By analyzing a decade of data on how artificial intelligence is used in cancer clinical trials, UF Health Cancer Institute researchers have provided a snapshot of how AI may improve and personalize cancer care.
By systematically reviewing completed U.S.-based cancer clinical trials that incorporated AI, the research team provided snapshots of how these technologies are being tested in practice and where more rigor is needed to ensure AI advances translate into evidence-based cancer care.
“AI has enormous potential to improve cancer care, but its benefits depend on rigorous clinical evaluation, transparent reporting, and training on diverse and representative datasets,” said Shama Karanth, Ph.D., an assistant professor in the UF Department of Surgery and senior author of the new study, published in the journal Cancers. “Our findings suggest that while momentum is growing, substantial work remains to ensure AI-based approaches are robust, generalizable and capable of benefiting all patients.”
Researchers conducted a review of completed U.S.-based cancer clinical trials registered on ClinicalTrials.gov to identify trials that incorporated AI. For each eligible trial, they pulled out information on study design, enrollment size, cancer type and specific AI methods used. Next, they categorized each trial using the Cancer Control Continuum, a framework that covers the spectrum of cancer care, from cancer etiology, to prevention and detection, to diagnosis and treatment, to survivorship.
“Although AI is widely promoted as transformative in oncology, there has been limited information on how it is being evaluated in clinical trials, for what purposes and across which cancer types,” said Karanth, a member of the UF Health Cancer Institute’s Cancer Control and Population Sciences research program.
The team identified 50 completed clinical trials that met the criteria. Most were interventional studies, meaning they tested a specific intervention like a drug, device or therapy to see how it affected outcomes.
They found machine learning was the most used AI approach and most trials focused on cancer detection. Colorectal cancer and unspecified cancer types were the most common focus, likely reflecting the availability of large, well-structured datasets suitable for AI development.
Importantly, the team also found significant gaps. Many trials did not report results or link them to published findings, making it difficult to assess how AI performed and its clinical impact. In addition, most studies were conducted at single institutions.
“That raises concerns about how well these tools will generalize to broader and more diverse patient populations,” Karanth said. “These findings underscore the importance of increased transparency, validation and multi-center collaboration in AI-driven cancer research.”
Next, the researchers plan to incorporate AI methods directly into their own cancer research. They plan to use insights from the review to guide rigorous study design, validation and reporting.
“We want to contribute evidence on where and how AI can meaningfully improve cancer care,” Karanth said.
UF Health Cancer Institute researchers Aline Fares, M.D., Ali Zarrinpar, M.D., Ph.D., and Dejana Braithwaite, Ph.D., are co-authors on the study, which received funding from the UF Health Cancer Institute and National Institutes of Health, National Institute of Dental and Craniofacial Research. Collaborators at the University of Miami Leonard M. Miller School of Medicine, UT Health San Antonio School of Dentistry and MD Anderson Cancer Center also contributed.
