Clinical Data Mining: Application and Lesson Learned
With the financial incentives provided by the Health Information Technology for Economic and Clinical Health (HITECH) Act for hospitals to adopt electronic health records (EHR), more clinical data in electronic format are expected to be available to public health agencies and researchers to improve quality of healthcare; however, there is limited research on comprehensive data analyses applied to various linked patient data types such as clinician reports, laboratory reports, nurse assessments, medications, etc. In this talk, we will present our recent research on automated influenza case detection from ED visits and hospital readmission prediction for heart failure and pediatric seizure from inpatient visits. We employed Bayesian networks, natural language processing, and ontology to identify predictive models from multiple linked patient data types. The results demonstrated that our approach outperformed methods from published literature and a commercial tool. We will discuss the lessons we learned from the applications.