This randomized controlled trial was carried out between 2014 and 2016 in the Netherlands. The PiB program was compared to a waiting-list control group receiving usual care. Following the waiting-list period participants were offered the opportunity to follow the PiB program. The Medical Ethics Committee of the Maastricht University Medical Center+ (MUMC+) approved this study (#12-4-059) and the study was registered in the Dutch trial register (NTR4748). The study protocol and supporting SPIRIT checklist are available [23].

From September 2014 to December 2015, family caregivers of people with mild dementia of all subtypes (Clinical Dementia Rating, score 0.5-1) [24] were recruited from memory clinics (MUMC+, Elkerliek Hospital Helmond, Catharina Hospital Eindhoven) and ambulatory mental health clinics (Virenze-RIAGG Maastricht, MET ggz Roermond) in the south of the Netherlands. In addition, caregivers were informed about the trial via caregiver support services, and the website of the Dutch Alzheimer Association. Caregivers were included if they had access to the internet at home, had basic computer skills, and provided written informed consent. Potential participants with insufficient cognitive abilities to engage in the online self-management program, who were overburdened or with severe health problems as determined by study staff, or who cared for people with dementia caused by HIV, acquired brain impairment, Down syndrome, chorea associated with Huntington disease, or alcohol abuse were excluded from participation. Inclusion and exclusion was based on the clinical judgment of the referrer, based on their experience with the target group. Both spouses and other caregivers (eg, children) could be included, as long as they met the criteria above and were >18 years. Details on the recruitment procedure are described in the study protocol [23].

Following the baseline assessment, participants were randomly assigned to either the PiB program or the waiting-list control group receiving usual care by the first author. Assignment was carried out using a computerized random-number generator for block randomization with variable sizes of 4, 6, and 8. An independent research assistant who was blinded to the allocation of the treatment conducted the postintervention assessments. It was not possible to blind the participants because of obvious differences between the interventions in content (PiB is a multicomponent intervention combining psycho-education, movie clips, assignments, and change plans and usual care often consists of psycho-education) and mode of delivery (PiB blends face-to-face contact with online modules and usual care often consists of face-to-face contact only).

For this study, self-reported data from the baseline visit (T₀) and after 8 weeks (T₁) were compared. These data were collected in writing by an independent research assistant who was blinded to the treatment, separately from the coach visits.

The primary proximal outcome was caregiver self-efficacy and primary distal outcome was depressive symptoms. Caregiver self-efficacy was measured with The Caregiver Self-Efficacy Scale (CSES) [27], measuring care management self-efficacy (4 items) and service use self-efficacy (5 items). Care management self-efficacy scores theoretically range from 4-40 and service use self-efficacy from 5-50. Higher scores on the CSES indicate higher levels of self-efficacy. The 20-item Centre for Epidemiological Studies Depression Scale (CES-D) [28] was used to measure depressive symptoms. Total scores range from 0-60; where higher scores indicate more symptoms. Secondary outcomes were mastery, psychological complaints (anxiety and perceived stress), and quality of life. Mastery was measured with the 7-item Pearlin Mastery Scale (PMS) [29]. The total score ranges from 7-35; where higher scores indicate higher levels of mastery. The 7-item Hospital and Anxiety Depression Scale-Anxiety (HADS-A) [30] rates symptoms of anxiety. Scores theoretically range from 0-21 with higher scores indicating more symptoms. Quality of life was measured on five attributes with the Investigating Choice Experiments for the Preferences of Older People CAPability measure for Older people (ICECAP-O) [31]. This index value indicates how good or bad the average person aged 65 or older considers a given state to be, for instance attributing to “attachment” (love and friendship) and “control” (independence). The value system for the 1024 (4⁵=1024) possible states uses a best-worst scaling valuation method, providing a single summary score, anchored at zero (“no capability”) and 1.0 (“full capability”) [32].

Demographics were obtained (sex, age, relationship to care recipient, level of education, sharing household, and care intensity in years). The Global Deterioration Scale (GDS) [33] measured dementia severity with the caregiver as the informant. The possible modifying effects of the following variables were measures. Quality of the relationship was measured using 4 self-rating items of the University of Southern California Longitudinal Study of Three-Generation Families measures of positive affect [34]. The 12-item Emotional instability domain of the NEO Five Factory Inventory (NEO-FFI) [35], was used to identifying individuals who are prone to psychological distress, by assessing 6 traits: anxiety, angry hostility, depression, self-consciousness, impulsiveness, and vulnerability. Scores ranged from 0-24; where higher scorers are likely to be sensitive, emotional, and more prone to experiencing feelings that are upsetting.

