Abstract
Purpose: :
Diagnostic test evaluation is difficult for glaucoma because of the absence of a completely accurate gold standard. However, it is important for clinical and public health practices to have the best possible estimates of test performance parameters. We aim to study the validity of using a Bayesian approach to estimating the performance of glaucoma diagnostic tests in the absence of a gold standard.
Methods: :
This cross-sectional, observational study included 600 subjects at high risk for glaucoma who underwent testing between January 2007 and October 2009 at a glaucoma clinic in Montreal, Quebec, Canada. Subjects underwent Matrix 24-2 frequency doubling perimetry (FDT), GDx variable corneal compensation (GDx-VCC) scanning laser polarimetry and confocal scanning laser ophthalmoscopy with the Heidelberg retina tomograph (HRT). The FDT was considered abnormal when there were 9 or more total deviation points less than 5%. A GDx-VCC nerve fiber index of 35 or greater was considered abnormal. The Moorfield’s regression analysis was used to define abnormality on HRT. We analyzed the data using a previously published Bayesian method whereby the marginal posterior densities of sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV) were estimated using the Gibbs sampler. Estimates were compared to those found in the literature using a gold standard.
Results: :
The data from 500 subjects were analyzed. The estimates for the median Se-Sp combinations for the Matrix FDT, GDx-VCC and HRT were respectively: 0.67-0.80, 0.66-0.98, 0.77-0.57. The range of PPV was between 0.31 to 0.80. The NPVs ranged between 0.95 to 0.96.
Conclusions: :
Using a Bayesian approach and no gold standard for glaucoma diagnosis, the HRTIII had the highest sensitivity, but the lowest specificity compared to Matrix FDT and GDx-VCC. The negative predictive values for all tests were high. Our estimates are similar to those found in the literature when similar test cutoffs and criteria are used. Bayesian approaches may be useful in estimating the parameters of diagnostic tests in glaucoma in the absence of a gold standard.
Keywords: clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • imaging/image analysis: clinical • perimetry