Effect Modifiers and Statistical Tests for Interaction in Randomized Trials

Abstract

Statistical analyses of randomized controlled trials (RCTs) yield a causally valid estimate of the overall treatment effect, which is the contrast between the outcomes in two randomized treatment groups commonly accompanied by a confidence interval. In addition, the trial investigators may want to examine whether the observed treatment effect varies across patient subgroups (also called 'heterogeneity of treatment effects'), i.e. whether the treatment effect is modified by the value of a variable assessed at baseline. The statistical approach for this evaluation of potential effect modifiers is a test for statistical interaction to evaluate whether the treatment effect varies across levels of the effect modifier. In this article, we provide a concise and nontechnical explanation of the use of simple statistical tests for interaction to identify effect modifiers in RCTs. We explain how to calculate the test of interaction by hand, applied to a dataset with simulated data on 1,000 imaginary participants for illustration.

Total hip arthroplasty versus progressive resistance training in patients with severe hip osteoarthritis: protocol for a multicentre, parallel-group, randomised controlled superiority trial

“…The statistical approach for this evaluation of potential effect modifiers will be a test for statistical interaction to evaluate whether the treatment effect varies across levels of the effect modifier. 89 …”

Who are likely to benefit from the Good Life with osteoArthritis in Denmark (GLAD) exercise and education program? An effect modifier analysis of a randomised controlled trial

“…For the analyses of the main trial (5) we applied repeated measures mixed linear models with adjustment for stratification facors. To simplify and facilitate interpretation (22), we have applied a simpler and more conservative statistical approach to this study (9).…”

“…A registry-based cohort study from the GLAD registry in Denmark (>6,700 patients) explored 51 patient characteristics as predictors of changes in knee OA pain following the GLAD programme but did not identify any relevant predictors (7). However, registry-based cohort studies are prone to bias and a more valid method for identifying effect modifiers is through secondary analyses of RCT data as the RCT allows for causal effect estimations (8) and less biased exploration of effect modifiers (9). Our recent RCT (5) is the only RCT of the GLAD programme that has been conducted and secondary exploration of effect modifiers has not been done before.…”

Effectiveness of an expert assessment and individualised treatment compared with a minimal home-based exercise program in women with late-term shoulder impairments after primary breast cancer surgery: study protocol for a randomised controlled trial

“…Subgroup analyses [ 82 ] will be used to examine whether the observed overall treatment effect varies across participants’ subgroups, and to whether the effect is modified by the value of a variable assessed at baseline: analysed by thresholds median age, median duration of shoulder symptoms, obesity (BMI ≥ 30 kg/m 2 ), dominant side affected (left vs right). This statistical approach to evaluate potential effect modifiers will be a test for statistical interaction on whether the treatment effect (net benefit SPADI score) varies across levels of the effect modifier [ 83 ].…”