Socioeconomic Position and Late-Onset Dementia: A Nationwide Register-Based Study

Study Population and Design

We used data from Danish nationwide registers and linked the relevant register data on the study population using a unique personal identification number assigned to all permanent residents in Denmark. This unique personal identifier is available in all national registers (Pedersen, 2011).

Due to the relatively low incidence and validity of dementia diagnoses among younger individuals (Salem et al., 2012), we considered individuals at risk of dementia from age 60 years. Thus, as end of follow-up was in 2017, only birth cohorts born ≤1956 were eligible for inclusion. Exposure information from birth cohorts born before 1934 was not available, and thus we included birth cohorts born between 1934 and 1956. We assessed highest attained educational level at age 55 in order to capture this information before the start of the follow-up period. We defined occupation-based SEP as the longest held position (of at least 3 years) during a 10-year period between age 30 and 55. Occupational information was available between 1980 and 1995. Therefore, the specific age range where occupation-based SEP could be assessed depended on the birth year of individuals in the study population (Figure 1).OPEN IN VIEWER

The long preclinical phase of dementia (Jack et al., 2010) could create a spurious relationship between SEP and dementia, as decline in cognitive ability is likely to affect attachment to the labor market. Therefore, we postponed start of follow-up at least 5 years from the latest obtained exposure assessment in order to reduce the risk of reverse causation, that is, that subclinical dementia pathology influenced educational and occupational choices (Figure 1).

We identified individuals eligible for inclusion based on the following criteria: (1) Survived, not having emigrated or been diagnosed with dementia up until age 60 years; (2) having non-missing information about exposures (educational attainment and occupation-based SEP), outcome (dementia), and the following covariates: age, sex, marital status, and comorbidities; (3) being registered as employed for longer than being registered as unemployed in a job with the same occupation-based SEP for at least 3 years. We did not include assisting spouses of self-employed individuals. The final study population consisted of 1,210,720 individuals (Figure 2).OPEN IN VIEWER

Educational Attainment

We obtained information on education from the Population Education Register (PER). PER contains information on the highest attained education of the entire Danish population from 1981 onward. The categories of highest attained education are based on the Danish version of the International Classification of Education (DISCED) (Jensen & Rasmussen, 2011). Nine original levels of educational attainment were aggregated into three levels based on number of years of education: high (>12 years, i.e., short-term further education, middle-range education, bachelor’s degree, and above), medium (10–12 years, i.e., upper secondary education, business high school, and vocational education and training), and low (≤9 years, i.e., primary school).

Occupation-based SEP

Information on occupation-based SEP was drawn from the Employment Classification Module (ECM) (Petersson et al., 2011). We identified levels of occupation-based SEP through the Danish Occupational Classification (DOC), which was used to classify occupation in the entire Danish population from 1980 to 1995. The main classifications of the DOC for individuals in the workforce were (1) self-employed, (2) assisting spouse to self-employed individual, (3) manager with ≥20 employees, (4) manager with <20 employees or senior official, (5) employee in management, (6) other employees, (7) skilled worker, and (8) unskilled worker. These classifications are based on the employment status in the registers defined as the primary source of income during a calendar year.

We operationalized the original main classifications of the DOC into six levels of occupation-based SEP: (1) High non-manual (manager with ≥20 employees, manager with <20 employees or senior official), (2) medium non-manual (employee in management), (3) low non-manual (other employees), (4) skilled manual (skilled worker), (5) unskilled manual (unskilled worker), and (6) self-employed (Self-employed). Table 1 lists the levels of our operationalization of occupation-based SEP and the corresponding categories in the DOC along with examples of specific occupations within each DOC category. The operationalization of occupation-based SEP was based on the hypothesized level of management responsibilities and/or required skill-level that each of the categories of the DOC represented. This hierarchy was inspired by the overview of occupation-based indicators of SEP by Galobardes et al. (2006b) and previous studies on occupation-based SEP and dementia (Fratiglioni et al., 1993; Prince et al., 2012; Russ et al., 2013). We included a separate level for self-employed individuals, as these have been shown to be difficult to classify according to occupation-based SEP (Galobardes et al., 2006a).

