A more rapid approach to systematically assessing published associations of genetic polymorphisms and disease risk: type 2 diabetes as a test case

摘要

Background: Comparative effectiveness research and research in genomic medicine are not orthogonal pursuits. Both require a robust evidence base, and each stands to benefit from applying the methods of the other. There is an exponentially growing literature reporting associations between single nucleotide polymorphisms (SNPs) and increased risk for diseases such as type 2 diabetes. Literature-based meta-analysis is an important method of assessing the validity of published gene-disease associations, but a traditional emphasis on exhaustiveness makes it difficult to study multiple polymorphisms efficiently. Here we describe a novel two-step search method for broadly yet systematically reviewing the literature to identify the "most-studied" gene-disease associations, thereby selecting those with a high possibility of replication on which to conduct abbreviated, simultaneous meta-analyses. This method was then applied to identify and evaluate the validity of SNPs reported to be associated with increased type 2 diabetes risk, to demonstrate proof of principle.Methods: A two-step MEDLINE search (1950 to present) was conducted in September 2007 for published genetic association data related to SNPs associated with risk of type 2 diabetes. The top 10 "most-studied" genes were selected for focused searches and final inclusion/exclusion determinations. To demonstrate the ability to efficiently update this two-step search for additions to the literature, an update of the second-step search was conducted 9 months later. Abstracted data were sorted based on study design, risk model, and specific SNPs. Meta-analyses were performed for individual SNPs, with separate analyses done for case-control and prospective studies, and were compared with the results of more recent genome-wide association studies.Results: The first-step search found 1116 articles covering 108 different genes. The top ten "most-studied" genes were: ABCC8 (or SUR1), ACE, CAPN10, KCNJ11 (or Kir6.2), HNF1 alpha, HNF4 alpha, IL-6, PGC-1 alpha, PPAR gamma 2, and TCF7L2. The second-step search found a total of 658 articles, yielding 124 articles for initial data abstraction and analysis. We also demonstrated the ability to update this search as newer studies appeared, using the same method almost a year later to find an additional 107 articles (77 were ultimately excluded), bringing the number of included studies to 154. From these studies, data on 90 different DNA variants within the ten genes were abstracted. Simultaneous meta-analyses found that higher-risk alleles for SNPs rs7903146 and rs12255372 in TCF7L2, rs1801282 in PPAR gamma 2, rs5219 in KCNJ11, rs3792267 in CAPN10, rs2144909 in HNF4 alpha, and rs1800795 in IL-6 appeared to be associated with increased type 2 diabetes risk. These findings were generally highly concordant with the results of traditional literature-based meta-analyses performed for individual genes.Conclusions: The methodology described in this manuscript represents a reasonable approach to more rapidly identifying and evaluating frequently studied genetic-risk markers for diseases such as type 2 diabetes. Comparison with results of traditional meta-analyses suggests that these gains in efficiency do not necessarily come at the price of reduced accuracy. Given the quickening pace of discovery of such markers, more efficient, unbiased, and readily updatable methods for systematically assessing and re-assessing a changing literature could prove valuable. Good methods for evidence evaluation are also important to the potential application of genetic markers to comparative effectiveness research, and vice versa.