Participants enrolled as caregiver singletons or caregiver–care recipient dyads. Eligibility criteria for caregivers were as follows: (1) aged >50 years; (2) is a primary caregiver for a patient with Alzheimer disease (AD) or an AD-related dementia; (3) experiencing stress or strain related to caregiving; (4) the person being cared for had problems with sleep or keeping a consistent routine; (5) experiencing at least mild depression symptoms defined as a score ≥5 on the 9-item Patient Health Questionnaire (PHQ-9); (6) living with their care recipient; and (7) willing to try the app and participate in assessments and calls. Eligibility criteria for care recipients were as follows: (1) aged >60 years; (2) having been diagnosed with AD, vascular dementia, mixed Alzheimer or vascular dementia, or frontotemporal dementia; (3) living with their caregiver; and (4) willing to wear the watch and participate in assessments and calls.

Enrollment and follow-up procedures are presented in Multimedia Appendix 1 [34]. There were 25 caregivers screened, of which 3 did not meet the inclusion criteria and 7 were not interested, leaving 15 eligible or interested caregivers. Note that the requirement of at least mild depression symptoms was added after the study began, based on stakeholder feedback that the program worked or was most appropriate for caregivers with at least some depression symptoms. This modified inclusion criteria allowed us to identify and enroll more caregiver–care recipient dyads. Caregivers enrolled after the change in PHQ-9 eligibility (n=3) reported more depressive symptoms (mean 8.83, SD 5.81) compared to caregivers enrolled before the change (n=12; mean 5.69, SD 5.28). We invited the 15 interested and eligible caregivers to involve their care recipient. A total of 7 of these said they were not interested or did not think the study would be appropriate for their care recipient, and 2 care recipients were ineligible, rendering a sample of 6 care recipients (people with dementia).

The University of Pittsburgh Institutional Review Board approved all procedures (STUDY21060122), and the trial was registered with ClinicalTrials.gov (NCT05309577). Participants provided informed electronic consent (e-consent) in REDCap (Vanderbilt University). To include individuals representative of the population of people with dementia, we allowed individuals who did not have the capacity to give informed consent to enroll, provided they gave verbal assent and proxy consent signed by a legally authorized representative. Proxy consent was described as the process of honoring the care recipient’s values and making medical decisions on their behalf when they can no longer do so themselves. Of the 6 care recipients (people with dementia), 3 (50%) signed their own e-consent in REDCap; the remaining 3 (50%) provided consent by proxy, and their caregiver consented on their behalf. All data were deidentified. The watches and the myRhythmWatch app did not record any identifiable data; they recorded only movement and acceleration intensity. Participants were compensated US $125 for completing study procedures.

This single-arm pilot trial enrolled participants between March 2023 and April 2024 in Pittsburgh, United States. Participants were recruited from a University of Pittsburgh research registry (Pitt+Me [35]). The registry includes more than 400,000 participants who agreed to be considered for research studies at the University of Pittsburgh. Targeted emails describing this clinical trial were sent to members who identified as adult caregivers. The study coordinator would receive a list of interested members, and we completed the screening and consent during the same initial phone call. Self-report and clinical health assessments were conducted with nonblinded research staff at 3 time points: baseline, 6 weeks after the intervention, and 3 months after the intervention (caregivers only). All assessments were completed over the phone or virtually. The completed TREND (Transparent Reporting of Evaluations with Nonrandomized Designs) checklist is provided in Checklist 1 [36].

After enrollment, all 21 participants (caregiver singletons and caregiver–care recipient dyads) received one-on-one orientation to wearing the Apple Watch and using the myRhythmWatch app. Of the 21 participants, 18 (86%) preferred virtual orientation; the remaining 3 (14%) requested an in-person orientation. Participants were provided an Apple Watch Series 8 via the United States Postal Service and asked to wear it over the 6-week trial, except when they were bathing or showering, when we asked them to charge the watch. As needed, we also followed up with participants 2 days after the technology orientation to troubleshoot any issues that arose. Of the 15 caregivers, 11 (73%) also completed pretreatment and posttreatment assessments and were included in the health outcome analysis. Incomplete assessments (4/15, 27% caregivers) were linked to scheduling conflicts and low caregiver availability. There were no statistical differences in age, mood, or insomnia between caregivers who did and did not complete the assessments. However, these findings should be interpreted with caution given the low sample size (Multimedia Appendix 2).

The myRhythmWatch system uses raw accelerometry data from the Apple Watch, as previously described [15-17], to compute standard RAR metrics that are displayed to users on their mobile phones (Figure 1). For technical details on handling raw accelerometer data, refer to the study by Zhang et al [16]. Briefly, the raw triaxial accelerometer data were bandpass filtered, aggregated across axes, and then converted to 30-second epochs. Standard RAR measures are then extracted and presented as standardized scores. Therapists accessed information on participants’ RARs via a Health Insurance Portability and Accountability Act (HIPAA)–compliant clinician dashboard, which displayed participants’ RAR metric data (Figure 2) and graphs including raw accelerometer data plots (Figure 3). Therapists used this information in 6 sessions, 1 per week, each lasting about 30 to 45 minutes. Caregivers facilitated sessions for themselves and for their care recipients; when care recipients were involved, sessions lasted approximately 60 minutes. All 21 participants (caregiver singletons and caregiver–care recipient dyads) received the 6-week intervention.