OPEN IN VIEWER

Table 1.

Occupation-Based Socioeconomic Position (SEP), Corresponding Levels of the Danish Occupational Classification (DOC) and Examples of Occupations Within DOC Levels.

Occupation-based SEP	DOC category	Occupations
High non-manual	Manager with ≥20 employees	Manager in agriculture, production, trade, or public services
	Manager with <20 employees or senior official	Manager in public administration, doctor, engineer, office manager
Medium non-manual	Employee in management	Teacher, nurse, social worker, foreman, production and building technician
Low non-manual	Other employees	Office assistant, nursing assistant, shop assistant, bank assistant, transport employee
Skilled manual	Skilled worker	Metal worker, builder, mechanic, construction worker
Unskilled manual	Unskilled worker	Home care worker, hotel and restaurant personnel, cleaner, factory worker
Self-employed	Self-employed	Self-employed within any industry

Assessment of Dementia

We obtained information on dementia from the Danish National Patient Register (NPR) (Schmidt et al., 2015), the Danish Psychiatric Central Research Register (PCRR) (Mors et al., 2011), the Danish Register of Causes of Death (RCD) (Helweg-Larsen, 2011), and the Danish National Prescription Registry (DNPR) (Kildemoes et al., 2011). Based on information from the registers, we defined the date of dementia diagnosis as the date of the first registration or date of the first redemption of an anti-dementia medication, whichever occurred first. We included information on main or secondary diagnoses from NPR and PCRR, as well as on main or contributing cause of death from RCD. Dementia diagnoses are based on WHO’s International Classification of Disease (ICD) criteria from ICD-8 in 1970–1993 and ICD-10 from 1994 onward. From DNPR, we included information on redemption of anti-dementia medication used in the treatment of AD, dementia with Lewy bodies, and dementia in Parkinson’s disease from 1995 and onward. Table 2 lists the ICD codes and Anatomical Therapeutic Chemical (ATC) codes, which defined dementia cases in this study.

OPEN IN VIEWER

Table 2.

Diagnoses of Dementia based on Diagnostic Codes in the International Classification of Diseases (ICD) and Anatomical Therapeutic Chemical (ATC) Codes for Anti-Dementia Medications.

ICD-8	ICD-10 code	ATC code
Alzheimer’s disease: 290.10	Alzheimer’s disease: F00.0–9, G30.0–9	Cholinesterase inhibitors: N06DA02 (Donepezil)
Vascular dementia: 293.09–19	Vascular dementia: F01.0–9	N06DA03 (Rivastigmine) and N06DA04 (Galantamine)
Frontotemporal dementia: 290.11	Frontotemporal dementia: F02.0	Glutamate receptor antagonists: N06DX01 (Memantine)
	Dementia with Lewy bodies: G31.8
Dementia without specification: 290.09–19	Dementia without specification: F03.9, G31.9

Covariates

Based on existing literature, we considered the following covariates potential confounders assessed prior to assessment of occupation-based SEP: age, birth cohort, sex, marital status, cognitive ability, and comorbidities. Information on age, birth cohort, sex, and marital status was obtained from the Civil Registration System (Pedersen, 2011). Marital status was assessed at first year of assessment of occupation-based SEP (see Figure 1). From the Danish Conscription Database, we obtained data on cognitive ability at approximately age 18 for men born in 1939 or later (Christensen et al., 2015). Cognitive ability was measured using the validated cognitive ability test, the Børge Priens Prøve (BPP) (Teasdale et al., 2011). We standardized the BPP score to have a mean of 0 and a standard deviation (SD) of 1. We assessed the presence of comorbidities based on the Charlson Comorbidity Index (CCI) (Charlson et al., 1987). A cumulative CCI score was determined based on diagnoses in the NPR and the PCRR.

