Remote activity monitoring for family caregivers of persons living with dementia: a mixed methods, randomized controlled evaluation

Background and contextual factors collected included caregiver and care recipient demographics, relationship between caregiver and care recipient, living situation, and duration of memory loss and caregiving relationship (see Table 1, below).

Primary objective stressors captured the nature and extent of the care recipient’s needs [28]. These included the level of assistance (no help, some help, or a lot of help) the care recipient required for six activities of daily living (ADLs) (α = .89 [34]; and six instrumental activities of daily living (IADLs) (α = .96 [35];. Eight items were used to measure level of care recipient cognitive impairment (e.g., ‘how difficult is it for the person with memory loss to know what day of the week it is?’) on a 5-point scale ranging from ‘can’t do it at all’ to ‘not at all difficult’ (α = .86 [28];. The Revised-Memory and Behavior Problems Checklist was used to measure frequency of 30 common dementia-related behaviors (α = .89 [36];.

An 8-item measure of socioemotional support assessed the help provided to the caregiver by family and friends on a 5-point scale (1 = ‘strongly disagree’ to 5 = ‘strongly agree;’ α = .68) [28]. A checklist of community-based long-term services and supports (LTSS), developed by the authors, further queried the caregiver’s use of paid services (e.g., housekeeping, transportation, respite programs) to support their care recipient. Finally, caregivers were asked if their care recipient currently used any monitoring technology apart from the RAM system such as an emergency call pendant or bracelet.

Self-efficacy was measured by asking caregivers to rate their confidence in achieving eight caregiving tasks on a 5-point scale ranging from ‘very unconfident’ to ‘very confident’ (e.g., ‘do something to keep the person with memory loss as independent as possible’) (α = .86) [37]. The Short Sense of Competence Questionnaire (SSCQ) was used to measure caregiver competence for seven items (e.g., ‘I feel stressed between trying to give to the person with memory loss as well as meet my other responsibilities’) on a 5-point scale ranging from ‘strongly disagree’ to ‘strongly agree’ (α = .74 [38];.

The Zarit Burden Interview was used to measure caregiver distress for 22 items (e.g., ‘do you feel the person with memory loss asks for more help than he/she needs?’) on a 5-point scale ranging from ‘never’ to ‘nearly always’ (α = .93 [39, 40];. Additionally, three items each were used to measure caregiving role captivity (α = .80) and overload (α = .81) (e.g., ‘how often do you wish you were free to lead a life of your own?’) on a 4-point scale ranging from ‘not at all’ to ‘very much’. Finally, depressive symptoms were measured using the 20-item Center for Epidemiological Studies-Depression scale (e.g., ‘during the past week, I was bothered by things that don’t usually bother me;’ on a 4-point response scale ranging from ‘rarely or none of the time’ to ‘most of the time;’ α = .94) [41].

Caregivers in the intervention arm were asked to complete a 22-item review checklist rating the acceptability and utility of the RAM system on a Likert-type scale at 6, 12, and 18 months. Example items include, “The alerts generated by the [RAM system] have helped prevent crises for the person with memory loss” and “I would recommend the [RAM system] to others in a similar situation as the person with memory loss” (on a 5-point scale range from 1 = ‘strongly disagree’ to 5 = ‘strongly agree’). Caregivers were also asked eight open-ended questions about difficulty/ease of use, utility of system alerts and the web-based dashboard, and perceived effects of the RAM system on their caregiving role.

Caregivers in the intervention arm who were administered semi-structured interviews at the conclusion of the study were asked to provide further detail on why they felt the RAM system did or did not reduce caregiving distress or help manage their care recipient’s daily function. They were also asked to provide feedback on the benefits and drawbacks of the system. All interviews were digitally recorded and professionally transcribed. The qualitative interview guide is included in the Online Supplemental Material.

We analyzed open-ended responses from the 6-, 12-, and 18-month follow-up surveys as well as transcripts from the semi-structured interviews. All qualitative data were organized in the software package NVivo (Version 12).

