Participants were recruited using physical flyers and existing digital institutional listservs containing individuals from previous studies who had explicitly consented to be contacted for future research opportunities. All concurrent physiological testing sessions were conducted at the University of California, San Diego Exercise and Physical Activity Center across various specialized laboratory rooms optimized for resting and exercise protocols. Data collection occurred between November 2021 and February 2022. The target sample size of 28 participants was established as a convenience sample for a pilot evaluation primarily focused on at-home usability, acceptability, and baseline laboratory validation of metrics. This cohort size aligns with established methodological guidelines for initial clinical device pilot studies, which recommend samples of 20 to 25 individuals to adequately assess instrument reliability and system feasibility prior to large-scale epidemiological deployment [18].

The study protocol and documents were approved by the institutional review board at the University of California, San Diego (800873). All participants in this study provided written informed consent prior to participating. To safeguard participant privacy and confidentiality, all collected data were managed within REDCap (Vanderbilt University), a secure, password-protected, and limited-access data management platform. Identifying information and deidentified analytical data were stored in completely separate databases within the platform. Cross-database access was strictly restricted to the principal investigator and project manager, who remained blinded to the data until active collection was complete. Participants received US $65 for completing the in-laboratory task and measurement portion of the project, with additional compensation of up to US $130 available for completing subsequent at-home and interview components.

Upon completion of the informed consent process, participants were fitted with experimental and control devices as follows.

The Withings ScanWatch was placed in “ECG” mode, and the participant was given a countdown to activate the “ECG” function on the watch. The researcher then noted the exact time (hh:mm:ss) of measurement on the Cardea SOLO patch and the Biopac at the beginning of the Withings measurement period. This ensured that temporally identical QRS complexes were analyzed on each ECG produced. The Withings measurement was collected for 30 seconds.

A standardized operating procedure for identifying clear ECG signals and placing start and end lines for each interval was developed by an exercise physiologist with extensive experience working with ECG, in consultation with a licensed physician.

The assessment of the individual cardiac tracings was completed by a single researcher with a clinical background who used a standard operating procedure. ECGs collected from each device (Withings, Cardea SOLO, and Biopac) were analyzed using the software program “EP calipers” (EP Studios). Crucially, no automated interval measurements provided by the devices’ internal software or associated applications were used for analysis. The electronic calipers were calibrated prior to each assessment following manufacturer recommendations. Manual measurements of RR, QT, and QTc intervals were gathered on 3 consecutive heartbeats. The analyzed beats were chosen by the reviewer, who selected the clearest ECG tracing from the ScanWatch-based tracing for each condition described above.

RR intervals were measured by placing the start line of the caliper at the beginning of the Q wave in the first QRS complex and the closing line at the start of the Q wave of the fourth QRS complex. The program then averaged the RR interval over those 3 complexes. The QT interval was measured by placing the start position at the initiation of the first Q wave and the completion line at the end of the first T wave after it had returned to the baseline amplitude. Identical procedures and manual caliper measurements were followed for both the Biopac (3-lead) and Cardea SOLO (1-lead) devices using the same temporal ECG complexes identified by the Withings ScanWatch trigger. The program used the Bazett formula to calculate QTc once RR and QT were recorded. An example of the measurements taken in EP calipers is shown in Figure 2.

Statistical analysis was performed using SPSS (version 26; IBM Corp). To assess error in the calculation of the intervals of interest, the mean error (ME), mean absolute error (MAE), and mean absolute percent error (MAPE) were calculated as the difference between the Withings ScanWatch measurement and the reference devices. MAPE does not have a standardized threshold for determining the accuracy or validity of measurements. Fokkema et al [19] suggest a MAPE threshold of ≤5%, whereas Nelson et al [20] used a MAPE threshold of ≤10% to classify a wearable device as valid. Studies that used the 5% cutoff when evaluating activity monitors were tracking step count, for which a difference of 5% or greater in step count was considered practically significant [19]. Given the relative difficulty of precisely identifying the end of the T wave, 10% was considered more appropriate than the 5% used by Fokkema et al [19] when considering variables with a relatively easy-to-identify signal, such as steps.

