Individualised NT-proBNP thresholds for the diagnosis of heart failure

Abstract Background Heart failure (HF) is underdiagnosed in the community. N-terminal pro B-type natriuretic peptide (NT-proBNP) has variable sensitivity for HF across different subgroups, particularly for HF with preserved ejection fraction (HFpEF). Purpose We calculated individualised NT-proBNP thresholds with 90% sensitivity for the detection of HF, adjusted for age and comorbidities. We investigated whether individualised thresholds had greater sensitivity for the detection of HF, and whether adverse event rates were similarly low in patients with NT-proBNP levels below individualised and fixed thresholds. Methods 1320 patients were recruited from a prospective clinical cohort referred for cardiovascular magnetic resonance (CMR). HF diagnosis was blinded to NT-proBNP and based on patient-reported symptoms and CMR imaging evidence of structural abnormalities and ejection fraction. Individualised NT-proBNP thresholds were derived using a novel nonparametric Bayesian inference method to model covariate-specific receiver operating characteristic (ROC) curves. 1-3 Baseline characteristics were considered for inclusion as covariates based on significant multivariable associations with NT-proBNP. The diagnostic performance of individualised and fixed NT-proBNP thresholds (125 and 400 pg/ml) for the detection of HF were compared. Outcome data were obtained from Hospital Episode Statistics ICD-10 codes. The associations between individualised thresholds and the primary composite outcome of first heart failure hospitalisation or cardiovascular (CV) death were investigated using Fine-Gray analysis, with non-CV death as a competing risk. Results Individualised NT-proBNP thresholds ranged from 31 to 2295 pg/ml, adjusting for age, estimated glomerular filtration rate (eGFR), body mass index (BMI), atrial fibrillation (AF) and ischaemic heart disease (IHD) (Figure 1). Individualised thresholds were more sensitive for the detection of all HF (0.90; 0.80 for 125 pg/ml; 0.55 for 400 pg/ml, Figure 2A), HFpEF (0.87; 0.74 for 125 pg/ml; 0.43 for 400 pg/ml), and HF with reduced ejection fraction (0.98; 0.94 for 125 pg/ml; 0.83 for 400 pg/ml), and achieved a high sensitivity of ~0.90 in all patient subgroups. The primary outcome occurred in 168 (12.7%) patients over a median follow-up duration of 2620 days (interquartile range 2433 – 2784). Individualised thresholds associated with adverse outcomes (hazard ratio and 95% confidence interval 6.88 (3.04 – 15.60)). Patients with NT-proBNP levels below individualised thresholds and 125 pg/ml had similarly low event rates (Figure 2B). Conclusions Individualised NT-proBNP thresholds, optimised for increased sensitivity, may improve the detection of HF in the community. This method simultaneously adjusts for five important covariates and can be readily incorporated into electronic health care records.Figure 1 Figure 2