This is an observational, single-arm study with no control group. The study is a sub-project from the “When Movement Moves”, described elsewhere (16). During a 16-week program, subjects wore a wrist-based HR monitor. The study was conducted in concordance with the Declaration of Helsinki (37), registered and approved by the Ethics Committee of Denmark’s Capital Region (Journal no.: H-20010668), including a pre-registration at Clinical Trials.gov (NCT04536779). All subjects gave written informed consent prior to inclusion.

The Team Twin organization provides an adapted running activity for PLWD (henceforth referred to as handiathletes) and has ten clubs across Denmark. It has approximately 450 members, 150 handiathletes and 300 volunteer-runners (38).

The activity has the same format across the individual Team Twin clubs. A standard session is comprised of three phases, the before, during, and after phases. Sessions are typically scheduled on Sunday mornings. Before; runners and athletes meet and handiathletes are assisted in transferring from their wheelchair to running chair. The route is planned, and members and associates have casual conversations. During; co-running, handiathletes are sat in their running chairs driven by a runner. After; casual conversation, eating, and re-hydrating before end of session. The organization furthermore participate in numerous official races (half- and full marathons and other running events) national and international during their season.

We invited all 150 handiathlete members to participate. Between January and March 2021, 22 were enrolled (15%), representing six clubs. One withdrew after enrolment, and three either dropped out and or were excluded for non-adherence. Thus, 18 handiathletes were included in the analyses of this study (flowchart in Supplementary Figure S1). Inclusion criteria were 1) affiliation to a Team Twin club and 2) 18 + years of age. Characteristics of the sample were obtained between April and June, 2021. The study period began mid-June and finished in October 2021.

All handiathletes were equipped with a Garmin Vivosmart 4 watch (39). The Vivosmart 4 was selected based on a practical approach taking the population into account. Following practical criteria was essential for the device selection, 1) suitable for relatively small wrists (due to disabilities) to fit all subjects properly, 2) comfortable to wear day and night for four months, to strengthen compliance, 3) long battery life (7 days) to avoid recharging daily and limit interference with the subject’s everyday life. The Vivosmart 4 did comprise the required features for the overall study (16) (sleep and HR monitoring) and the price accommodated the project budget. We personalized the devices based on the participants’ age, sex, height, and weight. All widgets (e.g., total steps per day and body battery (39)) were pre-emptively removed from the display to prevent interference. All obtained data were uploaded to individual Garmin profiles only accessible for the research team. All handiathletes, relatives and caretakers received a comprehensive walkthrough of the Vivosmart 4 (39). We employed the “Running” activity profile provided by Garmin (39) for tracking Team Twin sessions. We instructed the handiathletes and volunteer-runners to activate/deactivate to track only during the actual co-running (the “during”-phase).

The Vivosmart 4 was placed on the athlete’s preferred wrist. It was worn both day and night, the latter to estimate resting heart rate (RHR) with the highest accuracy (40). RHR was estimated from the lowest 30-minute average calculated over the 24 h (40). We excluded the first two weeks of the RHR-data to gain further accuracy to accommodate a subject-specific calibration period (41).

We estimated peak heart rate (HR_max) using Fernhall et al.’s formula (42), which takes disability into account. We calculated the individual heart rate reserve (HRR) by estimating a mean RHR across the whole period (excluding the calibration weeks described above) and subtracting it from the estimated HR_max (HRR = HR_max-RHR). Based on the HRR, we determined %HRR (resulting in intensity levels) by the equation from Karvonen (43). Intensity levels based on HRR instead of HR-dependent estimation (e.g., %HR_max) is recommended for avoiding over- or underestimation. The HRR method consider 1) individual assessment; which reflects the RHR, which differs from person to person, and thus less sensitive to over- or underestimation, 2) accuracy in submaximal exercise; more accurate to predict exercise intensity at submaximal intensity levels due to various factors like stress, fatigue, or medication use, and 3) cardiorespiratory level: individuals with higher fitness levels tend to have lower RHR resulting in a larger HRR and potentially higher exercise intensities. Thus, it is more appropriate for people with different levels of cardiovascular fitness to apply HRR estimation (21, 44).

To target individual HR intensity zones, we applied the intensity division from two guidelines (21, 45), and estimated HR for each participant reflecting the distribution from Table 1 (individual HR reflecting the intensity levels from Table 1 can be seen in Supplementary Table S1). We applied the predefined intensity zones and used the HR_max estimation from Fernhall et al. (42) as a frame to categorize HR responses in relation to the intensity of PA.

