eyeforpharma is now Reuters Events - LEARN MORE
The Patient Summit USA 2019
Attend this event to get real case studies which are moving the needle and changing the relationship pharma has with its ultimate customer - giving you practical techniques across the value chain (R&D and commercial) on inspiring cultural change, delivering ROI, tackling enrolment problems, boosting adherence and improving health.
Unleashing AI’s potential in clinical trial recruitment
Multiple innovations using AI are transforming the trial recruitment process
The onerous cost of clinical trials in time, money and effort is well documented. The average clinical trial process lasts between 7.5 and 12 years, studies estimate, with costs ranging from $161M - $2.6B per drug. Just 14% of clinical trials are successful and only one in ten drugs entering Phase 1 ends up being approved by the FDA.
Recruitment is a significant source of the problem. According to Christina Busmalis, director of global life sciences at IBM Watson Health, 
80% of clinical trials do not finish on time and the reason for this, in 86% of cases, is that they do not meet target recruitment on time. “It’s a massive problem across the board,” she explains, “which leads to delays and trial costs that can run into the multi-millions.”
80% of clinical trials do not finish on time and the reason for this, in 86% of cases, is that they do not meet target recruitment on time. “It’s a massive problem across the board,” she explains, “which leads to delays and trial costs that can run into the multi-millions.”
The growing complexity and volume of data involved in trials is a chief reason for this, says Michelle Longmire, CEO and co-founder of California startup, Medable. The clinical trial world has gradually reached a point over the last decade where the number of patients and the amount of data has increased to such an extent that it has become impossible to deal with it manually.
“When you look at clinical trial protocol, the inclusion/exclusion criteria and the variety of parameters one needs to meet to be eligible, it is a significant challenge for both patients and companies,” says Longmire. “Whereas a clinical trial in the past would have had 20 variables, it will now have an average of 150.”
For Robert Jan-Sips, chief technical officer at myTomorrows, a Netherlands-based treatment delivery company specialising in expanded
access treatments and compassionate use programmes, another of the main hurdles in successful clinical trial recruitment is the ‘information gap’, which often results in a lose-lose situation where patients cannot find clinical trials and pharma companies are unable to find patients.
access treatments and compassionate use programmes, another of the main hurdles in successful clinical trial recruitment is the ‘information gap’, which often results in a lose-lose situation where patients cannot find clinical trials and pharma companies are unable to find patients.
“Information is scattered across 17 different registries,” he says. “Patients have difficulty finding a concise overview of trials that are recruiting, many are out-dated and most are only available in the English language. Furthermore, it is very difficult for patients to understand the medical language in order to enroll.”
Building new solutions
A new approach is needed and Artificial intelligence has the potential to disrupt and transform the clinical trials landscape, ultimately speeding drug discovery and cutting costs. One of its most promising applications is in streamlining the recruitment process for doctors and pharma by extracting information from patient records and matching it to ongoing trials.
Several novel approaches are under development.
Medable is focused on building a unified platform, ‘the Human Digitome’, for clinical trial execution, enabling patient generated data to drive clinical research. “We refine and unify the data stream – medical record data, claims data, genomics data and lifestyle data - around patients,” Longmire explains, “and then use that to apply intelligence to match to treatment programmes.
“It’s extremely powerful. It is improving the process of identifying patients and the screening capability. We are the only enterprise cloud purpose built for life sciences with dedicated AI and machine learning,” she says. Medable’s technology is currently being used by over 750,000 patients across five continents, she says.
Matchmaking patients with trials
At IBM Watson Health, the initial focus was to build a solution that helped doctors find the right clinical trial for their patients. This involved a project with the Mayo Clinic in Minnesota during which the company ingested data from clinicaltrials.gov’s database of privately and publicly-funded clinical trial studies around the world and, using AI, combined it with specific information from patient health records to allow doctors to identify relevant trials.
Building on the success of that project, which went live in June 2016, IBM Watson Health started talking to pharma about the issue. “Many organisations were coming to us saying they were having trouble finding patients for their trials,” explains Busmalis.
“We conducted a study with Novartis and Highlands Oncology in Northwest Arkansas. Instead of taking data from clinicaltrials.gov we used Novartis’ clinical trials protocols. This is pages and pages of information which includes inclusion and exclusion criteria. We injected 10 of these into our system and merged them with patient data from Highlands.
