Abstract
Purpose :
To investigate the effect of criterion-bias on sensitivities measured using standard automated perimetry (SAP).
Methods :
One eye of 12 glaucoma patients (7 female; 8 OD; age: 47-77 years) were tested at 3 locations in the visual field (normal and damaged locations) where 0dB could be seen in at least 15 of 30 trials. Frequency of seeing (FoS) curves were measured on an Octopus 900 using a forced choice method (2IFC; observer responds after 2 intervals, one which contains the stimulus) and a yes-no method (YN; respond when a stimulus is detected). Sensitivity was taken as the 50% probability of seeing for the YN task and 75% probability for 2IFC. The maximum probability of seeing (max-prob) was given by the upper asymptote position of the FoS curve. Curves were fit as Cumulative Gaussians with 3 free parameters using maximum likelihood fitting, with the lower asymptote fixed at 0.5 for 2IFC, and the participant's false positive rate as measured by catch trials (20% of trials) for the YN task.
Results :
Of the 36 locations tested, 10 had equal max-prob of 1.0 for both 2IFC and YN tasks, 18 had higher max-prob for the 2IFC, and 8 for the YN task. 2 locations had a max-prob less than 75% for the 2IFC task and similarly 2 locations had less than 50% for the YN task (1 location in common). Of the remaining 33 locations, 21 had a higher 2IFC sensitivity than YN sensitivity, 5 were equal, and 7 were higher with the YN task. The slope of the FoS curve measured by the YN task generally agreed with previous literature, but showed wide variation as a function of sensitivity.
There was no correlation between the difference in sensitivity between 2IFC and YN and: eccentricity (Spearman p=0.74); baseline sensitivity measured as average of two HFA fields (p=0.33); sensitivity as measured by the 2IFC (p=0.79). Similarly, it did not seem possible to predict the difference in sensitivity per person (Kruskal-Wallis p=0.15).
Conclusions :
Generally, sensitivities and upper asymptotes of FoS curves measured with a yes-no paradigm in perimetry are under estimates of those collected when criterion-bias is removed using the 2IFC paradigm. Researchers should be cautious, therefore, in using perimetric data to model underlying vision processes free from this form of cognitive bias.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.