Development and Independent Validation of a Machine Learning–Based Non-Invasive Venous Waveform Analysis for Heart Failure (NIVAHF) Device to Estimate Pulmonary Capillary Wedge Pressure

Background Hemodynamic-guided management improves outcomes in heart failure (HF), but implantable pulmonary artery pressure monitoring is limited by cost and invasiveness. Non-Invasive Venous Waveform Analysis for Heart Failure (NIVA HF ) is a machine learning–based device estimating pulmonary capillary wedge pressure (PCWP) from peripheral venous waveforms. Objectives To describe the development and multicenter validation of the NIVA HF device by evaluating agreement between the NIVA Score and invasively measured PCWP. Methods In this prospective, multicenter study, peripheral venous waveforms were acquired non-invasively with the NIVA HF Device and analyzed by a supervised neural network using a 60:20:20 training, validation, and independent test allocation, with one site held out as the dominant contributor to the locked test cohort. Agreement between the NIVA Score and invasively measured PCWP was assessed by Bland–Altman analysis in the locked, independent, right heart catheterization (RHC) test cohort. Discrimination for elevated filling pressure (PCWP >15 mmHg) was evaluated by receiver operating characteristic (ROC) analysis, with PCWP–LVEDP and PAD–LVEDP agreement analyzed for context. Results In the independent RHC test cohort (n=122), NIVA Score showed a mean bias of −1.06 mmHg (SD 5.57) and 95% limits of agreement −11.98 to 9.87 mmHg versus PCWP, falling between the invasive surrogate comparisons (PCWP–LVEDP, −9.31 to 7.54; PAD–LVEDP, −14.85 to 8.63). The AUC for PCWP >15 mmHg was 0.71 (95% CI, 0.61–0.80; P<0.0001; 89% sensitivity, 36% specificity). Conclusions NIVA Score demonstrated agreement with PCWP comparable to accepted invasive surrogates and detected elevated filling pressures with high sensitivity, supporting NIVA HF as a non-invasive congestion marker in heart failure.