The PflegeTab study was designed as a two-arm prospective longitudinal cRCT. It took place from April 2016 to May 2017 and was carried out in ten nursing homes located in Berlin, Germany. The randomisation was performed at nursing home level to avoid contamination across groups (cluster-randomisation, parallel design) and was stratified according to total number of residents in each unit. Within each of the five strata, a member of the research team randomly assigned the units to the intervention group (TBI, five nursing homes) and to the active control group (CAS, five nursing homes) using opaque sealed envelopes (1:1 randomisation). Assessments of primary and secondary outcomes were conducted before the intervention (baseline) and immediately after the eight-week intervention (post intervention). Additionally, EMA were continuously collected in both groups during the intervention phase. Study assessors, participants and staff members of each unit were left blinded to the allocation until the collection of baseline data was completed. Effective blinding was not possible during and after the intervention, as TBI administers received tablet computers and training. The study was conducted and reported in accordance with the CONSORT guidelines for cRCT and was approved by the Ethics Committee of the Medical University of Berlin before enrolment of study participants (Charité – Universitätsmedizin Berlin; EA1/013/16). The trial was registered retrospectively during the recruitment phase with the ISRCTN registry (Trial registration number: ISRCTN98947160).
Participants and recruitment
All participants were long-term residents from the included nursing homes. In nine participating facilities, all residents with dementia were screened for eligibility. The remaining facility had a special dementia unit, in this case only PWD from this unit were included in the further recruitment process. In a first step, legal guardians of potential participants were contacted via telephone. Comprehensive verbal and written information was provided to guardians about the research project and the trial. After consent was obtained from legal guardians, PWD were approached and also asked to give consent. In the few cases of PWD without legal guardians, PWD were approached directly by members of the study team. We took great care to ensure the recruitment procedure was conducted in a comprehensive manner appropriate for PWD. PWD and guardians were thoroughly informed about each part of the trial and all written study information was provided in plain language. Formal inclusion criteria for participants were defined as dementia diagnosis: ICD-10 F00-F03, G30, G31.0, G31.82 and/or cognitive impairment: Mini-Mental State Examination (MMSE)-score of less than 24 points. Exclusion criteria were other mental and behavioural disorders: ICD-10 Diagnosis F10-29 (except F10.1, F17.1, F17.2), and short-term residency: nursing home residency for less than four weeks at baseline.
Intervention Group Nursing Homes
The multicomponent TBI comprised a total of seven basic applications specifically developed to engage PWD in stimulating and motivating activities. The TBI was designed (a) to target cognitive and functional abilities and (b) to support emotional regulation. These domains were selected based on the results of a pilot study (43) and existing research on suitable NPI for PWD. The individual components of the TBI were developed within a participatory and iterative design framework. Each application was designed for older and inexperienced tablet users with impaired cognitive, sensory and/or motor functioning. Several pretests were conducted with PWD, family and formal caregivers. As far as possible, their requirements were implemented in the further design process. Four applications (Quiz, Spelling game, Show me, Move me) targeted cognitive and functional abilities. Task difficulty within these applications was adapted individually based on participant’s task performance: exercises became more difficult as participant’s performance improved and vice versa, for details see Cha et al. (44). Two further applications (Interactive Cat, Colour and Sound game) were designed to support emotional self-regulation. A Picture Gallery addressed both cognitive and emotional aspects within the scope of life story work. Task difficulty was not adapted for the applications Interactive Cat, Colour and Sound game and the Picture Gallery, as these were mainly designed to enhance communication and promote quality of life. An application for communication with family members by means of videoconferencing was also included.
A trained member of the occupational therapy staff accompanied each intervention session. Training workshops were conducted prior to the intervention period to ensure staff members could administer the TBI correctly. The occupational therapy staff members were instructed to guide participants throughout each session and provide assistance whenever needed. Within each TBI session, several activities were selected by the occupational therapists according to the participants’ current preferences and needs. Instructions for each activity were provided both visually and auditory via tablet. After completing a task, participants also received motivational visual and auditory feedback via tablet. The main purpose of the intervention sessions was to engage PWD in a stimulating activity and to provide a positive and enjoyable experience. Therefore, the provided feedback was based on user interactions rather than user performance, meaning that every interaction with the tablet was rewarded, regardless of whether a user action was correct or not. To avoid unpleasant experiences, no direct error feedback was given. If a given response was not correct, the PWD were simply encouraged to try again. The TBI was executed on Apple iPads version Air 2 (Model A1567) and the PflegeTab application was programmed in Swift.
Active Control Group Nursing Homes
According to the study protocol, participants from the five CAS nursing homes were to receive the same amount of individual CAS as participants from the five TBI facilities. A member of the occupational therapy staff also accompanied the individual CAS in this group. No further specifications were made about the nature of the activities, except that no ICT devices were to be involved. Staff members were asked to document the activities conducted in each session in a logbook.
