Professional Documents
Culture Documents
Nrneurol 2017 63
Nrneurol 2017 63
Dementia has become an important public health, eco- suffice to provide an understanding of the full spectrum
nomic, social and political issue, and attracts increasing of dementia in the general population and to iden-
investment into research. According to estimates from tify risk factors across different populations and life
the World Alzheimer Report 2015, 46.8 million people courses4. These aspects can only be investigated fully
worldwide have dementia, and this number is expected with population-based e pidemiological research.
to increase to 74.7 million by 2030 and 131.5 million Investigation of the changes in dementia incidence
Correspondence to C.B. by 2050 (REF. 1). In light of this predicted dementia and prevalence over time is challenging, as changes
Department of Public Health ‘epidemic’ and consequent economic burden, the G8 in diagnostic criteria and other methodological vari
and Primary Care, Cambridge
dementia summit in 2013 and the WHO Ministerial ation can affect the prevalence and incidence esti-
Institute of Public Health,
Forvie Site, University of
Conference in 2015 resulted in calls for global action mates. Primary evidence must, therefore, be founded
Cambridge, School of Clinical against dementia. The summit established a goal to on population-based studies that have consistent study
Medicine, Cambridge identify a cure or disease-modifying therapy by 2025 designs and measurement methods over time. In the
Biomedical Campus, (REFS 2,3). To date, most dementia research has focused past 5 years, several population-based studies of demen-
Cambridge CB2 0SR, UK.
on the neurological features, pathophysiological mech- tia have been published in which consistent research
carol.brayne@
medschl.cam.ac.uk anisms, and drug discovery. This basic science approach methods were used, providing new insight into the
has provided knowledge about dementia at the individ- descriptive epidemiology of dementia and challenging
doi:10.1038/nrneurol.2017.63
Published online 12 May 2017; ual or biological level, but a predominantly reductionist the accepted forecasts of increasing prevalence and
corrected online 17 May 2017 approach that focuses on single mechanisms does not incidence.
and has been poorer still across time. These changes in owing to heterogeneous definitions and diagnostic
diagnostic boundaries combined with parallel increases methods across time and studies. Those using medical
in awareness of dementia among the public and med- records, healthcare administrative databases, systematic
ical professionals has led to earlier diagnosis on aver- reviews and meta-analyses are described briefly in the
age17. Furthermore, a consensus diagnosis, even when latter part of this review.
made with use of standardized data collection methods
and standardized diagnostic criteria, can be affected Trends in prevalence
by changes in a clinician’s perceptions of diagnostic Study details. Nine studies have investigated trends in
thresholds and criteria over time18,19. dementia prevalence in France (the Bordeaux farmer
Ultimately, diagnosis of dementia remains heavily study)19, Sweden (the Gothenburg, Nordanstig and
influenced by clinical judgement, medical records and Stockholm studies)24–26, Spain (the Zaragoza study)27,
characteristics of the individuals that make up a study the UK (the Cognitive Function and Ageing Study
population. For this reason, dementia remains a clinical (CFAS))28, the USA (the Health and Retirement Study
syndrome, with an emphasis on cognitive and functional (HRS) and the Indianapolis–Ibadan Dementia Project
states, and identifying subtypes is particularly challenging. (IIDP))29,30, and Japan (the Hisayama study)31 (TABLE 1).
For example, Alzheimer-type dementia can be diagnosed The earliest of these is the Gothenburg study (1976–
with varying levels of investigation, but regardless of the 1977)24 and the most recent is the HRS (conducted in
investigations used, diagnosis is based on the assump- 2012)30. The Zaragoza study, the CFAS and the IIDP had
tion that the underlying pathology is Alzheimer-type similar designs: two independent cohorts were recruited
pathology. Subtype analysis is, therefore, more difficult across two time points in defined geographical areas27–29.
