Our study shows good diagnostic performances of SD-OCT to discriminate glaucoma and control subjects in a general population of elderly subjects. We observed best AUC for average peripapillary, temporal-inferior, and temporal-superior RNFL thickness parameters, with values comparable with most previously published case-control studies. Quite a few case-control studies have already reported good discriminating performances of SD-OCT and have thus validated its usefulness in glaucoma diagnosis.
16–29 Despite differences in study methodology design, SD-OCT machines evaluated, population sample selected or severity of glaucoma disease in each glaucoma group, all these studies have validated high diagnostic performances of OCT imaging technology with best values of AUC ranging from 0.786 to 0.978. Furthermore, the average peripapillary RNFL thickness was the most frequently described discriminating parameter followed by temporal parameters. Results of our population-based study are in accordance with all these previously published studies and confirm the diagnostic performance of SD-OCT to discriminate between glaucoma and control subjects in a larger and representative population sample. Although case-control studies are necessary for an initial evaluation of a diagnostic test by providing very useful information on its accuracy, it has also been demonstrated that this methodology design is dependent on severity of cases recruited and of the representativeness of controls and thus might induce up to a 3-fold overestimation of the test power as compared with studies using a clinical cohort of population.
34 Furthermore, Michelessi et al.
33 reported in a systematic review of studies evaluating imaging technology for diagnosing glaucoma, that most of them had a high risk of bias mainly related to patient selection. Actually, spectrum bias, the most frequent cause of diagnostic accuracy overestimation, may be generated by an inappropriate selection of cases and controls and by a higher prevalence of the disease in a referral center where more severe cases are recruited.
44–47 Additionally, an unequal recruitment of case and control patients in the studied population may also improve the statistical power of the diagnostic test when there is more than one control patient per case patient selected. Our study was performed using an unselected population of elderly people randomly recruited from electoral rolls and individually contacted for enrollment in the study cohort with a longitudinal follow-up visit every 2 years.
39 Hence, the study design we used improved representativeness of our population sample of elderly people and allowed a more precise estimation of glaucoma prevalence in this age group. Thus, by limiting the risk of spectrum bias, we speculate our results may be more applicable in a real-life situation for glaucoma screening purposes in elderly populations. Some other population-based studies have reported diagnostic accuracy of OCT imaging technology.
30–32 By using a time-domain OCT, Li et al.
32 demonstrated moderate sensitivity but high specificity values of the machine to diagnose glaucoma in high-risk populations defined by either a family history of glaucoma or self-described from Caribbean and African origin; and Bengtsson et al.
31 found similar results in a younger cohort of glaucoma patients. More interestingly, Springelkamp et al.
30 observed high discriminating performances of SD-OCT by analyzing regional layer thicknesses of the peripapillary and macula area in a general population with AUC of 0.77 and 0.85, respectively. In our population sample, we observed higher AUC values of peripapillary RNFL thickness parameters, which was more likely related to a difference in the selection of the eye chosen for analysis of each participant. Hence, while Springelkamp et al.
30 analyzed a random eye of each participant, we selected the most affected eye of each participant defined by the lowest average RNFL thickness between the two eyes, if eligible, for analysis. Thus, we assume our methodology may be more appropriate for real-life screening purposes and may also have provided greater statistical power to the study. Indeed, a random selection strategy of only one eye for each participant is more likely to miss glaucoma patient. As glaucoma is a frequent and irreversible cause of blindness or visual impairment, we also speculate this screening strategy could improve sensitivity of SD-OCT to detect glaucoma and could better prevent visual disability in a general population of elderly people with a higher prevalence of the disease than population of lower age.