We aimed to enroll 80 participants (40 participants per group), based on previous online intervention studies in caregivers of people with dementia with the CSES as outcome measure, on the basis of repeated measures, within-between interaction with a mean effect size of 0.2 [36], assuming an alpha of .05, a power of 85%, and 25% loss to follow-up.

Prior to the analysis, data were checked for missing values, outliers, and normality. Possible differences between the study groups’ baseline characteristics were tested with t tests for continuous variables and chi-square tests for categorical variables. Nonparametric tests (eg, Mann-Whitney U Wilcoxon test) were used when necessary in case of nonnormality.

To examine the differences between outcomes for the intervention and the waiting-list control group during the intervention period, an analysis of covariance (ANCOVA) was conducted with outcome at post intervention as the dependent variable, intervention (PiB program, waiting-list control group) as the between-subjects variable and per outcome its baseline value, age, sex, emotional instability, quality of the relationship, educational level, and relationship to the care recipient as covariates. If significant, the intergroup effect size was calculated according to Cohen d. Effect sizes of 0.2 were considered small, 0.5 considered medium and 0.8 was considered high [37]. IBM SPSS statistics 22.0 for Macintosh was used and all tests of significance were two-tailed with alpha set at .05 and reported mean change.

As recommended by the Medical Research Council (MRC) Framework, a stepwise approach was adopted to explore potential user needs, followed by a pilot evaluation to test the feasibility of the intervention and the measurement tools prior to the effect evaluation. The iterative development and pilot evaluation of PiB as recommended by the MRC framework is described elsewhere [25]. The burden of the intervention was assessed in a process evaluation. Further, results were disseminated to study participants by means of a newsletter and PhD thesis.

A total of 163 caregivers expressed an interest to participate. See Figure 3 for the study flowchart, the details of which are described elsewhere [38]. Table 1 lists the baseline data for the included caregivers (N=81).

Between-group comparisons revealed no significant differences in demographics and main outcome measures at baseline. Care recipients of the included caregivers were 73.9 years old (SD 8.2), diagnosed with Mild Cognitive Impairment (MCI; 12/81, 15%), Alzheimer’s Disease (AD; 33/81, 41%), or other dementias (36/81, 44%). Dementia severity was rated as preclinical memory decline (55/81, 68%), mild dementia (24/81, 30%), or moderate dementia (2/81, 2%) on the GDS. At T₁, 13 caregivers were lost to follow-up. The completers did not differ from noncompleters at baseline in terms of age (t₇₉=0.19; P=.851), relationship to the care recipient (χ²₁=1.39; P=.238), same household as care recipient (χ²₁=0.82; P=.665), care intensity in years (U=377.5; P=.781), sex (χ²₁=2.80; P=.094), education (χ²₁=1.20; P=.550), self-efficacy service use (t₇₉=0.53; P=.599) care management (t₇₉=1.36; P=.177), depression (U=280.0; P=.266), stress (t₇₉=0.25, P=.806), anxiety (U=372.0; P=.497), mastery (t₇₉=–1.18; P=.253), and quality of life (U=775.0; P=.956).

The effects were compared between groups (intervention and waiting-list control) after 8 weeks. Table 2 shows the results of the ANCOVA at T₁ on self-efficacy (care management and service use), depression, mastery, perceived stress, anxiety, and quality of life. After controlling for age, sex, emotional instability, and quality of the relationship, significant effects in favor of the intervention group were found for self-efficacy care management (F_1,60=10.37; P=.002, d=0.85), and self-efficacy service use (F_1,60=11.47; P=.001; d=0.93), but not for depression (F_1,60=1.13; P=.293). Significant effects in favor of the intervention group were also demonstrated for mastery (F_1,60=12.66; P=.001; d=0.94), and quality of life (F_1,60=4.83; P=0.032; d=0.58), but not for perceived stress (F_1,60=3.40; P=0.071), and anxiety (F_1,60=0.80; P=.374).

This randomized controlled study evaluated the first blended-care intervention for caregivers of people with early-stage dementia developed together with potential users, following the MRC Framework, and demonstrated a significant improvement in care management self-efficacy, service use self-efficacy, mastery, and quality of life after receiving the PiB intervention; compared to a waiting-list control group receiving care as usual. Effect sizes were medium (>0.5) for quality of life to high (>0.8) for self-efficacy and mastery. No differences between groups were demonstrated for caregiver depression, anxiety, and perceived stress.

