The explosion of generative AI and large language models (LLMs) across diverse industries has placed significant demands on professionals seeking innovative applications to expedite their work processes. This trend extends to the realm of clinical research.
Throughout the year 2023, there has been extensive conjecture surrounding the potential for generative AI to revolutionize various aspects of clinical research, including data capture, medical editing, and medical writing, among other essential functions.
Yet, it is crucial to temper expectations, as the current state of this technology does not quite align with the lofty anticipation. AI may have certain limited applications in TMF handling, but the technology cannot currently replace today’s electronic data capture (EDC) and trial master file (TMF) software.
The Future Potential of Generative AI
Generative AI tools and large language models hold considerable promise for the future due to their ability to handle repetitive tasks, thus allowing researchers to dedicate their expertise to more intricate endeavors.
These innovations have the potential to introduce profound efficiencies. As Atgeir Solutions Cofounder and Chief Product Officer Manish Kumar writes, LLMs could study trial data and medical literature to “foresee and identify impediments and inefficiencies that potentially compromise trial quality,” as well as help identify underrepresented groups to boost recruitment.
AI has already found a niche in managing TMF documents, note Timm Pauli at PharmaLex and Jim Nichols at Phlexglobal. “Moreover, use of such technologies has the potential to help build the bridge between the pharmaceutical client, their vendor, and the health authorities.
“That’s because advanced technology integration, using a continuous bidirectional data feed, enables a more natural exchange of information between systems where work generated by an outsourcing partner is fed back into a client company’s systems, as well as seamlessly shared with regulators.”
Envisioning Clinical Research With Future Applications in Electronic Health Records (EHRs)
It’s easy to envision a future where data is automatically captured, analyzed, and shared among all participants in clinical research, streamlining processes and enhancing collaboration.
Further, there are exciting future applications in electronic health records (EHRs), data which is notoriously hard to scrub. This is happening at smaller levels — e.g. at the hospital or regional level — but it’s not ready to scale.
“An ideal system would use a single model that can extract many types of information, work well at multiple hospitals, and learn from a small amount of labeled data,” Communications Manager Rachel Gordon at MIT Computer Science and Artificial Intelligence Laboratory writes.
Current Limitations of Large Language Models
Despite the promise of LLMs, several hurdles must be overcome before clinical researchers can build new processes around them.
One prominent obstacle is the presence of biases in these tools, as highlighted by Manish Kumar. “These biases make LLMs prone to misinterpreting complex medical information and drawing poor conclusions,” he writes.
“If not addressed, it can impact clinical trial outcomes. Because of their complexity, LLMs are usually called ‘black boxes,’ making it difficult to comprehend how they obtain their conclusions. A lack of transparency makes it difficult to analyze outcomes and conceal[s] biases and errors. Because of its powerfulness, human researchers may depend too heavily on LLMs rather than their knowledge and experience.”
Another critical issue lies in the quality of inputs and outputs. As Martina Mršnik and Dr. Christoph Engler write in their two-part series at CliNFo.eu (see part two here), LLMs don’t necessarily work from up-to-date data sources. Unlike humans, LLMs won’t “realize when information is missing or is not robust and may require further verification.”
And on the output side, there is the ongoing issue that some outputs “ … are plain false, are purely made up by the AI tool, and provided references are incomplete,” Mršnik and Engler write. “Deceptively, texts are so coherently and confidently written that they can be easily mistaken as a text created by a competent human being.”
Until these tools mature to the point of basic trustworthiness, their applications in collecting or analyzing clinical data are limited.
A Promising Alternative for Clinical Data Management
In contrast, our TrialMaster software has been purposefully designed to elevate data quality and streamline workflows in clinical research. It accomplishes this through:
- Simplification and time reduction in study set-up, allowing for the creation of forms that can be applied across all devices.
- Provision of auto-generated and parameterized edit checks, offering immediate, in-line feedback during data entry.
- Scalability to accommodate tens of thousands of patients in a single study.
Ready to explore the future of clinical data management?
Don’t miss out on the opportunity to see our TrialMaster software in action. Schedule your demo today and discover how Anju can enhance your clinical research processes while others catch up with the hype.
Images used under license from Shutterstock.com.