Validating agent-based simulation model of hospital-associated Clostridioides difficile infection using primary hospital data

Abstract

As agent-based models (ABMs) are increasingly used for modeling infectious diseases, model validation is becoming more crucial. In this study, we present an alternate approach to validating hospital ABMs that focuses on replicating hospital-specific conditions and proposes a new metric for validating the social-environmental network structure of ABMs. We adapted an established ABM representing Clostridioides difficile infection (CDI) spread in a generic hospital to a 426-bed Midwestern academic hospital. We incorporated hospital-specific layout, agent behaviors, and input parameters estimated from primary hospital data into the model, referred to as H-ABM. We compared the predicted CDI rate against the observed rate from 2013–2018. We used colonization pressure, a measure of nearby infectious agents, to validate the socio-environmental agent networks in the ABM. Finally, we conducted additional experiments to compare the performance of individual infection control interventions in the H-ABM and the generic model. We find that the H-ABM is able to replicate CDI trends during 2013–2018, including a roughly 46% drop during a period of greater infection control investment. High CDI burden in socio-environmental networks was associated with a significantly increased risk of C. difficile colonization or infection (Risk ratio: 1.37; 95% CI: [1.17, 1.59]). Finally, we found that several high-impact infection control interventions have diminished impact in the H-ABM. This study presents an alternate approach to validation of ABMs when large-scale calibration is not appropriate for specific settings and proposes a new metric for validating socio-environmental network structure of ABMs. Our findings also demonstrate the utility of hospital-specific modeling.