We conducted a prospective observational study that integrated a longitudinal component with parallel cross-sectional cohorts. Data were collected at four sites spanning a broad altitudinal gradient in Western China: Chengdu (CD, 500 m), Kangding (KD, 2,560 m), Xinduqiao (XDQ, 3,460 m), and Litang (LT, 4,014 m). This study was prospectively registered in the Chinese Clinical Trial Registry (registration number: ChiCTR2100048780) and was approved by the Institutional Review Board of West China Hospital. All procedures were performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to any study procedures. An overview of the study workflow and sampling scheme is shown in Figure 1.

Participants were enrolled into two prespecified cohorts. The Migrant cohort comprised healthy lowlanders permanently residing at the 500 m site who underwent baseline assessment in Chengdu and subsequent repeated measurements at KD, XDQ, and LT to model acute high-altitude exposure. The Resident cohort comprised three independent cross-sectional samples of healthy long-term residents recruited locally at KD, XDQ, and LT to represent chronic adaptation to altitude. Eligible participants were 18–60 years old and willing to provide informed consent. Individuals were excluded if they reported cardiovascular, respiratory, or hematologic disease, current pregnancy, or regular use of medications known to affect cardiovascular function. Demographic data included age, sex, height, weight, and self-reported ethnicity (Han, Zang, or Yi).

All measurements were obtained in a quiet, temperature-controlled room (20 °C–25 °C) after at least 10 min of seated rest. Each session included paired assessments using a commercially available smartwatch and reference medical devices performed in close temporal succession. The smartwatch (HUAWEI Watch GT3 Pro) employs photoplethysmography to estimate peripheral SpO₂, systolic blood pressure (SBP), and diastolic blood pressure (DBP). Importantly, the GT3 Pro model used in this study was equipped with the TruSeen™ 5.0+ optical heart rate monitoring system with enhanced PPG sensors. For blood pressure estimation, a research-grade algorithm provided by Huawei for investigational purposes was enabled through a research collaboration; this algorithm is distinct from the consumer-facing features available on commercially marketed HUAWEI Watch GT3 Pro devices and should not be conflated with standard retail functionality. The device was worn snugly on the non-dominant wrist following the manufacturer’s guidance. Measurements were performed under controlled ambient lighting conditions with overhead fluorescent illumination and no direct sunlight. Watch band tightness was adjusted to allow one finger to fit snugly between the band and the wrist, ensuring secure contact without excessive compression of underlying tissues. Reference SpO₂ was measured with a medical-grade finger pulse oximeter placed on the non-dominant index finger. Reference SBP and DBP were obtained by a trained operator using a mercury sphygmomanometer and stethoscope (auscultatory method) on the non-dominant arm. Cuff size was selected according to the participant’s mid-arm circumference following international guidelines, with the bladder encircling at least 80% of the arm (See details in Supplementary Figure S1). Two consecutive readings were obtained at least 1 minute apart; if the difference between them exceeded 5 mmHg for either SBP or DBP, another two paired measurements would be taken until the difference was within 5 mmHg. The time interval between corresponding smartwatch and reference device measurements was kept within 2 minutes to minimize the influence of short-term physiological fluctuations. Participants were seated quietly with feet flat on the floor and the arm supported at heart level throughout all measurements.

Participants permanently residing at 500 m were classified as Migrants, whereas those with long-term residence at 2,560 m, 3,460 m, or 4,014 m were classified as Residents. For each parameter, measurement error was defined as the smartwatch value minus the corresponding reference value, such that negative values indicate underestimation by the smartwatch.

Prior to analysis, a data quality check was performed in accordance with the operational requirements of ANSI/AAMI/ISO 81060-2:2018 (ISO 81060-2: 2018 S nOf, 2018). For each participant at each altitude, we examined the consistency of the two reference blood pressure readings. Sessions with >5 mmHg difference between duplicate reference BP readings were flagged for exclusion. No sessions met this criterion (0/181 flagged), indicating high reference measurement consistency.

All analyses were performed in Python (version 3.10). Two-sided P-values <0.05 were considered statistically significant. Baseline characteristics were summarized as mean ± standard deviation (SD) for continuous variables and as counts and percentages for categorical variables, with group differences examined using one-way analysis of variance or the chi-square test, as appropriate.

Agreement between smartwatch and reference measurements was evaluated with Bland–Altman methods, reporting the mean difference (bias) and the 95% limits of agreement, calculated as bias ±1.96 times the SD of the paired differences. To quantify absolute error magnitude, we also computed the mean absolute error (MAE) and the root mean square error (RMSE) within each subgroup.

