LRTae: improving statistical power for genetic association with case/control data when phenotype and/or genotype misclassification errors are present
© Barral et al; licensee BioMed Central Ltd. 2006
Received: 19 September 2005
Accepted: 27 April 2006
Published: 27 April 2006
In the field of statistical genetics, phenotype and genotype misclassification errors can substantially reduce power to detect association with genetic case/control studies. Misclassification also can bias population frequency parameters such as genotype, haplotype, or multi-locus genotype frequencies. These problems are of particular concern in case/control designs because, short of repeated sampling, there is no way to detect misclassification errors.
We developed a double-sampling procedure for case/control genetic association using a likelihood ratio test framework. Different approaches have been proposed to deal with misclassification errors. We have chosen the likelihood framework because of the ease with which misclassification probabilities may be incorporated into in the statistical framework and hypothesis testing. The statistic is called the Likelihood Ratio Test allowing for errors (LRTae) and is freely available via software download.
We applied our procedure to 10,000 replicates of simulated case/control data in which we introduced phenotype misclassification errors. The phenotype considered is Ankylosing Spondylitis (AS). The LRTae method power was always greater than LRTstd power for the significance levels considered (5%, 1%, 0.1%, 0.01%). Power gains for the LRTae method over the LRTstd method increased as the significance level became more stringent. Multi-locus genotype frequency estimates using LRTae method were more accurate than estimates using LRTstd method.
The LRTae method can be applied to single-locus genotypes, multi-locus genotypes, or multi-locus haplotypes in a case/control framework and can be more powerful to detect association in case/control studies when both genotype and/or phenotype errors are present. Furthermore, the LRTae method provides asymptotically unbiased estimates of case and control genotype frequencies, as well as estimates of phenotype and/or genotype misclassification rates.
In the field of statistical genetics, phenotype and genotype misclassification errors can substantially reduce power to detect association with genetic case/control studies [1–4]. Misclassification also can bias population frequency parameters such as genotype, haplotype, or multi-locus genotype frequencies. These problems are of particular concern in case/control designs because, short of repeated sampling, there is no way to detect misclassification errors .
We developed a double-sampling procedure for case/control genetic association using a likelihood ratio test framework . Different approaches have been proposed to deal with misclassification errors . We chose the likelihood framework because of the ease with which misclassification probabilities may be incorporated into in the statistical framework and hypothesis testing.
The statistic is called the Likelihood Ratio Test allowing for errors (LRTae) and is freely available via software download . We applied our procedure to 10,000 replicates of simulated case/control data in which we introduced phenotype misclassification errors. The phenotype considered is Ankylosing Spondylitis (AS).
The program is compiled to work in UNIX Solaris, LINUX, and Windows (PC) operating systems. All commands are executed from the command line in UNIX and LINUX or DOS prompt in Windows. A full description of file requirements and program features is available via the web .
We simulated 10,000 replicates of case/control genetic association data to evaluate power and estimation of multi-locus genotypes frequencies for the LRTae and LRTstd (the likelihood ratio test that does not include double-sample information) methods. By double-sample we mean that some individuals are measured twice: once with the standard measurement instrument (called the fallible measurement) and once with a gold-standard measurement instrument (called the infallible measurement) [10, 11].
Generating multi-locus genotype frequencies for simulations
We set generating phenotype misclassification rates in each direction (case→ control, control→ case) at 10%. We did not generate genotype misclassification errors in these data because of limitations in determining permutation p-values when both categories are double-sampled. For each replicate, we randomly sampled 25% of the individuals to obtain infallible phenotype measurements. To estimate power for the LRTae and LRTstd methods, we computed permutation p-values. For each replicate, we computed the permutation p-value by randomly reassigning the multi-locus genotypes for all individuals, keeping the total count of each multi-locus genotype classification fixed. The proportion of permuted data sets whose LRTae or LRTstd statistic exceeds the observed dataset statistic is the permutation p-value for the corresponding statistic. We computed power by determining the proportion of replicates for which permutation p-values exceeded given thresholds.
Power for LRTae and LRTstd methods using permutation p-values
We also comment that the LRTae mean estimates of multi-locus genotype frequencies in cases and controls and misclassification error rates were very accurate (data not shown) with relatively small variances.
One of the main concerns for case/control genetic association studies is the presence of undetectable genotype and phenotype errors, since these misclassifications may lead to a loss in statistical power to detect association and may also lead to incorrect estimation of parameters such as multi-locus genotype- or haplotype- frequencies in cases and controls [1, 13, 14].
