Background: Meta-analysis is used to summarise data from several clinical trials by taking an average of the results from those trials and is often used in systematic reviews of medical literature. However, the methods must be adapted for diagnostic accuracy studies and take account of threshold effects. Threshold effects occur when diagnostic studies are using different cut-offs to decide whether a result is positive or negative, and mean that studies with very similar data can have very different estimates of diagnostic accuracy. Existing methods of meta-analysis for diagnostic studies either produce estimates of accuracy that are difficult to apply clinically, require specialist software, or are recently developed and hence not widely tested.

Objectives: To explore in depth the properties of these methods using individual patient data from existing diagnostic accuracy studies of carotid stenosis.

Methods: Data from nine diagnostic accuracy studies (totalling 1982 patients) will be used to assess the performance of these methods and the implications for their use with aggregate data from published diagnostic accuracy studies. The raw data will allow exploration of the effect of variation in diagnostic threshold and patient population.

Results: Preliminary analyses into the statistical properties of the methods suggest that they can be consistent but can also diverge when the data from the individual studies are atypical, for example, when they have low sensitivities. This will reflect on the overall accuracy of the models and the circumstances in which they should be used.

Conclusions: There is scope to develop the statistical models used for diagnostic meta-analysis to make them more accurate or usable as there is a need to create methods that could be readily used by people other than medical statisticians.