There is growing interest in developing methods to combine information from structural and functional tests to improve the assessment of glaucoma progression and a potential technique is to use information from OCT to individualize automated perimetry testing strategies.1,2 A major advance in automated perimetry was the introduction of Bayesian algorithms such as the Swedish Interactive Threshold Algorithm (SITA), which reduce test time, with minimal to no loss in accuracy, by estimating thresholds using prior knowledge concerning the distribution of sensitivity at each test point, modified according to individual test responses. In this study, Montesano and colleagues investigated whether information from OCT could be used as an additional 'prior' to improve visual field testing in the macula.
Thirty patients with glaucoma and 20 healthy controls performed a 10-2 visual field using a fundus perimeter equipped with scanning laser ophthalmoscopy tracking (Compass, CenterVue, Padua, Italy). Participants were tested using the standard 10-2 Zippy Estimation by Sequential Testing (ZEST) strategy and an independent validation group of 20 patients with glaucoma had high precision testing of eight visual field locations using a full-threshold strategy. Macular OCT was used to estimate ganglion cell density for corresponding visual field test points using a previously described model,3 accounting for lateral displacement of ganglion cells, and these estimates were used to construct a structure-function model. Logistic regression was employed to determine the probability of a stimulus being seen or not seen based on age, local ganglion cell density, eccentricity and stimulus intensity. A series of 'probability of seeing curves', which incorporated information from OCT, were then constructed and these were used to modify the ZEST visual field strategy by altering the starting prior distribution on a point by point basis depending on estimates of ganglion cell density. The structural-ZEST algorithm was then validated using simulations based on data obtained from the independent group of glaucoma patients.
Important findings were that the novel structural-ZEST algorithm led to a 13% reduction in the number of stimulus presentations in reliable simulated subjects and a 14% reduction in those with higher (≥ 20%) false positive or false negative rates. This likely means that test time could be significantly reduced by using information from OCT to modify the 10-2 visual test algorithm. The simulation also suggested improved precision when measurements from OCT were taken into account. The structural-ZEST algorithm was less susceptible to higher levels of error, as shown by mean absolute error rates 10.6% less for structural-ZEST compared to ZEST when there was a 30% false-positive rate, and 18.3% less when there was a 30% false-negative rate. Structural-ZEST therefore seemed particularly beneficial in patients with less reliable visual fields.
Structural-ZEST therefore seemed particularly beneficial in patients with less reliable visual fields
Overall, the study provides further evidence that combining information from tests of structure and function is likely to improve the sensitivity for detection of glaucoma and glaucoma progression and more specifically that information from OCT can be used to improve visual field test strategies. A major strength of the study was the use of a fundus perimeter to improve fixation stability, but further work is needed to evaluate the model in real patients.