Machine learning–based analysis of non-invasive measurements for predicting intracardiac pressures

Early detection of congestion has demonstrated to improve outcomes in heart failure (HF) patients. However, there is limited access to invasively haemodynamic parameters to guide treatment. This study aims to develop a model to estimate the invasively measured pulmonary capillary wedge pressure (PCWP) using non-invasive measurements with both traditional statistics and machine learning (ML) techniques.

The study involved patients undergoing right-sided heart catheterization at Erasmus MC, Rotterdam, from 2017 to 2022. Invasively measured PCWP served as outcomes. Model features included non-invasive measurements of arterial blood pressure, saturation, heart rate (variability), weight, and temperature. Various traditional and ML techniques were used, and performance was assessed using R² and area under the curve (AUC) for regression and classification models, respectively. A total of 853 procedures were included, of which 31% had HF as primary diagnosis and 49% had a PCWP of 12 mmHg or higher. The mean age of the cohort was 59 ± 14 years, and 52% were male. The heart rate variability had the highest correlation with the PCWP with a correlation of 0.16. All the regression models resulted in low R² values of up to 0.04, and the classification models resulted in AUC values of up to 0.59.

In this study, non-invasive methods, both traditional and ML-based, showed limited correlation to PCWP. This highlights the weak correlation between traditional HF monitoring and haemodynamic parameters, also emphasizing the limitations of single non-invasive measurements. Future research should explore trend analysis and additional features to improve non-invasive haemodynamic monitoring, as there is a clear demand for further advancements in this field.