Articles

Population attributable fractions for risk factors for

dementia in low-income and middle-income countries:
an analysis using cross-sectional survey data
Naaheed Mukadam, Andrew Sommerlad, Jonathan Huntley, Gill Livingston

Summary
Background Nine potentially modifiable risk factors (less childhood education, midlife hearing loss, hypertension, Lancet Glob Health 2019;
and obesity, and later-life smoking, depression, physical inactivity, social isolation, and diabetes) account for 35% of 7: e596–603

worldwide dementia, but most data to calculate these risk factors come from high-income countries only. We aimed See Comment page e538
to calculate population attributable fractions (PAFs) for dementia in selected low-income and middle-income Division of Psychiatry,
University College London,
countries (LMICs) to identify potential dementia prevention targets in these countries.
London, UK (N Mukadam PhD,
A Sommerlad MSc,
Methods The study was an analysis of cross-sectional data obtained from the 10/66 Dementia Research surveys of J Huntley PhD,
representative populations in India, China, and six Latin America countries (Cuba, Dominican Republic, Mexico, Prof G Livingston MD)
Peru, Puerto Rico, and Venezuela), which used identical risk factor ascertainment methods in each country. Between Correspondence to:
2004 and 2006 (and between 2007 and 2010 for Puerto Rico), all residents aged 65 years and older in predefined Dr Naaheed Mukadam, Division
of Psychiatry, University College
catchment areas were invited to participate in the survey. We used risk factor prevalence estimates from this London, London W1T 7NF, UK
10/66 survey data, and relative risk estimates from previous meta-analyses, to calculate PAFs for each risk factor. To n.mukadam@ucl.ac.uk
account for individuals having overlapping risk factors, we adjusted PAF for communality between risk factors, and
used these values to calculate overall weighted PAFs for India, China, and the Latin American sample.

Findings The overall weighted PAF for potentially modifiable risk factors for dementia was 39·5% (95% CI 37·5–41·6)
in China (n=2162 participants), 41·2% (39·1–43·4) in India (n=2004), and 55·8% (54·9–56·7) in our Latin American
sample (n=12 865). Five dementia risk factors were more prevalent in these LMICs than worldwide estimates, leading
to higher PAFs for dementia: less childhood education (weighted PAF of 10·8% in China, 13·6% in India, and
10·9% in Latin America vs 7·5% worldwide), smoking (14·7%, 6·4%, and 5·7%, respectively, vs 5·5% worldwide),
hypertension (6·4%, 4·0%, and 9·3%, vs 2·0%), obesity (5·6%, 2·9%, and 7·9%, vs 0·8%), and diabetes (1·6%,
1·7%, and 3·2%, vs 1·2%).

Interpretation The dementia prevention potential in India, China, and this sample of Latin American countries is
large, and greater than in high-income countries. Less education in early life, hypertension, hearing loss, obesity, and
physical inactivity have particularly high PAFs and could be initial targets for dementia prevention strategies.

Funding No funding.

Copyright © 2019 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND
4.0 license.

Introduction the nine potentially modifiable risk factors for dementia

Around two thirds of people with dementia worldwide live that had been identified by the UK National Institute
in low-income and middle-income countries (LMICs). The for Health and Care Excellence6 and the US National
number of people with dementia in these countries is Institutes of Health.7 We found that 35% (95% CI
predicted to rise more rapidly than in higher-income 34·1–35·9) of dementia was theoretically preventable
countries because an increasing number of people in through elimination of these risk factors;2 namely, less
LMICs are living to an older age.1 However, potentially education in childhood, hearing loss, hypertension, obesity
modifiable risk factors for dementia might drive or moder­ in midlife (age 45–64 years), depression, social isolation,
ate this increase, as they have in many higher-income physical inactivity, diabetes, and smoking in later life (age
countries where falling age-specific dementia incidence >65 years). To calculate this percentage, we used meta-
and prevalence have been reported,2,3 related to higher analyses of relative risk (RR) for dementia associated with
levels of education4 and reduced cardiovascular morbidity.5 each risk factor and prevalence of that risk based on
Population attributable fractions (PAFs) estimate the summary global prevalence estimates,8 which were studied
proportion of disease cases that would not occur in a predominantly in high-income countries. Whether PAF
population if an individual risk factor were to be eliminated. estimates are applicable to LMICs is therefore unclear,
We previously reported the overall PAF for dementia using given that the prevalence of some risk factors are likely to

www.thelancet.com/lancetgh Vol 7 May 2019 e596

 Articles

Research in context
Evidence before this study dementia and then we calculated the percentage of potentially
We searched PubMed on March 21, 2018, for any studies preventable dementia cases in these regions. To the best of
investigating population attributable fractions (PAFs) or our knowledge, this is the first such calculation. We searched
preventable dementia in low-income and middle-income the literature for other studies of risk factor prevalence for
countries (LMICs) using the search terms “dementia”, dementia and used the studies with the highest and lowest
“preventable”, and “low income”, with no limits on language or prevalence estimates in sensitivity analyses to calculate
date of publication. We found no papers investigating overall highest and lowest possible PAF estimates.
dementia risk, although there were some articles investigating
Implications of all the available evidence
individual risk factors and their link with dementia in LMICs.
The PAF percentage is higher in India, China, and our
We have previously published estimates of PAFs for dementia
Latin American sample of countries than worldwide estimates,
risk factors based on worldwide prevalence estimates from
indicating greater potential for dementia prevention in these
meta-analyses (primarily those that included studies from
regions. Low education was the risk factor with the highest
higher-income countries), and found that up to 35% of
PAF in all three regions, but other priority risk factors are
dementia cases are potentially preventable worldwide.
hypertension in China and Latin America and smoking in India.
Added value of this study
We used data from the 10/66 Dementia Research Group from
eight LMICs to estimate the prevalence of nine risk factors for

