Economic Evaluation of Diagnostic Tests in the Absence of a Gold Standard: Addressing Reference Standard Bias

Health decision-making must balance clinical benefits, costs, and equity in an environment where diagnostic tests are pivotal yet often lack a definitive reference standard. This thesis explores how imperfections in reference standards bias economic evaluations of diagnostic tests and potentially mislead decision-makers. First, it reviews statistical methods designed to mitigate errors arising from an imperfect reference, highlighting approaches such as latent class analysis and adjustment equations. Next, it examines how these methods have (or have not) been integrated into economic evaluations, underscoring gaps between methodological advances and applied practice.
A conceptual framework is then presented, illustrating how misclassified test accuracy parameters can alter key metrics like incremental cost-effectiveness ratios and net health benefits, resulting in suboptimal decisions. To quantify the magnitude of such bias, simulation studies are conducted across diverse scenarios, varying disease prevalence and the degree of reference standard imperfection. These simulations reveal that modest misclassification in the reference standard can sway conclusions about a test’s value, especially in borderline cost-effectiveness contexts.
Finally, the thesis synthesises these findings into pragmatic recommendations for decision-analytic modelling of diagnostics when test accuracy evidence is generated using an imperfect reference standard. Rather than proposing universal adjustment in all health technology assessments, it argues for a proportionate, decision-focused approach: transparent reporting of reference standard limitations, structured sensitivity and scenario analyses, and targeted quantitative adjustments when decision uncertainty is high and credible information on reference standard performance is available. However, the utility of such quantitative adjustments depends critically on whether the assumption of conditional independence between test and reference standard errors can be defended in the specific clinical context; when this assumption is violated, the correction methods themselves may introduce bias. By demonstrating when and how reference standard misclassification can alter economic conclusions, this work aims to support more reliable analyses and resource allocation decisions.