to keep consistent over time than is syndromic diagno- In the Gothenburg study24, age-specific prevalence at
sis, limiting the possibility of making valid comparisons ages 70 years and 75 years were compared over three
of prevalence or incidence between dementia subtypes. decades in random samples of local populations. In the
Nordanstig25 and Stockholm26 studies, the prevalence of
Primary evidence dementia from the earlier investigations (the Nordanstig
We identified 14 population-based studies in which the Project (1995–1998) and the Kungsholmen Project
study methods used at all time points were sufficiently (1987–1989)) was compared with the regional preva-
similar and in which the geographical areas were suffi- lence from a more recent nationwide cohort (Sweden
ciently well-defined to enable a meaningful comparison. National Study on Aging and Care (SNAC; 2001–2004)).
Of these 14 studies, nine investigated prevalence and five The Bordeaux farmer study included only farmers living
investigated incidence. The populations included were in the Bordeaux area19. The HRS is a dynamic study in
from the USA, Western Europe, Japan and Nigeria. which new cohorts are enrolled every 6 years to ensure
Five other studies have also investigated trends in a representative sample of older adults in the USA30.
dementia prevalence or incidence but were not included The only study from East Asia, the Hisayama study con-
because of methodological aspects that mean they can- ducted in Japan31, included all residents of the study area
not be meaningfully compared with the other 14 studies who were aged ≥65 years at four time points.
(BOX 1). Studies that focused on only Alzheimer dis- In the Zaragoza study, the CFAS, the IIDP and the
ease20,21 or cognitive impairment22,23 were also excluded, Bordeaux farmer study, considerable drops in response
rate were seen across the two cohorts27–29. A wide range
of sensitivity analyses was conducted in the CFAS and
Box 1 | Excluded studies of dementia prevalence and incidence the Bordeaux farmer study to address potential selection
Five studies have investigated trends in dementia prevalence and incidence but used
bias owing to this differential response rate, and revealed
methods that prevent them from being meaningfully compared with others; limited effects on the results. In the Zaragoza study, a sam-
consequently, these studies were not included in our Review. Two Swedish studies81,82 pling strategy was used to account for the population that
were excluded. In one, prevalence and incidence trends during 1947–1957 and did not respond, so the estimates are considered to be rep-
1957–1972 were investigated81, and findings from these periods are likely to have resentative of the entire older population in the study area.
limited relevance to contemporary older populations. The other focused on short-term The nine studies had either one-stage or two-stage
prevalence trends in populations aged ≥85 years82, but had methodological flaws: the designs: a one-stage design which included only a diag-
study cohorts were not sampled independently, the analysis did not account for the nosis phase that involved detailed examination and
overlapping of the study population, and medical records were used to support application of clinical criteria, and a two-stage design
dementia diagnosis, which might have led to bias owing to differences in the
included a screening phase to identify potential par-
diagnostic boundaries applied for patients across time.
One Japanese study70 that investigated prevalence trends between 1980 and 2000
ticipants with dementia as well as a diagnostic phase.
was excluded because the screening for dementia was based on self-reported cognitive Clinical diagnoses were mainly based on the Diagnostic
problems rather than objective cognitive testing, and clinical diagnosis was only and Statistical Manual of Mental Disorders, third edition
applied to those who reported cognitive problems. This approach might have led to revised (DSM‑III‑R) criteria32. Algorithmic diagnosis
biased prevalence estimates because a marked change in awareness of dementia was used in the CFAS, and algorithmic historical crite-
within the population is likely to have affected patient self-reporting. Another report, ria were used in the Gothenburg study; these approaches
from the Hisayama study in Japan69, was excluded because it focused on an autopsy were similar to the DSM‑III‑R criteria. In the French
subsample without clear representation of the older population in the study area. study, an algorithm approach based on Mini-Mental
One study has compared prevalence and incidence trends in dementia in China, but State Examination (MMSE)33 and Instrument Activity
used different diagnostic criteria at the two time points studies83, so was also excluded.
of Daily Living (IADL)34 scores was used in addition
to clinical diagnosis to identify potential participants interview, and was validated in a subsample of the HRS
with dementia. Diagnosis in the HRS was based on a cohort (the Aging, Demographics, and Memory Study
27‑point cognitive test, or a proxy assessment if the (ADAMS)); this approach had a 78% concordance with
participant was unable to complete the interview30. The clinical diagnosis made by medical professionals on the
assessment was conducted as a phone or face‑to‑face basis of c riteria developed for the study35.