We thematically analyzed the data using Braun and Clarke’s [42] six steps of thematic analysis: (1) familiarization; (2) generation of initial codes; (3) search for themes; (4) review themes; (5) define and name themes; and (6) write up themes. All authors read the data to familiarize and gain a sense of the whole. This supported immersion in the data to enable new insights to emerge and inductively develop categories without imposing preconceived categories [43]. After generating initial codes, authors JF, MN, LM, CR, CP, ZB, and EA compared interpretations and points of divergence to refine and clarify codes. We independently coded a sample of responses to check for consistency in meaning and application of the codebook and illuminate any differing interpretations via individual bias. After team discussions to finalize the codebook, authors JF, MN, LM, CP, ZB, and EA each coded a portion of the qualitative data. We reviewed each other’s coding to ensure completeness and accuracy and add any additional coding. Peer debriefing, referential adequacy, negative case analysis, and clear audit trails enhanced transparency and credibility [44]. Iterative analyses continually seeking interpretation, alternative understandings and linkages led to saturation, whereby the themes were well-described by and fitting with the data [45].

Peer debriefing occurred as experts in the field of dementia care met for extensive discourse to ensure alignment of understanding of themes. We employed referential adequacy by having coders each explore two interviews, which we used to create our initial set of themes. Coders were then assigned a distinct subset of interviews to apply those codes to.

Negative case analysis occurred through the revision of codes and themes based on disconfirming data, upon discovery of disconfirming data through consensus discussions among coders. We likewise updated our codebook to explicitly indicate when individual contexts were inappropriate for the use of a given code. Audit trails were created through the maintenance of extensive records including specific coders, how codes were applied, why a code was applied, and when it was applied, as well as when and how themes were newly created or modified to fit observations in the semi-structured interviews.

To determine whether use of the RAM system improved caregiver outcomes we used latent growth curve modeling. We tested intercept-only, linear, and quadratic unconditional models. We used fit statistics including Akaike Information Criterion (AIC), Bayesian Information Criteria (BIC) and root mean square error of approximation (RMSEA) to identify the best functional form. We created models for each of the six primary outcomes: competence, self-efficacy, burden, role captivity and overload, and depression. Based on available statistical power calculators for growth curve models (http://​saspower.​psychstat.​org) that consider both sample size and the number of data points available (n = 4), our analyses were powered to detect moderate to large changes in our selected outcomes.

Next, we added covariates to the best fitting unconditional model for each outcome. We modeled treatment (RAM system versus control), demographics and context of care variables as time-invariant covariates. Primary stressors identified in our preliminary 6-month effectiveness analysis were modeled as time-varying covariates. Using the results of the qualitative analysis (see Results), we mapped themes onto items measured in the survey where available. We modeled these items as dichotomous, time-varying covariates and added interaction terms for each covariate crossed with treatment. We adjusted for multiple tests using the Holm correction [46].

Caregivers and their surrounding support networks varied in their ability to customize the RAM system so that it suited their needs over time. Douglas (54 years old [y], male [M], 18 month [m] interview) recounted his initial high stress levels: “I kept getting calls and calls, and I’m going ‘which button do I hit?’ And then once we figured it out… I didn’t hear nothing for a while.” Douglas continued that he was able to relax after learning which buttons to press and reconfiguring the notification system for no calls from midnight to 6 am: “cause I needed my sleep too. And then they put [notifications] on an email to me.” Douglas’s perceptions of the system improved after a troubled start.

Multiple participants discussed the learning-intensive initial period with the RAM system. Dawn (57y, female (F), 6 m checklist) shared that “some of the charts were difficult to use at first but I learned how to read them.” Maria (61y, F, 6 m checklist) noted: “I have learned how to use [the RAM system] more effectively throughout the past six months with excellent training (emails, phone calls, webinars) from [the DHT]. It’s easy to use (if you have some familiarity with technology)... Initially there seems to be so much information but you learn.” Other participants moved sensors in response to false alarms and troubleshooted issues themselves. In contrast, Valerie (71y, F, 18 m interview) shared: “I tried and it’s like ‘oh this is too confusing, I’m lost’. It’s not user friendly.” Janice (51y, F, 12 m checklist) similarly noted that “I tried to use it in the beginning but found it too difficult so I gave up.... I felt like I should have been able to figure it out and when I couldn’t, I stopped asking for help.”

Participants varied over the follow-up period in their level of engagement and assistance from the RAM’s DHT. Kathleen (54y, F, 6 m checklist) shared the following:

DHT interactions were always positive and helpful- from troubleshooting to serving as a liaison when we first started with the study. Her overall knowledge of the living situation, and the wisdom added as I worked out a notification plan, meant that other than building-specific challenges, participation ran smoothly.

These experiences and ongoing support contributed to high system rating reviews from Kathleen (18 m interview):

[The DHT] was there when you needed her to be there. She would check in periodically and—because I would check in with her. There were times she didn’t need to check in with me cause she knew exactly where we were. So I just remember checking in with her only a couple of times, but it was well-timed, and if I needed to get a hold of her, she was there.

