Statistical pitfalls of “personalized medicine”

“Personalized medicine” has become a buzzword for many clinicians and geneticists during the past decade. Personalized medicine aims to match each individual with the (most appropriate) therapy that is “best suited to him/her for their condition” (this goal of course, is “FAR easier said, than done”). The author of the attached brief review was formerly at the Luxembourg Institute of Health. Before joining LIH in 2011, he had been Professor in Statistics at the University of Glasgow and at University College London. In addition to working as an academic, the author has also worked for the pharmaceutical industry in Switzerland, and the National Health Service in England. He is author of three books and claims to have expertise in statistical methods for drug development and statistical inference. He consults extensively for the pharmaceutical industry. The author [see attached] presents his view of personalized medicine.

Some of us prefer a broader, more lucid view of “personalized medicine”: Each person’s drug response should be considered holistic (i.e. characterized by treating the whole patient, taking into account mental and social factors, rather than just the physical symptoms of a disease). Every drug response is a phenotype (a trait) that reflects five contributing factors: genotype (differences in DNA sequence), epigenetic effects (independent of DNA sequence, which include DNA-methylation, RNA-interference, histone modifications, and chromatin remodeling), endogenous influences (e.g. age, disease, status of heart and kidney, hormones, exercise, stress), environmental factors (e.g. cigarette smoking, diet, consumption of alcohol and over-the-counter drugs, occupational chemicals), and microbiome differences (metabolites from gut bacteria are increasingly being realized to affect the immune system, brain, intestine, and just about all aspects of the patient).

In a recent review [Zhang G & Nebert DW, Pharmacol Ther Jul 2017; 175: 75–90 & Glossary, Mar 2018; 183: 205-206], genetic variations in inter-individual drug response were categorized in three classes, albeit as a gradient with overlapping categories: [a] monogenic traits (Mendelian; typically influenced by one or a few rare coding variants); [b] predominantly oligogenic traits (usually representing variability elicited largely by a small number of major pharmacogenes); and [c] complex pharmacogenomic traits (contributed mostly by hundreds or thousands of small-effect variants). This last category is by far the most common; each of these variants usually contributes such a small-effect to the trait, however, that they have no clinically useful predictive value, even when combined. The more that is learned about how complex the human genome is, and how unique each individual is, prediction of most drug responses — in each individual patient — seems unlikely in the foreseeable future.

Nature 29 Nov 2o18; 563: 619–621

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