Abstract
Purpose :
Retinal Nerve Fiber Layer Thickness (RNFLT), as measured using Optical Coherence Tomography (OCT), correlates well with functional measures from perimetry. However, a variety of data suggests that structural change within the optic nerve head may precede RNFL change. This study asks whether changes in the neuroretinal rim, as measured by the OCT parameter Minimum Rim Width (MRW; averaged around the neuroretinal rim), predict subsequent changes in RNFLT.
Methods :
A structural equation model (SEM) was formed, as illustrated in the Figure. The principle is that each eye has two possibly correlated latent (unobservable) variables; baseline retinal ganglion cell (RGC) count, and rate of RGC loss. The observable values of MRWn and RNFLTn at time n are predicted by their values at time (n-1), and the values of the latent variables for that eye at time n. This model was fit using the lavaan software package for R, on data from 160 eyes of 160 participants with glaucoma or suspected glaucoma in the Portland Progression Project longitudinal study, tested every 6 months using the Spectralis OCT.
Results :
The mean series length was 8.7 visits (range 5-11), over 4.3 years (range 2.5-5.1). The fit of the model was good, as seen in the Table. An increased baseline RGC count was correlated with more rapid RGC loss (r=-0.135, p<0.001). MRWn was predicted by MRWn-1 and RNFLTn-1 (both p<0.001). RNFLTn was predicted by RNFLTn-1 (p<0.001), but not MRWn-1 (p=0.251). Similar analyses using longer time lags of up to 3 years for MRW did not improve the fit of the model. When splitting the data into equal parts based on age, MRWn-1 did not improve predictions of RNFLTn in older eyes (p=0.301), but it is possible that it could help in younger eyes (p=0.075).
Conclusions :
There was no evidence that MRWn-1 improved predictions of RNFLTn measured 6 months later, compared with just using RNFLTn-1. It remains possible that the time lag between changes in these measures could be less than 6 months and so undetectable in this dataset, or much longer than was tested here. However, the results may be because MRW in its current form is more variable than RNFLT, reducing its utility as a predictor.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.