To compare dementia prevalence and how it varies by socioeconomic status (SES) across the USA and England.
Population-based comparative study.
Non-Hispanic whites aged over 70 population in the USA and England.
Data from the Health and Retirement Study and the English Longitudinal Study of Ageing, which are harmonised, nationally representative panel studies. The sample includes 5330 and 3147 individuals in the USA and England, respectively.
Between country differences in age-gender standardised dementia prevalence, across the SES gradient. Dementia prevalence was estimated in each country using an algorithm based on an identical battery of demographic, cognitive and functional measures.
Dementia prevalence is higher among the disadvantaged in both countries, with the USA being more unequal according to four measures of SES. Overall prevalence was lower in England at 9.7% (95% CI 8.9% to 10.6%) than the USA at 11.2% (95% CI 10.6% to 11.8%), a difference of 1.4 percentage points (pp) (p=0.0055). Most of the between country difference is driven by the bottom of the SES distribution. In the lowest income decile individuals in the USA had 7.3 pp (p<0.0001) higher prevalence than in England. Once past health factors and education were controlled for, most of the within country inequalities disappeared; however, the cross-country difference in prevalence for those in lowest income decile remained disproportionately high.
There is inequality in dementia prevalence according to income, wealth and education in both the USA and England. England has lower dementia prevalence and a less steep SES gradient. Most of the cross-country difference is concentrated in the lowest SES group, which provides evidence that disadvantage in the USA is a disproportionately high risk factor for dementia.
This is the first study to compare dementia prevalence across countries using the same survey methodology and the exact same measure of dementia. The surveys have similar sample selection and questionnaire design. We standardise our estimates by age and gender to the English population aged over 70 in 2016. Any differences in overall prevalence across the two countries should represent true differences.
We measure the socioeconomic status (SES) gradient of dementia across four different measures of SES: income; education; wealth and non-housing wealth.
Dementia disproportionately affects the most disadvantaged in both countries, although the gradient is steeper in the USA according to all four measures of SES.
We do not ascertain dementia directly but predict cases using a common battery of measures in English Longitudinal Study of Ageing and Health and Retirement Study. One of the SES measures, education, is also used as a predictor of dementia.
Dementia, a severe and irreversible decline in memory and other cognitive functions, is a major and increasing global health challenge. It is the fifth leading cause of death globally and is one of the most common comorbidities for COVID-19 morbidity.
Americans are more likely to be in poor health than their English counterparts in multiple dimensions, including heart disease and diabetes.
The Global Burden of Disease Study (GBD) reported that in 2017 among those aged over 70, the USA had a lower overall prevalence of dementia at 7.89% compared with the UK at 8.91%.
In this study, we compared dementia prevalence in England and the USA among non-Hispanic whites aged 70+, and how it varied across the SES gradient of each country. Location-specific inconsistencies caused by differences in diagnostic practices were not an issue in our study because we used an identical case definition for dementia, and the surveys in the analysis shared the same design and sampling techniques. More specifically, we used two large surveys, the US Health and Retirement Study (HRS) and the English Longitudinal Study of Ageing (ELSA), that contain a battery of the same demographic, cognitive and functional measures, and we applied the same prediction algorithm in both countries to detect undiagnosed as well as diagnosed cases. We compared dementia prevalence within and across England and the USA using important indicators of SES, specifically: income; education; wealth and non-housing wealth.
Data were extracted from the 2016 and earlier waves of the HRS and ELSA, which are nationally representative biennial surveys of the USA and English populations, respectively.
Our samples are restricted to non-Hispanic whites over the age of 70 years old that live in the community or in nursing homes in 2016. This provides a study sample of 5330 participants in the HRS and 3147 participants in ELSA. We restrict our sample to non-Hispanic whites to ensure estimates are comparable across countries. Summary statistics of both the raw and selected samples are displayed in
No study participants were involved in setting the research question or outcome measures, nor were they involved in any other area of the design, implementation and analysis of the study. There are no direct plans to disseminate the results of the research to study participants.
