Using Machine Learning to Rescue Clinical Trials in Trouble During the COVID-19 Pandemic
Artificial Intelligence helped researchers develop today’s life-saving COVID-19 vaccines in less than a year – a pace faster than was thought to be possible.
And building on that success, an AI-related technology – machine learning, which is used to train machines to learn – is now being used to rescue critical cancer clinical trials that are in danger of shutting down due to the ongoing coronavirus pandemic and low patient enrollment rates.
Cancer patients already faced hurdles enrolling in clinical trials long before COVID-19 battered the globe. The pandemic has highlighted the problem, however, leading stakeholders to question whether they should continue to use inefficient manual clinical trial selection and sign-up processes that have been used for years even though they have never been optimal.
Thanks to innovations in cancer research whose development began long before the pandemic, new trial-management solutions are now available — and scalable.
The challenge is simple. To jumpstart the development of new cancer cures and treatments, we need to find more patients who are eligible for clinical trials. Fortunately, machine learning has evolved to find them quickly, efficiently and without disrupting workflows in hospitals, cancer centers and other settings.
The progress came out of need. As clinical trials were suffering from a drop-off in enrollment due to COVID-19 and stay-at-home protocols around the nation, a community oncology center in the southwestern U.S. enrolled three times as many patients between March and June 2020 than they had in the five months prior. The numbers were small, with only 10 patients enrolling in trials at the center. But the machine learning technology that achieved the increase holds enormous promise to help clinical trials in trouble.
This technology uses Natural Language Processing parsers that sift through electronic medical records to identify patient characteristics that suit particular oncological clinical trials. These parsers discover the histology and behavior of patient cancers, their genetic makeup, their stages, whether or not patients have undergone a procedure and their potential Gleason scores, which are all factors that are taken into consideration for participation in trials.
Those who understand AI and machine learning know that capturing and normalizing data, not computational power, is the challenge when adapting technologies to the physical world. Parsers are useless without data to parse.
That’s where advancements in machine learning today have tackled that obstacle. Machine learning is particularly good at sifting through tens of thousands of permutations in the criteria to determine who fits best with thousands of clinical trials, a task that we already know is too much for doctors and their research assistants to tackle manually as they see scores of sick patients a day.
Traditionally, to join a trial, cancer patients visited ClinicalTrials.gov, answered advertisements or were lucky enough to have a doctor who believed they might qualify. Their chances of joining a trial then depended on a time-consuming, costly manual process – think phone calls and paper questionnaires – that occurs between hospital research directors and physicians on one side and biopharmaceutical trial sponsors and contract research organizations on the other.
The pandemic exposed the inefficiencies of this process. Clinical trials engaged 38 percent fewer patients in the US through June last year, further reducing what was already a small pool of patients eligible for trials. Under normal conditions, an average of only 5 percent of patients have a chance of participating in trials. The numbers have rebounded slightly in recent months. But continued lockdowns, social distancing and overwhelmed hospitals have made it harder than ever for cancer patients to connect with clinical trial sponsors, as the New York Times recently reported.
We can use existing technologies not only to assimilate electronic medical records (EMRs) and other good data, such as from tumor registries, but also to scan and interpret doctors’ notes, yielding a massive amount of quality, on-the-ground, clinically relevant insights into potential trial participants.
Directly accessing EMRs, doctors’ records and other data does not necessarily produce eligible patients instantly. But it can filter out ineligible patients quickly and narrow down searches so that qualified research directors and doctors can save time and focus on delivering care. Harnessing this world of latent data creates a wide new avenue through which machine learning can match patients to trials.
This matching process is exactly what advocates of AI and machine learning have been envisioning for years.
It doesn’t place a burden on busy clinicians. It amasses granular details for trials that need patients with unique conditions at specific times in their treatment. It gives more cancer patients access to new treatments and therefore, more choices and new hope.
Machine learning transforms valuable medical data that has been sitting untapped for years into potential gold that will fuel the creation of blockbuster pharmaceuticals in the future. We needed a public health crisis to see its value. Now that we know, it’s time to step into the future.
About the Author
Kirk F. Junker is the vice president of engineering at Inteliquet, which uses data to help match patients to clinical trials. Junker is responsible for software development at the company.