advertisement
Editors Selection IGR 14-1
Progression: Combination of CSLO and SAP data
Comment by Gustavo de Moraes on:
49013 Combining Structural and Functional Measurements to Improve Estimates of Rates of Glaucomatous Progression, Medeiros FA; Zangwill LM; Girkin CA et al., American Journal of Ophthalmology, 2012; 153: 1197-1205
Almost ten years ago, the first consensus meeting of the World Glaucoma Association on 'Glaucoma Diagnosis' recommended that patients with or suspected glaucoma should be monitored with a combination of structural and functional tests. There has been a rapid improvement in the resolution and repeatability of structural technologies since the advent of confocal scanning laser ophthalmoscopy (CSLO), a pioneering technology to provide in-vivo, objective measurements of the optic nerve head topography. The detection of optic disc and visual field changes simultaneously increases the likelihood that true glaucomatous progression has occurred, as opposed to measurement variability. However, these simultaneous structural and functional changes are not always seen and cases of disagreement pose as a major challenge for clinicians when tailoring glaucoma therapy. To overcome this challenge, Medeiros et al. tested the performance of a new method to detect glaucomatous progression which integrates longitudinal information from CSLO and visual field using Bayesian statistics.
In brief, this statistical method is used to determine the probability of an observation (e.g.: structural or functional progression), given that another observation has occurred. For the present study, a priori information from CSLO (rim area change) was used to improve the performance of a visual field testing (based on change in mean deviation, MD) to estimate rates of progression, and vice versa. The authors found that the Bayesian model was able to explain the variance in MD or rim area change better than the ordinary least squares (OLS) model, which employs data from each technology individually. Moreover, OLS method calculated rates of change underestimated the amount of future visual field damage (MD) by more than two dB in 30% of the visual fields compared to 13% when the Bayesian model was employed. Also, overestimation of future visual field loss by more than two dB occurred in 18% of visual fields with OLS while with the Bayesian model it occurred in 12% of cases. Although the two statistical methods correlated significantly and the error associated with each model decreased with more frequent testing, Bayesian predictions still outperformed OLS predictions for eyes with shorter and longer follow-up periods.
This approach could overcome the lack of consensus on which technology is the most accurate to detect glaucomatous progression and may potentially minimize the need of repeated testing - a serious burden for patients and clinicians
This advantage is essential since the algorithms each technology uses to define progression individually may often depict changes that do not reach statistical significance ‐ but which may correspond to subtle and true glaucomatous changes. This approach could overcome the lack of consensus on which technology is the most accurate to detect glaucomatous progression and may potentially minimize the need of repeated testing ‐ a serious burden for patients and clinicians.
Medeiros et al. have long been working on the development of more accurate methods to diagnose glaucoma and monitor progression and the present paper is an important step towards establishing consensual definitions in a time when the lack of a gold standard has critically limited progress in the field.
