IF YOU WANT to see a spiffy new protein structure, ask an AI. If you really need a fax machine, try a doctor's office. It will be somewhere in the corner under a clipboard with a variety of paper forms attached to it. Not true for all doctors, or all health-care systems, but true enough for a wry grin in many; the sector's digital transformation has been patchy at best.
Economists think technology has been responsible for between 25% and 50% of growth in health expenditure in OECD countries over the past 50 years, growth which has seen the sector's share of GDP grow relentlessly. In many of those countries it has achieved much. And yet, after decades of costly effort, stories still abound of incompatible IT systems, confidentiality breaches and paper records that need to be held on to in parallel to electronic health records. Is there any reason to think that AI can really sort this out?
There is. And it is offered, in part, by the sheer size of the problem. America spent $4.5 trillion on health care in 2022. That was considerably more than would be expected in comparable countries, and administrative costs accounted for 30% of the excess. Trillion-dollar opportunities can attract the attention of very large companies, such as America's tech giants. And those companies think their large language models (LLMs) and other big self-supervised-learning systems offer new tools particularly well suited to the job. The fact that the biggest companies in AI see health care as a place to compete is a genuine cause for optimism.
Among Google's health-care ventures is Med-PaLM2, a health-specific LLM being developed to answer health-care questions and summarise information during patient handoffs or staff shift changes. Amazon's investment in Anthropic, which provides an AI assistant called Claude, was in part intended to bolster what the company can offer in health care. Chinese giants are interested too. In 2022, a report by McKinsey, a management consulting firm, argued that using AI to predict diagnostic outcomes and support clinical decisions could create about $5 billion in economic value in China.
Then there is Microsoft, the tech company keenest on growth through acquisition. In 2021 it paid $19.7bn for Nuance, based in Burlington, Massachusetts, a company which makes AI that helps doctors with administrative tasks such as creating clinical notes and electronic health records.
The AI-backed voice transcription offered by Nuance and others, such as Amazon's Healthscribe, is a big deal. Harpreet Sood, a doctor who used to be NHS England's Associate Chief Clinical Information Officer, says the technology has been game changing for him. It saves him four to six minutes per patient, which means two to three hours a day. His patients have noticed that he is looking at them more and his screen less; he and they both like it that way.
Other roads to better efficiency abound--especially in America, which is not just the world's biggest health-care market but also a peculiarly inefficient one. One currently popular way of trying to improve efficiency and outcomes is the creation of hospital "command centres". The idea is something like an air-traffic-control system in which a wall of screens gives up-to-date information about key metrics such as bed availability, resource use and the status of patients across the hospital. Parts of that ensemble are replicated on tablets and mobile devices used by staff on the wards. Not only can such systems see problems as they happen, they can anticipate bottlenecks to come.
These systems are now found at more than 200 hospitals around the world. The command centre at Johns Hopkins Hospital in Baltimore has made transferring patients between locations 60% faster, cut the waiting times for emergency treatment by 25% and reduced the time in post-surgical beds by 70%. Tampa General Hospital has reported a gain in efficiency worth $40m since launching a command centre using 20 AI applications.
Another AI-led vision of the future involves keeping people out of hospital--or looked at in another way, taking hospitals to the people. Britain, which has a relatively low number of hospital beds, is an eager adopter of "virtual wards", which allow patients to be transferred from hospital to recover in their own homes with the aid of monitoring devices such as a tablet or a blood pressure cuff. In 2023 Britain reached 10,000 virtual ward beds. As yet, these systems are not showing as much benefit as they might. A recent study of one such scheme found them more expensive than hospital treatment. But AIs might help.
Doccla, one of a number of British virtual-ward tech firms, says it is working towards integrating LLMs into its clinical workflow. The vision is to bring together data from wearable devices, patient records and call transcripts into a system that provides a "co-pilot" which can keep the health-care provider abreast of what is happening among the patients. Such capacities will help not just in virtual wards, but across the system. They should let doctors get to grips with vital health information that can be buried in plain sight.
If this is to come about, though, systems will have to adapt, and that can prove hard. Robert Wachter of the University of California, San Francisco, and Erik Brynjolfsson, at Stanford, have recently argued that humans are generally unable to implement the profound changes in "organisational structure, leadership, workforce or workflow needed to take full advantages of new technologies, at least at first."
Countries with less established health infrastructure but good digital connectivity may lead the AI way
Take the decentralisation of care. To the extent that AI supports good decision making, its tendency will probably be to move care away from the centre and towards the edge: to allow more diagnosis in general-practice, perhaps through smarter instruments; to move other decisions to pharmacies; to increase the access patients have to advice and monitoring in the home. But patients often have set expectations about seeing a doctor in person, or having a nearby hospital.
It may be that countries which are still developing their health systems have a better chance at "reimagining the work" than those where institutions and patients are already set in their ways. Dr Sood thinks countries with less established health infrastructure but good digital connectivity may lead the AI way--he points to India, Kenya and Indonesia. These nations may be better able to build their systems around the technology patients are already using, for example by providing care on platforms such as WhatsApp.
No one should think that what AI offers in terms of greater efficiency can be taken as a given. There will doubtless be schemes that overpromise and overcharge. There will be a constant need for evaluation, supervision and updating. It is not just that AIs can "drift". Some of the advantages companies want their systems to offer, such as keeping abreast of research in a way doctors often cannot, require the systems to change over time, and this means retraining and recertification to ensure efficacy and safety.
But if constant updating and change bring challenges, their upside is clear. And the changes institutions need to make when adapting to new technologies should be made easier if those technologies can change too. That ability to both drive change and make it easier to cope with is one of the prizes that AI should, at its best, be able to offer.
None of those prizes will be won easily. Getting the most out of AI will require institutions which find change hard to undertake a lot of it. It will find regulators under proper pressure to ensure safety facing new challenges in terms of the scope of the technology and the speed at which it changes. And it will need economic incentives that realise the technology's potential to save costs and lives. But if people can bring these shifts and reforms about, the machines will pay them back bountifully. #
Editor's note: Due to an editing error, the print edition of this piece mis-states the proportion of America's health-care costs due to administration. That has been corrected in this online version. We regret the error.
This article was downloaded by calibre from https://www.economist.com/technology-quarterly/2024/03/27/can-artificial-intelligence-make-health-care-more-efficient
Previous | Articles | Sections | Next |