Although statistical methods incorporating genetic models of inheritance in association studies have verified potential power gains [15–18], we did not specify any genetic inheritance model for AS in our study since there is no consistent mode of inheritance suggested for AS . We note however that dominant and recessive modes have been reported for particular data sets [20, 21].
Regarding the impact of these misclassification errors on false positive rates (or Type I error), when non-differential misclassification is considered (that is, the misclassification rates are equal in the cross-classified groups), there is no increase in Type I error [1, 4, 13, 22]. In the case of differential misclassification, researchers  recently demonstrated in a diabetes association study that there is an increase in the Type I error rates when case and control groups have differential genotype error rates.
The observed increase in statistical power of the LTRae statistic is due the presence of double sample on some individuals; the infallible measurement establishes the correct classification for those individuals. Furthermore, the misclassification rates that are estimated from the double-sample data are used to "weight" phenotype classifications for the entire sample. More details are provided in our previous work . We note that, if we have correct misclassification rates but no double-sample information, our LRTae method is not necessarily more powerful than the LRTstd method . Double-sampling methods are among the few methods, along with Bayesian methods , that enable researches to detect and treat misclassification errors.
We also performed simulations using double-sampling proportions of 50% and 10% (full results not shown). As noted above, power gains for the LRTae method (Table 2) increased as the double-sampling proportion increased. However, in all simulations, LRTae mean estimates of multi-locus genotype frequencies in cases and controls and misclassification error rates were accurate. For the interested reader, we comment that we recently published the non-centrality parameter for our LRTae method . This parameter enables researchers to perform power and sample size calculations for the LRTae statistic and also to perform cost/benefits analyses. That is, researchers may ask whether, for a fixed cost, it is more powerful to genotype larger samples subject to misclassification error or to collect smaller samples, some proportion of which have been double-sampled.
The LRTae method has the flexibility to perform case/control association analysis when the genotype data are single-locus genotypes, multi-locus genotypes, or multi-locus haplotypes, and to determine significance via permutation methods when double-sampling has been performed on only one of the categories (phenotype or genotype).
When double sampling data are available on either phenotypes, genotypes, and/or haplotypes, we strongly recommended the LRTae method to: (i) improve the statistical power to detect genetic association in case/control studies in the presence of misclassification error; (ii) obtain estimates of misclassification probabilities from double sample information; and (iii) use these misclassification parameters to weight the estimates of population frequency parameters.
Availability and requirements
Project name: LRTae software.
Project home page: See .
Software availability: See .
Operating system: UNIX Solaris, LINUX, and Windows.
Programming language: C++.
Other requirements: None.
Any restrictions to use by non-academics: None.
The software is freely available via download from the ftp site listed above (Software availability).
Likelihood Ratio Test allowing for errors
Likelihood Ratio Test standard method
Single Nucleotide Polymorphism
Tumor Necrosis Factor
The authors gratefully acknowledge grant MH44292 from the National Institutes of Health. The authors also gratefully acknowledge Robert Inman, Professor of Medicine, University of Toronto, and Proton Rahman, Associate Professor of Medicine, Memorial University Newfoundland, for allowing access to their samples.
- Mote VL, Anderson RL: An investigation of the effect of misclassification on the properties of chisquare-tests in the analysis of categorical data. Biometrika. 1965, 52: 95-109. 10.2307/2333815.PubMedGoogle Scholar
- Zheng G, Tian X: The impact of diagnostic error on testing genetic association in case-control studies. Stat Med. 2005, 24 (6): 869-882. 10.1002/sim.1976.PubMedView ArticleGoogle Scholar
- Edwards BJ, Haynes C, Levenstien MA, Finch SJ, Gordon D: Power and sample size calculations in the presence of phenotype errors for case/control genetic association studies. BMC Genet. 2005, 6 (1): 18-10.1186/1471-2156-6-18.PubMedPubMed CentralView ArticleGoogle Scholar
