Abstract
Purpose :
Visual acuity(VA) is the primary visual function endpoint in most ophthalmic trials. However, the high test-retest variability(TRV) of VA testing indicates a lack of precision and reduced sensitivity for detecting vision changes. To enhance detection of VA signals, we developed three data analytic procedures to reduce variability of E-ETDRS testing (Beck et al., 2003), the electronic version of the gold standard ETDRS chart.
Methods :
We initiated post-hoc analyses with a VA behavioral function (VABF) as the generative model of trial-by-trial performance for observers in VA testing (Figure 1). The model comprises two parameters: VA threshold, representing the optotype size required to achieve a specific performance level, and VA range, specifying how rapidly VA behavior changes with increasing or decreasing optotype size. Three distinct procedures were developed to infer VA threshold and range from E-ETDRS testing:(1) A Bayesian Inference Procedure (BIP) that calculates the posterior distribution of VABF parameters independently for each E-ETDRS test, (2) A hierarchical Bayesian model (HBM) that computes the joint distribution of the VABF parameters and hyperparameters from all E-ETDRS data in the study (Zhao et al., 2021a), and(3) A hierarchical Bayesian joint model (HBJM)(Zhao et al., 2023) that computes the joint distribution of the VABF parameters and hyperparameters from both E-ETDRS and qVA (Lesmes & Dorr, 2018) data in the study. These procedures were applied to a VA dataset obtained from 14 eyes, with four repeated measures in each of 4 Bangerter foil conditions with both E-ETDRS and qVA (Zhao et al., 2021b). We assessed TRV(1.96×test-retest difference SD) of the estimated VA thresholds derived from the repeated E-ETDRS tests.
Results :
Figure 2 displays the Bland-Altman test-retest differences of VA thresholds from the original E-ETDRS procedure and three new analyses. The TRV values were 0.17 for E-ETDRS, 0.19 for BIP, 0.14 for HBM, and 0.12 logMAR for HBJM. While the TRV for BIP is comparable to that of E-ETDRS, the HBM and HBJM reduced the TRV by 22% and 30%, respectively.
Conclusions :
By integrating information across subjects, the HBM improved TRV of the E-ETDRS tests. Integrating information across subjects as well as additional tests, the HBJM exhibited the greatest reduction of TRV of the E-ETDRS test. Both post-hoc procedures can be employed to enhance the sensitivity of E-ETDRS testing.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.