Study Design and Participants
The study was based on analyses of cross-sectional data collected through the ongoing PROTECT (30) study in 2019. PROTECT is a 25-year longitudinal study launched in 2014 that assesses participants every year on measures of physical, mental, and cognitive health, lifestyle, and perceptions of aging through an online platform.
Individuals are eligible to participate in the PROTECT study if they are UK residents, English speakers, aged 50 years and over, have access to a computer and internet, and do not have a clinical diagnosis of dementia at the point of recruitment. Participants were recruited through national publicity and via existing cohorts of older adults. Potential participants enrolled through the PROTECT study website, downloaded the study information sheet, and provided consent online.
The PROTECT study has ethical approval from the London Bridge NHS Research Ethics Committee and Health Research Authority (Ref: 13/LO/1578). Ethical approval for the data analyses was sought through the ethics committee at the University of Exeter, School of Psychology (Application ID: eCLESPsy000603 v1.0).
Between 1st January 2019 and 31st March 2019, 14,797 participants took part in the PROTECT annual assessment. Among these, 9,410 participants completed the AARC questionnaires and were therefore included in the present study (mean (SD; range) age: 65.9 (7.1; 51-95) years). Only 0.4% of participants reported having been diagnosed with mild cognitive impairment. We estimated that a further 1.2% of participants had mild cognitive impairment (as they scored 1.5 SDs below the mean study sample score in two or more cognitive tasks). Those participants that we identified as having mild cognitive impairment were kept in the analyses. However, participants with higher levels of AARC losses on the AARC-10 SF and on the AARC-50 cognitive functioning subscale had poorer scores on the four objective cognitive tasks, indicating that participants were aware of their cognitive abilities (See supplementary material Table 1) and hence their answers to the AARC-10 SF and the AARC-50 cognitive functioning subscale can be deemed accurate.
The majority of study participants were of white ethnicity (98.5% of participants), married (79.1% of participants), completed a university education (75.8% of participants) and were not retired (42.6% of participants). Demographic characteristics for the study sample are reported in Table 1. Means and standard deviations stratified by age, gender, and education level for AARC gains and losses assessed both with the AARC-10 SF and the AARC-50 cognitive functioning subscale are reported in Tables 2a and 2b.
A high proportion of participants perceived their health as good (54.1%) or excellent (30.8%). On average participants did not report functional difficulties (IADL mean (SD) score = 0.16 (0.77)). Participants had minimal levels of current depressive (mean (SD) = 11.5 (3.0)) and anxiety symptoms (mean (SD) = 9.3 (8.5)), and low levels of both lifetime depressive symptoms (mean (SD) = 2.7 (3.3)) and lifetime anxiety symptoms (mean (SD) = 1.0 (2.1)).
Compared to those who did not complete the AARC questionnaires (N= 5,387), the study sample included a larger proportion of females (79.9% versus 71.3%) and participants who were better educated (75.8% versus 70.8%), and a lower proportion of individuals who were employed (42.6% versus 54.7%).
Measures assessing felt age, ATOA, mental and physical health, and objective cognitive functioning were used to explore construct validity for the AARC-10 SF. Measures assessing felt age, ATOA, and objective, self-reported, and informant-reported assessments of cognitive functioning were used to explore construct validity for the AARC-50 cognitive functioning subscale. Demographic variables (age, gender, marital status, employment, and university education) were assessed to explore their relationships with levels of AARC gains and losses assessed both with the AARC-10 SF and with the AARC-50 cognitive functioning subscale.
