We recruited 46 healthy adults (27 women and 19 men, age range 22-54 years) from Chongqing, China, based on the inclusion and exclusion criteria. All participants had lived at low altitudes (<500 m) for at least 10 years and had no recent history of high-altitude (>2500 m) exposure (in the last 6 months). Participants with any one of the following conditions were excluded: respiratory and cardiovascular diseases, malignant tumors, liver and kidney dysfunctions, and psychiatric disorders or neuroses that would not allow them to complete the questionnaires.

The study protocol (ChiCTR2200059900) complied with the Declaration of Helsinki and was approved by the ethics committee of Xinqiao Hospital of Army Medical University (approval: 2022-研第-060-01). Written informed consent was obtained from all the participants after the study details, procedures, benefits, and risks were explained.

This study consisted of 2 exercise tests at low and high altitudes (Figure 1). The participants were instructed to avoid heavy load training 7 days before the tests and during the recovery days and to abstain from caffeine and alcohol for 24 hours before testing. On the first day of the study, each participant underwent a routine blood test before the SWT. After a 24-hour break, CPET was performed. The 2 tests (SWT and CPET) were performed at a similar time of day (SD 30 minutes) and were completed in 2 days. After resting for 3 days, the participants ascended to a high altitude (3900 m, Shigatse, China) in 3 hours by plane from a low altitude (300 m, Chongqing, China). On the second day at high altitude, they took 1 day off to complete the 2018 Lake Louise score assessment. Unfortunately, 2 individuals had knee injury because of the trip; therefore, they did not undergo exercise tests. Besides, 1 participant had an ST segment depression in the electrocardiogram and 2 participants had chest pain; therefore, they could not make it to the end. Finally, the remaining 41 participants repeated the exercise tests completely.

The participants were required to avoid eating or drinking anything (fasting) apart from water for up to 12 hours. Approximately 5 mL of intravenous blood was collected from the inside of the elbow and mixed with 1 mL of dipotassium ethylenediaminetetraacetic acid anticoagulant by using a tight band (tourniquet). Blood samples were analyzed using a BC-3000 plus automated hematology corpuscle analyzer (Mindray). The details of the 19 different parameters are presented in Multimedia Appendix 1. Blood tests at low altitude were performed between 7 AM and 9 AM on the same day before the exercise tests. All biochemical parameters were measured in the blood samples at the Clinical Laboratory of Cardiology Science of Xinqiao Hospital, Army Medical University.

The CPET was performed on an electronically braked cycle ergometer (EC3000e, Customed) in an erect position with breath-by-breath measurements through a tightly fitted face mask of minute ventilation, O₂ uptake, and CO₂ output by using a cardiopulmonary exercise testing system (Metalyzer 3B, Cortex). Before performing CPET, the baseline physiological measures for all devices used in this study were measured for 5 minutes in a resting state and subsequently in a standing position. After the baseline measurement, the test was conducted immediately. The cycle ergometry test protocol included 3 minutes of free-wheel cycling and subsequently proceeded with a continual increase in resistance by 25 W/min (according to the prior known exercise capacity [21], so that the test would last 10-12 minutes) until test completion or exhaustion. VO₂max was defined as the highest 30-s average value within the last minute of exercise until the first 15 s of recovery at peak exercise [22]. Standard 12-lead electrocardiogram, blood pressure, and SpO₂ were obtained at rest, every minute during exercise, and for ≥4 minutes during the recovery phase throughout the procedure using a 12-lead connection (custo-Cardio 3000BT-A, Cortex) in real time, blood pressure cuffs (Suntech Tango M2, Cortex) in the upper arm, and a finger clip portable oximeter (Nonin wristOx2), respectively.

