OBJECTIVES: Health Technology Assessments (HTAs) considered by the National Institute for Health and Care Excellence (NICE) Diagnostics Assessment Program often concern the comparative or combined accuracy of two or more diagnostic tests. Their estimation requires modelling techniques that account for within-study correlations between the tests. We describe a novel application of a meta-analytic model with copulas to capture within-study dependencies between two tests assessed in the same patient group. The methodology is applied to a motivating example assessing the accuracy of emerging biomarkers in Alzheimer’s disease dementia.

METHODS: We developed Bayesian meta-analysis models for evaluating the accuracy of two diagnostic tests in the same patients, using bivariate copulas to flexibly capture within-study dependencies between the two tests. Five bivariate copula models capturing different dependence structures were described: Gaussian, Frank, Gumbel, Clayton and Clayton rotated 180°. The models were compared to the currently recommended meta-regression approach for modelling two cerebrospinal fluid biomarker tests: amyloid-β 42 (Aβ₄₂) and total tau (t-tau). Model fit was assessed using the widely applicable information criterion (WAIC).

RESULTS: Aβ₄₂ and t-tau demonstrated sensitivities of 80.9% (95% credible interval: 73.4%, 87.5%) and 76.4% (69.4%, 83.1%), respectively. Summary specificity was 70.3% (61.3%, 78.4%) and 72.5% (63.7%, 81.3%), respectively. The bivariate copula models resulted in a better fit compared to the meta-regression model, and increased precision in estimates of sensitivity and specificity by as much as a 15% reduction in the width of the 95% credible intervals.

CONCLUSIONS: The bivariate copula framework supports health technology assessment, enabling test comparisons while accounting for complex dependence structures arising between tests. Increased precision in sensitivity and specificity estimates aids the evaluation of clinical and cost-effectiveness of diagnostic tests, enabling more appropriate decisions regarding the most efficient use of health resources. This novel methodological development is applicable to a broad range of disease areas.