A nationwide Swedish study of age at retirement and dementia risk

The aim of this nationwide study was to examine the association between age at retirement and dementia risk, with a follow‐up period of up to 24 years.


| INTRODUCTION
It has been argued that an enriched environment throughout life may provide an individual with a higher cognitive reserve that may compensate for age-related cognitive decline and consequently postpone onset of dementia. 1,2 Suggested proxies for cognitive reserve include educational attainment, leisure activities, and social activities. A factor that is highly relevant to the cognitive reserve hypothesis is an individual's work life, but this factor has so far received less attention.
There are several reasons why job-related factors may be potential moderators of age-related cognitive decline and dementia onset. First, we typically spend a large portion of our adult life at the work place, and many people receive majority of their social contacts and mental stimulating activates from their works. Second, this type of activity (in modern society) declines, often abruptly, as we retire. At this point, it is relevant to point out that the typical timing of pension/retirement actually coincides with the average age of onset of decline that has been observed across a variety of process-based (or "fluid" as opposed to "crystallized") cognitive abilities, is discernible around age 60 or 65 years. 3,4 The potential protective effect of staying active longer for dementia is of fundamental importance also from a socioeconomic perspective.
We may need to work longer to cope with increased costs associated with population aging seen in Western countries, and this issue have consequently been addressed in a few previous studies. [5][6][7][8] For example, Lupton et al 5 conducted a retrospective study, based on probable AD

| Participants
The sample included all individuals born in 1930 that where alive in the year 1990. We excluded persons 17 514 who earned zero income due to unemployment or other reasons, 17 with dementia diagnosis prior to 1990, and 154 whose education level was missing, leaving a final sample of 63 505 individuals. Persons who were unemployment were excluded since this study aimed to examine the effect of leaving paid work on dementia.

| Retirement variable
Timing of retirement is defined as the first year when income from pension benefits is equal to or exceeds 50% of the total annual earnings (including employment income, income from self-employment, and disability benefits). 14 Pension benefits are the sum of income from old-age public pension, occupational pensions, early pension, or voluntary private pension. The reason for using this definition of retirement is that timing of retirement in Sweden is not always clearcut. For many people, retirement is a gradual transition, and some people continue to work part-time.

| Diagnoses of dementia and death
Dementia diagnoses were retrieved from both the National Patient Register and the Cause of Death Register. These registers have been shown to provide a high specificity for detecting dementia and moderate sensitivities (ie, missing dementia cases). 15,16 The following codes from ICD-10 were used for to confirm the dementia diagnosis; F00 (Alzheimer's disease), F01 (vascular dementia), and F03 (dementia of

Key points
• Prior research has suggested that the transition from work to retirement can lead to reduced cognitive stimulation, which may have a deteriorating effect on a person's cognitive health and indicate an increased risk of subsequently receiving a dementia diagnosis.
• This study showed that later-than-average age of retirement was associated with a decreased risk of dementia.
• These findings supports the idea that higher age at retirement is associated with reduced risk of dementia. However, it also highlight the complexity of retirement's relationship to dementia and suggest that underlying factors, such as premorbid cognitive level and genetic predisposition, may have influenced the findings and need to be considered in future studies before any causal inferences are drawn. unspecified type). Diagnosis of dementia from ICD-8 and ICD-9 (code 290) was also included to find dementia cases before 1997. Date of diagnosis of death (all causes) was derived from death records in the Swedish Cause of Death Registry.

| Covariates
In addition to age and sex, several sociodemographic variables collected at baseline (1990) were considered as potential confounders. The sociodemographic factors included were education (classified as low [≤ 9 years], intermediate [10-12 years], and high [≥13 years]), marital status (married vs not married), and employment (defined as in blue and white collar, self-employed, and unknown). We also adjusted for previous history of cardiovascular diseases (CVD) during the years 1980-1990.
CVD was defined by the first hospitalization caused by coronary heart disease, stroke, or heart failure and were coded according to the ICD-8 and ICD-9 codes (410-414, 428, 430-438, and 440-448).

