IBM’s Watson artificial intelligence supercomputer

Medicine looks to Big Data for targeted treatments

Image credit: IBM

Data mining offers new hope for precision cancer drugs

Graham Silk, the head of charity Empower: Data4Health, counts himself lucky to be alive and he wants to change the UK’s healthcare system so that everyone else winds up with the same degree of ‘luck’. Diagnosed with leukaemia in 2001, his prospects were not good. Only around 10 per cent of people survived five years with his condition.

“I got lucky because of many factors, the main one being that I was in the right place at the right time to get access to the drug Gleevec. I was lucky enough to be one of only 40 people out of 130,000 with leukaemia in the UK to get on to the clinical trial. The drug saved my life,” Silk explained at a conference last autumn organised by Google’s artificial-intelligence subsidiary DeepMind.

Silk’s form of leukaemia matched the profile of disease Gleevec was designed to treat. It does not treat other forms of the disease with the same name, and that is far from unusual in cancers – a disease for which variation is the norm.

A study published in Science in 2013 by researchers from the Johns Hopkins Kimmel Cancer Center in Baltimore, Maryland found that colon, breast, brain and pancreatic cancer tumours contain as many as 70 genetic mutations. Individual mutations found in one tumour will be seen in 10 per cent or fewer patients with a cancer of the same organ: a drug that attacks the key mutation in one of those cancers will mostly likely be ineffective against the other 90 per cent. The heterogeneity of cancer tumours goes a long way towards explaining the huge variation in survival rates among patients using a single, common treatment.

“Many diseases we think of as being common will, in the future, be seen as rare,” says Silk.

On top of the genetic variation in tumours is the variation in patients’ own genomes. This variation accounts for an estimated 20 to 95 per cent of the difference in response to treatments, according to a 1997 study by Kristine Crews and colleagues from St Jude’s Children’s Research Hospital in Memphis, Tennessee. A drug that works well in one patient can attack healthy cells of another, leading to painful and dangerous side effects. Even within the same patient, as tumours spread they mutate and become resistant to treatments that successfully attack others nearby.

One answer is to tie the treatment to the genetic make-up of the patient and the cancers. In the UK, the overall approach goes by the name ‘stratified medicine’ – some clinicians in the US call it ‘personalised medicine’. However, a report by the US National Research Council of the National Academies argued for the term ‘precision medicine’, as the group argued that medicine is not becoming truly personalised.

The US government has embraced the title precision medicine. In his State of the Union Address at the beginning of 2015, President Barack Obama launched the country’s Precision Medicine Initiative to “bring us closer to curing diseases like cancer and diabetes”. In December, US Congress passed the 21st Century Cures Act. The legislation backs billions of dollars in funding for precision-medicine research.

“You can take your pick of which term to use,” says Silk, but the aim of the charity that he set up with Professor Charlie Craddock CBE (the University of Birmingham clinician responsible for organising Silk’s access to the Gleevec trial) is to expand the use of precision medicine in the UK to a much wider range of patients.

Although the pharmaceutical industry has made much of its money out of one-size-fits-all ‘blockbuster’ medications, companies are changing their approach to fit the idea of precision medicine. Part of the reason is that the US is switching the way it pays for drugs. Outcomes determine how much the supplier receives.

“A lot of the payments in the future will be outcome-based. Technology will enable that,” claimed Johnson & Johnson digital innovation lead Laurent Morlet at the Genesis 2016 conference in London in early December.

Menelas Pangalos, executive vice- president of AstraZeneca’s innovative medicines unit, says: “By 2023 we will have over 50 drugs enabled by a personalised diagnostic. It’s part of our inherent DNA now. We don’t launch unless we understand which patients will respond to the therapy.”

The ideas behind precision medicine resurrected the ovarian cancer drug Olaparib. The drug is not just used on patients who have mutations in one of a pair of genes that go under the name BRCA. Those with prostate cancers who show similar mutations are also trialling it.

“The drug was terminated in 2012,” Pangalos says, and AstraZeneca wrote off $285m in development costs. Then in the same year CEO Pascal Soriot took over. “When Pascal came into the company, he was very surprised it wasn’t in Phase III [trials],” Pangalos claims.

Although the idea of choosing treatment based on patient characteristics looks promising, getting to the point where luck is taken out of the equation for all patients is a long way off.

In the short term, data is the key, Silk says: “One of the most important factors in all of this is the ability to link together hospitals within the NHS and to utilise their patient data, which in this case is quite literally the lifeblood to enable better health delivery.

“The ability to look at the cellular levels of the disease means you can get to the cure of it,” Silk adds.

Groups in the US began to build databases of genetic and patient profiles and treatment outcomes to try to find out what works for which cancer. Crunching that data calls for massive amounts of compute power. IBM’s Watson artificial-intelligence (AI)supercomputer, famous for beating two human champions of the US game show ‘Jeopardy!’ in 2011, now features prominently in a range of genomics programmes as part of a push into medicine.

“Some of the first solutions after ‘Jeopardy!’ were focused on healthcare,” Cory Wiegert, vice president of product management for IBM’s Watson Health, explained at the autumn 2016 World of Watson conference in Las Vegas.

In 2013, IBM started working with healthcare institutions such as the MD Anderson Cancer Center in Houston, Texas to work on data mining of genetic data and medical records. In June last year, IBM signed a deal with the US Department of Veterans Affairs (VA) to help treat combat veterans with cancer. The VA is ramping up DNA sequencing, with the aim of taking samples from 800 veterans per month over two years, feeding the results to Watson.

