Machine learning to optimize use of natriuretic peptides in the diagnosis of acute heart failure

Abstract

Aims: B-type natriuretic peptide (BNP) and mid-regional pro-atrial natriuretic peptide (MR-proANP) testing are guideline-recommended to aid in the diagnosis of acute heart failure. Nevertheless, the diagnostic performance of these biomarkers is uncertain.

Methods and results: We performed a systematic review and individual patient-level data meta-analysis to evaluate the diagnostic performance of BNP and MR-proANP. We subsequently developed and externally validated a decision-support tool called CoDE-HF that combines natriuretic peptide concentrations with clinical variables using machine learning to report the probability of acute heart failure. Fourteen studies from 12 countries provided individual patient-level data in 8493 patients for BNP and 3899 patients for MR-proANP, in whom, 48.3% (4105/8493) and 41.3% (1611/3899) had an adjudicated diagnosis of acute heart failure, respectively. The negative predictive value (NPV) of guideline-recommended thresholds for BNP (100 pg/mL) and MR-proANP (120 pmol/L) was 93.6% (95% confidence interval 88.4-96.6%) and 95.6% (92.2-97.6%), respectively, whilst the positive predictive value (PPV) was 68.8% (62.9-74.2%) and 64.8% (56.3-72.5%). Significant heterogeneity in the performance of these thresholds was observed across important subgroups. CoDE-HF was well calibrated with excellent discrimination in those without prior acute heart failure for both BNP and MR-proANP [area under the curve of 0.914 (0.906-0.921) and 0.929 (0.919-0.939), and Brier scores of 0.110 and 0.094, respectively]. CoDE-HF with BNP and MR-proANP identified 30% and 48% as low-probability [NPV of 98.5% (97.1-99.3%) and 98.5% (97.7-99.0%)], and 30% and 28% as high-probability [PPV of 78.6% (70.4-85.0%) and 75.1% (70.9-78.9%)], respectively, and performed consistently across subgroups.

Conclusion: The diagnostic performance of guideline-recommended BNP and MR-proANP thresholds for acute heart failure varied significantly across patient subgroups. A decision-support tool that combines natriuretic peptides and clinical variables was more accurate and supports more individualized diagnosis.

Study registration: PROSPERO number, CRD42019159407.

Keywords: Heart failure; Machine learning; Natriuretic peptide.