Comparison of Consumer Smartwatch and Research‐Grade Accelerometer‐Derived Step Counts in Amyotrophic Lateral Sclerosis Authors: Gordon Jewett, Reed Ferber, Colin B Josephson, Lawrence Korngut, Joon Lee Published: 2026 ABSTRACT ABSTRACTIntroduction/AimsObjective, scalable measures of function are needed in amyotrophic lateral sclerosis (ALS). Research‐grade accelerometers are promising but may be difficult to deploy for prolonged remote monitoring, whereas consumer smartwatches offer a practical alternative. We evaluated the feasibility of long‐term smartwatch monitoring in ALS and compared smartwatch‐ and accelerometer‐derived daily step counts.MethodsIn this single‐centre prospective observational study, participants wore a Fitbit Sense (smartwatch) on the wrist longitudinally and an ActiGraph GT9X Link (accelerometer) on the opposite wrist during clinic‐aligned assessment periods. Device comparison was restricted to concurrent valid days (≥ 10 h wear) and included paired comparison of mean daily step counts, Pearson correlation, Bland–Altman analysis, and proportional bias assessment.ResultsForty participants were enrolled and 39 contributed data across 8093 smartwatch and 915 accelerometer days. Step count comparison included 503 concurrent valid days from 34 participants. Mean daily step counts were 3480 (SD 4365) for smartwatch and 4066 (SD 4797) for accelerometer (p < 0.0001), with strong correlation (r = 0.91, p < 0.0001). Mean bias was −586 steps for smartwatch, with 95% limits of agreement from −4193 to 3021. Among participants with ≥ 6 months of concurrent data, devices showed similar decline over time (interaction p = 0.76).DiscussionLong‐term remote monitoring using a smartwatch was feasible in ALS. Smartwatch‐derived step counts were strongly correlated with accelerometer‐derived estimates but were systematically lower and showed wide day‐level limits of agreement. Smartwatch‐derived step counts may be useful for group‐level analyses and longitudinal monitoring, though device‐specific bias should be considered in interpretation.Keywords: actigraphy, amyotrophic lateral sclerosis, smartwatch, step count, wearable devices Abbreviations amyotrophic lateral sclerosis amyotrophic lateral sclerosis functional rating scale–revised application programming interface confidence interval forced vital capacity interquartile range research ethics board research electronic data capture standard deviation Introduction Developing robust clinical outcome measures in amyotrophic lateral sclerosis (ALS) remains challenging because onset, progression, and patterns of functional involvement are heterogeneous across individuals [1]. The ALS Functional Rating Scale–Revised (ALSFRS‐R) is the most widely used functional outcome measure in ALS research [2], but measurement‐property limitations, including ordinality, multidimensionality, and limited sensitivity to change, restrict its ability to detect progression [3, 4]. Although self‐administered ALSFRS‐R instruments enable remote assessment, they are typically collected intermittently [5]. There is therefore substantial interest in objective digital outcomes that can be captured passively and repeatedly in free‐living settings. Actigraphy‐derived metrics from research‐grade accelerometers have emerged as promising digital measures of function in ALS [6, 7, 8, 9]. These devices support passive monitoring and have demonstrated feasibility, associations with established clinical measures, and sensitivity to disease progression over time [10, 11, 12, 13, 14]. However, research‐grade devices are often deployed intermittently, require dedicated study infrastructure, and may be poorly scalable for long‐term decentralized monitoring. Consumer technologies may address some of these barriers [15]. Step counts are attractive as an outcome because they are intuitive, clinically interpretable, closely linked to ambulatory function, and routinely derived by many consumer wearables. Recent work using passively collected smartphone data supports the potential of consumer step counts to track ALS progression and predict survival [16, 17]. Consumer‐grade smartwatches (hereafter “smartwatches”) are now widely adopted [18, 19], and offer practical advantages for longitudinal remote monitoring, including integrated interfaces that may support adherence, multimodal sensing, and scalable designs [20]. Prior studies in non‐ALS populations suggest that consumer wearables can estimate step counts with reasonable agreement in free‐living settings, although systematic bias has been reported and performance varies by device and context [21, 22]. Step‐count