“Living Well” Trajectories Among Family Caregivers of People With Mild-to-Moderate Dementia in the IDEAL Cohort

Abstract Objectives Understanding whether and how caregivers’ capability to “live well” changes over time, and the factors associated with change, could help target effective caregiver support. Methods We analyzed 3 time points (12 months apart) of Improving the experience of Dementia and Enhancing Active Life (IDEAL) cohort data from coresident spouse caregivers of community-dwelling individuals who had mild-to-moderate dementia at baseline, using latent growth and growth mixture models. Capability to “live well” was derived from measures of quality of life, well-being, and satisfaction with life. Results Data from 995 spouse caregivers at Time 1, 780 at Time 2, and 601 at Time 3 were included. The mean “living well” score decreased slightly over time. We identified 3 classes of caregivers: one with higher baseline scores declining slightly over time (Stable; 66.8%), one with low baseline scores remaining stable (Lower Stable; 26.0%), and one with higher baseline scores showing marked decline (Declining; 7.2%). Scores on baseline measures differentiated the Lower Stable, but not the Declining, from the Stable class. Longitudinally, the Declining class was associated with care recipient cognitive decline and increasing hours providing care, as well as caregiver stress and depression. Findings were similar when caregivers with other kin relationships were included. Discussion The findings indicate the importance of prompt identification of, and support for, caregivers at risk of the declining capability to “live well” and may assist in identifying those caregivers who could benefit most from targeted support.


Appendix 1. Further details of measures included in the analyses
Except where otherwise indicated, all measures relate to the caregiver and are based on caregiver self-report.
Demographic and clinical characteristics. Demographic information was collected by interviewers including caregiver age, sex, relationship to the care recipient (spouse/partner or family/friend), whether the caregiver is co-resident with the care recipient, education (no qualifications; school leaving certificate at 16; school leaving certificate at 18; college/university), and social class (I/II, high; III-NM/III-M, middle; IV/V/armed forces, low; Office for National Statistics, 2010). The number of hours spent caring per day was collected and categorized into under 1 hour, 1-10 hours, and over 10 hours. Sex of the person with dementia was recorded.
Dementia diagnosis was obtained from the medical records of the person with dementia and classified into seven groups; Alzheimer's disease (AD); vascular dementia (VaD); mixed AD/VaD; frontotemporal dementia (FTD); Parkinson's disease dementia (PDD); dementia with Lewy bodies (DLB); unspecified/other. For analysis purposes, PDD and DLB subtypes were combined. Where the specific diagnosis changed over the course of the study, the last recorded diagnosis was used; this applied to 28 participants. Social situation. The MacArthur Scale of Subjective Social Status (Adler et al., 2000) was used to assess perceived standing in society and in the community, with participants asked to place themselves on a ladder ranging from 1 (low) to 10 (high). Social comparison was measured with a single question 'Do you think compared to most other people your age, your situation is….' with responses ranging from much worse to much better. Social isolation was measured using the sixitem Lubben Social Network Scale (score range 0-30; Lubben et al., 2006); higher scores indicate more social contact. To assess social capital, several measures are provided by the UK Office for National Statistics (Office for National Statistics, 2008). Frequency of social contact was measured using a 9-item measure of contact with friends, relatives, and neighbors, with a higher score indicating more contact (scale 0-32). Civic participation was a single item question 'In the last 12 months have you taken any of the following actions in an attempt to solve a problem affecting people in your area' with a list of seven actions responses which are summed to provide a score out of 7. Social participation is assessed in terms of involvement in unpaid help to groups, using the question 'During the last 12 months have you given any unpaid help to any groups, clubs, or organizations in the ways listed below', with a list of 12 responses which are summed to provide a score out of 12. For both civic and social participation, higher scores indicate more participation.
For analyses, scores were categorized into no participation (score of 0), low participation (score of 1), and high participation (score >1). Engagement in social activity was measured with the thirteenitem Cultural Capital Scale; higher scores indicate greater engagement (score range 13-65; Thomson, 2004).

