Nosocomial infection modeling and simulation using fine-grained healthcare data
Simulation has long been used in healthcare settings to study a range of problems, such as determining ideal staffing levels, allocating patient beds, and assisting with medical decision making. Some of this work naturally focuses on the spread of infection within hospitals, where the importance of...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | English |
| Published: |
University of Iowa
2011
|
| Subjects: | |
| Online Access: | https://ir.uiowa.edu/etd/4856 https://ir.uiowa.edu/cgi/viewcontent.cgi?article=4897&context=etd |
| Summary: | Simulation has long been used in healthcare settings to study a range of problems, such as determining ideal staffing levels, allocating patient beds, and assisting with medical decision making. Some of this work naturally focuses on the spread of infection within hospitals, where the importance of hospitals as loci and amplifiers of infection was demonstrated during the 2002-2003 SARS outbreak. Increasingly, fine-grained healthcare data is being collected (e.g., patient care data stored in electronic medical record systems, and healthcare worker data from sources including nurse locator badges), presenting an opportunity to develop models that can drive more realistic simulations. We seek to build a realistic hospital simulator that can be used to answer a wide variety of questions about infection prevention, the allocation and placement of expensive resources, and issues surrounding patient care.
Our simulation framework requires three primary inputs: architectural, healthcare worker, and patient data. We used data from the University of Iowa Hospitals and Clinics to build our virtual hospital. We manually constructed a weighted, directed, 19,000 node graph-theoretic representation of the facility based on printed architectural drawings. Using timestamped location information from electronic medical record system logins and algorithms inspired by prior work on location-aware search, each healthcare worker is modeled by one or more “centers” of activity. Centers are determined using a maximum likelihood approach to fit a location and appropriate decay parameters that best describe the observed data. Finally, we developed compartmental patient models of varying granularity, with each compartment representing some subset of patient care areas within the hospital. Transition probabilities and patient length of stay were fit using three years of patient data.
In designing our simulator, we were able to minimize assumptions about how healthcare workers and patients move, avoiding the “random mixing” assumption common to many infectious disease simulators. We translated techniques from location-aware search into the hospital environment, developed data structures for use in efficiently processing millions of location data points in tens of thousands of rooms for thousands of healthcare workers, improved the performance of the algorithm for identifying optimal single-center healthcare worker models, and introduced heuristics for training multi-center models. We validated our models by comparing the properties of simulated data to known quantities, and testing against expert expectations. To the best of our knowledge, this is the first agent-level hospital-wide simulator based on fine-grained location and interaction data for healthcare workers and patients. |
|---|