A substantial part of the care for people with dementia is provided by informal caregivers such as family, friends, and relatives [1]. Consequently, informal caring not only affects the well-being of the primary caregiver but also changes their roles and interactions with the social environment [2,3]. Although some involved caregivers report the benefits of caring, such as the strengthening and enrichment of mutual relationships, family cohesion, or personal growth [4], others face negative consequences of caring on their physical and mental health or experience increased burden due to the growing dependence of the people with dementia on their environment [2,5]. During the disease process, caregivers of people with dementia are at risk of becoming socially isolated since they might become homebound due to decreased mobility, memory problems, behavioral problems, denial of the disease, or experienced stigma [6-8].

To prevent social isolation and loneliness in caregivers and people with dementia, support is needed after a diagnosis has been made. Previous qualitative studies with carers of people with dementia have shown that there is a postdiagnostic care mismatch between the supply and demand of informal support. The authors advocated introducing early access to tools to improve open communication and facilitate positive social engagement in dementia care networks before carers might become overburdened and while they still have the resources to learn new caregiving skills [9,10]. Existing psychosocial interventions focus on psychoeducation, skill building, and psychotherapeutic counselling based on techniques derived from cognitive behavioral therapy [11-13]. It has been demonstrated that individually tailored, multicomponent interventions offered to both the caregiver and the people with dementia have positive effects on burden, anxiety, and depression [11-13]. However, the effects are generally small to moderate, and available studies are limited in their methodological quality [14].

With the introduction of the Internet and social media in daily life in recent decades, online interventions have become a new avenue for caregiver support. Recent online intervention studies that contained multiple individual tailored elements of psychoeducation, peer support, skills training, and health assessment have demonstrated improved caregiver well-being, including confidence, self-efficacy, and lower levels of depression [15-17]. Therefore, online caregiver support might be an alternative to traditional counselling and support interventions for several reasons. Online tools are always available, regardless of distance, time, and mobility constraints, and provide easily accessible, low-cost support and caregiver empowerment [18]. Using online tools might elicit support-seeking behavior and engagement in social activities to cope better with stressful situations [19] or enhance feelings of competence [20,21] to deal with the challenges faced in the caregiving process. Additionally, through accessibility and widespread reach, online interventions can lower the threshold to involve the caregiver’s social network and either seek or provide social support by increasing openness and positive interaction [9]. This might prevent social isolation and increase feelings of competence in caregivers. Although results from online network interventions are promising [22], the use in informal caregiver social networks is relatively new and not yet studied.

Therefore, the online social support intervention “Inlife” was developed for caregivers of people with dementia and made available in the Netherlands [23]. Inlife intends to help caregivers of people with dementia overcome barriers to seeking help while also removing barriers for loved ones and other individuals involved in the dementia care network to offer help. Using the “Inlife platform,” the primary caregiver is encouraged to invite family, friends, and significant others into their personalized support circles. The functionalities in the intervention include Profile (personal information), Circles (layers of caregivers with different privileges), Helping (general overview to place and receive responses to help requests), Timeline, Calendar, Personal Messages, Care Book, and Compass (information about dementia-related topics). Details of these functionalities are described elsewhere [24]. Inlife is currently being called “Myinlife” after being adopted by a societal partner, the Dutch Alzheimer’s association (Alzheimer Nederland).

The aim was to evaluate the effectiveness of Inlife in a randomized controlled trial (RCT) over a 16-week period [24]. Primary outcomes were feelings of competence and perceived social support. Secondary outcomes were received support, feelings of loneliness, psychological complaints, and quality of life. We hypothesized that, compared with care as usual, use of the Inlife intervention would lead to change in both the primary and secondary outcome measures.

