While the performance levels of BMO-MRW and RNFL in discriminating between early and moderate/advanced glaucoma and healthy disc-size-matched patients were comparable based on an ROC analysis of eyes with small (BMO-A < 1.95 mm2) and large optic discs (BMO-A > 1.95 mm2), the classification report using the 5th percentile as a cut-off RNFL value showed higher sensitivity in eyes with a large optic disc size as compared to that using BMO-MRW. Because an assessment performed by means of percentiles, as opposed to an ROC analysis, is used in clinical practice, the results of the study are highly relevant to everyday clinical practice. In the following, the aforementioned results are discussed.

In eyes with a small BMO-As (< 1.95 mm2), we found that BMO-MRW and RNFL thickness had comparable AUCs for discriminating between 50 eyes with early (0.83 and 0.84) glaucoma and 50 eyes with moderate/advanced glaucoma (0.98 and 0.95) and a healthy optic-disc-size-matched control group. This result was also confirmed in a subgroup analysis of eyes with a disc size < 1.69 mm2. These results are in line with the literature:

For microdiscs, Enders et al. found that RNFL and BMO-MRW had similar performance in discriminating 51 glaucomatous patients from 20 healthy controls with small optic discs, with AUCs of 0.81 and 0.87, respectively. Also, similar to our results, they reported comparable sensitivities (RNFL = 66.4%; BMO-MRW = 68.6%) at a fixed specificity of 95% [13]. However, optic disc size was determined via HRT (< 1.63 mm2), making a comparison with the present study difficult.

Similar to our work, Gmeiner et al. analyzed eyes with a small BMO-A (< 1.84 mm2) and found a comparable AUCs for BMO-MRW and RNFLT in discriminating between 23 eyes with pre-perimetric (0.904 and 0.857, respectively) and 23 eyes with perimetric glaucoma (1.00 and 0.987, respectively). The sensitivities at a fixed specificity of 95% were even higher for BMO-MRW as compared to RNFLT in both pre-perimetric (69.6% versus 47.8%) and perimetric glaucoma (100% versus 87.0%) [12]. Thus, based on the ROC analysis, there is comparable discriminative performance between glaucoma patients and healthy controls using BMO-MRW and RNFL.

In eyes with a large BMO-A (> 1.95 mm2), in our work, ROC curves demonstrated that RNFL tended to perform similar or slightly better than BMO-MRW. The subgroup analysis of optic discs with a BMO-A above 2.35 mm2 showed a similar result.

In the literature, contradictory results can be found regarding the performance of RNFL and BMO in detecting glaucoma in eyes with large optic discs. Similar to our work, Gmeiner et al. stated that the discriminative performance levels of BMO-MRW and RNFL in eyes with a BMO-Area > 1.84 mm2 were comparable in 27 eyes with pre-perimetric (0.764 versus 0.817, respectively) and 19 eyes with perimetric glaucoma (0.892 versus 0.934, respectively). The sensitivities at a fixed specificity were even higher for RNFL in both pre-perimetric (49.7% versus 25.9%) and perimetric glaucoma (78.9% versus 57.9%) [12], which can also be confirmed in our work, as seen in Table 5.

In contrast to that, Enders et al. concluded that the performance of BMO-MRW in discriminating 44 eyes with glaucoma from 70 healthy controls was higher (AUC = 0.96) than that of RNFL (AUC = 0.89) in eyes with BMO-A > 2.45 mm2 [17].

These contradictory results may be explained by the different age distributions of the study populations. The age of the control group used by Enders et al. was lower (38.2 ± 25 years) than the age of the glaucoma group (65.0 ± 11.0 years) [14]. In contrast, Gmeiner et al. found no difference between healthy controls and patients with glaucoma [12]. Because the age-related decrease in MRW is higher (1.77 µm/year) than that of RNFL (0.23 µm/year), this may have led to easier discrimination between controls and glaucoma patients in Enders et al.. Similarly, in our work, after adjusting the parameter based on age, RNFL showed better performance than MRW.

In clinical practice, the classification of healthy and diseased participants is based on using the manufacturer’s 5th percentile as a threshold. We found that, in eyes with small optic discs, RNFL and BMO-MRW had comparable results, whereas in eyes with large optic discs, RNFL showed higher sensitivity for both early glaucoma (46% versus 42%) and moderate/advanced glaucoma (84% versus 68%, respectively). This effect could also be shown amplified in the subgroup analysis of micro- and macrodiscs. This stands in contrast to the results of the ROC analysis in the above mentioned literature. Zheng et al. analyzed the performance of RNFL and BMO-MRW independent of disc size and also found that using the 5th percentile to classify 188 eyes from 137 patients with glaucoma showed a higher sensitivity of 88.83% for RNFL as compared to a sensitivity of 76.06% for BMO-MRW [7]. To the best of our knowledge no other comparable studies can be found in the literature.

