Contributed Commentary by Francis Kendall

October 23, 2019 | Artificial intelligence (AI) and machine learning (ML) are transforming data analysis and decision-making across a broad range of industries. In the pharmaceutical and healthcare sector, AI and ML offer enormous opportunities for delivering innovative medicines to patients faster and more cost-effectively by using real-world data (RWD) to accelerate the design and implementation of clinical trials.

The broader use of AI and ML to process RWD could, for example, enable designers of clinical trials to better tailor inclusion and exclusion criteria to target specific populations or widen them to optimize cohort sizes. These powerful data analysis tools may also be used to predict when patients may be eligible to meet enrollment criteria—this is particularly beneficial when stages of diseases are known, such as in cancer, rheumatoid arthritis and multiple schlerosis. Moreover, as the global aged population grows, the use of AI and ML models that consider comorbidities could be used to predict which patients may benefit most from a particular drug, potentially supporting the delivery of more targeted treatments that improve clinical outcomes.

However, despite the growing analytical capabilities of AI and ML strategies, their use in clinical trials has yet to be fully realized due to several key challenges.

Increasing access to real-world data and protecting patient privacy

Among the most significant issues currently limiting the greater adoption of AI and ML in clinical trials is the availability of RWD. AI and ML come into their own when used to analyze vast quantities of data, revealing insights that traditional analysis techniques simply cannot access. To fully capitalize on their analytical power—and justify their use—these approaches are most effectively applied when looking at large complex datasets; for example, when omics and patient-level data such as clinical records are used in combination.

Despite this, sourcing patient data on this scale is a major challenge, particularly when it comes to designing clinical trials for rare disease interventions. Obtaining all the information salient to an impactful clinical trial from a single data provider can be extremely difficult, with most sources offering specialized datasets.

Accessing data from dedicated providers is currently very costly, while extracting patient data from clinicians’ notes is time-consuming and resource-intensive.

To improve access to patient information and drive progress, greater data sharing within the pharmaceutical industry is essential, and there are a large number of initiatives demonstrating how successful collaboration can be. TransCelerate’s Placebo and Standard of Care Data Sharing Initiative, which seeks to maximize the value of existing clinical trials data to advance research, is just one example of how greater co-operation within the sector is helping to accelerate the design and execution of clinical trials.

Of course, by associating increasing amounts of personal data with clinical trial participants, maintaining patient anonymity becomes more challenging, especially when highly individualized genomic data is involved. However, with the advent of secure encryption and ID technologies such as blockchain, patient data is being made more secure. As these solutions become better established and more familiar in clinical research, I believe that concerns over patient privacy are likely to become less of an issue.

Overcoming perceptions around regulatory acceptance

Another factor limiting more widespread use of AI and ML in clinical trial design has been wariness over regulatory acceptance. Regulatory authorities such as the United States Food and Drug Administration (FDA) have, in fact, welcomed the use of RWD and tools such as AI and ML, and are open for discussion on how the industry can make better use of big data assets. The goal of the FDA's Information Exchange and Data Transformation (INFORMED) program, for example, is to promote the expansion of infrastructure for big data analytics and approaches for evidence generation to support regulatory decision-making. An ongoing collaboration between the FDA and Flatiron Health has already demonstrated how real-world evidence curated from electronic health records could be used to track off-label drug use, laying the groundwork to identify promising new indications for existing medicines.

Despite this, many companies are taking a wait-and-see approach towards adopting AI and ML strategies, and although some of the larger pharmaceutical companies are dipping their toes in the water, many others are holding back.

As the industry becomes more familiar with RWD in clinical trials, the success achieved by AI and ML pioneers is likely to encourage further growth. Here, information from wearable devices such as fitness trackers could be a promising launch pad for the wider inclusion of big data. Such tools offer exciting opportunities—analyzing location data, for example, could be used to better understand triggers of asthma attacks or link conditions to mobility. While this kind of information is likely to be used as secondary data at first, as confidence in RWD increases, it may well be used as a primary endpoint in future. Indeed, the growing use of pragmatic clinical trials and synthetic control arms—which draw on data collected in real-world clinical practice settings as well as sources such as electronic health records and wearables—looks set to normalize the handling and analysis of RWD for clinical research.

AI and ML offer enormous benefits for clinical trial design; however, their full potential is yet to be realized. Through greater collaboration between pharmaceutical companies, regulators and data providers, I believe that the barriers to the use of these powerful tools can, and will, be overcome.

Francis Kendall is vice president of FSP global services at Cytel. He has over 30 years of experience in leadership roles within the biometrics field, including as head of statistical programming at Roche/Genentech, and leading groups at Novartis, Sandoz and Nycomed. Francis is also a research affiliate at the MIT Media Lab, studying the impact of machine learning on clinical development. He can be reached at francis.kendall@cytel.com.