accuracy may be reduced in individuals with slow or abnormal gait and when assistive devices are used, which is particularly relevant to ALS and could lead to differential misclassification as mobility declines [23, 24]. Notably, no wrist‐worn step‐counting algorithm has been validated against ground‐truth step counts in ALS [25]. We conducted a prospective observational study in people with ALS to evaluate long‐term remote collection of consumer smartwatch data and to compare smartwatch‐derived daily step counts with step counts estimated from simultaneously worn research‐grade accelerometers. Our aim was to evaluate agreement between smartwatch‐ and accelerometer‐derived step counts in ALS, while also assessing the feasibility of smartwatches for longer‐term free‐living monitoring. Methods Study Design We conducted a single‐centre prospective observational study of wearable device use in people with ALS followed through the Calgary ALS and Motor Neuron Disease Clinic. Participants were enrolled between May 2022 and March 2023 and followed for 12 months. Some extended participation until study closure in April 2024. Eligible participants were adults aged 18 years or older with ALS diagnosed according to the Gold Coast criteria [26], who were within 4 years of symptom onset, able to walk 10 m independently without a walker or rollator on observed assessment at screening (other gait aids such as canes or orthoses were permitted), and able to complete online self‐assessments in English independently or with assistance. Participants were also required to use a compatible smartphone with internet access. Individuals were excluded if they had a non‐ALS condition substantially affecting mobility, cognitive impairment limiting their ability to complete assessments, or were imminently planning for medical assistance in dying. The study was approved by the University of Calgary Conjoint Health Research Ethics Board (REB22‐0375), and all participants provided informed consent. Reasons for declining participation were collected using a voluntary paper‐based survey offered to individuals who declined enrolment. Completion of the survey implied consent (REB22‐1627). Device Procedures At enrolment, participants were provided with a Fitbit Sense smartwatch (hereafter “smartwatch”; Fitbit Inc., San Francisco, CA, USA; approximate cost $300 USD in 2022) and instructed to wear it on their preferred wrist, typically the non‐dominant wrist, as continuously as possible, removing it only for charging or skin care. The smartwatch had a manufacturer‐rated battery life of 5–7 days under typical use. Participants were advised to charge it on two consistent days per week to support adherence, although charging was not directly monitored. Participants were assisted with Fitbit app installation and Bluetooth pairing to their smartphone. Smartwatch minute‐level and day‐level data, including step counts and heart rate, were retrieved regularly through the Fitbit Web API [27], with support at no cost from a technology partner (Lumiio, Calgary, Alberta, Canada). Participants were also provided with an ActiGraph GT9X Link accelerometer (hereafter “accelerometer”; Ametris, Pensacola, FL, USA; approximate cost $325 USD in 2022) at each in‐person clinic visit, which occurred roughly every 3 months according to routine clinical care schedules (day 1 of recording). They were instructed to wear the accelerometer continuously on the opposite wrist to the smartwatch, as pre‐study testing had identified same‐wrist wear as uncomfortable. Same‐wrist wear was permitted if preferred. Devices were worn until removal on waking on day 9, yielding 7 full days of recording. Participants could remove the device for bathing or personal care. Accelerometer devices did not require recharging and were returned by prepaid mail. Clinical Data Demographic and clinical characteristics were collected at enrolment. ALS Functional Rating Scale–Revised (ALSFRS‐R) scores were obtained monthly through online questionnaires based on the self‐assessed version developed and validated by Bakker et al. with minor wording modifications to align with North American/Canadian English [28], administered in REDCap (v14.5), hosted by the Centre for Health Informatics at the Cumming School of Medicine, University of Calgary [29, 30]. Each day was assigned the nearest ALSFRS‐R score within 30 days. Days without an assessment in that window were excluded from step count comparisons by ambulatory status. Change in ALSFRS‐R (ΔALSFRS‐R) was calculated per month from disease onset to enrolment, assuming a “normal” ALSFRS‐R score of 48 at symptom