AI prompts a scramble for healthcare data

Covid-19 has opened the eyes of the healthcare world to the promise of artificial intelligence. Two years ago, AI applications were largely seen as demonstration projects and start-up dreams. There were “nice to have” experiments in triaging mammograms, scheduling operating rooms and monitoring diabetes patients remotely.

The pandemic forced national and commercial health systems to rely on remote medicine and use real-time dashboards to spot looming needs. While neither focused on AI per se, both could be enhanced with its strategic use. Suddenly top executives understood the technology’s potential and bumped it up their list of priorities.

We are now seeing a mad rush to gain access to patient and hospital data and turn AI loose upon it. Last week’s deal that will see Google store HCA’s data and help the US hospital chain develop healthcare algorithms is one example. The UK NHS’s plan to consolidate 55m patient primary care records into a single database is another. Global fundraising for AI health start-ups has risen steadily since the end of 2019 and hit a new record of $2.5bn in the first quarter, says CB Insights.

In some ways, healthcare is following financial services. The 2008 financial crisis forced bankers to invest in better data collection and analysis to improve risk monitoring. The sector then started finding other ways to exploit it.

Healthcare has been slow to the data party, in part because so much of it is collected in ways that are hard to consolidate: in conversations, in different locations and using non-standard measurements and formats. Just having an electronic healthcare record system isn’t enough: it needs to be comprehensive and searchable.

“In a world where data is flowing in constantly [we need] something non-human to manage it,” says Robert Wachter, professor of medicine at the University of California, San Francisco, and author of The Digital Doctor.

Big cloud services providers such as Microsoft and Google are competing for this valuable business by offering healthcare users a growing range of services. Microsoft recently bought Nuance, a clinical intelligence service that translates doctor-patient visits into usable data. In studies, it let cardiologists see 24 per cent more patients a day by cutting administrative tasks.

Google has found potential in prediction: last year, it worked with HCA to build an online portal that combined data on Covid tests, hospital admissions and ventilator use from facilities across the US and used AI to track pandemic hotspots, anticipate surges and warn local officials. The new data deal could take such efforts beyond Covid.

Medical researchers think that AI mining of primary care records could revolutionise work on slow-moving diseases like dementia, Parkinson’s and heart failure. Previously, the need for repeated interviews made large, long-term studies prohibitively expensive and participants dropped out. “The public benefits are indisputable,” says Cathie Sudlow, who directs the British Heart Foundation’s data science centre. “It is impractical to conduct this kind of research without this kind of access.”

Still, medical records include some of the most sensitive personal data, and it should not be shared too easily. A 2019 collaboration between Google’s health arm and Ascension, another US healthcare system, sparked outrage from advocates who feared the tech group would misuse the information. More recently, some efforts to use smartphones to track coronavirus exposures foundered on privacy concerns.

Google says it is simply providing storage and tools to HCA and will not get direct access to the data. The NHS says that identifying details will be stripped out and it will audit users to make sure data is not misused. But privacy groups remain concerned.

Opening access without jeopardising privacy is key. Many imaginative researchers have smart ideas for healthcare algorithms. But they have little value until they are trained using real-world data, and the more of it the better. “I am a big believer that data saves lives,” says Microsoft’s Elena Bonfiglioli.

AI may provide a partial solution: synthetic data sets that mimic real ones with slight changes. A 44-year-old with a fatal blood clot might become 43 and live nearby. Syntegra and Vanderbilt University are building just such a diabetes patient database to share with commercial researchers.

The current scramble for healthcare data is cause for concern. Big players have already amassed outsized positions in social media data. Healthcare must be different. If we end up with a few gated kingdoms, we will all be the poorer — and sicker — for it.

brooke.masters@ft.com

Follow Brooke Masters with myFT and on Twitter