Issues Specific To Specialized Fields Genetic Association Studies

Genetic association studies evaluate the association between specific genetic polymorphisms and disease. They often assess relatively small effects against a noisy background of biological and social complexity, and consequently, tend to lack statistical power [45]. Power calculations are critical for genetics researchers who wish to map susceptibility genes, with chi-square being the most commonly used test statistic for association studies. The noncentrality parameter for the chi-square asymptotic distribution developed by Mitra [46] is the key to computing power and sample size for this type of study. Gordon and Finch present detailed instructions (including step-by-step use of a Web-based calculator) for this, as well as reviewing other concepts of statistical genetics and the best study/analysis design to optimize power for this type of study [17].

Synthesizing evidence from multiple studies has been used as a means of increasing power; however, it seems that metaanalyses in this field have been fraught with serious flaws. As well as more general concerns, there are genetic issues particular to molecular association studies, including checking Hardy-Weinberg equilibrium (HWE, discussed below), handling data from more than two groups while avoiding multiple comparisons, and pooling data in a way that is sensitive to genetic models [47]. As regards the latter two points, there are always at least three possible genotypes to compare, but in practice most studies reduce the number of comparisons by assuming a specific genetic model, such as dominant or recessive, even though there are often no biological justifications for assumption of the model. Using an inappropriate genetic model or inappropriately pooled data produce misleading estimates of odds ratios. Appropriate metaanalysis methods and avoiding assumptions about genetic models are discussed further by Minelli et al. [45] and Thakkinstian et al. [48].

Bias seems to be another serious problem in this type of study. Some methods for detecting and correcting it are discussed by Sterne et al. - 49] ; publication bias, for example, may be detected by a simple funnel plot of the data. Not just the metaanalyses, but also single studies in this field seem excessively prone to various sources of bias, with some critics going so far as to assert that most documented associations, even those that are replicated, represent nothing other than false positives based on bias - 50] - The list of prevalent bias sources is impressive and includes biological plausibility bias, ascertainment bias, publication bias, selective reporting bias, spectrum of disease bias, population stratification, biased selection of controls, lack of blinding in the genotyping process, and genotyping error.

Some of these error sources can be addressed simply through researchers being cognizant of them and striving to avoid them. For example, one can choose to blind genotyping rather than failing to do so. In another example, ascertainment bias can be avoided by scrutinizing controls with the same intensity as cases; this type of bias relates to the situation where affected individuals ("cases") have their DNA resequenced to identify rare variants. Identification of such rare variants in an affected group does not necessarily signify a role in disease (as is often assumed when only cases are ascertained by resequencing) since sequencing DNA from any group tends to turn up a few rare mutations. Instead, a strong and statistically convincing preferential presence of variants in the cases compared to controls (that had been scrutinized in the same manner) would support the involvement of the variants in disease [51].

In cases where being cognizant of the potential for bias does not allow us to avoid it, designing experiments and/or data analysis with an expectation of inevitable bias can dampen its influence. For example, designing studies to have higher than desired power can combat the power loss expected from genotyping misclassification errors, as discussed earlier. As another example, various statistical techniques that are robust to bias can be used, such as genomic control techniques or family-based methods to address problems of population stratification [17]. Population stratification refers to the situation in case-control studies where control groups are not well matched to case groups and in fact have different typical levels of what is being measured simply because they are from different populations. In this situation, association tests like chi-square may falsely indicate associations, even with as high as 100% probability.

In fact, having appropriate control groups is a key determinant of the validity of genetics association studies, and determining whether controls deviate from HWE is a standard way of evaluating this. Theoretically, disease-free control groups from outbred populations should follow HWE, as should combined cases and controls if they both have a particular disease (i.e., such as in studies where different treatments are evaluated). If they do not, it is a signal of some peculiarity, error, or problem with the data sets that could invalidate the key inferences from a genetic association study [52]. For example, if there is a recessive model and the control group has an excess or deficit of one group of homozygotes, this will directly affect calculation of the odds ratio (the control homozygotes divided by the other genotypes is the denominator of the odds ratio). Overall, if there is a significant deviation from HWE it should induce some thinking about the study; deviations could result from genotyping errors but may also arise from other sources. For example, HWE deviation may suggest that allele-based estimates of genetic effects are biased, or may give further insight into the population from which the data are derived.

A recent review of studies published in high -quality specialized genetics journals demonstrated that HWE is very commonly tested improperly or inadequately -3] . Of 776 associations tested, only 29% reported on HWE, introducing uncertainty as to if it was tested in the others. Furthermore, where HWE testing was described, the test was applied to the correct control group in only 50% of the associations tested. It is a common error to include the disease cases with the control cases in the HWE test when controls are disease-free. Combined cases and controls should be tested only when the controls have the same disease as the cases; otherwise, only controls should be tested. These authors recalculated HWE for the studies reviewed and noted that in most of the samples where HWE was actually violated, this was either not mentioned in the original article or HWE conformity was actually claimed.

Another common problem in the studies reviewed was the unjustified use of an inappropriate test. A number of different statistical tests could be used to test HWE (including chi- square, exact tests, and Bayesian methods), all based on the conditional probability that there would be the number of homozygotes that actually turned out to be in the sample. The chi-square test was the only test applied to HWE calculation in the studies reviewed, despite the fact that the chi-square asymptotic distribution is inadequate to deal with low genotype frequencies and is therefore not justified in studies involving them; in this case an exact test provides a simple and superior alternative [3,54].

A final serious concern in the studies reviewed was that only 7% of the studies had an acceptable power to detect HWE deviation, with most studies being much too underpowered to make any claim of lack of deviations. Whereas power to detect HWE is of secondary importance compared to the prime consideration of power to detect a genetic association, undetected modest HWE deviations could affect considerably the inferences of many genetic association studies [3].

Project Management Made Easy

Project Management Made Easy

What you need to know about… Project Management Made Easy! Project management consists of more than just a large building project and can encompass small projects as well. No matter what the size of your project, you need to have some sort of project management. How you manage your project has everything to do with its outcome.

Get My Free Ebook


Post a comment