Translating Trial Results in Clinical Practice: the Risk GP Model

Abstract

The medical research community is fairly unanimous in its agreement that Btranslating scientific discoveries^ and Btranslating research findings^ is important, but achieving translation remains a challenge [1]. There exists a research literature on strategies for knowledge dissemination or transfer [2]. In comparison, an equally important aspect of translation, the incorporation of research findings in clinical reasoning, has been all but ignored. What does a Brelative risk reduction of 20 %^ in a population of clinical trial participants tell us about whether the drug will work for a particular patient outside the trial? In this article, we examine the logic of translating therapeutic study results (e.g., clinical trial findings) in clinical practice. The kind of translation we have in mind is often called knowledge translation (KT) or T2 [1]. Although little explicit attention is given to the logic of translation, we argue that a particular model of translation is implicit in most KT activities; we call it the Risk Generalization-Particularization (Risk GP) Model. Here, we describe the model, including its limitations. Elsewhere, we argue that the Risk GP Model is the standard model of prediction in prognosis and therapy [3]. In therapy, prediction requires making an inference about a patient’s response to treatment. In the Risk GP Model, this involves translating study findings in two stages. In the first stage, generalization, translation involves transportation of the group-level effect size from the trial to a target population. The preferred measures of effect size, particularly in cardiovascular research, are derived from the absolute risk (AR), and include the relative risk (RR) and absolute risk reduction (ARR). For instance, in a meta-analysis of trials, the RR of cardiovascular disease (CVD) events due to a cholesterollowering statin was approximately 0.8 [4]. In the Risk GP Model, the evidence user generalizes this RR of 0.8 to a target population (e.g., patients at high risk of CVD in the USA). In the second stage, particularization, translation involves transformation of the effect size into a change in outcome for a patient in the target population. Thus, we call this model the Risk GP Model. The inference from populations to individual patients (particularization or Bindividualization^) has received virtually no attention, yet is presupposed whenever clinicians Bapply^ study results to particular patients. When the effect size is derived from the AR (which quantifies the relative frequency of outcomes/events), the output of particularization is the change in probability of the outcome for the patient. Thus, if a patient in the high risk target population has an untreated CVD risk of 25 %, we predict that a statin with a RR of 0.8 will lower their probability of CVD to 20 %. The slip in the meaning of Bmedical risk^ at this stage often goes unnoticed. The RR or ARR measured in the trial is Associate Editor Craig Stolen oversaw the review of this article