Between June 2016 and June 2017, informal primary caregivers of people with dementia of all subtypes and stages were recruited via online advertisement, newsletters, and social media channels of the Dutch Alzheimer Association; regional dementia community services; and memory clinics or other relevant care institutions. The inclusion criteria were (1) being a primary, informal caregiver of a person diagnosed with dementia of any subtype; (2) having Internet access; and (3) having basic (tablet) computer knowledge as assessed by the researcher. Participants were excluded if they were unavailable for longer than 4 weeks during the study period or had serious health problems incompatible with participation as assessed by the study staff.

After the baseline assessment, participants were randomly assigned to either the intervention or a waiting list control group. The intervention group participated in the Inlife intervention. The waiting list control group received care as usual and was able to start with Inlife after 16 weeks.

Randomization was performed using a computerized sequence generator for block randomization with variable sizes of 4, 6, and 8 (for details, see [24]). The follow-up assessments were completed online using a secure, custom-designed query system.

The study was approved by Ethical Committee of the Faculty of Psychology & Neuroscience of Maastricht University (ERCPN-172_20_03_2016_A1; Dutch Trial Register trial number: NTR6131) More detailed information about the study design is presented elsewhere [24]. The CONSORT eHEALTH checklist is presented as Multimedia Appendix 1.

Participants were screened by telephone to check eligibility. Subsequently, the participating caregivers provided online informed consent. Assessments were collected online at 3 time points: pre-intervention (T₀), 8 weeks (T₁), and 16 weeks (T₂). The 16-week follow-up assessment served as a primary endpoint to compare group effects [24]. Because 2 weeks was deemed an adequate amount of time to become familiar with the platform, 2 weeks after registration on the Inlife platform, participants were contacted by phone to reflect on user experiences. This was done to facilitate engagement, stave off attrition, and resolve any initial queries about the platform.

The intervention group had access to Inlife, an online social support platform for informal caregivers and people with dementia aimed at strengthening positive interactions and social support. All users had a secure username and password combination to access the website and complementary app for smartphones and tablets. Participants could use Inlife in a flexible manner and at their own pace. The platform remained accessible to them after the 16-week study period. Participants in the control group remained on the waiting list for 16 weeks and received care as usual. Care as usual entailed that the participants continued with any routine care they may receive, such as consultations with their general practitioner or dementia case manager. After the 16-week follow-up assessment (T₂), they had the opportunity to register on the Inlife platform.

A short overview of applied self-reported measurement instruments is provided in this section. More details can be found elsewhere [24].

The sample size calculation was based on previous intervention studies of caregivers of people with dementia with the SSCQ as an outcome measure [16,35], using differences between intervention and control groups at follow-up with an assumed mean effect size of Cohen d of 0.5 (medium effect). With an alpha of .05 and power of 80%, we aimed to include 102 primary caregivers (51 participants per group). Allowing for a 20% loss to follow-up, we aimed to enroll a total of 122 caregivers into the study.

Statistical analyses were conducted in SPSS Version 24.0 (IBM Corp, Armonk, NY). Before analysis, data were examined for missing values, outliers, and normality. Potential differences between the intervention and control groups in baseline characteristics and outcome variables at baseline and the 16-week follow-up, which might require adjustment for such differences, were tested using either t tests for continuous variables or Chi-square tests for categorical variables. Since there were missing values, we compared the baseline characteristics of study completers and the participants with missing values. A separate analysis revealed that missingness was related to the sex of the person with dementia. Since missing values were not completely at random, data were analyzed according to intention-to-treat (ITT) principles applying a multiple imputation strategy. We used the Markov chain Monte Carlo method in SPSS to produce 10 data sets. These were subsequently analyzed, and estimates were pooled using the Rubin rule [36]. Subsequently, we performed a per-protocol (PPT) analysis including only the caregivers in the analyses that used Inlife until 16 weeks in the intervention group. A subsequent sensitivity analysis was conducted by contrasting the high active versus the low active Inlife users with the control group. The intervention group was split into high active and low active Inlife users based on the median total number of clicks on the platform [24].

