Abstract
Background/Aims
The diagnosis of heart failure (HF) is frequently delayed until patients are symptomatic enough to require hospitalization. Earlier identification of these patients would allow for the aggressive initiation of preventive strategies, potentially resulting in a decrease in hospitalizations and improved outcomes.
Methods
Patient Electronic Health Record (EHR) data from 39 community practice clinics within the Geisinger Clinic were used. Among primary care patients, 4,644 incident cases of HF were identified between 2001 and 2010 with their diagnosis date determined by specific operational criteria. A validated natural language processing application was applied to primary care encounter progress notes to identify affirmations and denials of Framingham signs and symptoms for heart failure.
Results
During a mean duration of 3.4 years of observation preceding the HF diagnosis date, positive affirmations of HF signs/symptoms were frequently documented. The median duration of time between first documentation of a positive sign/symptom and the date of clinical diagnosis was over 2 years for several, and greater than one year for most signs/symptoms. Surprisingly, the majority of signs/symptoms were documented to come and go (affirmation followed later by a negation) multiple times. In particular, ankle edema, rales, dyspnea on exertion and hepatomegaly were all documented to come and go a median of 5 or more separate times before clinical diagnosis.
Conclusions
These results suggest that the waxing and waning course of HF signs and symptoms in the years prior to a clinical diagnosis of HF may pose challenges to the earlier diagnosis of HF in a primary care setting. The clinical application of automated tools to identify HF signs and symptoms within the EHR could substantially improve the early identification and treatment of these patients.
Keywords: Heart Failure, Natural Language Processing, Early Diagnosis