Cost-effective use of diagnostic information

Abstract

Decisions about the calibration and use of diagnostic tests are still often based on simple accuracy statistics - the absolute or relative likelihoods of false positive and false negative results in relation to some defined reference standard.  For example, test thresholds may be set to minimise errors, or the choice of test to maximise the area under the Receiver Operating Characteristic (ROC) curve.  But such non-economic decision rules fail to take account of the financial and health impacts of tests, including direct test-related costs and harms, and the opportunity costs of false positives and false negatives.  Decision analytic methods have been used for many years to weigh up these various factors and to seek economically optimal ways of using tests in clinical practice.  However, this can be difficult as the cost-effectiveness of tests depends on how they are used to influence downstream treatment and care, and their consequent impact on health outcomes and costs.  Direct evidence linking the use of tests with long-term outcomes is rarely available – randomised trials of diagnostic tests are expensive and difficult to design – so modelling is often used to link short-term diagnostic accuracy data with evidence about treatment effectiveness.  Such models can be difficult to design and to populate with data, but there may also be more fundamental problems due to the choice of reference standard for the diagnostic accuracy.  For example, if this standard is related to prognosis rather than to predicted response to treatment, diagnostic accuracy statistics will not be a reliable basis for treatment prioritisation.  In this talk I will explore the relationship between diagnostic accuracy data and cost-effectiveness analysis, and discuss alternative ways of identifying cost-effective ways to use diagnostic information to prioritise treatments.