To test the differences in the outcome variables in the intervention group and waiting list control group, we performed linear regression analysis on the imputed data sets, with outcomes from the T₂ assessment as dependent variables (ie, the primary endpoint at 16 weeks; after this period, the waiting list control group was able to start using Inlife). The primary and secondary outcome variables at the T₂ follow-up were included in the model as dependent variables, and group was included as the between-subjects variable. Statistically significant baseline differences between the treatment arms (eg, age of the person with dementia) were included as covariates. Each outcome measure was assessed as a dependent variable in separate analyses. For a variable that was positively skewed, a cubic transformation was applied to better approximate a normal distribution. Subsequently, to test the changes in the primary outcome measures over time, data were analyzed performing a linear mixed model (LMM) on the nonimputed data set. This analysis estimates the fixed effects of the regression slopes, indicating the changes during the intervals (T₀-T₁ and T₁-T₂) in the intervention and waiting list control groups. This procedure allows for modelling the rate of change in the primary outcome variables for the caregivers who did not receive the intervention compared with the caregivers who received the intervention (T₀-T₂). This analysis accounts for within-subject correlations between repeated measures using random (ie, individual-specific) effects, thus accounting for the hierarchical structure in the data (ie, time nested in individuals). Additionally, LMM handles missing values efficiently under the missing at random assumption if variables that are associated with missingness are included in the analyses using maximum likelihood estimates for the missing observations. Hence, it is suitable for ITT analysis [37]. Random effects for the intercept only were specified because likelihood ratio testing revealed that this model fit the data better than adding a random slope or adjusting for correlated residuals. To model the effect of the intervention on the primary outcome variables over time, we entered a group-by-time interaction term as a dummy variable for each of the follow-ups to allow for nonlinear effects. The model was adjusted for baseline differences (eg, age of the person with dementia) and associations with missingness (eg, sex of the person with dementia). All tests were 2-tailed with an alpha level of .05.

A total of 475 caregivers were approached to participate in the study. In total, 379 people were excluded: 124 people were excluded due to ineligibility, and 255 people were excluded due to the fact that they declined to participate. Subsequently, 96 informal caregivers who signed informed consent were included. In total, 96 caregivers signed informed consent and were randomly assigned to either the Inlife intervention group (n=48) or the waiting list control group (n=48). Reasons for declining participation are described elsewhere [24]. Of the 96 randomized participants, 89 completed the 16-week follow up (T₂). Figure 1 depicts the flowchart of study participation. The baseline characteristics for completers and noncompleters did not differ significantly. The baseline characteristics of the participants are presented in Table 1. For the majority of variables, there were no significant differences between the groups, except for age of the person with dementia (t₉₄=–2.05, P=.04) and baseline scores on the SSCQ (t₉₃=–2.65, P=.01) and ICECAP-O (t₉₃=–2.81, P=.006), which were significantly different. Therefore, these variables were included as covariates in the analysis. The high active Inlife users had a larger number of circle members in their network (mean 9.4, SD 5.2) compared with the low active users (mean 3.3, SD 3.7).

Primary and secondary outcome measures were compared between groups (intervention vs waiting list control) after 16 weeks (T₂). Table 2 shows the results of the ITT analysis. Overall, no significant effects in favor of the intervention group compared with the control group were found. The PPT analysis also did not yield any significant effects. Considering our heterogeneous group, we performed a post hoc sensitivity analysis splitting the data into low active and high active users (Multimedia Appendix 2), caregiver relationship status (spouse vs children or others at distance), and living situation of the person with dementia (home vs institution). This analysis revealed no significant differences between the caregiver groups at T₂ (see Multimedia Appendix 2).

For the primary outcome variables, we explored the rate of change over time, as shown in Table 3. For caregiver feelings of competence (SSCQ), the analysis demonstrated a significant group difference at baseline, but no significant overall interaction between group and time was found, indicating that change over time in caregiver feelings of competence was not explained by the intervention (see Multimedia Appendix 3). Similarly, the caregiver perceived support (MPSSS) analysis revealed no overall interaction between group and time, indicating that change over time in caregiver-perceived social support was not explained by the intervention (see Multimedia Appendix 4).