differ in these regions because of different educational risk factor are presented in panel 1. We dealt with missing
policies, health behaviours, genetic predisposition, and data by case-wise deletion given that the proportion of
health-care practice. PAF in LMICs might therefore be missing data was very low (0–2%) for all variables, except
higher or lower than in high-income countries, resulting social contact, which had 10–12·7% missing data.
in specific priority targets for dementia prevention in Some factors that are risks in middle age (eg, hyper­
particular countries. In this study, we aimed to estimate the tension and obesity) decrease before the onset of dementia
PAF of potentially modifiable risk factors for dementia in as part of the developing illness, and therefore attributing
LMICs for which we could obtain population-level data. risk factors to specific time periods within the lifecourse
is essential.2 However, for hypertension and obesity,
Methods prevalence data for midlife were not available. We there­
Data fore used the prevalence of pre-existing hypertension
The study was an analysis of cross-sectional data and obesity at the time of the survey as an estimate of mid­
For more on the 10/66 Dementia obtained from the 10/66 Dementia Research Group life prevalence. Although the 10/66 Dementia Research
Research Group see comprehensive one-phase prevalence surveys,9 given Group sample contained individuals who were older than
https://www.alz.co.uk/1066/
that these surveys reported data on the nine identified 65 years, 60% of those were younger than 75 years, and so
potentially modifiable risk factors for dementia. The for a substantial proportion of the sample, the diagnosis of
10/66 Dementia Research Group aimed to interview all hypertension and the development of obesity are likely to
residents aged at least 65 years in geographically have occurred during midlife. Obesity was measured by
defined catchment areas in selected LMICs, and used waist circumference (which is con­sidered to be a more
cross-culturally validated assessments for dementia.9 valid measure of obesity-related risk than body-mass
Participants were inter­viewed between 2004 and 2006 index)13 and was favoured over waist-to-hip ratio because of
(and between 2007 and 2010 in Puerto Rico). These data the relative ease of obtaining this measurement.11
are available for 17 031 participants in eight countries The PAF for a risk factor is defined as the percentage of
(Cuba, Dominican Republic, Mexico, Peru, Puerto Rico cases of a disease that would be eliminated if that
and Venezuela [grouped hereafter as the Latin American particular risk factor was eliminated. The value of the
sample]; China; and India). Sample size varied between PAF depends on the prevalence of the risk factor and the
1900 and 3000 for each country and more than 80% of strength of its association (RR) with the disease.
the target population responded in all areas surveyed.
Statistical analysis
Risk factor prevalence We calculated the prevalence estimates of the nine risk
We calculated prevalence for the risk factors of interest factors for each of the eight countries using
using 10/66 Dementia Research Group data,9 aiming 10/66 Dementia Research Group data. Given that the
to group them into early-life (age <45 years), midlife prevalence estimates were similar in each Latin
(age 45–64 years), and later-life (age ≥65 years) risk factors American country, we grouped these into one estimate
on the basis of previous literature10 that linked these for the Latin American sample by calculating the
factors to these particular age groups. Definitions of each prevalence from the combined data. As in our previous