onset. Smartwatch Data Processing Smartwatch data were processed in Python (version 3.9.21). Because missing smartwatch data could reflect device non‐wear, smartwatch‐to‐phone synchronization failure, or Fitbit Web API retrieval failure, wear time was assessed on days with available data, but daily compliance could not be directly assessed. Minute‐level step counts were derived from raw smartwatch accelerometry by Fitbit's proprietary consumer algorithm and retrieved as processed values. Daily step counts were calculated by summing minute‐level estimates over each 24‐h period. Smartwatch wear time was estimated from minute‐level heart rate data. A minute was considered wear time if a non‐zero heart rate value was present, and daily wear time was calculated as the total number of such minutes. Sustained zero step counts were not used to define non‐wear to avoid misclassifying sedentary behavior as non‐wear [22]. This method has not been formally validated against ground truth. It may underestimate wear time if heart rate fails to record (e.g., poor sensor contact or motion artifact), but is expected to perform better than classifying non‐wear by sustained zero steps, which has been shown to underestimate Fitbit wear compared with heart‐rate‐incorporating approaches [22]. Accelerometer Data Processing Accelerometer raw triaxial acceleration data recorded at 30 Hz were downloaded using ActiLife software (version 6.13.5, Ametris, Pensacola, FL, USA). Wear time was estimated using the Choi algorithm [31, 32] applied to vector magnitude units, with previously optimized parameters for low‐activity populations: interval length 210 min, spike tolerance 1 min, and window length 20 min [33]. These parameters, previously used in ALS by Unnik et al. [34] may overestimate wear time but reduce misclassification of sedentary wear time as non‐wear in less active people [35]. Minute‐level step counts were derived from raw wrist accelerometry using the Oxford Wearables self‐supervised foundation model toolkit in Python (version 3.9.21) [36, 37]. In the absence of a step‐counting algorithm validated against ground‐truth steps in ALS, this open‐source algorithm represents a well‐validated wrist‐worn approach and provides a transparent reference against which to assess proprietary consumer step counts. Daily step counts were calculated by summing minute‐level estimates over each 24‐h period. Accelerometer days representing device application, return, or mailing transit were excluded before analysis. Device Wear and Data Availability For both devices, daily step counts were obtained from valid wear days, defined as days with at least 10 h of estimated wear time [38]. Wear and data availability were summarized descriptively for the smartwatch and accelerometer. For the smartwatch, we reported the number of observed days with data relative to the number of possible study days between enrolment and study end. Accelerometer donning and doffing days (first and last day of each wear period) were excluded from analysis as partial days that may have captured non‐participant movement, and we reported observed days with data relative to possible recording days during all 7‐day recording windows. Because the two devices differed in wear protocol and wear‐time estimation methods, formal statistical comparison of wear time between devices was not performed. To explore the mechanism of missing smartwatch data, a mixed‐effects logistic regression model was fit with daily smartwatch missingness as the dependent variable and a random intercept for participant. Candidate predictors included age at symptom onset, sex, time from symptom onset to enrolment, total duration follow‐up, day of study, day of week, and ALSFRS‐R total score. These analyses assessed whether missingness was plausibly missing completely at random [39]. Step Count Comparison Agreement analyses were restricted to days on which the smartwatch and accelerometer were worn concurrently and both met the valid‐day definition of at least 10 h of wear. Mean daily step counts were compared using paired Student's t‐tests. Association between device step counts was assessed using Pearson correlation coefficients. Agreement was further evaluated using Bland–Altman analysis, with estimation of mean bias and 95% limits of agreement. Proportional bias was assessed by regressing the between‐device difference on the mean paired daily step count. To account for potential differences in between‐device wear time, a sensitivity analysis was performed restricting the primary comparison to concurrent valid days where between‐device wear time differed by 30 min or