Retinal nerve fibre layer thinning is associated with drug resistance in epilepsy

Objective Retinal nerve fibre layer (RNFL) thickness is related to the axonal anterior visual pathway and is considered a marker of overall white matter ‘integrity’. We hypothesised that RNFL changes would occur in people with epilepsy, independently of vigabatrin exposure, and be related to clinical characteristics of epilepsy. Methods Three hundred people with epilepsy attending specialist clinics and 90 healthy controls were included in this cross-sectional cohort study. RNFL imaging was performed using spectral-domain optical coherence tomography (OCT). Drug resistance was defined as failure of adequate trials of two antiepileptic drugs to achieve sustained seizure freedom. Results The average RNFL thickness and the thickness of each of the 90° quadrants were significantly thinner in people with epilepsy than healthy controls (p<0.001, t test). In a multivariate logistic regression model, drug resistance was the only significant predictor of abnormal RNFL thinning (OR=2.09, 95% CI 1.09 to 4.01, p=0.03). Duration of epilepsy (coefficient −0.16, p=0.004) and presence of intellectual disability (coefficient −4.0, p=0.044) also showed a significant relationship with RNFL thinning in a multivariate linear regression model. Conclusions Our results suggest that people with epilepsy with no previous exposure to vigabatrin have a significantly thinner RNFL than healthy participants. Drug resistance emerged as a significant independent predictor of RNFL borderline attenuation or abnormal thinning in a logistic regression model. As this is easily assessed by OCT, RNFL thickness might be used to better understand the mechanisms underlying drug resistance, and possibly severity. Longitudinal studies are needed to confirm our findings.


S2-OCT Methodology
The peri-papillary RNFL thickness was measured using the Optic Disc Cube 200x200 protocol. A 6x6mm grid of data is acquired with 200 horizontal scans, each composed of 200 A-scans that are centered over the optic disk. The participant was positioned at the machine and asked to fixate on an internal target. When necessary, polarization was optimized with the automatic tool or manually.
Once the optic disc was clearly visualised, the scan circle was centered on the optic disk. The position of the scan circle was analysed according to its relation to the optic nerve, and repositioned if necessary. We attempted to obtain at least three scans of both eyes, always starting with the right.
Scans were analysed according to quality, visualisation of optic nerve, movement artefacts and position of scan circle. The software provides a quality rating between 0-10 arbitrary units; scans were excluded if quality fell below 7. Quality measure variation above this threshold of 7 did not influence RNFL thickness measurement. The right eye scan with highest signal strength was selected for analysis for each individual. If more than one scan had equal signal strength, the last scan was selected.

S3-"One-eye design"
A "one-eye design" 4 was chosen for this study so that standard statistical techniques could be applied to analyse the data without increasing the risk of Type 1 errors. Analyses in which both eyes of an individual are used ("two-eye designs") require particular statistical approaches. This is because, within an individual, data from one eye are more likely to be similar to data from the other eye of the same individual, than it is from another eye from a different individual, i.e. within a subject the data from each eye will be highly correlated (unless there is unilateral eye disease) 4-5 .
This is due to multiple factors including genetic and environmental effects that will act within an individual to influence any structural or functional parameters 5 . In standard statistical analysis, data should be independent, and failure to take account of between-eye correlation in a "two-eye design" can lead to an overestimation of the precision of the statistical values [4][5] . When a "two-eye design" is used for the analysis data from both eyes should be averaged and the average value can be treated with the same standard statistical techniques as a "single-eye design". Alternatively, more complex statistical approaches are needed to determine the contribution to the outcome measure of variance arising from between-subject factors and those arising from within-subject factors 4 . A further complicating factor is added to both of these "two-eye design" approaches when some individuals only have data from one eye. In our study, whilst many individuals had data from both eyes, some individuals only had data from only one eye.
For a "one-eye" design, the eye to be entered into the analysis can either be chosen at random, or can be chosen to be always the right eye, or always the left eye 5 . Recently, this observation has also been made in a group of 248 healthy volunteers, where the average RNFL thickness in the right eye was 0.52µm thicker than the average RNFL thickness in the left eye, which reached statistical significance 6 . Thus only the right eye was chosen from each individual to use in the analysis.

S5-Study profile with exclusions
A total of 454 people with epilepsy were consecutively recruited and underwent OCT scan. Of these, 154 people were excluded according to the exclusion criteria (130 had previous vigabatrin exposure documented in medical records, 16 had diabetes, glaucoma or other known ocular disease, 3 had overt brain MRI evidence of occipital or visual pathway involvement, 2 had concurrent diagnosis of multiple sclerosis, 3 OCT scans below the minimum quality criteria or with significant movement artifact). Three hundred people with epilepsy therefore met inclusion criteria and were included in the analysis. Their scans were analysed and compared to results from 90 healthy controls participants who were consecutively recruited and included in the analysis.