Members of the occupational therapy staff in both the TBI and CAS group were instructed to engage participants in three 30-minute individual activity sessions per week over a period of eight weeks, resulting in a planned goal criterion of 24 activity sessions per participant. A two-hour staff training session was conducted on-site in each TBI facility and a written manual with crucial information for the implementation was provided. Additionally, a support hotline was manned permanently during office hours in case any problems or malfunctions occurred. Two trained research assistants visited each TBI and CAS nursing home and collected informant and self-rated outcome data at baseline and after eight weeks on site. Informant data on participants was assessed from members of the nursing home staff (care professionals) who had worked with the participant on a regular basis and thus knew them well. None of the staff informants participated in the CAS or TBI sessions. Additionally, EMA were collected before and after each activity session. These brief assessments were conducted by the occupational therapist and recorded via tablet in the TBI group and on paper in the CAS group, respectively.
Screening Variable: Cognitive Status
Cognitive status was measured using the MMSE (45). The MMSE is comprised of 30 questions on different aspects of mental functioning, such as orientation, memory, and ability to understand verbal instructions. Each correct answer is scored with one point resulting in a maximum score of 30 points. In the present study, internal consistency of the MMSE was excellent with Cronbach’s α = .91.
The primary outcome apathy was rated by professional caregivers with the informant version of the Apathy Evaluation Scale (AES-I) (46) at baseline and after eight weeks. The AES-I consists of 18 items rated on a 4-point Likert scale ranging from 1 = not at all characteristic to 4 = very characteristic. A sum score is computed for each item; higher scores reflect higher levels of apathy. In the present study, internal consistency was excellent with Cronbach’s α = .91. The subscale Apathy of the Neuropsychiatric Inventory – Nursing Home Version (NPI-NH) (47) was used to determine the convergent validity of the main outcome scale AES-I (48). In terms of convergent validity, in our sample, correlations between the NPI-PH subscale Apathy and the AES-I scores were at moderate levels (Spearman’s ρ = .52).
Quality of Life
Informant reports of dementia-related quality of life were assessed with the QUALIDEM scale (49). QUALIDEM consists of 37 items belonging to nine subscales. All items are rated on a 4-point Likert scale ranging from 0 = never to 3 = frequently. Scores are computed for each subscale. Although the original QUALIDEM does not include a total score, previous studies have created a comprehensive QUALIDEM score by summing the scores of the nine subscales, see (50) for an example. In accordance with this procedure, we also computed a total score for the QUALIDEM scale. Additionally, self-rated quality of life was measured with the Quality of Life in Alzheimer's Disease (QOL-AD) questionnaire (51) from PWD who were able to respond to the questions. Within the 13-item QOL-AD, participants are asked to rate different aspects of their lives on a 4-point Likert scale ranging from 1 = poor to 4 = excellent. The total score is the sum of all 13 items. Higher scores reflect higher quality of life levels in both measures. Internal consistency of both measures of quality of life was good with Cronbach’s α = .83 for QOL-AD and Cronbach’s α = .87 for QUALIDEM.
The informant-based NPI-NH questionnaire was used to measure neuropsychiatric symptoms (47). The NPI-NH evaluates 12 neuropsychiatric symptoms commonly observed in PWD using standardized interview questions. Professional caregivers are asked to rate the frequency and severity of each neuropsychiatric symptom. Scores are computed for each symptom by multiplying severity (1 = mild – 3 = severe) by frequency (1 = rarely – 4 = very often). A total NPI-NH score is computed by adding up the individual symptom scores. Higher scores represent higher degrees of neuropsychiatric symptoms. In the present study, internal consistency of NPI-NH was Cronbach’s α = .73. We assessed the intake of psychotropic medications as a further indicator of neuropsychiatric symptoms (52). The total number of psychotropic medications were recorded, as well as presence of antidementia agents (0 = not present, 1 = present) and neuroleptic agents (0 = not present, 1 = present).
Depressive symptoms were measured using the Geriatric Depression Scale (GDS). The GDS is a 15-item questionnaire in a yes/no format. Higher total scores indicate a higher risk of depression (53). In the present study sample, internal consistency of GDS was Cronbach’s α = .68.
