Overestimation of genetic risks owing to small sample sizes in cardiovascular studies

We sought evidence of publication bias to explain conflicting findings in studies of angiotensin-converting enzyme deletion polymorphism (ACE D) and glycoprotein IIIa PlA2 (PLA2) polymorphism and the risk of myocardial infarction. Factor 5 Leiden (F5L), a well-established thrombotic risk factor, served as an internal comparison. We conducted systematic reviews of published studies involving ACE D, PLA2, F5L and relevant outcomes, searching medline (January 1990 through February, 2001), bibliographies, and meta-analyses. Random effects pooled odds ratios (95% confidence interval) for cardiovascular outcomes were as follows: PLA2 (n = 13,167 subjects): 1.13 (1.02, 1.26); ACE D (n = 42,140 subjects): 1.22 (1.11, 1.35); and F5L (n = 27,277 subjects): 4.43 (3.65, 5.38). However, funnel plots of ACE D and PLA2, but not F5L, showed an inverse relationship between sample size and odds ratios for ACE D (p = 0.02) and PLA2 (p = 0.04) but not F5L (p = 0.65) by Egger's test for potential publication bias. Despite research-based genotyping of over 50,000 subjects, the overall risk for myocardial infarction as a result of PLA2 and ACE D remains doubtful. Our study provides a clear example of how publication of underpowered studies can spuriously implicate polymorphisms as genetic risk factors.