Clinical significance of a machine learning model based on short-term changes in NT-proBNP after TAVR

Abstract

Objectives: The purpose of this study was to investigate the short-term dynamic changes in NT-proBNP after transcatheter aortic valve replacement and to evaluate its availability in prognostic assessment based on machine learning. Methods: The differences in the NT-proBNP ratio between baseline, 30-day, and 6-month follow-up of patients in the internal derivation cohort (n = 1115) were recorded as D1 and D2; the difference ratio of the NT-proBNP ratio (D2/D1) was recorded as DR. According to the DR-based range, patients were divided into the DR-1 (DR ≥ 1), DR-2 (0 ≤ DR < 1), DR-3 (−0.5 ≤ DR < 0), and DR-4 groups (DR < −0.5). Results: In the internal derivation cohort, 351, 461, 186, and 117 cases were assigned to the DR-1, DR-2, DR-3, and DR-4 groups, respectively. The 4-year primary end-point rates in the four groups were 13.1 %, 18.0 %, 26.9 %, and 33.3 % (log-rank, p < 0.001). A model for predicting the incidence of the primary end point was based on the random forest algorithm. The calibration recognition of the area under the curve, accuracy, sensitivity, and specificity were 0.984, 0.993, 0.996, and 0.991, respectively. The model achieved a similar performance in the external validation cohort. Conclusions: The DR based on short-term follow-up after transcatheter aortic valve replacement was an effective indicator to evaluate prognosis quality, and the model was helpful for predicting adverse events and treatment management.

Keywords

• N-terminal B-type natriuretic peptide precursor
• Transcatheter aortic valve replacement
• Prognosis
• Machine learning