e597 www.thelancet.com/lancetgh Vol 7 May 2019

 Articles

Panel 1: Definitions of risk factors from 10/66 Dementia Panel 2: Standard method for the calculation of population attributable fractions and
Research Group data9 communality10
Less education Formula for individual population attributable fraction
Not receiving more than primary education in early life Population attributable fraction (PAF)=Pe(RRe – 1) / (1 + Pe[RRe – 1]), in which Pe is the
prevalence of the exposure and RRe the relative risk of disease because of that exposure.
Hearing loss
Self-reported hearing impairment Calculation of communality
Input data for all nine risk factors into our model.
Hypertension
Self-reported known diagnosis of hypertension Calculate the tetrachoric correlation to generate correlation coefficients and a correlation
matrix. This calculation establishes the correlation between unobserved and latent
Obesity variables and observed dichotomous variables.
Waist circumference measured by the 10/66 research team,
of at least 88 cm in women and more than 102 cm in men, Do a principal component analysis on the correlation matrix to generate eigenvectors,
according to WHO guidelines11 which are directions mapped onto the datapoints from which variance to the data is
measured. These eigenvectors represent unobserved factors underlying all the variables
Smoking that explain the variance observed.
Self-reported smoking in later life
Components with eigenvalues of at least 1 were retained in the model, as is standard
Depression practice, so that only eigenvectors that hold the most information about the data
Diagnosis of depression (according to the fourth edition of distribution are retained.
the Diagnostics and Statistical Manual of Mental Disorders) in Communality was calculated as the sum of the square of all factor loadings (ie, how much
later life following a structured Geriatric Mental State12 each unobserved component explained each measured variable).
interview or self-report of previous depression
Calculation of overall PAF
Physical inactivity We then calculated overall PAF:
Self-report of being either not at all or not very physically PAF=1 – [(1 – PAF1)(1 – PAF2)(1 – PAF3)…]
active in later life on a four-point Likert scale of level of physical
activity (categories are not at all, not very, fairly, and very Each individual risk factor’s PAF was weighted according to its communality using the
physically active) formula:
Weight (w)=1 – communality
Social isolation
Weighting was included in the calculation of overall PAF using the formula:
Social contact occurring less than once per month in later life,2
PAF=1 – [(1 – w*PAF1)(1 – w*PAF2)(1 – w*PAF3)...]
calculated using pooled self-reported contact frequency with
friends, relatives, and neighbours or attendance at social clubs
Diabetes We did sensitivity analyses to estimate variability in
Self-report of known diagnosis of diabetes in later life PAF depending on prevalence of each risk factor,
by using estimates from other high-quality studies
(appendix) of risk factor prevalence. We did a literature
study, we used previously published meta-analyses of search in PubMed for articles published from database
RR for individual risk factors.8 We then adjusted the inception to June 8, 2018, using the search terms for
results for com­ munality (ie, overlap between risk each risk factor, the country or region of interest, and the
factors). age group of interest—eg, “China”, “older adults”, and
The formula used to calculate PAF is available in “hypertension” or “high blood pressure”. We did not
panel 2. We chose to use RRs for risk factors on the basis restrict our searches by language or date of publication.
of previous meta-analyses.2,8,10 Details of how we In line with guidance14 for evaluating prevalence
calculated overall PAF are shown in panel 2. People studies, we defined a prevalence study as one that used
might have several risk factors and individual PAFs probability sampling with at least 70% response and
cannot therefore be summed to get the total PAF; thus, it reported a prevalence for the specified risk factor from
is important to consider communality and to calculate a the relevant countries, using valid tools for measuring
weighted PAF, taking communality into account. each risk factor.14 We used the lowest and highest
Individual weighted PAFs were calculated with the prevalence figures found to calculate possible lowest and
following formula: highest PAFs. In cases in which we found no differing
prevalence estimates for a specific risk factor from
individual weighted PAF= population surveys, we used 10/66 Dementia Research
Group prevalence figures in the sensitivity analyses to
(individual PAF) calculate the overall PAF. Therefore, in these calculations
× (overall PAF)
∑(individual PAF) the weighted prevalence still changed in our sensitivity
analyses despite using the same prevalence figures,

www.thelancet.com/lancetgh Vol 7 May 2019 e598

 Articles

Latin America* India China

Early life Early life Early life

Low education
11% Low education 14% Midlife 11% Low education
Midlife Midlife

Hearing loss Hearing loss

Hearing loss
8% 6% Hypertension 4%
Hypertension 4% Hypertension
9% 6%
3% Obesity
8% Obesity 6% Obesity

Later life Later life Later life

6% Smoking
6% Smoking 4% Smoking
7% Depression 2% 0·5%
5% Depression Physical inactivity 6% Depression
Physical inactivity 2% Physical inactivity
Social isolation 0·1% 0·7%Social isolation
Social isolation 2%
3% Diabetes 2% Diabetes Potentially 2% Diabetes Potentially
Potentially
modifiable modifiable modifiable
56% 41% 40%

Percentage reduction in cases

5 of dementia if this risk factor
is eliminated

Figure: Population attributable fractions for potentially modifiable risk factors in low-income and middle-income countries
*Our data for Latin America include the data for Cuba, Dominican Republic, Mexico, Peru, Puerto Rico, and Venezuela.