“This gave two perspectives. One gave Highlands a detailed view of which patient would fit which trial and the other gave Novartis an insight into how many applicable patients would be relevant to their trials.” The results of the study, which were published at ASCO in 2017, showed that screening time was reduced by 78% from 1hr 50mins to 24mins.
Democratising information access
Among the major challenges, explains Busmalis, were that clinical trial protocols were not written in a way that was easily machine readable and that inclusion and exclusion criteria were either too big or too small. So, the company worked with Novartis to adjust the protocols.
IBM Watson Health has since been working on a platform based on IBM MarketScan claims data in the US market, which it plans to bring to market later this year, that optimises the protocol authoring and development using AI and real-world data.
Meanwhile the priority for myTomorrows has been to set up a platform that engages directly with patients, says Sips. “We use AI to essentially translate medical concepts into lay language, so we try to democratise information access by opening up the information source. This enables us to provide information about treatment options to patients with unmet medical need, and their doctors, about treatment options worldwide and then to facilitate access to medicines in development.”
The company has an automated system that tries to predict the terms an average person would use to describe a disease or a medical need. “We are trying to be findable by patients who are looking for treatment options in whatever language they might use,” says Sips. “So, for example a patient might be diagnosed with malignant pancreatic neoplasm, which in lay terms is pancreas cancer but it would not be called that in a clinical trial.
“So, when a patient finds us on Google and types ‘pancreas cancer’ into our system it will automatically translate that into ‘malignant pancreatic neoplasm’ and also expand the search into all sub-forms of the disease. Based on that information all relevant clinical trials, drawn from clinicaltrials.gov and a variety of European registries will then be listed.”
Patient and physician decide together
Armed with a list of relevant clinical trials, patients are then offered the option of reaching out to myTomorrows’ medical team to guide them through the process of narrowing down the options on the list based on patient specific criteria. This requires their documentation - letters, formal diagnoses and whatever other information they might have – to be entered into the myTomorrows machine-learning system to extract information on what treatment has already been received, any genetic information, and other co-morbidities to filter down the list of trials.
“This is then reviewed by an in-house medical professional,” says Sips, “to make sure it is not solely an automated process. This information is then communicated to the patient who either makes contact with the treating physician or allows us to contact the treating physician on his/her behalf.”
myTomorrows then takes a step back from the process, as the company’s role is merely one of providing guidance and information on clinical trials and expanded access programs rather than advising and consulting on treatment specifics. The company has active partnerships with patient advocacy groups whereby it forwards patients looking for information from reliable sources and they return patients looking for treatment options.
This offers an unbiased overview of all treatment options for the patients while facilitating a shared decision-making process between the patient and the physician, says Dennis Akkaya, head of corporate development at the company.
“That sort of concierge service does not exist in the majority of pharma companies and if it does, they will obviously try to steer patients to their own clinical trial or programme. We give the best overview of options and facilitate the steps needed to get to that.”
Towards siteless trials
Busmalis believes that AI will ultimately empower patients and transform the way clinical trials are planned and executed. “Using AI and data-driven decisions can help us get to the point where clinical trial execution is faster and cheaper.
“I think patient empowerment will change things a lot and at some stage in the future patients will be empowered to find trials that best suit them and to work directly in a different way with organisations through a trial perspective.” She believes the landscape will shift towards siteless trials as more and more data from wearable devices is harvested and analysed, and that adaptive trial design will enable a more robust and flexible process.
“From an IBM point of view, AI stands for augmented intelligence, not artificial intelligence. All of our solutions are about empowering someone to do things better and faster. An AI system will guide through a process more efficiently and augment the capabilities of humans. You have machine learning and humans – you bring those together and that’s quite a partnership.”
Since you're here...
... and value our content, you should sign-up to our newsletter. Sign up here
The Patient Summit USA 2019
Attend this event to get real case studies which are moving the needle and changing the relationship pharma has with its ultimate customer - giving you practical techniques across the value chain (R&D and commercial) on inspiring cultural change, delivering ROI, tackling enrolment problems, boosting adherence and improving health.