Statistical Analyses

We estimated the risk time for each individual in the study population from the date they turned 60 years until registration with dementia, emigration, death, or end of follow-up, whichever came first. We estimated the number of additional dementia cases per 100,000 person years (PY) associated with levels of educational attainment and occupation-based SEP, accounting for the competing risks of emigration and death, using Aalen’s additive hazards model (Rod et al., 2012). The additive model provides estimates of absolute risk differences, conveying the public health impact of our socioeconomic exposures and the number of dementia cases potentially avoidable by intervention (Harper et al., 2010; Rod et al., 2012).

We performed the main analyses by first calculating the additional number of cases for each level of educational attainment and occupation-based SEP in a mutually adjusted model (reference: high educational attainment and high non-manual occupation-based SEP, respectively). Then, we calculated the additional number of cases for different combinations of educational attainment and occupation-based SEP (reference: combination of the highest level of educational attainment and occupation-based SEP, respectively). We estimated the associations using both an unadjusted model and a model including birth cohort, sex, and marital status. Age was the underlying time scale in all models. Additionally, we have plotted the unadjusted cumulative incidence of dementia in each level of educational attainment and occupation-based SEP (see Figure A.1 and A.2 in the Appendix).

We also performed a number of sensitivity analyses. First, we tested whether associations in the main analyses were different for men and women separately. Second, we tested if missing occupational data or unemployment during part of the study period could have affected our results by requiring ≥1, ≥5, and ≥10 years in the same occupation-based SEP (instead of ≥3 years as in the main analyses). Third, we adjusted for comorbidities throughout the life course to test whether this could explain associations observed in the main analyses. Fourth, in a subsample of 446,969 men, we tested whether the results from the main analyses were due to confounding by cognitive ability at conscription.

In order to test the underlying assumption of proportional hazards in Aalen’s additive hazards model, we tested whether the effects of educational attainment and occupation-based SEP varied over the follow-up period in the main analyses. Furthermore, because the educational attainment, occupation-based SEP, and cognitive ability at conscription are likely correlated, we tested for the presence of multicollinearity between these factors. Thus, in the entire study sample, we estimated Pearson correlation coefficients, tolerance levels, and variance inflation factors (VIFs) in a model only including educational attainment and occupation-based SEP (Dormann et al., 2013). We then performed the same tests while also including cognitive ability in the above-mentioned subsample of men. We used SAS 9.4 for data management and descriptive analyses and R for Aalen’s additive hazards model.

Main Findings

In the main analyses, we found a higher number of dementia cases among individuals with lower educational attainment and lower occupation-based SEP. These associations were present when adjusting for other sociodemographic factors and comorbidities. The main analyses also indicated a potential joint effect of low educational attainment and low occupation-based SEP on the number of dementia cases. Occupation-based SEP appeared to have the strongest association with dementia among individuals with low or medium educational level. It should be noted that the association between educational attainment and risk of dementia disappeared with adjustment for cognitive ability at conscription in a subpopulation of men. This adjustment also markedly attenuated the estimated effects of occupation-based SEP so that only unskilled workers had a significantly higher dementia risk compared with those in a high non-manual position.

Comparison with Previous Research

The results from the present study differ from those reported in the previous study by Karp et al. (2004) investigating the association between both the mutually adjusted and joint exposure to low educational attainment and low occupation-based SEP and risk of dementia. When mutually adjusted, the authors only found an association between educational attainment and dementia, and concurrent exposure to low occupation-based SEP was not associated with any additional risk (Karp et al., 2004). However, this previous study was limited by a small study population (n = 931), self-reported measures of educational attainment and occupation-based SEP at old age (≥75 years), and a short follow-up of 3 years (Karp et al., 2004).

