Moorfields Eye Hospital and DeepMind

In August 2018, researchers from Moorfields Eye Hospital, University College London and DeepMind published results in Nature Medicine showing that an AI system could analyse eye scans and recommend the right referral for more than 50 eye diseases.¹

The system worked from optical coherence tomography (OCT) scans, a common 3D imaging technique. It recommended the correct referral more than 94% of the time, a performance on a par with eye disease experts who reviewed the same scans.¹² It was trained on 14,844 scans from about 7,500 patients with sight-threatening retinal diseases.¹

Why the results mattered

Moorfields’ Pearse Keane explained that the technology was “designed to prioritise patients who need to be seen and treated urgently.”² Eye clinics face rising demand, and some conditions can cause irreversible sight loss if treatment is delayed. A system that could triage scans quickly and accurately could help clinicians see the most urgent patients first.

The partnership terms also stood out. If clinical trials succeeded and regulators approved the product, Moorfields could use it free of charge across its 30 UK hospitals and community clinics for an initial five years.²

The road from research to routine care

The 2018 study was research, not deployment. As the researchers said at the time, clinical trials and regulatory approval would be needed before wider use.² In September 2019, Moorfields transferred its agreements from DeepMind Technologies to Google Health.³

As of our last check, we have not found published evidence that the system is in routine clinical use at Moorfields. Its own updates page lists trials and approval as next steps and has no later update on either.³ That is not a criticism of the research, which was strong. It is a reminder of how long and uncertain the path is from a landmark paper to a tool in everyday care.

The contrast: the Royal Free

DeepMind’s first NHS partnership told a different story. In July 2017, the Information Commissioner’s Office (ICO) ruled that the Royal Free London NHS Foundation Trust had not complied with the Data Protection Act when it provided about 1.6 million patients’ records to DeepMind to test Streams, an app to alert clinicians to acute kidney injury.⁴

The ICO found that patients “were not adequately informed that the processing was taking place.” The Information Commissioner, Elizabeth Denham, said: “the price of innovation does not need to be the erosion of fundamental privacy rights.”⁴ The trust was required to establish a proper legal basis for the project and commission an audit of the trial.⁴

Lessons for leaders

1. Separate the research result from the operational result. A strong study is the start of the journey. Plan and fund the trials, approvals, integration and change needed to reach patients, and be honest about how long that takes.
2. Negotiate the value you will keep. Moorfields secured free use of the eventual product. Public bodies that contribute data and expertise should agree in advance what they will get back.
3. Earn data trust before you need it. The Royal Free ruling shows that a worthwhile clinical goal does not excuse weak transparency. Be clear with the public about what data is shared, with whom, and on what legal basis.

Questions to ask in your organisation

1. For our most promising AI pilot, what stands between the results and routine use, and who owns each step?
2. In our data partnerships with technology companies, what do we receive in return, and is it written down?
3. Would the people whose data we share recognise and accept how we are using it?