For more than two years, health-related news has been dominated by stories about COVID-19 vaccinations, deaths, and hospitalization rates. Amidst this focus on the pandemic, however, it can be easy to forget that flu season is also a normal part of our existence.
Both COVID-19 and influenza are respiratory viruses, which means they are passed on in very similar ways. Both viruses are mainly spread by inhaling droplets that are produced by an infected person when they sneeze, cough, talk, or even sing. Unfortunately, this means that their initial symptoms (including fever, coughing, congestion, fatigue, and muscle aches) can be quite similar.
To help distinguish between the two illnesses, a team of researchers from York University turned to machine learning techniques. The research was led by Suzan Farhang-Sardroodi, a postdoctoral fellow at York, and published Mathematical Biosciences and Engineering.
Why is it difficult to distinguish between the two?
COVID-19 will likely become a regular part of our existence, similar to the flu. Both viruses circulate year-round; however, the conditions created by the winter can exacerbate infections, leading to what we know of as flu season.
For example, people spend more time indoors during the winter, making it easier for the virus to be transmitted. The cold, dry air of winter also leads to a decrease in nasal mucus, which can make it more difficult to trap and clear viruses.
Since influenza (the virus that causes the flu) and SARS-CoV-2 (the virus that causes COVID-19) are spread in the same way, we will likely see peaks of flu season coinciding with peaks of COVID-19 infections. This has the potential to pose great difficulties to the healthcare world, especially in making initial clinical diagnoses.
In the context of emergency rooms, for example, doctors and nurses are often overwhelmed and have to make quick diagnoses. Wait times in emergency rooms can be long, and doctors may not have time to address every symptom. They also may not have the luxury of using multiple test types and medical readouts in order to make a diagnosis.
When we consider how diverse the symptoms of COVID-19 can be, and the fact that rapid antigen tests are not always accurate, a method to quickly differentiate between COVID-19 and influenza is of great interest.
Can a computer differentiate between COVID-19 and the flu?
To solve this problem, researchers at York University decided to turn to machine learning and artificial intelligence.
Machine learning is a method of teaching computers to use information provided to them in order to make intelligent decisions. In other words, it’s essentially a way of teaching computers to imitate human thinking and decision making.
The information provided to the decision-making software is the same information that a medical professional would collect. For example, demographics, body mass index, and vital signs are all information that can be collected in an emergency room context and help support an infection diagnosis.
Previous studies have shown that these qualities can be used by machine learning models to quickly distinguish between influenza and COVID-19. Taking guidance from this research, Farhang-Sardroodi and collaborators used a mathematical model to classify flu and COVID-19 patients based on information about how the virus behaved within the patient, as well as information on the patient’s immune response.
To do this, the researchers used information about how COVID-19 impacts our immune systems to create their model for telling the two illnesses apart. They used information such as which types of cells the virus targets, how many cells are infected, and how long it takes for an exposure to become an infection.
They also looked at the patients’ levels of Type 1 Interferon, which is a protein produced by infected cells to warn the body about an infection and help limit the spread of a virus.
Using their model, the team attempted to classify patients as influenza- or SARS-CoV-2-infected. By inputting these characteristics into their mathematical model, the researchers were able to predict a patient’s diagnosis.
Machine learning models can help
The team found that viral load (meaning the amount of virus detected in a person) and productively infected cells (meaning the number of cells that became infected) were the most important characteristics in telling influenza patients apart from COVID-19 patients.
Their research also showed that early on in the infection, influenza patients were more likely to be misdiagnosed as COVID-19 patients. This was because Type 1 Interferon levels and viral load measurements were initially similar across the two patient groups.
Despite this overlap, however, their model was still 91% accurate in distinguishing between patient types.
This research has applications in the context of clinical trials, where trial outcomes may be complicated by participants becoming infected with a virus that has similar characteristics to the infection of interest. Using machine learning models like this one could help researchers tell different viruses apart.
This research also has applications in emergency rooms, where diagnoses can occur quickly and with limited data. Furthermore, machine learning methods are much less susceptible to the biases, inconsistencies, and assumptions that can cloud human judgment.
Going forward, this work will help medical professionals distinguish between these two illnesses. With Canada entering a summer wave of COVID-19 and flu season on its way in the fall, quick and accurate diagnoses will be crucial in keeping everyone healthy.