less. As an additional descriptive analysis, mean daily step counts were compared between ambulatory and non‐ambulatory participants separately for each device using Welch's t‐tests. Participant days were classified as ambulatory if their ALSFRS‐R walking subscore on that day was ≥ 2. To assess whether devices tracked longitudinal change similarly, linear mixed‐effects models were fit with daily step counts as the outcome, day of study as a fixed effect, and a random intercept for participant, fitted separately for each device and restricted to participants with at least 6 months of concurrent valid days. A device‐by‐time interaction term was used to formally test whether rates of change differed between devices. Statistical Analysis Statistical analyses were performed in R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are reported as mean (standard deviation) or median (interquartile range), as appropriate, and categorical variables as counts and percentages. Two‐sided p‐values < 0.05 were considered statistically significant. Results Of 170 people with motor neuron disease screened, 64 were eligible and 40 were enrolled (Figure 1). Median follow‐up was 16.1 months (IQR 7.1–18.8), maximum follow‐up was 22.3 months, and 19 (47.5%) continued beyond 12 months in the study extension. Baseline characteristics are shown in Table 1. During the initial 12‐month follow‐up period, 7 participants (17.5%) withdrew early because of study burden and 8 (20.0%) died. One participant was lost to follow‐up immediately after enrolment and did not return devices or contribute device data. The remaining 39 participants contributed device data for analysis. Of these, 7 transitioned from ambulatory to non‐ambulatory status during follow‐up based on ALSFRS‐R walking subscore, at a median of 6.5 months after enrolment (IQR 6.4–9.1), with a median of 6.0 months of follow‐up after transition (IQR 3.5–12.8). Among the 27 participants with follow‐up clinical visit data, 5 (18.5%) acquired a new gait aid during follow‐up and 4 (14.8%) began using a wheelchair. Participant flow and reasons for non‐eligibility and non‐participation through screening, enrolment, and contribution of device data. MAiD, medical assistance in dying; MND, motor neuron disease. TABLE 1Baseline characteristics of the study cohort at enrolment.CharacteristicsStudy cohort (N = 40)Age, median years (IQR)66.0 (58.0–71.6)Female, n (%)9 (22.5%)Time from onset, median months (IQR)19.5 (7.7–32.9)FVC, % predicted mean (SD)85.9 (19.6)ALSFRS‐R, mean (SD)34.3 (8.0)ALSFRS‐R gross motor combined score (out of 12), median (IQR)8.5 (5.75–11)ALSFRS‐R walking subscore, median (IQR)3 (2–4)ΔALSFRS‐R, median (IQR)−0.56 (−1.15 to −0.37)Region of onsetRespiratory, n (%)1 (2.5%)Bulbar, n (%)21 (52.5%)Thoracic, n (%)1 (2.5%)Cervical, n (%)12 (30.0%)Lumbar, n (%)5 (12.5%)Wheelchair use at baseline4 (10.3%)Any gait aid at baseline14 (35.9%)Ankle foot orthosis2 (5.1%)Cane7 (17.9%)Walker12 (30.8%) Note: Participants may use more than one gait aid; individual gait aid counts sum to more than total.Abbreviations: ALSFRS‐R, Amyotrophic Lateral Sclerosis Functional Rating Scale‐Revised; FVC, forced vital capacity; IQR, interquartile range; SD, standard deviation. Baseline characteristics of the study cohort at enrolment. Note: Participants may use more than one gait aid; individual gait aid counts sum to more than total. Abbreviations: ALSFRS‐R, Amyotrophic Lateral Sclerosis Functional Rating Scale‐Revised; FVC, forced vital capacity; IQR, interquartile range; SD, standard deviation. Device Wear and Data Availability Data availability and valid days are summarized in Figure 2. After exclusion of 236 accelerometer application and return days, 875 of 915 remaining accelerometer days (95.6%) and 7572 of 8093 smartwatch days (93.6%) met the valid‐day definition. Per‐participant means (SD), across the 39 participants contributing device data, were 207 (192) smartwatch days and 23.5 (13.4) accelerometer days, of which 194 (183) and 22.4 (13.4) respectively met the valid‐day definition. The 503 concurrent valid days used for device‐comparison analyses were contributed by 34 of the 39 participants, with a mean of 14.8 (10.6) days per contributing participant. Across participants, the median proportion of observed smartwatch days relative to possible study days was 45.3% (IQR 17.1%–87.1%). By design, the accelerometer was worn only during clinic‐aligned 7‐day periods, within which 99.8% of expected days had recorded data across 131 wear periods. Among the 34 participants with concurrent device wear, 32 wore the smartwatch and accelerometer on opposite wrists and 