Personalized Genetic Prediction: Too Limited, Too Expensive, or Too Soon?

Genetic epidemiology has identified many common genetic variants that are associated with common diseases, and the list is growing monthly (1, 2). This success has boosted expectations for personalized genetic prediction. According to these expectations, genetic information can tell people about their risk for various diseases and which medications they should use or avoid. However, 2 articles in this issue (3, 4) suggest that this promise may be exaggerated and premature.

Paynter and colleagues (3) evaluated the predictive performance of rs10757274 for cardiovascular disease (CVD) in 22 129 white women. This gene variant has emerged from genome-wide association studies, and its association with CVD has been extensively and consistently replicated across many data sets (5). Although the authors confirmed that this polymorphism is associated with incident CVD risk, even after adjustment for other risk factors, no improvement in the prediction of CVD occurred. Adding the genotypic information to traditional risk factors—including family history—reclassified 2.6% of the women to different risk categories, but 63.4% of them were incorrectly reclassified. Knowing a patient's rs10757274 genotype would not help a clinician make better preventive or therapeutic decisions to reduce the future risk for heart disease.

The genetic effect of rs10757274, as for most associations discovered through current genome-wide association studies, is subtle, conferring an odds ratio for future heart disease of 1.25 for heterozygotes and 1.32 for homozygotes. Such effects typically explain a small proportion (<1%) of the risk variance for the disease of interest. Thus, 1 variant alone is unlikely to convey any important predictive information. The situation could improve if we discover many variants with small effects or at least a few variants with large effects.

Recent studies on the predictive performance …