Utility of isolated-check visual evoked potential technique in dysthyroid optic neuropathy

Abstract

Purpose: To analyze the utility of isolated-check visual evoked potential (icVEP) for discriminating between eyes with dysthyroid optic neuropathy (DON) and eyes with thyroid-associated ophthalmopathy (TAO) but not DON.

Methods: Forty-three eyes with TAO but not DON (as non-DON), fifty-three eyes with DON, and sixty healthy eyes (as controls) were included. Comprehensive ophthalmic examinations, including best-corrected visual acuity, refraction, color vision test, intraocular pressure measurement, slit-lamp biomicroscopy, ophthalmoscopy, RAPD, exophthalmometry measurements, pVEP test, icVEP test, standard automated perimetry, and clinical activity score classification of TAO, as well as demographic information, were collected and analyzed.

Results: In the DON group, the signal-to-noise ratio (SNR) value of icVEPs decreased significantly compared with that of the non-DON group as well as control (p < 0.05). The SNR values under 8%, 16% and 32% depth of modulation (DOM) were significantly negatively correlated with BCVA (p < 0.05, r = - 0.9 ~ - 0.6), papilledema (Y/N) (p < 0.05, r = - 0.8 ~ 0.4) and DON (Y/N) (p < 0.001, r = - 0.7 ~ - 0.5). The 8% DOM of icVEP had the largest area under the receiver operating characteristic curve (AUC) (0.842) for discriminating DON from non-DONs. Meanwhile, decision curve analysis (DCA) showed that patients clinically benefit most from 8% DOM of icVEP. Furthermore, the 8% DOM of icVEP combing with papilledema (Y/N) and BCVA (Model 1) has significantly larger AUC than the 8% DOM of icVEP (p = 0.0364), and has better clinical benefit in DCA analysis.

Conclusions: The SNR of 8% DOM from icVEP may represent a significant ancillary diagnostic method for DON detection. Furthermore, icVEP combined with papilledema (Y/N) and BCVA should be considered as a diagnostic model in future clinical practice.

Keywords: Diagnosis; Dysthyroid optic neuropathy (DON); Isolated-check visual evoked potential (icVEP); Thyroid-associated ophthalmopathy (TAO).

Fig. 1

The correlation of the visual functions among three groups. The Scatter plot graphs present the visual function among the control group, DON group and Non-DON group. (a–c) The value of SNR (8%DOM,16%DOM and 32%DOM) of icVEP in the three groups. (d) BCVA in the three groups. (e–f) The two parameters (MD and PSD) of visual field examination in three groups. (g–h) The two parameters (P100 Amplitude and P100 latent time) of p-VEP examination in three groups. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001

Fig. 2

Comparison of icVEP, VF, and pVEP in the diagnosis of DON. (a) Receiver operating characteristic curves for the 8%, 16%, and 32% DOM of icVEP. The AUCs for the 8%DOM, 16%DOM and 32%DOM were 0.842 (95% confidence interval [CI], 0.763–0.922); 0.788 (95% confidence interval [CI], 0.696–0.880) and 0.778 (95% CI, 0.686–0.870), respectively. (b) Decision curve analysis of six parameters for diagnosing DON. The diagonal light-gray solid line is shown when all patients are considered as having the outcome (which is DON), while the horizontal dark-gray solid line presents when all are considered as not having the outcome; the above two lines represent two extreme cases. On the abscissa line, the red square represents the high risk threshold range with highest NB of the red line (SNR 8%DOM), and the light blue square represents the high risk threshold range of the light blue line (MD VF) with highest NB, while the gray square indicates that multiple DCA curves partially overlap or basically overlap the high risk threshold range. The red curve (8%DOM) has widest range with highest NB. (c) Diagnostic power of the SNR (8%DOM, 16%DOM, 32%DOM) of icVEP, PSD(dB), MD(dB), and P100 latent time(ms). The 8% DOM had the largest AUC (0.842) for discriminating DON from non-DON subjects SNR, signal-to-noise ratio; DOM, depth of modulation of the check luminance; PSD, pattern standard deviation; MD, mean deviation

Fig. 3

Comparison of the diagnostic capabilities of different diagnostic models for DON. Model 1 = 6.92785 − 2.45686*SNR(8%DOM) + 18.12121*( Papilledema = 1) − 5.03063*BCVA, Model 2 =  − 2.17895 − 1.73444* SNR(8%DOM) − 0.76181* SNR(16%DOM) + 0.06535* P100 latent time(ms) − 0.41904* MD(VF) − 3.61406* BCVA + 18.27022*(RAPD = 1), Machine and learning shows that the factors that affect DON were: SNR(8%DOM), P100 latent time (ms), MD (VF), SNR (16%DOM) and BCVA, Receiver operating characteristic curves for the models prediction of DON. The AUC values of model 2 (red curve) had the largest AUC (0.924) for discriminating DON from non-DONs. Decision curve analysis of three models for diagnosing DON. On the abscissa line, the blue square represents the high risk threshold range with highest NB of the blue line (model 2), and the green square represents the high risk threshold range of the green line (model 1) with highest NB, while the gray square indicates that multiple DCA curves partially overlap or basically overlap the high risk threshold range. The blue curve (model 2) has widest range with highest NB