To the Editor:—
We read with great interest the article by Arbous et al. 1Although we commend the authors for designing an ambitious and creative study to address this important topic, we do have several reservations about the study design, interpretation of the results, and hence the conclusion.
Because anesthetic mortality and serious morbidity are rare events, the use of a case–control design is very appropriate. Case–control studies usually include several controls for every case to increase the power of the study and to increase the likelihood that the conclusions will be valid.2In this study, the investigators used only one control per case (807 cases vs. 883 controls). A higher number of controls would have increased the probability that the controls were a representative sample of the entire cohort.
The main outcome in the study was death or coma within 24 h of surgery, an important and indisputable outcome measure. The limitation of this outcome is that many patients who experience perioperative complications die within days to weeks rather than within 24 h of surgery.3Intraoperative management factors may impact their postoperative course and their eventual demise.
We understand that the investigators chose to limit the matching of controls only to age and sex to prevent bias and also to allow them to examine all factors affecting mortality in a multivariate model. The possible flaw with this approach in the context of their study is that American Society of Anesthesiologists (ASA) physical status classification is such a powerful confounding factor that it could have undermined their multivariate model. Most of the patients who died (> 90%) had an ASA physical status of III–V. Fewer than 30% of the control patients had an ASA physical status of III–V (table 1).
This brings into question whether there were enough ASA physical status III–V patients in the control category to validate a multivariate calculation for other factors relating to mortality. This could have been addressed in two ways: The study could prospectively have enrolled only ASA physical status III–V patients, or many more control patients could have been included.4
In addition to presenting odds ratios for rare events, it is important to present the number needed to harm. For example, the investigators presented an odds ratio of 10:1 for preventing death by adopting a universal practice of reversing muscle relaxants. We made some assumptions and calculated that the number needed to harm for the entire cohort might be 1 in 25,000. This means that 25,000 people on average would have to have muscle relaxation not reversed to result in one additional death. The odds ratio for preventing death by having two anesthesia providers present for all inductions and emergences was 10:6. This may translate to an even larger number needed to harm than for not reversing muscle relaxants. The cost of providing two anesthesia providers for every anesthetic—no matter how minor the surgery and no matter how healthy the patient—would be staggering and may not result in many lives saved. A similar study focusing only on patients with an ASA physical status of III–V might be more useful in identifying which anesthetic management factors are most important in decreasing the likelihood of death among the sickest patients presenting for high-risk surgery.
Despite our reservations, we appreciate the study of Arbous et al. and believe that it raises important issues. In particular, we believe that separating anesthetic from surgical death is a false distinction. This study highlights the contention that multiple aspects of perioperative care and management may impact on postoperative outcome. This is a seminal study that is likely to be extensively quoted. It is important to highlight some of its limitations and to avoid overinterpretation of the findings.
*Washington University Medical School, St. Louis, Missouri. bilenrog@msnotes.wustl.edu
