Quality of life in people with dementia living in nursing homes: validation of an eight-item version of the QUALIDEM for intensive longitudinal assessment

The PflegeTab (engl. CareTab) study aimed to develop and evaluate a tablet-based psychosocial intervention tailored to the needs of PwD. A tablet application was developed and combined with innovative care concepts in order to enable a flexible, patient-centered care approach. The app was tested in ten nursing homes in Berlin over an 8-week intervention period, in which participating residents received activation sessions 3 times per week for up to 30 min. The full 37-item version of the QUALIDEM questionnaire was used for the assessment of comprehensive QoL before and after the intervention period. In the present analyses in this article, we only regarded the individual measurements that were administered before the intervention (at baseline). As our main focus lay on momentary and comprehensive QoL, momentary QoL was measured via tablet during the intervention period before and after each activation session (For the present analyses, before session assessments were used only; Table 1).

The planned sample size of the PflegeTab study was N = 240 PwD across eight nursing homes [22]. The sample size calculation was based on the main study, which included a randomized trial design, where the sample size referred to a medium-large effect size of Cohen’s d = 0.5 (α = 0.05; 1 − β = 0.80; 20% dropout rate; between cluster variance ICC = 0.005; G*Power 3.1) between the two arms where we primarily used the assessments at baseline (see ISRCTN98947160). Although the number of participating care facilities was increased to ten during the planning phase, the target sample size could not be fully achieved. Participants were included if they were nursing home residents and had a medical diagnosis of dementia (International Statistical Classification of Diseases and Related Health Problems—ICD-10: F00-F03), including Alzheimer’s disease, vascular dementia, and unspecified dementia [23]. Participants were excluded if other serious chronic psychiatric diagnosis were given F10-29, exceptions: F10.1; F17.1; F17.2; F32.2; and F32.3 may be included. An admission to the nursing home less than 4 weeks beforehand was also a criterion for exclusion. A total of 203 people (eligible nursing home residents or their conservators) were contacted and 163 PwD took part in the PflegeTab study. For the current analyses, 13 of the participants could not be included because the baseline momentary QoL and momentary QoL could not be imputed. As QUALIDEM is a proxy-rated assessment tool, all participants were assessed by the nursing staff working in the nursing homes (Fig. 1).

Comprehensive QoL was assessed with a 37-item version of QUALIDEM suitable for PwD with mild-to-severe dementia using proxy rating [24]. Out of 40 existing items, Item 9: ‘Does not want to eat’, item 30: ‘Likes to lie down in bed’, and item 15: ‘Enjoys meals’ was not represented because we excluded the subscale J in our study; thus, 37 items remained. In order to test the main hypothesis, a sum score was created across all 37 items at baseline. Further, we studied the determinants of QoL according to the literature [25‐27] to establish nine subscales: care relationship (7 items), positive affect (6 items), negative affect (3 items), restlessness (3 items), positive self-image (3 items), social relationship (6 items), social isolation (3 items), feeling at home (4 items), and having something to do (2 items) (Table). These subscales are sum scores across the respective items, which were presented in a Likert format ranging from never (0), rarely (1), and sometimes (2) to frequently (3). Across all 37 items, a principal component analyses (PCA) extracted ten components with an Eigenvalue over 1 although the visual evaluation of the Scree plot suggested four components only. The varimax-rotated PCA solution accounted for 67.2% of the overall variance. Next to the subscales, we created a composite sum score across all 37 items, which has been used extensively in the literature as well [25‐30]. Thus, in our analyses, we look into both the comprehensive sum score (Table 2) as well as subscale results (Table 3) and its relation with momentary QoL.

