| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Ann Thorac Surg 2007;83:2016
© 2007 The Society of Thoracic Surgeons
21672 Montbury Dr, Lake Forest, CA 92630
(Email: wnilesanderson{at}aol.com).
The clinical issue studied in the paper by Parolari and colleagues [1] presents interesting statistical challenges. The gold standard for comparing two clinical methodologies (linear and geometric reconstruction techniques in this instance) is an adequately powered randomized trial. Quite often there are not randomized trials of sufficient size, and the techniques of meta-analysis allow for combination of the results of smaller trials; the meta-analysis also allows for inclusion of results from multiple centers. Unfortunately, no combination of trial results can improve on the quality of the included trials; for this reason selection of trials to use in a meta-analysis is generally limited to randomized trials.
The literature on reconstruction techniques contains results on over 2500 patients in 18 studies. Since none of these studies were randomized, it would be easy to dismiss the possibility of a meta-analysis on such a basis alone. But the important clinical question of comparing the reconstruction techniques would remain, and surely these data are sufficient to throw some light on the clinical issue.
What can be done is to combine some or all of the 18 observational studies using standard meta-analysis techniques, and recommendations for use of such meta-analyses in epidemiology are given by Stroup and colleagues [2].
The major problem with analysis of an observational study is assignment bias. Any observational study should take steps to account for the bias, and one important step is to isolate and compare the sources of bias. Temporal trends were a clear potential source of bias in the reconstruction studies, and the meta-analysis identified studies where the temporal effect could be analyzed. The standard method for overcoming assignment bias would be matching based on propensity scoring [3]. Since that was not done in the underlying studies there was no opportunity to do so in the meta-analysis.
Another important problem is publication bias, since studies that produce statistical significance seem to have greater acceptability. The standard method for analyzing publication bias is the funnel plot, and that was done in the Parolari paper.
When the meta-analysis has been performed, with as much accounting as possible for assignment and publication bias, one still has an observational study. The resulting study is larger than the studies that were combined; to the extent that the studies are consistent the meta-analysis will have a smaller error than the individual included studies. As discussed by Egger et al, there remains the possibility of over interpreting the results [4].
In spite of the problems of observational studies, the meta-analysis accomplishes two important goals. First it allows a systematic use of the many studies on reconstruction techniques. Second, the meta-analysis furnishes valuable information for use in designing a randomized trial; such a trial may or may not be feasible, but the meta-analysis will help the clinical community make an informed decision.
 |
References |
|---|
 Top References |
|---|
Related Article
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 |
