and was clearer to the reader. However, in retrospect, the original format was probably easier to understand and may have avoided potential analysis error.

This strategy used 2 separate 2 x 2 tables with the time cohorts of less than 15 minutes, and greater than 15 minutes, grouped separately. Here, the same proportional analysis found that 8 of 135 or 5.9% of the patients survived in the control group compared with 19 of 158 or 12% in the group that received bicarbonate. Clearly, although the overall proportional increase is roughly 2-fold, the 2 x 2 analysis revealed a P value of .07 v² with Pearson correction and P = .05 Fisher exact test. The relationship was described in the text as bapproaching significance.Q

It is then suggested that bcontrary to title this paper suggests that bicarbonate is not beneficial on prehospital arrest regardless of down time.Q There are comparisons in this study ranging from the best-case comparison that offers a P value of 0.03 versus the administration of bicarbonate administration and in the worst-case scenario, a P value of .07 which bapproachesQ statistical significance. The conclusions of the original article suggest that there were, overall, no benefits of bicarbonate administration across all times, but the cornerstone of the study was that bicarbonate would likely only work, if at all, in the prolonged Cardiac arrest scenario. Here, we conclude that in the worst-case scenario, there was an approximate 2-fold increase in survivorship that approached clinical significance. In this day and age, with research more difficult to perform and to complete properly, it is sometimes necessary to use subgroup analysis and evaluate proportions of raw numbers as opposed to strict

delineations of statistical significance.

In fact, if we were to use a precise statistical analysis, it would be with v² using the Mantel-Haenszel test with odds ratio of that analysis noting the 2 groups to be almost precisely equal ( P = .94). This control between the 2 separate 2 x 2 tables would eliminate the time factor from consideration. When in fact, the proper Clinical approach would incorporate specifically the time factor effect, not eliminate it from consideration.

Therefore, we feel that this statistical analysis of 2 x 2 proportional tables of survivorship allows a reasonable comparison of the clinical issue at hand. Although we have not conclusively proven the use of bicarbonate in this circumstance, it should be left to the reader to draw their own conclusions concerning the potential clinical utility in the prolonged cardiac arrest setting.

The entire discussion distills down to this question. From an overall study of 792 patients, there were

8 patients (5.9%) who survived in a prolonged arrest scenario without bicarbonate, and there were 19 patients (12%) who survived with the administration of bicarbonate. This difference can obviously be because of chance or near clinical significance. Let the reader decide.

Once again, the reviewer and editors comments are appreciated commenting on this research study.

Rade B. Vukmir MD, JD Critical Care Medicine Associates Sewickley, PA 15143, USA

E-mail address: [email protected] doi:10.1016/j.ajem.2006.05.005

References

1. Vukmir R, Katz L. Sodium bicarbonate study group: sodium bicarbonate improves outcome in prolonged prehospital cardiac arrest. Am J Emerg Med 2006;24:156 – 61.
2. Vukmir R, Bircher N, Radovsky A, et al. Sodium bicarbonate may improve outcome in dogs with brief or prolonged cardiac arrest. Critical Care Medicine 1995;23:515 – 22.

Sodium bicarbonate improves outcome in prolonged pre-hospital cardiac arrest

The Authors’ Reply,

I would like to thank Drs Vukmir and Katz for their article, bSodium bicarbonate improves outcome in pro-

longed prehospital cardiac arrest [1].Q It is impressive that they were able to conduct such a well-organized clinical trial in the chaotic setting of prehospital cardiopulmonary arrest. Their definitive finding of no difference in outcome among all comers who received bicarbonate or placebo is a significant contribution to our literature.

However, based on a subset analysis, they arrived at the conclusion that bicarbonate improves outcome in victims of prolonged arrest. This conclusion seems to be erroneous. I attempted to use the data they reported to generate the proportions and P values they reported, and I believe I have identified some important errors.

To begin with, there is the general problem that subset analysis is always risky, and often leads to spurious results, due to 2 problems. First, the problem of multiple hypothesis testing, which can be simplified as, bThe more questions you ask of one dataset, the more likely you are to find dsignificant findingsT by chance alone.Q Second, different confounders can be relevant for different outcomes, and a design that is well controlled for a particular main outcome measure may not be well controlled for a secondary outcome measure. In general, conclusions drawn from subset analysis should be presented with caveats and should be subjected to Prospective analysis in a new trial before a change in practice is considered.

Above I have mentioned a common general problem. However, in the present article, there also appear to be overt errors in the calculations and in the printing of the article. Most importantly, there seem to have been errors in the calculation of the magnitude of effect and P values, invalidating the subset analysis that led to the strong statement in the article’s title.

Problem 1. The wrong proportions were compared. The authors compared the probability of prolonged arrest among patients who got bicarbonate and survived, vs the probabil- ity of prolonged arrest among patients who got placebo and survived (19/58 = 32.8% vs 8/52 = 15.4%). They report a P value of .007 and declare a btrendQ toward survival among patients receiving bicarbonate after prolonged arrest.

To measure what they sought to measure, I think the correct comparison should have been the Probability of survival after bicarbonate therapy, vs the probability of survival after placebo therapy, among patients with pro- longed arrest. Using this latter comparison, 19 of 158 vs 8 of 135, the v² analysis yields a P value of .07, indicating no significant difference.

Problem 2. The wrong P value was calculated. Even if 19 of 58 vs 8 of 52 had been the correct comparison, the P value for this comparison should have been reported ( P =

.15 by my calculations). Instead a P value of .007 was reported. As shown in my calculations below, this seems to have been calculated from all the numbers in the article’s Table 5. This P value of .007 shows that the data in Table 5 are not distributed randomly in general, but says nothing about the particular substratum of interest.

Problem 3. There appear to be a number of typographic and other errors, such as the following:

• b5852Q survivors in the bicarbonate group on the left in Figure 1;
• incorrectly calculated percentages in the following quote from the abstract: b(7.4% [58/420]), compared with those who received placebo (6.7% [52/372])Q;
• incorrect titles in Table 5, the main results table: b% with ED survival and bicarbonateQ in the bnonsurvivalQ part of the table (emphasis mine).

The article describes a successfully conducted trial that provides good evidence that bicarbonate is not effective for the treatment of cardiopulmonary arrest in the prehospital setting. No other conclusion can be drawn from the trial. The finding of nonsignificant differences in survival in certain subgroups may justify further study, but does not justify a change in practice.

Daniel J. Pallin MD, MPH

Departments of Emergency Medicine Brigham and Women’s Hospital and Children’s Hospital

Boston, MA, USA Division of Emergency Medicine Harvard Medical School Boston, MA, USA

DOI of original article:10.1016/j.ajem.2005.08.016 doi:10.1016/j.ajem.2006.06.002

Reference

[1] Rade BV, Laurence K. Sodium Bicarbonate Study Group. Sodium bicarbonate improves outcome in prolonged prehospital cardiac arrest. Am J Emerg Med 2006;24(2):156 – 61.