This RCT evaluated Inlife, an online social support intervention for caregivers of people with dementia. No significant improvements in the primary or secondary outcome variables were demonstrated for the intervention group relative to the control group. Additional PPT and sensitivity analyses revealed no beneficial results for the high active Inlife users or specific subgroups of caregivers (ie, spouse vs children or community dwelling vs institutionalized). However, the results indicated that users in general were more active when they had a larger number of people in their Inlife network. Furthermore, active users tended to have slightly longer care duration.

A first important strength of Inlife includes its development through co-design with potential users and its feasibility testing in a pilot evaluation, as recommended by the Medical Research Council framework [45]. This improved the usability and face validity of the platform. Second, the effects of Inlife were evaluated with a robust research design and statistical approach via an RCT that applied both ITT analysis and PPT analysis.

However, this study also has several limitations (in addition to the issues relating to the study's RCT design and retrospective measurements discussed in the section “Considering Innovative Research Designs”). First, the waiting list control design of the study might have affected group differences, as the waiting list control group could have had a longer anticipatory experience than the intervention group. However, eventually, the design enabled all interested caregivers to gain access to the Inlife platform. Second, this study’s sample was heterogeneous in nature, consisting of both spousal caregivers and children of people with dementia living either in the community or in care institutions. We selected this broad population considering the exploratory nature of our study to increase the generalizability of our findings to the general population. However, the power of the present study was insufficient to reveal the effectiveness of Inlife for specific subsamples in separate analyses. This was because we were unable to recruit the intended (n=122) number of caregivers. Finally, although participants were recruited on a national level, they may reflect a distinct subgroup from the general population. For instance, the online nature of the study inevitably resulted in a highly educated sample with a relatively high computer literacy and with highly motivated individuals who potentially already had a special interest in online tools [46,47]. This recruitment method that relied on self-selection probably introduced a selection bias, in which individuals who were more motivated and proficient regarding the use of digital tools than average were sampled, potentially resulting in more positive effects than would be found in a more representative sample. However, given the lack of positive results in this study, the potential impact of this selection bias appears to be minimal.

First, future studies should determine methods to identify practical benefits, such as with qualitative research methods or momentary assessments in the flow of daily life. Efforts should be undertaken to develop research methods that can identify benefits in daily life that are ecologically valid for caregivers of people with dementia. Second, future studies should incorporate contextual factors, such as organizational implementation determinants and available social capital into the intervention design and implementation, to facilitate uptake and make use of the benefits of human interaction. It would be worthwhile to include other Inlife circle members and the people with dementia in the evaluation. Considering our promising qualitative findings in a caregiver subsample [38], we suggest adding more extensive qualitative research methods to gain more insight into the circumstances and factors that are required to make Inlife use effective. This might also enable tailoring the Inlife platform to individual needs by integrating persuasive design features, such as by providing personalized functionalities and tailored notifications that are relevant to the individual caregivers’ needs and are specific to their current stage in the caregiver process [48]. In this way, caregivers can become acquainted with the opportunities of the Inlife platform in accordance with their own needs and at their own pace.

The present RCT demonstrated no significant effects of Inlife on feelings of caregiver competence, social support, and measures of caregiver well-being. Nevertheless, online tools such as Inlife show promise to facilitate the care process in daily life, though researchers should be modest regarding their effectiveness in terms of quantitative measures of well-being and quality of life. Future eHealth studies should (1) exploit the power of human interaction in eHealth and facilitate use of the user’s social capital, (2) apply extensive qualitative process evaluations to unravel beneficial effects for specific subgroups of caregivers and gain insight into potential barriers for implementation in clinical practice, and (3) from the start of the intervention’s development, carefully consider how interventions should be implemented by including contextual factors into the design and evaluation process. Applying these lessons can help researchers develop eHealth interventions for caregivers of people with dementia (such as Inlife) that are better suited to both carer needs and their wider implementation contexts.