Participants provided demographic information through the PROTECT platform at baseline through an online assessment adapted from Office of National Statistics measures, which included data on age, sex, ethnic origin, marital status, employment, and university education. Ethnicity included the following categories: white, mixed (included white and black Carribean, white and black African, white and Asian, any other mixed multiple ethnic background), Asian, black, or other ethnic groups. Marital status was used as a dichotomous variable (individuals who were married, in a civil partnership, or co-habiting were grouped together versus individuals who were unmarried, divorced, separated, or widowed). Employment status was used as a dichotomous variable (employed versus not employed). University education was used as a dichotomous variable (university education versus no university education). Individuals without a university education were those participants that had completed secondary education (GCSE/O levels) or post-secondary education (college, A-levels, NVQ3, or below). Individuals with a university education were those participants that had completed vocational qualifications (diploma, certificate, BTEC, NVQ4, and above), undergraduate degrees (e.g., BA, BSc), post-graduate degrees (e.g., MA, MSc), and doctorates (PhD).
Awareness of Age-Related Change (AARC)
The AARC-10 SF (1) is a brief tool for capturing perceived age-related gains (AARC gains) and losses (AARC losses). It contains ten items, five assessing AARC gains and five assessing AARC losses. Each of these five items assesses a different AARC behavioral domain (health and physical functioning, cognitive functioning, interpersonal relationships, socio-cognitive and socio-emotional functioning, and lifestyle/engagement). All ten items start with the same stem “With my increasing age, I realize that…”. An example of an item capturing AARC gains is “…I appreciate relationships and people much more”, while an example of an item capturing AARC losses is “…I have less energy”. Respondents rate how much each item applies to them on a five-point Likert scale (1 = “not at all”, 2 = “a little bit”, 3 = “moderately”, 4 = “quite a bit”, and 5 = “very much”). Scores can be obtained for the AARC gains and AARC losses subscales by summing items that fall into the respective scales. Scales scores range from a minimum of five to a maximum of 25 with higher scores indicating higher levels of awareness of age-related change.
AARC-50 Cognitive Functioning Subscale
The cognitive functioning subscale of the AARC-50 questionnaire (8) includes ten items, five assessing AARC gains and five assessing AARC losses. An example item capturing AARC gains in the cognitive domain is “With my increasing age, I realize that I have become wiser”, while an item capturing losses is “With my increasing age, I realize that I am more forgetful”. Respondents rate how much each item applies to them on a five-point Likert scale (1 = “not at all”, 2 = “a little bit”, 3 = “moderately”, 4 = “quite a bit”, and 5 = “very much”). Scores on the AARC- cognitive functioning gains and AARC- cognitive functioning losses subscales are obtained by summing items that fall into the respective subscales. Subscales scores range from a minimum of five to a maximum of 25 and higher scores indicate higher levels of awareness of age-related change in the cognitive domain.
Attitudes Toward Own Aging (ATOA)
The ATOA scale is a valid and reliable five-item scale assessing participants’ attitudes toward their own aging taken from the Philadelphia Geriatric Center Morale Scale (24). For each statement respondents are asked to make temporal comparisons about changes in energy level, perceived usefulness, happiness, and quality of life and to respond on a binary response set (better versus worse, yes versus no). An example item is “things keep getting worse as I get older”. A proportion-based score can be obtained by summing the participant’s item scores and by dividing it by the number of responses, with a score of one indicating that positive attitudes are implied in all answers and a score of zero indicating that a negative response is implied in all answers.
Felt age was assessed with a single-item question (adapted from the National Survey of Midlife development in the United States; MIDUS; 23) asking participants to write the age (in years) that they feel most of the time. A proportional discrepancy score was calculated by subtracting the participants’ felt age from their chronological age, and by dividing this difference score by participants’ chronological age. A positive value indicates a youthful felt age, while a negative value indicates an older felt age.
Cognitive Functioning – Objective Assessment
Cognitive function was measured with the PROTECT Cognitive Test Battery (31-33) which includes four tests: (1) the Grammatical Reasoning task assesses verbal reasoning (34); (2) the Digit Span task (35) assesses verbal working memory; (3) the Self-ordered Search task measures spatial working memory (36); and (4) the Paired Associate Learning task (37) assesses visual episodic memory.
