Integration of meta-analysis and economic decision modelling for evaluating diagnostic tests

Tags: Poster
Sutton A, Cooper N, Goodacre S, Stevenson M

Background: Meta-analysis of diagnostic test performance data is more difficult than of effectiveness data for reasons including:

- statistical methods have to simultaneously combine both sensitivity and specificity information;

- since test performance is a function of threshold, and establishing optimal thresholds is important for clinical practice;

- tests are rarely used in isolation and assessment of tests in combination will often be necessary to inform practice usefully.

Objectives: To discuss how these issues can all be addressed in a joint synthesis and cost-effectiveness framework. Appraisal of how Cochrane reviews could be conducted to facilitate such an assessment will be presented.

Methods: A fully integrated synthesis and decision model is described for the evaluation of methods of diagnosing deep vein thrombosis. In this, the optimal performance location on each test's summary ROC curve is identified as a function of the willingness to pay per unit improvement in heath. This work is extended to consider the performance of sequences of tests.

Results: The model demonstrates how diagnostic test meta-analyses can be integrated into decision models. Several issues regarding the interpretation of meta-analysis results and the appropriateness of different modelling assumptions are raised. Establishing whether threshold is implicit or an explicit, controllable variable is critical. Interpretation of random effects adds further complexity. No current synthesis methods incorporate threshold values explicitly which is a limitation.

Conclusions: There is some way to go before rigorous evaluations of diagnostic tests will become commonplace. This is due to i) the necessarily complex statistical methods required; ii) the generally poor quality and incomplete reporting of primary studies; and iii) lack of information on other key quantities required, e.g. correlations between performance of different tests. By raising awareness, it is hoped this talk will help improve design, reporting and synthesis of diagnostic test studies in the future.