RR for dementia Risk factor Communality PAF Weighted

Role of the funding source
(95% CI) prevalence2,8,10 PAF* The funder of the study had no role in study design, data
collection, data analysis, data interpretation, or writing
Early life (<45 years)
of the report. The corresponding author had full access
Low education 1·6 (1·3–2·0) 75·9% 72% 31·3% 10·8%
(29·4–33·3) (9·5–12·2) to all the data in the study and had final responsibility
Midlife (45–64 years) for the decision to submit for publication.
Hearing loss 1·9 (1·4–2·7) 14·3% 46% 11·4% 3·9%
(10·1–12·8) (3·2–4·8) Results
Hypertension 1·6 (1·2–2·2) 38·1% 52% 18·6% 6·4% In this analysis, we used data from 12 865 respondents in
(17·0–20·3) (5·4–7·5) Latin America, 2162 respondents in China, and 2004
Obesity 1·6 (1·3–1·9) 32·0% 65% 16·1% 5·6% respondents in India to calculate PAFs. The median age
(14·6–17·7) (4·7–6·7)
was 73 years (range 65 to 110) and 62·5% of the sample
Later life (≥65 years)
was female.
Smoking 1·6 (1·2–2·2) 23·0% 34% 14·7% 4·2% The proportion of dementia cases that were theoretically
(13·3–16·3) (3·4–5·1)
preventable through elimination of the nine identified
Depression 1·9 (1·6–2·3) 1·5% 55% 1·3% 0·5%
(0·9–1·9) (0·2–0·9) risk factors (ie, overall weighted PAF) was 55·8% (95% CI
Physical inactivity 1·4 (1·2–1·7) 50·7% 55% 23·3% 5·8% 54·9–56·7) in Latin America, 39·5% (37·5–41·6) in
(21·6–25·1) (4·9–6·8) China, and 41·2% (39·1–43·4) in India (figure). Risk
Low social contact 1·6 (1·3–1·9) 3·4% 62% 1·3% 0·7% factor prevalence, communality, and weighted PAFs
(0·9–1·9) (0·4–1·1) for each region are presented in tables 1–3. For
Diabetes 1·5 (1·3–1·8) 9·4% 52% 4·5% 1·6% comparison, the previously published worldwide risk
(3·7–5·5) (1·2–2·2)
factor prevalence, communality, and weighted PAF are
Overall weighted PAF ·· ·· ·· ·· 39·5% presented in table 4. We have provided risk factor
(37·5–41·6)
prevalence esti­mates for each individual Latin American
PAF=population attributable fraction. RR=relative risk. *Weighted PAF is the relative contribution of each risk factor to
country and the associated PAFs in the appendix. The
the overall PAF when adjusted for communality.
PAFs for the Latin American countries in our sample
Table 1: PAF for dementia risk factors in China (n=2162) were: 52·7 (95% CI 50·9–54·5) for Cuba, 54·3 (52·1–56·4)
for the Dominican Republic, 50·2 (47·9–52·4) for Peru,
See Online for appendix because of the weighting adjustment. If other factors 54·8 (52·6–57·0) for Venezuela, 55·5 (53·5–57·7) for
were less prevalent, then a risk factor with unchanged Mexico, and 53·2 (50·9–55·3) for Puerto Rico. Less
prevalence contributed a relatively larger risk to the childhood education, smoking, hypertension, obesity, and
overall PAF (and the inverse for the highest PAF diabetes were more common than worldwide estimates
estimate). across all three regions, leading to higher PAF.

e599 www.thelancet.com/lancetgh Vol 7 May 2019

 Articles

The range of estimates for the overall PAF combining

RR for dementia Risk factor Communality PAF Weighted
all nine risk factors and using highest and lowest risk (95% CI) PAF*
prevalence2,8,10
factor prevalence estimates was 31–55% in China,
34–45% in India, and 44–58% in the Latin American Early life (<45 years)

sample. In particular, we found a trend for better Low education 1·6 (1·3–2·0) 92·2% 44% 35·6% 13·6%
(33·5–37·7) (12·2–15·2)
educational attainment over time, which reduces the
Midlife (45–64 years)
PAF for less education in each region. Variability in
Hearing loss 1·9 (1·4–2·7) 22·3% 37% 16·7% 6·4%
prevalence estimates were due in part to variations in (15·2–18·4) (5·4–7·6)
measurement (eg, in hearing loss) and in part due to the Hypertension 1·6 (1·2–2·2) 19·3% 60% 10·4% 4·0%
subjective nature of some assessments, such as physical (9·1–11·8) (3·2–4·9)
activity. Details of the studies that were used to calculate Obesity 1·6 (1·3–1·9) 13·7% 64% 7·6% 2·9%
the sensitivity analyses are shown in the appendix. (6·5–8·8) (2·2–3·7)
Later life (≥65 years)
Discussion Smoking 1·6 (1·2–2·2) 39·9% 68% 19·3% 6·4%
To our knowledge, this study is the first to estimate the (17·6–21·1) (5·4–7·6)
proportion of dementia cases that are attributable to nine Depression 1·9 (1·6–2·3) 5·2% 72% 4·5% 1·7%
(3·7–5·5) (1·2–2·4)
risk factors (ie, the PAF) in LMICs. Overall PAFs for
dementia were higher in China and India than our Physical inactivity 1·4 (1·2–1·7) 15·3% 78% 8·4% 2·2%
(7·3–9·7) (1·6–2·9)
previous worldwide estimate of 35%, and higher still
Low social contact 1·6 (1·3–1·9) 10·4% 57% 4·0% 2·3%
in our Latin American sample at 56%. The potential for (3·2–4·9) (1·7–3·1)
prevention of dementia is therefore even greater in these Diabetes 1·5 (1·3–1·8) 9·3% 52% 4·4% 1·7%
countries than in higher-income countries. The highest (3·6–5·4) (1·2–2·4)
PAF in each region was for less education in early life, but Overall weighted PAF ·· ·· ·· ·· 41·2%
smoking, hypertension, obesity, and physical inactivity (39·1–43·4)
PAFs were also high, which highlights the potential PAF=population attributable fraction. RR=relative risk. *Weighted PAF is the relative contribution of each risk factor to
priorities for prevention of dementia in these regions. the overall PAF when adjusted for communality.
Dementia is more prevalent overall in India and Latin Table 2: PAF for dementia risk factors in India (n=2004)
America than in developed countries (eg, the UK, when
culturally appropriate screening tools are used), with
prevalence estimates of 8·5% in India, 6·4% in China,
and 8·6% in our Latin American sample,15 as compared RR for dementia Risk factor Communality PAF Weighted
with 6·4% in the EURODEM study16 of 11 European (95% CI) prevalence PAF*
countries (to which the 10/66 data were directly Early life (<45 years)
standardised for age, sex, and education). Low education 1·6 (1·3–2·0) 68·8% 36% 29·2% 10·9%
The estimation of PAFs are dependent on accurate (28·4–30·0) (10·4–11·5)
prevalence data, which in turn are dependent on accurate Midlife (45–64 years)
measurement of a risk factor in the population. We did Hearing loss 1·9 (1·4–2·7) 28·8% 48% 20·6% 7·7%
sensitivity analyses to identify whether alternative (19·9–21·3) (7·3–8·2)

