|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 22, No 13 (July 1), 2004: pp. 2754 © 2004 American Society of Clinical Oncology. DOI: 10.1200/JCO.2004.99.033
Association Between TGFBR1*6A and Cancer: Is There Any Evidence?Hamilton Regional Cancer Center and The Department of Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario, Canada To the Editor: In the September 1, 2003, issue of the Journal of Clinical Oncology, Kaklamani et al1 performed a meta-analysis to evaluate whether TGFBR1*6A, a polymorphic allele of the type 1 transforming growth factor beta receptor (TGFBR), is over-represented in patients with cancer. The results showed that there is a higher frequency of TGFBR1*6A in cancer patients compared with controls, and subgroup analyses suggested that this positive association holds true in breast, ovarian, and colorectal cancers but not in bladder cancer. I thank the authors for their effort in improving the current evidence of TGFBR1*6A and cancer risk through a meta-analysis, but the methodology of this study could be improved if they were to conduct a formal test of homogeneity before the pooling of data.2 The authors used the fixed-effect model by performing the Mantel-Haenszel and the inverse variance methods, which required a prior nonsignificant test of homogeneity. This test is especially important here because the patient characteristics, the ways controls and cases were sampled, and the types of cancer included in the chosen articles are quite different among each other. Using the data the authors presented in Figure 1, the Breslow-Day test for homogeneity is significant (Q = 14.36; df = 6; P < .026); this means the studies are too different to be pooled. The authors may need to explain the heterogeneity and analyze the studies separately according to the source of heterogeneity, or they may consider using the random-effect model, which incorporates the heterogeneity into calculating the pooled estimate. If the random model is used, the pooled odds ratio is 1.13 (95% CI, 0.85 to 1.50; P = .38), which is nonsignificant. In light of this finding, the authors may need to reconsider their conclusion regarding TGFBR1*6A and cancer risk. Moreover, a sensitivity analysis may be necessary to evaluate the methodologic qualities of each study.2 The statement in the discussion "the quality of individual studies does not seem to bias the results of our meta-analysis" needs justification. The authors did not detect an association between *6A heterozygotes and colorectal cancer; they attributed the cause to an opposite effect of 6A heterozygosity on the European population, possibly from environmental factors. They reanalyzed the data on a post hoc basis, using 6A carriers from only the United States and found an increased association with colorectal cancer. The credibility of this type of subgroup analysis would be improved if the hypothesis were formulated a priori, given that sources of heterogeneity can often be postulated before analysis.2 Meta-analysis of observation studies often poses particular challenges because of the inherent biases in study designs; therefore, one must be rigorous in applying its methodology. Author's Disclosures of Potential Conflicts of Interest The author indicated no potential conflicts of interest. REFERENCES
1. Kaklamani VG, Hou N, Bian Y, et al: TGFBR1*6A and cancer risk: A meta-analysis of seven case-control studies. J Clin Oncol 21:3236-3243, 2003
2. Stroup DF, Berlin JA, Morton SC, et al: Meta-analysis of observational studies in epidemiology. JAMA 283:2008-2012, 2000 Related Reply
Related Article
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2004 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
