Andrew Turpin, School of Computer Science and Information Technology, RMIT University, Melbourne, Australia
Allison McKendrick, Department of Optometry and Vision Science, University of Melbourne, Melbourne, Australia
Balwantray Chauhan, Department of Ophthalmology Dalhousie University, Canada
This was our second application of the theoretical derivation of individual confidence limits presented in Vision Research, 2005.
Purpose: This study uses clinical data to examine the effectiveness of a theoretical technique we have previously published for incorporating an individual patient's variability information into estimates of probability of change in visual fields [1].
Methods: The clinical data set consisted of 9 primary open-angle glaucoma patients with fields measured twice a year for 5-12 years (mean 8.3 years) with 30-2 Full Threshold (Humphrey Field Analyzer). These patients also had frequency-of-seeing (FOS) curves measured with the same stimuli using a short MOCS procedure in six locations of their fields at their first visit. A cumulative Gaussian curve was fitted to each FOS, and fits with a correlation coefficient less than 90% were discarded. From the remaining FOS curves, linear regression on log-slope and threshold was used to elicit a slope-threshold relationship similar to that published by Henson et al. [2]. False response rates were determined from the average of the asymptotes of the FOS curves for each patient. Using this variability information for each patient, and specific knowledge of the Full Threshold algorithm, we derived individual probability of change (IPoC) maps similar to GCP maps, and also performed a pointwise linear regression (PLR) weighted by variability information (WPLR) for each patient. We compared these customised methods for determining change with standard techniques [3,4].
Results: The IPoC maps generally agreed with GCP on the classification of patients as progressing or stable, but discovered progression on average 2 visits earlier than GCP. IPoC flagged significantly less points than GCP. WPLR, however, classified all 9 patients as progressing almost immediately after their two baseline measurements, which disagreed with the PLR two-omitting criteria [3] and the GCP methods.
Conclusions: IPoC compared favourably to GCP because both used the same definition of baseline (average of first two fields). WPLR, however, suffered from the error characteristics of the Full Threshold algorithm used to gather the data, not allowing a solid baseline estimate from the first two fields. More work is required to integrate customisation of PLR using individual's variability data.
References