Purchase this article with an account.
Christopher K. Leung, Marco Yu, Robert N. Weinreb, Dennis S. Lam; Modeling Retinal Nerve Fiber Layer Thickness (RNFLT) Progression: A Comparison between Trend (TA) and Event Analyses (EA). Invest. Ophthalmol. Vis. Sci. 2011;52(14):2118.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
TA and EA are two fundamental methods to study glaucoma progression. However, TA may not agree with EA and it is unclear which type of analysis is more reliable to detect change. With mathematical modeling simulating different patterns of RNFL progression, we compared the performance between TA and EA.
RNFLT was measured weekly for 8 weeks with an FD-OCT from 46 subjects (19 had glaucoma) to compute intervisit within-subject standard deviations (SD). 15,000 sets of simulation data, each with 16 data points simulating 4 monthly serial measurements collected from an eye over 5 years, were then generated with reference to the normal distribution of an individual's SD and the pattern (linear, exponential or stepwise decrease) of RNFLT change (modeled at -2µm or -4µm/year) imposed on that particular dataset. Progression was defined as a significant negative slope in the linear regression between RNFLT and time (TA), or a decrease in RNFLT > reproducibility coefficient (2.77 x within-subject SD) for 2 serial visits (EA). Results were validated with a longitudinal dataset obtained from 107 glaucoma patients followed for 2 years with RNFLT measured 4 monthly.
Fig. 1 shows the sensitivity and specificity computed from the simulation dataset modeled at a linear rate of -2µm/yr. For individuals with a small (A) or moderate (B) intervisit SD, TA is more sensitive than EA in detecting progression (p<0.001) at specificity ≥95%. For those with a large intervisit SD (C), EA is more sensitive than TA in the initial period but at the expense of a lower specificity (85% vs. 95%, P<0.001). These findings were independent of the pattern, and the rate of RNFL progression. Analysis on the longitudinal dataset revealed that for patients with a within-subject SD<1.7µm, 18% and 12% progressed by TA and EA at 2 years, respectively. For those with a within-subject SD≥1.7 µm, 11% and 12% progressed, respectively.
Individual variability has a significant impact on progression analysis. For most individuals, TA is more reliable than EA to detect progression.
This PDF is available to Subscribers Only