A convenience subsample of 5 subjects (n = 5) were recruited for the evaluation of device agreement and accuracy. The sample consisted of subject who voluntary agreed, were accessible, and able to wear the HRM-DUAL™. We compared the wrist-worn photoplethysmography-based Vivosmart 4 with a Garmin HRM-DUAL™ chest electrocardiogram-based HR monitor, connected to a Garmin Forerunner 265. Both devices were worn simultaneously during a Team Twin session, the Forerunner 265 was placed on the running chair, not the athlete. Each athlete wore two monitors and acted as their own controls (concurrent validity). A researcher attached the wrist-worn and the chest-worn monitor as instructed (39, 46) and simultaneously activated/deactivated the monitors. During this study, we assessed the two devices using methods accommodating accuracy, agreement, and inter-rater agreement between the HR monitors (as described under 2.7.2. HR agreement—secondary objective).

When using commercial HR monitors, including the one used in this study, continuous measurements (every second or other predefined time-interval) were not a standard output (24). The Vivosmart 4 sampling frequency is HR-determined instead of a pre-defined time interval. Meaning that the time frequency keeps in the same timeslot until the HR exceeds ±3 bpm. Thus, we used an HTML script to transform the HR data into a second-by-second time-interval dataset (see Supplementary File S1). Data was then trimmed with a ±15 s moving average filter, rounded to the nearest whole number, to accommodate interference, potential time displacement (when activating/deactivating the wrist/chest HR monitors), and device-specific measuring errors. This approach was applied to both datasets (wrist and chest). For the wrist data, we also compressed HR data into a one-minute mean and transferred the HR into %HRR (intensity levels) to fit the frame of intensity zones (Table 1). All statistical analyses were conducted with STATA version 17 (StataCorp LP, College Station, TX, USA). We applied a two-sided test with a significance level of 0.05% and 95% confidence intervals (16).

Our sample consisted of seven females and eleven males aged 19–65 (mean 35.4 years). In-depth characteristics are found in Table 2. Cerebral palsy (CP) was the most frequently self-reported diagnose (72%). Other conditions reported were muscular dystrophy, physical and mental disability, inherited neurodegenerative diseases, and multiple sclerosis (28%). Of those with CP, most were diagnosed with quadriplegia (92%), with a Gross Motor Function Classification System (GMFCS) at IV-V (85%). Due to their various conditions, eight handiathletes (44%) used both non- and prescription-only drugs on a daily base. These included, but were not limited to, Sirdalud, Euthyrox, Buscopan, medicinal cannabis etc. Common side-effects of these drugs might influence coronary circulation, resulting in palpitation, frequent pulse, increased RHR and extrasystole. During the 16-week program, the handiathletes had an average of seven sessions (±4) with the Vivosmart 4 activated and a mean HR of 90 (±16.8) bpm (20 ± 13.2%HRR) across race and training. An average session lasted 77 (±24) minutes and covered 10.5 (±4) km. The subsample (n = 5) used for the agreement evaluation consisted solely of males with GMFCS IV-V, had slightly higher mean HR during sessions, and a lower-than-average-sample BMI and weight (Supplementary Table S2).

A total of 130 sessions, 113 training (162 h, 18 m, and 46 s), and 17 races (39 h, 51 m, 47 s), gave a total monitoring time of 202:10:33 (HH:MM:SS) recorded with the Vivosmart 4. Table 3 depicts time spent in each intensity zone (cf. Table 1). Most time (80%) was spent in the very-light intensity zone, which equals sedentary behavior. All subjects sustained an HR response consistent with light and moderate intensity zones; however, the duration in each level varied considerably across the group. Nine handiathletes (50%) reached the vigorous intensity zone at one or more points during the program period, and the duration in it varied between 0.02% and 6.6% of their total recorded time (Supplementary Figure S2).

We found a decline-in-HR-over-time tendency during both training and races. This tendency is depicted in Figure 1, which presents the %HRR-responses for all handiathletes grouped by race or training participation. Further, the plot illustrates a variance in the %HRR with increasing duration, meaning that dispersion increases with time. The median training duration was 79 min (25 p; 63 min, 75 p; 105 min), and the median race duration was 153 min (25 p; 105 min, 75 p; 175 min) (Figure 1). The greater dispersion after the 79th and 153rd min, respectively, are likely due to significantly fewer observations, as few handiathletes had training/race sessions lasting for more than 105 and 175 min (75% percentile), respectively. During race participation, we found a higher mean %HRR (intensity level), compared to training sessions, 26%HRR (95%CI 20.9; 31.0) vs. 19%HRR (95%CI 14.1; 24.0), respectively. A significantly higher mean of 6.8%HRR (95%CI: 6.1; 7.4, p < 0.000) adjusted for sex, age, and duration of the session, in favor of race participation was found (see graph in Supplementary Figure S3). A sensitivity analysis revealed the same results when only including handiathletes who had data from both race and training sessions (n = 13) (data not shown).