To evaluate interdevice agreement while accounting for repeated measurements within participants across the 4 experimental conditions, repeated-measures Bland-Altman analyses were conducted using linear mixed-effects models (LMMs) with a random intercept for participant ID [21,22]. Fixed intercepts from the models were used to determine the ME and associated P values, replacing standard paired t tests (2-tailed) to account for within-participant clustering. The 95% limits of agreement (LoA) were calculated by taking the ME plus or minus 1.96 times the total SD, where total SD was derived from the square root of the combined between-participant and within-participant residual variances [21,22]. To assess intraindividual trajectory association across repeated physiological states, repeated-measures correlation (r_rm) analysis was performed. Correlation values were interpreted as follows: 0.10 to 0.39=weak; 0.40 to 0.69=moderate; 0.70 to 0.89=strong; 0.90 to 1.00=very strong.

General characteristics of the participants are presented in Table 1 as means and SDs.

Strong agreement was observed for RR intervals, supported by a strong r_rm coefficient (r_rm=0.84; P<.001). The LMM confirmed that the fixed intercept for RR interval differences was not statistically significant (P=.36), with Bland-Altman analysis showing a slight ME of 9.4 milliseconds (Figure 3). Conversely, agreement was significantly weaker for both QT and QTc intervals, with r_rm coefficients indicating weak correlations (r_rm=0.37; P<.05 and r_rm=0.27; P<.05), respectively. The LMM fixed intercepts demonstrated that these mean discrepancies were statistically significant (P<.001 for both QT and QTc), revealing a systematic underestimation of −42.9 milliseconds for QT (Figure 4) and −51.4 milliseconds for QTc (Figure 5) by the Withings smartwatch (Table 2).

Evaluation against the Cardea SOLO single-lead patch demonstrated a highly similar trend. The RR interval showed a strong correlative relationship (r_rm=0.80; P<.001). The LMM indicated that the fixed intercept for RR interval differences was not statistically significant (P=.61), demonstrating strong clinical agreement with a low ME of 6.1 milliseconds (Figure 6). Agreement was notably lower for repolarization interval durations. The r_rm coefficient indicated a weak correlation for both the QT (r_rm=0.37; P=.001) and QTc (r_rm=0.27; P=.03) intervals. The LMM fixed intercepts demonstrated that these mean discrepancies were statistically significant (P<.001 for both QT and QTc), revealing a systematic underestimation of −15.8 milliseconds for QT (Figure 7) and −18.8 milliseconds for QTc (Figure 8) by the Withings smartwatch (Table 3). Consistent with the Biopac findings, Bland-Altman calculations highlighted a clear systematic pattern of interval underestimation by the wrist-worn consumer device.

This evaluation assessed the clinical agreement of intervals captured by a wrist-worn consumer smartwatch with those measured using both a continuous 3-lead laboratory ECG reference and an FDA-approved single-lead ambulatory patch monitor. Our primary findings demonstrated that while the wrist-worn device provided highly accurate, tightly correlated tracking of beat-to-beat RR intervals, it displayed pronounced systematic discrepancies when evaluating intracardiac microintervals. Specifically, the smartwatch consistently and uniformly underestimated both QT and QTc durations across all experimental testing conditions, demonstrating clinically meaningful absolute measurement errors. Crucially, the relatively wide Bland-Altman LoAs derived from our repeated-measures LMM underscored a substantial degree of individual-level variability, which further confirmed that the device could not be considered clinically interchangeable with standard diagnostic monitors at this time.

The strong correlative performance observed for the RR interval matches historical validation work verifying consumer smartwatches for ambient heart rate tracking and baseline rhythm surveillance. However, our data revealed that macrorhythm precision did not automatically equal accurate repolarization interval assessment. While previous studies exploring alternate consumer wearables reported tighter absolute interval alignment with clinical 12-lead ECG standards, the systematic underestimation found here limited this specific device’s immediate clinical interchangeability for acute arrhythmia screening.

This capability aligns well with the intended use of these devices for personal fitness tracking and general wellness monitoring [17]. Given the strong agreement at the individual level, our data suggested that smartwatches can be used to reliably track heart rate across multiple exercise intensity thresholds [23-25]. However, tracking capability has known boundaries; prior work indicates that consumer-grade device differences increase substantially at elevated heart rates (eg, heart rates ≥150 beats per min) [24], though participant heart rates in the current evaluation stayed within a window (≤130 beats per min) where this specific limitation was minimized.

