Pulse wave velocity from 165,549 smartwatches rose with age, male sex and BMI, and those at ≥10 m/s had higher cardiovascular disease prevalence

The study reports the distribution of algorithm-derived smartwatch pulse wave velocity (SW-PWV) in a large real-world Chinese population: median SW-PWV was 7.79 m/s (IQR 7.46–8.40) among 165,549 users, and 4.53% had SW-PWV ≥10 m/s. Prior SW-PWV data came mainly from small validation studies; this work describes its distribution at a scale of over one hundred thousand users across 34 provinces and cities. Cross-sectional design with 165,549 participants; records underwent plausibility screening for age, height, weight and BMI plus individual-level mean ± 3 SD outlier removal, and the distribution changed only minimally after outlier screening (mean 8.03 vs 8.04 m/s).

SW-PWV showed a moderate-to-strong positive correlation with age (r=.6323) and positive correlations with male sex (r=.1856) and BMI (r=.1981); an OLS model with age, sex and BMI explained 60.9% of the variance in SW-PWV. Age-stratified analysis further showed the sex association was stronger in younger participants (r=.5033 under 30) while the age association became more pronounced after 50 (r=.7212 at 60–69), with BMI positively associated across all age groups. Spearman correlation, Kruskal–Wallis tests and OLS regression, all P<.001; the authors explicitly note that age, sex and BMI are incorporated into the SW-PWV estimation algorithm, so these patterns should be read as characteristics of the algorithm-derived output rather than independent biological associations.

SW-PWV was significantly elevated in participants with hypertension, diabetes, dyslipidemia, coronary heart disease, stroke and carotid plaque, with hypertension (r=.28) and carotid plaque showing the strongest correlations and two or more concurrent diseases yielding r=.32. Among 35,402 questionnaire completers the study systematically compared multiple cardiovascular risk factors and diseases, and age-stratified analyses showed the strength of associations varied by age (for example, the hypertension association was no longer significant at ≥70 years). Questionnaire completion was voluntary and conditions were self-reported physician diagnoses; associations are unadjusted cross-sectional correlations, and the authors state that blood pressure, antihypertensive treatment, smoking, physical activity and socioeconomic status were not collected.

When the guideline cfPWV threshold of 10 m/s was applied, SW-PWV ≥10 m/s was associated with a higher prevalence of cardiovascular risk factors and prevalent cardiovascular disease (54.43% of that group were free of CVD versus 84.76% below 10 m/s), discrimination for prevalent CVD gave an AUC of 0.71, sensitivity was 13.6% and specificity 96.7% at 10 m/s, and the maximum Youden index identified an exploratory threshold of 8.53 m/s (sensitivity 56.5%, specificity 76.9%). No prior studies had examined an SW-PWV threshold; this work is the first attempt to transfer the guideline cfPWV threshold to smartwatch-derived values and report its ROC performance. ROC analysis and chi-squared testing (χ2=1,184.7599, P<.0001) within the 35,402-person questionnaire subsample; the authors stress that 10 m/s is only an exploratory reference threshold requiring direct validation against conventional cfPWV.

The study addresses Chinese adults who owned and used compatible Huawei smartwatches and consented to research use of their data, with measurements self-performed under real-world conditions (seated for 10 minutes, watch on the left wrist, right index finger on the collector for 30 seconds). Its direct use is to establish a large-sample distributional reference for SW-PWV and to suggest that the cfPWV-derived 10 m/s threshold might serve as a preliminary or exploratory reference, supporting the idea of SW-PWV as an initial community-based screening step to identify individuals who may benefit from formal cardiovascular evaluation. For a reader, this means the results apply to description and hypothesis generation in wearable-user populations, not to individual diagnosis or risk prediction.

Worth watching: the design is cross-sectional, so causal inference and prospective risk prediction are out of scope; age, sex and BMI are incorporated into the SW-PWV estimation algorithm, so associations with these variables cannot be attributed independently to vascular biology; disease information came from voluntary self-reported questionnaires not verified against clinical records; blood pressure, antihypertensive treatment, smoking, physical activity and socioeconomic status were not collected, leaving associations and relative risks unadjusted; women made up only 11.97% of participants and stroke cases numbered only 480; and only the first valid measurement per user was analyzed, so intra-individual variability and reproducibility were not assessed. In addition, the loaded text is an incomplete version: the supplementary tables and figures (for example Supplementary Tables S1–S5 and Figures S1–S3) are not included, so some stratified details can only be taken from the main text rather than checked item by item. The authors' proposed next steps are repeated measurements to assess reproducibility, prospective longitudinal studies to test outcome associations, and validation in more diverse populations.