We provided participants with a smartwatch (Huawei Watch GT Runner) and instructed them to wear it correctly on the left wrist, which enables reliable and persistent measurement of running speed, distance, and heart rate. Therefore, these measurements could be monitored continuously and automatically during each running activity, stored on the participant’s mobile device (Huawei MatePad 11 DBY-W09), and regularly transmitted to a secure cloud server, which was later transferred to the Huawei Health Center software through Bluetooth. Specifically, VO₂max estimation steps were as follows: (1) the personal background information (age, height, and weight) of the participant was logged in and the exercise type (running outdoors) was selected; (2) the participant started to run with a smartwatch that measured the heart rate and speed on level ground; (3) the start and end points were in the same place, and the smartwatch was stopped by the researchers uniformly with a timely click; (4) the researchers subsequently saved the participants’ running data to an album on the pad, facilitating further statistical analysis; and (5) the smartwatch and mobile device were formatted to prepare for the next test.

The signal processing by Huawei Watch GT Runner is licensed by the Firstbeat Technology’s Fitness Test, which is based on intelligent detection for both data reliability and exercise pattern during successive recording [13]. Briefly, the moving average filter was applied to both heart rate and physical activity data. After filtering the data, only data points at which both heart rate and physical activity data increased were selected as a period of physical activity. This was conducted by differentiating the data and selecting where both differentiated data were positive. The situations where the data series were excluded are listed below: (1) significant heart rate decreases and exceptional striding pattern (identified as a situation of running on a very steep downhill or soft surface automatically), (2) significant heart rate increases while the velocity remained 0 (identified as stopping suddenly in the middle), and (3) a short duration of highly increasing intensity (identified as insufficient effort level). After exclusions, the selected data series were further segmented as different heart rate zones according to the effort levels. Of them, the reliable data segments that belong to a long series of successive heartbeat intervals (in generally 20 s-10 minutes and preferably 30 s-4 minutes) and with a small heart rate change level were recognized as sufficient effort and used to calculate the VO₂max. In these reliable segments, speed was measured based on acceleration measurements by using a satellite navigation system. VO₂max estimates were made for each reliable segment by using the following theoretical VO₂ equation: theoretical VO₂ (mL·kg^–1·min^–1) = 3.5 * speed (km/h). The obtained VO₂max for each data segment was weighed and subsequently utilized to make a linear equation for calculating the final VO₂max (the detailed rule of weighting is shown in patent US9237868B2).

The presence of AMS at high altitude was assessed using the Lake Louise consensus scoring system 2018 version [23]. According to the 4 main symptoms, namely, headache, gastrointestinal symptoms, fatigue/weakness, and dizziness/vertigo, the scores were 0, 1, 2, and 3 in the order of none, mild, moderate, and severe, respectively. A total score of ≥3 combined with headache can be diagnosed as AMS.

Categorical variables were described as numbers and percentages. Descriptive statistics were presented as mean (SD) for variables with skewed distribution and median (IQR) for variables with normal distribution. The Mann–Wilcoxon rank-sum, independent-sample t test (2-sided test), Pearson chi-square test, and Fisher exact tests were used to compare the continuous and categorical variables statistically. The correlation magnitude and coefficient of determination between VO₂max-CPET and VO₂max-SWT were assessed using Pearson correlation. The intraclass correlation coefficient and paired-sample t tests (2-sided test) were performed to determine the agreement between VO₂max-CPET and VO₂max-SWT at low and high altitudes, respectively. We calculated the mean absolute error and mean absolute percentage error (MAPE) to evaluate the accuracy of the estimation. Furthermore, we used a Bland–Altman plot to investigate the level of agreement with 95% limits of agreement [24].

The relationship between the variables and AMS was examined by binomial logistic regression analysis with univariate analyses. The relationship between VO₂max-CPET, VO₂max-SWT, RDW-CV, and AMS was further examined by multivariate analyses. In the preliminary screening, we considered the variable with P<.05 as a potential risk factor, and an adjusted binary logistic regression model subjected the variable to identify the independent risk factors for AMS after the adjustment. Receiver operating characteristic (ROC) curves were constructed, and the Youden index was calculated. The optimal cutoff of variables for diagnosing AMS was determined at the point where the Youden index was maximum on the ROC analysis. We also compared the ROC curves of VO₂max-CPET, VO₂max-SWT, and RDW-CV alone or in combination. Differences were considered statistically significant at P<.05. Statistical analyses were performed using the SPSS Statistics software (IBM Corp) for Windows (version 26) and MedCalc software for Windows.

