AI in Clinical Trials – Hype vs. Reality

There’s been substantial Artificial Intelligence (AI) momentum growing over the years – some believe it will revolutionize clinical research and development and is the solution to many challenges and inefficiencies. But what can AI do in reality? This question was the motivation behind the panel, AI in Clinical Trials – Hype vs. Reality, at the latest CORE event produced by Halloran and PLG.  

Moderated by Karen Travers, Principal Consultant at Halloran, the panel featured: 

• Malachi Bierstein, CEO, ScienceMedia
• Jin Kim, CEO & Founder, Miracle 
• Michael Goedde, President, BioForum 
• Marc Bellemare, Chief Scientific Officer, Reliant AI 

Realistically, over the past 18 months, AI use in clinical research and development has accelerated. For companies without a data science or AI group in-house, AI can be leveraged practically and realistically throughout the various stages of a clinical trial, supported by concrete, definable results. While there are many solutions to various needs, this discussion sought to balance hype with realistic applications and outcomes. 

Biggest AI Challenges Over the Past 18 Months 

A panelist shared that we often overestimate the amount we’re going to get done in one year and underestimate the amount that we’ll get done in ten years. Does that resonate?  

While ChatGPT was launched almost three years ago, we’re now at a point where AI is really beginning to take off, offering specific tasks to alleviate workloads often at a faster, more efficient rate than current practices. 

Benefits of AI 

• Handle increased data volume and complexity 
• Manage diversity of data sources 
• Reduce repetitive or manual data processing steps 
• Address resource constraints and costs 

While there is great potential, there are still concerns propelling resistance, with the most common one being security and privacy. With that in mind, to what extent can ChatGPT be used? And when data is siloed (and intentionally not uploaded into ChatGPT for various reasons), how can AI truly provide a comprehensive insight?  

Another challenge is the hype curve – executive leaders pushing AI use on their teams for experimental purposes without a clear roadmap – but that approach can often lead to confusion, a lack of efficiency, and improper usage.  

Where to Begin? 

Given that many sponsor companies, especially small or emerging, are likely not going to build their own AI platforms and deploy a team of data scientists, how do they go about reaping the benefits of available AI tools? 

Here is a summary of their recommendations: 

• Define and articulate the problem, workshop the solution, and maybe the solution is an AI tool (or not) 
• Engage with vendors that listen to the real problem, not trying to force their solution 
• Make the most informed decision based on key factors that may impact internal teams and infrastructure 

While there may be pressures from executives (and investors) to use AI to generate more investments, the panelists shared recommendations to take a step back from that mindset. By focusing on assessing where teams are struggling and spending most of their time, paired with an open and curious mindset to accompany the right solution, teams are better equipped for the efficiency gains conversations investors may be looking to hold. 

Common Pitfalls with AI Use 

One of the panelists shared examples around AI-driven data review, often focused on mining for anomalies. However, it’s not just about outliers. Rather, it’s the data that looks normal that may cause an alert of being too normal. 

Another panelist shared how the rationale for AI pilot use is a key factor in its success. To reap the benefits of everyone’s time and efforts, it was encouraged to identify a single project or task that is vital or burdensome and truly focus on just that instead of more widespread experimental use. 

Demonstrating AI Tools are Compliant 

Furthering the discussion around trust with customers (and employees), the panelists shared insights on how to demonstrate that AI tools are both fit-for-purpose and compliant. In summary, factors include: 

• AI model must be reproducible as possible 
• The site and Clinical Research Organization (CRO) staff must be protocol compliant, designed and outlined by the sponsor 
• Protocol comprehension with human verification must be built in 

While regulatory bodies are open and encouraging AI use in clinical trials, the call-to-action for sponsors is to be mindful of their approach with AI, pairing the right problem with the right solution, whether that be AI or not. 

If we look ahead at the next five years when AI can be most useful, informational models will continue to provide a great opportunity. For most, sifting through extensive research and material just isn’t feasible with the time available. But going forward, human-in-loop validation is still essential for clinical research, development, and operations teams. 

To learn more about the transformative power of AI shaping the life sciences industry, click here to download PLG’s research.