When posture matters: Investigating cuffless BP estimation errors across postural changes

This study investigates how body posture influences cuffless blood pressure (BP) estimation using consumer wearables and evaluates whether incorporating simple contextual information can reduce estimation bias across postural transitions. A controlled laboratory experiment was conducted with fifteen subjects across six postures (sitting, supine, lateral left, lateral right, repeated supine, and standing). Multi-sensor data, including electrocardiogram (ECG), photoplethysmography (PPG), and accelerometer signals, were collected using wearable devices, synchronized, and processed to extract timing and morphological pulse features. A baseline model without contextual input was compared to a context-aware model incorporating posture and demographic covariates, with reference BP obtained using an oscillometric arm cuff. Results demonstrate a clear posture-dependent bias in BP estimation, with systematic underestimation in supine positions and overestimation in lateral right and standing postures. Incorporating posture as a contextual feature reduces these biases and improves estimation consistency across postural conditions with mean absolute error decreasing from 4.27 to 4.10 mmHg for systolic BP and from 3.92 to 3.29 mmHg for diastolic BP. While error levels fall within commonly referenced standard accuracy ranges, the findings highlight that such metrics alone are insufficient under changing conditions. The results reveal that cuffless BP estimation is influenced not only by static posture but also by transitional physiological effects. Overall, this study demonstrates that simple context-aware modelling can improve the robustness of wearable BP estimation under controlled postural changes, providing insight into the role of postural context in future generation cuffless monitoring systems.