A total of 46 participants were recruited for this study, of whom 20 (44%) participants developed AMS. The clinical characteristics of participants with AMS and without AMS are presented in Table 1. There were no differences in age, sex, BMI, baseline heart rate, SpO₂, and blood pressure between the 2 groups. RBC count; hemoglobin, hematocrit, and mean corpuscular hemoglobin levels; mean corpuscular hemoglobin concentration; and RDW-SD did not differ significantly between the participants in the 2 groups, whereas the AMS group had higher RDW-CV at low altitude than the non-AMS group (14.25 [IQR 12.75-21.03] vs 12.70 [IQR 12.25-13.53], respectively; P=.02). In addition, the AMS group had lower VO₂max both measured by CPET (27.80 [SD 4.55] vs 32.00 [SD 4.04], respectively; P=.004) and estimated by SWT (28.00 [SD 6.75] vs 32.00 [IQR 30.00-37.00], respectively; P=.001) than the non-AMS group.

Table 2 shows the VO₂max in the SWT and CPET at low and high altitudes. The values of VO₂max-SWT were significantly overestimated at both low (constant error=1.11 [SD 1.73] mL·kg^–1·min^–1; t₄₅=4.35; P<.001) and high (constant error=0.98 [SD 1.54] mL·kg^–1·min^–1; t₄₀=4.05; P<.001) altitudes. A 6% MAPE (mean absolute error=1.761 mL·kg^–1·min^–1) at low altitude and a 6.8% MAPE (mean absolute error=1.610 mL·kg^–1·min^–1) at high altitude were observed, indicating a low average deviation between the 2 methods (Table 2). Furthermore, a strong correlation was found between VO₂max-SWT and VO₂max-CPET values (low altitude: R²=0.889; P<.001; high altitude: R²=0.947; P<.001; Figure 2). The results of the intraclass correlation coefficient revealed that VO₂max-SWT had a good level of agreement with the directly measured VO₂max-CPET at low (0.942; P<.001) and high (0.973; P<.001) altitudes. Additionally, the Bland–Altman plots demonstrated a small bias of the VO₂max-SWT values compared to the VO₂max-CPET at low (bias=1.11 mL·kg^–1·min^–1, Figure 3A) and high (bias=1.00 mL·kg^–1·min^–1, Figure 3B) altitudes. VO₂max-SWT showed even a lower range of bias at high altitudes than at low altitudes (upper to lower limits of agreement: 6.0 mL·kg^–1·min^–1 vs 6.8 mL·kg^–1·min^–1, respectively).

There was a significant difference in the VO₂max values at low altitude between the participants with and without AMS. VO₂max-CPET in the AMS group was lower than that in the non-AMS group (27.80 [SD 4.55] vs 32.00 [SD 4.64], respectively; P=.004), and a similar result was revealed by SWT (28.00 [IQR 25.25-32.00] vs 2.00 [IQR 30.00-37.00], respectively; P=.001; Figure 4A). The distribution of the VO₂max values based on the AMS results is shown in Figure 4B. Approximately 90% (16/17) of the participants with VO₂max values <26 mL·kg^–1·min^–1 developed AMS compared with approximately 20% (3/15) of the participants who developed AMS with a VO₂max>35 mL·kg^-1·min^-1. Patients with AMS seemed to have a lower VO₂max, regardless of whether it was directly measured by CPET or estimated by SWT.

To further explore the association between VO₂max and AMS, a univariate analysis was performed. Table 3 shows that VO₂max-CPET (odds ratio [OR] 0.807, 95% CI 0.686-0.949; P=.01) and VO₂max-SWT (OR 0.765, 95% CI 0.635-0.922; P=.005) at low altitude and baseline RDW-CV (OR 1.177, 95% CI 0.999-1.386; P=.05) were potentially associated with AMS occurrence. Multivariate regression analysis identified VO₂max-CPET (OR 0.770, 95% CI 0.640-0.926; P=.006) and RDW-CV (OR 1.263, 95% CI 1.028-1.553; P=.03) as well as VO₂max-SWT (OR 0.720, 95% CI 0.578-0.898; P=.004) and RDW-CV (OR 1.273, 95% CI 1.027-1.577; P=.03) as independent factors associated with the development of AMS at high altitude.

