HYBRID ONTOLOGY AND MACHINE LEARNING BASED METHODS USING EMR DATA FOR EFFECTIVE CLINICAL DECISION SUPPORT (CDS); A SEPSIS CASE STUDY USING ICU DATA

DESCRIPTION provided by applicant Our research is focused on the use of advanced ontological models as a foundation for computerized Clinical Decision Support CDS systems that link hospitalized patient data routinely captured in electronic medical records EMRs with medical knowledge to effectively influence timely awareness and treatment choices by clinicians Our phase goal is to demonstrate how our advanced CDS technology has the potential to improve preventable mortality outcomes associated with sepsis in ICU patients We distinguish two types of CDS algorithms available today for detecting and or predicting sepsis using EMR data evidence based andquot knowledge drivenandquot detection algorithms and data driven andquot predictiveandquot algorithms based on machine learning ML techniques Recent studies indicate currently available CDS tools do not reduce risk of death in hospitalized patients We believe this may be because diseases such as sepsis are time sensitive complex syndromes and also due to the challenges of computerized reuse of unstructured EMR data Our sepsis ontology models this complexity to a provide enhanced knowledge driven sepsis risk stratified cohort classifications that help guide interventions b support accurate natural language processing NLP of free text clinical notes to enhance real time sepsis risk detection and c improve the accuracy of data driven prediction models when used in conjunction with ML training algorithms Our CDS is based on proprietary cluster computing technology we call andquot VFusionandquot designed to efficiently deal with the generation and practical use of large application domain specific ontologies Our sepsis ontology employs a family of upper level ontologies combined with granular evidence based domain ontologies configurable rule sets e g first order logic based and required components of reference terminologies Our research will use openly available ICU patient data to establish statistical detection prediction performance metrics using this ontology in modes of use as a knowledge based screening tool to detect subtle signs of sepsis in individualized hospitalized patients used in conjunction with ML to improve data driven predictive performance We will measure specificity sensitivity and positive negative predictive power of our hybrid ontology based technology to demonstrate dramatically improved performance over existing CDS algorithms In Phase II we plan a retrospective demonstration with a much larger sample of patients to include non ICU patients in collaboration with a major healthcare system Our product vision is an early inpatient sepsis detection algorithm with high accuracy embodied as a andquot plug in applicationandquot compatible with any modern EMR platform in use at a client hospital effective in both ICU and non ICU care settings PUBLIC HEALTH RELEVANCE Even with the advent of powerful new computer medical record technologies used in hospitals the risk of death has not been reduced in hospitalized patients The goal of this research effort is to combine medical record systems with advanced cognitive technologies that are commonly used today in products such as the iPhone andquot Siriandquot to improve clinician awareness and decisions that will dramatically reduce preventable mortality Our initial use case will be the management of sepsis the leading cause of death in non coronary intensive care units in hospitals