With its ability to ingest huge amounts of unstructured data and to see patterns, IBM’s Watson machine learning platform has the potential to revolutionise medicine. Martin Cooper AMBCS meets its makers and reports.

Here’s a question: what won a million dollars on a US quiz show, can dream up recipes like a Michelin chef and - if it’s makers have their way - will also help doctors spot cancer?

The answer is Watson - IBM’s natural language and machine learning platform. Outside the tech world Watson is probably best known as the machine that won $1m on the US quiz show Jeopardy! Despite the project’s high profile, it wasn’t just a vanity project, explains Dr Matthew Howard, European Lead for IBM Watson Healthcare. Rather, it was a valuable proof of concept.

Jeopardy!, for the unfamiliar, is a game where contestants are provided with the answer to a question and are challenged to guess what the question might be. To be good at the game, Dr Howard explains, contestants need good general knowledge and also the ability to understand unstructured information. This latter part is where Watson was designed to excel.

‘The game show,’ Dr Howards says, ‘allowed us to demonstrate the three key tenets of a cognitive solution. That is a solution that: understands natural language, a solution that can ingest a corpus of knowledge and a solution that gets better - or one that can be trained to get better - at how it answers questions.’ In short, he says, one that has machine learning.

Watson in healthcare

So why did IBM choose to turn Watson’s attentions on the healthcare sector? Along with clear and unmet need, the answer, it seems, comes down to data - or rather an abundance of unstructured data.

‘Anything up to 70 per cent of a medical record can be unstructured, ‘ Dr Howard says. ‘It’s referral notes, discharge letters, it’s radiological reports. So, using cognitive solutions, solutions that can be trained to read and to understand medical information... to structure that information very quickly... That has enormous power. You can generate a prognosis very quickly from 50 pages of referral letters.’

Part of a team

Watson isn’t, however, designed to be some kind of monolithic machine that ingests patients’ records and issues diagnoses, so rendering doctors redundant. Indeed, in many ways, Watson is quite the opposite. For starters, there is no single Watson.

Rather there are many - each is a discrete, cloud-based installation of the underlying cognitive technology. IBM is deploying Watson Healthcare systems in the US, internationally and it is in the process of bringing it to Europe. These systems aren’t generally bought outright either, clients tend to purchase the Watson capability.

Watson, within the healthcare context, is also envisaged very much as an augmentation - as an assistant to expert physicians. ‘If you think of the volume of information that a physician has to understand, to manage and to keep on top of, it’s huge,‘ Dr Howard explains. ‘And we have very high expectations of these individuals. So, a good use for a Watson technology is in clinical decision support.’

Watson, he says, is best placed to use its cognitive abilities to ingest the best practices, the guidelines, textbooks, case notes, and to surface that information and to help the physician make a decision. This might be finding the appropriate literature about how that patient could be treated or finding the exact guidelines about drugs. And it’ll do this instantly.

‘For cancer patients there’s a multidisciplinary team who reviews all the information around a patient,’ he says. ‘In this setting systems like Watson can be very powerful. It can present all the information in one place and it can make a recommendation which can then be reviewed by physicians. This ability to consume vast amounts of data and to make the data contextually relevant - to the patient and the physician - is very powerful.’

And indeed IBM has a solution called Watson Oncology and it specialises in clinical support. Physicians can use it as one of the inputs into their treatment plan.

An appointment with Dr Watson

From a patient’s perspective the presence of Watson promises clear benefits but, critically, IBM appears highly respectful of medicine’s heritage and history. ‘The patient-physician paradigm is paramount and it should remain paramount,’ asserts Dr Howard. ‘There’s a lot of other things a physician is doing when they talk to
a patient.’

‘A benefit that patients may not immediately perceive’, he says, ‘is the ability to improve clinical consistency. Watson helps clinicians understand very quickly how a patient may fit into a pathway. This means that, potentially, all patients get access to the very best care and they get improved outcomes.’

Cognitive solutions like Watson may also have applications within the sphere of prescribing drugs. ‘One of the use cases we’re seeing is: what is a patient eligible for, particularly in cancer treatment,‘ explains Dr Howard.

‘Cancer medicines can be very expensive and so knowing the approved drugs in your country is a powerful tool. So, helping physicians keep abreast of that information is very important. It also protects the patent as you don’t want to prescribe a drug that’s not reimbursed by your health authority.’

Helping patients understand

Visiting a physician and visiting a hospital can be very unsettling for patients. The medical world speaks its own language and its processes can appear rather Delphic to the uninitiated. What’s happening to me? Why? Diagnosis throws up many perfectly reasonable questions - questions physicians may not have time to answer during a consultation.

‘The beauty of systems like Watson’, Dr Howard says, ‘is that the information’s outputs can be very consumable. So, the physician can generate information for the patient. ‘Patients are very engaged with their care these days - people want to know more about how and why they are being treated. So, giving people the key literature around why a decision has been made is important.’

Another concrete example of how cognitive solutions may be felt by patients is a reduction in the need to google their ailments. Search engines can provide a wealth of medical information to people but there’s no guarantee that that information is correct or up-to-date. By extension, self-diagnosis through Google is a risky business. The information provided by a search engine is, in a way, uncontrolled.

Cognitive solutions like Watson do, however, provide physicians with a means of helping patients understand their diseases more readily. ‘In a way’, Dr Howard says, ‘the technology can improve patient care through the exchanges of information.’

Personal wellness management

More tantalising still, it may not be apples that keep doctors away in the future, it may be cognitive solutions.

‘On a macro level, there’s a big opportunity in population health - using data to spot patterns, to spot those patients at most risk and intervene with those patients,‘ Dr Howard says. ‘It’s a different model. Moving from disease management and helping people once they become sick. Instead it becomes about using data to promote wellness and disease prevention.’

‘Today’, he explains, ‘many healthcare systems are effectively sickness management processes. ‘The challenge,’ he says, ‘is managing wellness. One of the nice things about cognitive solutions - if we go beyond disease treatment advisors - is that the same technology that lets you understand the medical record is also very good at interacting with individuals. This means we can build much better wellness coaches, applications that people can interact with, applications that can help people manage their fitness and their diet.’

Looking further forward, he believes, there’s an opportunity to enrich those applications with, say, information about blood sugar levels and understand if there’s a diabetic risk. ‘So,’ he says summarising, ‘there are huge opportunities between cognitive and cloud technologies and the fact that everybody carries a smartphone.’ It’s just a matter of integration.

Governance of data

For Watson to achieve its potential and to continue in that central tennet of machine learning - to keep getting better - the system needs access to huge amounts of data. And providing the system with the information to learn from and to work on is a challenge in itself.

‘On our Health Cloud,’ Dr Howard says, ‘IBM now has over 300 million lives of anonymous data that we can use to improve our solutions. It’s anonymously extracted data, mostly US data.’

The fact that the data is American in origin isn’t accidental or an act of patriotism by an American company. The US, Dr Howard explains, ‘has a single data privacy framework which makes it a simpler place to do this type of work. So through acquisition, we have an incredible repository of data.’

Doctors with data infographic