The HRS included a detailed clinical substudy (ADAMS: Aging, Demographics and Memory Study) of 856 sample members aged 70+ who completed an in-depth in-home assessment of cognitive status conducted by experienced teams at the Duke University Dementia Epidemiology Research Center who diagnosed each participant as normal, cognitively impaired but not demented or demented.
We applied Hurd
Non-response for people unable or unwilling to participate in the survey is important when attempting to estimate dementia prevalence across the population. While attrition exists in both surveys, it is unlikely to significantly affect our estimates since among older ELSA and HRS respondents, there is no statistically significant correlation between attrition and prior health or the SES indicators of education, income and wealth.
We considered four measures of SES: income; education; wealth and non-housing wealth. Income is measured as current household income from all sources. For education, we used total years of schooling. Wealth is measured as the sum of all household reported savings, stocks, bonds, business wealth, other assets and the value of housing assets (eg, properties) after financial debt and mortgage debt has been subtracted. Non-housing wealth is the same measure as wealth but excludes housing assets and mortgage debt and therefore measures wealth that can be more easily converted to cash. Wealth and non-housing wealth are both measured from 4 years prior to minimise reverse causality, as medical expenses associated with dementia are high and may run down wealth.
We created a pooled dataset of the two surveys. In our statistical analysis, we used HRS and ELSA sampling weights to adjust for non-response and for the sampling design of the surveys. To make both within country estimates along the SES gradient and cross-country estimates directly comparable, estimates were age-gender standardised to the English population aged over 70 in 2016 using direct standardisation, categorising the population into 10 groups: five age bands (70–74; 75–79; 80–84; 85–89 and 90+) by gender. We estimated the prevalence of dementia in each country, their difference and compared the prevalence along the four SES gradients. For each estimate presented, we computed the corresponding 95% CI, and for any differences we computed the corresponding p values. For each SES factor, as well as estimating the age-gender standardised prevalence along the gradient, we calculated the Relative Index of Inequality (RII) and Slope Index of Inequality (SII) using generalised linear models (log binomial regression) with logarithmic and identity link functions, respectively. The RII can be interpreted as the relative likelihood of dementia prevalence of those in the lowest SES group compared with those in the highest, and the SII can be interpreted as the absolute effect on dementia probability of moving from the lowest SES group to the highest.
Prevalence of dementia, USA versus England, 2017
England | USA | Difference | P value | |
Age-gender standardised | Age-gender standardised | |||
All | 0.097 (0.089 to 0.106) | 0.112 (0.106 to 0.118) | 0.014 | 0.0055 |
