Information Extraction from EMRs to Predict Readmission Following Acute Myocardial Infarctions
Each year, nearly 635,000 people in the United States will have their first acute myocardial infarction (heart attack) and about 20% will be re-hospitalized within 30-days of their incident discharge. Re-hospitalizations are costly for health systems, insurers, patients and their families. As such, there is great value and emphasis placed on readmission reduction programs, many of which are centered around a predictive tool. The Brown Lab developed a clinical decision support tool to enumerate the risk for 30-day hospital readmissions among patients hospitalized with an AMI. The project harmonized electronic health record data from Dartmouth Hitchcock Health and Vanderbilt University Medical Center to the Observational Medical Outcomes Partnership (OMOP) common data model. In addition, a natural language processing (NLP) model called Moonstone was deployed on clinical notes from both health systems, which extracted information on social risk factors. Together, the structured clinical data and NLP-derived social risk factors were used in the development and external validation of five prediction models. Best performing models achieved good discriminatory and calibration performance.
Study status: Final analysis
To learn more about this project, please see major publications below:
- Development of electronic health record-based prediction models for 30-day readmission risk among patients hospitalized for acute myocardial infarctionOpens in a new window
- Adaptation of an NLP system to a new healthcare environment to identify social determinants of healthOpens in a new window
- Information extraction from electronic health records to predict readmission following acute myocardial infarction: Does natural language processing using clinical notes improve prediction of readmission?Opens in a new window