THE B-SCORE IS A NOVEL METRIC FOR MEASURING THE TRUE PERFORMANCE OF BLOOD PRESSURE ESTIMATION MODELS

Objective: Many novel blood pressure (BP) measurement technologies have been recently introduced into the market. Evaluating their measurement performance reliably and detecting important features of distinct technologies and devices is crucial for them to potentially reach clinical usability. To facilitate this, we developed a novel metric for the relative performance of blood pressure estimation systems (B-Score). Design and method: The B-Score sets absolute blood pressure estimation model performance in contrast to the dataset the model is tested upon. We calculate the B-Score based on inter- and intrapersonal variabilities within the dataset. To test the B-Score for reliable results and desired properties, we designed generic datasets with differing inter- and intrapersonal blood pressure variability. We then tested and validated the B-Score’s real-world functionality on large scale clinical datasets (MIMIC IV database). Results: The B-Score demonstrated reliable and desired properties. The real-world test provided allowed the direct comparison of different datasets and revealed insights hidden from absolute performance measures. The B-Score discriminates reliably between differing BP measurement performances and can detect similarities between measurement devices and their performances even when evaluated in different datasets. Conclusions: The B-Score demonstrated reliable and desired properties. The real-world test provided allowed the direct comparison of different datasets and revealed insights hidden from absolute performance measures. The B-Score discriminates reliably between differing BP measurement performances and can detect similarities between measurement devices and their performances even when evaluated in different datasets.