Smartwatch-Based Blood Pressure Measurement Demonstrates Insufficient Accuracy

Abstract

Background: Novel smartwatch-based cuffless blood pressure (BP) measuring devices are coming to market and receive FDA and CE labels. These devices are often insufficiently validated for clinical use. This study aims to investigate a recently CE-cleared smartwatch using cuffless BP measurement in a population with normotensive and hypertensive individuals scheduled for 24-h BP measurement.

Methods: Patients that were scheduled for 24-h ambulatory blood pressure monitoring (ABPM) were recruited and received an additional Samsung Galaxy Watch Active 2 smartwatch for simultaneous BP measurement on their opposite arm. After calibration, patients were asked to measure as much as possible in a 24-h period. Manual activation of the smartwatch is necessary to measure the BP. Accuracy was calculated using sensitivity, specificity, positive and negative predictive values and ROC curves. Bland-Altman method and Taffé methods were used for bias and precision assessment. BP variability was calculated using average real variability, standard deviation and coefficient of variation.

Results: Forty patients were included. Bland-Altman and Taffé methods demonstrated a proportional bias, in which low systolic BPs are overestimated, and high BPs are underestimated. Diastolic BPs were all overestimated, with increasing bias toward lower BPs. Sensitivity and specificity for detecting systolic and/or diastolic hypertension were 83 and 41%, respectively. ROC curves demonstrate an area under the curve (AUC) of 0.78 for systolic hypertension and of 0.93 for diastolic hypertension. BP variability was systematically higher in the ABPM measurements compared to the smartwatch measurements.

Conclusion: This study demonstrates that the BP measurements by the Samsung Galaxy Watch Active 2 show a systematic bias toward a calibration point, overestimating low BPs and underestimating high BPs, when investigated in both normotensive and hypertensive patients. Standards for traditional non-invasive sphygmomanometers are not met, but these standards are not fully applicable to cuffless devices, emphasizing the urgent need for new standards for cuffless devices. The smartwatch-based BP measurement is not yet ready for clinical usage. Future studies are needed to further validate wearable devices, and also to demonstrate new possibilities of non-invasive, high-frequency BP monitoring.

Keywords: Taffé method; ambulatory blood pressure monitoring; blood pressure; blood pressure variability; cardiovascular disease; digital health; hypertension; smartwatch.

FIGURE 1

Bias and precision plots according to Taffé method. Figures were constructed according to the Taffé method. (A) Bias plots for systolic and diastolic blood pressure (BP). The x-axis indicates the best linear unbiased prediction (BLUP) or true value of BP. Smartwatch measurements (blue dashed lines) are compared to 24-h BP measurements (solid black line); both values are read from the leftmost y-axis. The bias (difference of smartwatch measurement compared to the 24-h BP measurement) is depicted as a solid red line; the value of bias is read from the rightmost y-axis. (B) Precision plots for systolic and diastolic blood pressure (BP). The x-axis indicates the best linear unbiased prediction (BLUP) or true value of BP. The standard deviation of measurement errors for both measurements are depicted with the 24-h BP measurement depicted in black and the smartwatch method depicted in blue.

FIGURE 2

Scatterplots and Bland–Altman plots. Scatterplots (A) and Bland–Altman plots (B) for systolic and diastolic blood pressure measurements. In the scatterplots, cut-offs of 135 and 85 mm Hg are depicted as dashed lines. 24-h ABPM: 24-h ambulatory blood pressure monitoring. SD, standard deviation.

FIGURE 3

ROC curves (A) and PPV/NPV optimization curves (B). ROC curves and PPV and NPV optimization curves in predicting systolic (left) or diastolic (right) hypertension. AUC, area under the curve. BP, blood pressure. PPV, positive predictive value. NPV, negative predictive value. ROC, receiver operating characteristic.

FIGURE 4

Scatterplots for blood pressure variability per patient based on daytime measurements. ARV, average real variability. ABPM, ambulatory blood pressure monitoring.

FIGURE 5

Scatterplot for pulse pressures measured by ABPM vs. smartwatch measurements per patient. ABPM, ambulatory blood pressure monitoring.