The intention in each of these studies was to imple- non-overlapping sub-cohorts, and the FHS40 involved
ment the same diagnostic methods over time, but analysis of participants from two generations. In the
changes in subjective clinical opinion cannot be ruled Rotterdam study, the 1990 cohort included all residents
out as a major factor that might have influenced case of the study area aged 60–90 years, and the 2000 cohort
identification and prevalence estimates. Measures were included a non-overlapping sample of residents who had
taken in several of the studies to address this issue. In since entered that age range or were of the correct age
the Nordanstig and Stockholm studies, the same physi- and had moved into the study area. The FHS combined
cians made the diagnoses in the two cohorts within each data from an original cohort and a cohort of their off-
study. In the IIDP, a clinical consensus process was used, spring, and divided them into four epochs to compare
in which the same group of clinicians made diagnoses incidence across these periods. The follow‑up periods
in the two cohorts. In the CFAS and the Gothenburg and intervals varied between studies. To address differ-
study, algorithmic diagnosis with a structured psychi- ential response rate and potential effect of missing data,
atric interview was used to avoid variation in subjective several s ensitivity models were tested in the Bordeaux
opinions. Small changes in study designs and method- study and CFAS.
ologies were made in the Hisayama study31, Stockholm Different sets of diagnostic criteria were used in
study26 and the CFAS28; to ensure that these changes had the Rotterdam study (DSM-III‑R), IIDP (DSM-III‑R,
minimal effects on prevalence estimates, the new meas- ICD‑10) and FHS (DSM‑IV)32,41,42. In the Bordeaux
urements used in these studies were tested and validated study and the CFAS, algorithmic diagnosis was used.
before their application to subsequent cohorts. In the Bordeaux study37, MMSE33 and IADL34 scores
were used to define dementia, and in the CFAS38, a dif-
Findings. In contrast to the increase in prevalence of ferential diagnosis procedure derived from a structured
dementia that is predicted on the basis of projection psychiatric interview was used. The Bordeaux study also
methods, most of the nine population-based studies included clinical diagnosis, but different clinical criteria
reported stable or declining prevalence over time (FIG. 1; were applied to the two cohorts, so these data were not
age-specific and sex-specific prevalence estimates are used to assess temporal trends.
in Supplementary information S1 (table)). The three
Swedish studies indicate generally stable prevalence Findings. Despite different study designs and meth-
within the total population, albeit with wide confidence ods, all five studies suggest a decrease in the incidence
intervals in all but the Stockholm study. The CFAS of dementia in the total population across cohorts and
demonstrates a 23% reduction in prevalence over two time periods (FIG. 2; age-specific and sex-specific inci-
decades in the total study population in England28, and dence estimates are in Supplementary information S2,S3
the HRS suggests a 26% decrease in prevalence over (tables)), although notable differences were seen in
12 years in the older US population30. The Bordeaux subpopulations, particularly between the sexes. In the
farmer study found conflicting results when the two Bordeaux study, the decrease was mainly driven by an
diagnostic approaches were used: with the algorithmic effect in women, whereas in the CFAS, the reduction
diagnosis, a 40% decline in prevalence was seen, but when was confined to men. In the FHS, the reduction occurred
clinical diagnosis was used, an increase in prevalence earlier and was sustained across the three epochs in
by more than twofold was seen19. The Nordanstig and women, but occurred only in the last epoch in men.