Confounder Assessment. Given the imbalanced sex distribution observed at 3,460 m (23.3% male), we assessed whether sex was a potential confounder for measurement error. At this altitude, independent-samples t-tests were used to compare the mean error (Watch − Standard) between male and female participants for each physiological parameter. For DBP, a statistically significant association between sex and measurement error was identified (P = 0.029). Consequently, all subsequent analyses involving DBP error employed sex-adjusted residuals derived from a linear regression model (DBP Error ∼ Sex), ensuring that the primary comparisons (e.g., Migrant vs. Resident, Han vs. Zang) were not confounded by the unequal sex distribution. For SpO₂ and SBP, no significant sex effect was detected (P = 0.73 and P = 0.17, respectively), and unadjusted errors were used.

Benchmarking Against International Standards. We benchmarked device performance against relevant international standards. For blood pressure, we assessed compliance with the ANSI/AAMI/ISO 81060-2:2018 Universal Standard (including Amd 1:2020) (EN, 2011) using its two validation criteria. Criterion 1 requires that, across all paired determinations, the absolute mean difference does not exceed 5 mmHg and the SD does not exceed 8 mmHg. Criterion 2 requires that the SD of the per-subject mean differences remains below a bias-dependent limit, which is approximately 6.9 mmHg when the overall bias is near zero. Both criteria were calculated within each altitude stratum and for the combined dataset to contextualize accuracy across conditions.

For oxygen saturation, we compared RMSE values with the accuracy requirement specified in ANSI/AAMI/ISO 80601-2-61:2017(33) for pulse oximeters, which summarizes accuracy using the root-mean-square error (A_rms) across the SaO₂ range of 70%–100%, with a pass threshold of A_rms ≤3%. Because arterial co-oximetry was not performed in this field study, RMSE served as a pragmatic proxy for A_rms relative to a medical-grade finger oximeter; compliance with the standard is therefore interpreted as benchmarking rather than formal certification.

Hypothesis Testing and Multiple Comparison Correction. Hypothesis testing followed the study design. Within the longitudinal Migrant cohort, we assessed whether the distribution of measurement error changed across the four altitudes using the Friedman test for repeated measures.

At each high-altitude site, mean errors were compared between Migrants and Residents using Welch’s t-test, which does not assume equal variances and is robust to unequal sample sizes. Within the Resident cohort, ethnic differences (Han vs. Zang) were explored at each altitude. Given the small sample size of Han participants at the highest altitude (n = 7 at 4,014 m), the Mann-Whitney U test was additionally employed as a sensitivity analysis to verify the robustness of the parametric findings.

To quantify the magnitude of group differences independent of sample size, Cohen’s d was calculated for each comparison. Effect sizes were interpreted as small (|d| < 0.5), medium (0.5 ≤ |d| < 0.8), or large (|d| ≥ 0.8).

Because multiple comparisons were performed across three parameters (SpO₂, SBP, DBP) and three high-altitude sites, all P-values from pairwise group comparisons were adjusted for multiple testing using the Benjamini-Hochberg procedure to control the false discovery rate (FDR) at 5%. Adjusted P-values (P_adj) are reported alongside raw values where relevant.

Across the four altitudes, 181 measurement sessions were obtained. Following data quality screening per ISO 81060-2:2018 requirements (reference BP measurement consistency check), all 181 sessions were retained for analysis (0 exclusions). The Migrant cohort included 24 healthy lowlanders (14/24 male, 58.3%; mean age 31.50 ± 7.53 years) assessed at all sites. The Resident cohort comprised three independent samples totaling 85 participants: 30 at 2,560 m (16/30 male, 53.3%; mean age 45.87 ± 10.86 years), 30 at 3,460 m (7/30 male, 23.3%; mean age 33.77 ± 7.78 years), and 25 at 4,014 m (10/25 male, 40.0%; mean age 37.04 ± 10.06 years). The ethnic composition of Residents shifted with altitude, with a greater proportion of Zang participants at higher elevations. Detailed demographics are provided in Table 1.

As altitude increased, standard measurements showed a progressive decline in SpO₂ and a general increase in blood pressure, particularly in the Migrant group (Table 2; Figures 2A–C). The wearable device successfully tracked these physiological trends for SpO₂, SBP, and DBP in both Migrant and Resident groups (Figures 2A–C). At all high-altitude sites, the Resident group maintained a higher mean SpO₂ (Figure 2A) and a lower mean SBP (Figure 2B) and DBP (Figure 2C) compared to the Migrant group (Table 2).

Bland–Altman analysis demonstrated that agreement between the smartwatch and reference devices varied by parameter and altitude (Figure 3). For SpO₂ (Figures 3A–D), agreement at 500 m was high, with a narrow 95% LoA (−0.55%–0.99%) and low MAE (0.40%) (Table 3). At higher altitudes, LoA widened and both MAE and RMSE increased, particularly in Migrants at 4,014 m (MAE 0.75%, RMSE 0.81%; Figure 3D). Notably, across all sites and groups the SpO₂ RMSE ranged from 0.19% to 0.81%, which is within the ANSI/AAMI/ISO 80601-2-61:2017 accuracy requirement for pulse oximeters (A_rms ≤3%).