However, an exact comparison between ROC analysis and classification based on percentiles is difficult because they are fundamentally different. The performance of the ROC analysis depends on the age and optic disc size distribution in the control group and the severity of the disease in the glaucoma group. In contrast, both age and optic disc size are included in the calculation of the 5th percentile. This has the advantage being more independent from the study population analyzed.

The poorer performance of BMO-MRW with large or very large ONHs may be due to various factors. In large optic discs, the BMO-MRW is thinner overall, as the axons entering the optic nerve head are distributed over a wider rim area [13, 14]. Therefore, large but healthy discs may be erroneously classified as glaucomatous. This, however, leads to a decrease in specificity, not in sensitivity, and only explains why many patients with large optic discs are classified as diseased in clinical practice [18], even though they are healthy, whereas a decrease in sensitivity is not explained.

As the sensitivity of BMO-MRW is lower with large optic discs, apparently, BMO-MRW falls below the 5th percentile later in the course of glaucoma than RNFL; thus, more patients with glaucoma are classified as healthy when using BMO-MRW as compared to RNFL. This may be explained by the reactivity of the tissue to the progression of glaucoma. Amini et al. hypothesized that the remodeling of neural tissues could occur more slowly in large optic discs as glaucoma advances [19]. Similar, Vianna et al. found that higher baseline values of BMO-MRW, because they were present in small optic discs, were related to faster reductions [20]. Thus, in large optic discs, slower reduction would be expected, which could lead to a subsequent drop below the 5th percentile. Furthermore, larger discs could contain a higher proportion of non-neuronal tissue, as the axons are distributed over a wider rim area [19]. A decrease in neural tissue could be distorted by a larger proportion of non-neuronal tissue, leading to falsely high values. Similar, Mardin et al. found that the accuracy of BMO-MRW could be worsened by retinal blood vessels, in contrast to that of RNFL measurements [21]. This could, however, explain why RNFL shows a higher sensitivity than BMO-MRW in diagnosing glaucoma in eyes with large optic discs.

Interestingly, when analyzing microdiscs in early glaucoma, sensitivities of RNFL were superior (60%) to that of BMO-MRW (44%). We hypothesize that there is more neuronal tissue in small optic nerves and that the disease has not yet progressed to the point of falling below the 5th percentile.

Our work has several limitations. First, due to the characteristics of the study population and the use of CSLT or BMO area for the calculation of optic disc size, it is difficult to compare our work with studies in the literature. Ideally, there would be an even distribution of glaucoma, as well as an age- and optic-disc-size-matched control group. Therefore, in our study, a BMO-A matched control group was used, which is a strength of our study. However, there was a significant age difference between the healthy and glaucoma populations, which clearly influenced the results of our work. In a second ROC analysis, we adjusted the parameters based on age-related changes [16]. Because the adjustment of age is only a theoretical calculation and the effect of age on the parameter may also differ between small and large optic nerves, this limited the results to a certain extent. However, no contradictory results were found after adjusting the parameters based on age.

Second, in using the 5th percentile to analyze the performance of the parameters to detect glaucoma, a mode of analysis was chosen that was more independent of the study population, compared to ROC analysis. This is an advantage of the study. However, the cut-off values for the 5th percentile were calculated approximately, by means of a regression. Because the manufacturer’s exact formula is not known, this is a limitation of the study, as it could have influenced the results.

Third, our classification of the severity of glaucoma is based on the mean deviation as proposed by Mills et al. [15]. However, we only classified into early (MD: < -5.0 dB) and moderate or advanced glaucoma (MD: > -5.0 dB). The results of the study in advanced glaucoma should therefore be interpreted with caution.

Fourth, both the intraocular pressure and axial length were not analysed in our study using OCT data. As this significantly influences the RNFL and BMO-MRW measurements [22], this might have influenced the results to a certain extent. In particular, the influence of high myopia or hyperopia on the accuracy of RNFL and BMO-MRW classification based on the 5th percentile needs further investigation. The results of this study should be applied with caution to highly myopic and highly hyperopic eyes.

Fifth, one major limitation is the retrospective design. Further prospective studies are needed. Finally, our work addresses the topic of disc dependency in glaucoma detection using BMO-MRW and RNFL, which has been addressed using BMO-MRA in eyes with small and large optic disc sizes [11]. However, this parameter is not widely used in everyday clinical practice. A promising approach is the use of ganglion cell layer thickness [18], but threshold values have not yet been established for both BMO-MRA and ganglion cell layer thickness.

In conclusion, based on an ROC analysis, the discriminative performance levels of BMO-MRW and RNFL between patients with early and moderate/advanced glaucoma and a healthy control group matched based on optic disc size were comparable in eyes with BMO-As smaller and larger 1.95 mm2. Using a classification based on the 5th percentile, as is used in clinical practice, RNFL was shown to be superior to BMO-MRW in terms of sensitivity in glaucoma diagnosis with large optic discs. Our findings underscore the importance of RNFL imaging and measurement in the diagnostic evaluation of glaucoma, especially in eyes with large optic discs and a BMO-A above 2.35 mm2.