Flexible Clinical Modeling: How Advanced Analytics and AI/ML Can Help Ensure Effective Patient-Centered Drug Development

Patient-first clinical trial design is a priority for research and development stakeholders in a post-pandemic era. Gathering intelligence about patient preferences and considerations can help shape trial programs, leading to trials designed with patients rather than for them. Offering patients flexibility in trial participation helps address some long-standing recruitment, engagement and retention issues.

Flexible modeling via tech-enabled advancements is a tangible way pharmaceutical and biotechnology sponsors can enhance patient-centered trial design and allow for adjustments during a trial. Though sponsors have been interested in flexible modeling for some time, it is through current advanced analytics and artificial intelligence (AI) and machine learning (ML) capabilities that sponsors can genuinely view patient participation as a journey, potentially improving trial success.

For one, sponsors, clinical research organizations (CROs) and study teams can gain a better understanding of the types of challenges patients face (i.e. physical, emotional, financial or logistical). Also, within the end-to-end trial process, sponsors need to gauge what areas need improvement for stronger participation, such as an easier-to-follow informed consent process. Building trust with patients is paramount for all in R&D. Valuing patients as partners in the process ensures they have a more meaningful experience and are more likely to continue engaging and even discuss their experiences with others.

Additionally, by having options to shift the trial course quickly, sponsors are offered flexibility that can improve operational efficiency and research outcomes that positively impact patients. Reducing sponsor burden is critical in a complex R&D landscape with impacts from broader economic changes.

To take full advantage of flexible modeling, understanding how advanced analytics and AI/ML solutions can provide the granular insights needed to course correct during trials is key. Below are a few examples of how these comprehensive modeling tools and expert guidance can help determine what specific adaptations can optimize trials.

Validating assumptions on trial feasibility

One of the biggest concerns for sponsors and CROs when planning studies and designing trial protocols is ensuring each decision has a positive impact on study outcomes. When asking themselves, “How likely is it that the trial – as designed – is going to meet enrollment targets and timelines?”, it has been historically difficult to gauge feasibility outside of reviewing similar trial programs that were completed. But for new classes of treatments and other unique therapies, there may not be previous trial programs to reference.

Applied analytics and AI methodologies can mitigate potential risk prior to operationalization by providing critical insights about variables that can impact trial success, including:

• Identifying optimal country and site mix to inform study enrollment strategies
• Finding pools of potential patients likely to be eligible for the trial
• Accurately assessing the potential time and resources needed for optimized trial execution
• Minimizing challenges for trial delays by:

            – Assessing protocols to mitigate avoidable amendments and deviations

            – Identifying the right patient subpopulations to reduce screen failure rates                and keep enrollment on track

            – Incorporating diversity and inclusion models, identifying targeted ranges                of patients to meet regulatory requirements for underserved populations

Researchers and study teams can then make evidence-based decisions about how feasible it will be to run the trial and what adjustments are necessary to ensure enrollment targets will be met. For example, predictive insights from modeling can shed light on how difficult patient enrollment may be, so adjustments in eligibility criteria can be made if needed. The study team may determine that targeted recruitment in specific geographies or sites is a better approach. Or, if modeling predicts that a trial is too resource-intensive, the study team can consider the pros and cons of added budget or resources.

Finessing trial site selection

As sponsors gauge what is needed for a study’s operational success, trial sites – a key factor in the process – must be considered. As sites differ in many aspects, fine-tuning site selection processes early in trial planning is beneficial.

Variables to site selection can include:

• Relevant patient population at the site or nearby facilities
• Availability to participate
• Sites’ historical performance, including healthcare professional team and providers, individual site recruitment rates and capacity for related resources needed to conduct trials

The multilayers of considerations in site selection are peeled with flexible modeling, uncovering details regarding each of the variables noted above. Applying AI/ML-trained models to site-level data allows for deeper insights, including:

• Pinpointing potential sites in geographies with higher concentrations of target patient populations
• Assessing each site’s capability to effectively screen, enroll and retain patients
• Extracting from the above findings to identify sites that may have the necessary resources, such as adequate staffing, equipment (e.g., phlebotomy supplies and data-collection tools), facilities, etc. to manage trial activities

From a solid list of sites to consider, the study team can develop an accurate risk assessment for each site, comparing potential challenges that could impact trial operations and ability to capture quality data outcomes and provide stronger patient care.