Momentary QoL was assessed with an eight-item version from the 37-item long version of QUALIDEM. All eight items had been executed on tablet computers at repeated time points. Aiming to use a brief scale that was considered as suitable for the tablet-based momentary QoL assessment in real life nursing home environment by nurses, we chose eight items: restlessness [item 19 of the QUALIDEM], mood [item 10], anxiousness [item 6], body language [item 22], communication [item 12], happiness [item 05], sadness [item 11] and sociability [item 34] belonging to the four subscales: ‘positive affect’, ‘negative affect’, ‘restlessness’, and ‘social relationships’ of the full 37-item version of QUALIDEM (i.e., subscale B, subscale C, subscale D, and subscale F) [8, 31]. As part of a workshop, the study team was selecting the items based on the following considerations: Two items from four subscales of the German QUALIDEM version were used for the short version. In addition, some scales were not suitable for nursing home residents with severe stages of dementia. The item length played a minor role and was relevant only when selecting the two items from a scale. In that case, shorter items were preferred, since we assumed that all items belonging to the same scale were assumed equivalent in content and therefore largely interchangeable.

Dementia stage was measured with the Functional assessment staging (FAST) which is scored from 1 to 16 for 7 consecutive stages and 9 substages [32]. As PwD progress in severity, the numerical value of the FAST increased, with stage 4 corresponding to mild dementia, stage 5 to moderate, stage 6 to moderately severe, and stage 7 to severe dementia. According to Arons et al. (2017), the QUALIDEM is applicable in PwD with all stages of dementia severity [7].

Functional status was measured with the Barthel Index [33]. Barthel Index is scored from 0 to 100, served as a covariate and activities of daily living (ADL) related to self-care activities such as bathing, dressing, grooming, and other activities. Barthel Index ranged from 0 to 95.

Depressive symptoms were measured with GDS-SF-15 (Geriatric Depression Scale– 15 items Short Form which served as a covariate [34]. Total score ranged from 0–15. A score of 5 or more indicated probable depression. GDS ranged from 0 to 15 points.

Additional covariates were ID (i.e., unique identifier of facility), age, and gender of the residents.

The univariate and multivariate associations of momentary and comprehensive QoL were estimated using mixed modeling which takes the nested data structure of measurement occasions in individuals in nursing homes into account. Momentary QoL (level-1) was regressed on momentary session (time trend; level-1) and comprehensive variables, i.e., comprehensive QoL, age, gender, functional and cognitive status, and depressive symptoms (all level-2). The ID of the nursing home of each person with dementia was used as a clustering variable (level-3). Intercepts were allowed to vary across individuals and facilities (intercept only model; variance components). Furthermore, an autoregressive covariance matrix was assumed for the time variables to account for autocorrelation (based on inspection of fit indices [AIC; Akaike's Information Criterion [35], where smaller values indicate better fit: AIC_{autoregressive} = 6272; AIC_unstructured = 6544, without reaching convergence; AIC_identity = 6617; AIC_{variance components} = 6622] and theoretical plausibility). Prior to the estimation of the multivariate models, we inspected the trends graphically as well as compared the AIC values. Based on the inspection, we decided for a fixed slope random intercept model (AIC = 6272) as quadratic and cubic fixed and random effects did not provide improvement in the fit of the model with the data (AIC = 6278 to 6307). All variables have been standardized allowing the coefficients to vary between − 1 and 1 and, thus, can be interpreted as beta coefficients. All data analyses were conducted with MIXED procedures in SPSS (IBM SPSS Statistics for Windows, Version 25.0, released 2017. Armonk, NY: IBM Corp.) using the Restricted Maximum Likelihood estimator (REML), which accommodates for unbalanced datasets, i.e., missing cases in repeated measure designs [35]. Scale missing values were imputed using Expectation–Maximization algorithm for Barthel Index (n = 1), FAST (n = 1), GDS (n = 38), and comprehensive QoL (n = 1). Models with and without imputation did not differ in the direction or magnitude of the main findings.

For the evaluation of the internal consistency, at each session (time point), Cronbach's alpha was used (Kuder-Richardson Reliability Coefficients KR20 in the case of binary items). Internal consistency was considered as excellent if Cronbach’s alpha was α ≥ .90 or higher, as good if α ≥ .80, as acceptable if α ≥ .70 and as questionable to unacceptable if α ≥ .60 [36]. For test–retest reliability, lagged momentary QoL variables (lag 1, lag 2) were calculated and regressed on momentary QoL in a model that accounted for the nested structure as described above (except for using a variance components covariance structure to capture autocorrelation merely within the coefficients). According to Dichter et al. (2011), test–retest reliability can be seen as the response stability over time, which was operationalized as the regression coefficient of the same test across subsequent time points [37, 38]. Thereby, lag 1 refers to the regression of the value of momentary QoL in one session onto the value of momentary QoL in the prior session. Accordingly, lag 2 refers to the regression of the value of momentary QoL in one session onto the value assessed two sessions before. In the tested models, momentary QoL was compared with each lagged momentary QoL for the nested data structure (Supplementary Table S2).

