20 September 2007
It has been suggested that, in the future, it may be possible to use ‘genomic profiling’ – that is, the simultaneous testing of many different gene variants – for the prediction of common diseases such a coronary heart disease or Type 2 diabetes. Continuing their careful theoretical analysis of genomic profiling, Janssens et al have analysed the clinical validity (that is, the ability to predict disease) for a variety of simulated genomic profiles that included 40 independent gene variants, tested on a hypothetical population of 1 million people in whom the average risk for the disease was set at 10% [Janssens AC et al. (2007) Genet Med. 9(8):528-35]. Using a parameter called the area under the receiver-operating characteristic curve as a measure of the discriminative accuracy of each profile (that is, its ability to distinguish those who will from those who won’t get the disease), they looked at the effect of varying the odds ratios for the genetic variants from 1.1-2.0, and varying the population frequency of the higher-risk gene variant from 1% to 50%.
Their results show that the feasibility of using genomic profiling to predict disease depends both on the odds ratio of the variants and on their population frequency, but that the frequency is particularly crucial for variants with odds ratios in the range typical for variants associated with common disease (1.1-2.0). The discriminative accuracy of a profile was low, even for variants with odds ratios between 1.5 and 2.0, if their population frequency was low (a few per cent). Only when the population frequency rose above about 30% was adequate discriminative accuracy achieved for variants with odds ratios in this range.
Janssens et al also analysed a real example in which the variants used in the profile had different odds ratios as well as different population frequencies. They found that a profile using five known variants associated with Type 2 diabetes, with odds ratios from 1.10 to 1.51 and frequencies from 6% to 78%, had poor discriminative accuracy.
Comment: This study underlines the importance of avoiding the premature use of ‘predictive’ tests based on one or a few weak susceptibility genes. Odds ratios of 1.5-2.0 are very impressive in epidemiological terms, but they only translate to a test with high discriminative accuracy if the variants that have these odds ratios are also present at high frequency in the population.
It is clear that the evaluation of genomic profiling for the prediction of common disease – or even for identifying population subsets that are at increased risk – will be an extremely complex process that will have to be carried out separately for each proposed profile. Adding even further complexity, the effects of gene-gene and gene-environment interactions will also eventually have to be taken into account.