Four studies investigating the mutually adjusted associations of both educational attainment and occupation-based SEP with the risk of dementia reported conflicting results. Two studies found an association between low educational attainment, but not low occupation-based SEP, and higher risk of dementia (Russ et al., 2013; Takasugi et al., 2019). Conversely, Rusmaully et al. (2017) only found a higher risk of dementia among individuals with low occupation-based SEP and not among individuals with low educational attainment. However, this association disappeared when adjusting for cognitive performance in midlife, which is a potential mediator between occupation-based SEP and dementia (Rusmaully et al., 2017). Finally, Prince et al. (2012) found a lower risk of dementia among individuals with low educational attainment and no association between occupation-based SEP and dementia. This latter study was based on data from middle income countries, which complicates the comparison with the results of the present study (Prince et al., 2012).

Our findings regarding the independent associations between lower educational attainment and higher risk of dementia may be explained by an effect of education on cognitive reserve (Staff et al., 2004) and cardiovascular risk factors and diseases (de Mestral & Stringhini, 2017). Similar mechanisms could explain the higher risk of dementia among individuals with low occupation-based SEP that we found in this study. Occupations of higher socioeconomic status often involve a higher level of complexity and autonomy at work (Dieker et al., 2019), potentially reducing the risk of dementia by contributing to the cognitive reserve (Hussenoeder et al., 2019). In addition, these types of occupations are less likely to encompass chemical and physical exposures (Dieker et al., 2019), which could otherwise increase the dementia risk (Hasselgren et al., 2018; Jalilian et al., 2018; Rovio et al., 2007; Yan et al., 2016). Differences in the working environment across levels of occupation-based SEP (Dieker et al., 2019) may also explain why individuals with low or medium educational attainment tended to have a lower risk of dementia, if they attained a higher occupation-based SEP. Conversely, the importance of the above-mentioned mechanisms should be considered with caution in light of the reduction of associations between the exposures and dementia risk when adjusting for cognitive ability. Cognitive ability in adolescences is likely a strong predictor of later educational attainment and occupation-based SEP (Foverskov et al., 2019) and should be viewed as an important confounder of the association between our main exposures and risk of dementia. However, cognitive ability at conscription did not explain the entire association between occupation-based SEP and dementia risk in our study.

Strengths and Limitations

The strengths of the study include the large study population of more than 1.2 million individuals, a long follow-up time (mean 10.5 years, maximum 24 years) from age 60 with low loss to follow-up, and the temporally separated assessment of educational attainment and occupation-based SEP during midlife and dementia after age 60. A major strength is the adjustment for cognitive ability at age 18 in a subgroup of men. According to a validation study, clinical dementia diagnoses in the Danish national registers have a high validity, with a positive predictive value of 85.8% (Phung et al., 2007). We also included data on anti-dementia medication in order to use all available sources to identify dementia cases. We did not test whether results were different for different subtypes of dementia for two reasons: First, although the overall validity of a diagnosis of dementia as such is high, the validity of the diagnoses of dementia subtypes in the Danish registers has been reported to be low (Phung et al., 2007) and >73% of dementia patients are registered with dementia without specification (Phung et al., 2010). Second, those identified through the drug register do not have a specific diagnosis (Kildemoes et al., 2011). This leaves us with insufficient data to perform separate analyses for each of the dementia subtypes.

It is up for debate whether the hierarchy in our measure of occupation-based SEP accurately captures this latent construct. Our operationalization is inspired by previous research placing non-manual occupations above manual occupations (Fratiglioni et al., 1993; Prince et al., 2012; Russ et al., 2013). The estimated independent effects of occupation-based SEP reported in our study empirically support a socioeconomic gradient in dementia risk across the proposed levels.