Household income decile | ||||
1 (lowest) | 0.114 (0.089 to 0.139) | 0.187 (0.166 to 0.208) | 0.073 | <0.0001 |
2 | 0.113 (0.090 to 0.136) | 0.141 (0.119 to 0.163) | 0.028 | 0.090 |
3 | 0.124 (0.097 to 0.151) | 0.111 (0.095 to 0.127) | −0.013 | 0.42 |
4 | 0.099 (0.071 to 0.126) | 0.118 (0.099 to 0.137) | 0.019 | 0.26 |
5 | 0.094 (0.072 to 0.116) | 0.086 (0.069 to 0.102) | −0.008 | 0.56 |
6 | 0.098 (0.070 to 0.127) | 0.108 (0.088 to 0.128) | 0.010 | 0.59 |
7 | 0.068 (0.042 to 0.093) | 0.100 (0.078 to 0.122) | 0.032 | 0.060 |
8 | 0.083 (0.053 to 0.114) | 0.082 (0.066 to 0.097) | −0.002 | 0.92 |
9 | 0.082 (0.041 to 0.122) | 0.093 (0.071 to 0.116) | 0.012 | 0.62 |
10 (highest) | 0.059 (0.035 to 0.083) | 0.077 (0.052 to 0.103) | 0.019 | 0.30 |
Years of schooling | ||||
9 or fewer | 0.128 (0.101 to 0.154) | 0.190 (0.162 to 0.218) | 0.062 | 0.0015 |
10 | 0.095 (0.074 to 0.116) | 0.137 (0.109 to 0.165) | 0.042 | 0.018 |
11 | 0.096 (0.077 to 0.115) | 0.109 (0.080 to 0.139) | 0.013 | 0.471 |
12 | 0.071 (0.042 to 0.100) | 0.124 (0.114 to 0.133) | 0.053 | 0.0006 |
13 | 0.061 (0.038 to 0.083) | 0.116 (0.090 to 0.141) | 0.055 | 0.0013 |
14 or more | 0.056 (0.039 to 0.073) | 0.085 (0.076 to 0.093) | 0.029 | 0.0031 |
Household wealth decile | ||||
1 (lowest) | 0.165 (0.132 to 0.198) | 0.187 (0.162 to 0.211) | 0.022 | 0.31 |
2 | 0.117 (0.092 to 0.143) | 0.149 (0.129 to 0.169) | 0.031 | 0.061 |
3 | 0.100 (0.073 to 0.127) | 0.107 (0.091 to 0.122) | 0.006 | 0.68 |
4 | 0.093 (0.071 to 0.115) | 0.115 (0.098 to 0.132) | 0.022 | 0.12 |
5 | 0.110 (0.085 to 0.134) | 0.091 (0.077 to 0.106) | −0.018 | 0.21 |
6 | 0.080 (0.057 to 0.103) | 0.089 (0.074 to 0.103) | 0.008 | 0.55 |
7 | 0.070 (0.046 to 0.094) | 0.103 (0.084 to 0.123) | 0.034 | 0.034 |
8 | 0.092 (0.066 to 0.118) | 0.089 (0.072 to 0.106) | −0.003 | 0.85 |
9 | 0.082 (0.048 to 0.116) | 0.103 (0.084 to 0.123) | 0.021 | 0.28 |
10 (highest) | 0.060 (0.038 to 0.081) | 0.067 (0.051 to 0.084) | 0.008 | 0.58 |
Household non-housing wealth decile | ||||
1 (lowest) | 0.136 (0.103 to 0.168) | 0.201 (0.176 to 0.226) | 0.065 | 0.0019 |
2 | 0.109 (0.084 to 0.134) | 0.137 (0.119 to 0.156) | 0.029 | 0.074 |
3 | 0.123 (0.095 to 0.151) | 0.131 (0.111 to 0.150) | 0.008 | 0.66 |
4 | 0.096 (0.073 to 0.118) | 0.101 (0.085 to 0.116) | 0.005 | 0.71 |
5 | 0.108 (0.082 to 0.134) | 0.107 (0.091 to 0.123) | −0.001 | 0.95 |
6 | 0.093 (0.067 to 0.119) | 0.086 (0.070 to 0.101) | −0.007 | 0.64 |
7 | 0.099 (0.072 to 0.125) | 0.086 (0.070 to 0.102) | −0.013 | 0.42 |
8 | 0.078 (0.051 to 0.105) | 0.090 (0.073 to 0.106) | 0.011 | 0.49 |
9 | 0.058 (0.036 to 0.081) | 0.092 (0.074 to 0.111) | 0.034 | 0.024 |
10 (highest) | 0.063 (0.043 to 0.084) | 0.079 (0.062 to 0.095) | 0.015 | 0.26 |
Sample includes non-Hispanic white population aged 70+ only. The sample size is 3147 participants in England and 5330 participants in the USA. All estimates are age-gender standardised to the overall 2016 aged 70+ white population in England. The difference is calculated as the prevalence in the USA minus prevalence in England. All deciles are calculated within country.
Overall prevalence and prevalence according to four measures of socioeconomic status.
