Clinical trials have a dirty little secret. For all the careful work that goes into randomizing and blinding participants just so, the criteria that determine who can enter a trial can be unexpectedly arbitrary. Patients can be nixed because of age, lab values, medication history, and a laundry list of other factors that may not always be necessary.
“It was certainly surprising to us that these clinical trial criteria designs are fairly ad hoc and quite anecdotal,” said James Zou, who leads Stanford’s Laboratory for Machine Learning, Genomics, and Health. The unwitting result can be that women, older patients, and people of color are excluded from studies at higher rates. That’s why Zou, in collaboration with Genentech and colleagues at Stanford, started looking at how to design smarter eligibility criteria that can boost enrollment without compromising on safety.
In a paper published Wednesday in Nature, they describe an artificial intelligence tool to run simulated clinical trials for cancer drugs applying different eligibility criteria. Using data from real-life patients, they found that in most cases, they could loosen the criteria for trial entry — making it possible to include more and more diverse participants — without any impact on safety.