Despite the advantages of using register-based data, such data may also miss non-registered educational activities or suffer from imprecise registration of jobs, which bias the risk estimates toward unity. Furthermore, register data underestimate the general dementia prevalence, as it has been estimated that only 40% of the total number of dementia cases in Denmark are registered with a diagnosis (Jørgensen & Waldemar, 2014). To improve the validity of our outcome, we added information on prescriptions of anti-dementia medication in addition to diagnoses registered during patient contacts at hospitals or on death certificates. However, the underreporting of cases may still have biased our results. One proposed reason for not being diagnosed is that many patients and relatives lack information about the dementia syndrome and possibilities for treatment (Jørgensen & Waldemar, 2014). Dementia cases with short education and/or low level of occupation-based SEP may be less likely to seek out treatment options (Berkman et al., 2011) due to a low level of health literacy (Bostock & Steptoe, 2012). Thus, the underreporting of dementia may be inversely correlated with education and occupation-based SEP (Kokmen et al., 1996) resulting in an underestimation of the rate differences in the present study. The duration of our study period may also have caused an underestimation in rate differences across socioeconomic positions as studies indicate an increased incidence of dementia beyond the maximum age in this study of 83 years (Livingston et al., 2017; Taudorf et al., 2019). This assumed underestimation may be exacerbated by those with higher socioeconomic position being diagnosed earlier as indicated by a recent Danish study (Petersen et al., 2021). Conversely, our findings may also be overestimated due to social inequality in morbidity and thereby contact to the health-care sector, that is, detection bias. Nevertheless, adjusting for comorbidities during the life course only resulted in minor changes in the estimates.

Individuals with poor health could be more likely to become unemployed (the healthy worker effect), in particular when they are in a lower occupation-based SEP (Chowdhury et al., 2017). This could lead to an underestimation of the differences in dementia rates across occupation-based SEP. However, we do not consider this a major issue, as we found similar results when restricting the study population to individuals with the same occupation-based SEP for at least 1, 5, and 10 years.

We adjusted for a wide range of potential confounders including other morbidities than dementia (i.e., CCI), which did not affect the results. We did not, however, adjust for epilepsy, hearing loss, and traumatic brain injuries, as well as parental educational attainment or occupation-based SEP. Additionally, we were only able to adjust for cognitive ability for a subgroup of men in our study population.

In the eligible birth cohorts, there was a slightly higher proportion of missing data on cognitive ability among those with the lowest level of educational attainment and occupation-based SEP compared with their respective reference groups (3 and 5 percentage points more individuals with missing data, respectively). One reason for missing information on cognitive ability is non-participation at conscription due to chronic health conditions. Thus, selection based on health status may have biased the estimated effects from the sensitivity analyses adjusting for cognitive ability. However, in our study, we did not have the data to elucidate the direction and magnitude of this potential source of bias. Additionally, a Danish study has found that men from immigrant backgrounds scored lower on cognitive ability tests at conscription (Teasdale, 2009). Thus, bias due to ethnicity may have affected the results of the cognitive ability–adjusted analyses.

The generalizability of our results is strengthened by the register-based information on educational attainment and occupation-based SEP from the entire working population in Denmark dating back to 1970 and 1980, respectively. However, due to the age-related nature of dementia, we used data on exposure from older birth cohorts, which may not be representative of the current educational system and labor market. The risk estimates adjusted for cognitive ability were limited by only being based on men. This is a particular drawback of the data considering the impact of adjusting for cognitive ability on the risk estimates and the differences in results between men and women in the sex-stratified analyses.

Finally, our large study sample increased the likelihood of type I error. In the main analyses, differences of roughly ≥20 dementia cases were statistically significant and the largest effect size we observed was 78.3 additional dementia cases per 100.000 PY among unskilled workers with low educational attainment compared with high non-manual occupation-based SEP and high educational attainment. For comparison, the incidence of dementia in the entire study sample during the entire study period was 237.6 cases per 100.000 PY. While the estimated differences in this study may seem small, we expect that the actual social gradient in number of dementia cases is larger based on the assumed underreporting discussed above. Furthermore, due to the high economic, societal, and individual burden of dementia, we find it relevant to identify risk factors for even small numbers of additional cases, in order to inform future strategies for prevention of dementia in old age at a population level.