The identification of bias in studies of the diagnostic performance of imaging modalities.

The demand for evidence-based healthcare is increasing nationally and internationally and it is equally necessary in both diagnostic and therapeutic practice. Evidence may be collected and combined by means of a systematic literature review of published and unpublished data on a well-defined topic. The output of such reviews is then available to guide health policy, influence good practice or direct research. Published guidelines are available on the performance of systematic reviews, especially those of randomized controlled trials. Although there is an extensive literature base of research data in diagnostic imaging there are few such trials, but it is still possible to perform systematic reviews. With the alternative study designs encountered it is important to be aware of the main threats to study validity. In this paper the biases likely to be encountered in studies of diagnostic performance are reviewed, with particular reference to diagnostic imaging tests. The biases are sub-divided into three categories. The first category is patient selection and covers the validity of generalizing results beyond the study population. The other two, concerning study design and execution and the interpretation of results, affect the likely validity of the results of a study. An understanding of these factors is an essential prerequisite for those undertaking or using a systematic literature review in the field of diagnostic imaging. The definitions form the foundations of a defensible review protocol.