prevalence data changed our PAF estimates. We found Hypertension 1·6 (1·2–2·2) 55·6% 59% 25·0% 9·3%
(24·3–25·8) (8·8–9·8)
that all regions had a similar or higher overall PAF than
Obesity 1·6 (1·3–1·9) 44·8% 53% 21·2% 7·9%
our previous worldwide estimates, even when the lowest (20·5–21·9) (7·5–8·4)
prevalence figures were used, indicating the potential for Later life (≥65 years)
greater targeting of dementia risk factors in these Smoking 1·6 (1·2–2·2) 30·0% 60% 17·8% 5·7%
regions. In general, the prevalence estimates of low (17·2–18·5) (5·3–6·1)
education, hypertension, obesity, and diabetes were Depression 1·9 (1·6–2·3) 23·9% 55% 17·7% 6·6%
higher than worldwide estimates across all three LMIC (17·1–18·4) (6·2–7·0)
regions (although the diabetes estimate was similar to Physical inactivity 1·4 (1·2–1·7) 34·2% 37% 17·0% 4·5%
the US prevalence),10 leading to higher PAFs for all of (16·4–17·7) (4·2–4·9)

these risk factors than previously calculated PAFs, which Low social contact 1·6 (1·3–1·9) 0·5% 69% 0·2% 0·1%
(0·1–0·3) (0–0·2)
were mostly based on data from higher-income countries.
Diabetes 1·5 (1·3–1·8) 18·5% 35% 8·5% 3·2%
The relative importance of each risk factor, as indicated (8·0–8·9) (2·9–3·5)
by their individual PAF, also remained the same in most Overall weighted PAF ·· ·· ·· ·· 55·8%
sensitivity estimates. (54·9–56·7)
Previous estimates of the proportion of people with low PAF=population attributable fraction. RR=relative risk. *Weighted PAF is the relative contribution of each risk factor to
education17 found that, in the 1960s, 57% of people older the overall PAF when adjusted for communality.
than 15 years in Latin America, 80·3% in south Asia, and
Table 3: PAF for dementia risk factors in the Latin American sample (n=12 865)
68·5% in east Asia had no education or only completed