In the first session, and as needed thereafter, therapists reviewed standardized educational materials about circadian rhythms and health. They also used open-ended questions to understand the relevance of rhythms to the participant’s life or health goals. The rest of session 1, and all subsequent sessions, focused on reviewing findings from the myRhythmWatch app and used behavioral activation techniques to plan activities focused on supporting a healthy rhythm [37]. There was an explicit focus on adhering to 4 Rules for Healthy Rhythms: (1) waking up at the same time every day, (2) engaging in rewarding morning activity (preferably with exposure to sunlight and physical activity), (3) limiting naps to <30 minutes and ending before 4 PM, and (4) maintaining active periods of around 15 to 16 hours per day.

Feasibility was assessed with process outcomes including (1) the proportion of users providing sufficient data for assessing RARs (ie, ≥3 consecutive valid days defined as days with at least 20 hours of data) and (2) the total number of valid days of accelerometer data collected. Program acceptability was measured using the total number of completed health coaching visits. Participants also rated satisfaction with intervention components on a 5-point Likert scale, ranging from “very dissatisfied” (score=1) to “very satisfied” (score=5). Scores ≥4 indicate adequate levels of intervention acceptability.

Initially, the intervention was designed to focus on promoting self-care and reducing caregiving burden. However, in designing the trial, the focus was switched to the circadian-informed, behavioral activation–based approach (described above) that aimed to improve mood and sleep in caregivers. Mood and sleep measures were retained as secondary outcomes in the final trial that included (1) depression symptom severity assessed using the PHQ-9 [38] (scored on a 4-point Likert scale [scores=0-3] and summed to a maximum severity score of 27) and (2) insomnia symptom severity assessed using the Insomnia Severity Index (ISI [39]; scored on a 4-point Likert scale [scores=0-3] and summed to a maximum severity score of 28). The measures were chosen to be consistent with prior research on mood, sleep, and circadian rhythms in dementia caregivers [40].

Secondary outcome measures of mood and sleep were also included. Depression was measured with the 20-item Center for Epidemiological Studies Depression Scale [41]. Item responses range from 0 (rarely or none of the time) to 3 (most of the time), with a scale range of 0 to 60 (higher scores reflect increased depressive symptom severity). Anxiety symptoms were assessed using the 7-item Generalized Anxiety Disorder Scale [42]. Item responses range from 0 (none of the time) to 3 (most of the time), with a scale range of 0 to 21 (higher scores reflect increased anxiety symptom severity). Measures of sleep/circadian functioning were also included. Sleep quality was assessed using the 19-item Pittsburgh Sleep Quality Index [43]. Item responses range from 0 (no difficulty or problem) to 3 (severe difficulty) with a scale range of 0 to 21 (higher scores indicate worse sleep quality). Daytime sleepiness was assessed using the 8-item Epworth Sleepiness Scale [44]. Item responses range from 0 (would never doze) to 3 (high chance of dozing) with a scale range of 0 to 24 (higher scores indicate more excessive daytime sleepiness). Circadian preference was assessed using the 13-item Composite Scale of Morningness [45]. Item responses are time increments that range from 1 to 4 or 5 with a total scale range of 13 to 55 (higher scores indicate a greater degree of morningness).

Feasibility and acceptability outcomes were analyzed descriptively for caregivers and care recipients (people with dementia). We used paired-sample 2-tailed t tests to report preintervention and postintervention change scores in caregivers’ mood and sleep over 6 weeks. Hedges g effect sizes were calculated and interpreted according to established standards (0.3 indicates a small effect, 0.5 indicates a medium effect, and >0.8 indicates a large effect). Caregivers with incomplete assessments (n=4) were excluded from the pre-post analyses. Our sample size was determined (approximately 20 participants) to ensure it met practical resources and timeline constraints for determining feasibility and acceptability. All analyses were conducted using SPSS (version 31; IBM Corp).

Consistent with the known gender distribution of dementia caregivers [46], caregiver participants were mostly women (12/15, 80%) providing for a loved one with dementia (9 caregiver singletons and 6 caregiver–care recipient dyads). In total, 24% (4/15) of participants were Black or African American (Table 1). Of the 21 participants (caregiver singletons and caregiver–care recipient dyads) enrolled in the study, all of them completed the myRhythmWatch intervention.

Hedges g effect size estimates indicate medium to large effect size changes for all outcomes except for anxiety symptom severity (Table 2). Total PHQ-9 scores were reduced by an average of 58% from baseline to week 6. ISI scores were also reduced by an average of 58% from baseline to week 6. Statistically significant improvements were observed in multiple dimensions of sleep including improved sleep quality, less daytime sleepiness, and an increased circadian preference for morningness. To further illustrate key changes observed, we plotted individual and group preintervention and postintervention PHQ-9 (Figure 6) and ISI scores (Figure 7).

