| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
| ||||||||||||||
| ||||||||||||||
|
|
|
|
||||||||||||
|
||||||||||||||
Department of Anesthesia, Critical Care and Pain Medicine, Royal Infirmary of Edinburgh, Edinburgh, Scotland, UK
To the Editor:
The article of Yoshitani et al. (1) regarding cerebral hypoperfusion and cognitive outcome after cardiac surgery has some limitations (1). The arterial carbon dioxide tension (PaCO2) values reveal a large degree of variance. As PaCO2 is a fundamental determinant of SjvO2, it should have been entered as varying covariate into the analysis of variance model. As a result, most of the between-group and within-group differences would likely have been eliminated (2,3).
A statistical tenet is that variables should be orthogonal, i.e., independent of each other. All the predictor variables in Table 4 are nonorthogonal, being related by being either mathematically coupled, highly intercorrelated, or repeated measures (3–5). Being robust and independent estimates, either jugular venous oxygen saturation or tension (SjvO2, PjvO2) should have been used, and given the orthogonal nature of repeated measures, a summary should have been generated for all the time points (5,6).
Although some advocate the use of categorical definitions of cognitive decline (7), they are arbitrary (5). As SjvO2, PjvO2, and cognitive decrement are continua, linear (not logistic) regression should have been used (5). In taking this approach, we have been unable to relate any level of SjvO2, high or low, to either cognitive or neurological outcome 3 mo after coronary artery bypass grafting surgery (5,6).
References
Department of Anesthesiology, Nara Medical University, Nara, Japan Department of Hygiene, Nara Medical University, Nara, Japan
In Response:
Our study (1) aimed to clarify the role of SjO2 for cognitive decline after hypothermic cardiopulmonary bypass. We used the logistic regression model for this purpose. Dr. Alston, however, states that the linear (not logistic) regression model should be used because SjO2 and other parameters involved are continuous variables. We do not agree with this opinion. Logistic regression model can accept continuous variables as explanatory variables in nature, and examples using continuous variables are shown in the standard textbook of logistic regression analysis (2). Our outcome is dichotomous: either "decline" or "not decline" of cognitive function. Such outcomes agree with the logistic model better than with the linear model. A more important issue is whether the model suits the data when adopting a statistical model. In our case, Hosmer-Lemeshow goodness-of-fit test revealed that the logistic regression model fit to the observed data shown in Tables 4 and 5. We therefore conclude that our analysis did not have any statistical problems.
In our article (1), Table 4 showed significant predictors of postoperative cognitive decline. Although large correlations among the predictors were recognized, we presented the basic information for the multivariate logistic regression analysis. Although variables should be independent of each other, highly intercorrelated variables are generally eliminated in multivariate regression analysis. As a result, multivariate logistic regression analysis in our study indicated SjO2 and PjvCO2, which are reciprocally independent, as significant predictors.
We recognized that PaCO2 is a fundamental determinant of SjO2. PaCO2 was entered as varying covariate into the univariate logistic regression analysis in our study. PaCO2 was not a significant predictor, although we did not show the result in Table 4.
To compare the cognitive decline of two divided groups, such as cardiopulmonary bypass (CPB) group and non-CPB group, noncategorical definition may be used. However, most previous studies (3–10) advocated the use of categorical definitions of cognitive decline. It is controversial whether linear regression is better than logistic regression analysis as a statistical method for investigating the associated factors of cognitive decline. We should complete an even more detailed examination on cognitive decline.
References
This article has been cited by other articles:
|
|
 |
|
  R. P. Alston and Y. Kadoi Rewarming Rate, Diabetes, Jugular Bulb Saturation, and Cognitive Outcome from CABG Surgery * Response Anesth. Analg., March 1, 2003; 96(3): 914 - 915. [Full Text] [PDF]
|
|
|
|
|
|
Y. Kadoi, S. Saito, F. Goto, and N. Fujita Slow Rewarming Has No Effects on the Decrease in Jugular Venous Oxygen Hemoglobin Saturation and Long-Term Cognitive Outcome in Diabetic Patients Anesth. Analg., June 1, 2002; 94(6): 1395 - 1401. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
