As is typical of decision-making in clinical diagnosis, choices in clinical genetics are never straightforward and always lead to subsidiary questions. Similarly in genetics itself, as in physiology or biochemistry, an answer to a question leads to other questions about that answer, and investigators and clinicians are led deeper into the maze.
One undertakes the journey into the labyrinth, of course, if one is interested in pursuing science. If one’s concerns are pecuniary, then the most superficial explanation suffices.
On the science side, our deep understanding of genetic mechanisms – the relationships between genotype and phenotype – continue to unfold. The facile explanations of the hallowed Central Dogma no longer seem so certain. The complex interactions of regulatory networks and the expanding roles of RNA (as we’re beginning to learn them) make the simple sequence of gene/transcript/protein seem naive.
Genome-wide association studies (GWAS) are useful tools that may provide valuable information, but the data they generate are derived from statistical analyses and there’s the opportunity for misinterpretation or even mischief. It’s well-known there are “lies, damn lies, and statistics”.
The promise of GWAS is described as (1) increasing the understanding of disease pathophysiology and etiology and (2)refining an individual’s risk of developing a disease. These are promises. These are hopeful outcomes. It is early days yet for genetic science, despite the fevered claims of CEOs of biotechnology startups.
There are several challenges regarding the appropriate and effective use of information derived from GWAS. The first and primary concern will always relate to the validity of the data. Many potential sources of bias exist and require strict analysis. True associations need to be uncovered using a “drill-down” approach and not merely read off the top of the list. Suspected associations must be confirmed by comparison with data from other studies. Identifying real associations is a multistep process.
On the raw statistics alone, across a panel of 50 diseases most people will be ranked in the top 5% of risk for at least one disease. So the possibility of false positives must be considered. This is similar to methods for deriving “normal” ranges for laboratory testing. Samples are taken from a large sample of healthy people. The top and bottom 2.5% are deleted, and the remaining 95% of the distribution is retained as the normal range. Thus, 5% of the normal population will be evaluated as “abnormal” for any given test. If a panel of 20 lab tests is done on a normal person’s blood sample, she will be “abnormal” for one test.
If the physician is not familiar with statistical analysis, she may incorrectly interpret the laboratory data. The same opportunity for misinterpretation exists for analysis of GWAS data.
