How a statistical model created with artificial intelligence would predict COVID resistance

At this point, there are already more than 621 million the cases of COVID-19 worldwide, with more than 6.5 million deaths. Despite the high rate of transmission of COVID-19 in shared homes, some exposed people seem “immune” to contracting the virus. Even, it is still not known, in depth, whether the emergence of resistance to the coronavirus differs between people, according to health characteristics.

Now researchers from Johns Hopkins Medicine and Johns Hopkins University have created and preliminarily tested what they believe is one of the first models to predict who is most likely to be wearing to COVID-19, despite exposure to the virus. The scientists’ findings were published in the journal PLOS ONE.

“If we can identify which people can naturally avoid SARS-CoV-2 infection, we can learn, in addition to behavioral and social factors, what genetic and environmental differences influence their defense against the virus. This idea could lead to new preventative measures and more targeted treatments,” said study lead author Karen Yang, a graduate student in biomedical engineering at the Johns Hopkins Translational Informatics Research and Innovation Laboratory.

The research team set out to determine if a statistical model machine learning could use the health features stored in health records electronic data, providing patient data such as comorbidities and prescription medications, as a means of identifying subjects with a natural ability to avoid SARS-CoV-2 infection. “These people could be studied to better understand the factors that enable their resistance,” Yang said.

A machine learning model is a computer system that uses algorithms mathematical methods to find statistical patterns and, once identified, apply them in the future. This allows the technology to constrain human thinking and reasoning and, like the brain, to learn over time.

“Using a machine learning system to recognize complex patterns in large numbers of people with COVID-19 allowed another team of Johns Hopkins Medicine researchers in 2021 to predict the course of the case of a patient and to determine the probability to get serious. Given its success, our team wondered if the same approach could also be applied to predict who might be exposed to it. SARS-CoV-2 inside and not get infected,” said study co-lead author Stuart Ray, vice president of medicine for analytics and data integrity, and Johns professor of medicine. Hopkins University School of Medicine.

To demonstrate the model’s ability to predict COVID-19 resistance, the researchers first acquired data from a clinical registry called Johns Hopkins COVID-19 Precision Medicine Analytics Platform Registry (JH-CROWN). The registry contains information about patients cared for by the Johns Hopkins Health System who are suspected or confirmed to have SARS-CoV-2 infection.

The scientists only included people who received a COVID-19 test between June 10, 2020 and December 15, 2020 and reported “potential exposure to the virus” as the reason for the test. The end date chosen was the one that coincided with the start of large-scale vaccination against COVID-19 in the United States. This inflection has allowed specialists to avoid in their discoveries the effects of vaccines that prevent infection instead of natural resistance.

The bone 8536 attendees of the study who reported exposure as a reason for getting tested for COVID were divided into two groups: those who did not share a residence (called a “household” in this study) with COVID-19 patients or their residence had 10 or more patients; and those who shared a residence with 10 or fewer people, at least one of whom was a COVID-19 patient.

The first group, with 8,476 participants, was named the “Training and Testing Set”, while the second, called the “Household Index (HHI) Set”, had 60 members and was used as a test. According to the researchers, determining the number of 10 or fewer people in the household excluded people living in higher density multi-unit housing areas where exposure to a particular SARS-CoV-2 positive person would be less intense.

To identify patterns y agrupar a los modo participants that destaquen aquellos naturally resistentes al SARS-CoV-2, ambos conjuntos de estudio se analizando el algoritmo de agrupamiento basado en models de selección de patrones de maximum frequency y confianza total (MASPC, por sus siglas in English).

MASPC is specifically designed for analysis of electronic health record data that combines patient demographic information (age, sex, and race), International Statistical Classification of Diseases and Related Health Problems (ICD), and diagnostic codes relevant medical. , and the number of comorbidities (other diseases) present. All are tools commonly used in medical centers in the United States and capable of sharing information openly.

“We hypothesized that MASPC would allow us to group patients with similar patterns in their data to define them as resistant and non-resistant to SARS-CoV-2, and in the hope that the algorithm would learn with each analysis how to improve the accuracy and reliability of future missions. . This initial study using data from JH-CROWN was conducted to test this hypothesis, a proof-of-concept trial of our statistical model to show that resistance to COVID-19 could be predictable based on the clinical and demographic profile of a patient. Ray explained.

“In the training and test set, we identified 56 ICD code patterns divided into two groups: whether or not associated with resistance. Statistical analyzes of how these patterns differentiated between resistance and non-resistance yielded five that did the best job for our small, localized study population in defining who was most likely exposed to SARS-CoV-2.” , Yang said.

“Looking for these models, people most likely to have been exposed to SARS-CoV-2 indoors, and then statistically analyzing the results, our model’s best performance was 0.6. Since a score of 0.5 only shows a random association between prediction and reality, and 1 is a 100% association, it shows that the model is promising because tool identify people resistant to COVID-19 that can be investigated further,” Ray said.

Along with Yang and Ray, study team members from Johns Hopkins Medicine and Johns Hopkins University are graduate and undergraduate students Yijia Chen, Jacob Desman, Kevin Gorman, Chloe Paris, Ilia Rattsev, Tony Wei and Rebecca Yoo; and faculty co-lead authors Joseph Greenstein and Casey Overby Taylor.

Continue reading