The illusion of polygenic disease risk predictio

These GEITP pages continue to discuss genome-wide association studies (GWAS) which, since ~2006, are designed to detect genes or genetic loci correlated with whatever trait (phenotype) the researchers have chosen to study. Rather than Mendelian traits (usually reflecting defects in one or very few genes) — GWAS are performed in order to gain insight (i.e. find the cause, predict future outcome, and discover possible treatment) into multifactorial traits — which include complex diseases (e.g. schizophrenia, type-2 diabetes), quantitative traits (e.g. height, body mass iindex), as well as response to a drug or environmental toxicant or mixture of toxicants. A problem at the interface of genomic medicine and medical screening, however, is that “genetic associations of etiological significance” are often interpreted as “having predictive significance” (i.e. percent risk of developing schizophrenia, type-2 diabetes, etc.).

GWAS have identified many thousands of associations between common DNA single-nucleotide variants (SNVs) and hundreds of diseases, as well as quantitative traits. This knowledge has generated many publications — with the expectation that it can be used to derive polygenic risk scores for predicting disease (or outcome) to identify those at sufficiently high risk, so that they might benefit from preventive intervention (i.e. how to prevent schizophrenia or type-2 diabetes from occurring in that individual). However, this expectation rests on the incorrect assumption that odds ratios (statistical quantification of the strength of an association between two events, A and B; the OR is defined as the ratio of the odds of A in the presence of B, and the odds of A in the absence of B) derived from polygenic risk scores (that can be important, etiologically) — are also directly useful in risk prediction and population screening.

As summarized [see attached 3-page Commentary], two widely publicized recent papers (Nat Genet 2018; 50: 1219 and Am Coll Cardiol 2018; 72) illustrate the problem; these papers show associations between polygenic risk scores and a number of common disorders, including coronary artery disease. Their results demonstrate the importance of genetic variation in the etiology of these disorders — but not the value of the risk score proposal in disease prediction (i.e. screening) — contrary to what is suggested by the authors of these two papers. Those authors suggested that polygenic risk scores could be used to initiate preventive intervention among individuals with a high score, but not among those with a low score. Authors of this Commentary state that this suggestion is based on a misconception that the estimates of relative risk (e.g. odds ratios or hazard ratios) can adequately assess the discriminatory value of polygenic risk scores as screening tests.

The problem is that an odds ratio or hazard ratio does not directly indicate the discriminatory value of a screening test. To assess the discriminatory value, it is necessary (whenever possible), to specify the detection rate (sensitivity), and risk score cut-off, for a given false-positive rate (a test result that incorrectly indicates that a particular condition is present, or will become present) — or the false-positive rate and risk score cut-off for a given detection rate. For those interested in further details, please read carefully and study this very worthwhile Commentary. 😊


Genet Med Aug 2019; 21: 1705-1707

This entry was posted in Center for Environmental Genetics. Bookmark the permalink.