As both VO₂max and RDW-CV were closely related to AMS, we constructed the combined predictive models for AMS. Table 4 and Figure 5 show the area under the curve (AUC) of VO₂max-CPET (AUC 0.743, 95% CI 0.597-0.889), VO₂max-SWT (AUC 0.785, 95% CI 0.646-0.923), and RDW-CV (AUC 0.708, 95% CI 0.547-0.868).

Either the AUC of the VO₂max-CPET or VO₂max-SWT was higher than that of RDW-CV (both P>.05, Table 4). For the VO₂max-SWT, a sensitivity of 65% and a specificity of 88.46% were observed at the optimal cutoff value of 29.5 mL·kg^–1·min^–1, with a higher positive predictive value of 81.25% and a negative predictive value of 76.67%. However, no significant difference was found in AUC when compared to the VO₂max-CPET (0.785 vs 0.743, respectively; P=.25). Although the independent indicators were effective and significant, this combined predictive model was more accurate when VO₂max and RDW-CV were combined. In other words, the combined model 2 enhanced the diagnostic power with modest AUC gains of 0.03-0.06, although not statistically different from model 1 (0.839 vs 0.804, respectively; P=.27) or VO₂max-SWT alone (0.839 vs 0.785, respectively; P=.28). The combined model 2 also improved the prediction of AMS by increasing sensitivity from 65% to 80% compared with VO₂max-SWT alone while attaining high specificity.

Our study comparatively evaluated the VO₂max (CPET vs SWT) of individuals at a low altitude and subsequently at a high altitude. We demonstrated that the smartwatch device was a feasible and accurate tool for assessing cardiorespiratory fitness at both altitudes. We also proposed a novel model based on smartwatch-derived VO₂max with good performance in predicting AMS. Our easy-to-use approach for estimating VO₂max can be more widely applied for screening individuals susceptible to AMS on a large scale.

Previous clinical trials [25-27] have shown that VO₂max ranged from 20 mL·kg^–1·min^–1 to 50 mL·kg^–1·min^–1 according to the sex, age, or ethnicity of participants. In our study, the value of VO₂max measured either in CPET (low altitude: 30.17 [SD 5.01] vs high altitude: 25.20 [SD 6.46]; P<.001) or SWT (low altitude: 31.28 [SD 5.17] vs high altitude: 26.17 [SD 6.71]; P<.001) was within a fair level range, reflecting the sedentary fitness of this cohort. The 10 pairs of participants’ VO₂max at low altitude were nearly the same. The high overlap rate of our data may be because the measurement output from CPET and SWT are integers, which reduce the numerical difference. In other words, we consider that when the difference between the 2 measurements is less than 1, this difference may not show up. Such a high overlap rate outcome has been reported in a previous study [12]. Besides, repeated measures may help reduce the repetitive rate.

Similar to that shown in a previous report, VO₂max was lower by approximately 16.5% at 3900 m evaluated by both CPET and SWT compared to that at low altitude in our trial, which can be attributed to the reduction of atmospheric PO₂ at high altitude [28]. Interestingly, the VO₂max-SWT was slightly higher than the VO₂max measured by the breath-to-breath method at both low and high altitudes. Therefore, we considered that the following issues may be associated with the differences: (1) the exercise mode in SWT was accelerative running, which requires a larger number of muscle groups, whereas the mode of CPET was a cycle ergometer, mostly relying on endurance, and (2) CPET required maximal performance; however, SWT only required submaximal exercise intensity. In addition, the overestimated VO₂max of the SWT was more evident at high altitude, as highlighted by the higher MAPE (6.80% vs 6%, respectively) and wider 95% limits of agreement criterion. Interestingly, an increase in altitude did not significantly affect the R² and intraclass correlation coefficient values of VO₂max assessed by the SWT, suggesting its high compatibility for hypobaric hypoxic conditions. To keep up with the exercise pattern used in the SWT, it is more rigorous to perform the CPET program on a treadmill than on a cycle ergometer to minimize the impact of the different muscular factors. However, there are several distinct differences between a treadmill and a bicycle ergometer (ie, space, safety, and costs)—all of which are conducive to a bicycle ergometer. It is only possible to change the slope and speed in the treadmill exercise. However, a cycle ergometer attains the linearity in workload increment well [29]. Thus, a cycle ergometer program should be preferred over a treadmill program at high altitudes. Hence, we finally opted to use a cycle ergometer for CPET. Nevertheless, the VO₂max measured by the 2 methods has good correlation and consistency.

