Abstract
Purpose :
Find the probability of detecting a small decrease in sensitivity from an already low sensitivity at one location in the visual field using computer simulation of practical and ideal perimetric procedures for conventional automated perimetry (CAP: white-on-white, Size III targets).
Methods :
Using the Open Perimetry Interface, we simulated the measurement of a true sensitivity of θ and θ-2 where θ can vary from 0 to 35dB. The procedures that performed the measurement included the Full Threshold algorithm (which has similar error properties as SITA approach) with various starting values, and the ZEST algorithm with a variety of priors. We also included a new algorithm that combines ZEST on a reduced domain (12dB to 40dB) with a check for an absolute scotoma. Success was recorded when the measured sensitivity of θ-2 was less than the measured sensitivity for θ. Each pair of measurements was repeated 1000 times, estimating the proportion of tests in which such a decrease in sensitivity would be noticed in practice.
We also used an ideal procedure that tried all possible stimuli values and chose the best single value at which to test (generally θ-1, where the difference between frequency-of-seeing curves is maximised), using 5 presentations at this dB value and counting the number of 'seen' responses. If the number of 'seen' responses for true sensitivity θ was greater than that for θ-2, then this was counted as a success.
Results :
For θ < 30dB, the Full Threshold variants measured θ-2 as less than θ 60% of the time. For the ZEST variants, the value ranged from 60% when θ was near 1dB to 70% when θ was 30dB. For the ideal procedure, the proportion was about 55% for 1 < θ < 21 dB. Our new procedure performed similarly to established ZEST variants, but used less presentations, particularly when θ < 17dB.
Conclusions :
Once CAP sensitivities drop below about 20dB the chance of detecting a 2dB decrease in sensitivity using short perimetric procedures is only slightly above chance level. This analysis confirms the utility of the clinical practice of requiring confirmation visual field measurements before reliably detecting change, particularly in moderate and advanced glaucoma. It also provides further evidence that stimuli presentations used to test visual field locations that have dropped below 20dB might be better spent at locations that are still above this level.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.