We consider the COX-2 gene highly interesting in the search for susceptibility genes for PCa. Previous study results suggested that single nucleotide polymorphisms (SNPs) are the most common sources of human genetic variation, and they may contribute to an individual's susceptibility to cancer [27
]. In the recent years, interest in the genetic susceptibility to cancers has led to a growing attention to the study of polymorphisms of genes involved in tumourigenesis. Since the identification of COX-2 rs2745557 polymorphism, growing number of studies suggested that COX-2 rs2745557 polymorphism plays an important role in the development of PCa. Epidemiological studies of the rs2745557 polymorphism in COX-2, if large and unbiased, can provide insight into the in vivo relationship between the gene and PCa risk. However, these studies have appeared in the literature either supporting or negating the significant association. Some reviewed studies are limited by their sample size and subsequently suffer from too low power to detect effects that may exist. But the pool ORs generated from much larger population can increase the statistical power. Combining data from many studies has the advantage of reducing random error [28
In order to provide the comprehensive and reliable conclusion, we performed the present meta-analysis of 8 independent case-control studies [18
], including 11356 patients and 11641 controls. According to the study design, 3 studies were conducted in a population-based design [20
], and 5 in a hospital-based design [18
]. Some studies reported insufficient information about recruitment methodology and study participant characteristics, particularly for controls. The control populations were not uniform. Healthy populations as well as non-cancer patients were included. Some individuals in the control group are likely to develop cancer in subsequent years though they had no clinical symptoms at the time of investigation. Our results indicated that the rs2745557 was associated with a lower PCa risk under dominant model in overall population. Nevertheless, considering that rs2745557 polymorphism may play different roles in PCa susceptibility among different ethnic subgroups and the frequencies of rs2745557 polymorphism might be different among different ethnic groups which might contribute to the possible presence of heterogeneity between the studies, we further conducted subgroup analysis by ethnicity in current meta-analysis. In the stratified analysis by ethnicity, our results suggested that COX-2 rs2745557 polymorphism was associated with a lower PCa risk among subjects of Caucasians. We found that COX-2 rs2745557 polymorphism was associated with a trend of decreased PCa risk under dominant model and allelic model in African Americans, however significant relation was absent in Asians. There may be many factors influencing the result, such as differences in populations, selection factors and so on. The rs2745557 was associated with a lower PCa risk under dominant model. However, the significant association is completely lost under homogeneous co-dominant model or allelic model. So we can speculate that rs2745557 may associate with a lower PCa risk under over-dominant model. However, the significant association was absent. The reason for this phenomenon may be caused by a lack of sufficient genotype data in several studies. Considering the limited studies and population numbers of African Americans and Asians included in the meta-analysis, this may increase the risk of false negative findings, any conclusions at overall population level should be interpreted with caution. Therefore, we are not sure whether there is a significant association between the COX-2 polymorphism and decreased PCa risk in the whole population due to low statistical power.
Heterogeneity is a potential problem when interpreting the results of the present meta-analysis. In overall analysis, significant between-study heterogeneity existed in dominant model, heterogeneous co-dominant model, over-dominant model and allelic model comparisons. After subgroup analyses by ethnicity, the heterogeneity was removed under allelic model in Caucasians. In this meta-analysis, high levels of heterogeneity were observed in some comparisons. There are some factors that could have contributed toward the high heterogeneity. First, there is likely to be considerable genetic heterogeneity between the samples that were drawn from geographically diverse populations. It is known that genotype distributions differ across populations, and genotype-phenotype associations may also depend on population stratification. Second, definition of control group is different in different studies, the definition differences of the controls could have contributed to the high heterogeneity observed in our meta-analysis. Third, we attempted to determine if the high heterogeneity might also be explained by other variables such as stages of PCa, smoking status, and environmental factors included in the different studies, but are unable to provide a reliable answer to this question because we did not have access to individual level data for these variables.
Some limitations of this meta-analysis should be acknowledged. First, because only published and English articles were included in the meta-analysis, publication and potential English language biases may have occurred, even though it was not found by making use of statistical test. Second, the result should be cautiously interpreted because controls were not uniformly defined. Non-differential misclassification bias was possible because these studies may have included controls that had different risks for developing PCa. Third, only one study analyzed Asian population and another one African American population in this study. So it is quite important to have more studies and sample of Asians, Africans, and African Americans in the future so that more precise conclusion about the associations between rs2745557 polymorphism and PCa risk could be achieved. Fourth, our results were based on unadjusted estimates, while a more precise analysis should be conducted adjusted by other factors like smoking, drinking status and environmental factors. In addition, our analysis did not consider the possibility of gene-gene or SNP-SNP interactions or the possibility of linkage disequilibrium between polymorphisms. Further investigations of the haplotypic effect of a gene and the study of multiple polymorphisms in different genes are needed.