However measured, OverDiagnosis is an expected value – ‘average’ – concept. For example, it is often the average percentage of persons diagnosed with a condition who would not have experienced a specified outcome if it had been left undiagnosed and untreated. Evidence-based Guidelines (EBGs) do not reflect expected values, so are they likely to produce OverDiagnoses, if implemented? Two methodological sources suggest: probably. The simpler one is that, at any and every decision node in a guideline, some degree of risk aversion is likely to be built in by the panel or organisation responsible. Whether mild or severe, embedding risk aversion will, by definition, have the effect of encouraging the implementation of interventions that have below expected value success at a group level. Secondly, much less simply, the evidence underlying an EBG will frequently be from meta-analyses that establish the changes in the target outcome produced. The researchers report, as confidence/credible intervals, the distributions for the central point change in the individual studies, and, hence, for the overall average effect change size. But in practice, EBGs can and will utilise only this average relative change, albeit accompanied by verbal acknowledgement and warning of the surrounding variance. Thus, despite widespread formal condemnation of the use of relative risks, as opposed to absolute risks, in healthcare decision making, they become embedded in practice. It does not matter whether the relationship is expressed as an arithmetic difference (e.g., Absolute Risk Difference) or as a ratio (e.g., Absolute Risk Reduction), the trial distributions have been reduced to a single number, then implemented in the EBG on the assumption that the relative effect size is independent of the base figure (risk) and is in fact equal irrespective of the base. The second argument then relies on the heterogeneity in a practice population subjected to the resulting EBG being greater – crucially with lower average base - than that in the constrained meta-analytic dataset. The effect of the consequently overestimated relative gain is compounded when the relative formulation is used because it ‘communicates more effectively’. To illustrate. In a clinical trial, statin prescription redemption was only 5.4% when patients were given an individualised Life Expectancy of 16 years and 4 months with statins vs.16 years without. Critically, the latter provides an absolute base for the change. In contrast, no less than 25% redeemed when given individualised Absolute Risk Reductions (5.7% chance of dying within ten years with statins vs. 6.6% without). Critically they were not provided with the absolute base to which these reductions apply. A profound ethical question is raised by trialling to establish which communication format maximises uptake, if any of the alternative formats presents true-but-biased-because-unbased information. But it is especially worrying if the resulting relative measures are incorporated in clinical EBGs, as a result of their supposed communicability as well as methodological practicality. Does their incorporation of risk averse preferences and overestimated relative effect gains doom EBGs to produce OverDiagnoses?.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.