The Expanding Role Of Artificial Intelligence In Clinical Research

In recent years, access to patient medical information, coupled with rapid advancements in data analytics tools and technologies, has significantly altered many areas of healthcare, from early-stage discovery and research to patient treatment. One of the most significant applications of new technology is in efforts to streamline and advance clinical research. Advanced technologies, including artificial intelligence (AI), offer the promise of addressing many of the most challenging aspects of drug development, with potential benefits for pharmaceutical companies, investigators, patients, regulators, and payers. While protocols and standards in clinical research have become increasingly complex, slowing progress and increasing costs, companies from startups to Big Pharma are identifying opportunities to apply AI to enhance trial efficiency, patient enrollment, and outcomes targeting. The fact that we are at the dawn of AI technology indicates that its role in clinical research could grow exponentially in the years ahead.

To assess the applications of these technologies in research now and in the future, the Life Sciences Practice team at the global consulting firm CRA recently conducted an analysis of trends in the use of AI in drug development and identified several areas where manufacturers now consider AI to be essential, including in discovery of new drugs and drug targets, identification of eligible patients for clinical trials, screening and diagnosis of patients, and optimization of drug administration and dosing regimens. The analysis also highlighted the need for drug developers to keep pace with innovations in AI to ensure that advanced technologies are applied as quickly and cost-efficiently as possible and that specialized expertise, resources, and infrastructures are in place to support their use.

Identifying Drug Candidates And Patients

On a global scale, the number of investigational drugs in development has increased dramatically in recent years. Many represent potentially historic advances in the treatment of a range of serious diseases, including different types of cancer and infectious and autoimmune diseases. The fact that there are many drugs in clinical development, including some targeting the same or similar indications, also introduces new levels of competition both for patients to participate in clinical research and in planning for commercialization.

In many cases, these development programs follow the long-established approach of selecting and targeting cells with higher proliferation activity associated with disease. More recently, however, manufacturers have focused clinical research on targeting the root cause of disease – the underlying biological pathways that are associated with disease onset and progression – in efforts to deliver optimal and potentially curative benefit. This shift has also introduced the need for more advanced protocols in patient screening and in execution of clinical research.

Within this environment, many drug manufacturers are now aggressively exploring the use of automated algorithms and advanced predictive models to help identify potential molecular targets faster. These models also are being used to project the potential of new drugs to advance through regulatory reviews and ultimately succeed commercially. In some cases, the application of these technologies and approaches requires drug developers to collaborate with other industry stakeholders, including academic researchers and technology suppliers. Groups such as the Machine Learning Ledger Orchestration for Drug Discovery (MELLODDY) and the Machine Learning for Pharmaceutical Discovery and Synthesis (MLPDS) consortia were created from such partnerships in efforts to develop and apply more accurate predictive models and make drug discovery more efficient.

In addition to supporting drug discovery, AI tools also are being used to identify patients for clinical research, with the potential to deliver many significant advantages in drug development. Companies are using options including advanced neural networks and Bayesian algorithm-powered software systems to mine large datasets of patient information to identify enrollment candidates. Many are also using AI to implement predictive models to improve the statistical significance of data collected from more targeted candidates as well as help reduce trial costs.

In just one example, Novartis, Pfizer, and Eli Lilly recently leveraged a technology platform from the U.K.-based AI firm Antidote to help streamline clinical trial processes and identify appropriate patients.¹ Antidote’s technology works to comb through trial data and make patient information, including relevant eligibility criteria, more accessible to pharmaceutical companies. In this effort, Antidote has reviewed data associated with approximately 14,000 clinical trials related to more than 700 diseases and conditions to date.

For companies developing therapies for rare and ultra-rare diseases, it can be especially difficult to identify appropriate patients for clinical research within predetermined timeline goals. AI tools can improve the ability to assess which patient candidates might respond better to drug treatment and also forecast their trial dropout rates.