Psychological health. The Center for Epidemiologic Studies Depression Scale-Revised
(CESD-R) was used to measure depression (score range 0-60), with higher scores indicative of more depressive symptoms (Eaton et al., 2004). Loneliness was measured using the six-item De Jong-Gierveld Loneliness Scale (score range 0-6; De Jong Gierveld & Van Tilburg, 2010); higher scores indicate greater loneliness. Neuroticism was measured using the mini-IPIP neuroticism measure, where a higher score indicates higher rates of neuroticism (score range 4-20; Donnellan et al., 2006). Self-esteem was measured using the ten-item Rosenberg Self-Esteem Scale (score range 10-40; Rosenberg, 1965); higher scores indicate greater self-esteem. Perceived self-efficacy was measured using the general sense of perceived efficacy scale (score range 10-40), with higher scores indicative of better self-efficacy (Schwarzer & Jerusalem, 1995). The six non-filler items from the Life Orientation Test-Revised scale (Scheier et al., 1994) were used to measure optimism (score range 0-24); higher scores indicate greater optimism.
Physical health. The Charlson Comorbidity Index (CCI) age-adjusted score (Charlson et al., 2008;Charlson et al., 1987) identified the number of chronic conditions. Subjective health was assessed with the question "How would you rate your health in the past four weeks?" with six ordinal response options ranging from very poor to excellent (Bowling, 2005).
Experiences of caregiving. The Relative Stress Scale is a 15-item measure assessing the degree of distress and social upset experienced by a relative as a result of caring for a person with physical or behavioral difficulties (score range 0-60); a higher score indicates more severe stress (Greene et al., 1982). Role captivity (score range 0-12) is measured using three items assessing the extent that caregivers feel trapped in their role; higher scores are worse (Pearlin et al., 1990).
Management of Meaning is a 9-item measure assessing the extent that caregivers of people with dementia feel they have lost aspects of their personality due of caring (score range 9-36; a higher score is worse; Pearlin et al., 1990); higher scores are worse. The Modified Social Restriction Scale (MSRS) contains two items asking how easy it is to find someone to look after the care recipient if the caregiver needs a break or is unwell (Balducci et al., 2008); a higher score indicates more difficulties. Caregiver competency is measured using three items assessing the extent to which caregivers of people with dementia feel they are doing an adequate job as a caregiver (Robertson et al., 2007); possible scores range from 0-12, with higher scores indicating greater competence.
Measures relating to the person with dementia. Caregivers used the modified 11-item Functional Activities Questionnaire (FAQ; Martyr et al., 2012;Pfeffer et al., 1982) to rate the current functional ability of the care recipient (score range 0-33) and the Dependence Scale (Brickman et al., 2002), a 13-item questionnaire measuring the amount of assistance needed by the person with dementia, to rate dependence (score range 0-15). In both cases, a higher score indicates greater functional impairment. Caregivers rated their level of emotional distress in response to specific symptoms they identifed as experienced by the person with dementia on the Neuropsychiatric Inventory Questionnaire (score range 0-35; Kaufer et al., 2000;Morris & National Alzheimer's Coordinating Center, 2008); higher scores indicate higher distress. People with dementia were administered the Addenbrooke's Cognitive Examination-III (ACE-III; Hsieh et al., 2013) and the total score was used as a measure of cognitive ability. Scores ranged from 0-100, with higher scores indicating better cognition.
Relationship quality. The Positive Affect Index provided a measure of current relationship quality (Bengtson & Schrader, 1982). Responses to five questions on a 6-point scale are summed for a total score (range 5-30); a higher score indicates better relationship quality.
'Living well'. Caregivers' capability to 'live well' was defined using three individual selfrated measures covering caregivers' quality of life, well-being, and satisfaction with life. Quality of life was measured with the World Health Organization Quality of Life-BREF (WHOQOL-BREF), which is designed to measure multiple components related to quality of life (Skevington et al., 2004). It includes two single indicators (overall quality of life and general health) and four domains (physical health, psychological health, social relationships, and environment). There is no total score, and to create one a factor analysis was conducted to estimate factor scores for those with complete data as previously described (Clare et al., 2019;Wu et al., 2021). Well-being was measured by the World Health Organization-Five Well-being Index (WHO-5; score range 0-100), which includes items on positive mood, vitality, and general interests (Bech, 2004). Satisfaction with life was measured using the Satisfaction with Life Scale (SwLS; score range 5-35), which is designed to measure global judgements of satisfaction with life (Diener et al., 1985).

