Generalising the results of clinical trials

Article type
Authors
Glasziou P, Irwig L
Abstract
To which groups of patients can we apply the results of clinical trials? Authors and readers often answer this question by reference to the trial entry criteria, that is, the sample frame from which subjects came. As an alternate and more rational answer to the question, we extend a model suggested by Lubsen and Tijssen: the separate assessment and individual prediction of the benefit of treatment and its harm. Their model can be expressed by the following equation:

Net Benefit = Risk Level x Risk Reduction � Adverse Effects

The model suggests potential patient benefit increases with risk � those most at risk have most to gain - but that harm will remain relatively fixed. Thus at low levels of risk, the benefits will not outweigh the harm and we should refrain from treatment, but at higher levels the benefit will outweigh the harm.

To complete this model generally requires several sources of data: the relative risk reduction should come from randomised trials, the adverse event rates may come from both randomised trials and epidemiological studies; the individualised patient risk level will usually come from multivariate risk equations derived from large cohort studies, thus allowing individualised predictions of net benefit from the above equation. Before making firm conclusions, the model assumptions, of fixed adverse effects and constant relative risk reduction, need to be checked. In addition, it will be useful to search for modifier of either the benefit and harm by a subgroup analysis. We illustrate these methods with the use of examples and compare this model to how authors approach generalisation in recently published trials and meta-analyses.