www.thelancet.com/lancetgh Vol 7 May 2019 e600

 Articles

depression were similar to previous findings for India20

RR for dementia Risk factor Communality PAF Weighted
(95% CI) prevalence PAF* and China,21 but for our Latin American sample were
much higher than those reported in a WHO survey20
Early life (<45 years)
(although only Mexico was included in their Latin
Low education 1·6 (1·3–2·0) 40·0% 64% 19·1% 7·5%
(18·3–19·9) (7·0–8·0) American sample). Notably, the prevalence estimates for
Midlife (45–64 years)
depression vary substantially between these regions
Hypertension 1·6 (1·2–2·2) 8·9% 57% 5·1% 2·0%
(1·5% in China vs 23·9% in our Latin American sample),
(2·9–3·6) (0·6–0·9) which is in keeping with a meta-analyses that showed a
Obesity 1·6 (1·3–1·9) 3·4% 60% 2·0% 0·8% higher prevalence of depression in South America than
(1·7–2·3) (2·9–3·6) in Asia.22 These results possibly reflect a true difference
Hearing loss 1·9 (1·4–2·7) 31·7% 46% 23·0% 9·1% in prevalence. However, previous studies that made use
(22·2–23·8) (8·6–9·7) of 10/66 data have highlighted that the low prevalence of
Later life (≥65 years) depression reported in China might in part be due to
Smoking 1·6 (1·2–2·2) 27·4% 51% 13·9% 5·5% contextual and cultural factors.23 If, as previous authors
(13·2–14·6) (4·1–6·0)
have suggested, the divergent prevalence of depression
Depression 1·9 (1·6–2·3) 13·2% 59% 10·1% 4·0%
(9·5–10·7) (3·6–4·4)
in these cultures reflect different cultural expressions
Physical inactivity 1·4 (1·2–1·7) 17·7% 27% 6·5% 2·6%
of the same underlying depressive process, which
(6·0–7·0) (2·3–2·9) is inadequately captured by traditional diagnostic ap­
Low social contact 1·6 (1·3–1·9) 11·0% 46% 5·9% 2·3% proaches, then the PAF for depression in China might
(5·5–6·4) (2·0–2·6) be an underestimate and the Latin American PAF might
Diabetes 1·5 (1·3–1·8) 6·4% 70% 3·2% 1·2% be an overestimate.
(2·9–3·6) (1·0–1·4) Social isolation was very uncommon in China and our
Total adjusted for ·· ·· ·· ·· 35·0% Latin American sample, but in India was similar in
communality (34·1–35·9)
prevalence to our UK estimates.2 This finding led to social
Data were obtained from Livingston and colleagues.2 PAF=population attributable fraction. RR=relative risk. *Weighted isolation having the lowest PAF out of all the risk factors
PAF is the relative contribution of each risk factor to the overall PAF when adjusted for communality.
across the three regions, with a value of close to zero in
Table 4: Worldwide PAFs for dementia risk factors China and our Latin American sample. In Health Survey
England, 11% of people had less than monthly social
primary school. The 10/66 population sample, who were contact.24 The higher amount of social support might be a
65 years or older, would therefore have been older than strength of LMICs, and could be a useful model for how to
15 years at the time these data were collected in the 1960s, build better connected communities in higher-income
and included in these schooling estimates. Estimates of countries. Physical inactivity was more common in China
low education from 2010 are lower in all regions than and less common in India than worldwide estimates. Self-
the data from the 1960s,17 which suggests that progress to reported prevalence of physical inactivity was also higher
improve education over the past 50 years might lead to a in all countries than estimates from other cross-sectional
reduction in the prevalence of dementia, and the relative representative surveys,25 but lower than the studies in the
importance of this risk factor is likely to be lower for sensitivity analyses,26,27 which could reflect the inaccuracy
future generations than the present generation. of self-reported activity.28
Lifecourse analysis is important, given that, for The prevalence of diabetes in the 10/66 surveys was
example, hypertension and obesity, which are risk similar to other cross-sectional surveys in India29 and
factors in midlife, decrease as part of the development of China,30 but higher for Latin America.31 The PAF for
dementia; therefore, tackling these risk factors in midlife diabetes was relatively low in all three regions, but the
years before the probable development of dementia is fact that, in previous surveys, almost 50% of people
appropriate and important as a management strategy.18,19 surveyed in India,29 nearly 40% in China,30 and 20% in
Although the 10/66 prevalence estimate might have Latin America32 were unaware that they had diabetes
included some people with later-life hypertension, it indicates that more needs to be done to diagnose and
does not include those with undiagnosed hypertension treat diabetes, to further mitigate any risk to dementia
or those who did not report their diagnosis. Sensitivity development it might pose.
analyses based on other available prevalence estimates Hearing loss prevalence was lower in China and India
gave similar PAFs for all regions, indicating that than worldwide estimates and our estimates from the
hyper­tension is an important potentially modifiable risk UK.2 The 10/66 survey data only documented self-reported
factor, and so public health programmes to increase hearing impairment and not hearing impairment from
identification and management of hypertension in audiometry, as done in cohort studies of hearing loss.
midlife might reduce dementia prevalence. Around two thirds of people with mild hearing loss and a
Depression was relatively uncommon in China and third of those with severe hearing loss on audiometry
India but more prevalent in Latin America than in report normal hearing.33 This might have led to an
higher-income countries. Our prevalence estimates for underestimate of hearing loss in the current study, given

