Healthcare-associated infections (HAIs) brought about by multi-drug resistant organisms (MDROs) prove a daunting obstacle for healthcare systems due to their impact on patient safety and cost implications. These infections could already be present in patients upon arrival at the hospital or might be acquired during their stay. Identification of such infections remains complicated due largely to testing limitations and delayed results. Though recent innovations in mathematical modeling and machine learning intended to identify at-risk patients, these models often fail to utilize valuable electronic health record (EHR) data, and machine learning methods typically lack in-depth understanding of underlying processes.
Addressing these gaps, a novel framework has been proposed – NeurABM. This framework combines neural networks with agent-based models (ABM), drawing on the strengths of both methods. Notably, NeurABM has shown to significantly outperform current methods, representing a considerable advancement in accurately identifying importation cases and foreseeing future nosocomial infections in clinical practice.
This improvement in identification and forecasting could facilitate better hospital management. As practices to prevent the spread of MDROs within hospitals – such as testing, quarantine, and isolation – take up precious healthcare resources, informed decisions on prioritizing these resources are essential to ensure effective healthcare service delivery. In this respect, agent-based models and machine learning have proven to be key tools for optimization. Building on this, the NeurABM framework aims to identify HAI importation and predict future nosocomial infection cases by amalgamating a neural network and an ABM and training both simultaneously. This approach mitigates the issues encountered when either component is used individually.
NeurABM was demonstrated using electronic health record data for patients in the University of Virginia (UVA) hospital intensive care units. The results underscore that NeurABM not only identifies importation cases but also forecasts future nosocomial infection cases more effectively than other machine learning or modeling-based standards.
Ultimately, the success of the NeurABM framework is a testament to the efficacy of integrating neural networks and mechanistic models in a cohesive manner. Its applicability is not just limited to predicting and preventing HAIs, it represents a versatile framework that could be adapted to other ABMs or EHR data, thereby potentially improving the healthcare landscape.