Abstract
Purpose: :
To spatially characterize the retinal nerve fiber layer (RNFL) thinning that corresponds with loss of perimetric sensitivity at each visual field location.
Methods: :
Data were taken from 1385 eligible visits of 213 participants enrolled in the ongoing Portland Progression Project study. Each visit consisted of an Ocular Coherence Tomography scan together with a Standard Automated Perimetry visual field (using the SITA testing algorithm, with no more than 33% fixation losses or false negative errors). Linear mixed effects models were created to predict the sensitivity at each field location from the average RNFL thicknesses within each clock-hour sector of the optic nerve head, accounting for longitudinal and inter-eye correlations. Single backwards elimination was used to find the most predictive sectors. The correlations between these predictions and the observed sensitivities were calculated. First, a map was created using the entire dataset, repeating the process using different numbers of RNFL sectors in the final predictive model. To assess robustness, maps were then created using data from a randomly-chosen half of the participants, and used to predict the sensitivity at each location within data from the remaining half of the participants, repeating this process for different bootstrapped subsets of the dataset.
Results: :
Using the entire sample for both derivation and validation, the pointwise correlations averaged 0.40 using only one sector per field location; 0.44 using two sectors; 0.46 for three sectors; asymptoting to 0.47 using all twelve sectors. Henceforth three sectors were used per field location. Predictability was highest in the SupraNasal quadrant (mean correlation=0.53), followed by the SupraTemporal (mean=0.46), InferoNasal (mean=0.45) and InferoTemporal (mean=0.40) quadrants. RNFL sectors found to be predictive of sensitivity (for a right eye) were most commonly 6-8 o’clock for locations in the superior visual field, and 11 and 7 o’clock for the inferior visual field. Averaged over the bootstrapped subsamples, the correlations between the predicted and observed sensitivities at each field location had mean 0.48 in the SupraNasal quadrant, 0.41 SupraTemporal, 0.40 InferoNasal and 0.35 InferoTemporal.
Conclusions: :
RNFL thickness within three clock-hour sectors predicted observed pointwise sensitivity well enough to justify using it to help reduce test variability, but not well enough to justify using it as a surrogate for functional testing.
Keywords: perimetry • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)