Purchase this article with an account.
William H. Swanson, Victor E. Malinovsky, Mitchell W. Dul, Julie K. Torbit, Bradley M. Sutton, Rizwan Malik; Agreement Between Contrast Sensitivity Perimetry (CSP) And Clinical Measures Of Glaucomatous Damage: Validation Of A Neural Model For A Longitudinal Study. Invest. Ophthalmol. Vis. Sci. 2012;53(14):5622.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We developed a neural model relating ganglion cell responses to perimetric sensitivity (IOVS 2004;45:466; IOVS 2011;52:764). Using baseline data from a longitudinal study of contrast sensitivity perimetry (CSP) whose design was based on this model, we tested its predictions about agreement between CSP and clinical measures of glaucomatous damage.
One eye each of 59 patients with glaucomatous field loss and 38 age-similar controls was tested with CSP, CAP (conventional automated perimetry), FDP (frequency-doubling perimetry) and ocular coherence tomography for RNFL (retinal nerve fiber layer) thickness. For each person, arithmetic means were computed for both RNFL thicknesses and perimetric sensitivities using maps of visual field locations associated with superior temporal (ST) and inferior temporal (IT) optic nerve sectors (IOVS 2002;43:2213). For each patient, defect depth for each sector was computed as log difference from mean of the controls. A ceiling of 1.5 log unit for depth of defect was used to control for differences in dynamic range (IOVS 2011;52:3759-60). RNFL thicknesses were used both as raw thickness values and "neural" values computed by subtracting a "non-neural" basement (1/3 of mean normal; IOVS 2007;48:3662) before calculating log defect. Predictions of the model were tested for IT and ST separately: 95% confidence limits for agreement will be ~ ±0.3 log unit, agreement with CSP will vary with depth of defect for CAP but not FDP, and will vary for RNFL but the effect will be altered by removing the non-neural basement effect. Predictions about effect of defect depth on agreement between measures were tested by linear regression of difference versus mean for each pair of measures. Z-scores were computed for slope, and the test for statistical significance was set to | Z | > 2.73 (p < 0.006) to control for repeated tests.
Correlations were strong for all comparisons with CSP (r2 from 51% to 83%), except for the "neural" value for ST RNFL thickness (r2 = 40%). Confidence limits for agreement ranged from ±0.27 to ±0.35 log unit for CSP/CAP, CSP/FDP and CSP/RNFL. For both sectors (IT, ST), values for Z were in accord with the predictions: (-5.3,-4.5) for CSP/CAP, (+8.8,+6.4) for CSP/RNFL, and (+0.3,-2.2) for CSP/FDP. When RNFL thicknesses were converted to "neural" values, confidence limits increased (±0.54, ± 0.56 log unit) and the effect of depth of defect was reversed (Z = -1.5, -3.1).
The predictions of the neural model were confirmed in these baseline data, validating the use of the model for analysis of longitudinal fluctuations in agreement between CSP data and clinical data from these patients and controls.
This PDF is available to Subscribers Only