Purchase this article with an account.
J. Bandi, D. L. Budenz, W. Feuer, D. R. Anderson, R. Feldman, L. Herndon, D. Rhee, A. Tanna, J. Whiteside-DeVos; Agreement of GPA Progression Program With Gold Standard of Glaucoma Progression. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1095.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To determine whether the GPA progression program agrees with gold standard determinations of glaucoma progression made by glaucoma experts.
Five serial Humphrey 24-2 visual fields from 100 eyes were collected during clinical monitoring of patients with glaucoma. Determinations of progression were made with both the Humphrey GPA progression analysis software and independently by five glaucoma experts. The glaucoma experts rated each sequence of fields with respect to progression as none, questionable, probable, and definite. These were subsequently dichotomized as progressed (rated probable or definite) or not. One month later the glaucoma experts rated the field sets again after they were assigned new subject identifiers and shuffled. For the first and second readings separately, the glaucoma-expert "gold standard" for each eye was determined by the majority dichotomized designation of the five readers. The gold standard from the first readings was compared to GPA and to the second readings.
The first expert gold standard determination identified 51 eyes which progressed and 49 which did not. The second independent ratings agreed with the first 89% of the time, and the kappa statistic was 0.78 (SE=0.05). (Kappa ranges of agreement ≤ 0.4 are considered poor; >0.4 to <0.75, fair to good; and ≥0.75, excellent.) GPA determinations made from these fields classified 30 eyes as likely progression, 17 eyes as possible progression, and 53 eyes as non-progressed. 78% of gold standard and GPA determinations agreed, kappa=0.56 (SE=0.08) when the GPA analysis possible progression was combined with likely progression. If possible progression was combined with no progression, only 69% of gold standard and GPA determinations agreed, with kappa=0.39 (SE=0.08).Although the repeat majority gold standard expert determinations agreed well (kappa=0.78), individual intra-rater agreement was only fair to good (kappa values ranged from 0.58 to 0.71) and inter-rater kappa values within a single set of gradings were only fair (kappa=0.45).
Only if the GPA analysis possible progression was combined with the likely progression, the kappa statistic suggests that the GPA software was equivalent to the majority determination of 5 readers.
This PDF is available to Subscribers Only