Artificial Intelligence (AI) is often described as the next major transformation in clinical research. Industry conference agendas and exhibit halls are filled with the promise of AI-powered solutions. But when our AI Working Group recently discussed how AI is actually being used today, a different picture emerged.

The most mature and widely adopted AI use cases were not specific to clinical trials at all. They were healthcare and operational applications that happened to support research activities.

Clinical care is leading the way

Participants from health systems, academic medical centers, and research organizations described a growing number of AI implementations already in use for clinical care at their institutions.

Ambient listening tools are capturing provider-patient conversations and generating clinical notes. AI imaging tools are being used to make image review faster and more accurate. AI assistants are helping staff summarize meetings, draft communications and review documents. Many institutions are building internal knowledge repositories to help staff find answers quickly and improve consistency across staff. AI scheduling capabilities, smart patient messaging technology and EHR search tools are used to improve clinical care outcomes. These solutions commonly streamline administrative work and reduce documentation burden. 

In many situations, these tools can help research teams work more efficiently even when research was not the primary use case.

Research-specific AI tools remain limited

Implementation of AI applications developed specifically for clinical research was very low, with many sites not having adopted any AI clinical research tools. Instead, they are leveraging generic or clinical AI tools to support clinical research use cases, such as:

• Searching EHR data for certain inclusion and exclusion criteria
• Drafting regulatory and operational documentation based on institutional templates
• Reviewing contracts and supporting contract negotiations
• Assisting with communications and correspondence with sponsors/CROs
• Supporting quality review processes and document checks

These use cases offer incremental efficiency gains rather than transformational changes, but they are relatively easy to implement and typically do not require sponsor/CRO approval.

Why the Gap?

In most organizations, AI adoption is being driven by enterprise healthcare priorities rather than research-specific initiatives. This is very logical considering the differences in scale and revenue between clinical care and clinical trials. The discussion also highlighted external considerations for clinical trials that complicate AI adoption.

Oversight and Data Governance Challenges

Under ICH E6(R3), sponsors and CROs are responsible for ensuring that all software systems used in a clinical trial, even those used by sites, are fit for purpose and validated. They are also fully responsible for the governance of clinical trial data throughout the lifecycle of a study. AI raises new concerns about data access and privacy controls when AI models are used in an uncontrolled way. These responsibilities and concerns have led many sponsors to prohibit sites from using AI in clinical trial activities, or to severely limit the study materials that sites are allowed to input to AI tools they are using.

Regulatory Guidance

Research teams operate in environments where documentation, traceability, and audit readiness are paramount. While global regulators have begun issuing guidance related to AI, the guidance is not yet specific to clinical trial operations. Most organizations are still determining how to apply these principles within clinical trial operations.

Validation 

Several participants emphasized the importance of human oversight and consistently described AI as a tool that augments human decision-making rather than replaces it. The question is often not whether AI can perform a task, but whether organizations can trust the output enough to use it in regulated research processes. Guidance for how to validate AI systems and vendors is still in its infancy, leaving organizations to determine their own requirements and leading to a cautious approach.

What Comes Next?

The consensus from the discussion is that the clinical research industry is still in the early stages of AI adoption. 

The most successful applications at clinical research sites today use generic AI tools to solve practical operational problems, fitting naturally into existing workflows, and earning the trust of the people who use them. Leveraging AI to improve staff efficiency on administrative tasks is the right first step, even if it is not the most sophisticated.

There is still a broad range of adoption of these basic AI use cases across organizations. In the near term, experimentation with these internal use cases will continue to grow. 

Sophisticated, research-specific implementations will be the next stage of AI adoption. This stage will be driven by sponsor/CRO adoption and distribution of AI tools built for clinical research. Implementation of these tools will bring new challenges of determining how to adopt tools responsibly, validate them appropriately, manage data privacy and security between the many parties participating in a clinical trial, and ensure they deliver measurable value without creating additional burden. The future of AI in clinical trials may ultimately depend less on what the technology can do and more on what the industry is prepared to adopt.