To the Editor:
We read with great interest the recent systematic review and meta-analysis by Hovaguimian et al.1 on the advisability of intraoperative administration of a high inspired oxygen fraction (Fio2). The authors analyzed randomized trials comparing high supplemental oxygen (defined as Fio2 ≥50%, preferably 80%) with a lower fraction (Fio2 <50%) for reducing surgical site infection (SSI) and postoperative nausea and vomiting.
Previous meta-analyses had reached inconsistent conclusions regarding the effect on SSI.2–7 This new meta-analysis1 found evidence (in nine trials including 5,103 patients) that SSI risk decreased by 23% when a high Fio2 was applied during all types of surgery (risk ratio [RR], 0.77; 95% CI, 0.59 to 1.00). However, moderately large heterogeneity (I2, 54%; P = 0.03) was also detected.1 Moreover, in a subgroup analysis of colorectal surgery (1,977 patients in eight trials), the effect of supplemental oxygen was statistically borderline (RR, 0.78; 95% CI, 0.60 to 1.02), but as heterogeneity was not significant (I2, 39%; P = 0.11), the authors did demonstrate a statistically significant overall effect (RR, 0.79; 95% CI, 0.65 to 0.97) under a fixed-effects model.
As a result, we feel that these data should not be taken to suggest that additional trials or meta-analyses are unnecessary, that is, we believe that we should not take the reduction in SSI to have been demonstrated. Nothing could be farther from the truth. We believe deeper analysis, with greater power to account for the detected heterogeneity may still call into question the translation of these findings to clinical practice. It is noteworthy that all the included trials enrolled relatively small numbers of patients with colorectal surgery, and most (six of eight) lacked sufficient power to determine potential differences. To illustrate this concern, we note that in a hypothetical meta-analysis that assumes a baseline SSI rate of 17.5% and an effect size of 22% (RR, 0.78), more than 3,200 patients would be required for an alpha of 0.05 and 80% power.8 The implication is that to detect such an effect, 1,200 more cases would be needed in a future meta-analysis.
The extent of heterogeneity in analyzed studies partly determines how difficult it is to draw overall conclusions. Analysts do not generally explore potential causes of heterogeneity by incorporating study-level covariates in random-effects regressions, sometimes because of limited data provided in the selected trials. Conflicting conclusions on SSI in the analyzed studies2–7 may derive from variations in patient characteristics (e.g., baseline differences in risk factors for SSI) or varying surgical procedures or outcome assessments across trials. The implication of such differences in past trials is that future analyses should look at the distribution of potential effect modifiers when the body of evidence permits.
Results can also be usefully reported with a prediction interval (PI), a novel statistical proposal which assesses the degree of heterogeneity and its likely effect by predicting a range for the treatment effect in future studies.9 To differentiate the information provided by the more familiar CI from the information given by the PI, consider that the CI confirms that a sample mean has been reasonably well estimated whereas the PI tells us where 95% of the future data points should lie. Thus, the PI gives us a more complete picture of the degree of uncertainty around our current inferences. When we calculated the PI for the data provided by Hovaguimian et al.,1 the upper and lower limits included the null value of 1, and so we can expect that in some settings applying a high Fio2 may actually be ineffective for reducing SSI (RR, 0.77; 95% PI, 0.38 to 1.57). Given such a large PI, further research is needed until the causes of heterogeneity can be controlled for. Meanwhile, we can only tentatively conclude, based on these results and a solid pathophysiological foundation,10,11 that applying a high Fio2 is very likely to reduce SSI and may provide protection in many surgeries.
In summary, it seems to us that further research is still needed if we are to clarify the specific effects of perioperative high Fio2 so that we can confidently generalize results to a range of populations and clinical situations. However, it is already clear that new trials have to reduce heterogeneity by including standard measures of SSI12 and postoperative nausea and vomiting13 and by recruiting primarily patients at high risk for these events. Meanwhile, this meta-analysis brings us closer to closing the debate and helping us realize what is still needed.
The authors declare no competing interests.