Results on caregiver self-efficacy, mastery, and quality of life are in line with previous results in an uncontrolled study [25] and results of previous eHealth interventions for dementia caregivers [22]. Furthermore, the results of the present study fit the Stress and Coping paradigm by Lazarus and Folkman [8] and the Social Learning theory by Bandura [9], suggesting that taking charge of the changes in one’s life increases self-efficacy and general wellbeing. Learning to positively manage life with dementia instead of managing the dementia itself in a self-management program may have facilitated caregivers’ adaptation to their new caregiving role. The program’s focus on enhancing positive, intact experiences that are tailored to the individual caregiver’s situation could explain the positive effects on caregiver self-efficacy [11]. In addition, the relationship between the participant and the coach may have influenced the outcomes. The process evaluation of the present study showed that both participants and coaches mentioned that their relationship with each other had deepened [38], which was also demonstrated in a previous blended-care intervention for depression [18]. The opportunity to reflect on one’s feelings anonymously in one’s personal safe environment is easier than face-to-face, but the face-to-face contact increased caregiver openness, and therefore coach empathy with their situation [19]. However, we expected that higher levels of wellbeing or quality of life could be the result of a decrease in stress [8,9], which could not be derived from the results of the present study. It is conceivable that interventions aimed at the early stages may not be capable to decrease burden and stress, as these are relatively low during the early stages [7], leaving little room for improvement. Previous caregiver interventions demonstrating positive effects on burden and stress were not specifically aimed at early-stages of dementia [20,39-41]. The process evaluation also revealed that the intervention period and dose varied between participants. Moreover, the discussion forum was not used because caregivers mentioned that sharing their story felt like a betrayal to the care recipient and reading about other people’s “misery” was considered undesirable [38]. These process characteristics may have influenced the intervention effectiveness [42]. Future follow up of PiB effects could clarify if an increase in self-efficacy results in a decrease or prevention of increased stress and depression on the long term.

High face validity was demonstrated as the program was evaluated in multiple institutions with multiple coaches of different backgrounds. Development together with the potential users and a pilot evaluation following the MRC Framework may have increased its effectiveness.

The waiting-list period may have affected the differences in outcomes between both groups. The effects of waiting are highly variable and depend on the characteristics of the sample and of the trial [43]. However, this design allowed all potentially interested participants to participate in the intervention program, which may have increased their motivation to participate given that usual care for mild dementia caregivers often either does not include counseling or includes only infrequent counseling [44]. Furthermore, the waiting-list group was not deprived of usual care. An alternative would be a pseudo-intervention in which only psycho-education or only attention of the coach is provided, but the aim of this study was not to evaluate merely the online aspect of the intervention, but the effect of the blended-care intervention of which psycho-education and face-to-face contacts are integral parts.

Intention-to-treat analyses was not fully possible, as intervention noncompleters refused to participate in further assessments. However, we did include participants that were not completely compliant (completed only 2, 3 or no modules at all) in the analyses [38]. Drop-out was higher in the intervention group compared to the control group, which could have resulted in inflated effect sizes. However, selective drop-out was not demonstrated as completers did not differ from noncompleters at baseline. Often mentioned reasons for drop-out were no need for help or refusal by the care recipient, which was demonstrated previously as reasons of nonuse of formal services [45,46]. Furthermore, a higher rate of drop-out in the intervention group has previously been reported. Previous randomized controlled trials even controlled for any possible loss of power beforehand by increasing the sample of the intervention group. Nevertheless, the current effect sizes should be interpreted with caution. Although the power of our group was not jeopardized based on our power calculation, future studies could consider controlling for a higher rate of drop-out in the intervention group to prevent loss of power.

Our sample was not limited to memory clinics only, but the included participants may represent a subgroup of all dementia caregivers in the early stages. Caregivers in the early stages often decline formal care and it is conceivable that many were not familiar with the care parties involved in recruitment and were therefore overlooked in this study [45,46]. This could have resulted in a highly motivated sample more open to support [47]. Furthermore, only computer-literate caregivers could be included, which represents only around 59% of dementia caregivers [48]. However, seniors’ use of internet is expected to rise in the near future [49], increasing the accessibility of PiB.

Future research could consider combining all resources used during the intervention period with the intervention costs and outcomes in a cost-consequence analysis to aid decision makers. Furthermore, future research should evaluate sustainability of improvements at long-term follow-up. The higher rate of drop-out in the intervention group showed that this group feels overwhelmed but is perhaps most in need of the intervention. Some eHealth interventions show dropout rates of up to 80% [50-52] and therefore suggest blending face-to-face contacts with online modules, like the PiB program, to prevent these high drop-out rates. We found a relatively high response and participation rate [18], indicating that there is a need for at least having the option to choose for this type of caregiver support.

In conclusion, this study showed that a blended care self-management program for dementia caregivers in the early stages is effective in increasing caregiver self-efficacy, mastery, and quality of life on the short-term. The program could provide accessible care to future generations of caregivers of people with early-stage dementia and strengthen the primary caregivers.