Validated and curated datasets are essential for studying the spread and control of infectious diseases in hospital settings, requiring clinical information on patients’ evolution and their location. The literature shows that approaches based on Artificial Intelligence (AI) in the development of clinical-support systems have benefits that are increasingly recognized. However, there is a lack of available high-volume data, necessary for trusting such AI models. One effective method in this situation involves the simulation of realistic data. Existing simulators primarily focus on implementing compartmental epidemiological models and contact networks to validate epidemiological hypotheses. Nevertheless, other practical aspects such as the hospital building distribution, shifts or safety policies on infections has received minimal attention. In this paper, we propose a novel approach for a simulator of nosocomial infection spread, combining agent-based patient description, spatial-temporal constraints of the hospital settings, and microorganism behavior driven by epidemiological models. The predictive validity of the model was analyzed considering micro and macro-face validation, parameter calibration based on literature review, model alignment, and sensitive analysis with an expert. This simulation model is useful in monitoring infections and in the decision-making process in a hospital, by helping to detect spatial-temporal patterns and predict statistical data about the disease.
In recent years, Artificial Intelligence (AI) techniques have demonstrated their potential to implement effective data-driven clinical decision support systems. These techniques are strongly dependent on the volume and quality of available clinical data. However, its development is limited due to the problems associated with the use of sensitive and sometimes incomplete clinical data and, for example in the case of machine and deep learning, the low interpretability of the models generated (“black box problem”). The growing concern of the medical AI community by the observed lack of reproducibility and interpretation of these models have entailed the study of more trustworthy and ethical guidelines for AI-based systems.
Some approaches have been proposed to address this issue. Among them, we find the anonymization of real health data and the development of digital twins – dynamic virtual representations of physical objects -. However, both of these techniques share many of the same issues and challenges faced by AI and data analytics, such as availability of quality data, risk of bias, the privacy of individuals (e.g., through data triangulation techniques), ethical behavior in the collection and use of data, confidentiality and consent, among others.
Realistic data simulations can be part of the solution to this problem. Although it is not possible to simulate all the real-world factors that can influence a parameter of study or a specific situation (e.g. changes in decision making, errors, etc.), they have a twofold benefit: from a public health perspective, they enable predictive analysis and an early evaluation of hospital policies; from a medical AI perspective, simulated datasets are helpful for building and evaluating future AI techniques in a more fair and trustworthy way.
The constant increase in the prevalence of healthcare-associated infections (HAIs) caused by multi-resistant microorganisms is currently posing a challenge and is one of the main concerns in public health. Bacteria and other pathogens are capable of evolving and becoming resistant to the drugs that are used to fight them, turning into what is known as multidrug-resistant microorganisms (MDR-microorganism). This resistance complicates the treatment and increases its severity, mortality, and risk of spread. Therefore, a priority issue is to control and prevent MDR bacterial infections, since they involve rises in healthcare costs and a threat to society. Healthcare systems must have the necessary means to be able to evaluate the presence of these infections within hospitals. To this end, the spatial structure of a hospital and the physical distribution of patients over time play important roles in detecting outbreaks and preventing their spread.
The increasing need to study MDR-bacterial infections has led to a variety of computational model implementations of different types. The ones that stand out in the literature are network-based, agent-based, and compartmental models.
Network-based models are developed to study the movement and contact between patients, and therefore, the transmission of diseases, but space does not play any role other than informative. From an epidemiological perspective, compartmental models are classic frameworks for quantifying disease transmission and studying the application of intervention strategies. The population is divided into labeled compartments, they can progress from one to another and, depending on the labels, there are different approaches (e.g. SIR, SIS, SEIR, etc.). However, these approaches do not represent explicitly individual contact within a population but rather show dynamics on a large scale.
In contrast to these, agent-based models are used to study the dynamic processes that involve agents’ interactions with each other and with the environment. In such dynamic processes, both individuals and environments can change and adapt over time, which makes these models suitable for discovering spatial patterns derived from the results of interactions at individual level. They can also be applied to the study of dynamic processes related to the effects of space on health or to the specific processes that are believed to lead to the observed empirical regularities. In a study conducted by Willem et al., they identified 698 papers about agent-based implementations with infectious diseases, of which 89 worked with bacterial infections for different purposes. Another systematic review studied 372 papers on different simulation approaches applied in COVID-19 research and found out that agent-based models were the most used and covered more research areas than the others.
Several studies have used agent-based approaches to analyze the transmissions of infectious diseases in hospitals and the effects of control strategies. Codella et al. developed an agent-based model combined with a Markov model to study the transmission of Clostridioides difficile (CD) and analyzed the performance of several control measures in a mid-size hospital. Nelson et al. developed an economic analysis of the strategies applied in a hospital to control the transmission of CD and conducted probabilistic sensitivity analyses in which all parameter values were allowed to vary simultaneously through 2nd order Monte Carlo simulations. Lee et al. presented a software tool that generates an agent-based simulation to study the spread and control of infectious diseases in any healthcare ecosystem, and they evaluated the performance using real datasets. Haber et al. developed a simulation to explore different regimes that use second-line antibiotics – those given when the initial treatment is not effective or is no longer effective – to successfully treat and reduce the resistance frequency to other drugs. They evaluated this model with several runs to predict the effectiveness of various treatment strategies.
In this paper we present a simulation model with the aim of obtaining a reliable spatial-temporal dataset on the activity of hospitalized patients. This realistic simulator is a goal of a research project on eXplainable AI (see Acknowledgements) applied to the monitoring of infection spreads in hospitals caused by relevant bacteria. This intersection between spatial, temporal and epidemiological information is not easy to achieve by other means and is, in turn, necessary for studies in this field. The contribution of this work is a simulation model that combines (1) a compartmental model to represent the evolution of bacterial infections (macro-model); (2) an agent-based model for the dynamics and spread of the infection as well as the individual actions (micro-model); and (3) spatial-temporal constraints defined by the hospital infrastructure, through the representation of its layout, cleaning policies, and staff shifts.