Leveraging AI agents to eliminate ‘white space’ in clinical trials

For decades, clinical trials have played an integral role in developing and determining the safety and efficacy of new and often lifesaving drugs. However, the clinical trial process has become increasingly complex in recent years, alongside the growing burden of human disease – creating a broadening range of inefficiencies and operational challenges.

This issue was quantified in 2023 in a study by Tufts Research, which attributed expanding trial complexity to a “continuing upward trend across all protocol design variables,” particularly in phase II and phase III trials, which “average more endpoints, eligibility criteria; protocol pages; investigative sites; countries and datapoints collected” than other phases. Not only has this complexity made research more challenging, but it has exacerbated the amount of ‘white space’ across the clinical trial lifecycle, an already pervasive and long-standing challenge for the industry.

In fact, one study shows that nearly half (45%) of a new drug’s development time is spent on white space, defined by the study’s authors as the time between trial completion and reaching the next phase of regulatory submission. Worse, the common definition of white space does not even account for the full spectrum of unproductive periods in between and throughout individual trial activities, which can be harder to pin down and study, yet which have an equally disruptive impact on trials. In fact, it has likely contributed to overall lengthening trial cycle times (i.e., the duration from approval of a trial protocol to database lock) which have jumped, on average, by at least seven months just since 2020.

Fortunately, while it seems clear that we’ve finally reached the limits of human-only clinical development, it’s also true that pharmaceutical technology is now entering a new era defined by rapid innovation and the integration of artificial intelligence (AI), with agentic AI standing out as uniquely suited to address the white space conundrum. Human-AI collaboration has the potential to mitigate gross productivity gaps and establish a new standard of clinical trial efficiency to save more lives and capture the true potential of advancing breakthroughs in modern medicine.

Understanding white space within trial execution

White space has plagued the clinical research industry for decades. However, despite the implementation of new clinical trial processes and technologies, productivity gaps continue to unnecessarily extend timelines, with white space for clinical trials over the past five years, increasing by 14 months and effectively offsetting recent progress.

But why, more specifically, is this happening? For one, technological solutions to this point have still failed to address the more fundamental and systemic issues plaguing clinical development operations.

For example, there are still multiple manual processes in place that depend entirely on human availability and handoffs, including tracking regulatory submission status, coordinating site feasibility assessments, and resolving data queries following patient visits. Moreover, even when supportive technology is being utilised, the systems being leveraged are often maintained and managed in silos, resulting in an irrevocably fragmented data ecosystem, poor collaboration between operational teams, and further exacerbating delays across the trial lifecycle.

Additionally, most clinical trials today are still being performed around sequential workflows, forcing the linear execution of activities that can and should occur in parallel, and creating a compounding effect in which delays in one area end up negatively impacting the efficiency of dependent processes, to the point that trial timelines are often needlessly extended by several months or, in some cases, even years.

The combined inefficiency of manual processes, siloed systems, and sequential workflows not only hinders individual tasks, but creates unproductive, self-replicating white space across all stages of the clinical trial process – whether it’s a delay in training or IRB approvals between protocol finalisation and site activation; stagnation due to protocol amendments between activation and last patient last visit (LPLV); or data cleaning and query resolution bottlenecks between LPLV and database lock.

Regardless of when or where white space emerges, the fact is there are deeply entrenched challenges surrounding how labour-intensive tasks are planned and executed. However, the silver lining is that white space is not the result of scientific or technological limitations, and rapid breakthroughs in agentic AI are actively delivering new and more effective ways to tackle these issues.

What is agentic AI?

To better recognise its potential to address the challenge of white space in clinical trials, it helps to first understand what exactly agentic AI is, as well as what makes it distinct from more mainstream generative AI tools, such as OpenAI’s popular large language model (LLM), ChatGPT.

