To the Editor:—
I read with great interest the recent article by Nisanevich et al. 1suggesting that restricted fluid therapy for intraabdominal surgery reduces postoperative morbidity. My comments focus on the description of the statistical methods and their application to the data.
The authors mention the use of both the chi-square test and Fisher exact test for the analysis of categorical data. However, it seems that only the results for the chi-square test are reported: For the number of patients with complications, the Fisher exact test gives a nonsignificant P value of 0.056. The authors should explain why they report the results of one test and not the other.
No follow-up is given for the four patients who were withdrawn after randomization because their surgeries were not considered extensive. Assuming no complications with these patients, the P value would not be statistically significant by either the chi-square (P = 0.057) or the Fisher exact test (P = 0.086). The postrandomization exclusion of patients without any analysis is a serious error because the reader can never be sure why these patients were excluded.
The authors state that exact confidence intervals were calculated for the overall rate of complications, but I am unable to find these in the article. An exact 95% confidence interval for the odds ratio of an increase in complications with liberal fluid therapy is 0.95–5.14. This confidence interval includes 1 and so would not be taken to indicate a statistical difference between the two therapies.
The authors mention the use of the Newman-Keuls adjustment, but that correction only applies if the group means are independent, which is clearly not the case here.
No advanced statistical methods are used to model the data and explain the impact of relevant covariates. In particular, logistic regression could be used to model the presence of a complication on the number of fluid boluses, the degree of hypotension, the duration of surgery, or American Society of Anesthesiologists physical status. Based on these outstanding statistical issues, I agree with the authors that additional studies are needed.
Dartmouth Medical School, Lebanon, New Hampshire. email@example.com