Properties of staircase procedures for estimating thresholds in automated perimetry.

The properties of the staircase procedure as applied in automated perimetry were examined. Two computer simulation models were used to vary different test- and patient-related parameters in clinical perimetry. One model was based on the KRAKEN computer simulation program; the other computer simulation was based on stimulus-response data sets from 11 normal subjects. The results were analyzed in terms of efficiency and accuracy. It was found that: (1) in general, there was an efficiency-accuracy trade-off; (2) increases in response fluctuation produced substantially greater errors in threshold estimates; (3) little or no improvements in accuracy were achieved by increasing the number of reversals; (4) the starting position of the staircase relative to the threshold influenced the efficiency of threshold determinations but not their accuracy; (5) a single-response error reduced the efficiency of staircases; (6) the position of a single-response error in a staircase sequence influenced the accuracy and efficiency of the threshold determination; and (7) more than one response error during a staircase sequence always resulted in a marked reduction in accuracy and/or efficiency. Current perimetric strategies appear to be at or near optimal levels, and therefore, strategies in the future may need to depart from a staircase-style procedure to achieve a significant increase in both accuracy and efficiency. Computer simulation studies can provide an effective means of evaluating perimetric test procedures and defining optimum strategies, which then can be verified clinically by subsequent testing in patient populations.