Descriptive results
The numbers of those living in the community who provided a score from cognitive tests or self or proxy information on dementia were 9,826 for Wave 6, 9,602 for Wave 7, and 8,387 for Wave 8. In the pooled data there were 405 with dementia, 4,520 with low cognition, and 10,264 with intact cognition (no dementia or low cognition). These numbers excluded those who reported no ADL, IADL or mobility difficulties and were therefore not asked about receipt of help (11% of those with dementia, 33% with low cognition, and 50% with intact cognition). Table 2 shows the distribution of variables between the three cognition groups. People with dementia or low cognition tended to be older, less likely to own their homes, to have no qualifications, to belong to a lower occupational social class (NS-SEC), to be in the lower wealth quintile, and to have more functional limitations than those with intact cognition.
The cognitive scores in orientation and recall tasks were understandably lower among those with dementia and those with low cognition compared to those with intact cognition (Table 2). Fewer people with dementia participated in the cognitive tests (n.b. the cognitive scores were not used to identify the group with dementia and are shown for descriptive purposes only). A higher proportion of people with dementia or low cognition reported unmet need and unrequired help compared to those with intact cognition (Table 2). Quality of life was lowest among those with dementia and highest among those with intact cognition. The distributions of the individual unmet need and unrequired help items are shown in Supplementary Figure 1.
There were no missing data on gender, age, and having partner/children in the sample; 0.03% had missing data on functional limitation, 1-2% on education, home ownership and receipt of help, 6% on wealth quintile, and 18% had missing quality of life scores. The last two items were more often missing because determining wealth quintile required detailed information on financial assets which proxies did not necessarily know, and quality of life was a self-completion task which was not always competed by those with limited functioning. Missingness resulted in some differences between the distributions, for example the analytic sample with no items missing used for the model for unmet need included a higher proportion of people with dementia living with a partner and a lower proportion with unmet needs compared to all available data (see Table 2). However, when we ran the regression models with data using multiple imputation, the results for the associations between the variables were very similar between the imputed and unimputed datasets. Missingness in the predictors was generally low and affected only one of the outcomes, quality of life, in which case multiple imputation has very little effect and missing data can be ignored (37). In this paper, we therefore present the regressions for the unimputed data.
Differences in receipt of help by cognitive status
Unmet need was more likely (logit model) to be reported by those participants who were: older, living in rented housing, in the lower wealth quintile, and with more functional limitations (Table 3). The number of unmet needs (GLM model) was higher those who were: male, without a partner, living in rented housing, in a lower wealth quintile, and who had more functional limitations. People with dementia were less likely on average to have an unmet need and tended to have a lower number of unmet needs than those with no dementia. However, there were interactions between cognitive status and age (Wald test = 19.69, degrees of freedom (df) = 4, p < 0.001), and cognitive status and number of functional limitations (Wald test = 328.86, df = 4, p < 0.001) (see the estimates for the interaction model in Supplementary Table 1). Unmet need increased with age among people with intact cognition, but not among those with dementia or low cognition. The apparent contradiction in results on unmet need and cognitive status is explained by the finding that people with dementia at the lowest levels of functional limitations (1-2 limitations) were more likely to have unmet need, whereas at the higher levels of functional limitations (7+ limitations) those with no dementia had a higher number of unmet needs (Figure 1). The profiles of unmet need at the lower levels of limitations suggest that some ADL limitations (help with walking, stairs, bathing and getting in and out bed) were more often met for those with no dementia compared to those with dementia (Supplementary Figure 2). At higher levels of limitations, the profiles of unmet need show that those with dementia received help more often in all 13 limitations compared to those with no dementia (Supplementary Figure 3).
Unrequired help (logit model) was more common among: women, older people, those with a partner, those in the routine/manual compared to intermediate occupational social class, those who were not home owners, those in lower wealth quintiles, and those with more functional limitations (Table 4). The number of areas for which unrequired help was received (GLM model) was higher among: those with a partner, and those in the lower wealth quintile. People with low cognition or dementia were more likely to receive unrequired help compared to people with intact cognition. The interactions showed that, although women and those with a partner more often received unrequired help in the cognitively intact group and among those with low cognition, there were no differences by gender (Wald test = 9.81, df = 4, p = 0.044) or partnership status (Wald test = 11.07, df = 4, p = 0.026) among those with dementia. This was due to a large variation in the receipt of unrequired help by gender and partnership (see the estimates and marginal effects for the interactions in Supplementary Table 2a and 2b). The interaction between functional limitations and cognitive status (Wald test = 83.40, df = 4, p < 0.001) showed that a higher number of functional limitations among those with no dementia increased the likelihood of unrequired help, whereas among those with dementia there was very little difference in the frequency of unrequired help by the number of functional limitations (Figure 2). The profiles of unrequired help showed that at lower levels of functional limitation unrequired help for IADL tasks, especially help with shopping and housework, was more frequent among those with dementia compared to those with no dementia (Supplementary Figure 4). At higher levels of functional limitations, those with dementia tended to have somewhat higher levels of unrequired help, apart from help with dressing and housework, compared to people with intact cognition (Supplementary Figure 5).
Associations with quality of life (CASP-19)
Both unmet need and unrequired help were associated with lower quality of life (Table 5). Quality of life was lower among those with dementia or low cognitive functioning compared to those with intact cognition. Being female, higher age, having a partner, higher socioeconomic status and lower levels of functional limitations were associated with higher quality of life. However, among those with dementia or low cognitive functioning only female gender, older age and, to some extent, higher levels of wealth and functional limitations were associated with higher quality of life. The analysis of interactions between cognitive status and having a partner (Wald test = 4.81, df = 2, p = 0.008), occupational social class (Wald test = 3.64, df = 4, p = 0.006), home ownership (Wald test = 8.72, df = 2, p < 0.001), and wealth (Wald test = 9.37, df = 2, p < 0.001) showed that statistically significant associations were most evident among those with intact cognition (see the estimates and marginal effects for the interactions in Supplementary Table 3a and 3b). The interaction between cognitive status and functional limitations showed that poorer cognitive status was associated with lower quality of life at lower levels of functional limitations but this association disappeared at higher levels of functional limitation (Wald test = 4.02, df = 2, p = 0.018) (Figure 3).