This was followed by the start of a roll-out across the US by IBM and Quest Diagnostics of a genomics analysis system based on Watson. The service uses laboratory sequencing and analysis of a tumour’s genetic make-up in the hope of revealing mutations susceptible to targeted therapies. Watson compares the mutations identified by sequencing to those mentioned in medical research papers, clinical studies and rules created by oncologists. Through this partnership, Watson for Genomics alone is sucking in around 10,000 scientific articles and the output of 100 clinical trials monthly.

IBM is far from alone. In the US, the Huntsman Center at the University of Utah has taken advantage of a state-wide health database to assemble its own genomics-driven database. Starting in 2014, the American Society of Clinical Oncology (ASCO) started to build its own CancerLinq system, using experience derived from a 170,000-record prototype. By October 2016, the ASCO team had added one million patient records to the CancerLinQ database and recruited 1,500 oncologists working across 70 practices in the US.

In the UK, the government-owned group Genomics England launched a project in 2013 to sequence genomes of 70,000 patients with cancers or rare diseases. AstraZeneca aims to push its collection of genomes to two million after signing a deal in April 2016 with Human Longevity, a company formed by shotgun DNA-sequencing pioneer J Craig Venter. Pangalos says: “It’s not so much sequencing two million genomes, but enabling us to deal with the massive amounts of data in electronic health records. If we can handle two million genomes, we can handle all the data that comes from outside our world. Creating a genomic database married up with electronic health records will help tell us what biology is driving health processes.”

It’s not just genomes. “What’s living in your gut has an impact on side effects. It might also govern the efficiency of treatments,” said Mike Burgess, head of exploratory clinical and translational research at Bristol-Myers Squibb.

Bristol-Myers Squibb signed a deal with gut-flora biomarker specialist Enterome in November to develop tests for microbiome samples that could be used to determine whether patients are compatible with the company’s antibody-based cancer drugs.

Although the databases will provide doctors and medical researchers with more clues as to what might be used to treat diseases, they will not lead to immediate medicines for all. In many cases, the treatments will involve combinations of existing drugs, partly because attacking cancers on multiple fronts can prevent resistance building up. They still need to be tested together to ensure they do not lead to harmful side effects.

Indications of success with combinations or biomarkers that point to new drugs will drive creation of many more trial programmes. Pharmaceuticals companies hope the trials will be quicker, more focused and, on average, more successful. In the short to medium term, many sufferers may not have a more effective treatment than palliative care.

“That is the way precision medicine will go. It takes time, money, research. And it takes some genius and bravery,” says Silk. “The data can help inform and educate. That’s the way it will go increasingly if you want to see survival rates increase.”

In the UK, doctors and developers of treatments face the problem of convincing the public to accept that some of their medical data will be used to drive drug and treatment research, much of which is carried out by commercial organisations. A quarter of the respondents to the Wellcome Trust’s ‘One Way Mirror’ survey said they did not want their data to be used for commercial drug development. The name of the report itself came from one patient’s response: “It’s a one-way mirror: they know everything about you, but we don’t know what they’re doing with that [information].”

In December 2016, pressure group MedConfidential claimed Health Secretary Jeremy Hunt had broken a promise to allow patients to opt out of having anonymised versions of their records released to research. MedConfidential and other privacy campaigners argued that the anonymisation techniques proposed are not sufficient to prevent people being identified through a combination of available records.

“People have justifiable worries over privacy,” Silk argues, but that should not stop the exchange of data that could drive health research into the future. “No-one is saying they want to compromise data. As a patient you are screaming for these treatments to be ready. And if it’s not them, it’s someone they love. I think we should bring these millions of health records together. You could argue for it forever and someone will make a very good argument against. In the end, we just have to make a judgement.

“Pharma companies are looking for market share. People find that to be anathema because ‘my data is being used in a quid pro quo arrangement’. When Novartis developed Gleevec, the outcomes for the 50 or so people in the trial made their share price rise considerably. They saved my life. I thought it was a pretty good deal.” 

Identifying the disease you are going to have

Precision medicine is only in its initial phase, but governments and medical organisations are looking at data as a way of reducing treatment costs. One direction is prevention – changing habits in the hope of reducing the number of people who contract chronic diseases.

There is also interception. “Don’t just detect a disease before clinical manifestation. It is possible to intervene before the disease manifests itself,” says Laurent Morlet, Johnson & Johnson’s digital innovation lead.

This is different from prevention, Morlet stresses. The aim is to pick up signs of the disease well before the worst symptoms hit: “Some mechanisms will have started. Can those mechanisms be interfered with and can they be stopped? It opens up areas in medicine. Yet it presents a challenge: how do you intercept people who are not yet sick?”

Interception will be based on the same data-mining techniques as those proposed for precision medicine. “Diagnosis will be based on panels of biomarkers that will make you think that the [disease] process has started.  But as they are not patients they will have to be found in a different way.”

Jennifer Laird, senior director of search and evaluation at Eli Lilly says the medical community will have to be sure that it is prescribing drugs to people whose biomarkers indicate they have a disease but show none of the symptoms and that “there are big ethical issues. You have to care for many different aspects before you even get to who pays for the treatment.”

Laird points to vaccinations and the campaigns around them as providing a guide to what may happen with an interception policy. “Parents decide not to vaccinate their children because they think ‘what’s in it for us’ or they have other concerns.”

Morlet says: “The industry will have to prove that when a panel of biomarkers has been identified you will have the disease in 10-20 years.”

The process of demonstrating reliability will be to mine data extensively to “find reproducible results and that the correlations are there,” he adds.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles