A new analytical tool can improve a hospital's ability to limit the spread of antibiotic-resistant infections over traditional methods like contact tracing, according to a new study led by researchers at Columbia University Mailman School of Public Health and published in the peer-reviewed journal Nature Communications . The method infers the presence of asymptomatic carriers of drug-resistant pathogens in the hospital setting, which are otherwise invisible.
Antimicrobial resistance (AMR) is an urgent threat to human health. In 2019, 5 million deaths were associated with an AMR infection globally.
The inference framework developed by Columbia Mailman School researchers is the first to combine several data sources—patient mobility data, clinical culture tests, electronic health records, and whole-genome sequence data—to predict the spread of an AMR infection in the hospital setting. In the study, the researchers used five years of real-world data from a New York City hospital. They focused on carbapenem-resistant Klebsiella pneumoniae (CRKP), an AMR bacterium with a high mortality rate. The framework draws on the four data sources to model the spread of CRKP infections, from individual to individual over time.
Levels of CRKP colonization in healthcare facilities vary by location but can reach up to 22 percent of patients. However, hospitals do not routinely screen for CRKP, and surveillance relies on testing patients who are either symptomatic or suspected of coming into contact with symptomatic patients, overlooking asymptomatic colonizers.
"Many antimicrobial-resistant organisms colonize people without causing disease for long periods of time, during which these agents can spread unnoticed to other patients, healthcare workers, and even the general community," says the study's first author, Sen Pei , PhD, assistant professor of environmental health sciences at Columbia Mailman School. "Our inference framework better accounts for these hidden carriers."
The researchers used the inference framework to estimate CRKP infection probabilities despite limited data on infections. They found that combining the four data sources led to more accurate carrier identification. Furthermore, using data simulations, they found that the framework was more successful at preventing the spread of infections after isolating carriers than traditional approaches based on an individual's time in the hospital, the number of people they came in contact with, and/or whether the people they came in contact with were identified as having infections.
Using the inference model, isolating 1 percent of patients on the first day of each week (10– 13 patients per week) reduces 16 percent of positive cases and 15 percent of colonization; isolating 5 percent of patients on the first day of each week (50– 65 patients per week) reduces 28 percent of positive cases and 23 percent of colonization. For comparison, using contact tracing —a typical approach in clinical settings (i.e., screening close contacts of positive patients)—isolating 1 percent of patients reduces 10 percent of positive cases and 8 percent of colonization; isolating 5 percent of patients reduces 20 percent of positive cases and 16 percent of colonization.
The new study builds on a study in the journal Proceedings of the National Academy of Sciences (PNAS) that introduced a method that more accurately predicts the likelihood that individuals in hospital settings are colonized with methicillin-resistant Staphylococcus aureus (MRSA) than existing approaches. The new study is a significant advance over the previous study because it now includes patient-level electronic health records and whole-genome sequence data, which allows more precise identification of silent spreaders. While the inference model improves on traditional methods, it remains challenging to eliminate AMR pathogens in hospitals due to their widespread community circulation, limited hospital surveillance, and high false-negative rates in clinical culture tests. However, there is room for improvement; a future study aims to look at the spread of AMR using ultra-dense sequencing.
The new study is the result of a collaboration between computational researchers and physician scientists from different hospital systems. Additional authors include Dwayne Seeram and Anne-Catrin Uhlemann of Columbia University Irving Medical Center; Seth Blumberg of the University of California, San Francisco; Bo Shopsin of New York University Grossman School of Medicine; and Jeffrey Shaman of Columbia Mailman School and Columbia Climate School.
This study was supported by funding from the U.S. Centers for Disease Control and Prevention (U01CK000592, U01CK000590), National Institutes of Health (AI180492, GM156799, GM147702, AI140754, AI137336, AI183182, AI183668, DK104309), and funds from the NYU Langone Health Antimicrobial-Resistant Pathogens Program. The project described was supported by cooperative agreements (U01CK000592, U01CK000590) from the CDC. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC.
Shaman and Columbia University disclose partial ownership of SK Analytics. Other authors declare no competing interests.
