Design and evaluation of an associative classification framework to identify disease cohorts in the electronic health record.

摘要

With the growing national dissemination of the electronic health record (EHR), there are expectations that the public will benefit from biomedical research and discovery enabled by electronic health data. Clinical data are needed for many diseases and conditions to meet the demands of rapidly advancing genomic and proteomic research. Many biomedical research advancements require rapid access to clinical data as well as broad population coverage. A fundamental issue in the secondary use of clinical data for scientific research is the identification of study cohorts of individuals with a disease or medical condition of interest. The problem addressed in this work is the need for generalized, efficient methods to identify cohorts in the EHR for use in biomedical research.;To approach this problem, an associative classification framework was designed with the goal of accurate and rapid identification of cases for biomedical research: (1) a set of exemplars for a given medical condition are presented to the framework, (2) a predictive rule set comprised of EHR attributes is generated by the framework, and (3) the rule set is applied to the EHR to identify additional patients that may have the specified condition. Based on this functionality, the approach was termed the 'cohort amplification' framework.;The development and evaluation of the cohort amplification framework are the subject of this dissertation. An overview of the framework design is presented. Improvements to some standard associative classification methods are described and validated. A qualitative evaluation of predictive rules to identify diabetes cases and a study of the accuracy of identification of asthma cases in the EHR using framework-generated prediction rules are reported. The framework demonstrated accurate and reliable rules to identify diabetes and asthma cases in the EHR and contributed to methods for identification of biomedical research cohorts.