Free
Correspondence  |   May 2011
A High Significance Level after Analysis of Covariance in a Small-group Study?
Author Notes
  • Chung Shan Medical University and Chung Shan Medical University Hospital, Taiwan, Republic of China.
Article Information
Correspondence
Correspondence   |   May 2011
A High Significance Level after Analysis of Covariance in a Small-group Study?
Anesthesiology 5 2011, Vol.114, 1243-1244. doi:10.1097/ALN.0b013e318212b928
Anesthesiology 5 2011, Vol.114, 1243-1244. doi:10.1097/ALN.0b013e318212b928
To the Editor:
In the study by Bellani et al.  ,1 changes of oxygen consumption (V̇o2) in patients, who succeeded or failed in weaning from mechanical ventilation support, were addressed. The authors rejected their original hypothesis and constructed a new theory after they analyzed their results. However, some statistical issues should first be clarified by them to better support their discussion and conclusion.
There were two main findings in their study. (1) There were no significant differences in the maximum V̇o2readings between the success and the failure groups during the weaning pressure support trials. In addition, the minimum V̇o2readings (when adequate pressure support was provided) in the failure group were significantly higher than in the success group (P  < 0.05). (2) The authors further analyzed the group and pressure support effects on patients' successive V̇o2data. By analysis of covariance (cited as a two-way ANOVA by the authors), significant differences were found both in the group and pressure support effects at P  < 0.001. Accordingly, the authors concluded that the patients able to successfully complete their weaning trials were those who reacted to the decrease of ventilatory assistance with a greater increase in V̇o2.
A paradox exists between these two results. The statistical values increased significantly after the analysis of covariance. With an increasing P  value, their analysis of covariance model probably omitted the patients' effects.2 In other words, they probably treated a patient's successive V̇o2data (these data were related) as independent V̇o2data from different patients. Thus, their statistical values reached levels of less than 0.001 in such a small-group study (16 patients in the success group and 12 in the failure group). This criticism seems reasonable, especially after considering the diverse V̇o2trend patterns in response to the withdrawal of pressure support (as shown in their second figure). The diverse patterns would add complexity to the determination of the pressure support effect and should decrease, rather than increase, the statistical significance.
Another statistical issue is that the authors used a correlation coefficient to access the reproducibility of V̇o2measurements. The correlation coefficient is misleading. The Bland–Altman analysis is much more appropriate for assessing reproducibility.3 
In conclusion, the authors produced an impressive and interesting study. However, after rechecking their statistics, it is clear that too many conclusions were drawn from the limited results.
Chung Shan Medical University and Chung Shan Medical University Hospital, Taiwan, Republic of China.
References
Bellani G, Foti G, Spagnolli E, Milan M, Zanella A, Greco M, Patroniti N, Pesenti A: Increase of oxygen consumption during a progressive decrease of ventilatory support is lower in patients failing the trial in comparison with those who succeed. Anesthesiology 2010; 113:378–85Bellani, G Foti, G Spagnolli, E Milan, M Zanella, A Greco, M Patroniti, N Pesenti, A
Rutherford A: Introducing ANOVA and ANCOVA: A GLM approach. Edited by Wright DB. London, SAGE Publications, 2001, pp 43–77Rutherford, A Wright DB London SAGE Publications
Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1:307–10Bland, JM Altman, DG