| Statistical methods
Differences in background characteristics between participants who developed dementia or remained dementia-free were tested used chi-square tests (categorical variables) and student's t-test (continuous variables).
We estimated the risk of developing dementia during follow-up by retirement age by using competing risk regression (CRR) modeling treating mortality as a competing event. 17 In competing risk data, an individual can potentially fail from several competing events, here, the competing events are dementia and death. In such analysis, an individual who die is no longer at risk of dementia and is treated essentially as censored, in opposed to the ordinary Cox regression where censoring is assumed unrelated to an individual's risk of developing dementia (ie, censoring is uninformative). 18 For the CRR models, proportional hazards (PH) assumption was assessed by fitting models including time-by-covariate interactions for covariates showing a nonrandom patterns against time by plotting Schoenfeld's residuals. Significant interactions were found to for sex, marital status, and history of cardiovascular disease and were thus included in the final model to account for the violation of the PH assumption. A significant parameter estimate indicates that the hazard ratio associated with this factor is not constant over time.
Time to event was calculated as date of entry into the study (1990) to the year of dementia diagnosis, being lost-to-follow-up, death, or date of final follow-up (2014), depending on which event came first. Statistical analysis was performed using the cmprsk package in R (R Core Team, 2018). 18,19 3 | RESULTS During the 24-year follow-up, 5181 (9.6%) individuals had developed dementia and 21 065 (39.1%) died without a dementia diagnosis.
Mean (SD) time of censoring was 20.9 (5.2) years and mean time to dementia diagnosis was 19.2 (4.4) years. Those who had developed dementia at follow-up were significantly more often women, less likely to be married, less highly educated, and less likely to have white-collar jobs at baseline. They were also more likely to have died at follow-up, see Table 1.
The distribution of age at retirement in this sample is shown in Figure 1. Most persons (44.1%) retired at age 65 years old, an age that corresponds well with the most common retirement age in Sweden.
We first evaluated the association between retirement age and risk of dementia for the full sample. A competing risk regression model, adjusting for sex, showed a trend, although not significantly, toward decreased risk of dementia with increase age of retirement (HR = 0.99, 95% CI 0.97-1.00, P = .065). Additional adjustment for marital status, education, employment, history of cardiovascular disease, and covariate-by-time interactions for sex, marital status, and history of cardiovascular disease revealed a significant association, whereas increase age of retirement with 1 year was associated with a 1.02-fold lower risk of dementia (HR = 0.98, 95% CI 0.97-1.00, P = .045), see Table 2.
We thereafter analyzed the three different categories of retirement age: early (aged ≤64 years of age), usual (65 years of age; reference category), and late (≥ 66 years of age). In a fully adjusted analysis, the results show that those who retired at 66 years and over had a 2.9-fold lower risk of dementia (HR = 0.35, 95% CI 0.24-0.49, P < .001), than those who retired at 65 years. However, the effect was attenuated over time as indicated by a significant time-by-covariate interaction for late retirement age (HR = 1.05, 95% CI 1.03-1.07, P < .001). This suggests that the strong positive effect of later retirement age on dementia risk was more pronounced at younger ages and decreases over time. There was no significant association among those who retired earlier (between 61 and 64 years of age). The results are found in Table 3.
To take into account the possibility that retirement was in part a result of dementia, a sensitive analysis was done by excluding individuals diagnosed with dementia or died within 3 years after retirement (N = 1528). The results yield similar to those of the primary analyses, in that later retirement was associated with a decreased risk of dementia (HR = 0.78, 95% CI 0.76-0.80, P < .001). However, the effect was attenuated over time as indicated by a significant time-bycovariate interaction for retirement age (HR = 1.01, 95% CI 1.01-1.01, P < .001). This suggests that the positive effect of later retirement age on dementia risk was more pronounced at younger ages and that after the age of 80 years only small and nonsignificant effects were found.

