A tenet of personalized medicine says that knowing genotypes will influence how physicians treat patients. Massachusetts General Hospital's James Meigs and his colleagues investigated whether knowing patients' genotypes could predict later development of type 2 diabetes as well as common risk factors can predict it. GT's Ciara Curtin spoke with Meigs about his work and its implications for personalized medicine.
Genome Technology: What do you think is the appeal of having a genotype score to predict whether people will develop a disease such as type 2 diabetes?
James Meigs: We know a lot of adverse consequences of diabetes can be avoided by preventing the disease in the first place. The question arises as to how best to find people at risk to develop diabetes and help them to make changes. If we knew genetic information, that could identify people at future risk of developing diseases like diabetes.
GT: How did you develop the genotype score you used?
JM: In the literature, there were about 18 different genetic loci that were generally considered to be true, positive diabetes risk loci. We selected these and took a simplistic attitude and asked, can you add risk alleles to generate [a] genotype score and, with this simple genotype score, discriminate future risk of diabetes? The reason for taking a relatively simple approach was that [our] overarching approach was a clinical approach. I'm a primary care doctor — simple is better. That said, we were cognizant of the fact that some of these risk alleles are better proven and [have] stronger evidence or have stronger individual effects than others.
GT: You found that the genotyping score could predict new diabetes cases, though only slightly better than knowing common risk factors. Was that surprising?
JM: It wasn't surprising, partly because we've understood for decades that phenotypic risk factors like body mass index are just super, super powerful diabetes risk factors. We did not expect genetics to be more powerful. Because the genetic determinants of diabetes appear to be independent of genetic determinants of body mass index, there was a good chance that genes could add something more. The question was, how much more?
There's two viewpoints that you can take on the predictiveness of the genotype score. The first would be individual viewpoint. No matter whether I have phenotypic diabetes risk factors or not — I could be a lean young person or obese older person — each additional risk allele would increase my relative risk to develop diabetes by 11 or 12 percent.
[And] there's this sort of population viewpoint, or how well does this score sort people in a group from lowest to highest risk? The value that we use to describe that sorting capacity is called the C-statistic. If I had a diagnostic test or a measurement I could make, like body mass index, that perfectly sorted, that would give us a C-statistic of one. If the new test was only as good as a coin toss, then the C-statistic would be 0.5.
The genotype score, all by itself, sorts properly about 58 percent of the time. The phenotype score, or the combination of body mass index and age and sex and family history and cholesterol and blood pressure, gives a C-statistic of 0.9. Clearly, phenotype score is the better sorter than the genotype score, just on the face of it. Now, if you add the two together, the predictive capacity goes from 90 percent to 90.1 percent. It doesn't really add much more to your sorting. In other words, if you sorted people on the basis of their phenotype, knowing the genotypes doesn't cause re-sorting to occur at any appreciable magnitude.
GT: Based on this and other studies, do you think that personalized medicine will come to fruition?
JM: Short answer, yes I do. It remains to be figured out how this would work in a prevention framework given the findings of our study.