| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Ann Thorac Surg 2005;79:957-958
© 2005 The Society of Thoracic Surgeons
Department of Cardiothoracic Surgery, Washington University School of Medicine, 3108 Queeny Tower, One Barnes-Jewish Hospital Plaza, St. Louis, MO 63110–1013
(E-mail: meyersb{at}msnotes.wustl.edu).
Meyer and colleagues present an analysis of the United Network of Organ Sharing database regarding outcomes after lung transplantation for pulmonary fibrosis (PF) with special attention given to the association between transplant type and survival. The authors point out that a simple comparison of the procedures would be biased because the two operations were not applied randomly but were at the discretion of the treating surgeons, in part due to differences in recipient characteristics. Notably, the bilateral sequential lung transplantation (BSLT) patients are younger but have higher pulmonary artery (PA) pressures and lower forced expiratory volume in 1 second (FEV1) scores. The authors concluded that patients < 60 years of age transplanted for PF appear to have better survival with single-lung transplantation (SLT) than with BSLT.
In the results, in the multivariable logistic regression model for 30-day survival in Table 4, the authors state that no advantage to BSLT in this patient population could be found. The p value associated with this odds ratio is 0.08 and the 95% confidence interval of the odds ratio includes 1. Therefore, using the usual rules of inference, we cannot exclude that there is no association between survival and type of transplant. This is acknowledged in the text by calling it a "trend toward worse survival," which is far less certain than the conclusion paragraphs of the abstract or the paper would suggest.
A major problem with this article is seen with the use of the propensity analysis. The authors note that there was a paucity of SLT with high propensity scores, so the matching technique could not be used. In choosing another strategy for analysis, they have not fixed the problem in the data; the patients who have the highest propensity scores generally received BSLT, and a high propensity score is associated with a higher risk of mortality regardless of procedure performed. If the matching technique does not work, it is the result of a problem with the data rather than a problem specific to the matching technique.
Once the stratified analysis has been used (Table 6), the superiority of SLT over BSLT still remains less certain than the authors describe. In the first two quartiles, or with half of the patients studied, there were no BSLT deaths, whereas in the SLT group there was a 6% and 10% death rate in the first and second quartile. The lack of statistical power in this half of the subjects leads to the claim of "no difference." In the third quartile in which there appears to be sufficient cases in each group for comparison, there is no difference in the mortality rate. In the fourth quartile, the mortality rate is higher for BSLT, but once again, the highest propensity scores and the highest risk patients are BSLT, causing confounding within this quartile.
Even when the propensity score was added to the regression model, the p value of the BSLT factor was still 0.06 and the 95% confidence limits for the odds ratio included "no difference." Again, using the typical rules for inference testing, we must accept the null hypothesis of no association. In the text, the authors report another "trend" that seems less dramatic than either the abstract or the conclusions of the article would suggest.
Other problems here include missing data and the lack of evaluation for interaction effects. Tables 1 and 2 show the large number of patients with missing data for many predictor variables. Only 568 of 821 patients (69%) had mean PA pressure scores, which is one of the main predictor variables. The authors used simple imputation of the mean scores into missing cells in this and other variables, but this imputation would diminish the estimated risk differences between the SLT and BSLT groups and undermine the effectiveness of the multivariable models and the propensity score adjustments.
There has been no evaluation for interaction terms. For instance, does the effect of BSLT versus SLT change at different levels of other important variables? Is there a difference in the odds ratio for death after BSLT versus SLT depending on the level of pulmonary hypertension in the recipient? This interaction effect seems to be suspected by the surgeons who did these operations; in the data described, SLT in the setting of extreme pulmonary hypertension is rarely performed. Another potential interaction effect is that of transplant type and transplant center volume. That is, was there a difference in the effect of BSLT versus SLT when one looks at different strata of hospital experience?
This is a carefully analyzed and clearly written article, but it raises doubts with missing data and confounding that cannot be overcome with regression techniques or propensity scores. These doubts challenge the conclusion that SLT is preferred over BSLT for PF.
Related Article
Ann. Thorac. Surg. 2005 79: 950-957.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ANN THORAC SURG | ASIAN CARDIOVASC THORAC ANN | EUR J CARDIOTHORAC SURG |
| J THORAC CARDIOVASC SURG | ICVTS | ALL CTSNet JOURNALS |
