Researchers from Georgia Tech, University of Virginia, and the University of Iowa have teamed up to prevent, control, and intervene against hospital acquired infection (HAI) outbreaks.
Detection and control of HAIs, such as Clostridioides difficile infection (CDI), is a fundamental public health problem and a resource intensive challenge for hospitals. And with the spread of Covid-19 on the rise, the need to combat HAI outbreaks is more critical than ever.
Despite the huge importance for hospitals, and the interest from both clinical and epidemiological researchers, these problems remain poorly understood and all too common. According to the Centers for Disease Control and Prevention (CDC), one in thirty-one hospital patients in the United States are infected with at least one HAI on any given day. HAIs are particularly challenging because of the high cost of patient treatment and disinfection of hospital facilities, as well as penalties against hospitals if HAIs occur.
In an effort to combat the rate of HAIs, the cross-institute group of researchers, led by Georgia Tech School of Computational Science and Engineering (CSE) Associate Professor B. Aditya Prakash, are creating a holistic approach to better understanding, preventing, and treating HAI outbreaks by developing a network-based framework to improve hospital infection control.
Equipped with a $1.2 million grant from the National Science Foundation (NSF), this three-year project aims to create a countermeasure toolkit to aid infectious disease experts.
“This will simultaneously improve care for current patients, make work safer for healthcare workers, and help prevent the incursion of Covid-19 into hospitals,” said Prakash.
“Our research brings together researchers from different backgrounds - data scientists, epidemiologists, hospital infection control experts and clinicians - in order to adopt a very interdisciplinary methodology. Through this, we aim to develop a new network-based approach that improves hospital infection control using data driven models and data science algorithms.”
A Holistic Approach
Given that each hospital is unique, the focus of the project is to design fundamental strategies and provide guidance for hospital infection control decision makers to determine what exact policies are best for each individual location.
Currently, there is limited data on HAI outbreaks, and the dynamics of HAI spread are more complex than other diseases due to several compounding factors. In response to these issues, the team says it is seeking a paradigm shift and will pursue a holistic view for this problem. The team will do so by modeling the disease spread using high dimensional clinical and mobility network data of healthcare workers and patients.
This type of data includes onsite surveys, RFID-type sensors, manual check-ins, anonymized electronic medical records, and more, to determine which areas are at higher risk for contagion exposure, who is most likely to come into contact with that area and with each other, and who may get eventually infected.
This research will use a unique fine-grained, large-scale dataset of operations from the University of Iowa Hospitals and Clinics collected over 10 years, supplemented with data collected from other hospitals. Results will be validated with the help of domain experts including epidemiologists and clinicians involved in hospital infection control.
These varying datasets will be combined to build a complete picture of disease transmission pathways to help hospitals quickly detect, understand, and control future HAI spread. To accomplish this, researchers will focus on multiple aspects of the infection control cycle: developing better surveillance techniques, more informed and carefully designed interventions, and more accurate exposure risk assessment tools.
“Building a comprehensive framework is very challenging because it is dependent on the layout of the hospital, the personnel, and how they all interact, in addition to the transmission characteristics of the disease itself. So, there are many combinations, which is why we need nimble models that can ingest heterogeneous dynamic data, leverage global information, and yet be useful at an individual level,” said Prakash.
To overcome these issues, researchers created a new class of two-mode cascade models to use throughout this project. This particular class has very different dynamics than current standard models used in heterogenous data analysis and has not been previously studied in data mining.
“These are difficult problems on networks, and we will invent rigorous and scalable methods using tools from data science, machine learning, and combinatorial optimization,” said Prakash.
The team expects their research will lead to novel computational methods `and algorithms, in addition to guiding the next stage of advances in infectious disease practice.