•  
  •  
 

Specificity of the Framingham Heart Failure Signs and Symptoms: Bridging the Gap Between Epidemiological Research and Clinical Practice

Publication Date

8-10-2017

Keywords

cardiovascular disease, primary care, natural language processing, epidemiology

Abstract

Background: While the Framingham Heart Failure Signs and Symptoms (FHFSS) have been around for decades, little is known about the specificity of the relation between presence of each FHFSS and early detection of heart failure. The FHFSS continue to be used in epidemiological research; however, their clinical relevance has been criticized. We examined variation in the ability of individual FHFSS to predict heart failure diagnosis by assertion or denial of each feature, by counts, and by a composite score.

Methods: We extracted electronic health record data from 2001 to 2010 from a single health system. A total of 1,684 incident heart failure cases were identified, and 13,525 matched controls were selected from the same primary care practices. We performed LASSO logistic regression analysis to determine which FHFSS are the strongest predictors of heart failure. We explored predictive value by observation window (12 vs 24 months) and by prediction window (6, 12, 18 and 24 months). FHFSS score statistics were used to determine if prediction was improved and to determine if changes in score over time improved the model performance.

Results: Our findings indicate that: 1) Less features yield better prediction performance (LASSO model), with the following FHFSS features being most predictive: negative/positive acute pulmonary edema, negative/positive bilateral ankle edema, positive dyspnea on ordinary exertion, positive neck vein distension, negative/positive pleural effusion, and negative/positive radiographic cardiomegaly; 2) More patient encounters perform better than less encounters and also have a greater average number of FHFSS mentions; and 3) Different features are more predictive during specific prediction windows.

Conclusion: This study aims to bridge the gap between epidemiological research and clinical practice as it relates to a highly prevalent, serious, costly disease. In practice, providers are left with the difficult task of basing their decisions to act on the examination of a patient in the moment, not based on a sophisticated quantitative assessment of longitudinal patient data. Our findings show that a smaller number of FHFSS can be used individually to predict early onset of heart failure.

Share

COinS
 

Submitted

June 23rd, 2017

Accepted

August 10th, 2017