Abstract
Purpose :
To assess longitudinal variability of central (10-2) visual field (VF) measurements as a function of baseline sensitivity.
Methods :
124 eyes (from 124 patients) with central or moderate to advanced glaucoma, more than 2 years of follow-up and 4 or more 10-2 VFs were included. Threshold sensitivity (TS) measurements ≥35 dB and visits within 0.2-years from the prior visit were excluded. Our newly developed Bayesian longitudinal functional model uses longitudinal TS at 68 locations of 10-2 VF to estimate pointwise rates of change. We used pointwise simple linear regression analysis (SLR) as a comparison. The main outcomes of interest were the slopes (rates of change) and residual SDs (indicating TS variability) at individuals VF locations. We also conducted a simulation study to assess the ability to detect change as a function of residual SDs.
Results :
The average (SD) mean deviation at baseline and rate of change were –8.9 (5.3) dB and –0.31 (SE: 0.06) dB/year, respectively. The mean (SD) global subject intercept (baseline TS measurements averaged across 68 locations) and rate of change (estimated across subjects with SLR) were 24.2 (4.5) dB and –0.38 (0.69) dB/year, respectively. Sampling variability for TS for subject-profiles whose estimated baseline TS was in the range of 5 to 20 was markedly (4 times) higher compared to that around 30 dB either from the Bayesian (SD: 6.7 vs. 1.5 dB) or SLR (SD: 6.5 vs. 1.6 dB) models, respectively. Figure 1 illustrates that locations with baseline TS values around 5-20 dB exhibit very high variance, whereas at both ends of the threshold spectrum, the variance is much smaller. In the simulation study, the percentage of slopes that could be identified as significantly decreasing diminished with increasing residual SD: 91% at SD=2 dB; 62% at SD=4 dB; 13% at SD=7 dB and 1% at SD=10 dB. Either model consistently identified 0% significantly positive slopes across all SD groups.
Conclusions :
Our results confirm cross-sectional findings that once threshold sensitivity at a test location falls below 20 dB, the variability is so high that it is unlikely for change to be detected over time; therefore, such locations are unreliable for detection of longitudinal change.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.