Is blood pressure monitoring using photoplethysmography on a smartwatch a reliable method?

Background Noninvasive blood pressure (BP) measurement has become more accessible with wearable photoplethysmography (PPG) devices. However, the reliability of these devices has been questioned in independent studies, such as one that found a high-end smartwatch tended to overestimate low BP values and underestimate high BP values [1]. Objective To assess the reliability of blood pressure monitoring using a budget-friendly smartwatch (BPSWm), both at rest and during physical exertion, based on 93 training sessions with a 78-year-old normotensive, healthy volunteer. Methods BP measurements were manually activated on both wrists, held stably in the same position at rest, during 30 minutes of steady effort at a constant load (using a Technogym bicycle ergometer), and after ten minutes of recovery. These measurements were taken simultaneously with a clinically-validated Omron wrist BP monitor (OBPm) and a BPSWm, analyzed and compared. BPSWm measured BP sequentially with (cBPSWm) and without (ucBPSWm) personal calibration. The variability of WBPm across sessions and between devices was calculated for systolic (SBP), diastolic (DBP), and pulse pressures (PP) using the coefficient of variation (CV) for both OBPm and BPSWm. Descriptive statistics are presented as means ± standard deviations. The Bland-Altman method was used to assess bias and precision, with differences calculated as OBPm minus ucBPSWm. Results A total of 744 BP measurements were analyzed. Table 1 summarizes the comparison between OBPm and ucBPSWm. The average coefficient of variation (CV) for SBP and DBP was 5% and 5.8%, respectively, at rest (relative error [RE] of 6.7% and 7.8%). During effort, the CV was 7.9% (SBP) and 19.6% (DBP), with corresponding RE of 12.1% and 33.5%. The average pulse pressure (PP) was 36.9 ± 4.9 mmHg for OBPm compared to 36.3 ± 2.9 mmHg for ucBPSWm at rest (p = n.s.), and 61.0 ± 10.4 mmHg for OBPm compared to 52.8 ± 6.1 mmHg for ucBPSWm during effort (p < 0.05). Examples of Bland-Altman plots at rest and during effort are shown in Figure 1. In contrast to OBPm and ucBPSWm, BP measurements from cBPSWm did not significantly differ between rest and effort, resulting in an underestimation of BP increase during effort. On the other hand, SBP and DBP increases during effort were systematically overestimated by ucBPSWm (p < 0.05). Conclusion Acceptable agreement between OBPm and BPSWm was observed only at rest. As in [1], a systematic bias introduced by personal calibration of BPSWm led to an underestimation of significant BP variations induced by effort, whereas ucBPSWm overestimated effort-induced increments in both SBP and DBP. Although this study confirms present insufficient reliability for clinical use of wearable BP monitoring based solely on photoplethysmography sensors [1], further progress is foreseen with novel smart wearable hardware technology and more advanced AI-assisted signal analysis [2-4].