SES Gradient of Dementia, USA versus England, 2017, according to four measures of SES. Absolute and relative inequality shown with 95% CI. The points in this figure represent the mean age-gender standardised dementia prevalence for each country by SES, along with 95% CI for these predictions. The solid lines represent the fitted Slope Index of Inequality (SII: absolute inequality) for each country. The values of the SII and the corresponding Relative Index of Inequality (RII: relative inequality) are listed in the top right of each figure, with 95% CI in brackets. For education, individuals are ranked based on their years of schooling within each country, and as the USA has higher educational attainment, individuals with 14+ years of schooling are the 80th percentile of the US education distribution, but at the 90th percentile of the English educational distribution. SES, socioeconomic status.
Next, we attempted to understand the potential drivers of these gradients and the differences in the gradients across countries. We extended the analysis to account for cardiometabolic diseases (ie, diabetes, heart disease and stroke) and behaviours (ie, smoking and body mass index) as dementia risk factors.
Difference in prevalence of dementia, USA vs England, 2017
Whole sample | |||||
Percentage point difference | 1.43 | 1.34 | 1.15 | 1.21 | 1.18 |
p value | 0.0055 | 0.0091 | 0.025 | 0.020 | 0.034 |
% Difference from baseline | – | −6% | −19% | −16% | −17% |
Excluding lowest income decile | |||||
Percentage point difference | 0.89 | 0.88 | 0.74 | 0.75 | 0.81 |
p value | 0.11 | 0.11 | 0.18 | 0.18 | 0.18 |
% difference from baseline | – | −2% | −17% | −16% | −10% |
Lowest income decile | |||||
Percentage point difference | 7.27 | 6.14 | 5.45 | 5.93 | 4.85 |
p value | <0.0001 | 0.0003 | 0.002 | 0.0011 | 0.012 |
% difference from baseline | – | −15% | −25% | −18% | −33% |
Control for | |||||
Past cardiometabolic diseases | ✓ | ✓ | ✓ | ✓ | |
Past psychiatric conditions | ✓ | ✓ | ✓ | ||
Ever smoked | ✓ | ✓ | |||
Past BMI | ✓ |
Sample includes non-Hispanic white population aged 70+ only. The sample size is 3147 participants in England and 5330 participants in the USA. All estimates are age-gender standardised to the overall 2016 aged 70+ white population in England. The difference is calculated as the prevalence in the USA minus prevalence in England. Differences are displayed as percentage points. ‘Past Cardiometabolic Diseases’ and ‘Past Psychiatric Conditions’ control for whether an individual says they had the conditions 4 years prior. ‘Cardiometabolic Diseases’ include diabetes, heart disease and stroke. ‘Smoking’ controls for whether an individual has ever smoked. ‘Past BMI’ includes dummy variables to control for whether an individual is classed as underweight,
Estimates of the between-country difference taking account of antecedent health status, smoking and degree of adiposity.
BMI, body mass index.
In the
Education has also been shown to be a risk factor for dementia.
In our main analysis, we exclude ethnic minorities, who have higher dementia prevalence and comprise a larger share of the USA than the English population. Including minorities increased the estimated difference across countries, from 1.43 pp to 2.42 pp, as can be seen in the
Using nationally representative samples of older individuals from England and the USA and applying the same algorithmic procedure to predict dementia in both samples, we showed that in both the USA and England dementia is more prevalent among the disadvantaged, and the SES gradient of dementia is steeper in the USA. The steeper gradient in the USA is largely driven by those in the lowest decile. In both countries, most of the SES gradient disappeared when we controlled for past health related factors, although prevalence for those in lowest income decile in the USA remained disproportionately high. If the lowest income decile is excluded from our sample the difference in dementia prevalence across the countries is statistically insignificant, and the remaining SES gradient of dementia is remarkably similar across countries.
