Skip to main content Skip to secondary navigation
Page Content

AI Expands the Reach of Clinical Trials, Broadening Access to More Women, Minority, and Older Patients

Using artificial intelligence to scour medical records, researchers were able to double the number of patients who qualified for clinical trials, while expanding the pool’s gender, age, and racial profile.

A health care worker stands over a patient at a desk, going over paperwork during a clinical trial.

A health care worker helps a patient with paperwork during a clinical trial of tests for the coronavirus. | Carl Recine/Reuters

Pharmaceutical companies spend years and billions of dollars developing new drugs. While most drugs never make it out of the lab, a precious few reach clinical trial — a real-world test on human patients. And yet, despite that hard work and long odds, many clinical trials must be abandoned or sent back to the drawing board because they fail to enroll enough patients in the study.

“A failed or delayed clinical trial is costly on many fronts,” says James Zou, a professor of biomedical data science at Stanford and a member of the Institute for Human-Centered Artificial Intelligence, who co-led a collaboration between Stanford and researchers at Genentech that set out to see if AI could rectify the problem.

Trial delays can add hundreds of millions of dollars to the price tag for a new drug and also be a profound emotional disappointment for the researchers working on it. But perhaps most important, Zou notes, delays mean that many needful patients cannot access new, potentially life-altering, medicines.

The team focused on a deadly form of cancer, known as advanced non-small cell lung cancer — aNSCLC. Eight in 10 aNSCLC patients did not qualify for drug trials and almost 90 percent of clinical trials did not complete target enrollment by preset deadlines, according to the study.

Their result, recently published in the journal Nature, is “Trial Pathfinder,” an AI algorithm that helps to design appropriate eligibility rules for clinical trials. Trial Pathfinder combs electronic health records and examines the details that allow some patients to be eligible for a trial and other patients to be excluded. Trial Pathfinder not only doubled the number of potential trial enrollees but it also broadened the pool to include more women, minority, and older patients as well.

Exclusive by Design

A key step in the design of any clinical trial comes in setting the medical criteria for patient eligibility. Simply having aNSCLC is not enough. The overall pool of patients is culled down using data from lab tests — levels of biological indicators such as bilirubin, platelet counts, hemoglobin levels, blood pressure, prior treatments, and many others contained in the medical records.

Should patients fail to meet the thresholds set forth in the study design, they do not qualify for the trial.

“It’s a major bottleneck to treatment. Clinical trials are usually quite restrictive in the criteria. We think overly so in many cases,” says Ruishan Liu, a doctoral candidate and an expert in computational health in Zou’s lab who helped develop the algorithm.

Restrictive criteria are, in theory, believed to protect vulnerable patients from the potential side effects of untested drugs. Interestingly, the researchers found that whether a trial used relatively aggressive or weak restrictions, patient withdrawal rates from trials due to adverse side effects were virtually the same. That is, the restrictions had little effect in preventing adverse events they are designed to avoid.

Further, in examining records from 22 prior oncology drug trials, comprising over 11,500 patients, Trial Pathfinder found that the criteria used to exclude patients were inconsistent from trial to trial — and perhaps even arbitrarily set.

“This was true even for identical cancers and trial stages,” Zou adds. “Sometimes the criteria were just copied and pasted from previous studies.”

“These factors led us to believe the criteria could be relaxed to make more patients eligible for a trial without increasing risk of harmful side effects,” Liu notes.

Trial Pathfinder researchers then proposed more inclusive trial criteria by examining the nationwide electronic health records of 61,094 patients with aNSCLC. They found that several commonly used exclusion criteria had “minimal effect” on patient risk and that loosening restrictions could double the number of patients qualifying for a trial without introducing additional risk to the patients.

A Powerful Tool

With their results now published, Zou, Liu, and their industry collaborators believe that Trial Pathfinder’s scope can be expanded to many other types of cancer, and even other diseases, helping to open the doors for patients into clinical trials. Zou pointed to lupus specifically, but the possibilities are virtually wide open.

“We are super excited about the potential,” Zou says. “There is power in the electronic health records, and the core AI mechanisms of Trial Pathfinder should be applicable across many different diseases to make clinical trials more inclusive.”


Stanford HAI's mission is to advance AI research, education, policy and practice to improve the human condition. Learn more

More News Topics

Related Content