Even though US hospitals have widely adopted electronic health record (EHR) documentation of patient care, interoperability of these systems remains an issue, leading to challenges in data integration. In the operating room (OR) setting, during surgery, physiological waveforms (arterial pressure, EKG, SpO2, central venous pressure, etc.) represent a large source of information used by clinical monitors to extract and display information in order for healthcare providers to make clinical decisions. Integration and synchronization of high-quality EHR and physiological waveform data in large datasets of surgical patients would allow machine learning and deep learning approaches to plumb these datasets for clinically relevant signatures that would promote advanced OR patient monitoring systems to define present state, predict state trajectory, suggest effective counter measures to minimize patients decompensated states, and define the usefulness and efficacy of new monitoring devices. The objective of this proposal is to focus the resources of an interdisciplinary team from academia (University of California Los Angeles (UCLA), University of California Irvine (UCI), and Carnegie Mellon University Computer Sciences), industry (Edwards Lifesciences Critical Care), and clinical medicine (anesthesiology, surgery, and critical care at UCLA, UCI, Beth Israel, and University of Pittsburgh Medical Center) to create, develop, and organize large surgical datasets combining EHR and high fidelity physiological waveform data, to make these datasets freely accessible, and to develop new predictive/forecasting monitoring systems for the surgical patients. The study will begin with the development of a machine learning algorithm to predict cardiovascular collapse during surgery. This algorithm development will be based on physiological signatures predictive of cardiovascular collapse identified in the animal models of shock. The study hypothesis is that the combination of two separate OR databases containing EHR and physiological waveforms will allow for training and development of monitoring solutions, predictive and/or prescriptive analytics tools, clinical decision support, and validate them on an independent, external validation database. The surgical setting is relevant because although 5.7 million Americans are admitted annually to an Intensive Care Unit, more than 50 million undergo surgery. OR databases are unique in medicine because: 1) Changes occur quickly and the lead-time before an event is compressed; 2) Knowledge of baseline/pre-stress status of surgical patients allows normalization, calibration, and markedly enhances prediction; 3) Continuous and immediate presence of dense skilled acute care practitioners allows faster implementation of complex treatment algorithms in the OR; and 4) Defined stages, procedures, and stressors allow building large common relational database registries. By helping to focus the provider's attention on significant events and changes in the patient's state and by suggesting physiological interpretations of that state, such systems will permit early detection of complex problems and provide guidance on therapeutic interventions improving patient outcomes.
The objective of this proposal is to focus the resources of an interdisciplinary team from academia (University of California Los Angeles (UCLA), University of California Irvine (UCI), and Carnegie Mellon University Computer Sciences), industry (Edwards Lifesciences Critical Care), and clinical medicine (anesthesiology, surgery, and critical care at UCLA, UCI, Beth Israel, and University of Pittsburgh Medical Center) to create, develop, and organize large surgical datasets combining electronic health record and high-fidelity physiological waveform data, to make these datasets freely accessible, and to develop new predictive/forecasting monitoring systems for the surgical patients. We hypothesize that the combination of two separate OR databases containing electronic health record and physiological waveforms will allow us and other investigators to train and develop monitoring solutions, predictive and/or prescriptive analytics tools, clinical decision support, and validate them on an independent, external validation database. We will use these datasets to develop a new physiological predictive tool of cardiovascular collapse during surgery and to validate the new data's potential utility to a broader community of researchers.
