Regression-based strategies to reduce false-positive glaucoma diagnoses when using OCT and OCT angiography

Abstract

Purpose : To correct refractive error-associated bias in optical coherence tomography (OCT) and OCT angiography glaucoma diagnostic parameters.

Methods : OCT and OCT angiography imaging were obtained from participants selected from the Hong Kong FAMILY cohort, a population-based study. Only normal eyes were included in this study. Software on the Avanti AngioVue system were used to measure the peripapillary nerve fiber layer thickness (NFLT) and nerve fiber layer plexus capillary density (NFLP CD), and macular ganglion cell complex thickness (GCCT) and superficial vascular complex vascular density (SVC VD). The glaucoma diagnostic threshold was set at the 5^th percentile of emmetropic (-1 to +1 D) eyes.

Results : A total of 1346 eyes from 792 participants were divided into four subgroups for data analysis (Table 1). After accounting for age, gender, and signal strength, multiple linear regression showed strong dependence for NFLT and GCCT on eye axial length (AL), spherical equivalent (SE) refraction, and apparent optic disc diameter (DD) on the OCT scans. Compared to structural parameters, NFLP CD had less dependence and SVC CD had no significant dependence on AL, SE, and DD. Compared to the emmetropic group, the false positive rate was significantly (Chi square test p < 0.003) elevated in both the high and low myopia groups for NFLT, NFLP CD, and GCCT (Table 2). Regression-based adjustment of diagnostic parameter with AL or SE significantly (McNemar test p < 0.04) reduced the elevated false positive rate. Adjustment using DD was ineffective.

Conclusions : Myopic eyes are biased to have lower NFLT, GCCT, and NFLP CD measurements, leading to elevated false positive rate of glaucoma diagnoses. Regression-based adjustment using AL and SE, but not DD, were effective in eliminating this bias. SVC CD was unbiased by refractive error even without compensation and may be more reliable for glaucoma assessment in high myopes.

This is a 2021 ARVO Annual Meeting abstract.

 

Table 1: Demographic and ocular characteristics of study participants including subgroups displayed in mean ± standard deviation form.

 

Table 2: False positive rate of glaucoma detection stratified by refractive error. Adjustments in diagnostic parameters were made using multiple linear regression against age, gender, and signal strength index in the original row. Below that, additional adjustment of either AL, SE, or DD was performed.