Big data enabled keratoconus care

Abstract

Keratoconus is a disorder affecting the cornea in which its normal round shape progressively thins causing a cone‐like bulge to develop. Because the cornea does most of the focusing of the eye, this change in shape causes significant problems with vision. As the disease gets worse, the cornea gets steeper, and vision continues to decline. It is a significant cause of visual loss in the UK, affecting up to 1 in 450 young people and is the commonest indication for corneal transplantation in young people. Corneal cross‐linking (CXL) is the only proven treatment to prevent keratoconus getting worse. However there is limited reports of its long term safety in large numbers and we reported one of the largest series in the world.We used data from the Early Keratoconus Clinic over the past 10 years to generate our results, looking at how many patients progressed despite treatment. What we also did that was innovative was to create personalized progression threshold definitions. We know that early progression can be quite hard to detect, and in later disease the imaging is poorly repeatable. Therefore a personalized threshold, derived from repeated scans from our dataset, will be able to pick up disease earlier, before more progressive visual loss.We used real world data from our patients to create a predictive model, a calculator, that predicts when a patient is likely to need treatment. The reason for creating this calculator is that it allows patients to better understand their risks in a visual format, gives clinicians an objective marker for an individual's risk, and also allows the hospital to plan services based on needs.We were able to also use genetic data to see how it influenced disease progression, based on the gene segments that were identified by the Institute of Ophthalmology and Moorfields. This work has resulted in an open access calculator available on https://beta.moorfieldscxl.com.This talk will also look at the opportunities and challenges of deploying artificial intelligence algorithms in the management of keratoconus, and how we can adapt our own routine practices to contribute to the development of big datasets that can further train and validate AI algorithms.