This is a protocol for a Cochrane Review (Diagnostic test accuracy). The objectives are as follows:
To determine the diagnostic accuracy of HRT, OCT and GDx for detecting ONH and RNFL damage in patients suspected of glaucoma.
To determine which morphometric measure or diagnostic algorithm yields the highest diagnostic accuracy within each instrument.
To compare the relative diagnostic accuracy of the three instruments.
To explore potential causes of heterogeneity of diagnostic performance across studies.
We will investigate the following sources of clinical heterogeneity:
A. Heterogeneity related to the choice of reference standard:
type of reference standard (optic disc assessment, visual field, or both);
definitions of visual field damage.
B. Heterogeneity related to characteristics of the study population:
severity of glaucoma.
C. Heterogeneity related to specific methodological aspects of included studies (Appendix 6):
inclusion of a representative spectrum of patients;
reporting of uninterpretable results;
choice of unit of analysis.
Since we expect that a large number of included studies will be case-controls, we will consider a peculiar type of bias resembling incorporation bias for these studies. In fact, when the investigator assessing the presence of glaucoma uses not only reliable and valid perimetric criteria, but also optic disc appearance such as cupping to allocate patients to the glaucoma group, the results will be that diseased patients will have larger cups than expected, thus enhancing the ability of imaging methods to detect disease based on disc morphology algorithms. Therefore, we will investigate heterogeneity between case-control studies using visual field only versus field plus optic disc as a reference standard.
Finally, we will consider an exploratory subgroup analysis based on the overall level of missing data regardless of their cause (including withdrawals and any patients who may have been excluded because of uninterpretable index test results) using the median level of missing data across studies to define better versus worse quality, as well as a level of 10% missing data for the same purpose. We will include studies not mentioning missing data, but also not stating that there were not any missing data, among better or worse studies as a further subgroup analysis.