Optimizing clinical trial site performance: A focus on three AI capabilities

This article, part of the IBM and Pfizer’s series on the application of AI techniques to improve clinical trial performance, focuses on enrollment and real-time forecasting. Additionally, we are looking to explore the ways to increase patient volume, diversity in clinical trial recruitment, and the potential to apply Generative AI and quantum computing. More than ever, companies are finding that managing these interdependent journeys in a holistic and integrated way is essential to their success in achieving change.

Despite advancements in the pharmaceutical industry and biomedical research, delivering drugs to market is still a complex process with tremendous opportunity for improvement. Clinical trials are time-consuming, costly, and largely inefficient for reasons that are out of companies’ control. Efficient clinical trial site selection continues to be a prominent industry-wide challenge. Research conducted by the Tufts Center for Study of Drug Development and presented in 2020 found that 23% of trials fail to achieve planned recruitment timelines¹; four years later, many of IBM’s clients still share the same struggle. The inability to meet planned recruitment timelines and the failure of certain sites to enroll participants contribute to a substantial monetary impact for pharmaceutical companies that may be relayed to providers and patients in the form of higher costs for medicines and healthcare services. Site selection and recruitment challenges are key cost drivers to IBM’s biopharma clients, with estimates, between $15-25 million annually depending on size of the company and pipeline. This is in line with existing sector benchmarks.^2,3

When clinical trials are prematurely discontinued due to trial site underperformance, the research questions remain unanswered and research findings end up not published. Failure to share data and results from randomized clinical trials means a missed opportunity to contribute to systematic reviews and meta-analyses as well as a lack of lesson-sharing with the biopharma community.

As artificial intelligence (AI) establishes its presence in biopharma, integrating it into the clinical trial site selection process and ongoing performance management can help empower companies with invaluable insights into site performance, which may result in accelerated recruitment times, reduced global site footprint, and significant cost savings (Exhibit 1). AI can also empower trial managers and executives with the data to make strategic decisions. In this article, we outline how biopharma companies can potentially harness an AI-driven approach to make informed decisions based on evidence and increase the likelihood of success of a clinical trial site.

Tackling complexities in clinical trial site selection: A playground for a new technology and AI operating model

Enrollment strategists and site performance analysts are responsible for constructing and prioritizing robust end-to-end enrollment strategies tailored to specific trials. To do so they require data, which is in no shortage. The challenges they encounter are understanding what data is indicative of site performance. Specifically, how can they derive insights on site performance that would enable them to factor non-performing sites into enrollment planning and real-time execution strategies.

In an ideal scenario, they would be able to, with relative and consistent accuracy, predict performance of clinical trial sites that are at risk of not meeting their recruitment expectations. Ultimately, enabling real-time monitoring of site activities and enrollment progress could prompt timely mitigation actions ahead of time. The ability to do so would assist with initial clinical trial planning, resource allocation, and feasibility assessments, preventing financial losses, and enabling better decision-making for successful clinical trial enrollment.

Additionally, biopharma companies may find themselves building out AI capabilities in-house sporadically and without overarching governance. Assembling multidisciplinary teams across functions to support a clinical trial process is challenging, and many biopharma companies do this in an isolated fashion. This results in many groups using a large gamut of AI-based tools that are not fully integrated into a cohesive system and platform. Therefore, IBM observes that more clients tend to consult AI leaders to help establish governance and enhance AI and data science capabilities, an operating model in the form of co-delivery partnerships.

Embracing AI for clinical trials: The elements of success

By embracing three AI-enabled capabilities, biopharma companies can significantly optimize clinical trial site selection process while developing core AI competencies that can be scaled out and saving financial resources that can be reinvested or redirected. The ability to seize these advantages is one way that pharmaceutical companies may be able to gain sizable competitive edge.

AI-driven enrollment rate prediction

Enrollment prediction is typically conducted before the trial begins and helps enrollment strategist and feasibility analysts in initial trial planning, resource allocation, and feasibility assessment. Accurate enrollment rate prediction prevents financial losses, aids in strategizing enrollment plans by factoring in non-performance, and enables effective budget planning to avoid shortfalls and delays.

It can identify nonperforming clinical trial sites based on historical performance before the trial starts, helping in factoring site non-performance into their comprehensive enrollment strategy.
It can assist in budget planning by estimating the early financial resources required and securing adequate funding, preventing budget shortfalls and the need for requesting additional funding later, which can potentially slow down the enrollment process.

AI algorithms have the potential to surpass traditional statistical approaches for analyzing comprehensive recruitment data and accurately forecasting enrollment rates.

