We've updated our Privacy Policy to make it clearer how we use your personal data. We use cookies to provide you with a better experience. You can read our Cookie Policy here.

Advertisement

AI and GenAI Innovation at the Asset, Clinical Program and Individual Study Levels

Female scientist holding a tablet. She is surrounded by icons depicting drug discovery and AI.
Credit: iStock.
Listen with
Speechify
0:00
Register for free to listen to this article
Thank you. Listen to this article using the player above.

Want to listen to this article for FREE?

Complete the form below to unlock access to ALL audio articles.

Read time: 7 minutes

As pharmaceutical and biotech companies continue to hunt for new approaches for improving clinical trials, ChatGPT has made “generative artificial intelligence” (GenAI) a mainstream term. GenAI is transforming how those in drug discovery and development approach decision-making. It is becoming a powerful tool for automating what have been manual documentation processes in clinical trials and reducing administrative workloads for study teams. R&D machine learning scientists and engineers are building evidence-based use cases for GenAI and large language models to push the boundaries of traditional R&D even further.

 

In lockstep with clinical trial sponsors, therapeutic experts, clinical trial experts and others, machine learning (ML) scientists and engineers are seeing that the true innovation of GenAI is in how it can be used to call on forms of artificial intelligence (AI) to create a multi-agent framework, a “dream team” of large language models and others working with human experts to offer more depth and precision to resolving longstanding and complex R&D challenges. This human-machine collaboration will allow trial sponsors to dive deeper into looming questions about critical unknowns in R&D strategies – which indications to pursue, how to design a clinical trial program for technical and commercial success and how to ensure operational success of clinical trials. Being able to extract layered, connected insights for analysis offers sponsors the prospect of evidence-driven answers to better guide these decisions.

 

Below, we discuss how the toolbox of GenAI agents can make drug development decision-making more informed about assets, clinical development programs and individual studies.

 

Understanding types of AI-driven outcomes: What do you need to know?

GenAI models can only provide thorough, accurate and useful outputs if they align with what R&D stakeholders are aiming to achieve and if they are based on curated, connected data. If done properly, AI may help uncover insights that sponsors did not even realize were possible. It is also important that trial sponsors work with ML scientists and engineers who are experienced in devising solutions for clinical development to bring together subject-matter expertise, knowledge of domain-relevant data and AI skills.

 

With this, AI can help sponsors with three types of analysis and outcomes: 

Descriptive outcomes

Deploying AI agents on historical data can answer factual questions about past events and trajectories related to assets, indications and clinical trials. For example, when evaluating the effectiveness of a trial protocol, seeing how trials with similar protocols have fared is useful. AI can help pull and organize all trials run with this indication and help sponsors see how their protocol compares to similar designs, as well as which design elements have the greatest impact on trial complexity and burden.

 

Predictive outcomes

We have historically built supervised learning models to predict technical, regulatory and commercial success of assets. With GenAI, we can improve these models by accessing features from unstructured data. Additionally, we can use GenAI agents to make these predictive models answer questions about the future trajectory of an asset, mechanism of action or indication of interest. These agents can provide insight into questions like: 

  • What is the probability of our product’s technical and regulatory success?
  • What may be the real-world uptake of our therapy once available?
  • What are the likely regulatory or commercial outcomes for alternative clinical development programs?

 

Generative outcomes

A GenAI-driven multi-agent approach is particularly useful for multi-step analyses. For example, if sponsors are interested in understanding what the standard of care within a given therapeutic indication may be in five years, they can use different agents to answer this question step-by-step: 

  • An agent that plans the answer to the question by breaking it into parts to be answered by different sources of data.
  • Sub-agents that get insights from each source of data, such as from treatment guidelines to understand current standard of care or from clinical trial databases to identify trials running in the indication that will launch within five years.
  • An agent that synthesizes the answers from the sub-agents to provide a complete answer to the user.

 

This demonstrates the power of agentic AI to help with different types of analyses and outcomes. However, this is not easily done and requires a combination of domain expertise, ability to curate and connect data and the ability to build an AI agentic system that works.

 

GenAI for clinical R&D optimization 

It is essential that stakeholders keep in mind that using GenAI for enhanced drug development is an ongoing journey. The promise it holds to optimize R&D strategies will grow with time and with dedicated effort by experts to fine-tune how models work together.

 

From the overall picture of the asset all the way downstream to individual trials, multi-agent GenAI frameworks will offer unique benefits to inform smarter decisions.

 

At the asset level

Drug developers use a multitude of information to place values on their assets. Traditionally, this was a lengthy manual process that was heavily dependent on time, access to data and subject-matter expertise. Leveraging AI to learn from the totality of historical data across multiple sources can help sponsors prioritize their pipelines and business goals better.

