Tools & Strategies News

Machine-Learning Models Can Help Predict COVID-19 Test Results

Florida Atlantic University researchers leveraged ML-based predictive analytics to identify the key symptom features associated with COVID-19 infection.

an illustration of the COVID-19 virus

Source: CDC

By Shania Kennedy

- In a study published this month in Smart Health, Florida Atlantic University (FAU) researchers used machine learning (ML)-based predictive analytics to evaluate the correlation between serology and molecular tests for COVID-19 testing and determine which COVID-19 symptoms play a key role in producing a positive test result.

According to the press release published alongside the study, serology, or blood-based, tests, and molecular tests are the two most commonly used methods for rapid COVID-19 testing. However, both types of tests use different mechanisms to determine whether an infection is present. Results can vary depending on the type of test.

Serology tests detect the presence of antibodies triggered by the SARS-CoV-2 virus, while molecular tests measure the presence of viral SARS-CoV-2 RNA. But immune response and viral load are continuously changing and can vary significantly within an individual. As a result, it is possible to observe positive and negative test results in the same person based on the type of test used and the number of post-symptom onset days, the press release notes.

Until now, there have been no studies on the correlation between serology and molecular tests and which factors are the most useful in distinguishing positive COVID-19 test results from negative ones, the researchers state.

To address this research gap, a team from FAU’s College of Engineering and Computer Science trained five ML classification algorithms to predict COVID-19 test results based on easy-to-obtain symptom and demographic features, such as the number of days post-symptom onset, fever, temperature, age, and gender.

Data were collected from 2,467 individuals, each tested using one or multiple types of COVID-19 tests. Based on their test results, donors were labeled as positive or negative for COVID-19. Researchers created symptom feature “bins” to represent each donor.

“Because COVID-19 produces a wide range of symptoms and the data collection process is essentially error prone, we grouped similar symptoms into bins,” explained Xingquan “Hill” Zhu, PhD, senior author of the study and a professor in FAU’s Department of Electrical Engineering and Computer Science, in the press release. “Without a standardization of symptom reporting, the symptom feature space greatly increases. To combat this, we utilized this binning approach, which was able to decrease symptom feature space while keeping sample feature information.”

Using these feature bins, the research team was able to cross-check participants' test types, results, and features to explore the correlation between test types.

“One unique feature of our testbed is that some donors may have multiple test results, which allowed us to analyze the relationship between serology tests versus molecular tests, and also understand consistency within each type of test,” said Zhu.

Model performance was evaluated using classification accuracy, F-1 score, and area under the receiver operating characteristic curve. Overall, the models achieved more than 81 percent area under the receiver operating characteristic curve scores and more than 76 percent classification accuracy. The findings also indicated that the number of days post-symptom onset is crucial for a positive COVID-19 test.

The researchers concluded that their approach might provide a method for rapid COVID-19 screening and cost-effective infection detection.

“Predictive modeling is complicated by many puzzling questions unanswered by research. The testbed created by our researchers is indeed novel and clearly shows correlation between different types of COVID-19 tests,” said Stella Batalama, PhD, dean of FAU’s College of Engineering and Computer Science, in the press release. “Our researchers have designed a new way to narrow down noisy symptom features for clinical interpretation and predictive modeling. Such AI based predictive modeling approaches are becoming increasingly powerful to combat infectious diseases and many other aspects of health issues.”