A total of 11:26:25 (HH:MM:SS) (equaling 41185 HR values) were collected for the second-by-second analysis, evaluating the agreement between the wrist and chest monitors. The mean time subjects wore the devices was 01:55:04 (min; 01:02:09, max; 02:37:38) (HH:MM:SS). We conducted six wrist and chest evaluations on five handiathletes. Two out of the six evaluations were conducted during a race.

Across the evaluation of agreement and second-by-second analysis, the wrist monitor showed an acceptable mean accuracy with the chest HR monitor [Rc = 0.86 (CI 95% 0.863; 0.868)] (Figure 2A). The Bland & Altman plot reveals that dispersion occurs during the whole spectrum of HR intensities; however, more excellent dispersion with increasing HR. That means, with an increase in HR, the agreement between the two monitors decreases, resulting in greater heteroscedasticity. The limits of agreement between the wrist and the chest monitor fall within −16 and +13 bpm. Despite an overall acceptable accuracy, the individual variation from Lin’s Rc-coefficient and the agreement from the Bland & Altman plots vary notably (Rc between 0.05–0.91 and Bland & Altman plot between −27 and +18, see Supplementary Figure S4). Cohens kappa’s coefficient reveals a substantial inter-rater agreement between the two HR monitors of the intensity zones (k = 0.75), with an overall agreement of 92%. The chest monitor generally registers more time at higher intensity zones than the wrist monitor (e.g., total time in moderate-intensity zone intercepted for the wrist HR monitor; 0:53:12 (HH:MM:SS), and for the chest HR monitor; 1:01:46 (HH:MM:SS)). The distribution and inter-rater agreement between the Vivosmart 4 and the HRM-DUAL™ are available in Supplementary Table S3.

Lin’s Rc coefficient revealed a mean acceptable accuracy between the wrist and the chest HR monitor across all subjects, consistent with similar studies (24, 25, 27). Our study, however, is the first evaluation of this specific wearable (Vivosmart 4) and population (24–30, 35). The limit of agreement falls within −16 and +13 and does not reveal any systematic bias or variability. Sartor et al. state that ±15 bpm is a great limit of agreement considering a second-by-second comparison (27) while the present study shows greater accuracy than other comparison studies (24, 25, 58, 59). HR accuracy studies among the general population imply that an increase in HR is possibly associated with increased upper-body movement (24, 32) and types of exercise involving more movement in the limbs (e.g., use or no use of arms on an elliptical trainer), highly affects accuracy among different wrist HR monitors (25, 26). However, some studies do not find the same inaccuracy-error regarding intensity and type of exercise (28, 29). These inconsistencies are likely due to methodology, protocol, and individual differences. In the present study, we observed a great individual variance in accuracy among the subjects included in the evaluation (see Supplementary Figure S4). This finding could be due to errors caused by movement, and in this case, by disability-induced upper-body (uncontrolled) movement and flexion of the wrist. The difference between the handiathletes with the most and least wrist/chest agreement was seen among handiathlete C and F (see Supplementary Figure S4). Their individual disability-specific conditions and diagnoses differed significantly. One lives with a disability prognosticate great hypokinesia (no muscle activity and thus no movement in limbs), and the other with an appearance similar with CP induced dyskinesia (characterized by rapid involuntary and abnormal movement (52)). Thus, it is likely that the disability-specific conditions affecting locomotion result in major subject-specific-agreement and variation (22), coupled with other factors affecting accuracy, as described by Garmin (e.g., non-proper fit on the wrist, physical characteristics and type of exercise, and intensity of the activity (60)). Although the chest HR monitor seems more sensitive to higher intensity levels compared to the Vivosmart 4, most of the time was captured within the same intensity zones (92% agreement). Further, the Vivosmart 4, used among PLWD, does not reach the same accuracy as other wrist-worn HR devices have demonstrated among the general population (e.g., Apple Watch and Mio Fuse (24–26, 29)). Thus, HR accuracy and agreement seems to be highly affected by individual variation and disability characteristics and can, for some individuals, cause measurement errors.

A main strength of this study is the second-by-second comparison of HR-data (27). Additionally, applying the HRR to estimate intensity levels, rather than a standard formula (e.g., % of HR_max) strengthens the validity of each individual (21, 34). Applying the chest-worn monitor as the reference device seems reliable due to a high overall accuracy between chest monitors and golden standard electrocardiograms for HR monitoring (24–30, 35, 36), especially under the circumstances of “real world settings” where other HR validation studies also apply commercial chest worn HR monitors (61–63). However, a limitation of the study may be that no validation studies have examined the Garmin HRM-DUAL™ among the general population nor among PLWD.