2 wore both devices on the same wrist. Missing smartwatch data varied according to participant characteristics, days since enrolment, and ALSFRS‐R total score, but not day of week. Flow diagram of device data collection and valid‐day selection. The Venn diagram shows mutually exclusive regions of recorded days. Total smartwatch days were 8093 (smartwatch‐only + concurrent) and total accelerometer days were 1151 (accelerometer‐only + concurrent). Invalid accelerometer days included device application and return days at the beginning and end of wear periods, which contained incomplete data and capture of researcher or shipping movements. Among all days with recorded device data, mean estimated daily wear time was at least 21 h (87% of the day) for both devices (Table 2). Estimated daily wear time was higher for the accelerometer than the smartwatch, although formal comparison was not performed because of differences in wear‐time estimation methods and device protocols. On the 503 concurrent valid days, mean estimated wear time was 91.1% of the day (SD 14.5) for the smartwatch and 96.6% (SD 10.3) for the accelerometer. The smartwatch had lower wear time than the accelerometer on 72.0% of concurrent valid days. TABLE 2Device wear and data availability for smartwatch and accelerometer among days with recorded device data. Accelerometer donning and doffing days were excluded as partial days that may have captured non‐participant movement. Device‐specific totals include concurrent wear days.DeviceDays, n Wear time per day, mean % ± SD (95% CI)AccelerometerSmartwatchSmartwatch8093NA87.4 ± 21.6 (86.9–87.9)Accelerometer91592.0 ± 20.9 (90.6–93.4)NAAccelerometer and Smartwatch simultaneously53495.4 ± 14.3 (94.2–96.6)87.8 ± 20.9 (86.0–89.6)Abbreviations: 95% CI, 95% confidence intervals; SD, standard deviation. Device wear and data availability for smartwatch and accelerometer among days with recorded device data. Accelerometer donning and doffing days were excluded as partial days that may have captured non‐participant movement. Device‐specific totals include concurrent wear days. Abbreviations: 95% CI, 95% confidence intervals; SD, standard deviation. Daily smartwatch missingness increased with day of study (OR 6.39 per SD, 95% CI: 5.85–6.98; p < 0.001) and higher ALSFRS‐R total score, indicating less impairment (OR 1.17 per 1‐point increase, 95% CI: 1.15–1.18; p < 0.001), but was not predicted by age, sex, time from onset to enrolment, total duration follow‐up, or day of week. Device Step Count Comparison The smartwatch recorded fewer steps than the accelerometer on concurrent days and step counts were strongly correlated between devices (Figure 3A). Bland–Altman analysis showed a mean bias of −586 steps (95% CI: −747 to −425) as shown in Figure 3B. A small but statistically significant proportional bias was present, with greater between‐device differences at higher mean daily step counts (β = −0.10, p < 0.0001). Mean daily step counts were substantially lower in non‐ambulatory than ambulatory participants for both devices (Table 3). Fewer device days were available for non‐ambulatory participants, reflecting that eligibility required baseline ambulation and only six participants contributed device data after becoming non‐ambulatory. High daily step counts (> 10,000 either device) occurred on 69/503 days (13.7%), contributed by 9 participants, and were associated with intact gross motor function (median ALSFRS‐R gross motor domain 12/12, IQR 12–12; median walking subscore 4). (A) Correlation between daily step counts measured by the smartwatch and accelerometer on concurrent wear days (≥ 10 h). (B) Bland–Altman plot comparing daily step counts shows that the smartwatch recorded a mean of 586 fewer steps than ActiGraph (bias = −586), with 95% limits of agreement ranging from −4193 to 3021 steps. The shaded region represents the limits of agreement. TABLE 3Daily step counts from the smartwatch and accelerometer overall and by ambulatory status on valid days (≥ 10 h wear).DeviceDays n Step counts mean ± SD (95% CI)AccelerometerSmartwatchSmartwatch7572NA3498 ± 4273 (3402–3594)Ambulatory6090NA3665 ± 4310 (3557–3773)** Non‐ambulatory436NA137 ± 290 (110–164)** Accelerometer8753830 ± 4390 (3539–4121)NAAmbulatory6974023 ± 4350 (3700–4346)*** NANon‐ambulatory5186 ± 145 (45–127)*** NAAccelerometer and Smartwatch simultaneously5034066 ± 4797 (3646–4486)* 3480 ± 4365 (3098–3862)* Ambulatory4154357 ± 4786 (3895–4819)3750 ± 4486 (3317–4183)Non‐ambulatory2893 ± 163 (30–156)47 ± 72 (19–75) Note: Days without valid ALSFRS walking subscore were included in total day counts but not ambulatory stratification. Device‐specific totals include days when the smartwatch and accelerometer were worn separately or concurrently.Abbreviations: 95% CI, 95% confidence intervals; SD, standard deviation.