Latent Growth Curve Model
To determine how 'living well' changes over time for caregivers of people with dementia, latent growth curve modelling (LGCM) was conducted in Mplus Version 8.2 (Muthén & Muthén, 1998-2017 using the first three timepoints of IDEAL data (T1-T3). The LGCM comprised a measurement model which was then extended to a second order growth model allowing estimation of the mean intercept (baseline) and the mean slope (change over time) of 'living well', with random effects to account for variation across individuals (Wickrama et al., 2016).
The measurement model involves building the latent factor 'living well' from measures of SwLS, WHOQOL-BREF, and WHO-5 at each year by longitudinal confirmatory factor analysis (LCFA). For WHOQOL-BREF factor scores were generated as previously described (Clare et al., 2019;Wu et al., 2021). SwLS was selected as the marker variable with loading fixed to 1 at each timepoint, and the intercept fixed to zero to allow for model identification. The scale of 'living well' took on the same scale as SwLS and the variance of each latent factor and covariance among latent factors were defined by SwLS (Brown, 2006). The associations among WHOQOL-BREF and WHO-5 were estimated relative to their association with SwLS. Variances were estimated for each subdomain indicator and autocorrelated errors specified and retained in the model to avoid misspecification (Little, 2013).
A good model will have a Comparative Fit index (CFI) and Tucker-Lewis index (TLI) greater than 0.90, and a root mean square error of approximation (RMSEA) less than 0.08 (Hu & Bentler, 1999). As shown in Supplementary Table 2, the unconstrained measurement model (configural model) was a good fit to the data indicating that each factor is defined by the same variables and that the same general pattern of factor loadings hold across time (Millsap & Cham, 2012;Millsap & Olivera-Aguilar, 2012). In order for meaningful comparisons to be made in a LCFA, the assumption of longitudinal measurement invariance should be met (Byrne & Watkins, 2003;Chen, 2007). Three levels of measurement invariance were tested imposing additional restrictions at each step: metric invariance (constrained factor loadings across measurement occasions), scalar invariance (constrained factor loadings across measurement occasions and intercepts across time to be equal), and strict invariance (constrained factor loadings across measurement occasions, and intercepts and residual variances across time to be equal). Each level of measurement invariance was applied, and a range of model fit indices examined to ensure that the model fit did not weaken when each level of constraint was applied. Studies have suggested that CFI, RMSEA, and standardized root mean squared residual (SRMR) are the most important indicators when it comes to testing measurement invariance, with cut offs of <0.01 change in CFI, <0.015 change in RMSEA, and <0.030 change in SRMR (Chen, 2007;Cheung & Rensvold, 2002). ΔX 2 was also examined but is sensitive to sample size so was not relied upon (Chen, 2007;Kline, 2011;Schermelleh-Engel et al., 2003). As shown in Supplementary Table 2, the changes within the model fit indices for each more constricted model were within these limits supporting the view that metric, scalar, and strict measurement invariance held. Further analyses were conducted with the strict invariance model. The intercept loadings were fixed to 1 for each latent intercept, and 0, 1, and 2 for time based on the yearly measurement occasions. Due to only having 3 timepoints a linear trend was assumed.