Overall, findings suggest that personalizing a behavioral intervention using consumer-based RAR monitoring is feasible and acceptable in older adults experiencing stress related to their role as dementia caregivers. This conclusion is supported by the high and sustained engagement with the myRhythmWatch app observed among caregivers in this sample. There were also significant preintervention and postintervention improvements in mental health, with dementia caregivers reporting fewer depression symptoms, less insomnia, less daytime sleepiness, and more “morningness” at the end of the trial.

Another novel finding is that RAR monitoring using a consumer wearable was feasible in people with dementia, albeit with significant caveats. First, most caregivers (9/15, 60%) did not think the study would be appropriate for their care recipients who had dementia. Second, among the 6 people with dementia who volunteered for the study, engagement was lower and for a shorter term compared with that among caregivers. Thus, it is feasible to use the Apple Watch to perform RAR assessments in people with dementia but not in all patients and typically not over longer periods.

Most prior research using accelerometers has used recordings, over periods of 1 to 2 weeks, as a measurement tool in association studies [4,5]. A few prior studies have demonstrated success measuring RARs using raw accelerometer data from the Apple Watch [15-17]. The present study is novel in that it is the first to demonstrate the feasibility of using this approach for (1) longer-term RAR monitoring and (2) tailoring the delivery of a behavioral intervention. Our findings are also novel in that the myRhythmWatch system’s clinician dashboard provides the first real-world or scalable approach to monitor measures of RAR disturbances in clinical practice.

The approach here also differs notably from prior mobile phone app–based research in this area. As noted in the Introduction section, similar apps have relied on phone use data [21] or step counts [18,19]. An advantage of using raw epoch-by-epoch accelerometer data here is that this allows us to characterize 24-hour behavioral patterns using the same RAR measures that have established health relevance in research [5,13,37]. We observed consistent engagement with the myRhythmWatch app for more than 6 weeks, which compares favorably with other mobile health apps. While approximately 40% of people drop out of mobile health interventions [47,48], all participants in the present study were retained, and caregivers remained engaged with the app after 6 weeks.

Prior studies have shown that people with dementia tend to have more fragmented and unstable RARs [22,23,30]. These results raise the possibility that myRhythmWatch may be useful for long-term monitoring in subgroups of patients with dementia. However, given the relatively lower engagement among people with dementia, future studies should consider integrating even less obtrusive measurement approaches, such as a wearable pin or pendant, as used by Figueiro et al [49]. In contrast, despite the stressors of being a dementia caregiver, nearly all caregivers were consistently able to use this Apple Watch–based system. This suggests that myRhythmWatch may be useful for long-term monitoring for RAR disruptions that are associated with poor health outcomes in caregivers [30-33].

Some limitations and caveats to our interpretations should be noted. First, due to cost restraints and the early-phase nature of this study, we chose a single-arm design to maximize the number of trial participants addressing our primary aims (eg, acceptability of enhancing a behavioral intervention with myRhythmWatch). However, this precludes inference on the extent to which the observed health benefits were due to personalization via real-time RAR monitoring, rather than other aspects of the intervention, nonspecific therapeutic effects, or spontaneous improvements. Second, because our early participants had low or no depressive symptoms, we adjusted our inclusion criteria to include those with at least mild depressive symptoms so that we could study feasibility and acceptability in those experiencing caregiving stress. Third, our sample consisted of older dementia caregivers and people with dementia who were willing or able to use digital technologies. Our findings only speak to the feasibility and acceptability of the myRhythmWatch app among these older adults under caregiving stress or with a serious medical condition (dementia), and results will not necessarily generalize to other subgroups in the broader population. The feasibility and acceptability of the myRhythmWatch app is likely lower among individuals with lower digital literacy. Because the sample included older dementia caregivers, who tend to be predominantly female, future studies in samples with more older men are also warranted [46]. Separate studies will also be needed to confirm feasibility or acceptability in younger people. Finally, due to the small sample size, effect size estimates should be replicated in larger studies.

As discussed above, we have shown that personalizing interventions via real-time RAR monitoring is feasible and acceptable in older dementia caregivers and people with dementia. The main immediate implication is that future large-scale randomized controlled trials, using the myRhythmWatch system to enhance behavioral interventions, are feasible in studies with similar older adult target populations. Personalizing treatments based on RAR monitoring is theoretically relevant not only for enhancing the rhythm-focused program used here but also for several behavioral approaches that depend on monitoring activity patterns such as behavioral activation [50], interpersonal and social rhythm therapy [51], and behavioral sleep interventions [52]. Given the feasibility and acceptability demonstrated in subgroups of older adults in this pilot study, future controlled studies are warranted to determine whether adding RAR monitoring to behavioral interventions improves their efficiency, efficacy, or durability of effect.