VO₂max evaluation using CPET is inconvenient in practice. In the past few decades, several new methods for estimating VO₂max have been investigated through a submaximal exercise protocol, including the Queen college step test [30], 20-m shuttle run test [31], and PWC170 [32,33]. Although they are easy to perform, the accuracy of the indirect method in estimating VO₂max remains controversial. Thus, more variables such as basic parameters (age, sex, BMI) [34] and exercise indicators (maximal heart rate, speed, and covered distance) [35,36] were utilized in discrepant equations for more accuracy in subsequent studies. For instance, Marsh [37] found a 4-stage incremental running program estimating VO₂max well (standard error of estimate=3.98-4.08 mL·kg^–1·min^–1; r=0.642-0.646). However, this equation cannot be applied to the general population because correlation data were obtained for male athletes [37]. Instead of simply substituting variables into the equation, the smartwatch employed an algorithm called Firstbeat to evaluate VO₂max in daily life. One of the key features of this patented technology is monitoring the running speed along with the heart rate continuously during each workout and automatically excluding the data without a linear relationship. A white paper of Firstbeat claimed that based on a database of 2690 freely performed runs by 79 individuals, its accuracy was up to 95% (MAPE<5%) and the error was below 3.5 mL·kg^–1·min^–1 [14]. For perspective, it is superior to most other indirect submaximal tests (10%-15%) and approaches the direct laboratory test (approximately 5%). Thus, the FFT method can be commonly used to estimate VO₂max when high-intensity exercise is limited or laboratory equipment is unavailable. Düking et al [38] found a coefficient of variation of 4% between Garmin watch and the criterion measure over the VO₂ peak range from 38 mL·kg^–1·min^–1 to 61 mL∙ kg^–1·min^–1 through a small sample study. In this regard, the smartwatch with FFT is a portable device with satisfactory accuracy for estimating VO₂max in the submaximal exercise protocol. Importantly, they also reported that the MAPE between the smartwatch and criterion measure was 7.1% when analyzing VO₂ peak below 45 mL·kg^–1·min^–1 whose discrepancy was less inaccurate in our study. The MAPE in this study was 6% and 6.8% at low and high altitudes, respectively. In [18], the 23 participants were of Caucasian origin, while in our study, the 46 participants were Chinese healthy adults. Hence, different race, gender, age, height, body weight and physical activity level of individuals may well explain differences in VO₂max, both between the other reports [39] and this study.

VO₂max represents the maximum oxygen utilization capacity of an individual. In normoxic conditions, a higher VO₂max indicates greater exercise capacity and better cardiorespiratory fitness [40-42]. However, due to PO₂ decrease (approximately 63% of low altitude at 3700 m) at high altitude, arterial SpO₂ and the amount of oxygen that eventually reaches the tissue and organ are reduced [43]. Thus, aerobic metabolic capacity and VO₂max are significantly inhibited. To cope with hypoxia at high altitude, the body undergoes a series of physiological compensations in the cardiovascular and respiratory systems, such as increased heart rate [44,45], blood pressure [46], and respiratory rate [47]. Although these compensatory responses can compensate for oxygen insufficiency in a short period, long-term exposure may lead to irreversible changes such as pulmonary hypertension, chronic pulmonary disease, and heart failure [48,49]. VO₂max has been an effective predictive indicator of mortality and rehospitalization in patients with chronic cardiovascular and respiratory diseases [50-52]. However, it has rarely been reported that VO₂max contributes to the prognosis and rehabilitation of acute and chronic mountain illnesses [50]. In this study, we present the first evidence of VO₂max as a predictor of AMS in a clinical trial.

