A group of researchers at Stanford University has developed a way to make clinical trials more inclusive, using a program that combines artificial intelligence with real-world data.

The team developed a tool that can enroll more patients in clinical trials — namely women, older people and minority groups, who are often excluded. Their study, published Wednesday in Nature, used the records of 61,094 patients with advanced non-small-cell lung cancer to model more inclusive trials. The researchers created a "Trial Pathfinder" program that can simulate tens or even thousands of clinical trials using electronic health records.

Clinical trials are considered the future of medicine: They determine if new treatments are safe in patients and lower the risk of severe reactions in the public, but are both expensive and inefficient. Improving clinical trials, which are "one of the main bottlenecks of medical and health care systems in the U.S.," can have a really large impact, said James Zou, a corresponding author and assistant professor at Stanford.

The study used data from approximately 280 cancer clinics in the U.S., with an average of 5,167 patients in 10 different trials. The research followed the model of current trials, which have both a control arm and a treated arm for new drugs. The Trial Pathfinder uses electronic health records and the survival rate of patients to hypothetically relax the eligibility criteria in a clinical trial.

According to Zou, "the power of data" and machine learning "could potentially double the size of current clinical trials" and allow far more patients to benefit from the treatments. Zou and his colleagues found the survival rate of patients in all 10 trials to be comparable to — and at times smaller than — the survival rate of patients eligible for the trial, implying that relaxed criteria could help more patients.

The researchers focused on lung cancer because cancer is incredibly "expensive, affects so many people, and is the main focus of many companies" offering drug treatments. Zou told The Academic Times that broadening a few of the commonly used criteria in trials would make cancer drugs more effective on the market. Both doctors and the U.S. government agree.

Certain groups of people are excluded without "solid justification," said the Food and Drug Administration, and the smaller sample size leads to less accurate results. For example, over 80% of patients with the type of cancer studied in this paper did not meet the current eligibility criteria for a clinical trial. 

"There's often a mismatch," Zou told The Academic Times, as, "Patients in the clinical trial may look very different than those on the market." Some current cancer trials are excluding Black people, and vaccine trials consistently underrepresent African Americans and Hispanics, leading to less-effective treatments.

The researchers worked closely with the pharmaceutical company Roche, which is the world's leading provider of cancer treatments. They used data from 11,602 patients in existing oncology trials to test the effects on safety when broadening eligibility, to ensure that broader criteria would still benefit real-world patients. The authors were shocked to see that, "Trials that targeted the same cancer, in the same phase, and that involved similar treatments used a number of different thresholds of laboratory values to exclude patients."

The research team is able to use data-driven algorithms and AI because many health care records are now online. There is a lot of momentum from hospitals and insurance companies to have digital records, which offer "data across diverse patients and longitudinal data over many years," Zou said.

AI is also useful in predicting which patients are likely to enroll in trials. Clinical trials are notorious for having strict requirements, and it can be "quite picky and inefficient to manually go through patients and match them to trials," Zou said. AI is used as a natural language processing tool to read through potential participants' files and match them to trials based on their stage of cancer or type of disease.

In the future, the researchers plan to build interfaces that will make it easier to use the Trial Pathfinder and data from this tool. A few years ago, a study that used AI to improve trials would have been "very difficult because of the lack of standardized data," Zou told The Academic Times, so the potential "to address health care challenges is very exciting."

Zou and his fellow scientists at the Stanford AI Lab will next look into the "priority list of diseases" to see where they can make the earliest impact, such as auto-immune diseases and other types of cancer. They can create new clinical trials with similar questions and methodologies to this study using the Trial Pathfinder.

Zou said he would like to thank Ruishan Liu, a Ph.D. candidate in electrical engineering at Stanford, for doing the "heavy lifting and developing the algorithms" for this study.

The study, Evaluating eligibility criteria of oncology trials using real-world data and AI, published Wednesday, April 7 in Nature, was authored by Ruishan Liu and Ying Lu, Stanford University; Shemra Rizzo, Samuel Whipple, Navdeep Pal, Arturo Lopez Pineda, Michael Lu, Brandon Arnieri, William Capra, and Ryan Copping, Genentech; and James Zou, Stanford University and Chan Zuckerberg Biohub.