* p < 0.0001 for difference in daily step counts between smartwatch and accelerometer on concurrent days (paired t‐test).** p < 0.0001 for the difference in mean daily smartwatch step counts between ambulatory and non‐ambulatory participants (Welch's t‐test).*** p < 0.0001 for the difference in mean daily accelerometer step counts between ambulatory and non‐ambulatory participants (Welch's t‐test). Daily step counts from the smartwatch and accelerometer overall and by ambulatory status on valid days (≥ 10 h wear). Note: Days without valid ALSFRS walking subscore were included in total day counts but not ambulatory stratification. Device‐specific totals include days when the smartwatch and accelerometer were worn separately or concurrently. Abbreviations: 95% CI, 95% confidence intervals; SD, standard deviation. p < 0.0001 for difference in daily step counts between smartwatch and accelerometer on concurrent days (paired t‐test). p < 0.0001 for the difference in mean daily smartwatch step counts between ambulatory and non‐ambulatory participants (Welch's t‐test). p < 0.0001 for the difference in mean daily accelerometer step counts between ambulatory and non‐ambulatory participants (Welch's t‐test). Bias direction varied within and across participants. Twenty‐three of 34 participants (67.6%) had a negative mean bias and 11 (32.4%) had a positive mean bias. Eight (23.5%) showed consistently lower smartwatch step counts on every concurrent day and 6 (17.6%) consistently higher. The primary comparison was repeated restricting to days where between‐device wear time differed by 30 min or less (n = 254 days, 28 participants). Mean bias was −460 steps (10.8% underestimation; 95% CI: −670 to −250; p < 0.001), Pearson r = 0.94 (95% CI: 0.92–0.95), and 95% LoA were −3790 to 2870 steps. Among 18 participants with at least 6 months follow up of concurrent valid days (403 days total), both devices showed a similar rate of decline in daily step counts over time. The estimated decline was −4.59 steps per day (95% CI: −6.08 to −3.10) for the smartwatch and −5.34 steps per day (95% CI: −6.84 to −3.85) for the accelerometer. The difference in slopes between devices was not statistically significant (interaction p = 0.76), suggesting both devices tracked longitudinal change similarly despite persistent mean bias. Discussion In this prospective study of wearable device use in ALS, long‐term remote collection of smartwatch data was feasible, although data availability was incomplete over follow‐up. On concurrent valid days, smartwatch‐ and accelerometer‐derived step counts were strongly correlated, but the smartwatch recorded fewer steps on average and showed wide day‐level limits of agreement relative to the accelerometer. Step counts from both devices were markedly lower in non‐ambulatory than in ambulatory participants. Together, these findings suggest that consumer‐grade smartwatch step counts can capture ambulatory activity in ALS and align with research‐grade accelerometer‐derived estimates at the group level, while highlighting important limitations in between‐device agreement for individual day‐level interpretation. The smartwatch enabled prolonged remote monitoring in people with ALS in a free‐living environment, generating a substantial number of observation days. However, smartwatch data availability was incomplete, in part because consumer wearable data capture depends on a multi‐step ecosystem (device wear, device‐to‐smartphone syncing, and smartwatch server to research data transfer) that we could not directly monitor. This resulted in substantial between‐participant variability in the proportion of observed days. Missing smartwatch data should therefore not be interpreted simply as non‐adherence. Data completeness was higher for accelerometers, likely reflecting the limited 9‐day wear protocol (7 days, plus donning and doffing days) after each clinic visit rather than continuous long‐duration wear. Higher completeness has been reported in other continuous long‐duration studies: Johnson et al. reported a median of 158 valid days out of 178 observation days (~89%) over 6 months in ALS using a research‐grade wrist accelerometer [40], and a 12‐month community study of 1314 non‐ALS adults wearing a smartwatch reported 77% of days met the same 10‐h threshold used here [41]. Our lower completeness likely reflects disease‐related attrition compounded by the absence of real‐time compliance