Precision management to revolutionise glaucoma diagnosis and treatment
Artificial intelligence and genetic testing are expected to facilitate choosing the best treatment for each unique glaucoma patient within the next decade. This will not only lead to the best outcomes for individuals but will also spare them the costs and side effects of suboptimal treatments.
I spoke to Dr Anthony Khawaja of University College London and Moorfields Hospital, London, UK, about the exciting realm of precision medicine in glaucoma.
How do you define ‘precision management’?
In looking after our patients, most of us review the evidence in the literature. For example, when considering glaucoma patients, I might look at the UK glaucoma treatment study (UKGTS),1 a randomised controlled trial that showed that patients treated with a prostaglandin analogue do better on average than those who are not treated.
Evidence-based medicine tells us how best to treat people on average, but everyone is different. Not everybody in the UKGTS who received latanoprost necessarily benefited from it; in fact, 15% of such patients progressed over 2 years, so they were under-treated. Yet three-quarters of the placebo group did not progress, so some may have been over-treated.
Precision management is about going beyond simply what is best for the average group of patients and instead being able to predict, the first time you see a patient, their risk of worsening, and then titrating their treatment accordingly — maybe starting off more intensively in some high-risk patients so that you do not have to observe further vision loss before stepping up treatment. And it is it about choosing no treatment at all for other patients.
This means resources and treatments can be directed only to the people who will benefit, thus sparing the costs and side effects of treatments that will not work. The key here is understanding that one size does not fit all.
Could you expand on what this would look like for glaucoma?
One example of where precision glaucoma management could be useful is trying, the first time you see the patient, to make an assessment about whether they need treatment or not. Take the patient who has high ocular pressure but not glaucoma as of yet. We know from the Ocular Hypertension Treatment Study (OHTS)2 that if you treat such patients, they will do better on average than people who do not receive treatment: fewer of them will convert to glaucoma. But we also know that most of the group that is not treated do not develop glaucoma, so some are being treated unnecessarily.
An OHTS calculator that predicts the risk of conversion to glaucoma already exists,3 which takes into account the age of the patient, baseline pressure, corneal thickness, visual field mean deviation and cup-disc ratio. Unfortunately, this calculator, while helpful, does not really go far enough.
And so what we are looking to do is bring in other new technologies using genetic information or perhaps artificial intelligence (AI) assessments of images to help decide whether a patient will very likely benefit from treatment or, actually, are at such low risk that we can save the costs and the hassle of treatment. This will in turn also free up health service resources so that we can focus on the patients at highest risk.
You can apply this to other aspects of glaucoma care. For somebody with established glaucoma, do you treat or not, and how aggressively do you treat?
Another area which in my opinion can really benefit from the precision management approach in glaucoma is choosing the right treatment. There are so many different types of treatments available, including four major classes of drug, selective laser trabeculoplasty for open angle disease, and a whole spectrum of surgery from the minimally invasive glaucoma surgeries to the traditional incisional surgeries such as trabeculectomy and tube implants.
And if we are honest with ourselves, we do not know who is going to do well with each treatment. As mentioned, some people respond very well to latanoprost, but others do not respond at all and some experience side effects.
Where does genetic testing come in?
In recent years, we have really started to understand what causes glaucoma. In fact, glaucoma is one of the most hereditary of the common diseases in humans; it is about 70% heritable.
This means, in part, that some glaucoma is inherited through generations in an autosomal dominant fashion due to just a single mutation, most commonly in the myocilin gene. But this comprises only about 4% of primary open angle glaucoma (POAG).
The majority of cases are more complex because they are due to a lot of genetic effects, each one small and insufficient in causing glaucoma on its own but when put together cumulatively reach a threshold that causes disease. For years we were unable to identify these small effects, but recently, with large studies, we have been able to identify many more.
For example, 2 years ago, by looking at the pressure in the eyes of around 140,000 people, we identified over 100 points in the genetic code that underlie some people developing high pressure and then glaucoma.4 And what is exciting is that, when you combine the genetic information at all of these points, you can start to predict who is going to get glaucoma, even from the point of birth because our genetic code does not change after that.
Given that it has got such a genetic basis, what we can start to do next is pick out different types of glaucoma, for example, which are currently indistinguishable. So, POAG to us looks pretty uniform. However, genetically speaking there is a subgroup that seems to have a problem particularly located in Schlemm’s canal and collector channels; for another group the problem seems to particularly affect the trabecular meshwork.
If we could work out which of these patients had primarily a trabecular meshwork problem then we would know that treatment targeted to the trabecular meshwork, such as selective laser trabeculoplasty, or the iStent Inject (Glaukos) and trabecular bypass procedures with microstents (Hydrus), is more likely to work.
Conversely, if the issue is predominantly due to a problem with a patient’s collector channels and Schlemm’s canal, then maybe you are better off not targeting the trabecular meshwork. In that way, we can start to use genetic risk scores to be able not only to tell us about the risk of disease progression but also to predict response to treatment.
How exactly would artificial intelligence assist?
