Machine learning to optimize the diagnostic performance of natriuretic peptides for acute heart failure across age groups

BACKGROUND AND AIMS: N-terminal pro-B-type natriuretic peptide (NT-proBNP) concentrations are influenced by age, which may influence the diagnostic performance of this peptide. Machine learning approaches incorporating NT-proBNP and age as continuous measures may have improved diagnostic performance. METHODS: We pooled individual patient-level data for 10 369 patients [median age 73 years (25th-75th percentile: 59-82)] with suspected acute heart failure across fourteen studies. The diagnostic performance of guideline-recommended NT-proBNP thresholds (uniform rule-out threshold of 300 pg/mL and age-stratified rule-in thresholds of 450, 900, and 1800 pg/mL for patients 75 years, respectively) and the Collaboration for the Diagnosis and Evaluation of Heart Failure (CoDE-HF) machine learning model were evaluated using random effects meta-analysis across age groups. RESULTS: Overall, 43.9% (4549/10 369) of patients had an adjudicated diagnosis of acute heart failure. The negative predictive value (NPV) of the rule-out threshold of 300 pg/mL was lower in older patients [NPV 88.7% (confidence interval (CI) 84.2-92.1%) in patients ≥80 years vs 98.9% (97.6-99.5%)