Skip to main content
Clinical Pharmacology and Therapeutics logoLink to Clinical Pharmacology and Therapeutics
. 2026 Oct 30:10.1002/cpt.70539. Online ahead of print. doi: 10.1002/cpt.70539

From Information to Insight: A Decision Architecture for Informed Governance

Jack A Cook 1,✉
PMCID: PMC13721024  PMID: 42960755

For effective drug development governance, understanding risks and success likelihood is crucial for informed decision‐making. Rayner et al. proposed a structured decision architecture that elevates risks to governance bodies and employs AI and expert judgment. This structure is valuable regardless of company size or portfolio. While it may not guarantee optimal decisions, it ensures decisions are made with comprehensive information and clear understanding of risks and assumptions.

THE CHALLENGE OF DRUG DEVELOPMENT

Rayner et al. 1 called the current process of governance of drug development “the problem we keep not solving”. Drug development is a complex and expensive process with a cost to bring a new drug to market estimated to be $2.23 billion dollars. 2 Once a molecule enters drug development, it will either become a drug or not. Thus, one role of governance is to separate, as early as possible, those that will not become drugs from those that will. The more efficient the process, the better the use of capital and resources. For those that are successful, the accompanying label should ideally provide for optimal use. There are opportunities for improvement in the process; indeed, the United States Food and Drug Administration recognized this need and initiated a research project to characterize the reasons for the high number of postmarketing dose changes to support their efforts to get the right dose to the right patient at the time of approval. 3

For drugs that are eventually approved, speed to market is a key driver of project value and thus a strong incentive for sponsors to accelerate development, with each day of delay estimated to cost approximately $500,000 in lost sales. 4 The experience of regulatory agencies has been that Phase 2 studies are simplified (e.g., a reduced number of doses studied) to reduce development time. The result is that dose selection for Phase 3 is often empirical and not as scientifically sound as one derived on a well‐characterized dose–response. 5 The result can be costly, resulting in new Phase 3 trials at alternative doses 6 , 7 or post‐marketing changes previously noted. 3

APPRECIATION OF THE RISK—A TOTALITY OF INFORMATION APPROACH

The challenge for good governance is at least twofold. Development risks need to be identified, and these risks need to be quantified in a way that governance bodies can fully appreciate their impact. The proposed structure of creating an interrogation taxonomy, use of threshold‐based escalation criteria (deterministic safeguards), and constrained artificial intelligence (AI) synthesis to identify and assess risk helps ensure that the totality of what is known about an asset is used to evaluate and guide decisions and that pertinent risks are identified and evaluated. Use of AI as an “independent evaluator/third party” can help guard against the bias of development teams and those making decisions, a desirable component of decision‐making. 8 As pointed out by the authors, such a structure can surface issues as they occur, a practice consistent with the desire for speed in drug development.

This tool has broad applicability across drug development and is applicable to companies regardless of their size or resources. While the AI component is valuable, it will be challenged by limits on data access. Hwang et al. found that pivotal trial results were never published in peer‐reviewed journals for 60% of drugs that failed in late‐stage development, rising to more than 90%. 9 Such information would be valuable for both identifying risks and assessing their importance.

HOW WILL GOOD GOVERNANCE MANIFEST? WHAT TO EXPECT AND WHAT NOT TO EXPECT

Implementation of such a tool will likely not yield a major improvement in the success rate for compounds entering human studies. For that to happen, an improvement in the selection of candidates to take into clinical development is needed. It might be that an analogous system for that space could be developed.

As proposed, the tool may improve the speed of development of some candidates. For example, identification of dose selection risks might avoid prolonged late‐stage development due to having to repeat pivotal trials as well as making posology changes after approval. Further, early identification of risks could result in quicker resolution of problems that can be overcome, for example, formulation changes. It should be noted that having the AI evaluation search not only for risks, but ways to accelerate programs has the promise of substantial benefits. There are significant examples of time (and cost) savings coming from application of model‐informed drug development (MIDD) strategies. 10

This tool may also reduce the cost of failure of unsuccessful candidates. Early identification of risks and assessment of the associated probability of success may allow for quicker termination of programs likely to fail, reducing costs and allowing redeployment of resources to other potentially successful candidates. If realized, this would result in acceleration of new medicines reaching patients.

Finally, the implementation can help create a learning organization. Drug development decisions are often complex and are not always driven strictly by financial considerations. For example, when considering which asset(s) to take forward, a company may choose a compound with a lower estimated rate of return because it is in a disease area where it wants to invest or choose a compound with a lower probability of success but in an area where, if successful, a much higher return is expected. A governance that not only keeps track of decisions but also records their rationale can learn. It allows organizations to distinguish good decisions from good outcomes. While the democratization of decision‐making does not guarantee optimal decisions, it will help ensure that decisions are made with full information and explicate assumptions.

FUNDING

No funding was received for this work.

CONFLICT OF INTEREST

JC is an employee of A2‐Ai, LLC, Ann Arbor, MI, USA, and in that capacity consults for multiple pharmaceutical companies. He also holds stock in multiple pharmaceutical companies.

DISCLOSURE

This work has not been previously presented.

References


Articles from Clinical Pharmacology and Therapeutics are provided here courtesy of Wiley and American Society for Clinical Pharmacology and Therapeutics

RESOURCES