The DHT provided support and check-ins throughout the process, as well as reacting to acute instances such as replacing monitors or fixing system malfunctions. Joyce (56y, F, 12 m checklist) noted the DHT’s patience, helpfulness, compassion, and availability to answer questions and make system adjustments during multiple interactions.

Not all participants had positive DHT experiences or assistance across the entire follow-up period. Janice (51y, F, 18 m checklist) noted that “[The DHT] was great but I think had a lot on her plate making it difficult for follow-up.” Peg (74y, F, 18 m checklist) similarly expressed: “[The DHT was] very helpful initially. Service seemed to wane a bit as time passed.” These experiences contributed to diminishing expectations and use of the RAM. Donna (64y, F, 6 m checklist) noted: “Training too short. No check-ins.” Julie (50y, F, 18 m checklist) relayed: “We have never gotten an alert. Could use a refresher course perhaps to make sure we are properly utilizing all the [RAM] can do for us.” Participants expressed the need for ongoing training and refresher courses to continue using the system beyond the startup period.

Some care recipients experienced changing reactions to RAM sensors over the follow-up period. Both positive and negative changes in attitudes toward the sensors were reported. For example, Debra (54y, F, 6 m checklist) stated that her husband “doesn’t really want to be monitored...I get the impression, he can’t wait till [the sensors] are gone!!” However, when the same participant was asked about her husband’s perceptions of the sensors in the semi-structured interview, she reported, “I think he quickly forgot about them.” These statements suggest that the care recipient’s initial wariness of the sensors wore off over time, until he no longer paid attention to them. In contrast, Angela (67y, F, 18 m interview) reported that her care recipient eventually became increasingly frustrated with the system over time: “After a while [CR] just saw it as an intrusion, you know, that they were watching him constantly, and why do they have to know how many times I go to the bathroom? And why do they have to know when I open the refrigerator door?” Rather than acclimating to the system, this care recipient found it more intrusive as time went on.

Caregivers discussed how the progression of care recipients’ ADRD influenced which features of the RAM system were or were not beneficial over time. Maria (61y, F, 12 m checklist) mentioned that “the alerts are the most useful tool in the [RAM] system for me right now because my mother’s Alzheimer’s has advanced so that I have to be at her home every day… The day-to-day monitoring from my computer isn’t as crucial but the alerts let me know about the situations of concern.” Theresa (56y, F, 12 m checklist) concurred that the alerts were vital as the disease increased in severity, sharing “with dad, progressively, the alerts are more important than ever.”

As care recipients’ ADRD progressed, some caregivers found that the RAM system was modifiable to better fit their caregiving needs over time. Debra (54y, F, 18 m interview) said that “because things changed for me and for [CR], … we don’t need to have all these things in place. Especially the refrigerator [sensor] and things like that… I said the front door [sensor], I’d like to keep that one.” However, not all caregivers had time available to change the RAM system to better meet their needs or even learn how the system worked. Heidi (67y, F, 18 m interview) also shared how the RAM system was not beneficial as her care recipient’s ADRD progressed. “His disease progressed in a way where that wasn’t particularly helpful. He’s not a wanderer. He still has really good judgment. He’s declined, but in ways different than I anticipated when I enrolled [in the study].”

Participants reported bringing the RAM data to physicians and other medical professionals as evidence of their care recipient’s medical needs. For example, Randy (74y, M, 18 m interview) showed how the care recipient was not sleeping in order to start a care plan: “To be able to have some discussion with a physician about, you know, lack of sleep.... We had something to back up our request for, let’s do something about this business of sleeping.” Caregivers could use the data rather than rely on their care recipients to connect their behaviors to new medical problems that arose over the follow-up period, as Julie (50y, F, 18 m interview) explained, “If we ever saw him going to the bathroom a lot we’d know that he might have some sort of bladder infection or something like that, so that he might not be able to communicate to us.”

Participants could also monitor the change in their care recipient’s habits following medical intervention. Randy, the 74-year-old male who brought the sleeping issues to their physician was able to see that their care recipient, who had been diagnosed with sleep apnea, “sleeps much better with the aid of a CPAP machine. Has not had to get up in the middle of the [night]” (6-month checklist). Amy (58y, F, 18 m interview) was able to monitor improvement in their care recipient’s sleep after they changed medication: “And as a result of being on [the RAM], we could see that the medication allowed her to sleep for longer periods. Prior to that she was up and down every night.” Participants used RAM data to validate their concerns to medical professionals and observe changes in their care recipients’ behavior before and after medical treatment across the 18-month follow-up period.