While physicians themselves previously took the lead in combing through data to highlight patterns in clinical trial patient populations and use that information to predict outcomes related to drug efficacy and safety, AI is well positioned to assume this role at a much larger and more efficient scale. As technology and AI companies continue to expand their networks of healthcare organizations, pharmaceutical companies, and contract research organizations, there will likely be more opportunities for manufacturers to use AI to identify and recruit optimal patients for clinical trials.

Assessing The Impact Of Different Dosing Regimens

New AI tools also are allowing drug manufacturers to better assess issues related to drug administration and optimal dosing regimens. Customized algorithms can determine the impact of different dosing levels and schedules – particularly in combination therapies – on drug efficacy and safety, which could ultimately reduce the risk of adverse events, trial delays, and patient discontinuations. The data rules for these algorithms are generally tailored to specific patient populations by observing how individual patients are responding to treatment and the impact of dose adjustments on both efficacy and side effects.

Zenith Epigenetics recently leveraged one such algorithm, called the CURATE.AI technology platform, to continuously identify the optimal dose for its investigational drug ZEN-3694 in treatment of a patient with metastatic castration-resistant prostate cancer (MCRPC).² In the clinical trial, a MCRPC patient was given a novel drug combination consisting of ZEN-3694 and enzalutamide, an approved prostate cancer drug. The CURATE.AI platform used the patient’s own clinical data – including drug dose and corresponding changes to tumor size or level of cancer biomarkers in the blood – to calibrate the unique response to treatment. This calibration was then used to create an individualized CURATE.AI profile, or map, that identified the drug doses that would lead to the best possible treatment outcome at any given point during the trial.

While there has been some success in achieving AI-enabled dosing regimens as seen in Zenith’s clinical trial, many industry insiders agree that additional research is needed before these types of analytics can be used more broadly. For example, to support development of investigational precision-based therapies including cell and gene therapies, manufacturers may need high-caliber, self-driven AI models to improve drug efficacy while decreasing the risk of dosing-associated side effects. These data analytics strategies could offer a major improvement over traditional “eye-balling” techniques used for dosing, which involves observing how patients respond to a specific dose and adjusting dose levels as needed.

Conclusion

Most industry stakeholders expect that the use of AI and advanced data analytics in drug development will continue to expand in the years ahead, with major implications for the speed, efficiency, and costs of these programs. Collaboration with key stakeholders such as government agencies and data providers will be essential to gain access to the most innovative and advanced analytics tools and relevant patient information. These partnerships also may enhance data integrity, drive best practices in predictive modeling, and help build consensus on challenging ethical issues associated with patient data collection. After commercialization, leveraging AI also may expedite and enhance patient care at screening, diagnosis, and treatment, including better management of all stages of a patient’s journey.

References:

1. Megan Molteni, “Meet the Company Trying to Democratize Clinical Trials With AI,” Wired, January 30, 2018, available at https://www.wired.com/story/meet-the-company-trying-to-democratize-clinical-trials-with-ai/.
2. “NUS researchers use AI to successfully treat metastatic cancer patient,” National University of Singapore, August 31, 2018, available at http://news.nus.edu.sg/press-releases/nus-researchers-use-ai-successfully-treat-metastatic-cancer-patient.

About The Authors:

Lev Gerlovin is a vice president in CRA’s Life Sciences Practice. He has more than 12 years’ experience in life sciences strategy consulting, focused on commercial and market access strategies. He leads CRA’s Life Sciences 2030 issues leadership platform.

Brooke Bonet is an associate principal at CRA with more than 10 years’ experience in life sciences consultancy, specializing in commercial strategy.

Inderpreet Kambo is an associate principal in CRA’s Life Sciences Practice with eight years’ experience focusing on analytical tool development and new product design.

Giuliano Ricciardi is an associate in CRA’s Life Science Practice, focusing on market access, commercial due diligence, and opportunity assessment projects.

The views expressed herein are the authors’ and not those of CRA or any organizations with which the authors are affiliated.