Abstract
Purpose :
Recently we have shown that a compensation formula based on multiple linear regression analysis reduced the interindividual variation of retinal nerve fiber layer (RNFL) measurement by 18% on average. This regression model can be applied to generate patient specific individualized normal limits (INL). The purpose of the present study is to analyze whether the number of false positive (FP) sectors or the number of subjects displaying FP sectors is reduced when using INL as opposed to group normal limits (GNL).
Methods :
This is a retrospective analysis of 202 healthy subjects having been collected for another prospective study. All subjects underwent FD-OCT (Cirrus ®, Carl Zeiss Meditec) of the macula and the circumpapillary area. The retinal vessels was automatically segmented using both scanning protocols, and the vessel thickness values at a 3.4 mm diameter circle were used to calculate the retinal vessel density function (RVD). The RVD reflects the locations and thickness values of the retinal vessels according to a Gaussian convolution. The factors considered for multiple linear regression were: RVD, distance between fovea and optic disk (OD) center, fovea angle, OD size, OD ratio (=elongation), OD angle and refractive error. The normal limits were calculated twice: for the GNL we used our sample’s mean values and standard deviations, and for the INL we applied the regression formula, using the RNFL estimate and the variance of the model. For each clock hour sector we evaluated the number below the 5% normal limit (NL5), or below the 1% normal limit (NL1). We also evaluated the number of subjects with at least one sector below NL5 or NL1. The reduction of the number of FP sectors and subjects when applying the INL, as opposed to the GNL was compared using Chi-square tests.
Results :
The number of clock hour sectors below the NL5 and NL1 was reduced from 82 to 23 (p<0.00001) and from 13 to 1 (p=0.00132) out of 2424 sectors for the patient specific INL compared to the GNL. The number subjects with at least one sector below NL5 and NL1 was reduced from 48 to 19 (p=0.00010) and from 12 to 1 (p=0.00187) out of 202 subjects.
Conclusions :
Application of patient specific INL resulted in a very pronounced reduction of FP sectors and of subjects with FP sectors, compared to standard GNL.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.