It can be concluded from our results that individuals with a higher VO₂max are unlikely to develop AMS. ROC analysis demonstrated that VO₂max-CPET and VO₂max-SWT showed a similar predictive value, particularly for VO₂max-SWT, with a specificity up to 88.46% for a cutoff value of 29.5 mL·kg^–1·min^–1. The high specificity ensures a low incidence of AMS when VO₂max is below the cutoff value of lowlanders unsuitable for acute high-altitude exposure. In addition, we found that RDW-CV was more closely related to AMS than other routine blood parameters. We believe that the low RDW-CV represents a uniform distribution of erythrocyte developmental states, indicating a more effective compensation response of RBC under acute hypoxia stress [53,54]. Although RDW-CV was an independent predictor for AMS and RDW-CV combined with VO₂max showed a higher AUC, it is more convenient for individuals to obtain information on the potential suffering probability by using a smartwatch in daily life. Moreover, models 1 and 2 in Table 4 showed no statistical difference in AMS prediction compared to the single VO₂max model. Therefore, SWT-based VO₂max estimation can conveniently identify AMS-susceptible individuals and help evaluate the cardiorespiratory function and working capacity, which can benefit high-altitude travelers and workers and reduce the consumption of medical resources at high altitude. Although we cannot extend our validity claims to the entire population due to the small sample size and insufficiently diverse population characteristics, the conclusion that people with lower oxygen intake are more likely to develop AMS is well-founded and has been recently reported [55].

The VO₂max estimated by SWT and measured by CPET has high consistency, indicating that smartwatches may replace the CPET system to obtain VO₂max accurately and objectively by monitoring the common physical activities with a portable, low-cost system. After each exercise, the smartwatch can measure and record the exercise information, integrate the calculation, and update the VO₂max value. In addition, VO₂max measured at low altitudes is highly correlated with the occurrence of AMS, with satisfactory prediction performance. Therefore, it is feasible to use a smartwatch to measure VO₂max at low altitudes to evaluate the possibility of AMS. In the future, this will benefit tourists, temporary workers, and other individuals who plan to travel at high altitudes and will help in identifying participants susceptible to AMS before high-altitude exposure.

To advance this field, several measures are required. First, there is an urgent need for validation standards for smartwatch devices to enable standardized research. Second, the open disclosure of commercial validation studies can enable better resource usage, as studies will not have to be repeated unnecessarily. Third, further development of smartwatch devices will allow new possibilities in the field of VO₂max monitoring. Finally, subsequent trials should continue to focus on validating these devices compared to conventional standards and broaden their use and demonstrate new possibilities for accurate VO₂max monitoring.

This study had some limitations. Notably, the Lake Louise consensus scoring system (2018 version) is subjective; therefore, we described each symptom as clearly as possible and provided necessary instructions before the participants completed the questionnaire to deal with the subjectivity. Second, running exercise instead of cycling should be implemented by CPET to minimize the inconsistencies in VO₂max with SWT, which was not applied here due to the great safety risk at high altitude. Further studies and repeated measures are required to develop and investigate the predictive models of the SWT method based on submaximal running programs in terms of validity and reliability [11,56]. The present predictive models of the SWT method cannot be extended to the entire population. In the long term, conducting long-term validation research within a large and representative population is the scope of future studies, as the smartwatch will be extensively used by people. In addition, this study has insufficient reliability, as repeated measures analysis was not performed. It is contradictory to perform repeated measurements because measuring on the same day is limited by physical strength. Repeated measurements on different days also affect the assessment of AMS.

Our findings demonstrate that VO₂max estimated by SWT and CPET have good accuracy and agreement at both low and high altitudes. Importantly, smartwatch-based VO₂max at low altitudes was a convenient and effective approach to predict AMS and to identify susceptible individuals following acute high-altitude exposure, particularly by combining the RDW-CV at low altitudes.