Conversely, making informed clinical decisions regarding the management of conditions such as long QT syndrome requires a high level of both accuracy and precision because the normal range for repolarization intervals is exceptionally narrow [26,27]. Identifying true interval prolongation is crucial for guiding clinical management and pharmacotherapeutic choices, as it prevents the accidental administration of medications that could further extend the QT interval and precipitate life-threatening ventricular arrhythmias, including torsade de pointes [4,28].

This systematic error is fundamentally explained by distinct device architecture and lead geography constraints. Standard clinical practice relies on multilead arrays to actively identify the single vector capturing the maximum absolute QT duration [26,27], a diagnostic requirement that a single wrist-to-hand lead pathway cannot anatomically fulfill. Furthermore, while this evaluation focuses strictly on ventricular repolarization intervals (QT/QTc), consumer wearables face inherent diagnostic limitations regarding broader conduction disorders. For instance, detecting atrioventricular blocks relies entirely on the precise identification of the P wave and the subsequent PR interval. Because single-lead wrist-worn devices frequently suffer from baseline wandering and lower-amplitude signal-to-noise ratios, isolating distinct P-wave morphologies can be challenging, meaning the feasibility of QT tracking cannot be extrapolated to infer reliable atrioventricular block screening capabilities.

Certain structural limitations also extend to ambient signal processing. Consumer wearable electrode recordings are highly susceptible to early artifact noise and baseline wandering caused by initial skin-contact motion errors [29]. Selecting the single “clearest” tracing segment within a postexercise window can preferentially isolate periods of rapid heart rate deceleration during recovery. Because the Bazett formula is exceptionally sensitive to rapid heart rate fluctuations, this brief temporal mismatch relative to continuous tracking introduces substantial artifacts into calculated QTc metrics. Because the directional error is highly predictable and systematic, future software calibrations or corrective predictive algorithms could potentially offset this bias, transforming the device from an acute diagnostic screen into a highly valuable tool for passive, long-term intraindividual trend tracking.

Several critical limitations qualify these findings. First, the evaluation sample size (N=28) was limited and established as a convenience sample for a pilot study. While this cohort size provided sufficient statistical power to detect robust, high-amplitude physiological trends, such as the strong beat-to-beat agreement observed across RR intervals, it restricted our overall statistical power when assessing highly variable microintervals such as the QT and QTc segments. Consequently, our findings may have limited generalizability to broader clinical populations. Furthermore, because the study comprised healthy, younger adults free from underlying cardiac disease states, the device was not evaluated in a true pathological cohort exhibiting baseline repolarization abnormalities or manifest conduction defects. Therefore, its performance in true high-risk clinical populations remains unverified. Second, data across the distinct physiological testing environments were aggregated for the primary repeated-measures statistical analyses. While using LMMs allowed us to account for within-participant clustering, pooling data across conditions may have inadvertently masked subtle accuracy variations caused by active postural changes or ambient motion artifact. Crucially, because all postexercise smartwatch triggers were executed while participants were seated quietly at rest to ensure tracing viability, this protocol did not capture the severe motion and muscle-activation artifacts associated with active, ambulatory movement. Finally, all manual interval calipers were placed by a single independent investigator. Although standardized operating procedures were strictly enforced, microinterval end point determination remains inherently operator dependent.

While the evaluated smartwatch serves as a convenient and precise method for monitoring heart rate regularity and tracking long-term trends, its systematic underestimation of QT and QTc intervals makes it currently unreliable as a standalone clinical tool for acute arrhythmia risk stratification. In a broader public health context, introducing uncalibrated consumer-facing interval metrics can trigger significant unintended health care use consequences. Minor, benign interval variations reported as abnormal by a device can generate unwarranted patient anxiety, increase psychological stress, and drive an influx of false-positive emergency or specialist consultations, imposing a substantial cost and burden on health care systems without definitively improving clinical outcomes [30]. Future iterations of consumer ECG technologies must prioritize lead-specific algorithmic calibration to safely unlock the vast potential of accessible, population-level cardiac tracking.

In summary, the Withings ScanWatch offers a convenient and reasonably accurate method for tracking heart rate and detecting basic rhythm abnormalities. However, its current limitations in accurately measuring QT and QTc intervals render it unreliable for screening for many of the most common and deadly arrhythmias. As wearable technology continues to advance, future iterations of these devices may address these limitations, paving the way for more comprehensive and clinically valuable cardiac monitoring tools.