Abstract
Asymptomatic carriers of antimicrobial-resistant organisms (AMROs) can unwittingly transmit these pathogens in hospitals, contributing to the burden of healthcare-associated infections (HAIs). Surveillance in hospitals can involve different types of observations; however, a framework to coherently synthesize these datasets to identify AMRO carriers is lacking. Here, we develop a new inference framework combining a data-driven mechanistic transmission model and multimodal observations from clinical cultures, electronic health records, patient mobility, and genomic data. Using extensive simulated outbreaks, we validate the inference framework for AMROs with various levels of community importation and hospital transmission and evaluate the utility of different combinations of data sources. Inference results show that using multimodal observations consistently improves the accuracy in identifying AMRO carriers. We apply the inference framework to carbapenem-resistant Klebsiella pneumoniae (CRKP) at an urban quaternary care hospital in New York City, United States and find that the addition of even sparsely sampled genome sequence data to patient characteristics supports more accurate identification of CRKP carriers. Model simulations suggest that inference-guided targeted isolation leads to a greater reduction of AMRO burdens compared to alternative, heuristic approaches. Thus, the synergistic effect of utilizing multimodal observations for estimating AMRO carriage risk may inform improved interventions in hospital settings.