Ecological Momentary Assessments of Quality of Life
We used an eight-item version of the QUALIDEM scale to conduct EMA of quality of life before and after each activity session. This short version consists of eight bipolar items based on items from the original QUALIDEM scale. The bipolar items belong to four QUALIDEM subscales (two items per scale) and address the following behaviours: body language, communication, happiness, anxiety, companionship, general satisfaction, restlessness, sadness. Each item was rated on a seven-point Likert scale by a staff informant immediately before and immediately after every single activity session in both the TBI and CAS group. Ratings of the QUALIDEM short version were associated with ratings on the original QUALIDEM scale and showed good internal consistency with Cronbach’s α with values ranging from .88 to .93 across sessions. Additional psychometric properties of the QUALIDEM short version have been published elsewhere (54).
Further covariates included the variables age, gender, functional status assessed with Barthel Index (BI) (55), and dementia stage measured with the Functional Assessment Screening Tool (FAST) (56).
Sample size calculation
A sample size estimate was calculated with G-Power (Version 3.1; test family: two-sample t-test). Based on previous findings of Hoe et al. (57) , the sample size calculation was conducted for expected differences in QOL-AD scores. Although the primary outcome in the present study was apathy, previous research suggests that the effects of NPI on quality of life and apathy are comparable (7, 8). Furthermore, the study by Hoe et al. was conducted under similar conditions to those in our study. The final sample size estimate was N = 240 PWD (i.e., 120 per group). This calculation was based on a significance level of 5% (two-sided), 80% power, a medium effect size of Cohen’s d = .5, and an expected attrition rate of 20% (57). Taking the nested structure of the data into account (i.e., PWD clustered in ten nursing homes), we anticipated a small intracluster correlation between nursing homes with an intraclass correlation coefficient (ICC) of ICC = .005 (58).
Descriptive statistics of baseline data are reported as means, standard deviations and score ranges for continuous variables, and as absolute and relative frequencies for nominal or ordinal variables. Cronbach’s α was calculated to determine the internal consistency and reliability of the outcome scales at baseline. Nonparametric Spearman correlations were computed for preliminary analyses of relationships between the primary outcome apathy and other clinical study variables. We used an Intention-To-Treat (ITT) approach for analyses of primary and secondary outcomes, meaning that all available data was included in the analyses regardless of loss to post intervention follow-up while assuming dropouts were missing at random (MAR). Multiple data imputation based on chained equations was performed to account for missing data using ten imputed data sets. The predictive mean matching (PMM) procedure was applied to ensure that imputed values were plausible. With PMM, the imputed value always matches one of the observed values of a participant without missing and a similar estimated value from the imputation model as the participant with missing. A random seed number (123) was set to allow reproducibility. Imputation was performed at item-level for primary and secondary outcome measures and covariates, scale scores were then computed. The number of scale scores including imputed data at item-level for each outcome were: AES-I (n = 28; 17%), QOL-AD (n =71; 44%), QUALIDEM (n = 28; 17%), GDS (n = 102; 64%), NPI-PH (n = 28; 17%), MMSE (n = 74; 46%) and FAST (n = 31; 19%). The following variables were used for the imputation process: all individual scale items, age, gender, group (TBI vs. CAS), years of education, nursing home and medication.
Linear mixed-effects models (LMM) fit by Restricted Maximum Likelihood Estimation (REML) were applied to analyse group differences for the primary and secondary outcomes. Change scores served as dependent variables and were computed for each outcome by subtracting baseline scores from post intervention scores. Baseline outcome measures were included as fixed covariates in each model and a random intercept was added at nursing home-level to account for clustering of participants in ten nursing homes. P-values are reported for unadjusted models and additionally for models adjusted for age, gender, neuropsychiatric symptoms (NPI-NH) and dementia stage (FAST). Generalized estimating equations (GEEs) were used in some cases where more robust estimation methods lead to more stable models.
For the purpose of a sensitivity check, differently specified LMM analyses were conducted based on a three-level hierarchy with the repeated measure time points (level 1) nested in participants (level 2) who were grouped in different nursing homes (level 3). Accordingly, nursing home-level random intercepts were also included in each model. Fixed factors in the unadjusted models included group (TBI vs. CAS), time (baseline vs. post intervention) and a group x time interaction term. Time was modelled as a repeated measure, an autoregressive covariance structure (AR1) was chosen to account for dependencies between both time points.
In LMM for analysing EMA of momentary quality of life measured before and after each individual activity session, the time-varying covariate session was added. Models for EMA also included the factor group (TBI vs. CAS) and covariates for pre-session EMA measurements (baseline EMA), age, gender, neuropsychiatric symptoms (NPI-NH) and dementia stage (FAST). Clustering at nursing home level was again accounted for (random intercept), resulting in three level models. All statistical analyses were performed using IBM SPSS statistics software (IBM SPSS Statistics for Windows, Version 25.0. Armonk. NY: IBM Corp). No adjustment for multiple testing was applied since all secondary outcomes were analysed within an exploratory framework. Therefore, p-values have to be interpreted with caution for secondary analyses.