Publicatie
Publicatie datum
Bayesian network models for the management of ventilator-associated pneumonia = Bayesiaanse netwerkmodellen voor de diagnose en behandeling van beademingsgerelateerde longontsteking.
Visscher, S. Bayesian network models for the management of ventilator-associated pneumonia = Bayesiaanse netwerkmodellen voor de diagnose en behandeling van beademingsgerelateerde longontsteking. Utrecht: Universiteit Utrecht, 2008. 177 p. Proefschrift Universiteit Utrecht.
The purpose of the research described in this thesis was to develop Bayesian network models for the analysis of patient data, as well as to use such a model as a clinical decision-support system for assisting clinicians in the diagnosis and treatment of ventilator-associated pneumonia (VAP) in mechanically ventilated ICU patients. Both construction of the models and use of such a decision-support system were expected to be guided by clinical expertise and patient data collected within the hospital. For this research, we were in particularly interested in patient data collected routinely by the electronic patient record system, EPR for short. Although not yet used in all hospital wards, EPR has become relatively common in ICUs. The exploitation of routinely collected ICU data, therefore, defined the context of the research. Typical for EPR data is that it includes both clinical and laboratory data that is time oriented. In this thesis, methods, techniques and tools are developed for using these temporal data to support clinical decision making. To this end, graphical statistical models, in particular Bayesian networks, have been chosen as starting point, since these formalisms enable the deduction of uncertain knowledge, both on the basis of knowledge from a domain and on the basis of factual data. Moreover, dynamic Bayesian networks provide the opportunity to represent both temporal and atemporal relationships. The validity of such associations should then be clinically judged. In this project specific attention has been paid to the structure and content of both atemporal and temporal Bayesian networks, the learning of networks from data and to the role of expert knowledge in this. A Bayesian decision-support system (BDSS) for diagnosing and treating VAP was developed that, in principle, can be used by intensivists. A previously collected database of data from mechanically ventilated ICU patients was used for improving and assessing the performance of the BDSS.
The purpose of the research described in this thesis was to develop Bayesian network models for the analysis of patient data, as well as to use such a model as a clinical decision-support system for assisting clinicians in the diagnosis and treatment of ventilator-associated pneumonia (VAP) in mechanically ventilated ICU patients. Both construction of the models and use of such a decision-support system were expected to be guided by clinical expertise and patient data collected within the hospital. For this research, we were in particularly interested in patient data collected routinely by the electronic patient record system, EPR for short. Although not yet used in all hospital wards, EPR has become relatively common in ICUs. The exploitation of routinely collected ICU data, therefore, defined the context of the research. Typical for EPR data is that it includes both clinical and laboratory data that is time oriented. In this thesis, methods, techniques and tools are developed for using these temporal data to support clinical decision making. To this end, graphical statistical models, in particular Bayesian networks, have been chosen as starting point, since these formalisms enable the deduction of uncertain knowledge, both on the basis of knowledge from a domain and on the basis of factual data. Moreover, dynamic Bayesian networks provide the opportunity to represent both temporal and atemporal relationships. The validity of such associations should then be clinically judged. In this project specific attention has been paid to the structure and content of both atemporal and temporal Bayesian networks, the learning of networks from data and to the role of expert knowledge in this. A Bayesian decision-support system (BDSS) for diagnosing and treating VAP was developed that, in principle, can be used by intensivists. A previously collected database of data from mechanically ventilated ICU patients was used for improving and assessing the performance of the BDSS.