While poorer individuals face a higher burden of dementia in both England and the USA, the extremely poor in the USA face a disproportionately high burden of dementia. Controlling for past health-related factors can explain some, but not all, of the cross-country difference. It can explain up to around one third of the difference for those in the lowest income decile. While past health factors such as adiposity and smoking are correlated with dementia, those in the lowest income decile in the USA do not smoke more or have higher body mass indexes (BMIs) than their English counterparts. Therefore, this cannot explain their disproportionately higher prevalence of dementia. Education also cannot explain the difference, as the US population is more educated at every income decile, and in fact the educational difference masks some of the underlying dementia risk difference between countries.
Adding minorities increased estimated dementia prevalence, especially in the USA, because dementia prevalence is higher among minorities, who comprise a higher share of the USA than the English population. This fits with prior research which showed dementia prevalence is higher for non-whites and Hispanics.
Previous studies have shown cross-country variation in dementia prevalence. However, substantial heterogeneity in case-ascertainment methods across countries and studies makes interpreting any observed differences difficult. We believe this is the first study to compare dementia prevalence in England and the USA using the exact same measure of dementia, thus overcoming previous difficulties in making comparisons across the two countries due differences in diagnostic practices and case definitions. We also compared the SES gradient of dementia in both countries. While some studies have shown in both England and the USA those with lower education and less wealth have been found to have higher rates of dementia,
We showed that risk factors for dementia such as cardiometabolic diseases, psychiatric conditions, high BMI, smoking have similar affects across countries. Accounting for these risk factors removes most of the SES gradient for both countries, but disproportionately high prevalence remains for the most disadvantaged in the USA.
Much research has shown that low-income Americans are more likely to be in poor health and die younger than their high-income counterparts.
While risk factors contribute to higher prevalence among those who are more disadvantaged, those in the USA appear to have an undue burden that is caused by risk factors for which we cannot account. One possible explanation is differential access to healthcare. The NHS provides broadly equitable care according to education in the older population after accounting for health status.
The implications of our results are that interventions designed to attempt to prevent dementia should be targeted towards the most disadvantaged. This is especially true in the USA. As yet, we are unable to advocate specific measures as we do not yet understand the specific nature of disadvantage in respect to dementia risk.
This study has a strong design. Results are directly comparable across England and the USA. The same predictive algorithm was applied to both countries, addressing the problem of heterogeneity in case ascertainment which has affected the literature.
This study had three limitations. First, we do not ascertain dementia directly, but predict cases using a common battery of measures in ELSA and HRS. Importantly, Hurd’s prediction algorithm has high accuracy and although our case definition lacks a clinical point of reference in England, it is based on a detailed clinical substudy in the USA.
Given the large social and economic costs of dementia, there is great value in understanding the scope and burden of dementia in the population along the SES gradient. This study indicates that more disadvantaged individuals face a higher burden of dementia and that the poorest individuals in the USA face a disproportionately high burden. The high burden faced by these individuals can be partly but not fully explained by past health factors. We lacked data on other important possible contributing factors such as habitual drug use. Further research is needed to fully understand this issue using data from multiple sources.
All authors made a substantial contribution to study conception and design. EF and JM developed the original idea. JM and KA prepared and analysed the data, with inputs from EF and EB. All authors contributed to interpreting the results and drafting the manuscript. JM and KA serve as guarantors and affirm that the manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted and that any discrepancies from the study as planned have been explained. All authors had full access to all of the data (including statistical reports and tables) in the study and take responsibility for the integrity of the data and accuracy of the data analysis. criteria and that no others meeting the criteria have been omitted.
This work was supported by grants from the British Heart Foundation (RG/13/2/30098, RG/16/11/32334), Michigan Retirement Research Center, ESRC, and the Alexander S. Onassis Foundation.
None declared.
Not commissioned; externally peer reviewed.
This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Data are available in a public, open access repository. Appendix with details on the formula and calculations is available online. Code and results are available on request from the corresponding author. Appendix with details on the formula and calculations is available online. Code and results are available on request from the corresponding author.
Not required.