Equitable and Interpretable Cuffless Blood Pressure Monitoring Using Distributed Wearable Spectral Sensors in a Multi-Cohort Trial

Ambiguous signal sources and algorithms currently impede wearable cuffless blood pressure (BP) monitoring and limit the efficacy of cardiovascular screening technologies. This results in inadequate validation, data distortion, physiological variability and restricted applicability across diverse populations, consequently biasing BP assessment outcomes. To mitigate these biases and facilitate equitable evaluation in cuffless BP inference, we present a smartwatch-based 16-channel multispectral multidirectional photoplethysmography (MSMD-PPG) array with synchronized ECG sensing and benchmarking platform in-depth studies of longitudinal and transverse cohorts. This system reduces depth and direction variability, capturing richer vasomotor information. Furthermore, we developed a hierarchical and interpretable feature selection framework that reduced redundant features by over 80% while preserving accuracy. In a multicentre clinical study of 5,190 participants, a calibration-based AdaBoost model achieved MAEs of 5.42 mmHg (systolic) and 4.15 mmHg (diastolic), meeting AAMI and IEEE 1708 criteria and delivering a 26.56% relative improvement in SBP MAE over a representative state-of-the-art baseline under consistent evaluation settings. Finally, the system demonstrated robust hypertension risk stratification across demographic subgroups, supporting scalable screening.