For SBP (Figures 3E–H), the measurements demonstrated wider limits of agreement across conditions. Among Migrants, LoA ranged from −16.40 to 13.64 mmHg at 500 m (Figure 3E) to −18.41 to 10.65 mmHg at 4,014 m (Figure 3H). The smartwatch showed a persistent negative bias (underestimation), most pronounced in Migrants at 2,560 m (bias −5.29 mmHg, SD 9.88 mmHg; Figure 3F), which exceeded both the mean-bias threshold (|bias| ≤ 5 mmHg) and the SD threshold (≤8 mmHg) specified by ANSI/AAMI/ISO 81060-2:2018 Criterion 1. Additionally, the Combined group at 2,560 m exhibited an SD of 8.06 mmHg, marginally exceeding the 8 mmHg limit (Table 3).

For DBP (Figures 3I–L), the smartwatch also exhibited a consistent negative bias across groups and altitudes. LoA were narrower than for SBP but wider than for SpO₂. The largest bias was observed in Migrants at 4,014 m (bias −5.28 mmHg, exceeding the ±5 mmHg threshold; Figure 3L), with an SD of 7.61 mmHg that approached but did not exceed the 8 mmHg limit (Table 3).

In summary, when benchmarked against ANSI/AAMI/ISO 81060-2:2018, the smartwatch met the mean-bias criterion (|bias| ≤ 5 mmHg) in most strata but showed three specific instances of non-compliance (Table 3, bolded values): (Charlton et al., 2022) SBP in Migrants at 2,560 m exceeded both bias and SD thresholds (Bayoumy et al., 2021); Combined SBP at 2,560 m exceeded the SD threshold; and (Sastimoglu et al., 2025) DBP in Migrants at 4,014 m exceeded the bias threshold. These findings indicate that measurement consistency, particularly for SBP under acute high-altitude exposure, did not fully align with the standard’s requirements.

A longitudinal analysis of the Migrant group showed no statistically significant change in the distribution of measurement error for SpO₂ (Friedman test, P = 0.199; Figure 4A) or SBP (P = 0.130; Figure 4B) across the four altitudes. However, a trend towards a more negative bias in DBP error was observed with increasing altitude, although this did not reach statistical significance (P = 0.080; Figure 4C).

When comparing the Migrant group to the Resident group at each high altitude (Figure 5), no statistically significant differences were found in the mean measurement error for SpO₂ or SBP at any altitude after FDR correction (all P_adj >0.05; Figures 5A–F). For DBP (sex-adjusted), a trend towards a more negative error in the Migrant group was observed at 4,014 m (Figure 5I), but this was not statistically significant after correction (raw P = 0.086, P_adj = 0.586).

Notably, while mean errors did not differ significantly between groups, the variability of errors showed a marked disparity. At 4,014 m, the variance ratio (Migrant/Resident) for SpO₂ error was 17.5, indicating substantially greater measurement variability in the acutely exposed group despite similar mean bias (Figure 5C). This is further reflected in the wider interquartile range observed in the Migrant group. Effect size analysis revealed a medium effect for DBP (Cohen’s d = −0.51), suggesting a potentially meaningful difference that may have been underpowered to detect given the sample sizes. For SpO₂ and SBP, effect sizes were small (Cohen’s d = −0.33 and −0.27, respectively).

Within the chronically adapted Resident group, there were no statistically significant differences in measurement error for SpO₂ or DBP (sex-adjusted) between participants of Han and Zang ethnicity at any of the high-altitude sites after FDR correction (all P_adj >0.05; Figure 6). Sample sizes for Han participants were limited at higher altitudes (n = 15 at 2,560 m, n = 11 at 3,460 m, n = 7 at 4,014 m; Figure 6, inset annotations), which may have reduced statistical power to detect small differences. Sensitivity analyses using the Mann-Whitney U test yielded consistent results with parametric tests.

For SpO₂ (Figures 6A–C), mean errors were comparable between Han and Zang participants across all altitudes (P_adj = 0.81, 0.68, and 0.62 at 2,560 m, 3,460 m, and 4,014 m, respectively). For SBP (Figures 6D–F), a trend towards a difference in mean error was observed at 4,014 m, with the device tending to underestimate in Han participants and show less bias in Zang participants, but this did not reach statistical significance after correction (raw P = 0.056, P_adj = 0.525; Figure 6F). For DBP (Figures 6G–I), errors were similar between ethnic groups at all altitudes (all P_adj >0.67).