It offers enhanced capabilities to analyze complex and large volumes of comprehensive recruitment data to accurately forecast enrollment rates at study, indication, and country levels.
AI algorithms can help identify underlying patterns and trends through vast amounts of data collected during feasibility, not to mention previous experience with clinical trial sites. Blending historical performance data along with RWD (Real world data) may be able to elucidate hidden patterns that can potentially bolster enrollment rate predictions with higher accuracy compared to traditional statistical approaches. Enhancing current approaches by leveraging AI algorithms is intended to improve power, adaptability, and scalability, making them valuable tools in predicting complex clinical trial outcomes like enrollment rates. Often larger or established teams shy away from integrating AI due to complexities in rollout and validation. However, we have observed that greater value comes from employing ensemble methods to achieve more accurate and robust predictions.

Real-time monitoring and forecasting of site performance

Real-time insight into site performance offers up-to-date insights on enrollment progress, facilitates early detection of performance issues, and enables proactive decision-making and course corrections to facilitate clinical trial success.

Provides up-to-date insights into the enrollment progress and completion timelines by continuously capturing and analyzing enrollment data from various sources throughout the trial.
Simulating enrollment scenarios on the fly from real time monitoring can empower teams to enhance enrollment forecasting facilitating early detection of performance issues at sites, such as slow recruitment, patient eligibility challenges, lack of patient engagement, site performance discrepancies, insufficient resources, and regulatory compliance.
Provides timely information that enables proactive evidence-based decision-making enabling minor course corrections with larger impact, such as adjusting strategies, allocating resources to ensure a clinical trial stays on track, thus helping to maximize the success of the trial.

AI empowers real-time site performance monitoring and forecasting by automating data analysis, providing timely alerts and insights, and enabling predictive analytics.

AI models can be designed to detect anomalies in real-time site performance data. By learning from historical patterns and using advanced algorithms, models can identify deviations from expected site performance levels and trigger alerts. This allows for prompt investigation and intervention when site performance discrepancies occur, enabling timely resolution and minimizing any negative impact.
AI enables efficient and accurate tracking and reporting of key performance metrics related to site performance such as enrollment rate, dropout rate, enrollment target achievement, participant diversity, etc. It can be integrated into real-time dashboards, visualizations, and reports that provide stakeholders with a comprehensive and up-to-date insight into site performance.
AI algorithms may provide a significant advantage in real-time forecasting due to their ability to elucidate and infer complex patterns within data and allow for reinforcement to drive continuous learning and improvement, which can help lead to a more accurate and informed forecasting outcome.

Leveraging Next Best Action (NBA) engine for mitigation plan execution

Having a well-defined and executed mitigation plan in place during trial conduct is essential to the success of the trial.

A mitigation plan facilitates trial continuity by providing contingency measures and alternative strategies. By having a plan in place to address unexpected events or challenges, sponsors can minimize disruptions and keep the trial on track. This can help prevent the financial burden of trial interruptions if the trial cannot proceed as planned.
Executing the mitigation plan during trial conduct can be challenging due to the complex trial environment, unforeseen circumstances, the need for timelines and responsiveness, compliance and regulatory considerations, etc. Effectively addressing these challenges is crucial for the success of the trial and its mitigation efforts.

A Next Best Action (NBA) engine is an AI-powered system or algorithm that can recommend the most effective mitigation actions or interventions to optimize site performance in real-time.

The NBA engine utilizes AI algorithms to analyze real-time site performance data from various sources, identify patterns, predict future events or outcomes, anticipate potential issues that require mitigation actions before they occur.
Given the specific circumstances of the trial, the engine employs optimization techniques to search for the best combination of actions that align with the pre-defined key trial conduct metrics. It explores the impact of different scenarios, evaluate trade-offs, and determine the optimal actions to be taken.
The best next actions will be recommended to stakeholders, such as sponsors, investigators, or site coordinators. Recommendations can be presented through an interactive dashboard to facilitate understanding and enable stakeholders to make informed decisions.

Shattering the status quo

Clinical trials are the bread and butter of the pharmaceutical industry; however, trials often experience delays which can significantly extend the duration of a given study. Fortunately, there are straightforward answers to address some trial management challenges: understand the process and people involved, adopt a long-term AI strategy while building AI capabilities within this use case, invest in new machine learning models to enable enrollment forecasting, real-time site monitoring, data-driven recommendation engine. These steps can help not only to generate sizable savings but also to make biopharma companies feel more confident about the investments in artificial intelligence with impact.

IBM Consulting and Pfizer are working together to revolutionize the pharmaceutical industry by reducing the time and cost associated with failed clinical trials so that medicines can reach patients in need faster and more efficiently.

Combining the technology and data strategy and computing prowess of IBM and the extensive clinical experience of Pfizer, we have also established a collaboration to explore quantum computing in conjunction with classical machine learning to more accurately predict clinical trial sites at risk of recruitment failure. Quantum computing is a rapidly emerging and transformative technology that utilizes the principles of quantum mechanics to solve industry critical problems too complex for classical computers.