How Causal Machine Learning is Transforming Clinical Trials
Welcome to an insightful discussion on the future of clinical trials and the role of causal machine learning in drug development. In this interview at DPHARM 2024, we speak with Raviv Pryluk, CEO and Co-founder of PhaseV, to explore how this technology is reshaping the landscape, addressing challenges, and enhancing the precision of medical treatments.
Moe Alsumidaie: Raviv, how do you see causal machine learning reshaping drug development, especially for niche therapies or rare diseases, over the next decade?
Raviv Pryluk: Before discussing its impact, let’s address the challenges. Different patients respond differently to treatments. You can release a drug to the market, and some patients will respond while others will not. The same is true in clinical trials. Many trials fail, not because the drug isn’t working, but because it’s only effective for a subset of patients. Identifying who will respond and who won’t is highly beneficial,
especially in the precision medicine era. Causal machine learning, which intersects causal inference and machine learning, can uniquely identify responders and the sources of heterogeneity. This is crucial in oncology, immunology, and neurodegenerative diseases, as well as in rare diseases where patient recruitment is challenging. For example, in rare oncology indications, where treatments can be particular to genetic markers, causal machine learning can help pinpoint which patients will benefit from a new therapy, thereby increasing the trial’s success rate and speeding up the drug’s time to market.
Moe Alsumidaie: How do you plan to scale PhaseV’s machine learning platform to smaller biotechs that lack the infrastructure of larger companies?
Raviv Pryluk: We provide either tech-enabled services or SaaS licensing of our software. Larger companies, like big pharma, often prefer to license and use the software themselves with our support. We use the technology for to provide services, reanalyzing data retrospectively and supporting the design and execution of adaptive, dynamic clinical trials. This approach makes our technology accessible across many therapeutic areas, regardless of company size. For instance, a small biotech working on a rare disease might not have the resources to build a robust data infrastructure. By offering our platform as a service, we enable them to leverage advanced machine learning capabilities without significant upfront investment. This democratization of data and technology ensures that even smaller players can compete on a level playing field with larger companies.
Moe Alsumidaie: How do you fully address concerns from stakeholders who are hesitant to embrace machine learning and AI in this highly regulated environment?
Raviv Pryluk: The hesitation is understandable, especially when dealing with patients and complex experiments. We spend a lot of time on validation. Our algorithms are designed with validation in mind, using train-test-validation frameworks, external datasets, and sophisticated statistical tests to ensure validity, accuracy and robustness. We have a proprietary data lake for additional validation and use bootstrapping to test many scenarios in-silico. This rigorous validation process gives us confidence in our findings, and we only recommend following AI insights when the confidence level is high. For example, we might validate our machine learning models against multiple clinical trial datasets to ensure that the patterns we identify are consistent and reliable. This thorough validation helps build trust among stakeholders, including regulators, who must be assured that the AI-driven insights are robust and actionable.
Moe Alsumidaie: What are the most significant barriers to adopting adaptive trial designs, and how does PhaseV lead this shift, especially in big pharma?
Raviv Pryluk: There are many barriers, including complexity, less intuitive designs, and operational challenges. We’ve developed an intuitive software platform that designs trials, unravels insights, and explains adaptations. This helps stakeholders, from biostatisticians to commercial teams, understand the implications. The same platform supports design and execution, making the process more reliable and streamlined. Clear communication with regulators about the logic behind adaptations also increases confidence and acceptance. For example, our platform can provide a detailed rationale for why a particular adaptation is being made in a trial, backed by data and statistical analysis. This transparency helps all stakeholders, including regulatory bodies, understand and trust the adaptive design, facilitating its adoption.
Moe Alsumidaie: What steps is PhaseV taking to establish itself as a thought leader in clinical trial optimization and influence regulatory frameworks?
Raviv Pryluk: We are humble and focused on bringing value to sponsors and regulators. We build valuable technology grounded in solid science, publish papers, and patents. Engaging with the community through conferences, panels, and meetings with regulators helps us provide insights and get feedback. We apply our technology in real-world scenarios, sharing successes and failures transparently. This collective approach helps us become an essential player in the space. By presenting our findings and methodologies at industry conferences, we showcase our capabilities and contribute to the broader dialogue on clinical trial optimization. This helps us influence regulatory frameworks by demonstrating our technology’s practical benefits and reliability.
Moe Alsumidaie: As AI becomes more integrated into clinical trials, do you foresee it aiding rather than replacing people? What do you think is the future here?
Raviv Pryluk: AI in clinical development is about augmenting and empowering experts, not replacing them. Our platform makes biostatisticians (clin-ops/clin-dev experts) more productive, accurate, and faster. Tasks that took weeks can now be done with a click of a button. We aim to increase the success rate of clinical development, making drugs more available to patients. The future is about increasing productivity and throughput in clinical development, not replacing anyone.