Well, in terms of personalising care, there is an awful lot of information that we collect as glaucoma specialists: images that provide detailed information and potential patterns that we recognise. We learned through our training that we can sometimes look at an optic nerve and see when there is a lot of damage and when a person is at high risk, meaning we need to treat them aggressively.
But there are other times where we just get it wrong and do not predict at baseline that somebody is at high risk of getting worse, and they worsen despite our best efforts. Conversely, we might treat other people that are at low risk and we can never know whether they needed that treatment.
The real power of AI in medicine so far is the ability to classify images, which it can do better than any single human grader. AI has the potential to see patterns in an image that no human being, no matter how expert they are, can see: an interesting study that supports this concept is one from the Google Brain group.5
They trained an AI algorithm to look at photos of the retina of healthy participants in a couple of studies, including the UK Biobank study, and to predict sex. The algorithm is 97% accurate in identifying whether a fundus photo is from a man or a woman. Is there is one glaucoma specialist or retinal specialist or human in the world who can do that?
The enticing thought is that if AI can see patterns beyond what humans can appreciate currently, it can predict the future. I can definitely see AI working together with genomics in the future, providing much more individualised care.
When will this all become a reality?
I have set my research group an ambitious target: by 4 years’ time we will have developed proven algorithms that can predict the risk of glaucoma progression and response to two different treatments. Once we have got this tool, we need to make it easily accessible and understandable to general glaucoma doctors and physicians, and even patients.
We will need to test it prospectively to see whether it does improve decision making. Glaucoma trials can take quite a long time—at least 2 years—and so I think we are looking at an estimated 5 years to get the proof of principle and a pilot tool that will then need to be proven over a few years before something can be rolled out.
In parallel to this, we need genetic data to be available. Currently, most of our patients do not come with genetic data, although increasing numbers of people are obtaining it from direct-to-consumer genetic tests. There are already platforms where healthcare services can interact with genetic databases to pull out the information they need, with the patient’s consent. So, this is one possibility.
This type of testing can cost less than £50 and when done on scale might be nearer to £20 per person, which is excellent value. And if that one £20 test can tell you whether you should be screened for glaucoma, which treatment you should have and whether you need treatment at all, it could provide a similar picture for other diseases.
For example, it could also tell you, say, whether you should take statins if you have got high lipid levels. Should you be screened for breast cancer or should it be delayed? And for prostate cancer? This is just scratching the surface.
The £20 test can inform a whole gamut of health-related decisions. It will really become a question of when rather than if you test and obtaining genetic information for all people who interact with healthcare services will become routine.
What are the main challenges?
The people in the aforementioned study were all of European descent and we can predict with a certain degree of accuracy someone’s risk of glaucoma if they are European. This prediction works quite well for Asian people too, but not so well for people of African descent.
If we are to develop a new, better way of managing glaucoma, which means that people get the best treatment for them from the get go and that we spare people who do not need treatment, can we really offer it if it only works in people of certain ethnicities? And I think we all know the answer to that is no.
So, one of the challenges is the need to make sure that we are getting enough data and information on people of all ethnicities. And this is why, having started the Moorfields Glaucoma BioResource, I am pleased to be able to recruit patients from all backgrounds due to London’s very diverse population.
The other challenge is to come up with a better way of managing glaucoma so that expertise is widely distributed and not clustered in certain centres of excellence where there are genetics experts. Such care needs to be delivered by any doctor and to an extent that even patients can understand.
It needs to be accessible and rolled out everywhere, equitably, to people of all ethnicities and backgrounds. And that is the key: we need to make this an equitable innovation.
---
Anthony Khawaja, MA (Cantab), PhD, FRCOphth
E: anthony.khawaja@gmail.com
Dr Khawaja is associate professor at UCL Institute of Ophthalmology and honorary consultant ophthalmic surgeon at Moorfields Eye Hospital NHS Foundation Trust, London, UK. He reports acting as consultant to Aerie, Allergan, Google Health, Novartis, Reichert and Santen and has received lecturing fees from Grafton Optical and travel support from Thea.
References
- Garway-Heath DF, Crabb DP, Bunce C, et al. Latanoprost for open-angle glaucoma (UKGTS): a randomised, multicentre, placebo-controlled trial. Lancet. 2015;385:1295-1304.
- Gordon MO, Beiser JA, Brandt JD, et al. The Ocular Hypertension Treatment Study: baseline factors that predict the onset of primary open-angle glaucoma. Arch Ophthalmol. 2002;120:714-720; discussion 829-830.
- Gordon MO, Torri V, Miglior S, et al. Validated prediction model for the development of primary open-angle glaucoma in individuals with ocular hypertension. Ophthalmology. 2007;114:10-19.
- Khawaja AP, Cooke Bailey JN, Wareham NJ, et al. Genome-wide analyses identify 68 new loci associated with intraocular pressure and improve risk prediction for primary open-angle glaucoma. Nat Genet. 2018;50:778-782.
- Poplin R, Varadarajan A V., Blumer K, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018;2:158-164.