For each task a summary score can be obtained by subtracting the number of errors from the number of correct answers. Hence for each task a higher score indicates a better performance. For digit span the summary score can range from 0 to 20. For paired associate learning the summary score can range from 0 to 16. For verbal reasoning the summary score is also obtained by subtracting the number of errors from the number of correct answers, but the score has no set upper or lower limit as the participants can attempt as many trials as they can manage within a specific timeframe. Finally, the summary score for the self-ordered search task can range from 0 to 20.
Cognitive Functioning - Informant Rating and Self-Rating
The Informant Questionnaire on Cognitive Decline in the Elderly short form (IQCODE; 38, 39) was administered to an informant close to the participant. The IQCODE is a valid and reliable 16-item questionnaire that asks respondents to rate the cognitive change of someone close to them over the last 10 years. Items describe both cognitive improvement and cognitive decline (an example item is “Remembering things that have happened recently”) and can be answered on a five-point scale (1 = “much improved”, 2 = “a bit improved”, 3 = “not much change”, 4 = “a bit worse”, and 5 = “much worse”). The final score is the mean of the item scores. A parallel version of the IQCODE was administered to the participant (IQCODE - Self; 38).
Patient Health Questionnaire-9
The Patient Health Questionnaire-9 (PHQ-9; 40) is a valid and reliable nine-item scale capturing depressive symptoms over the previous two weeks. It is based directly on the diagnostic criteria for major depressive disorder described in the Diagnostic and Statistical Manual Fourth Edition (DSM IV; 41). Respondents are asked to indicate how frequently they experience each symptom on a four-point Likert scale (1 = “not at all”, 2 = “several days”, 3 = “more than half the days”, and 4 = “nearly every day”). The total score is the sum of the item scores and can range from 9 to 36.
Composite International Diagnostic Interview-Short Form
The Composite International Diagnostic Interview-Short Form (CIDI-SF; 42) is a reliable and valid measure for assessing lifetime symptoms of depression and anxiety. Nine items assess depressive symptoms and eight items assess anxiety symptoms. An example of a depressive symptom question is “did you lose interest in most things?”. For each item, participants can answer “yes” if they have the symptom or “no” if they do not have the symptom. For both depression and anxiety a total score can be calculated by summing the items where the participants answer yes. For depression and anxiety the total score can range from zero to nine and from zero to eight, respectively.
The Generalized Anxiety Disorder-7 (GAD-7; 43) is a valid and reliable seven-item measure assessing symptoms of generalized anxiety disorder. Respondents are asked to indicate the frequency of occurrence of a list of symptoms over the past two weeks on a four-point scale (1 = “not at all”, 2 = “several days”, 3 = “more than half the days”, and 4 = “nearly every day”). The overall score is the sum of the item scores and ranges from 7 to 28.
Instrumental activities of daily living
Lawton’s Instrumental Activities of Daily Living Scale (IADL; 44) is a reliable instrument to assess everyday functional status. It describes seven activities including preparing meals, managing medications, and using the telephone. For each activity respondents have to rate how difficult they find performing the activity (0 = “no difficulty”, 1 = “some difficulty”, and 2 = “great difficulty”). The total score ranges from a possible 0 to 14.
We assessed perceived health with a single-item question (taken from the SF-36; 45) asking participants to rate their own health on a four-point scale ranging from excellent to poor (“excellent”, “good”, “fair”, and “poor”).
As the validation of the AARC-10 SF (1) in US and German samples supported a two-factor structure (one factor for each of AARC gains and AARC losses), we used confirmatory factor analysis (CFA) to confirm this structure in the UK population. We tested whether the five items assessing gains and the five items assessing losses (of the AARC-10 SF) are related to the respective hypothesized underlying factors of AARC gains and AARC losses. The two factors AARC gains and AARC losses were allowed to correlate in the CFA model. Error terms were allowed to correlate for the pair of gains and losses items for the same AARC behavioral domain (Figure 1a).
