Abstract
Purpose :
The ETDRS chart (Ferris III, et al., 1982) and its computerized version, e-ETDRS (Beck et al., 2003), remain the standard for testing visual acuity in clinical trials, but testing acuity with precision remains a challenge. Lesmes (2018) introduced a novel Bayesian visual acuity test (Bayesian VA) that provides the advantages of high-density sampling of optotype size (.02 logMAR), adaptive stimulus optimization (Lesmes, et al, 2006), and the post-hoc analyses of Bayesian credible intervals (68.2% half-width credible interval - HWCI). Here, a proof-of-concept psychophysical study evaluates the accuracy and precision of the Bayesian VA method.
Methods :
For six subjects with normal or corrected-to-normal vision, monocular visual acuities were measured for both eyes in four acuity conditions (three with different levels of Bangerter foil to degrade vision and one with no foil). In each session, subjects were tested monocularly with Bayesian VA and e-ETDRS methods 4 times in each eye in one of four foil conditions, with random ordering of the two methods and two eyes in different blocks. For each trial of Bayesian VA testing, three optotypes were presented, with their size selected by an adaptive maximization of information gained about the threshold and range of the full acuity function.
Results :
There was excellent agreement between the estimated thresholds from the novel VA approach and the e-ETDRS standard, with a correlation coefficient of .99 (p<0.001) across all subjects and foil conditions. Across conditions with average acuities ranging from -0.02 to 0.50 logMAR, the average HWCI of the estimated acuity from e-ETDRS was 0.096 ± 0.019 decimal log units. In comparison, the precision of Bayesian VA increased with the number of trials. The average HWCI of estimated acuities after 10, 15, and 45 trials were 0.091 ± 0.028, 0.070 ± 0.022, and 0.037 ± 0.011, respectively. Because of different stimulus sampling resolutions (.02 vs .10 logMAR), we compared the test-retest precision of the two methods using Fractional Rank Precision (FRP, where higher values indicate higher precision; Dorr, et al., 2017): for Bayesian VA, FRP=0.852 ± 0.025; for e-ETDRS: FRP=0.826 ± 0.027.
Conclusions :
After 10 trials (30 letters~40 sec), the precision of Bayesian acuity estimates matched that of the e-ETDRS in 30.8 ± 5.4 letters. With more trials, the precision of Bayesian VA continued to improve over that of e-ETDRS.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.