Just as artificial intelligence (AI) becomes an ever-more common part of drug discovery, its potential role in clinical trials is slowly becoming more visible, too. Efforts to improve trial recruitment, efficiency and decision-making are underway at companies of all sizes as organizations look to better the oft-daunting odds of clinical success, industry executives told BioWorld.
"AI has the potential to transform key steps of clinical trial design from study preparation to execution towards improving trial success rates, thus lowering the pharma R&D burden," IBM scientist and inventor Stefan Harrer and colleagues recently wrote in Trends in Pharmacological Sciences.
Particular areas in which the application of AI might improve pharma's track record include patient cohort selection and the recruitment of patients best suited to a particular study. Failures in those areas, "paired with the inability to monitor and coach patients effectively during clinical trials, are two of the main causes for high trial failure rates," they wrote.
At contract research giant Covance Inc., for instance, staff are applying AI-driven analyses to massive datasets to identify not just patients eligible for certain trials, but also criteria for helping identify those most likely to respond to trial recruitment efforts and complete the studies in which they're enrolled, Jonathan Shough, chief information officer for the Princeton, N.J.-based business told BioWorld. Using AI with a large de-identified dataset sourced from the diagnostic side of its corporate parent, Laboratory Corporation of America Holdings (Labcorp), "we can better identify those potential subjects with that wider-net approach," he said.
AI and machine learning (ML), a branch of AI employing algorithms to make predictions or decisions without prior instruction, are also helping the company get "beyond the noise" in data to focus on signals that are important for patients' safety and determining next actions on the clinical and financial fronts, he said.
Shough described today's applications of AI and ML as a "compound addition" to the traditional work of running trials. In the not-too-distant future, though, elements of the work, such as pharmacovigilance, may advance toward AI-driven autonomy, with human involvement reserved for verification of findings, he said.
Wilmington, N.C.-based Pharmaceutical Product Development (PPD) LLC, another major global contract research organization, has also made strides toward adopting AI. In February, it signed an exclusive agreement with Beijing-based Happy Life Tech designed to speed patient recruitment at high-performing sites and to more effectively incorporate Chinese patients into global trials.
"We will intelligently leverage real-world data in China to recruit patients with a high potential for participation in our clients' studies and to generate real-world insights needed for market access and product uptake," said PPD chairman and CEO David Simmons.
Finding the right patients
Oxford, U.K.-based Sensyne Health plc, is focusing on using ML to analyze vital signs and other patient data in an effort to aid its life sciences and pharma partners. Using the company's System for Electronic Notes Documentation (SEND) app deployed with Oxford University Hospitals NHS Foundation Trust, for instance, it's looking for novel subpopulations within specific diseases. Success, built on its strategic research agreements in Oxford and with five additional NHS trusts, could enable smaller and more targeted trials, as well as more specific enrollment criteria and endpoints.
For example, instead of thinking of heart failure (HF) as a monolithic indication, Sensyne's team can apply its approach to reveal groups that traditional diagnostic systems can't, Rabia Khan, Sensyne's chief of translational medicine told BioWorld. They're doing exactly that as part of a two-year collaboration with Bayer AG announced July 31. Sensyne's team will apply its algorithms to analyze 3 million medical records, with the aim of identifying disease subtypes that support patient stratification in phase III trials in cardiovascular disease. (See BioWorld, Aug. 1, 2019.)
"One of the things we know is that the way clinical trials are designed currently," she said, is that they "don't capture the real- world scenario." A better understanding of people with HF built through analysis of anonymized patient records – including vital signs, lab tests, prescription information and genetic analyses – could change that.
Drug "resurrection" is another interest at Sensyne, Khan said. "Looking at failed phase II and phase III assets, can we use machine learning to go back and look at the data and say, 'Actually, maybe there was a population here that did respond to the drug that you didn't pick up before and can you submit a new application based on that analysis.'" That could lead to the design and initiation of additional trials comprised of patients selected using criteria defined by the initial analysis.
Recognizing from afar
Recognizing the value of real-world data in a different context, California-based Bluejay Mobile Health Inc. is building AI capabilities into telehealth consulting. What began as a platform to support video-guided exercises for people in physical therapy has since grown into a sophisticated remote assessment tool for clinicians, CEO Tony Zhang told BioWorld.
"Current telehealth technology is pretty much like FaceTime," Zhang said, referring to Apple Inc.'s consumer video conferencing service. For rehab though, you need to assess a patient's movements accurately, he said. To measure a joint's range of motion, health care providers typically use a goniometer, an instrument designed to measure angles. With support from technology from Santa Clara, Calif.-based Aifi Inc., Bluejay created a tool capable of obtaining virtually the same information, but via video. That involved training an algorithm to recognize people's joints and joint-movement in a variety of clothing in near-real time, a capacity it is already preparing to use in a partnership with New York-based Better PT Inc. to support connecting patients with the right clinics. "Digital measurement eventually will replace manual measurement, like in all other fields," he said.
Though its AI-enabled capabilities have yet to be leveraged in clinical trials, it's easy to imagine the day isn't far off. The company has already been working with the Colorado Integrated Care Network to support follow-up video visits and digital patient messaging, achieving what it said was a significant reduction in the average number of patient visits when compared to conventional clinical methods.
In the future, Zhang said, he hopes to leverage the work his team has done to enable the remote assessment of tremor in people with Parkinson's disease and balance in people at risk of falling, areas of research his team has already embarked upon.
There's plenty of work still needing to be done. In particular, Covance's Shough identified a need to understand and drive out the effects of bias in the application of AI, to access data that is standardized and contextualized with, for example, diagnostics data. Global regulatory frameworks, too, need to advance, so they can "move where technology is today," he said.