While regulatory science can lag behind technology advances, the FDA has for the past few years been exploring ways to harness the potential of artificial intelligence (AI) to streamline drug development and the approval process.
A nexus for its efforts is the Information Exchange and Data Transformation (INFORMED) initiative anchored in the agency's Oncology Center of Excellence (OCE). At its inception in 2016, INFORMED was designed to tap into the power of big data and advanced analytics to improve disease outcomes.
Today, INFORMED has expanded its focus in line with President Donald Trump's American AI Initiative, serving as an incubator focused on driving innovations in agile technology development and advanced analytics.
In a Q&A with BioWorld, Sean Khozin, associate director for oncology regulatory science and informatics at the OCE, discussed how the FDA is approaching the use of AI in drug development and areas in which it holds near-term promise.
BioWorld: Is the FDA using AI or machine learning in drug reviews? If not, does it plan to in the future?
Khozin: FDA is experimenting with several advanced analytical and predictive machine learning methods that can potentially streamline the drug review process. However, these methods require validation, new technical capabilities, appropriate human capital and process reengineering prior to integration into regulatory review workflows. Therefore, AI doesn't currently play a substantial role in the review process.
BioWorld: Several drug companies are using AI in both drug development and clinical trial design. Do these uses require the FDA's blessing? How can companies validate these tools? What concerns does the FDA have about the use of AI to develop drugs and adaptive trials?
Khozin: FDA applies a risk-based approach to the clearance and approval of AI-based platforms. FDA's regulatory authority covers certain aspects of AI in clinical trial design and in drug development. FDA recently proposed a framework for the use of AI-based applications. [The framework is for modifications to AI/machine learning-based software as a medical device.]
FDA is conducting research to identify, and in some cases de-risk, the use of AI in clinical trial designs as a drug development tool. For example, FDA's INFORMED Program is conducting foundational research on the use of AI for advancing clinical development programs supporting product development in areas of high unmet need. . .
INFORMED is aligning its overall approach in this domain with the new national AI strategy aimed at catalyzing new innovations and market opportunities that drive economic growth and advance our national interests and priorities, including the FDA's mandate of promotion and protection of public health.
BioWorld: What do you see as the biggest challenges to the use of AI or machine learning in regulatory science, drug development and clinical trials?
Khozin: AI requires hard-to-find specialized expertise that should be incorporated into clinical development teams as multidisciplinary units, which requires significant organizational commitment and a new cultural orientation. Furthermore, validation of AI algorithms requires new methods and standards that are not yet fully established.
BioWorld: Is the FDA meeting with regulators in other countries to work through these challenges?
Khozin: FDA has held public meetings and discussion with international regulators on general themes related to the use of AI in regulatory review and drug development and seeks to coordinate such efforts to the extent feasible with the global community of regulators.
BioWorld: How far off do you think we are from AI becoming a common tool in the FDA's decision making, drug development and trial design?
Khozin: AI will undoubtedly play an increasingly larger role over the next five to 10 years in all aspects of drug development and the regulatory review process. In collaboration with academia and relevant stakeholders, FDA scientists are actively involved in research and knowledge sharing as we continue to develop new processes for capturing the value of AI in not only the regulatory review process but across the full spectrum of R&D efforts.
BioWorld: How else might we see AI used in the biopharma field?
Khozin: One of the most promising uses of AI in clinical trials is biomarker discovery and development. We typically think of biomarkers as objectively measured patient and disease characteristics that can be used as clinically meaningful endpoints or prognostic and predictive variables.
Most biomarkers (e.g., the lipid panel as a risk factor for cardiovascular disease and EGFR positivity as a predictor of response to targeted small molecules in advanced non-small- cell lung cancer) have led to the successful development of several safe and effective therapies.
As we begin to gain a better understanding of the biological and pathophysiological complexities of disease using modern tools of measurement such as next-generation sequencing and digital medicine devices, traditional bioinformatics tools and analytical approaches can fall short of sorting out the growing volume and diversity of the data into appropriately validated biomarkers to support the development of safer and more effective therapies.
AI is helping tame the data we're already generating in clinical trials to develop complex multi-omics phenotyping strategies that go beyond the assumption that, for example, a single somatic mutation like EGFR tells the full story of oncogenesis and response to treatment. We're seeing AI-based platforms for complex genomic data being co-developed as companion diagnostics.
In addition, incorporation of digital medicine tools such as wearable single-lead ECGs are providing new vehicles for continuous measurements leading to very large datasets. Analyzing these complex datasets using AI-based modalities is revealing new insights into the experience of patients and biological intricacies of disease, making the evidence base from clinical trials more comprehensive, precise, clinically meaningful and individualized.
AI can be a transformative tool to catalyze a new generation of discoveries and clinical evidence-generation modalities in the service of patients and the health and wellbeing of the public. Its appropriate use in biopharma is an iterative process requiring active experimentation and collaboration in addition to data sharing for validation of the tools and algorithms arising from the existing body of research and investments in the field.