Automated pupillometric quantification of relative afferent pupillary defect: comparison with neutral density filter grading in unilateral/asymmetric optic neuropathy.

Abstract

Purpose: To evaluate the diagnostic performance of automated pupillometric relative afferent pupillary defect (RAPD) scores based on constriction amplitude and latency and to assess their relationship with conventional neutral density filter (NDF) grading in patients with unilateral or asymmetric optic neuropathy.

Methods: Dynamic automated pupillometry was used to calculate RAPD amplitude and latency scores in 100 patients with clinically evident asymmetric optic neuropathy and 97 controls. RAPD was clinically graded using NDF. All participants underwent pattern visual evoked potential and retinal nerve fibre layer thickness assessment by optical coherence tomography. Diagnostic discrimination was assessed using receiver operating characteristic (ROC) analysis. Agreement with NDF grading was assessed using weighted Cohen's κ and exploratory Bland-Altman analysis. Correlations with inter-eye differences in RNFL thickness and pVEP latency were also evaluated.

Results: Both RAPD amplitude and latency scores were significantly higher in patients than in controls (p < .001). The area under the ROC curve was 0.850 (95% CI, 0.793-0.906) for RAPD amplitude and 0.799 (95% CI, 0.735-0.863) for RAPD latency. An exploratory amplitude cutoff of approximately 0.5 LU yielded 83% sensitivity and 80% specificity. The corresponding latency cutoff of approximately 0.24 LU yielded 87% sensitivity and 68% specificity. RAPD amplitude showed stronger correlation with NDF grading (Spearman ρ = 0.78) and greater quadratic weighted agreement (κ = 0.42) than latency (ρ = 0.68; κ = 0.36). RAPD amplitude showed weak correlations with inter-eye differences in pVEP latency (ρ = 0.25) and RNFL thickness (ρ = - 0.22). The difference between amplitude and latency AUCs was not statistically significant (ΔAUC = 0.051; p = 0.162).

Conclusion: Automated RAPD amplitude and latency measurements demonstrated good and fair discriminatory performance, respectively, in this predefined case-control cohort. Amplitude showed numerically better discrimination and stronger agreement with NDF than latency, but superiority was not demonstrated statistically. The approximately 0.5-LU amplitude threshold is consistent with thresholds reported in previous studies but should be considered exploratory and hypothesis-generating, given the two-gate design, lack of independent validation, and absence of repeatability assessment. Automated pupillometry may have potential as an objective adjunct to clinical RAPD assessment, but further prospective validation using standardised protocols and an independent reference standard is required.