Evaluation of blood pressure estimation models based on pulse arrival time

Abstract

Several models in the literature correlate blood pressure (BP) with the electrocardiogram (ECG) and photoplethysmogram (PPG) signals. These studies show substantial differences and have not been adequately compared. The MIMIC database, containing intrusive BP measures collected in a hospital setting, has been used to perform an extensive study of different models and variables extracted from the PPG and ECG signals. The best BP estimator model obtained is α/PAT2 + β HR + δ, where PAT is the Pulse Arrival Time and HR is the heart rate. This model allows the monitoring of fast BP trends and fluctuations. If absolute values of BP were required, the model would have to be calibrated with real BP measures. The model, with periodic recalibrations, meets the Association for the Advancement of Medical Instrumentation (AAMI) requirements for diastolic BP but not for systolic BP, for which the mean error is close to 8 mmHg.