Treating patients with the never-before-seen Covid-19 is forcing doctors to choose between two equally unpalatable options: try an unproven therapy and hope it works, or treat patients with standard supportive care for severe respiratory disease until a randomized controlled trial establishes the best treatment.
But maybe the choice doesn’t have to be so stark.
A randomized controlled trial ramping up in dozens of hospitals around the world proposes a third way — fusing those two approaches together by using artificial intelligence to home in on the most effective treatments on the fly.
The REMAP-CAP study seeks to turn frenetic attempts to save lives on the front lines into a running international experiment, with the goal of quickly identifying optimal treatments for desperately ill patients. By analyzing data on outcomes from more than 50 hospitals, organizers hope to supply fast answers to pressing questions, such as whether the antimalarial drug hydroxychloroquine is an effective therapy and, if so, for which types of patients. The trial will also allow the researchers to test multiple therapies at once.
“This is a way to optimize how to learn while doing something for patients at the same time,” said Derek Angus, an intensive care physician at the University of Pittsburgh Medical Center who is the senior investigator of the trial. “One of the ways to get through this is to thread that needle.”
While the approach seems ideally suited to providing answers during the pandemic, it faces a number of challenges, including the need to rapidly compile and analyze data from dozens of hospitals with different record-keeping systems on three continents, and then update protocols in unison during a crisis that is straining clinical resources.
It is, to say the least, a high-wire act.
“It puts a high burden on the operational realities of getting data in real time from all of these sites so that you can then learn from it,” said Ed Seguine, chief executive of Clinical Ink, a developer of software to collect clinical trial data.
REMAP-CAP is what’s known as an adaptive trial, in which researchers modify the treatment protocols or statistical procedures based on the outcomes of participants, like a chef tweaking a recipe during dinner service as feedback comes in from diners. It is seen as a way to more quickly identify promising treatments and make trials more flexible than traditional randomized trials that bucket participants into control and intervention groups and force patients, and trial sponsors, to wait for an outcome that often turns out to be disappointing.
The shortcomings of that latter approach have been brought into sharp relief during the pandemic, as thousands of patients cannot wait for gold-standard science to play out as they lay dying in intensive care of units. The World Health Organization and the U.S. Food and Drug Administration, along with groups like the Gates Foundation, have offered increasing support for adaptive trial designs in recent years, particularly as a way to evaluate therapies during epidemics.
But that doesn’t mean this particular effort is going to produce results in time to save the first wave of extremely ill patients.
The study is designed to randomize patients with severe pneumonia caused by the coronavirus to receive different treatments within four categories: antibiotics, antiviral therapy for influenza, steroids, and a class of antibiotic called macrolides that are often used to treat patients with skin and respiratory infections. The trial will also seek to evaluate different strategies for delivering oxygen and mechanical ventilation; the primary outcome being measured is 90-day mortality.
Once a promising regimen is identified within a category of treatment, more patients will be assigned to receive it during each successive round of therapy. So far, about 130 ICU patients with Covid-19 have been enrolled, in addition to hundreds of other hospitalized patients.
“It takes advantage of the fact that we’re simultaneously randomizing across several [treatments] all in one go,” said Angus. “That allows us to explore interactions between the interventions and means a far smaller proportion of the patients when you first start the trial are getting nothing other than usual care.”
Because of the number of different treatments being tested, carrying out these trials is particularly complicated. But advances in computing resources needed to share data and analyze it swiftly using artificial intelligence have begun to make these designs more practical.
The analysis of the data is powered by a common type of artificial intelligence known as machine learning, in which the software learns from the data, and pinpoints the most successful treatments, without being explicitly programmed to do so. It is the same technique Amazon uses to identify which advertisements to target to customers based on their shopping histories. The goal in the REMAP-CAP trial, once all the trial sites are up and running, is to analyze results and adjust treatments on a weekly basis.
The hospitals in the trial will share data through a common online portal, but the ability to coordinate care between sites, and verify adherence to the trial protocols, is limited by the fact that participating hospitals use different types of patient record-keeping systems. The U.S. is farther along in adopting digital record keeping than other countries, but is still limited because health systems have adopted different systems that cannot easily communicate with one another.
Angus said efforts are underway to make it easier for U.S. hospitals that use different systems to join the trial. The hope is to quickly expand the reach of the trial to more than 200 hospitals, which would dramatically increase the amount of data that could be collected and analyzed.
“We have multiple sites in the U.S. that are interested in joining,” Angus said. “If they want to embed this inside their EHRs, we will be sharing with them our tools, tips and triumphs, but they will have to do the hard wiring themselves.”