e601 www.thelancet.com/lancetgh Vol 7 May 2019

 Articles

that a substantial proportion of older people (aged 10 years ago, and the current prevalence of some risk
≥65 years) with hearing loss are unaware of their factors might have changed. This limitation is, however,
impairment.34 Hearing loss has been previously classed common within research that involves large amounts of
as a midlife risk factor given that the lowest mean age of data collection and periods of time between analyses.
participants in studies that showed this link was 55 years. Although overall numbers of people with dementia are
However, hearing loss is also a risk factor in adults aged increasing globally, the age-specific incidence and preva­
65 years or older, and preliminary evidence suggests that lence of dementia has reduced in many high-income
hearing aid use can mitigate the risk of dementia from countries over the past two decades.2,3,41 This reduction
hearing loss, indicating some potential for prevention.35 has been attributed to reduced frequency of dementia
To our knowledge, this study is the first to estimate risk factors, particularly low education and cardiovascular
PAFs for the widely accepted dementia risk factors in risk, in successive generations of older people. Our study
LMICs. We used data from a study whose methodology suggests that, because these risk factors are more
was identical in the eight countries it surveyed, making common in Asia and Latin America, there is greater
these figures comparable between countries. Our study dementia prevention potential in these LMICs than in
has some limitations. Because the quality of the high-income countries. Low education, hearing loss,
10/66 survey’s data on the prevalence of some risk factors, obesity, and physical activity had particularly high PAFs
such as hearing loss, was uncertain, we did sensitivity and so might be initial targets for policy makers devising
analyses to model the potential effect of different risk dementia prevention strategies. Public health strategies
factor prevalence estimates. The population sample was are likely to need to be specific to the setting in which
aged 65 years and older, so we could not measure midlife they are used, and future research should establish
hypertension or obesity. We were also unable to measure whether such strategies affect the prevalence of risk
exercise in midlife, but our calculations are based on the factors and subsequent dementia prevalence. As
risk of physical inactivity in later life. The data relied on dementia is forecast to become the leading public health
self-report for some of the risk factors, which is not ideal challenge globally, capitalising on the potential for
(eg, previous studies29,30,32 have shown that awareness of prevention is an urgent priority.
diabetes can be very low). Self-report is likely to have led to Contributors
the underestimation of PAFs. Although 10/66 study NM obtained and analysed the data. All authors designed the study,
participants had their blood pressure measured, we used interpreted the data, and wrote the manuscript.
self-report data, given that blood pressure might decrease Declaration of interests
in people who are developing dementia, which could lead We declare no competing interests.
to the under­estimation of its prevalence. We did not use Acknowledgments
country-specific studies for risk associated with factors but This is a secondary analysis of data collected by the 10/66 Dementia
Research Group. NM and JH are funded by the University College
rather available meta-analysis data, more of which London Hospitals (UCLH) National Institutes of Health Research
was from higher-income countries than from LMICs. (NIHR) Biomedical Research Centre. AS is funded by the Wellcome
However, these meta-analysis-based estimates are likely to Trust. GL is supported by the Economic and Social Research Council and
be more precise than individual studies in specific NIHR (ES/L001780/1) and the UCLH NIHR Biomedical Research Centre,
and receives funding from the NIHR Collaboration for Leadership in
countries, given that they combine the findings from Applied Health Research and Care, North Thames at Bart’s Health National
several studies, and they make the PAF more directly Health Service Trust, and through an NIHR Senior Investigator Award.
comparable with previous estimates.2 There is also no References
consistent evidence of gene–environment interaction with 1 Prince M, Wimo A, Guerchet M, Ali GC, Wu YT, Prina M.
regard to dementia, which might cause the effect of a risk World Alzheimer report 2015. The global impact of dementia.
An analysis of prevalence, incidence, cost and trends. London:
factor on dementia to vary globally.36–38 The causal direction Alzheimer’s Disease International, 2015.
underlying the association between these risk factors and 2 Livingston G, Sommerlad A, Orgeta V, et al. Dementia prevention,
dementia is uncertain, especially for associations found in intervention, and care. Lancet 2017; 390: 2673–734.
3 Matthews FE, Stephan BC, Robinson L, et al. A two decade dementia
later life. For example, depression might be a prodromal incidence comparison from the Cognitive Function and Ageing
feature or a consequence rather than a cause of dementia.39 Studies I and II. Nat Commun 2016; 7: 11398.
Additionally, a PAF is a theoretical construct that assumes 4 Langa KM, Larson EB, Crimmins EM, et al. A comparison of the
prevalence of dementia in the United States in 2000 and 2012.
a reduction in risk on the basis of elimination of the JAMA Intern Med 2017; 177: 51–58.
risk factor. Risk factors are unlikely to be completely 5 Ahmadi-Abhari S, Guzman-Castillo M, Bandosz P, et al.
eliminated; however, reduction in risk factors is still likely Temporal trend in dementia incidence since 2002 and projections
to delay the onset of dementia, and thereby reduce the for prevalence in England and Wales to 2040: modelling study.
BMJ 2017; 358: j2856.
number of dementia cases.40 However, even a partial 6 National Institue for Health Care and Excellence. Dementia,
reduction in some of the factors could, at a population disability and frailty in later life: mid-life approaches to delay or
level, make an enormous difference in the future prevent onset (NG16). London: National Institute for Health Care
and Excellence, 2015.
prevalence of dementia in LMICs, and we have indicated 7 Daviglus ML, Bell CC, Berrettini W, et al. NIH state-of-the-science
which factors could potentially have the largest effect. conference statement: preventing Alzheimer’s disease and cognitive
Finally, the data collected were recorded more than decline. NIH Consens State Sci Statements 2010; 27: 1–30.

