Using statistical decomposition to improve the identification of the photopic negative response of the erg

Abstract

AbstractPurpose The Photopic Negative Response (PhNR) is known to be correlated with disease severity in animal models of glaucoma. The technique has found little utility in the clinical setting due to the difficulty involved in obtaining repeatable measures. . Inter‐Session reliability has been poorly successful because of the difficulty to consistently identify and define the PhNR. In this study we propose a robust PhNR measure based on curve fit modelling of the three components of the electroretinogram (ERG) response. The robustness of this measure comes from the use of a global 'shape' as a PhNR measure which is less sensitive to variations than measuring a single value on the ERG.Methods 39 subjects with ‘normal’ eyes were recruited. Each subject underwent 2 sessions of bilateral ERG tests at five brightness levels. Raw data obtained from the tests was exported to the R statistical program. A parametric function consisting of a linear combination of three Gaussian functions was fitted to each averaged response . For each of these components, the optimal parameters were estimated using non‐linear least squares. The amplitude of the 3rd Gaussian was used as a measure of the PhNR.Results Rapid convergence and successful fitting was achieved for 446 out of 457 traces. The amplitude of the third Gaussian correlated well with the manually measured PhNR (r=0.79). Bland Altman plots showed good agreement between modelled waveforms and raw data.Conclusion By decomposing the ERG into three components the effect of variability of the baseline and the b wave can be decreased. Agreement between the 3rd Gaussian amplitude and manually estimated PhNR was reasonable. Inter‐session reliability of this new measure will be further assessed.