Visual Field Endpoints Based On Subgroups Of Points May Be Useful In Glaucoma Clinical Trials: A Study With The Humphrey Field Analyzer And Compass Perimeter

Precis: Visual field (VF) endpoints based on average deviation of specific subsets of points rather than all points may offer a more homogeneous data set without necessarily worsening test-retest variability and so may be useful in clinical trials. Purpose: The purpose of this study was to characterize the outcome measures encompassing particular subsets of VF points and compare them as obtained with Humphrey [Humphrey visual field analyser (HVF)] and Compass perimeters. Methods: Thirty patients with imaging-based glaucomatous neuropathy performed a pair of 24-2 tests with each of 2 perimeters. Nonweighted mean deviation (MD) was calculated for the whole field and separate vertical hemifields, and again after censoring of points with low sensitivity (MDc) and subsequently including only "abnormal" points with a total deviation probability of <5% (MDc5%) or <2% (MDc2%). Test-retest variability was assessed using Bland-Altman 95% limits of agreement (95%LoA). Results: For the whole field, using HVF, MD was -7.5 +/- 6.9 dB, MDc -3.6 +/- 2.8 dB, MDc5% -6.4 +/- 1.7 dB, and MDc2% -7.3 +/- 1.5 dB. With Compass the MD was -7.5 +/- 6.6, MDc -2.9 +/- 1.7 dB, MDc5% -6.3 +/- 1.5, and MDC2% -7.9 +/- 1.6. The respective 95%LoA were 5.5, 5.3, 4.6, and 5.6 with HVF, and 4.8, 3.7, 7.1, and 7.1 with Compass. The respective number of eligible points were 52, 42 +/- 12, 20 +/- 11, and 15 +/- 9 with HVF, and 52, 41.2 +/- 12.6, 10 +/- 7, and 7 +/- 5 with Compass. With both machines, SD and 95%LoA increased in hemifields compared with the total field, but this increase was mitigated after censoring. Conclusion: Restricting analysis to particular subsets of points of interest in the VF after censoring points with low sensitivity, as compared with using the familiar total field MD, can provide outcome measures with a broader range of MD, a markedly reduced SD and therefore more homogeneous data set, without necessarily worsening test-retest variability.