Latent class growth analysis and growth mixture modelling
So far, our approach assumes growth trajectories of all individuals can be adequately described using a single estimate of growth parameters. We employed latent class growth analysis (LCGA) and growth mixture modelling (GMM) to examine whether multiple growth trajectories of 'living well' exist in the IDEAL caregiver population (Jung & Wickrama, 2008;Muthén, 2004;Muthén & Shedden, 1999). Different assumptions were tested. The LCGA fixes variances of the global growth factors to zero across classes (assuming trajectories within a class are homogeneous).
The GMM-CI (class-invariant) constrains the variances of the global growth factors across classes to be equal, and the GMM-CV (class-varying) freely estimates all variances of the global growth factors across classes. A frequent problem with these models is non-convergence and local solutions (Hipp & Bauer, 2006), with the more complex models, particularly those with free variances, more likely to experience convergence difficulties (Grimm & Ram, 2009). Between 1 and 5 class solutions were tested for each assumption, with 1,000 random starts and 20 iterations for each model in order to avoid local solutions. Following successful convergence, the optimal number of distinct trajectories was determined using Bayesian Information Criterion (BIC), sample size adjusted BIC (ssBIC), the Lo-Mendell-Rubin likelihood ratio test (LMR-LRT), and the bootstrapped likelihood ratio test (BLRT) which provide between model comparisons (k vs k-1), and entropy (Jung & Wickrama, 2008;Nylund et al., 2007;Tein et al., 2013). Entropy is a standardized index of a model-based classification accuracy based on the average posterior probability, with higher values indicating clearer class separation (Muthén, 2004). Substantive criteria were based on a class size greater than 1% and theoretical and practical interpretability of the classes. Upon finding an optimal solution, the model was repeated with double the number of starts and iterations to ensure a global solution.
Due to their person-centered approach, GMMs allow for examination of predictors of class membership (Wickrama et al., 2016). The categorical latent class is related to the covariates by way of multinomial logistic regression which assigns each individual fractionally to all classes using posterior probabilities. Predictors of class were examined using the '3-step' approach in Mplus (R3STEP) in order to protect the latent class structure from influences of the covariates (Asparouhov & Muthen, 2013;Vermunt, 2010).

Latent Trajectory Model selection
For the cohort of co-resident spouse caregivers, three models were tested as described above: LCGA, GMM-CI and GMM-CV. One to 5 class solutions of each model were conducted, and results are displayed in Supplementary Table 3 for those models that converged, along with the associated plots in Supplementary Figure 2. The LCGA and GMM-CI models are less computationally intensive, and all classes ran with no convergence issues. The GMM-CV model with free intercepts and slopes experienced convergence issues. For the full sample of caregivers (spouse and family/friend), the LCGA and GMM-CI models also ran with no issues, with the GMM-CV model failing to converge. Given all available information including model fit indices, interpretability, and theoretical considerations, the 3-class solution was selected for both co-resident spouse caregivers and all caregivers.
Appendix 3. Sensitivity analysis to explore non-ignorable missing data models

Models that account for non-ignorable dropout
Growth modelling based on maximum-likelihood estimation tends to assume that data are missing at random (MAR; Little & Rubin, 2002). MAR is suitable when dropout is predicted by covariates or observed outcome values but if dropout does not fulfil the MAR assumption, missing data techniques that handle non-ignorable missingness may be more appropriate. Sensitivity analyses were conducted to compare the 3-class MAR growth mixture model for two of the measures used to estimate 'living well', SwLS, and WHO-5, with two non-ignorable dropout models described by Muthén and colleagues; Roy latent dropout pattern mixture models and Beunckens selection models (Muthén et al., 2011). The Roy latent dropout pattern-mixture model divides participants into groups based on their dropout pattern (Roy, 2003), and the Beunckens selection model combines the linear model for the observed responses with a logistic regression model for the non-ignorable dropout process (Beunckens et al., 2008).