We found that the internal consistency of the comprehensive sum scale of the QUALIDEM was good with a Cronbach’s alpha of α = .86. Likewise, the internal consistency of the eight-item momentary QoL was excellent as indicated by Cronbach's alpha with a range from α = .88 to α = .93, indicating decent reliability at each time point (Supplementary Table S1). Regarding unidimensionality of the momentary QoL scale, a PCA extracted one factor with an Eigenvalue over 1 and the evaluation of the Scree plot suggested one factor as well. The PCA solution accounted for 62.6% (sums of squared loadings) of the overall variance. Regarding test–retest reliability, momentary QoL was associated with each lagged momentary QoL (i.e., lag 1) with B = .40 (CI .36/.44, p < .001) when accounting for the nested data structure (Supplementary Table S2). Additionally, the lagged association of momentary QoL lagged across two sessions of assessment (i.e., lag 2) was B = .36 (CI .31/.40, p < .001). Finally, in a model where both (consecutive lag 1 and lag 2 time points momentary QoL) were entered, lag 1 association was B = .32 (CI .27/.36, p < .001) and lag 2 was B = .23 (CI .19/.27, p < .001; Supplementary Table S2).

Regarding univariate associations, momentary QoL was not significantly related to session of measurement (B = .05, CI − .01/.11, p = .123), age (B = − .06, CI − .12/.1, p = .081), gender (B = .01, CI − .05/.07, p = .72), Barthel Index (B = − .006, CI − .07/.06, p = .86), and FAST score (B = .04, CI − .03/.10, p < .240; Table 2). However, we did find a significant positive association of momentary QoL with comprehensive QoL (B = .17, CI .11/.23, p < .001) and a significant negative association of momentary QoL with GDS (B = − .16, CI − .22/− .10, p < .001).

Regarding the random effects in the univariate model, where momentary QoL was regressed on comprehensive QoL, the random variance of session of measurements between PwD (autoregression, diagonal) was significant in the univariate model (B = .90, CI .84/.96, p < .001), as well in the multiple model (B = .90, CI .82/.95, p < .001; Table 2). Furthermore, the residual correlation between two sessions of measurement was significant in the univariate model (autoregression, rho; B = .39, CI .35/.43, p < .001), as well as in the multiple model (B = .37, CI .33/.41, p < .001), indicating that a higher-than-average rating in one session is associated with a higher-than-average rating on the consecutive session. However, the random variance between facilities was neither significant in the univariate model (B = .08, CI .03/.22, p = .062), nor in the multiple model (B = .09 CI .03/.25, p = .058).

For the multiple associations of momentary QoL with the comprehensive QoL, momentary QoL was not significantly related to session of measurement (B = .03, CI − .02/.09, p < .264); Barthel Index (B = − .02, CI − .11/.06, p < .602); FAST (B = .04, CI − .04/.13, p < .358); age (B = − .06, CI − .13/.00, p = .054); and gender (B = .03 CI − .03/.09, p < .321). However, the momentary QoL was significantly positively related to comprehensive measured QoL at baseline (B = .14, CI .08/.20, p < .001) and significantly negatively related to GDS (B = − .13, CI − .19/− .06, p < .001).

Finally, concerning associations of QUALIDEM subscales with comprehensive QoL, positive (B = .17, CI .11/.23) and negative affect (B = .13, CI .07/.19), restlessness (B = .07, CI .01/.14) ,as well as social relationships (B = .16, CI .09/.22) and isolation (B = .07, CI .01/.14) were significantly related (all p > .05; see Table 3).