By Kellie Rademacher, Pharm.D., Precision for Value

Former FDA Commissioner Dr. Scott Gottlieb stressed the need for modernizing the clinical trials process in a speech to the Bipartisan Policy Center in January of this year.¹ He is quoted as saying, “digital technologies are one of our most promising tools for making healthcare more efficient.” Improving efficiency in clinical trial development is only one potential enhancement that can result from the use of machine learning. Machine learning and artificial intelligence (AI) are often used interchangeably, but that assumption is incorrect. Machine learning is the subset of AI that is related to the development of algorithms that can make accurate predictions of future outcomes via pattern recognition and rules-based logic. Such use of logic and algorithms can improve patient selection, provide predictive long-term outcomes, and reduce the time and cost in the execution of clinical trials.

Globally, the healthcare industry is experimenting with and reporting on a myriad of ways that machine learning may impact drug development. University College London recently published a study on how machine learning could determine whether a new drug works in the brain, which, according to the authors, could be missed by conventional statistical tests.² Another article discusses the impact that machine learning can have (and has had) in clinical trial design, patient selection, data interpretation, and FDA approval.³ The authors report the approval of two devices that use machine learning to support diagnosis and treatment. The same article highlights how a team of researchers used machine learning to stratify patients into subgroups based on different rates of disease progression and risk of complications.

I recently participated in a payer advisory board for a pharmaceutical manufacturer that was looking to gain insight into additional trial data that would support enhanced access with U.S. payers for a drug for a rare disease for which there are limited treatment options. Several of the proposed studies involved machine learning. One of the recommended studies would be using machine learning to estimate long-term efficacy and safety. Another proposed study supported by the payer advisors on the board was a machine learning algorithm that would identify early predictors of this rare disease using large claims databases. The task of the payers on the board was to prioritize the studies that could potentially move the dial for the agent; the studies referenced above were well supported by the attendees as impactful data.

The opportunity to use machine learning for drug development and approval seems limitless. Clearly, some manufacturers are looking into AI to enhance their data packages and testing those outputs in the industry. Why, then, has the incorporation of AI into the workflow for drug developers been slow to progress more broadly?

Recent healthcare industry articles highlight the cautious use of machine learning among pharmaceutical manufacturers.^3-5 Understandably, risks to patient safety, patient privacy, and data integrity have been the overwhelming rationale for the limited application of machine learning into clinical trial development.³ I suggest an additional factor is limited experience in how payers will assess and interpret data generated by AI. Gottlieb has stated that the FDA is developing a new regulatory framework intended to support the use of AI. Specific to clinical trials, the FDA has developed Master Clinical Trial Protocols (MAPs) intended to increase efficiency and lower costs of clinical trial development.⁶ So, it’s evident that progress is being made in the incorporation of machine learning in clinical trials, drug development, and regulatory approval.

Payers and providers have been using data from a variety of sources — clinical trials, real-world evidence, registries, etc. — to assess the impact of medications on patients. Data from a survey of payers and providers supports these stakeholders’ views on the continued use of data: 93 percent of survey respondents indicated that predictive analytics represent the future and will be required as part of the healthcare industry’s shift from volume to value.⁷ The Academy of Managed Care Pharmacy, managed care’s primary trade organization, recently held a webinar updating members on the use of AI, Big Data, and technology in the healthcare industry.⁸ The AMCP Foundation also produces a Trends in Health Care report, the latest of which highlighted how payer-owned claims data can assist manufacturers with clinical trial planning. Examples of areas where claims could enhance trial planning included targeting on disease prevalence via geography, investigator profiling with patient counts, and physician profiling for referrals, including patient count and distance from sites.⁹

Partnerships between payers and manufacturers in clinical trial planning could be a huge win for all stakeholders. Payers and providers alike are investing in infrastructure, including software and talent, to be able to conduct the necessary analyses for the shift from volume to value. Payer investment is intended for data generated and analyzed internally. Payers are notoriously skeptical of manufacturer-provided data. Collaborations such as Janssen and Aetna’s outcomes-research subsidiary Healthagen, described  in the AMCP Trends in Health Care Report referenced above,⁹may increase trust in the data generated, potentially improve trust between partnering organizations, reduce the time and cost of the research, and bring important therapies to patients more rapidly.

The majority of pharmaceutical manufacturers, regulatory authorities, payers, and providers are currently using and looking to expand the use of machine learning to support the cost and efficiency needs of treating patients in our evolving healthcare sector. It’s no secret there is a general lack of trust between manufacturers and payers, payers and providers, and providers and manufacturers when it comes to data generated in-house. It remains to be seen if machine learning-generated data can have a positive impact on these relationships. It is unlikely that the availability of machine learning data alone will be sufficient to move the dial toward robust efficiency and cost savings. It is this author’s opinion that much work will need to be done across stakeholders to improve transparency, data integration, data integrity, and partnership to truly realize the full potential of machine learning.

While the healthcare industry is taking baby steps into using machine learning in clinical trials and for other data-generating objectives, the value of this data will depend on the transparency of algorithms, robustness of data sources, and extrapolation to real-world outcomes. It will be imperative that manufacturers, payers, and providers look at each other as partners in evolving the affordability of improving the outcomes of patients in our healthcare system.

Source: https://www.clinicalleader.com/doc/machine-learning-in-clinical-trials-what-will-the-future-hold-and-what-s-holding-us-back-0001