Modeling the relative influence of fixation and sampling errors on retest variability in perimetry

Background: Our previous studies have shown that in standard automated perimetry (SAP) undersampling occurs if sensitivity varies across a visual field faster than the Nyquist rate (Nq) for the standard sampling interval of 6°. This undersampling was shown to be a major source of test–retest variability. This study first tests some of the assumptions of the undersampling idea, and then determines the relative contributions to test–retest variability of normal eye movements and undersampling. Methods: In all models fixational jitter was at normal levels. The first part investigates the effects of the jitter on the Fourier spectra of fields, and stimulus size effects. In the second part fields are smoothed in six gradations up to and beyond the point where no undersampling could occur. The spatial smoothing gradations covered nil to < Nq/4. For each smoothing level the resulting retest variability was determined for each of 11 bands of scotoma depth (0 to −28.5 dB). Results: As is commonly reported, and as undersampling predicts, retest variability was largest for deeper scotoma depths. When smoothing suppressed all undersampling effects, the inter-quartile range of the residual retest variability averaged only 2.3 ± 0.33 dB, much smaller than for unsmoothed fields (p < 0.003). For the five deepest scotoma depth bands (range, −16.5 to −28.5 dB) retest variability was smaller by 6.0 ± 0.5 dB (p < 0.0005). Conclusions: Retest variability appears in large part to be driven by undersampling. In real fields, the remaining variance would come from fixation errors and physiological sources.