Staircase scoring procedures for automated perimetry

Abstract

Conventional psychophysical studies have used a variety of analysis methods for determining thresholds from staircase procedures. In general, these methods demonstrate very good agreement in their threshold estimates when a large number of stimulus presentations (50–200) are employed. However, staircases in automated perimetry typically employ between 4 and 12 stimulus presentations. The present study compared five different analysis procedures (average of all presentations, median of all presentations, mode of all presentations, average of reversals at the minimum step size, and the value of the last seen target) for staircases used in automated static perimetry. Test-retest data for 14 normal eyes and 14 eyes with moderate visual field loss were obtained using a 6–3–3 staircase procedure on the Digilab 750 automated perimeter. Threshold estimates were determined for the five analysis procedures for 66 target locations in each visual field (blind spot region excluded). All of the analysis procedures exhibited good agreement in threshold estimates, except for the ‘last seen’ procedure, which was approximately 1.5 dB different from all other threshold estimates. The average of all stimulus presentations and the average of reversals at the minimum step size both displayed the best test-retest reliability. This was a small but consistent effect, suggesting that these staircase analysis procedures should be employed to improve the test-retest reliability of automated perimetry tests.