The number of validated gene-disease associations has increased enormously over the last few years, largely due to improvements in genotyping technology which have enabled large genome-wide association studies
to be conducted based on single nucleotide polymorphisms (SNPs).
In turn, this has fuelled the creation of a ‘consumer genetics’ industry, where companies offer genome-wide scans on a direct to consumer (DTC) basis, to predict the risk of various common diseases. There are broadly three steps involved in this process: first, an individual’s genome is scanned to detect many hundreds of thousands of SNPs, some of which have been associated with disease; second, an individual’s relative
risk of disease is calculated, based on combining multiple disease-associated SNPs; and finally, this relative risk is combined with the population average risk (i.e. the incidence or prevalence of the disease) to predict the individual’s absolute
risk of getting a disease over a defined time period.
However, because the calculated risk is updated every time a new association is discovered, the prediction for an individual can change from being above average risk to below average risk overnight. This is particularly problematic where it might result in opposing recommendations. Using a cohort of 5,297 people to predict the risk of type 2 diabetes, researchers have now quantified how often this kind of ”reclassification” is likely to occur [Mihaescu R et al. (2009) Genet Med 11(8):588-594
]. When the risk predictions were updated from using just a single SNP in the TCF7L2
gene, to using 18 SNPs, and finally to using 18 SNPs plus age, sex and body mass index, around 39% of individuals changed their risk category once relative to the average risk (defined as 20%, the actual prevalence of the disease in the cohort), and 11% changed twice; nearly half the participants switched risk categories at least once when the risks were updated after every SNP. In keeping with earlier findings (see previous news
), there was a small but significant increase in the predictive accuracy of the model, and its ability to discriminate those individuals who actually had diabetes from those who did not, as the number of risk factors increased.
Comment:The fact that updating risk factors often changes an individual’s genetic risk prediction seems to present an unusual philosophical paradox. On the one hand, it is rather counterintuitive, because DNA itself is immutable – your genome doesn’t really change with time – so the genetic contribution to most diseases shouldn’t change either. On the other hand, it is absolutely to be expected, as we are currently experiencing an explosion in human genetics research; science is a dynamic process, in which hypotheses are continually tested and updated based on the best available evidence. For this reason, many contend that the application of genetics to personal risk profiling is premature, as the science is still developing rapidly and there is very little evidence of clinical validity or utility for these tests.
However, far from supporting calls to forbid such tests being available DTC, this highlights the need for transparency in the provision of information.Companies offering genome-wide risk prediction services should ensure that their customers understand that, whilst the measurement of the DNA sequenceitself (the assay) will remain constant, the interpretation of the result (the test) is likely to change as the science develops.