Zaragoza studies did not indicate significant reductions The results of the IIDP suggest that the incidence was
in prevalence in the total population, but demonstrated reduced in African American people over a 10‑year
decreases of >50% among men25,27. Taken together, and period, and indicate a 20% reduction in incidence in
given the fact that the longevity of people with dementia Nigerian cohorts, but this effect was not statistically
is increasing26, these results suggest an actual decline in significant. In the Bordeaux study, the results obtained
the age-specific risk of dementia. with clinical diagnosis differed from those obtained with
algorithm d iagnosis; only the latter resulted in a decrease
Trends in incidence in incidence37.
Study details. Five studies have investigated trends in the
incidence of dementia in the Netherlands (the Rotterdam Secondary evidence
study)36, France (the Bordeaux study)37, the UK (the Secondary evidence for trends in dementia prevalence
CFAS)38, the USA (the IIDP and the Framingham Heart and incidence is provided by studies that are based on
Study (FHS))39,40, and Nigeria (the IIDP)39 (TABLE 2). The medical records, healthcare and insurance administra-
IIDP examined trends in African American people in tive databases, systematic reviews and meta-analyses,
Indianapolis, USA, and in a Yoruba population in Ibadan, as these types of research do not provide the option of
Nigeria39. The Bordeaux incidence study focused on controlling for changes in diagnostic methods, subjective
urban residents rather than farmers, but used the same clinical opinions and public awareness. Several studies
reference cohort as the prevalence study37. The CFAS conducted in Western Europe and North America
and IIDP reported both prevalence and incidence trends have reported trends in the prevalence or incidence of
within the same study cohorts28,29,38,39. dementia on the basis of analysis of medical records or
In the Bordeaux study, the CFAS, and the IIDP, healthcare administrative databases22,43–52. These studies
incidence was measured in two independent cohorts, tend to include large populations and are often based on
whereas the Rotterdam study36 involved analysis of patients’ contact with medical services over time. These
Figure 1 | Odds ratios and prevalence ratios reported in nine studies of dementia prevalence. Nature Reviews
Reported | Neurology
figures are the
ratios of the prevalence estimate for new cohorts to those for old cohorts. If prevalence estimates remain the same across two
cohorts, the ratio is 1.0. If estimates are higher in the new cohort than the old cohort, the ratio is greater than 1.0. Estimates
are adjusted as follows: Gothenburg study and HRS, unadjusted; IIDP, adjusted for age; Nordanstig, Zaragoza, Bordeaux and
Hisayama studies, adjusted for age and sex; Stockholm study, adjusted for age, sex and education; CFAS, adjusted for age,
sex, area and deprivation. In the Bordeaux study, clinical diagnosis was made by neuropsychologists using criteria in the
Diagnostic and Statistical Manual III revised, and algorithmic diagnosis was based on cognitive and functional ability tests.
CFAS, Cognitive Function and Ageing Study; HRS, Health and Retirement Study; IIDP, Indianapolis–Ibadan Dementia Project.
analyses have mainly focused on short-term trends over cohort across 22 provinces in China. One study reports
fewer than 10 years, and advanced analytical strategies declining incidence between 1998 and 2014 in people
have been required to estimate prevalence or incidence aged ≥60 years61, whereas the other suggests increasing
over continuous time periods and overlapping study prevalence across the oldest old cohorts aged ≥80 years62.
populations. Ascertainment bias, owing to inclusion These differences might be related to selection of study
of patients who approached medical services, and dif- populations and variation in analytical approaches.
ferences in diagnostic practice between clinical settings Whether any changes reported in East Asia are attribut-
(such as different departments and centres) cannot be able to heterogeneity of study designs and implementa-
fully addressed in these analyses, making interpretation tion or differences between high-income and low-income
of the findings challenging. Some of these studies have countries, or eastern and western society remains unclear.
