In studies that applied artificial intelligence and modern diagnostics to 30-year-old blood samples pulled from freezers, scientists showed how tests tracking levels of a virus in the serum of recovering bone marrow transplant patients reliably charted the eventual course of their disease.
This research also has implications for urgent efforts to test quickly — using similar methods — the efficacy of new drugs against COVID-19, where speedy assessments of potential treatments are desperately needed.
The studies at Seattle’s Fred Hutchinson Cancer Research Center were focused on cytomegalovirus, a common infection that can become lethal to patients as they try to rebound from transplants.
While antiviral drugs can prevent or dampen some CMV infections, the virus remains one of the most dangerous complications for cancer patients receiving bone marrow or stem cell transplantation.
Results of a pair of related studies using the frozen blood samples were published online in the Journal of Clinical Investigation. The research team was led by Dr. Michael Boeckh, head of Infectious Disease Sciences at Fred Hutch, his colleague, Dr. Joshua Schiffer and Dr. Elizabeth Duke, a physician-researcher who also practices at clinical-care partner Seattle Cancer Care Alliance.
Their paper describes how measures of viral load — the amount of virus in a milliliter of blood drawn in 1989 and 1990 — could have served as effective “surrogate markers” of drug effectiveness.
“One of the cool things about this paper is that we were using the most modern techniques on the oldest data available,” Duke said.
Three decades ago when those samples were drawn, the 72 transplant patients were participating in a clinical trial to test the effectiveness of an antiviral drug, ganciclovir, in preventing or treating CMV infection.
At that time, viral load testing was just beginning to be applied as a way to gauge the health and infectiousness of people infected with HIV, the virus that causes AIDS. It simply was not available to researchers in the cytomegalovirus studies, so drug effectiveness was measured solely by clinical outcomes.
Those outcomes comparing death rates, development of tissue-invasive CMV disease and other complications were tallied over months and years. Modern tests that measure viral load — actually, a handful of tests that comprise a measure called viral load kinetics — can produce answers within days if needed.
These cutting-edge viral load tests were paired with artificial intelligence — also known as machine-learning — and applied to the archived samples retrieved from Fred Hutch freezers. Machine learning is a use of computer programs that can tease out, from huge sets of data, patterns that would otherwise be difficult to detect.
'You put that data in, and it spits out a percentage likelihood that patient will get CMV disease.' — Dr. Elizabeth Duke
The results of the new surrogate marker analyses more or less mirrored the clinical findings made in the early 1990s. Applied to patients today, they can give physicians, in almost real time, important clues how to treat each patient most effectively.
“With machine learning, we are able to predict the outcomes of individuals better,” Duke said.
Using this system, doctors load the data from each patient into a machine-learning algorithm. It might require a person’s viral load, age, and risk factors such transplant complications like graft-vs-host disease, where donor stem cells start attacking the patient’s healthy tissues. The computer then predicts outcomes.
“You put that data in, and it spits out a percentage likelihood that patient will get CMV disease,” Duke said. “That helps the clinician decide, for instance, whom we should test more frequently, who should get prophylactic drugs.”
The Hutch studies produced encouraging results on the use of viral load kinetics to evaluate the effects of ganciclovir in treating patients who have developed CMV disease and for giving it to others deemed at risk, to protect them from developing it.
In 2017, the early findings of the viral load study on ganciclovir were used in a clinical trial that helped win approval of letermovir, a less toxic antiviral than ganciclovir, for prevention of CMV in immune-compromised patients.
As a result, researchers now have a proven method for use in future CMV drug trials, and a model that might be applied to other antiviral drugs against different diseases.
“It means we can conduct smaller and more efficient trials to measure new, promising CMV therapies,” said Fred Hutch virologist Schiffer, who is a senior author of the paper. “Doing so is a major priority to allow for more rapid assessment of promising treatments.”
Schiffer noted that viral load has yet to be established as a surrogate marker for COVID-19 transmission or risk. However, he said that the machine learning approach is “a fantastic way to extract maximal information from complex datasets,” and he hopes eventually to use the technique in coronavirus clinical trials.
At Fred Hutch’s newly established COVID-19 Clinical Trials Research Center, Duke and Boeckh will oversee participation in a phase 2 clinical trial of an oral antiviral drug that is designed to inhibit SARS-CoV-2 — the virus responsible for the disease. Originally developed at Emory University as a potential flu treatment, the experimental drug known as molnupiravir is now licensed to Ridgeback Biotherapeutics and pharmaceutical giant Merck. It targets a different mechanism to block viral replication than that exploited by remdesivir, the only approved COVID-19 treatment.
Although there are no plans at the moment to use surrogate markers to gain approval for molnupiravir, the Hutch scientists are hopeful that the approach will eventually be adopted in the testing of antivirals aimed at the coronavirus.
“I know how to do a surrogate marker analysis,” Duke said. “If you can do a trial based on viral load rather than clinical outcomes, you can do your trials more quickly and with fewer participants.”
Sabin Russell is a staff writer at Fred Hutchinson Cancer Research Center. For two decades he covered medical science, global health and health care economics for the San Francisco Chronicle, and wrote extensively about infectious diseases, including HIV/AIDS. He was a Knight Science Journalism Fellow at MIT, and a freelance writer for the New York Times and Health Affairs. Reach him at email@example.com.
Are you interested in reprinting or republishing this story? Be our guest! We want to help connect people with the information they need. We just ask that you link back to the original article, preserve the author’s byline and refrain from making edits that alter the original context. Questions? Email us at firstname.lastname@example.org