The potential of AI as a strategic lever for innovation and regulatory decision-making

While regulatory authorities publish guidelines that describe their expectations with submissions, what was often missing in the past was a way for manufacturers to interpret how these guidelines were applied to their products, especially products that deviate from standardized clinical and manufacturing models such as advanced therapy medicinal products (ATMPs). 

This is where predictive AI, which uses statistical analysis and machine learning to identify patterns and make predictions about potential outcomes, can transform the decision-making process. 

Predictive AI allows manufacturers to understand the potential impact of their products -- efficacy, safety, regulatory, and manufacturing – to support data-driven decision-making across the product lifecycle. In doing so, regulatory intelligence evolves from a passive knowledge repository of documents and information into an active part of the scientific and regulatory infrastructure that enables, guides, and defends scientific decision making.

Using AI tools – large language models (LLMs), semantic search tools, machine learning, generative AI, etc. – manufacturers can review publicly available information from the health authorities on the outcome of a regulatory activity – be that a list of questions, deficiency letters, or approvals – that document how rules and guidelines have been applied to those products.  

An example is the US Food and Drug Administration’s (FDA) complete response letters (CRLs)¹, which document deficiencies that can appear late in the review cycle and delay authorization. By anticipating hurdles based on past regulatory actions, manufacturers can mitigate risk with their own submissions. 

AI tools can help to unlock key insights to support more strategic decision-making across the product lifecycle. Properly applied and supported by statistical analytics, AI can help to inform the clinical design to better align it with regulatory strategies and health authority requirements to improve the odds of regulatory success. The goal is to have trial designs that improve the likelihood of regulatory approval, and don’t result in extensive questions and limitations from the health authorities. 

AI tools, particularly statistical and machine learning for predictive modeling, can help to enrich evidence packages, increasing confidence and clarity in clinical data, thereby improving regulatory submissions. 

In regulatory, safety, and quality, AI can help to build stronger dossiers with clearer data on benefit risk and more robust data on quality and consistency. AI together with statistical modeling could be leveraged to turn clinical trial data into evidence by identifying key features and quantifying potential outcomes (what that outcome could be, how likely it is, and what the uncertainty is). By using these capabilities to review the submission, manufacturers could gain insights into potential inconsistencies between modules, detect weak comparability links, or flag areas likely to generate major review questions based on patterns observed – all before regulatory submission. 

Equally, AI can play an important role in supporting economic modeling decision-making and shaping health technology assessment (HTA) evidence generation. ISPOR’s Value in Health journal has published a working group report that discusses the uses of generative AI in HTA, including systematic literature reviews, real-world evidence, and health economic modeling.²

Already, AI is revolutionizing the initial stages of economic modeling by enabling teams to test multiple model structures simultaneously, identifying the most efficient and credible options to support a product’s value story. AI is also being put to use to consolidate disparate data sources and simulate missing input, helping to tackle a common issue with rare diseases. This allows health economists to focus on higher value tasks. Report authoring is another area where AI is proving to be valuable, where it automates repetitive tasks such as formatting and data integration, enhancing efficiency and ensuring consistency across documents.  

AI can be used to simulate likely outcomes later in the R&D and regulatory submission stages. In an OECD paper on governing with AI in core government functions, the authors propose using AI simulations to model and predict potential impact of regulatory choices to support policy assessment.³ The goal, the authors note, would be to understand the drivers of innovation and their potential impact before implementing policy change. 

In the highly complex and regulated life sciences industry, such simulations should be conducted together with subject matter experts in regulatory, pharmacovigilance, quality as well as in market access.

AI has the potential to transform the regulatory process and engagement with payers. It can discover regulatory patterns, predict questions and concerns from regulators, and support evidence-based decision-making.

However, AI models require human expertise combined with effective digital capabilities. Manufacturers will need to consider their partners, including digital solution providers and regulatory experts, and ensure they continually invest in people and capabilities, not just tools.