Abstract
Purpose :
We previously showed that assessing glaucoma progression with visual field (VF) data post-processed with the dynamic structure-function (DSF) model resulted in a higher positive rate compared to the original (observed) data (Abu and Racette, IOVS 2020;61: ARVO E-Abstract 1991). The aims of this study were to validate this finding in an independent dataset and to assess the false positive rate of this approach.
Methods :
We used VF data from 139 patients with open-angle glaucoma (203 eyes) with at least 15 visits selected from the Rotterdam Eye Study. The DSF and ordinary least square linear regression (OLSLR) models were applied to the observed mean deviation (MD) values to derive two post-processed datasets: DSF-predicted and OLSLR-predicted. Specifically, MD values from visits 1–3 were used to predict MD at visit 4, then MD values from visits 1–4 were used to predict MD at visit 5. This procedure was repeated until MD for visit 9 was predicted. Linear regression was then used to assess progression within the observed, DSF-predicted and OLSLR-predicted datasets. Positive rate was determined for each dataset at a specificity of 95%. To examine the impact of VF post-processing on false positive rate, we performed 1000 permutations of the MD series in each dataset and reassessed progression. The permutations disrupted the temporal order of the MD series; therefore, significant negative slopes were deemed to be false positives. Mean false positive rates were computed for the observed, DSF- and OLSLR-predicted datasets.
Results :
The positive rate obtained with DSF-predicted data was 2.9% higher than OLSLR-predicted data and 12.8% higher than observed data at the 9th visit (Fig 1). DSF-predicted data flagged 72.9% and 59.7% of the eyes identified as progressing with observed data at the 12th visit and 15th visit, respectively. The false positive rates obtained for each of the datasets were similar to the proportion of eyes expected to be identified as progressing by chance (Fig 2).
Conclusions :
Our results confirm that post-processing visual field data using the DSF model yields a higher positive rate without compromising specificity. The higher sensitivity of this approach may lead to earlier detection of progression.
This is a 2021 ARVO Annual Meeting abstract.