, et al. The minimum alveolar concentration of nitrous oxide in man. Anesth Analg 1982;61:553-6.
2. Hartung J. Twenty-four of twenty-seven studies show a greater incidence of emesis associated with nitrous oxide than with alternative anesthetics. Anesth Analg 1996;83:114-6.
3. Chalmers TC. Problems induced by meta-analyses. Stat Med 1991;10:971-80.
4. Sorenson RM, Pace NL. Anesthetic techniques during surgical repair of femoral neck fractures. A meta-analysis. Anesthesiology 1992;77:1095-104.
5. Pace NL. Prevention of succinylcholine myalgias: a meta-analysis. Anesth Analg 1990;70:477-83.
6. Dexter F, Tinker JH. Comparisons between desflurane and isoflurane or propofol on time to following commands and time to discharge. A metaanalysis. Anesthesiology 1995;83:77-82.
7. Begg CB, Berlin JA. Publication bias and dissemination of clinical research. J Natl Cancer Inst 1989;81:107-15.
8. Cote CJ, Zaslavsky A, Downes JJ, et al. Postoperative apnea in former preterm infants after inguinal herniorrhaphy. A combined analysis. Anesthesiology 1995;82:809-22.
9. Wrigley SR, Fairfield JE, Jones RM, Black AE. Induction and recovery characteristics of desflurane in day case patients: a comparison with propofol. Anaesthesia 1991;46:615-22.