Modelling Hospital Acquired Clostridium difficile Infections And Its Transmission In Acute Hospital Settings

Abstract

The thesis explored a number of fundamental issues regarding the development of predictive models for hospital acquired Clostridium difficile infection (HA CDI) and its outbreaks. As predictive modeling for hospital acquired infection is still an emerging field and the ability to analyse HA CDI and potential outbreaks are in a developmental stage, the research documented in this thesis is exploratory and preliminary. Predictive modeling for the outbreak of hospital acquired infections can be considered at two levels: population and individual. We provide a comprehensive review regarding modeling methodology in this field at both population level and individual level. The transmission of HA CDI is not well understood. An agent based simulation model was built to evaluate the relative importance of the potential sources of Clostridium difficile (C. difficile) infection in a non-outbreak ward setting in an acute care hospital. The model was calibrated through a two stage procedure which utilized Latin Hypercube Sampling methodology and Genetic Algorithm optimization to capture five different patterns reported in the literature. A number of aspects of the model including housekeeping, hand hygiene compliance, patient turnover, and antibiotic pressure were explored. Based on the modeling results, several prevention policies are recommended. One widely used tool to better understand the dynamics of infectious disease outbreaks is network epidemiology. We explored the potential of using network statistics for the prediction of the transmission of HA CDIs in the hospital. Two types of dynamic networks were studied: ward level contacts and hospital transfers. An innovative method that combines time series data mining and predictive classification models was introduced for the analysis of these dynamic networks and for the prediction of HA CDI transmission. The results suggest that the network statistics extracted from the dynamic networks are potential predictors for the transmission of HA CDIs. We explored the potential of using the “multiple modeling methods approach” to predict HA CDI patient at risk by using the data from the information systems in the hospital. A range of machine learning predictive models were utilized to analyse collected data from a hospital. Our results suggest that the multiple modeling methods approach is able to improve prediction performance and to reveal new insights in the data set. We recommend that this approach might be considered for future studies on the predictive model construction and risk factor analysis.