The study also faced other limitations. We only managed to recruit five subjects and conducting a total of six evaluations due to limited resource and time and difficulties of recruiting the handiathletes. Some of the reasons for the sparse subsample (n = 5) was 1) few available days (only Sundays) due to not all planned visits were allocated to conduct the evaluation (some visits where scheduled for interviews and observation), 2) difficulties attaching the strap to all subjects due to their many layers of clothes and disabilities, and 3) some subjects refused to partake while the felt uncomfortable getting undressed to get the strap attached. Thus, our data basis is quite small and expected to be selection bias.

Despite applying the most valid methods available for the HRR estimation, a total of 7 h and 19 min was registered below the estimated RHR (<0%HRR) (data not shown). This is why we see 95% CI below zero in Figure 1. This questions 1) the accuracy of the estimated RHR from the Vivosmart 4, 2) the accuracy of the Vivosmart 4 during Team Twin activities, 3) the methods we applied to easily (and accurately) estimate HR responses among PLWD, and 4) the influence of the potentially falsely increased RHR due to the side-effects of drugs among 44% of the handiathletes (Table 2) and the well-known CP-induced increased RHR (64). During a post hoc estimation, we observed that handiathletes who used drugs with potential side-effects on the coronary circulation and pulse obtained significantly higher HR intensity zones than those who have not been prescribed those drugs (data not shown). This elaborates further the questions of HR as a measurement for PA-induced energy expenditure among PLWD with increased use of medications (with potential effects on HR).

A significant limitation of the study is that HR_max could not be directly measured, as no handiathletes reached peak HR. HR_max estimations are acceptable when no direct HR_max measure is feasible (21). Thus, the equation from Fernhall et al. (42) was used to estimate HR_max in relation to the selected intensity thresholds (21, 45). Using more wide-spread equations for HR_max estimation (e.g., Gellish et al. (65) or Fox et al. (66)), the proportion of time spent in higher intensity zones decreased (see Supplementary Table S4). However, apart from the very hard intensity zone, all intensity levels were generally alike, regardless of the method used to estimate HR_max. In general, quantifying intensity zones and interpreting HR response as an assumed function of PA among this population, with the accessible tools, applied methods, and taken approaches, are highly challenging, however, we believe we have optimized the use of the chosen techniques and methods. Ultimately, the level of accuracy will depend on the specific use and activity performed by the user.

Subjects faced significant challenges in operating the Vivosmart 4 due to the incompatibility between its design (tiny and touch screen (39)) and disabilities impairing both fine and gross motor function. This meant that turning on/off the device during sessions relied on runners’ and relatives’ help. Still, remembering to activate the watch was challenging, and much data was not monitored compared to the actual participation rate (Table 2). A larger device with buttons instead of a touchscreen would likely have better accommodated the participants’ motor function. However, a wrist-worn wearable seems more suitable, compatible, and accessible for this population than a chest-worn monitor due to the challenge of attaching it on severely PLWD.

Some handiathletes did not wear the watch precisely as described in the manual (39). During observations at Team Twin training sessions, we noted an improper fit of the watch (too loose/tight and wrong position on the wrist (39)). Sometimes, the PPG lens was dirty—which could impact its accuracy. The placement on either the left or right wrist could also affect the accuracy, although prior studies on the general population did not find accuracy variation between the left and right wrist (25). However, Garmin highlights that “wrist flexing” can prohibit accuracy (60), which inhibits the agreement and accuracy Thus, in a population with hyperreflexia due to CP attaching the HR monitor to the less affected side should be considered.

Garmin also highlights some essential factors to accommodate optimal conditions for the chest-monitor-accuracy (e.g., being wet to maintain electrical connection, correct positioning, warming up, not wearing synthetic fabrics due to the development of static electricity etc (46)). Some of those issues were difficult to accommodate among the PLWD, e.g., optimal chest positioning, due to abnormal posture and contact between monitor strap and running chair-back changing its fit over a session.

The findings from this study are limited to the specific population and activity form of recreational indirect PA regarding the external validity. However, our reflections based on the use of HR-monitors among PLWD, and the methodological challenges can be generalized to other groups of PLWD, with specific conditions that could affect the validity of a HR monitor. Despite that, generalize conditions across various disabilities in relation to this specific topic may be challenging, due to all the different factors affecting HR monitoring. Thus, applying and selecting HR monitors (either wrist- or chest-worn) among population with specific conditions should be considered according to the issues addressed in the study.