| DISCUSSION
The present study examined whether age of retirement is associated with incidence of dementia in a nationwide Swedish cohort study with up to 24 years of follow-up. After controlling for several potentially confounding factors (ie, sex, education, marital status, occupation, and previous history of cardiovascular diseases), we found age of retirement to be associated with decreased risk of dementia.
According to the literature, a later age of retirement is related to decreased dementia risk, including later onset of AD, 5-8 and better preserved cognitive function. 20 The results from the present study, showing that a higher age of retirement is associated with a reduced risk of dementia, are consistent with these earlier findings. A common explanation for why delaying retirement may be beneficial for cognitive health is that retirement may be associated with a drop in mental and social stimulation. Accordingly, it has been proposed that people who work for a longer time keep up their cognitive level for a longer period, thereby reducing the risk of dementia diseases or delaying its onset. [5][6][7][8] As such, the results are consistent with a "use it or lose it" perspective and the cognitive reserve hypothesis. 2,21 However, before strong conclusions related to causality can be drawn, further research focused on potential mediating factors and including additional uncontrolled factors is required. With regard to Baseline characteristics of participants who remained or did not remain dementia free at follow-up The complexity of one's work is a suggested mediator of the association between retirement and cognitive change; it has been proposed that retirement has a more adverse effect on cognitive performance among individuals with more complex occupations, 25 although recent work from our research group 26 30 The low incentive to retire T A B L E 2 Cox regression competing risk estimates of the association between age at retirement and dementia, fully adjusted model late may result in selection bias in that those who still continue to work past the average age of retirement have better health and also find their work more meaningful than those who retire at a more typical age-factors that may be protective of dementia. A recent study from Sweden also supports these lines of reasoning in that those who continued working after age 65 years were found to have a 6.8% greater chance of reporting better health (subjective) than those who retired at 65 years. However, the reported positive effect on subjective health was only transitory and disappeared after 6 years. The authors suggest that the disappearance of the beneficial effect on subjective health depends on the idea that the positive impact of work environment in the form of social contact and interaction is no longer apparent. 31 A similar pattern was observed in this study, as the positive effect of later retirement age decreased over time.
There are several strengths of the present study. A major strength is the large population-based sample that includes a complete Swedish birth cohort, thereby minimizing the risk of selection bias and allowing for a powered analysis. A follow-up time of up to 24 years and the possibility of adjusting for multiple possible confounding variables are also strengths. However, the database that was used has its limitations; for one thing, there is a lack of information concerning occupational complexity and mental and social leisure activities, both before and after retirement. In addition, all dementia diagnoses were provided by the national registry covering data from specialist healthcare at the hospital, but did not include diagnoses made in primary care, likely resulting in missed dementia cases. Moreover, the results need to be viewed in a national context. More specifically, countries differ in pension systems, employment policies, labor market conditions, possibilities to engage in activities outside the workforce, and cultural norms. It is possible that such variations moderate the effect of timing of retirement on dementia. Finally, in our study we defined age of retirement as the first year when income from pension benefits is equal to or exceeds 50% of total earnings. However, other definitions of age of retirement may have produced different results. This study also included a rather small variation in retirement age; a larger range in retirement age may have provided stronger findings.
In summary, the results from this nationwide study show that a later age of retirement is associated with a decreased risk of dementia even when major demographic and socioeconomic and health-related factors are considered. More research is needed to validate this link and study the underlying mechanisms, such as how job characteristics and involvement in leisure activities, as well as premorbid cognitive level and genetic predisposition, affect the association between retirement and dementia.

ACKNOWLEGEMENTS
The study is part of the research on Ageing and Living Conditions at . This study was approved by the Regional Ethical Committee at Umeå University (Dnr 07-142Ö). The founders had no role in the study design, data analysis, interpretation of data, writing of the report, or the decision to submit the article for publication.