Lowell Reed Lecture: Post-Genomic Age and the Unfulfilled Promise of Personalized Medicine: What's Missing? What Can Biostatisticians Do to Help?

A decade ago, the decoding of the human genome raised high expectations for unlocking the causes of complex diseases and devising personalized approaches to medicine and risk prediction. Aside from a few stunning successes, that promise has not been realized. As we should have suspected from twin studies, the inherited genome often acts as a relatively minor player for complex diseases. The missing pieces likely involve complex interactions between genetic variants and environmental exposures, possibly involving epigenetic modifications that distort gene expression. Methods to identify and characterize those complex interactions have proved elusive.

The challenges are both biological and statistical. Suppose a genetic variant and an environmental exposure are both risk factors for a dichotomous phenotype. Methods to characterize joint effects should help to uncover the causal processes. Do complex diseases have shared pathways or co-regulated families of genes? What can we learn from intermediate phenotypes for those who are not yet sick, and from more finely-defined disease phenotypes for those who are?

Joint effects on risk are statistically hard to identify and characterize even for simple questions. The rule of thumb is that a factor-of-four sample size is needed, compared to that needed to demonstrate the separate effects of either genes or exposures. So ingenious approaches have been developed, either through risk-based sampling for a cohort, or case-control sampling, extreme phenotype sampling, pooling, two-stage sampling, exposed-only, or case-only designs. Bias can arise when assessing joint effects, and emerging approaches that offer relative protection against such biases will be discussed.