A SIMPLE CLINICAL DECISION TOOL FOR DIAGNOSING ACUTE HEART FAILURE IN THE UNDIFFERENTIATED DYSPNEIC PATIENT USING A BAYESIAN APPROACH

Abstract

Diagnosing acute heart failure (AHF) in the undifferentiated dyspneic emergency department (ED) patient can be challenging, even with natriuretic peptide binary testing. Structured on a Bayesian approach, we had retrospectively derived and validated a promising mathematical diagnostic model (SoB-HF) for potential AHF cases based simply on initial clinical impression, patient age and absolute NT-proBNP value. We sought to prospectively compare the diagnostic accuracy of SoB-HF with that of emergency physician (EP) pretest probability for AHF, NT-proBNP binary value and final EP diagnosis (EPDx). Undifferentiated dyspneic ED patients with an EP assessed initial indeterminate (21-79%) pretest probability for AHF were enrolled across 4 international sites. At time of ED disposition, the EP recorded EPDx as “AHF” or “no AHF”. Receiver-operator characteristic (ROC) curves were constructed and area under the curve (AUC) calculated to illustrate both pretest AHF and SoB-HF posttest value agreement with gold standard adjudicated diagnosis by two blinded cardiology experts (AdjDx) with post index visit 60 day records For model agreement, optimal cut-points using sensitivity, specificity and likelihood ratios (LR) were calculated. If the SoB-HF model showed superior performance, then an App would be created to facilitate its use. One hundred ninety seven patients were enrolled (43% male, mean age 64 years, 53% with a history AHF, 49% with a history COPD), EP pretest AHF had a ROC AUC of 0.76 vs. 0.93 for SoB-HF (p = 0.000001). Adj Dx was 40% prevalence AHF. EPDx accuracy was 75%. NT-proBNP testing with standard cut-points had 88% (95% CI 79, 94) sensitivity, 78% (95% CI 70, 84) specificity. SoB-HF posttest optimal single cutpoint of 0.515 had 90% (95% CI 82, 96) sensitivity, 88% (95% CI 82, 93) specificity. SoB-HF posttest values > 0.65 had a +LR>10 and posttest values < 0.47 had a -LR of <0.1, together redirecting 84% of uncertain cases with 92% accuracy. An App was created to calculate the SoB-HF probability value and its corresponding LR. This simple clinical decision tool demonstrates superior diagnostic accuracy in a prospective cohort of dyspneic ED patients with an indeterminate pretest probability for AHF. With clinical implementation, further studies of its sensitivity and impact analyses are warranted.