CFA was also conducted to confirm the two-factor structure of the AARC-50 cognitive functioning subscale (8) (Figure 1b). For both the AARC-10 SF and the AARC-50 cognitive functioning subscale, to confirm the need for a two-factor model (above described), we also fitted a model in which a single factor loaded on all ten items. For both the AARC-10 SF and the AARC-50 cognitive functioning subscale, we compared goodness of fit indices (GOF) of the two-factor model with those of one-factor model. Because the Chi-squared statistic is often significant for well-fitting models in large samples (46) alternative goodness of fit measures including the Comparative Fit index (CFI), the Tucker-Lewis index (TLI), the Root Mean Square Error of Approximation (RMSEA), and the Standardized Root Mean Square Residual (SRMR) were examined. Criteria for acceptable model fit were CFI and TLI > .90, RMSEA < .08 (90% CI: between 0 and .08), and SRMR < .06 (47). The CFA models were fitted using the sem command in Stata. Analyses included only participants that provided complete data on all items.
We used Cronbach’s alpha (α) to quantify reliability for the gains and losses subscales of the AARC-10 SF and the AARC-50 cognitive functioning subscale (48). We considered α values between .65 and .95 to be satisfactory.
For both the AARC-10 SF and the AARC-50 cognitive functioning subscale, we used CFA to test measurement invariance (18, 49, 50) between males and females, between two groups characterised by university education (vocational qualification, undegraduate degree, post-graduate degree, or doctorate) and no university education (secondary or post-secondary education, and among three age groups (middle age = 50 to 65 years; early old age = 66 to 75 years, advanced old age = 76 years and over). To explore measurement invariance, we fitted three CFA models: (a) Model 1 placed no equality constraints across groups on factor loadings, item intercepts, the error variances, the variances of the latent variables, or the covariances of the latent variables (assumes configural invariance); (b) Model 2 constrained the factor loadings to be identical across subgroups (assumes metric invariance); (c) Model 3 constrained the factor loadings and item intercepts to be identical across subgroups (assumes strong invariance).
To evaluate the fit of a model compared to a less restrictive one, the traditional approach involves assessing the differences in the Chi-squared fit statistics of the two examined CFA models by conducting likelihood ratio tests (LRT). However, as LRTs often result in statistically significant differences in large samples for models that are not markedly different in fit (46) and alternative fit indices are less sensitive to sample size (51), we explored model differences using alternative GOF indices including the Comparative Fit index (CFI), the Root Mean Square Error of Approximation (RMSEA), and the Standardised Root Mean Square Residual (SRMR). We concluded that a model had a worse fit than a less constrained model when the difference in CFI (ΔCFI) was larger than -.01 (52, 53), the difference in RMSEA (ΔRMSEA) was larger than .015 (54), and the difference in SRMR (ΔSRMR) was larger than .03 (54).
Construct validity for the AARC-10 SF was explored by estimating correlations between the AARC-10 SF and each of felt age, ATOA, measures of mental and physical health, and objective assessments of cognitive functioning (Grammatical Reasoning, Digit Span, Self-ordered Search, and Paired Associate Learning). Construct validity for the AARC-50 cognitive functioning subscale was explored by estimating correlations between the AARC-50 cognitive functioning subscale and each of felt age, ATOA, and objective (Grammatical Reasoning, Digit Span, Self-ordered Search, and Paired Associate Learning), self-reported, and informant-reported assessments of cognitive functioning. We used Pearson’s r to quantify correlations (55). Correlation coefficients under .10 were considered negligible, between .10 to .29 were considered small, between .30 to .49 were considered moderate, and .50 or above were considered large (56).
To explore whether age, gender, marital status, employment status, and university education explain variability in levels of AARC gains and/or AARC losses, we fitted multiple linear regression models for each of the AARC-10 SF and the AARC-50 cognitive functioning gains and losses. We also conducted simple regressions in which the predictive role of each demographic variable (age, gender, marital status, employment status, and university education) on levels of AARC gains/losses was explored without controlling for the predictive role of the remaining demographic variables.