monitoring infrastructure to support participants. Balbim et al. identified infrequent device syncing as a common issue across Fitbit‐based studies [42], recommending careful study preparation, compliance monitoring dashboards and notifications, and adequate study staff to provide technical support and individualized compliance strategies. These measures equally apply to long‐term studies using research‐grade devices, and we suspect they would have reduced missingness in our study. Consumer wearables place practical burdens on participants (donning and doffing, charging, syncing) that may be particularly demanding in people with hand weakness. In our cohort, missingness increased with study duration, suggesting that interventions to support participation are even more important in longer studies. Unexpectedly, missingness was also associated with higher ALSFRS‐R score (indicating less impairment), which may reflect greater motivation or caregiver support among more severely affected participants. Day of the week did not affect missingness, possibly reflecting consistent weekday‐weekend routines in older, often retired adults. Importantly, smartwatch wear time was not reduced on days when participants also wore the accelerometer, supporting the feasibility of pairing consumer and research‐grade devices over short monitoring periods. Smartwatch step counts showed strong correlation with accelerometer‐derived step counts on concurrent valid days, supporting their ability to rank relative ambulatory activity across days and participants. However, the smartwatch recorded fewer steps on average than the accelerometer and demonstrated wide limits of agreement, indicating substantial day‐level variability between devices. The presence of proportional bias suggests that disagreement increased at higher activity levels. These findings indicate that while smartwatch and accelerometer step counts are directionally aligned, they should not be assumed to be interchangeable for individual day‐level measurement. Group‐level comparisons and within‐participant longitudinal monitoring may still represent useful outcomes for research and clinical use, and aggregation across multiple days may help stabilize individual‐level estimates. Despite persistent mean bias, both devices showed a similar rate of decline in daily step counts over follow‐up, suggesting that either device could capture longitudinal change in ALS at the group level. Although smartwatch and accelerometer hardware and outputs are similar [43], several factors may explain the observed between‐device differences. Smartwatch step counts are generated using proprietary consumer algorithms, whereas accelerometer step counts were derived from raw wrist accelerometry using the Oxford Wearables self‐supervised foundation model [36, 37]. Importantly, no wrist‐worn step‐counting algorithm has been validated against ground‐truth step counts in ALS. This comparison therefore evaluates agreement between two distinct step count estimates but cannot establish whether either reflects true walking activity. The most rigorous ALS‐specific step‐counting validation work to date was conducted exclusively at the ankle, with those authors citing wrist‐worn placement as unreliable due to altered arm swing and intermittent use of assistive devices [25]. Heterogeneous gait changes in ALS pose unique challenges for step counting algorithms; altered gait and arm swing, and intermittent use of mobility aids may affect step‐detection performance differently across devices and algorithms [23, 24, 25, 44]. More than half of our participants used a gait aid during the study period. Activity trackers underestimate steps in older adults and clinical populations [23, 45, 46], although overestimation at low to moderate cadences has also been reported [47], and similar limitations may exist for algorithms applied to raw accelerometry data. Asymmetric weakness may also have contributed, particularly as 32 of 34 participants wore the smartwatch and accelerometer on opposite wrists. Additionally, estimated wear time was lower for the smartwatch than the accelerometer on concurrent valid days, and step counts were not adjusted for wear time. This may have been due to differences in wear time estimation algorithms, or other factors such as smartwatch charging requirements during periods of concurrent wear, while the accelerometer did not require charging. However, sensitivity analyses restricting comparisons to days with matched wear durations showed persistent negative bias, suggesting that factors other than wear time contribute to the observed bias. This study has