To customize site selection based on therapeutic needs, such as for Alzheimer’s disease (AD) patients, modeling can help find sites in areas with higher patient populations and then identify potential challenges to address. Typically, AD trials involve intricate study designs, and modeling can help identify sites that have successfully delivered protocols with biomarkers, imaging techniques and multiple endpoints. Also, due to the slow progression of AD, trials may require long follow-up periods to assess treatment efficacy. Modeling can help identify sites adequately prepared for long study durations.

Customizing patient outreach

Ideally, sponsors, CROs and study teams aim to reduce guesswork within many trial activities, with recruitment, enrollment and engagement at the top of the list. What traditionally is not feasible in a timely manner through manual efforts can be done rapidly by using flexible modeling to collect and extract insights to build out trial design, optimize recruitment and ongoing engagement and gather details on patient demographics as well as needs for the trial, therapeutic nuances, etc.

Flexible modeling leveraging cutting-edge technologies (e.g., large language models (LLMs), ML, causal inference and simulation techniques), allows trial design to change as needed. Using ML or LLMs, sponsors can use new data collection methodologies to extract patterns of patient outcomes that would not be seen otherwise. Patterns of interest may be early-warning signs of upcoming adverse events, unfavorable toxicity profiles or the potential treatment impact on patients’ quality of life. This information helps secure the most clinically relevant and accurate data to drive smarter decisions for patient care.

From there, study teams can further customize outreach to patients from trial recruitment, enrollment and retention. Everything from educational trial information, gauging interest in participation and asking how to improve their experiences can be tailored. Examples include:

• Adapting email content to address patients’ needs, interests and concerns
• Developing engaging surveys to gauge patients’ interest in trial participation
• Creating dedicated social media channels for the study, where patients can ask questions, share experiences and connect with other participants, fostering a sense of community and support
• Organizing online support groups or webinars where patients can discuss their experiences, ask questions and receive guidance from experts and fellow trial participants
• Developing user-friendly mobile apps that allow patients to track progress, receive trial activity reminders and access educational resources

Where does flexible modeling go from here?

Flexible modeling can be complicated, especially when leveraging evolving tech-enabled methodologies. It requires the right expertise to gauge what insights are most helpful to extract, identify data patterns and guide design direction accordingly. Integrating flexible modeling approaches into the clinical trial plan before the study begins may mean a bit more work and resources upfront for sponsors. However, this legwork can create significant value for all stakeholders in the long run as evidence-driven decisions also incorporate patient feedback to improve trial outcomes for them. The examples above are just a sampling of how these technologies can help create quality-driven experiences for patients while improving outcomes.

Future developments in this space may include AI-powered chatbots and virtual assistants that improve overall patient engagement and satisfaction throughout the trial, providing patients with tailored answers to questions and addressing concerns in real-time. LLMs could analyze trial data as it’s collected, identifying trends and potential issues earlier, helping study teams make data-driven decisions and adapt trial design accordingly. LLMs may also analyze unstructured data (e.g., patient feedback and clinical notes) to identify patterns and insights that can inform trial design and improve patient experiences.

About the author:

Greg Lever is director of AI solutions delivery at IQVIA. With more than 13 years of life sciences and technology experience, Greg currently helps clients discover innovative ways to bring life-changing therapies to patients faster within IQVIA’s Applied Data Science Center’s consulting sales team. Previously, he led a team of machine learning engineers within the Analytics Center of Excellence. 

Greg has worked with several technology startup companies in London and helped see Genomics England’s 100,000 Genomes Project through project completion. He received his PhD at the University of Cambridge, combining quantum physics and ML to develop new approaches for small-molecule drug discovery and has worked as a postdoctoral associate at MIT.