Subject–Treatment Interaction

That the effect of a treatment will vary among subjects is not surprising, nor is it a recent concept. Roses provided an 1892 quote by Sir William Osler, ‘‘If it were not for the great variability among individuals medicine might as well be a science and not an art.’’ Subject–treatment (S–T) interaction is, as the term implies, an interaction of specific subjects with applied treatment(s). The result of such interaction is a variability of ‘‘individual treatment effects’’ or ‘‘individual treatment heterogeneity’’ in a population of interest. Effects of treatment may be assessed as a measurement of efficacy or toxicity because variability in either may have important consequences. Although such variability has often been acknowledged as an important consideration, medicine today generally makes use of statistical information gathered about the general population (often about the ‘‘average’’ subject) and then applies it to the individual. When there is a high degree of S–T interaction in a population, there may be a nonnegligible proportion of the population responding differently to a treatment, and possibly in the opposite direction, from the average subject. Even within a carefully designed clinical trial, the validity of results may be compromised if there is a wide variability of individual treatment effects, prompting at least one author to suggest that inference about the mean treatment effect be supplemented with information about suspected variability of effects. Variability in magnitude has been called a noncrossover interaction as long as the direction of the effect is the same across subjects, and variation in direction of individual effects has been termed a crossover interaction. The latter is usually of more concern to researchers, possibly playing a role in adverse drug reactions. The degree of S–T interaction in a population can be quantified by a parameter that is related to the variance of individual treatment effects, although this parameter cannot be directly estimated using observable data. However, S–T interaction can produce effects that are testable using such data. Methods to detect these effects focused on identifying subsets or covariates to explain perceived treatment heterogeneity. OVERVIEW