Abstract 12035: Text and Data Mining of Longitudinal Electronic Health Records (EHRs) in a Primary Care Population Can Identify Heart Failure (HF) Patients Months to Years Prior to Formal Diagnosis Using the Framingham Criteria
Introduction: The societal and individual health burden of HF is enormous. While its early recognition offers the possibility of initiating preventive action, earlier detection of HF can be challenging in primary care due to its complexity and overlapping symptoms with other disease states. We sought to evaluate whether sophisticated text and data mining of longitudinal EHRs might allow for the diagnosis of HF prior to its clinical recognition.
Methods: We extracted complete data from 6355 incident HF patients from between 2003 and 2010 seen in the Geisinger Clinic primary care practices. The well-validated Framingham criteria for HF, which identifies definite HF based on the presence of either 2 major, or 1 major and 2 minor criterion, were utilized. Text mining software was validated for detection of positive affirmation of the diagnostic criteria. The software was then applied to ∼575,000 clinical notes for the 6355 cases. The date of formal diagnosis for HF cases was the date that HF was documented in their EHRs.
Results: All patients were followed in their primary care clinic for an average of 41 months prior to being diagnosed with HF. Already at 2 years prior to receiving a formal diagnosis with HF 21% of patients met the Framingham diagnostic criteria for definite HF. Over time the percentage of patients meeting the Framingham criteria for the diagnosis of HF continued to increase such that by 3 months prior to diagnosis 44% met diagnostic criteria. (Figure) At the time of clinical diagnosis of HF 60% of these patients met the Framingham criteria for definite HF, similar to what was found in the original Framingham cohort.
Conclusions: In the primary care setting, the use of sophisticated text mining techniques of EHRs may allow for the diagnosis of HF months to years prior to their clinical diagnosis. Prospectively applying these tools clinically would be expected to substantially impact HF morbidity and mortality.
- © 2011 by American Heart Association, Inc.