Mathematical analysis, computational statistics, and machine learning are increasingly being deployed to understand and predict the dynamics of healthcare associated infections (HAI) and antimicrobial-resistant infections (ARI). However, the utility of these models to guiding clinical and health policy decisions often remains unclear. One challenge is that model calibration quickly becomes obsolete as the epidemiology of HAI and ARI changes. To address this gap, we propose to use mathematical modeling and machine learning approaches to build decision-making technologies that improve the risk assessment, prevention, and control of HAI and ARI. Our proposed technologies account for spatial and temporal dynamics, provide continuous, real-time feedback to clinicians and are robust to changes in risk factors and disease prevalence over time. We anticipate that implementation of these technological improvements will help healthcare institutions to substantially reduce the burden of HAI and ARI. We concentrate our efforts on two of the most important HAI: methicillin-resistant Staphylococcus aureus and Clostridioides difficile infections. To conduct these studies, we assembled a team of mathematical modelers, machine learning specialists, health economists, clinical informaticists, infectious disease physicians, and hospital epidemiologists based in California, New York, and Texas. Clinical, microbiological and environmental data to train our models will come from three academic quaternary medical centers and an expanding network of community hospitals.
The first aim i s to calculate the patient-specific risk of acquiring or transmitting a HAI or ARI. We hypothesize that the risk of acquiring an HAI or ARI is more accurately determined when data on patient movement and pathogen exposure are integrated into predictive models. This type of analysis is also expected to improve the risk assessment of automated systems used to detect HAI and ARI outbreaks.
The second aim i s to prevent invasive methicillin-resistant Staphylococcus aureus (MRSA) infections. One objective is to show that cost-effective reduction of invasive MRSA infections and hospital-based transmission can be achieved via personalized decisions for who should be screened for asymptomatic carriage and decolonized.
Our third aim i s to control the spread of Clostridioides difficile infections (CDI). We hypothesize that by calculating the number of CDI averted and the cost saved, models of disease transmission will demonstrate the benefit pre- emptive adoption of contact precautions for patients who are at high risk of transmitting CDI. We also expect to identify environmental pathways that contribute to the risk of CDI superspreading and would benefit from enhanced surveillance and decontamination. Finally, to better understand the importance of antibiotic stewardship programs, we characterize the specific role a patient's antibiotic, infection, social, exposure and colonization history plays in the personal risk of acquiring an invasive MRSA infection and CDI.
We propose to use mathematical modeling, agent-based simulation and machine learning approaches to build decision-making technologies that improve the risk assessment, prevention and control of healthcare- associated infections and antibiotic-resistant infections. Our proposed technologies will account for spatial and temporal dynamics, provide continuous, real-time feedback to clinicians and are robust to changes in risk factors and disease prevalence over time. We concentrate our efforts on two of the most important healthcare- associated infections: methicillin-resistant Staphylococcus aureus and Clostridioides difficile infections.
