Background:

Standard diagnostic accuracy studies present results as a 2 x 2 table of test performance. Methods for the analysis, and meta-analysis, of such data are well developed and are commonly used. There has been little work on how to analyse non-standard diagnostic accuracy studies.

Objective:

To present an example of a non-standard diagnostic systematic review and to discuss the methods used to analyse the results.

Methods:

We conducted a systematic review on the localisation of the epileptic focus in patients with refractory epilepsy considered for surgical treatment. This review is a diagnostic review in the sense that the studies included in the review compare the index test to a reference standard (generally the site of surgery). However, the studies differ from standard diagnostic accuracy studies as rather than simply attempting to identify whether a disease is present or absent, these imaging tests try to identify the site of the epileptic focus.

These studies do not report simple 2 x 2 data (i.e. numbers of true positive, false positive, true negative and false negative results), instead data are provided in a number of different categories. Within patients in whom the reference standard localised the seizure focus (who would be considered positive in a standard evaluation) the index test has the following possibilities: correctly identified the seizure focus, failed to identify a seizure focus, wrongly identified a seizure focus or partially identified the seizure focus. In patients in whom the reference standard failed to localise the seizure focus (considered negative in a standard evaluation) the index test can either localise a seizure focus, or not localise a seizure focus. In this example, there is also a third possibility for the reference standard: it could identify the patients as having a multi-focal seizure focus. Within this group of patients the index test can: correctly identify the patients as having a multi-focal seizure origin, fail to identify a seizure focus, or identify patients as having a single seizure focus.

Results:

The methods used to analyse the results will be discussed, focusing on how these differ from standard methods of meta-analysis for diagnostic data. In particular we will discuss why the analysis should not try to force the data into a 2 x 2 table so that standard methods of analysis may be used. The relative proportions in each of the outcome categories will be summarised. Such an analysis will make this type of data more suitable for inclusion in a decision analytic model than forcing the data into a 2 x 2 table.