several limitations. The cohort was modest in size and recruited from a single centre. Eligibility criteria required baseline ambulation and smartphone access, which may limit generalizability to people with low mobility or lower technology access, and risks excluding groups already underrepresented in digital health research [20, 48]. Female participants were also underrepresented relative to males, a common pattern in ALS digital device studies [1, 6]. Missing smartwatch data could not be disentangled into non‐wear versus technical failure. The two devices also differed in wear protocol, wear‐time estimation, and step‐count algorithms, and were typically not worn on the same wrist. The valid‐day definition did not distinguish between waking and sleep wear; however, mean wear on valid days exceeded 91% of the 24‐h period for both devices, suggesting substantial daytime wear was captured in most cases. Future studies could consider a daytime‐only wear threshold. Smartwatch step counts were generated by a proprietary algorithm whose update history is not publicly disclosed, so any changes during the study cannot be characterized. Finally, daily step count observations were clustered within participants, violating the independence assumption of the paired t‐test and Pearson correlation, though a sensitivity analysis using per‐participant mean step counts yielded consistent findings. The choice between devices depends on the intended use case. Despite technical and regulatory challenges, smartwatches may be well suited to selected clinical and research applications in ALS. Practical advantages include participant use beyond research, bring‐your‐own‐device study designs, and the potential to support decentralized monitoring in settings where in‐person assessment is less accessible. Smartwatches may detect changes between in‐clinic assessments that could prompt earlier follow‐up or targeted investigation. A key limitation of consumer devices is restricted access to raw accelerometry, which constrains independent data processing and the development of transparent algorithms. Additionally, consumer wearables are typically restricted to wrist placement, while accelerometers worn at the torso or ankle provide more accurate step counts [25]. Research‐grade devices remain important by retaining raw data access, offering greater methodological control and likely a stronger position for regulatory consideration as digital clinical outcome measures. With further validation, there may be opportunities for collaboration with industry to improve access to raw data or deploy more stable and transparent clinical algorithms on consumer platforms. In people with ALS, long‐term remote monitoring with consumer‐grade smartwatches was feasible. Smartwatch‐derived step counts were strongly correlated with research‐grade accelerometer‐derived step counts, although the smartwatch systematically recorded fewer steps and showed wide day‐level limits of agreement. These findings support the use of smartwatch‐derived step counts as a pragmatic measure of ambulatory activity in ALS, particularly for group‐level analyses and within‐device longitudinal monitoring, while emphasizing that device‐specific biases must be considered in interpretation. Author Contributions Gordon Jewett: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, project administration, data curation, software. Reed Ferber: supervision, writing – review and editing, conceptualization, methodology. Colin B. Josephson: conceptualization, methodology, writing – review and editing, supervision. Lawrence Korngut: conceptualization, investigation, writing – review and editing, methodology, validation, project administration, supervision, resources. Joon Lee: conceptualization, investigation, methodology, validation, writing – review and editing, project administration, supervision, resources. Funding This work was supported by ALS Canada. Ethics Statement We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. Conflicts of Interest Gordon Jewett has received salary support from ALS Canada, research support from Tanabe Pharma and Amylyx Pharmaceuticals, speaker fees from Biogen, and advisory board honoraria from Amylyx Pharmaceuticals, Argenx Canada, and Tanabe Pharma. Joon Lee is a co‐founder and major shareholder of Symbiotic AI Inc. Lawrence Korngut has received advisory board honoraria and research support from Amylyx Pharmaceuticals, Biogen and MT Pharma Canada. The other authors declare no conflicts of interest. Acknowledgments Funding for this study was provided by ALS Canada. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.