Critically ill patients in the intensive-care unit (ICU) are treated by ICU staff, who must reassess patients frequently and interpret patient observations from bedside monitors, mechanical ventilators and laboratory tests. The prodigious amount of real-time data causes information overload, which leads to errors in patient care. ICU-patient care will be improved if we develop and apply better methods of ICU-data analysis and interpretation. We propose to develop, validate, implement, and test a state-of-the-art automated method to automate an ICU-patient monitor. The patient monitor will suggest therapeutic interventions, offer critiques of established therapies, set patient-specific alarm conditions dynamically, highlight clinically meaningful changes in patient state as they occur (and before they turn into emergencies), and provide clinicians with """"""""what-if?"""""""" simulations of alternative treatment options. The ICU-patient monitor will gather multiple measurements as they become available in the ICU, determine the meaning of the measurements with reference to the predictions of a patient-specific integrated model of physiology, and suggest appropriate settings for the mechanical ventilator-settings that maximize the patient's respiratory function and minimize the risk of adverse events. We have built prototype programs that demonstrate the feasibility of our approach. To combine and develop these programs into a system that is suitable for clinical use, we propose the following steps. l. Create an ICU-Patient Simulator. In years 01 and 02, we will create an integrated prediction model of cardiopulmonary physiology, validate the predictions of this model, and implement the model in a patient simulator for the MICU of the Beth Israel hospital. The integrated model will combine two existing models: a belief/network model (VPnet) and a quantitative mathematical model of cardiopulmonary physiology (VentSim). 2. Develop a Model-Based ICU-Patient Monitor. In years 03 and 04, we will expand the simulator to a fully functioning ICU-patient monitor by adding a decision-theoretic preference model for adjusting the controls of the ventilator. We will then explore methods to apply this model to identify critical patient events. Before we implement the automated monitor in the ICU, we will test the hypothesis that the ICU-patient monitor makes recommendations for the settings of the ventilator that are as good as, or better than, the settings that are recommended by specialists in critical- care medicine. 3. Implement and Test the Automated Monitor in the Medical ICU. In year 05, we will evaluate the performance of the real-time monitor in the MICU environment. We will assess the impact of the computer-generated ventilator settings, warnings and alarms, and test the relative effectiveness of the model-based intelligent alarms relative to that of the conventional bedside monitor alarms.