www.thelancet.com/lancetgh Vol 7 May 2019 e602

 Articles

8 Barnes DE, Yaffe K. The projected effect of risk factor reduction on 26 Poggio R, Serón P, Calandrelli M, et al. Prevalence, patterns,
Alzheimer’s disease prevalence. Lancet Neurol 2011; 10: 819–28. and correlates of physical activity among the adult population in
9 Prince M, Ferri CP, Acosta D, et al. The protocols for the Latin America: cross-sectional results from the CESCAS I study.
10/66 Dementia Research Group population-based research Glob Heart 2016; 11: 81–8.e1.
programme. BMC Public Health 2007; 7: 165. 27 Anjana RM, Pradeepa R, Das AK, et al. Physical activity and
10 Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C. Potential for inactivity patterns in India–results from the ICMR-INDIAB study
primary prevention of Alzheimer’s disease: an analysis of (Phase-1)[ICMR-INDIAB-5]. Int J Behav Nutr Phys Act 2014; 11: 26.
population-based data. Lancet Neurol 2014; 13: 788–94. 28 Visser M, Brychta RJ, Chen KY, Koster A. Self-reported adherence to
11 WHO. Waist circumference and waist-hip ratio. Report of a WHO the physical activity recommendation and determinants of
expert consultation. Geneva, 8–11 December 2008. Geneva: misperception in older adults. J Aging Phys Act 2014; 22: 226–34.
World Health Organization, 2011. 29 Anjana RM, Deepa M, Pradeepa R, et al. Prevalence of diabetes and
12 Copeland J, Dewey ME, Griffiths-Jones H. A computerized prediabetes in 15 states of India: results from the ICMR–INDIAB
psychiatric diagnostic system and case nomenclature for elderly population-based cross-sectional study. Lancet Diabetes Endocrinol
subjects: GMS and AGECAT. Psychol Med 1986; 16: 89–99. 2017; 5: 585–96.
13 Janssen I, Katzmarzyk PT, Ross R. Waist circumference and not body 30 Wang L, Gao P, Zhang M, et al. Prevalence and ethnic pattern of
mass index explains obesity-related health risk. Am J Clin Nutr 2004; diabetes and prediabetes in China in 2013. JAMA 2017;
79: 379–84. 317: 2515–23.
14 Boyle MH. Guidelines for evaluating prevalence studies. 31 Aschner P, Aguilar-Salinas C, Aguirre L, et al. Diabetes in South
Evid Based Ment Health 1998; 1: 37–39. and Central America: an update. Diabetes Res Clin Pract 2014;
15 Rodriguez JJL, Ferri CP, Acosta D, et al. Prevalence of dementia in 103: 238–43.
Latin America, India, and China: a population-based cross-sectional 32 Irazola V, Rubinstein A, Bazzano L, et al. Prevalence, awareness,
survey. Lancet 2008; 372: 464–74. treatment and control of diabetes and impaired fasting glucose in
16 Lobo A, Launer LJ, Fratiglioni L, et al. Prevalence of dementia and the Southern Cone of Latin America. PLoS One 2017; 12: e0183953.
major subtypes in Europe: a collaborative study of population-based 33 Kim SY, Kim H-J, Kim M-S, Park B, Kim J-H, Choi HG.
cohorts. Neurology 2000; 54: S4. Discrepancy between self-assessed hearing status and measured
17 Barro RJ, Lee JW. A new data set of educational attainment in the audiometric evaluation. PLoS One 2017; 12: e0182718.
world, 1950–2010. J Dev Econ 2013; 104: 184–98. 34 Rawool VW, Keihl JM. Perception of hearing status,
18 Kivimäki M, Luukkonen R, Batty GD, et al. Body mass index and communication, and hearing aids among socially active older
risk of dementia: analysis of individual-level data from 1.3 million individuals. J Otolaryngol 2008; 37: 27–42.
individuals. Alzheimers Dement 2018; 14: 601–09. 35 Amieva H, Ouvrard C, Meillon C, Rullier L, Dartigues J-F.
19 Singh-Manoux A, Dugravot A, Shipley M, et al. Obesity trajectories Death, depression, disability and dementia associated with
and risk of dementia: 28 years of follow-up in the Whitehall II Study. self-reported hearing problems: a 25-year study.
Alzheimers Dement 2018; 14: 178–86. J Gerontol A Biol Sci Med Sci 2018; 73: 1383–89.
20 Kessler RC, Birnbaum HG, Shahly V, et al. Age differences in the 36 Ghebranious N, Mukesh B, Giampietro PF, et al. A pilot study of
prevalence and co-morbidity of DSM-IV major depressive episodes: gene/gene and gene/environment interactions in Alzheimer disease.
results from the WHO World Mental Health Survey Initiative. Clin Med Res 2011; 9: 17–25.
Depress Anxiety 2010; 27: 351–64. 37 Wirth M, Villeneuve S, La Joie R, Marks SM, Jagust WJ.
21 Gu L, Xie J, Long J, et al. Epidemiology of major depressive disorder Gene–environment interactions: lifetime cognitive activity, APOE
in mainland China: a systematic review. PLoS One 2013; 8: e65356. genotype, and beta-amyloid burden. J Neurosci 2014; 34: 8612–17.
22 Lim GY, Tam WW, Lu Y, Ho CS, Zhang MW, Ho RC. Prevalence of 38 Quinn J. Comment: gene-environment interactions in dementia—
depression in the community from 30 countries between 1994 and not just another fish story. Neurology 2016; 86: 2069.
2014. Sci Rep 2018; 8: 2861. 39 Singh-Manoux A, Dugravot A, Fournier A, et al. Trajectories of
23 Guerra M, Ferri CP, Sosa AL, et al. Late-life depression in Peru, depressive symptoms before diagnosis of dementia: a 28-year
Mexico and Venezuela: the 10/66 population-based study. follow-up study. JAMA Psychiatry 2017; 74: 712–18.
Br J Psychiatry 2009; 195: 510–15. 40 Jorm AF, Korten AE, Henderson AS. The prevalence of
24 Scholes S, Mindell J, Craig R. Health Survey for England 2014: dementia–a quantitative integration of the literature.
health, social care and lifestyles. Leeds: Health and Social Care Acta Psychiatr Scand 1987; 76: 465–79.
Information Centre, 2014. 41 Matthews FE, Arthur A, Barnes LE, et al. A two-decade comparison
25 Guthold R, Ono T, Strong KL, Chatterji S, Morabia A. of prevalence of dementia in individuals aged 65 years and older
Worldwide variability in physical inactivity. A 51-country survey. from three geographical areas of England: results of the Cognitive
Am J Prevent Med 2008; 34: 494. Function and Ageing Study I and II. Lancet 2013; 382: 1405–12.

e603 www.thelancet.com/lancetgh Vol 7 May 2019