In simple terms, whereas generative LLMs are purely reactive – i.e., responding to a single prompt in real-time – agentic AI can be programmed to autonomously perform a series of steps to achieve a larger goal. This means when you give an AI agent a specific goal, it will begin thinking, planning, and taking actions all on its own to make it happen. For example, if you asked an AI agent to plan a family vacation, it would be able to break down the prompt into smaller sub-tasks, such as booking flights, hotels, and restaurants, creating a detailed itinerary, finding a boarding facility for pets, and any other relevant tasks on a traveller’s checklist – all while adapting its strategy to new information in real-time.

In this way, the use of agentic AI can be seen as uniquely compelling in an area like clinical development, as many tasks across the trial lifecycle are repeatable, rules-driven, and data-intensive. In fact, AI agents are specifically designed to address this combination of complexity and repetition, which also happens to be one of the primary drivers of white space and exactly what makes human-only processes so notoriously challenging to scale.

Toward a collaborative, framework-driven approach to agentic AI

With their ability to pull information from multiple sources and systems, apply relevant rules and guidelines, and execute decisions and workflows with or without human initiation or intervention, it’s easy to see how AI agents could be leveraged to automate routine, well-defined trial tasks. In fact, agentic AI has already proven effective in the automated management of various administrative and follow-up activities, as well as data processing and analysis, eliminating the need for humans to individually aggregate and analyse information stored across multiple disparate systems.

While agentic AI systems are incredibly powerful, it’s also crucial to remember that they come with their own set of inherent limitations, many of which make their fully autonomous operation in highly regulated contexts such as clinical trials risky, if not entirely unfeasible, without a human in the loop.

To ensure regulatory compliance and preserve the integrity of the research process, there are several mission-critical activities across the trial lifecycle that still require human oversight. For example, highly critical tasks surrounding regulatory submissions and documentation, clinical data management, and quality assurance and compliance all require the kind of medical and clinical expertise and judgement that AI agents simply aren’t currently capable of delivering. After all, agentic AI is still in its relatively early stages of development, and mistakes made in the absence of human oversight can lead to additional product delays, regulatory violations, and even potentially serious harm to patients.

Importantly, this does not mean that AI agents cannot be used to make these processes more efficient; it simply means there are certain areas in which AI should be used as a collaborative tool, rather than a replacement for human expertise. And, due to an agentic AI’s capacity to learn and enhance its performance over time, the need to optimise human-AI collaboration isn’t so much of a disadvantage as it is an opportunity to make further, more consequential reduction to white space in the future. Agentic AI solutions should be seen more as an augmentation of the humans operating within clinical development, freeing them from burdensome, administrative work to focus on higher-value, strategic tasks that accelerate delivery of new therapies to more patients.

Put simply, the benefits are there for the taking, but it comes down to choosing the right technology and strategy. In an ideal scenario, the agentic AI platform being used would be specifically designed for clinical development and able to incorporate proprietary, vertical-specific data, something that can’t be achieved using a more generic LLM like ChatGPT. Additionally, the platform would be leveraged alongside clear governance frameworks that define decision-making boundaries, approval hierarchies, escalation procedures, and the broader collaborative relationship between humans and AI agents.

Overall, if executed properly, this framework-driven, “human in the loop” approach to agentic AI can allow research teams to not only ensure regulatory compliance, but also maximise an agent’s capacity to learn, leading to the progressive expansion of intelligent automation and, most importantly, the near-permanent transformation of idle time into robust productivity, accelerated decision-making, and an unwavering confidence in delivering lifesaving therapies when patients need them most.

About the author

Andrew Mackinnon oversees the Medable Customer Value Team, ensuring that customers can successfully run efficient digitally enabled clinical trials. Mackinnon has more than 20 years of experience in managing clinical trials at large pharmaceutical, biotech, and CRO companies, most recently as a senior executive at Covance. From his key role in one of the earliest deployments of decentralised methodologies, Mackinnon has remained passionate about their benefits and looks to leverage his broad operational expertise to improve how this approach is utilised globally.