Epidemiology of eye disease in the UK: the Bridlington Eye Assessment Project

Poostchi, A. (2023) Epidemiology of eye disease in the UK: the Bridlington Eye Assessment Project. PhD thesis, University of Nottingham.

PDF (epidemiology of eye disease in the UK) (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Available under Licence Creative Commons Attribution Non-commercial.
Download (5MB) | Preview



To determine the prevalence of and impact of eye disease in an elderly population and the diagnostic accuracy of a novel artificial intelligence algorithm for the detection of glaucoma

Design; population based, cross sectional study


3549 Caucasian individuals over the age of 65 years


A directed general and ophthalmic history was obtained from all subjects. Slit lamp eye examination including applanation tonometry and dilated examination of the fundus was performed by one of four specially trained optometrists and supplemented with fundus photography, visual field testing and Heidelberg

Retinal Tomography (HRT). Those with reduced vision, raised intraocular pressure, visual field defects or other abnormalities were referred for further assessment by a consultant ophthalmologist and followed longitudinally until a

definitive diagnosis was made. All diagnoses of glaucoma were made retrospectively using at least 5 years of longitudinal data to determine status at incident examination. All fundus photographs were reviewed by a single ophthalmologist for signs of age related macula degeneration (AMD) and other

retinal disease. HRT outputs were analysed using the device’s proprietary software which produced results for normative based Moorfield’s regression analysis (MRA) and the shape analysis tool Glaucoma Probability Score (GPS).

We used a bespoke Matlab based machine learning classifier to providing two further measures based on shape analysis which were termed shape abnormality score (SAS) and abnormal disc score (ADS).

Statistical Analysis: Outcomes and associations were explored using t-tests, chi- squared tests and Mantal Haenzel methods. Linear and logistic regression was sed for multivariate analysis. Agreement was measured using kappa, intraclass correlation coefficient and concordance correlation coefficient and plotted using Bland Altman plots. Covariate effects on diagnostic performance were examined using a combination of maximum likelihood probit models and bootstrap analysis. All data analysis was performed using Stata v14


Cataract; Significant lens opacities were present in 45% of individuals of whom 12% went on to have cataract surgery. Women were 29% more likely to have significant cataract than men. 9.5% of eyes showed signs of previous cataract surgery of which 17% either required or had received treatment for subsequent posterior capsular opacification. In the absence of thresholds for surgery 18 cataract operations per thousand ( 95% CI 14 – 23 ) were required for those aged 65-75 years old. For those over 75 years, 76 cataract operations per thousand ( 95% CI 66 – 86 ) were required

AMD; Geographic atrophy (grade 4a) occurred in 2.5%, and neovascular AMD (grade 4b) in 1.8% of eyes. Prevalence increased with age with grade 4 (advanced) AMD in 2.2% of those aged 65–69 years, 15.8% for those aged 85–

90 years, and 21.2% for over 90 years. There was significant asymmetry between eyes of individuals with advanced AMD (P<0.001). After correction for age and co-pathology, those with advanced AMD in the better eye were 4 times more likely to be disattisified with their vision.

Glaucoma; For tests with a specificity of > 90% for new OAG, intraocular pressure was the least sensitive (48%), while clinical CDR ≥ 0.7 was the most sensitive (76%) test. Optometric impression showed the best specificity (98%) with acceptable sensitivity (51%) but may have been subject to verification bias since final diagnosis was based on clinical impression albeit with the reference to longitudinal results. Because of the low relative prevalence of new glaucoma, the test specificity of 98% still resulted in referral of nearly twice as many false positives as new patients with glaucoma. There was moderate agreement between individual optometrists and the HRT in measuring CDR but wide limits of agreement precluding effective comparison between approaches.

Of the disc based measures, SAS and optometric assessment were found to be the most specific but MRA showed the best overall performance. In our subgroup analysis, we found a drop in sensitivity for detection of new disease by HRT using automated shape analysis and by optometrist using Jonas criteria. MRA performed well across all groups and showed similar sensitivity in detection of new and previously diagnosed glaucoma.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Vernon, S.A.
Dua, H.S.
Keywords: epidemiology; eye; ophthalmology; glaucoma; cataract; macula degeneration; artificial intelligence; AI
Subjects: W Medicine and related subjects (NLM Classification) > WW Ophthalmology
Faculties/Schools: UK Campuses > Faculty of Medicine and Health Sciences > School of Medicine
Item ID: 72448
Depositing User: Poostchi, Ali
Date Deposited: 06 Oct 2023 12:41
Last Modified: 06 Oct 2023 12:41
URI: https://eprints.nottingham.ac.uk/id/eprint/72448

Actions (Archive Staff Only)

Edit View Edit View