Predicting Variability and Reliability in Visual Field Testing: Short- and Long-Term Approaches.

Purpose: To predict intrinsic measurement variability and reliability (any cause of data loss) in visual field (VF) results using a computer simulation model. Design: Computer simulation study. Subjects: One hundred thousand subjects simulated with empirical mean deviation, progression rate, variability, and reliability characteristics. Methods: One hundred thousand subjects were simulated to undergo 4 VF tests per visit, 3 monthly, over 20 years (long-term condition) and 4 VF tests per visit daily over 28 days (short-term condition). Permutations of 1-4 tests per visit over 3-, 6-, and 12-monthly (long-term) and 1-, 2-, and 4-daily (short-term) review intervals were used. Visual field variabilities were estimated sequentially until 3 consecutive visits returned variabilities within 5% of each other using a rolling window. The same was applied to reliability. The last visit of the window denoted the critical time to estimating variability using the consecutive clinical criterion (TcV) and critical time to estimating reliability using the consecutive clinical criterion (TcR) estimation. Additionally, we identified the critical time at which 3 consecutive visits were within 5% of the ground truth (critical time to estimating variability using the consecutive clinical criterion and comparison with the ground truth [TgV] and critical time to estimating reliability using the consecutive clinical criterion and comparison with the ground truth [TgR]). Main Measures: Critical time to estimating intrinsic variability and reliability. Results: The most intensive long-term approach (4 tests/visit, 3 monthly) required a median of 6 years to reach TcV. In the long-term, most subjects arrived at TcR within 2 years, but short-term testing (even with 1 test per visit) required only 5 days of daily testing. More tests per visit and more frequent reviews shortened the critical time. Average differences between the estimated variability and reliability at TcV and TcR and their ground truth results were clinically small (within 1 decibel and 10%, respectively). Mean deviation, progression rate, and variability were significant predictors of TcV and TgV for long-term follow-up, with no clinically significant predictors for short-term variability (R Conclusions: Longitudinal estimates of variability are not likely achievable in clinical practice, but short-term intensive VF testing unaffected by progression can return variability and reliability rates within reasonable timeframes. We provide a framework for the effect of variability for the likelihood of detecting differences in VF results over time, given reliability rates. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.