Anesth Analg 2001;92:797
© 2001 International Anesthesia Research Society
LETTERS
Power Analysis with More than Two Groups
Hernán R. Muñoz, MD, and
Jorge A. Dagnino, MD
Department of Anesthesiology Catholic University of Chile Santiago, Chile
We have read with interest some studies published in Anesthesia & Analgesia last year in which power analysis is performed to calculate the sample size with more than two groups (13). We have not been able to find a reliable method to perform this calculation. Some statistical packages do this by introducing a mathematical artifact consisting basically of a reduction of several groups into only two groups: one deemed as control and the second including all the other interventions in the study. The calculated n for this last group is then divided by the number of potential groups, except the control, to give the sample size. Evidently, this is not correct as the final power turns out to be only a fraction of the one originally defined.
In addition, in some studies, the magnitude of change in the variable of interest that is considered significant is not clear (1,2). This last value, however, is absolutely necessary to calculate the sample size with two groups (4).
We think that it is essential that the authors give all the assumptions used to calculate the sample size with more than two groups and the appropriate reference for the applied method.
References
-
Larsen B, Seitz A, Larsen R. Recovery of cognitive function after remifentanil-propofol anesthesia: a comparison with desflurane and sevoflurane anesthesia. Anesth Analg 2000; 90: 16874.[Abstract/Free Full Text]
-
Gesztesi Z, Sa Rego MM, White PF. The comparative effectiveness of fentanyl and its newer analogs during extracorporeal shock wave lithotripsy under monitored anesthesia care. Anesth Analg 2000; 90: 56770.[Free Full Text]
-
Zarate E, Watcha MF, White PF, et al. A comparison of the costs and efficacy of ondansetron versus dolasetron for antiemetic prophylaxis. Anesth Analg 2000; 90: 13528.[Abstract/Free Full Text]
-
Altman DG. Clinical trials. In: Altman DG, ed. Practical statistics for medical research. London: Chapman & Hall, 1991: 44076.
Response
Brita Larsen, MD
Department of Anesthesiology and Intensive Care Medicine University Hospital Homburg, Germany
In Response: It is necessary to determine the number of patients per group by using a power analysis for a statistical examination of a relevant clinical difference.
Initially, a priori power analysis indicated that 17 patients were needed to be enrolled in each group (n1 = n2 = 17) to detect a 20% difference in the digit symbol substitution test (90% for remifentanil and 70% for sevoflurane, assuming the standard deviations taken from preliminary investigations).
Subsequently, we transferred this calculated sample size to the third group (desflurane, n3 = 17).
To take into account possible loss of patients as a result of noncompliance, we decided to increase the target sample size to 20 patients per group.
We consider the sample size of this calculation procedure to be appropriate to detect a relevant clinical difference.
This article has been cited by other articles:

|
 |

|
 |
 
I. Dobrydnjov, K. Axelsson, L. Berggren, J. Samarutel, and B. Holmstrom
Intrathecal and Oral Clonidine as Prophylaxis for Postoperative Alcohol Withdrawal Syndrome: A Randomized Double-Blinded Study
Anesth. Analg.,
March 1, 2004;
98(3):
738 - 744.
[Abstract]
[Full Text]
[PDF]
|
 |
|
|