Network Meta-Analysis Modeling for Diagnostic Accuracy Studies
Wei Cheng, Brown University
Keywords: network meta-analysis, diagnostic tests, Bayesian inference, technology assessment
Constructing a network of diagnostic test accuracy studies to compare multiple tests is more complex than doing so for studies of treatment efficacy. Synthesizing diagnostic accuracy studies may focus on summarizing the diagnostic performance of each test, rather than the pairwise contrast, and include information from eligible subjects with single-, paired-, and triplet-test studies for each test. Also, the TPF (true positive fraction, which equals sensitivity) and FPF (false positive fraction, which equals 1-specificity) of test(s) in a diagnostic accuracy study are correlated. We propose a joint modeling framework for networks of diagnostic accuracy studies with mixed study-types (single-, paired-, and triplet-test studies). The model assumes that true and false positive counts follow binomial distributions independently among diseased and non-diseased individuals. The underlying true and false positive fractions for each test are decomposed on the logit scale into components that represent their overall average across study-types for each test, study-type--specific effects to reflect inconsistency, and within-study--type random effects.