- Gordon D, Finch SJ, Nothnagel M, Ott J: Power and sample size calculations for case-control genetic association tests when errors are present: application to single nucleotide polymorphisms. Hum Hered . 2002, 54 (1): 22-33. 10.1159/000066696.PubMedView ArticleGoogle Scholar
- Rice KM, Holmans P: Allowing for genotyping error in analysis of unmatched cases and controls. Ann Hum Genet. 2003, 67: 165-174. 10.1046/j.1469-1809.2003.00020.x.PubMedView ArticleGoogle Scholar
- Gordon D, Yang Y, Haynes C, Finch SJ, Mendell NR, Brown AM, Haroutunian V: Increasing power for tests of genetic association in the presence of phenotype and/or genotype error by use of double-sampling. Stat Appl Genet and Mol Biol. 2004, 3 (1): Article 26Google Scholar
- Gustafson P: Measurement error and misclassification in statistics and epidemiology: impacts and Bayesian adjustments. Boca Raton: Chapman and Hall/CRC. 2004Google Scholar
- Haynes C, Gordon D: LRTae software, version 1.3.1. 2004, [ftp://linkage.rockefeller.edu/software/lrtae/]Google Scholar
- Gordon D: User's Guide for the LRTae software. 2004, [http://linkage.rockefeller.edu/derek/UserGuideLRTae.htm]Google Scholar
- Tenenbein A: A double sampling scheme for estimating from binomial data with misclassifications. J Am Stat Assoc. 1970, 65 (331): 1350-1361.View ArticleGoogle Scholar
- Tenenbein A: A double sampling scheme for estimating from misclassified multinomial data with applications to sampling inspection. Technometrics. 1972, 14 (1): 187-202. 10.2307/1266930.View ArticleGoogle Scholar
- van der Linden S, Valkenburg HA, Cats A: Evaluation of diagnostic criteria for ankylosing spondylitis. A proposal for modification of the New York criteria. Arthritis Rheum. 1984, 27 (4): 361-368.PubMedView ArticleGoogle Scholar
- Bross I: Misclassification in 2 × 2 tables. Biometrics . 1954, 10: 478-486.View ArticleGoogle Scholar
- Gordon D, Finch SJ: Factors affecting statistical power in the detection of genetic association. J Clin Invest. 2005, 115: 1408-1418. 10.1172/JCI24756.PubMedPubMed CentralView ArticleGoogle Scholar
- Schaid DJ, Sommer SS: Genotype relative risks: methods for design and analysis of candidate-gene association studies. Am J Hum Genet. 1993, 53 (5): 1114-1126.PubMedPubMed CentralGoogle Scholar
- Slager SL, Schaid DJ: Case-control studies of genetic markers: power and sample size approximations for Armitage's test for trend. Hum Hered. 2001, 52 (3): 149-153. 10.1159/000053370.PubMedView ArticleGoogle Scholar
- Freidlin B, Zheng G, Li Z, Gastwirth JL: Trend tests for case-control studies of genetic markers: power, sample size and robustness. Hum Hered. 2002, 53 (3): 146-152. 10.1159/000064976.PubMedView ArticleGoogle Scholar
- Wang K, Sheffield VC: A constrained-likelihood approach to marker-trait association studies. Am J Hum Genet. 2005, 77 (5): 768-780. 10.1086/497434.PubMedPubMed CentralView ArticleGoogle Scholar
- Milicic A, Lindheimer F, Laval S, Rudwaleit M, Ackerman H, Wordsworth P, Hohler T, Brown MA: Interethnic studies of TNF polymorphisms confirm the likely presence of a second MHC susceptibility locus in ankylosing spondylitis. Genes Immun. 2000, 1 (7): 418-422. 10.1038/sj.gene.6363701.PubMedView ArticleGoogle Scholar
- Robinson WP, van der Linden SM, Khan MA, Rentsch HU, Cats A, Russell A, Thomson G: HLA-Bw60 increases susceptibility to ankylosing spondylitis in HLA-B27+ patients. Arthritis Rheum. 1989, 32 (9): 1135-1141.PubMedView ArticleGoogle Scholar
- Tsui HW, Inman RD, Paterson AD, Reveille JD, Tsui FW: ANKH variants associated with ankylosing spondylitis: gender differences. Arthritis Res Ther. 2005, 7 (3): R513-525. 10.1186/ar1701.PubMedPubMed CentralView ArticleGoogle Scholar
- Gordon D, Ott J: Assessment and management of single nucleotide polymorphism genotype errors in genetic association analysis. Pac Symp Biocomput. 2001, 18-29.Google Scholar
- Clayton DG, Walker NM, Smyth DJ, Pask R, Cooper JD, Maier LM, Smink LJ, Lam AC, Ovington NR, Stevens HE, Nutland S, Howson JM, Faham M, Moorhead M, Jones HB, Falkowski M, Hardenbol P, Willis TD, Todd JA: Population structure, differential bias and genomic control in a large-scale, case-control association study. Nat Genet. 2005, 37 (11): 1243-1246. 10.1038/ng1653.PubMedView ArticleGoogle Scholar
- Ji F, Yang Y, Haynes C, Finch SJ, Gordon D: Computing Asymptotic Power and Sample Size for Case-Control Genetic Association Studies in the Presence of Phenotype and/or Genotype Misclassification Errors. Stat Appl Genet and Mol Biol. 2005, 4 (1): Article 37Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.