Using AI to Boost Brain Tumor Clinical Trial Enrollment

When too few people enroll in clinical trials testing new therapies for brain cancer, the findings are less likely to apply to the broader population of people living with the disease. 

Now researchers at UC San Francisco have developed a new AI tool that predicts which factors are most likely influencing whether an individual with glioma decides to enroll in a clinical trial. 

The findings, which appear in Science Advances, could help guide how physicians approach discussing trials with their patients and how clinics develop new interventions to meet their accrual goals. 

“Many trials fail because of inadequate enrollment,” said Shawn Hervey-Jumper, MD, a UCSF neurosurgeon, the Mitchel S. Berger Endowed Professor, and the study’s co-senior author. “This work takes us a step closer to solving the problem of who gets into a clinical trial and how we can enroll more people.”

Obstacles to participating in clinical trials

Hervey-Jumper and his colleagues at the UCSF Brain Tumor Center previously reported a discrepancy between how many patients are diagnosed with glioma and how many enroll in a clinical trial. In a retrospective study, they found that less than 18 percent of patients who are newly diagnosed with glioma enroll in a trial, and at recurrence, only about 26 percent of patients enroll. His team further showed in 2022 that neuro-oncology clinical trials accrued significantly fewer women and nonwhite people. 

“We thought machine learning techniques could help us understand the complex interaction between the variables influencing trial enrollment,” said Mulki Mehari, MD, a recent medical student graduate from the Hervey-Jumper lab and this new study’s first author.

She and her colleagues trained AI models on a dataset of 1042 patients who enrolled in therapeutic clinical trials at UCSF over a 20-year timespan. The scientists then validated the models using clinical trial participation data from Duke University, the University of Michigan, and the Dana Farber Cancer Institute.

This approach helped them identify which factors are most influential for patients with glioma as well as for specific groups of patients. Clinical factors – like the tumor location and a history of seizures and treatment with chemotherapy – played the biggest role in whether patients discussed clinical trials with their physician and decided to participate. But for women and nonwhite patients, factors like the patient’s employment status, their occupation, and their health insurance were more important.

The researchers then created an open-access tool called GPREDICT that estimates the probability that a patient will enroll in a clinical trial based on their health history as well as factors like insurance status, the median household income in their neighborhood, and the distance from the home to the hospital. In addition to the likelihood of enrollment, the tool offers specific recommendations to address the patient’s circumstances.

Finding data-driven interventions for patients

Hervey-Jumper and his team are now interested in seeing how using this enrollment tool in clinical settings could improve clinical trial accrual. Mehari, who received awards for this research from the Society for Neuro-Oncology and the Joint Section on Tumors of the American Association of Neurological Surgeons and the Congress of Neurosurgeons, will continue working this project as she begins her neurosurgery residency at Tufts University.

She notes that they have already identified several potential interventions. For example, the researchers found that physicians don’t talk with all their patients about clinical trials and were less likely to discuss trials with women and minorities.

“This may mean eligible patients who would otherwise be interested in enrolling never do so because they are unaware of the possibility,” Mehari said.

She and her colleagues think that by adopting a “just-ask” policy and keeping abreast of all open clinical trials at their institution, physicians could increase enrollment rates. 

Patients whose first language was not English were less likely to enroll in a clinical trial. Mehari says these findings suggest the physicians need to more cognizant of potential language barriers. 

“It is critical to always use medical interpreters or certified bilingual providers for these complex clinical trial screening discussions,” she said.

The researchers also found that a patient is more likely to enroll the more times their physician talks about clinical trials with them. Bringing up trials at different time points over the course of their treatment could be a key strategy to drive accrual. 

“This is an issue of scientific accuracy,” Hervey-Jumper said. “The patients who enroll in trials currently do not represent the geographic makeup of patients with the disease, which is a major limitation in making any progress for this universally fatal diagnosis.”

Reference: Mehari M, Warrier G, Dada A, Kabir A, Haskell-Mendoza AP, Tripathy A, Jha R, Nieblas-Bedolla E, Jackson JD, Gonzalez AT, Reason EH, Flusche AM, Reihl S, Dalton T, Negussie M, Gonzales CN, Ambati VS, Desjardins A, Daniel AGS, Krishna S, Chang S, Porter A, Fecci PE, Hollon T, Chukwueke UN, Badal K, Molinaro AM, Hervey-Jumper SL. Predicting therapeutic clinical trial enrollment for adult patients with low- and high-grade glioma using supervised machine learning. Sci Adv. 2025 Jun 6;11(23):eadt5708. doi: 10.1126/sciadv.adt5708.