MASSACHUSETTS INSTITUTE OF TECHNOLOGY
The overall scope of the parent project (5R01 EB 001659) is to develop and evaluate advanced patient-monitoring and clinical decision-support concepts that will improve the efficiency, accuracy, and timeliness of decision-making in critical care. In pursuit of this goal, a key achievement of the project has been the development of the Multi-parameter Intelligent Monitoring in Intensive Care II (MIMIC II) database, the first version of which was publicly released in early 2009. The very rich 30,000+ patient records of the MIMIC II database enable a large number of predictive patient-monitoring algorithms to be developed and tested. Having additional personnel take advantage of this unique resource and aid in the development and evaluation of candidate patient-monitoring concepts will accelerate progress of the parent grant. Our Administrative Supplement aims to advance the project goals of the parent grant by supporting two additional graduate student research assistants and increasing the percent-effort of a postdoctoral associate on the project. With the supplement, we target exclusively the development and evaluation of candidate monitoring concepts, predictive algorithms, and clinical alerts. Since the first version of the MIMIC II database has just been released, the additional research staff will find an existing (and growing) infrastructure for their research. Aided by this ARRA Supplement, we focus our efforts on three projects: 1. Continuous and non-invasive cardiac output and total peripheral resistance estimation (with extensions to ejection fraction, cardiac contractility, and end-diastolic ventricular volume estimation). During the first funding cycle, we have made progress toward continuous monitoring of left ventricular ejection fraction. The enhanced MIMIC II database contains continuous waveform records during cardiac surgery along with thermodilution measurements of cardiac output and intermittent trans-esophageal measurements of ejection fraction, stroke volume, and end-diastolic volumes. In addition to the MIMIC II ICU patient records, the enhanced database provides for an excellent resource to test these algorithms on clinical data. 2. Noninvasive, continuous estimation of intracranial pressure (ICP). ICP is a key variable to track in many neurovascular pathologies, such as stroke and cerebral hemorrhage, as well as in traumatic brain injury. Current measurement modalities are quite invasive and require the penetration of the skull and the placement of a pressure-sensitive transducer in the ventricular space, the brain parenchyma, or the subdural space. Because of the attendant risks for infection and damage to vital brain structures, ICP measurements are currently restricted to the sickest of patients only, though a much larger patient pool would undoubtedly benefit from assessment of ICP. Our group has developed very promising algorithms to estimate ICP from noninvasive measurements of cerebral blood flow velocity and arterial blood pressure. Under ARRA funding, we will develop these algorithms further and validate the ICP estimates against gold standard measurements in a sizable patient population. 3. Improved patient state/mortality model. An elaborate (and maybe partially mechanistic) approach to predicting patient state of mortality is to assume that there is an underlying patient 'state' that is represented by a small set of variables. Under this assumption, we could use a partially observable Markov process model to model the state evolution. Work in other areas of engineering has shown that both the approximation methods to solve such models and model-simplification methods have reduced the computational burden associated with these Markov models to the point that they can be applied to the problem sizes we are facing with the ICU population.