J Clin Epidemiol. 2015 Apr;68(4):460-5. doi: 10.1016/j.jclinepi.2015.01.001. Epub 2015 Jan 09.
P < 5 × 10(-8) has emerged as a standard of statistical significance for genome-wide association studies.
Journal of clinical epidemiology
Anne-Sophie Jannot, Georg Ehret, Thomas Perneger
Affiliations
Affiliations
- Division of Clinical Epidemiology, University Hospitals of Geneva, rue Gabrielle Perret-Gentil 4, 1211 Geneva, Switzerland. Electronic address: [email protected].
- Division of Cardiology, University Hospitals of Geneva, rue Gabrielle Perret-Gentil 4, 1211 Geneva, Switzerland.
- Division of Clinical Epidemiology, University Hospitals of Geneva, rue Gabrielle Perret-Gentil 4, 1211 Geneva, Switzerland.
PMID: 25666886
DOI: 10.1016/j.jclinepi.2015.01.001
Abstract
OBJECTIVES: In genome-wide association studies (GWASs), the practice regarding the choice of thresholds of significance and of thresholds used to include single nucleotide polymorphisms (SNPs) in a further validation stage is not well known. Here, we performed a systematic analysis of all GWASs published in two recent but nonconsecutive periods to assess whether any consensus was emerging regarding the choice of these thresholds.
STUDY DESIGN AND SETTING: We identified 167 GWASs published during the first semester of 2011 and 105 published in the third trimester of 2012 and retrieved the genome-wide threshold of significance and the thresholds to include SNPs in a replication stage if applicable.
RESULTS: The proportion of studies using 5 × 10(-8) as a genome-wide significance threshold increased between 2011 and 2012 (40% vs. 64%, P < 0.001), whereas the proportion of articles that gave a justification for this threshold decreased. The distribution of thresholds used to include SNPs in the replication stage remained stable over time (median 10(-5)), and the variance remained large [interquartile range (10(-6), 10(-4))].
CONCLUSION: Although 5 × 10(-8) has become the de facto standard genome-wide threshold, practice regarding the choice of inclusion threshold for the replication step remains heterogeneous and did not ensure the best trade-off between power and type 1 error.
Copyright © 2015 Elsevier Inc. All rights reserved.
Keywords: Evaluation studies as topic; False discovery rate; Genome-wide association study; Multiple testing; Significance thresholds; Validation studies
MeSH terms
Publication Types