To the Editor:

The recent article by Mihaescu et al.1 makes important points regarding the impact of updates to risk factors and the limitations of disease risk estimates derived from genetic variants during a time of active discovery. However, there is a fundamental assumption made, both in this work and in the presentations of disease risk from commercial companies offering genome-wide scans, that is worthy of challenge. The threshold used is the population average. By presenting risks in relationship to the average in the overall population, clinical utility of the population average is implied, though this assumption most often is not supported.

In the Rotterdam Study population used by Mihaescu et al., the average risk of type 2 diabetes, calculated using both incident and prevalent cases, is reported as 20%. Based on the single TCF7L2 variant, predicted risks were 17.6%, 20.8%, and 28.0% in the CC, CT, and TT genotype groups, respectively. So, consistent with the risk reporting used by direct-to-consumer companies offering full genome scans, risk of type 2 diabetes is deemed “below average” for CC individuals and “above average” for CT and TT individuals. However, what is the meaning of “above average” risk, particularly for the CT individuals, who make up about 40% of the population,2 and in whom the predicted risk is 20.8%? In addition, what are the implications of comparing the risk in one subgroup of a population to the risk in the full population, when variants are common and subgroups make up a substantial portion of the total population (in this case 40%)?

Some well-studied, clinically developed risk scores have corresponding thresholds used in clinical care. The Framingham risk score, for example, provides estimates of the 10-year risk of heart attack or dying from coronary heart disease, based on a patient's age, gender, smoking status, diabetes status, blood pressure, and cholesterol.3 Current guidelines for prescribing cholesterol lowering therapy from the National Cholesterol Education Program-(ACT III) incorporate ranges of Framingham 10-year risk (<10%, 10-20%, and >20%),4 and physicians may use the threshold of 20% when making treatment decisions. Hence, in this setting, a threshold of 20% for Framingham 10-year risk has utility, and classifications and reclassifications based on this cutoff are consequential.

Returning to the example of type 2 diabetes, one might attempt to put the risks of disease based on genetic variants within clinical context by referring to comparable risk estimates in the clinical literature. The relationship between body mass index (BMI) and risk of diabetes has been established. Also, Narayan et al.5 estimated the remaining life-time risk of type 2 diabetes at age 18 years to be 19.8% for men of average weight (18.5 ≤ BMI <25 kg/m2), 29.7% for overweight men (25 ≤ BMI <30 kg/m2), and 57.0% for obese men (30 ≤ BMI <35 kg/m2). Similar estimates of remaining life-time risk of type 2 diabetes at age 18 years in women were 17.1%, 35.4%, and 54.6%, respectively. Based on these estimates, remaining life-time risks at age 18 years of 30-35% (overweight) and greater (55-57%, obese) provide a clinical framework, and possible thresholds, for reported risks. (To apply these cutoffs to the predicted risks presented by Mihaescu et al., however, one would have to demonstrate that the predicted risks, which might be interpreted as life-time risks of developing disease from birth to an average age of 69.5 years, are comparable with remaining life-time risk at age 18 years.)

Another approach to presenting risk estimates within a clinical context is to present risk of disease due to genetic and nongenetic factors side-by-side. This option, which our research team is currently pursuing, requires knowledge of an individual's medical, lifestyle, and family history. Granted for some diseases, risks due to lifestyle behaviors or other nongenetic factors that are relevant for all demographic groups are not available. However, presenting genetic risk alone and applying a threshold of the population average most often does not place reported risks within a meaningful clinical context.