Combining perimetric suprathreshold and threshold procedures to reduce measurement variability in areas of visual field loss.

PURPOSE

Several current perimetric test strategies use information from neighboring test locations to seed the procedure. When this initial estimate is close to the true threshold, the algorithms terminate quickly with low error. However, when neighboring thresholds are dissimilar, seeding procedures in this manner results in poor accuracy and repeatability. This study aims to develop a test procedure that avoids assumptions based on neighboring locations yet terminates with an acceptable number of presentations. We explore the utility of a combined screening and threshold approach.

METHODS

Our approach [estimation minimizing uncertainty (EMU)] first applies suprathreshold screening and then in those locations that fail the screening, a ZEST procedure commencing with a uniform probability density function (pdf). EMU was compared with full threshold (FT) using computer simulation. Input to the simulation was empirical standard automated perimetry data (FT) from 265 normal subjects and 163 observers with glaucomatous visual field loss. Test performance was assessed using four patient error models designed to represent patients who respond with no errors, typical false-positive errors, typical false-negative errors, and extremely unreliable patients.

RESULTS

When patients made typical false-positive errors, EMU required about 20% fewer presentations than FT averaged across the visual field. EMU required a greater number of presentations than FT when false-negative errors were made. The percentage of locations misclassified as either normal (defined as the thresholds greater than the lower 95% confidence limit of age-corrected norms) or abnormal by EMU was lower than for FT. Point-wise analysis demonstrated that in the presence of false-positive errors, the average error and SD of error of thresholds returned by EMU were reduced compared with FT.

CONCLUSIONS

EMU enabled accurate estimates of threshold for situations in which neighboring locations are a poor predictor of true threshold. Combining suprathreshold and threshold strategies enables an acceptable number of total visual field presentations, while improving the accuracy and repeatability of threshold estimates in regions of abnormal sensitivity.