 

For one, AI helps shed light on indication identification and prioritization. To determine which indications to consider pursuing, sponsors can use AI to learn from similar assets, indications and trials and rank indications of interest based on potential for technical success, how strong the unmet need is and the likelihood, speed and magnitude of commercial success. For example, using AI for an oncology asset, the developer may wish to understand what types of cancer the molecule is best suited to target. Until now, drug developers and CROs have evaluated indication insights by manually selecting a few assets for comparison. With AI, they can take the data and understanding available from a breadth of historical trials to compare systematically across multiple scenarios, offering faster, more accurate and comprehensive options for consideration.


From there, trial sponsors can create scenarios to determine the optimal sequence of trials for the program.

 

At the clinical development program level

Via AI methodologies, trial sponsors can simulate outcomes for different clinical development pathway scenarios to evaluate more options to inform a better and more grounded clinical development program. This includes: 

  • Likely differentiation based on current and emerging competitive landscape.
  • Technical and regulatory outcomes.
  • Costs and time to completion outcomes.
  • Feasibility outcomes based on country, site and investigator performance.
  • Reimbursement, pricing and anticipated product uptake outcomes.

 

At the individual study level

Sponsors can inform better trial design at an individual study level by using AI to understand current standard of care and likely future updates, relevant endpoints, inclusion/exclusion criteria, development costs and timelines, and likely reimbursement, price and uptake outcomes. Below, we pinpoint a few tangible examples at various stages of a study:   

  • During study design development, generating “what if” scenarios can help drug developers review different eligibility criteria and parameters to determine what may help achieve recruitment goals and to quantify the impact of each option. For example, when aiming to increase the patient pool of a type 2 diabetes study, AI-driven analytical modeling using real-world data insights can generate scenarios for what likely variables will help meet that goal and by how much. The analysis may show that decreasing the lower limit of the required HbA1c range by just 0.5% may increase the number of eligible patients considerably. Examining variations of that modeling can help sponsors select criteria to maximize enrollment potential.

  • Conducting protocol analysis before finalization can be key to reducing amendments and related time delays. It is critical to know more about patients’ willingness to participate based on what they believe is expected of them to improve recruitment and retention. With patient-centric insights derived from clinical journeys, real-world data, etc., advanced analytics applications can assess several layers of insights that help predict patient satisfaction and recruitment success:

o   Using those analytics and other insights, it is possible to score protocol elements to quantify patient burden via an AI-driven algorithm. Using burden scoring for each element, sponsors can examine whether protocols are at the expected level of burden when compared to trial protocols for a similar phase, therapeutic area, disease, etc.

o   From there, it is possible to understand patient burden based on specific trial design elements that impact willingness to participate in trials, which a sponsor may want to consider changing to minimize recruitment or retention risks downstream. For example, in a lung cancer study, key study demographics may indicate that a fresh biopsy is overly burdensome. Patient burden analytics can identify this and demonstrate the impact on burden if an archived biopsy is used instead.

o   Going even further, protocol burden can be broken down according to race, ethnicity and country to understand how study design elements might impact recruitment and retention across various demographics and geographies. For example, patients from some countries may consider a fresh biopsy very burdensome, which could impact enrollment if lung cancer is more prevalent in those populations.

  • Historically, sponsors and CROs have depended on the predictability and reliability of core countries and sites for conducting their studies. But this can lead to overdependence. To ensure diversified country and site selection, AI-driven insights generated by global data are helping to develop selection strategies based on standard of care, trial competition, regulatory requirements and disease prevalence:

o   GenAI-powered automated strategies enable quick iterations of trial scenarios. For example, a sponsor can consider the impact various parameters have on constraints like trial duration and cost. Then, adjustments can be made using predictive analytics to optimize the country and site mix.

  • Currently, trend analysis can be used to support site and enrollment predictability. In the future, both historical and recent performance trends may be simultaneously assessed, creating real-time data and feedback loops that enable the models and increase accuracy while also accounting for potential risks.

  • Once the trial strategy is in place and start-up has been initiated, active trial management leveraging predictive analytics that generate enrollment models allows sponsors to monitor their studies in real-time. GenAI can alert study teams when enrollment is lagging within particular regions and sites, which allows sponsors to course-correct earlier. 



Building on multi-AI agent frameworks to answer diverse and deeper questions  

When discussing the potential of GenAI, especially for connecting models to secure the layers of insights needed to drive stronger development strategies, it can get complicated. But it is through this level of thoroughness and depth that R&D stakeholders advance with more confidence, knowing they have reduced or eliminated key unknowns in a demanding landscape.

 

AI/ML is creating building blocks of insights to further enhance end-to-end drug development efforts. This will allow drug developers and CROs to drill down even further and expand potential scenarios for evaluation to make better decisions to improve healthcare worldwide.