Changing life situations tended to translate to less need for the RAM system. Shifting life contexts often focused on more care being offered by persons in the care recipients’ lives. For instance, Betty (53y, F, 12 m checklist) shared that her mother is “now in assisted living, so we are using [the RAM] less since they are more involved.” Likewise, Terry (60y, F, 6 m checklist) indicated that they “had live-in help all summer” and had not set-up any alerts, “because our son is living with us now.” Terry (12 m checklist) might have needed the RAM more when their son moved out, but “I was laid-off from my job in September 2016, and I currently work part-time from home, so I am caregiving almost 100% of the time now. So, I have not used [the RAM] at all.” This trend continued through 18 months. Similarly, Debra (54y, F, 18 m interview) reported less time needed for care after stepping away from their job: “But after I had to retire and everything, I mean we’re almost pretty much together. You know. And I pretty much watch him closely.”

One exception to these patterns was when Dorothy (88y, F, 18 m interview) experienced a disruptive health event of their own and was not able to rely on the RAM at that crucial time, “We’d go up to the lake three, four days at a time. We’d go out [doing] other things. I had a heart attack. I was in the hospital for six days. There was never any kind of reaction from that system. We never heard anything, never called, never anything.” In a critical instance of needing the RAM, it did not react and subsequently led to diminished confidence in and use of the RAM.

Caregivers described that their comfort level and trust in the system was only achieved after long-term use and familiarity with the sensors and function of the system. Initially caregivers reported some challenges in explaining the system to the care recipient and family: “And I remember one time after I set it up I went there and she had unplugged it because I think she still didn’t quite grasp what it was doing but now I almost think it’s just more they don’t even really think about it” (Ellen, 59y, F, 18 m interview). In the beginning, caregivers reported not trusting how the system functioned or whether it was functioning appropriately. To better trust the system’s functioning, Theresa said she initially “kept the notepad open and logged in to watch movement/activity at night. I then felt confident that the system WAS watching while I slept...I just needed to know that it was working at the beginning. I would log into my iPad and check movement etc. It was very accurate, so I relied and trusted the system to alert me of changes” (56y, F, 6 m & 18 m checklists). Caregivers described that over time, in-depth understanding of how the RAM system functioned enabled the caregiver to maximize its use. For example, Theresa noted in her semi-structured interview, “If I was doing bookwork or something I felt like I could still track him and so I just felt more confident that I knew what and where he was... So I think it gave me, like I said, those few minutes that I needed to prevent-- hopefully prevent a disaster.” Trust in how the system worked affected how this caregiver responded to the person with dementia’s behavior. Additionally, trusting the system enabled the person with dementia to live independently and at home for longer than the family had anticipated: “It has also built our confidence to leave our loved one in her home and living independently longer than we might have had we not had the system in place so that we know she is safe” (Amy, 58y, F,18 m checklist). Additionally, reliance and trust in the system over time enabled the person with dementia to live at home in their familiar surroundings and was perceived to delay entry into residential long term care settings.

Model fit statistics for unconditional models are reported in Table 2. For role overload, role captivity, and depressive symptoms, the linear growth model provided an optimal balance between fit and parsimony, so this functional form was retained. For all three outcomes, the unconditional models described an average trajectory starting with a relatively low level of role overload, role captivity, and depressive symptoms with slight increases in each of these outcomes over time (for role overload, I = 7.55, p < .001; S = .12, p = .08; for role captivity, I = 6.25, p < .001; S = .30, p < .001; for depressive symptoms, I = 11.03, p < .001; S = 1.12, p < .001). All coefficients reported are unstandardized, and a Holm correction was applied to account for multiple tests across the conditional models.

The conditional growth models (see Table 3) indicated that the RAM treatment had no significant effect on the rate of change in role overload (B = .06, p = .90), role captivity (B = .39, p = .37), or depression (B = .71, p = .62) over time.

Although the qualitative data indicated several dynamic changes in context and situation, many of these domains were not specifically measured in the close-ended survey items. Moderators initially identified in our 6-month analysis as well as the complete longitudinal qualitative analysis here were included (difficulty navigating the home; caregiver lives with care recipient)) [26]. However, no significant interactions were found between the RAM treatment and any of the covariates that emerged as potential moderators of interest from the qualitative findings or our prior 6-month analysis of effectiveness (see Table 3).