Abstract 12217: A Hierarchical Model for Stage B Heart Failure Screening- Results of a Classification and Regression-Based Decision Tree

Abstract

Listen

Translate article

Purpose: In pts with stage A heart failure (SAHF), the identification of stage B HF relies on echo, but application of this to community based screening is limited by access and cost. We sought to use clinical characteristics and functional capacity to select at risk subjects for echo. Methods: 308 asymptomatic subjects ≥65 years with the presence of ≥ 1 HF risk factor (hypertension, type 2 diabetes, obesity, previous chemotherapy, family history of heart failure and history of cardiac disease but not in HF) were recruited from the local community. All subjects underwent baseline risk quantification using ARIC risk score, functional assessment by 6 minute-walk test (6MW) and a comprehensive echocardiogram including conventional diastology and speckle tracking analysis for global longitudinal strain (GLS) and diastolic strain (DS). New HF was assessed at 14±4 m, based on 3 independent cardiologists using Framingham criteria. A classification and regression-based decision tree (CART) was built using baseline characteristics as independent variables. Results: In the CART model, a 3-level decision tree model was chosen to predict 19 pts (6%) who developed new HF at 1 year. The best predictor was 6MW followed by ARIC risk score (Figure). Using this model, 190 individuals (6MW≤330, or ARIC score>4.75%) were classified as high risk (9.5% events) and 118 (38%) were classified as low risk (0.8% events). GLS, DS and conventional diastolic measures (e’, E/e’) were significantly impaired in the high risk group (Figure). The AUC of the combined CART model with GLS was 0.72 (95% CI: 0.62-0.82, p=0.001) for new HF. Conclusion: Pts with SAHF with low HF risk and preserved exercise capacity have preserved LV function and a low event rate. GLS assessment in SAHF may be considered in pts with >4.75% 4 year risk, or 6MW≤330m.