AI predicts eye damage from autoimmune drug, years before symptoms appear

A novel AI tool could transform screening for patients taking hydroxychloroquine by spotting early signs of eye damage, according to a study published in Ophthalmology Retina.

Researchers have devised an artificial intelligence (AI) system that can detect and predict serious eye damage caused by a common autoimmune medication years before patients or doctors notice any symptoms or warning signs. On average, the system flagged patients who would develop retinopathy two and half years before doctors made the diagnosis.

The research algorithm could change how millions of people taking hydroxychloroquine are monitored for a potentially blinding side effect that currently affects up to 7.5% of users. Development was led by a team at Moorfields Eye Hospital NHS Foundation Trust and University College London (UCL) Institute of Ophthalmology, using data from Moorfields and other eye centres in the UK and US.

In the UK, there are an estimated 77,000 long-term users of hydroxychloroquine (according to British Ophthalmological Surveillance Unit figures from 2022). In the US around 1.2 million people take the medication (ClinCalc figures 2013-2023).

Hydroxychloroquine, widely prescribed to treat rheumatoid arthritis, lupus, and other autoimmune conditions, can cause irreversible damage to the retina - the light-sensitive tissue at the back of the eye. This condition, called hydroxychloroquine retinopathy, typically develops after years of use of the drug and can lead to permanent vision loss if not caught early.

The challenge is that significant harm may have already occurred by the time doctors can reliably spot the damage using current screening methods, which involve an annual ophthalmic check-up for anyone who has taken the drug for five years. Currently, the only way to prevent damage is to stop taking the medication, which can mean worsening the symptoms of arthritis.

Trained on over 8,000 eye scans from 409 patients in the US and UK, the algorithm, called HCQuery, can analyse retinal images captured by optical coherence tomography (OCT), which is part of standard screening for hydroxychloroquine patients. The algorithm correctly identified 100% of patients with retinopathy, up to 2.74 years earlier than doctors, achieving 91% accuracy in ruling out patients without the condition. 

These images show scans of the retina on three dates prior to a clinician diagnosing retinopathy due to hydroxychloroquine use. A. 2,382 days before the clinician diagnosed retinopathy, the algorithm shows a low (2.4%) probability of retinopathy. B. 1,221 days prior to clinical diagnosis, the algorithm shows a 77.9% likelihood. C. 477 days before a clinician diagnosed retinopathy, the algorithm is 98.1% confident that retinopathy is developing.

HCQuery demonstrated its effectiveness across different patient populations, including self-reported White, Black and Asian ethnicities, who develop different patterns of retinal damage. It worked equally well on eye scans from varied machines and hospitals, suggesting the potential for wide adoption.

Lead author Peter Woodward-Court said, "This novel approach to screening could significantly improve care for the millions of people who depend on hydroxychloroquine. Early detection would prevent irreversible vision loss, while allowing patients to continue benefiting from this important medication for longer."

Peter is a PhD researcher at UCL Institute of Ophthalmology and an honorary research fellow at Moorfields Eye Hospital.

In the next phase of AI development, authors plan to investigate how the algorithm performs in a real-world setting, exploring how the current care pathway can be optimised for earlier detection of hydroxychloroquine retinopathy.

The algorithm was enabled by data infrastructure at the INSIGHT Health Data Research Hub at Moorfields Eye Hospital, and the authors have made the code publicly available on GitHub in order to accelerate its development and encourage further testing across diverse patient communities.

---

Image credits: top right, Canva; bottom: reproduced with author permission from Ophthalmlogy Retina under Creative Commons Licence Deed - Attribution 4.0 International - Creative Commons