A new study, published in the journal JAMA Network Open, looked at the risk factors for severe cases of COVID-19 analyzing the progression of the disease over time. Everything was achieved thanks to the support of machine learning models to predict which hospitalized patients would develop a severe course based on information gathered during the first day of hospitalization.
Covid-19, a complex study with machine learning to better understand the disease
To better explain what the study is about, it is good to start from the beginning. A centralized archive of medical records related to Covid-19 was created last year and is now starting to show first results. We are talking about the largest disease tracking set ever created so far, managed by a team of researchers and experts who wanted to see clearly using the latest technologies.
The centralized database is called National Covid Cohort Collaborative Data Enclave, or more simply N3C, and used data from 34 medical centers including information from over 1 million adults: of these, 174,568 tested positive for the virus and 1,133,848 tested negative. The archive contains information for the entire calendar year 2020.
Now, the analysis focuses on how the treatment to fight Covid-19 infection has changed over the twelve months of last year, as doctors tried new treatments and gained more experience as a result. During 2020, treatments have changed, survival rates have responded to increasingly effective treatments, and the various complications experienced in patients have become easier to combat over time.
The research team then built machine learning models using this and other information that could predict which patients are at risk of becoming seriously ill once infected. These machine learning models can eventually be used as a basis for making decisions about additional tests.
A resource like N3C isn’t perfect, but it eludes them typical limitations of studies that occur in cases like this, often packaged locally and therefore unable to have an overall picture of the situation. The archive used by the researchers now includes Ben’s information 73 health facilities, for over 2 million Covid-19 patients. From this information, more than 200 research projects were born.
As mentioned, it is not a perfect system, but having such a vast and varied set of data available is absolutely important to design new tools with which to eradicate the disease. Last fall, University of Rochester health economist Elaine Hell admitted to the microphones of The Verge that this system allows you to conduct studies that are impossible to tackle with your own resources alone institution.