Predicting Patient Response In COVID-19 Drug Trials

By Deborah Borfitz

August 16, 2021 | Based on cellular-level insights from previous COVID-19 clinical trials, mathematical modelers at the University of Waterloo (Ontario) have simulated how the body deals with the virus as a starting point for predicting how patients will respond to new experimental vaccines and treatments—including those targeting variants of concern. While this is a generic model meant to reflect an “average human being,” it can be customized to more precisely forecast drug effectiveness in individuals according to variables such as their gender, age, and comorbid conditions, says Anita Layton, professor of applied mathematics.

Layton has previously done simulations for pharmaceutical companies to predict drug action in other disease areas, including diabetes and hypertension, she says. But use of in silico modeling to pre-assess the efficacy of potential COVID-19 therapies breaks new ground.

“When we started the project, not a lot was known about our reaction to different treatments,” says Layton. Many of the model parameters are therefore derived from what is known about human immune response to influenza A viral infection, which has been studied for decades.

As recently detailed in Viruses (DOI: 10.3390/v13061141), the new mathematical model depicts control of a SARS-CoV-2 infection by innate and adaptive components of the human immune system. Among the factors considered are the role of different groups of immune cells, interaction between the virus and relevant systems in the body, and the specific physiology being targeted by a treatment. Viral load was one of the key predictive variables, she says.

For the study, the model was applied to three potential COVID-19 therapies—remdesivir (previously shown to inhibit the transcription of SARS-CoV-2), a “hypothetical” therapy that inhibits the virus’ entry into host cells, and convalescent plasma transfusion therapy. Simulation results indicate efficacy is strongly associated with intervention within a day or two after the onset of symptoms.

Simulated results are remarkably consistent with live trial data on COVID infections and treatments. In both instances, for example, remdesivir was shown to be biologically effective but clinically questionable unless administered shortly after viral infection.

Model Refinement

Sophisticated mathematical modeling has already been used to understand the global spread of COVID-19 and measures that might help slow the process, Layton says. In fact, thanks to a rapid response grant from the Canadian Institute of Health Research, she is now calculating the expected spread of concerning COVID variants across the country based on geographic region, demographic factors (e.g., household size and income level), and travel habits.

Computer models can be built to look at immune response to disease but requires a level of computational power that was unavailable even 10 years ago, says Layton. Disease modeling today frequently factors in the host immune response and might also couple genomic or chemical data with artificial intelligence to aid in drug discovery.

What’s different about the application of the approach to in silico testing of potential COVID-19 therapies and vaccines is primarily the pressure to accomplish the work quickly, she continues. With the help of co-author and applied mathematics Ph.D. student Mehrshad Sadria, the study was completed rather quickly.

Modifying the model for real-world drug simulation—for example, to establish exclusion criteria for a COVID-19 clinical trial—would take a bit longer. Different sub-models could be created to signify the immune system of people with diabetes and other comorbidities, over the age of 70, male or female, or of a certain ethnic group, Layton says.

As more COVID-19 data emerges, Layton herself will be refining the model to improve its overall accuracy and make it less generic, she adds. It’s not a trivial undertaking, given limited understanding of the individual-level differences between immune system pathways.

The model could be used to predict viral shedding, which has been associated with viral load, to characterize the contagious period and predict disease spread at a population level, says Layton. If it is made more patient-specific, it could also serve as a diagnostic tool to estimate response to a drug or predict how sick people are going to be.

“A lot of the diagnostic tools that are being used are not even sex-specific,” she notes, mirroring the gender-agnostic way drugs are developed despite differences in pharmacological response in women and men.

Mind To Machine

Drugs have various potential ways of stopping the SARS-CoV-2 virus, including preventing it from entering host cells and making it more recognizable to the body’s built-in defense system, says Layton. “To hit back at a virus, your body has to have a strong enough immune response but not too strong that it also causes all kinds of inflammation and kills you instead. The drug you take has to walk a rather fine line.”

The human immune system is quite complicated and, in many ways poorly understood, she continues. In another recently published study, she has modeled the unexpected impacts of disrupted circadian rhythm (e.g., jetlag and working the night shift), time of day, and gender on immune system response to infection.

Deciding what to look at is the tricky part, says Layton, whose philosophy is to start small and with a hypothesis that “works beautifully in my mind” to test its translation into if-then machine logic.

A population model recently developed by Layton offers predictions on the variable impact of the delta variant based in part on vaccine distribution in a region. The situation doesn’t look good for Ontario, and is even worse for other places, like the U.K., where the AstraZeneca vaccine (60% protection against the variant after two doses) rather than the Pfizer vaccine (88% with full vaccination) was the shot going into most arms, she says.

Load more comments