Human-Enabled AI Driving Enhancements in Clinical Trial Site Selection

Effective clinical trial site selection is crucial for the success of any clinical trial. It plays a pivotal role in ensuring timely patient recruitment, which, in turn, underpins overall trial success. The convergence of big data and artificial intelligence (AI) has opened new avenues for enhancing this critical aspect of clinical research. Here,  Travis Caudill,  Vice President, Feasibility, Site Identification & Clinical Informatics at ICON, explores the improvements that human-enabled AI is delivering.

How does AI contribute to improving clinical trials?

Healthcare and clinical research have embraced the value of data. Over the last decade, we’ve seen an explosion of digital technologies and an emphasis on generating healthcare data. With more data available than ever before, the approach has evolved to focus more on data management and facilitating more meaningful insight generation for specific use cases. Recent advancements in AI allow us to efficiently integrate, interrogate and interpret large datasets from diverse sources to produce valuable insights that guide optimised drug development decisions. In other words, it can take us from data overloaded to data empowered.

AI is contributing to improvements in clinical trials at various integration points across the development continuum. One promising contribution early in development is in enhanced site selection. Effective, accelerated site selection can improve a range of key milestones and contribute to overall project success. With a strategic approach, AI can turn mountains of data into actionable insight for intelligent site selection.

How does AI impact site selection processes for a clinical trial?

Employing AI in site selection has been transformative. It has reshaped the fundamental approach we use to understand and select sites for sponsors. As technology advances, companies are able to custom build AI and machine learning systems to fit specific niches and fulfil strategic needs. For example, we have built our AI site selection tool to ingest multiple data sources, provide visualisation and other tools necessary to manage and make sense of the abundant data. Through our multidisciplinary work to bring this technology to life for our specific purposes, we realised that starting with the study and building out a network of connections was key to discerning the meaningful data.

Using a bespoke AI system revealed the patterns within the clinical research ecosystem. It rapidly processes millions of overlapping data connections and how they fit together to tell us more about sites’ potential role in a study. Through the ecosystem approach, we determined new parameters for a site’s core clinical fit. Connectedness is a true marker of clinical fit as evidenced by the correlation between connectivity in the research network and the site’s likelihood of enrolling patients in clinical trials.

AI also enhances and accelerates our evaluations of potential sites through in-depth network feasibility analysis, which is guided by human analyst rankings of the value of potential relationships. AI can process all this data and weighted criteria exponentially faster than humans alone, and it clusters the sites based on enrolment performance, speed of start-up, and quality, etc. We use a four-tier system wherein the sites that meet requirements and have data determining they are more likely to enrol patients are categorised as tiers 1-2 and are fast-tracked for selection.

What is the importance of human-augmentation when using AI for site selection?

AI exponentially speeds up the site analysis, but human expertise is required for the more nuanced assessments required for selecting the most optimal sites. Although AI is highly powerful and quite elegant, it can fail to grasp the critical components of raw protocol details especially with rapidly emerging biomarkers and rare diseases. Human experts must identify the nuances in the protocol and prioritise key features that would define the best clinical fit from an investigative perspective.

Human expertise needs to bookend the AI computation. We set the parameters, defining site feasibility that AI uses to guide the computational processes, and analysts also review and refine the outputs. For example, in our four-tier system for site selection we are combining human-augmented intelligence and machine learning to rank and analyse this massive clinical research network. Tiers 1-2 are typically well evidenced to be highly suited for the trial, though tiers 3-4 may have significant potential unrecognised by the algorithm. Tier 3 sites may fit key requirements but, if they are new, they may have less data to provide on enrolment and therefore score lower. Given that we have a deeper capacity for understanding these up-and-coming sites and what they may need to realise their potential, it is reasonable to allocate a portion of a trial’s sites to this category.

What other aspects of site selection can AI assist in evaluating? 

After the initial core clinical fit assessment, we can layer additional data to refine the tiering system and build a more representative site profile. These layers prioritise elements that are important to the operation and success of the trial. Previous performance and enrolment speed, quality, diversity data and key population connectivity, and equipment and capabilities data from our existing site profile database can all be layered onto the initial analysis to provide a more complete picture of a site’s potential.

How do patients, the end-users of the development process, benefit from AI?

Human-augmented AI is enabling earlier and more impactful solutions to a significant challenge in clinical trials: diversity. In site selection, we can assess how connected sites are to target populations through various additional layers of data including claims, census, and third-party data. Selecting sites connected to underrepresented communities improves clinical trial access and trial diversity. Improving diversity affords more options for potentially meaningful treatment while also diversifying the clinical research data, in turn contributing to therapies that provide better outcomes for more diverse populations.

What tangible impacts has AI made when applied to site selection?

AI rapidly scales our ability to integrate, interrogate and interpret large datasets from diverse sources and produces valuable insights for site selection more efficiently. This efficiency has resulted in numerous significant improvements across all the expected metrics that set a clinical program up for success.

Compared to sites selected with traditional processes, sites selected through AI-enabled workflows demonstrated higher performance. We had a 50% reduction in non-enrolling sites and improved investigator engagement. They were 26% more likely to recruit subjects, and 24% more likely to hit their first patient in target, with similar performance for last patient in targets (21%).

Acceleration was a key factor since site selection and patient enrolment have high potential for delays with costly knock-on effects. However, the rapid processing power meant we had data-driven site proposals in as little as 48 hours and a demonstrated ‘time to first site selected’ averaging 25% faster, shaving weeks off the overall process.

What’s next?

Data is integral to clinical research and artificial intelligence will continue to evolve and advance alongside it. As AI becomes increasingly integrated into clinical trial operations, we must continue to develop our understanding of its uses and how to optimise its implementation to drive progress in the industry.

Looking ahead to the next steps for AI-enabled site selection, we are building on the success we’ve achieved thus far. We anticipate more granular analysis as we incorporate additional suitability layers for specific therapeutic areas or indications. For example, network proximity to predicted upcoming key opinion leaders in rare or ultra rare diseases, or incorporation of analysis components that can predict interrater variability. A clear data strategy powered by AI and human expertise that can adapt to the evolving data and technology landscape is critical to leveraging the full potential of these innovations.