Abstract
Purpose:
To compare the agreement of measured and predicted retinal nerve fiber layer thickness (RNFL) by using different equations based on peripapillary RNFL measurements obtained by time-domain OCT (TD-OCT) and spectral-domain optical coherence tomography (SD-OCT).
Methods:
The study included 138 eyes of healthy volunteers, 414 eyes of patients with glaucoma. All patients underwent standard clinical examination, TD-OCT (Stratus OCT) and SD-OCT (Spectralis OCT) measurements of peripapillary RNFL. Two groups were matched for diagnostic subgroup, eye side, gender and age. TD-OCT measurements of the first group were used to predict SD-OCT average and six sector vertical split RNFL measurements of the second group and vice-versa. Agreement between predicted RNFL thickness calculations of conversion equations and measured RNFL thickness of the second group was assessed by intraclass correlation coefficients (ICCs) and Bland-Altman plots.
Results:
Agreement was excellent for all investigated equations to predict average RNFL thickness with ICCs ranging from 0.966 to 0.970. Bland-Altman plots showed good agreement between measured and predicted average RNFL thickness for both devices. Systemic biases between -0.7 and +1.1 µm and no proportional biases were observed. To predict Spectralis average RNFL thickness the regression equation method showed the lowest absolute mean difference and the lowest median percentage difference between actual and calculated RNFL thickness with +0.9 and 4.7%, respectively. To predict Stratus RNFL thickness the regression method showed the lowest absolute mean difference with +0.6 µm and the ratio equation showed the lowest median percentage difference between actual and calculated RNFL thickness with 4.6%. The ratio method showed highest ICC values for all regional sectors (range between 0.770 for the nasal sector and 0.969 for the temporal-inferior sector).
Conclusions:
Prediction of average RNFL thickness values by equations derived from SD-OCT and TD-OCT data can be conducted with high levels of agreement. Methods reported in this study can be applied for different study populations from both Stratus OCT and Spectralis OCT to determine the best method for conversion and provide longitudinal comparability. Nevertheless in individual cases and singular sectors high prediction errors may occur.