Discovering Care Coordination Practice Patterns in the EMR: Interpretation and Impact on Patient Outcomes

点击翻译按钮获取中文摘要

Healthcare expenditures in the United States reached $3.5 trillion in 2017, up 4.6 percent from 2016. It has been recognized that prolonged length of stay (LOS) and unplanned readmission are two of the primary causes of higher healthcare costs. Determining which factors are associated with prolonged LOS and unplanned readmission will provide valuable knowledge about how to reduce costs and improve care delivery. The Agency for Healthcare Research and Quality (AHRQ) has recognized that care fragmentation under a fee- for-service system can lead to various problems, including poor harmonization of services and unnecessary testing and procedures, all of which have the potential to extend LOS and unplanned readmissions. Effective care coordination, has been proposed to resolve many of these problems, and is a priority of the National Quality Strategy, which is led by AHRQ. Yet, there are numerous challenges to measuring the effectiveness of care coordination. In particular, there is a lack of a clear relationship with a patient’s outcome (e.g., prolonged LOS or unplanned readmission). Electronic medical record (EMR)-based care coordination measures have been highlighted by AHRQ for three potential advantages: i) minimal data collection burden, ii) rich clinical context and iii) longitudinal patient observation. However, current EMR-based measures focus on an assessment of EMR systems (e.g., meaningful use) and compare effectiveness of care at a coarse-grained level (e.g., the relation between meaningful use of an EMR system and reduction in LOS or unplanned readmission rates). Unfortunately, such measures neglect the specific drivers (e.g., variations of interactions between healthcare professionals) of variability in LOS and unplanned readmission rates. In this project, we will develop an EMR-based framework to characterize care coordination at a fine-grained level, which accounts for the interaction network between two or more healthcare professionals (e.g., doctors, nurses, social workers, care managers, and supporting staff) involved in a patient’s care - and measure its impact on LOS and unplanned readmission. To achieve the goal, we will design i) data mining algorithms to automatically learn care coordination patterns and analyze LOS and unplanned readmission from the EMRs of ~2.3 million patients at a large academic medical center with a long history of EMR use; ii) hypothesis-driven approaches to quantify the relationship between a learned pattern and LOS and unplanned readmission, where a patient’s demographics (e.g., age, race and sex) will be considered as confounding variables; and iii) an interpretation process to translate the inferred patterns into actionable criteria for HCOs. This research is notable because methods created in the project can be served as a scientific basis to automatically i) learn care coordination patterns across a wild range of healthcare services and health conditions; and ii) measure the effectiveness of these patterns via their relationships with various patient outcomes (e.g., LOS and unplanned readmission).