suggested stability or reductions in the annual preva- The different results obtained from primary investiga-
lence or incidence of dementia22,43–47, whereas others have tions and systematic reviews emphasize the substantial
reported increases in prevalence or incidence47–52. impact that changes in diagnostic methods, study designs
Owing to a lack of comparable data, trends in the and social contexts over time can have on estimates of
prevalence of dementia outside of western countries have epidemiological measurements.
been estimated mainly on the basis of systematic reviews
and meta-analyses, which aggregate estimates from indi- Current evidence on dementia trends
vidual studies conducted around the same time period. Diagnosis of dementia, as of any disorder, is contextual
Systematic reviews of numerous prevalence studies con- and can change across time and geographies. The second-
ducted in East Asian countries have suggested that the ary evidence discussed above was based on the analysis of
prevalence of dementia is increasing in Japan53, South healthcare administrative databases, medical records, sys-
Korea54, Hong Kong55, Taiwan56 and China57,58, although tematic reviews and meta-analyses, and the mixed find-
the increase in China is no longer significant when ings emphasize the fact that comparisons should not rely
methodological factors, including changes in diagnostic on an overview of reported numbers, but must include
criteria, are controlled for1,59,60. Two recent studies have careful appraisal of methodologies and study contexts.
used different analytical methods to investigate changes Changes in diagnostic methods, medical knowledge and
in cognitive impairment (measured by the MMSE) in the public awareness all influence identification of who meets
Chinese Longitudinal Healthy Longevity Surveys, a large and does not meet study diagnostic criteria for dementia.
Bordeaux
(clinical diagnosis, age ≥65 years)
1999 vs 1988 Total 0.9 (0.7, 1.1)
1999 vs 1988 Men 1.2 (0.8, 1.9)
1999 vs 1988 Women 0.9 (0.7, 1.2)
Bordeaux
(algorithmic diagnosis, age ≥65 years)
1999 vs 1988 Total 0.6 (0.5, 0.8)
1999 vs 1988 Men 1.1 (0.7, 1.8)
1999 vs 1988 Women 0.6 (0.5, 0.8)
IIDP
(African American age ≥70 years)
2001 vs 1992 Total 0.4 (0.3, 0.5)
The different results obtained with use of clinical and the individual studies, these population-based studies
algorithmic diagnoses in the two Bordeaux studies19,37 dis- generally report stabilization of, or a reduction in, the
cussed above further emphasize the effects of changes in prevalence and incidence of dementia. Given that mor-
diagnostic boundaries and the substantial effect they can tality is declining in the general population, stabiliza-
have on estimates of prevalence and incidence over time. tion of dementia prevalence also suggests a decline in
Identification of true prevalence and incidence the incidence. This primary evidence indicates a reduc-
trends, therefore, requires population-based studies that tion in dementia occurrence in more recent generations,
use similar research methods across different time peri- particularly in western countries.
ods; the 14 primary studies discussed above are all such
studies. In many of these studies, declines in response Possible explanations
rates and changes in diagnostic boundaries used to make Western countries
consensus diagnoses were reported, factors that are The primary evidence from western countries discussed
likely to affect the results. Nevertheless, despite differ- in this Review suggests a reduction in the risk of devel-
ent study designs, methodologies and settings between oping dementia and improved health in old age across
Box 2 | Factors related to decreasing dementia occurrence dementia, although any positive effects would be subject
to time lags, the lengths of which are unknown. Most
The four studies listed here investigated the effects of risk and protective factors on medications for cardiovascular disease and other chronic
decrease in dementia prevalence or incidence. The common factors included in the diseases (antihypertensives that act on systolic blood pres-
analysis for these studies were education, smoking, hypertension, cardiovascular disease sure, antidepressants and statins) have only been widely
(stroke or heart disease), diabetes mellitus, BMI and cholesterol levels.
available since the 1990s, so their effect on the observed
Bordeaux study, France26 changes in dementia prevalence and incidence might
• Adjusted for factors to test whether the observed decreases in dementia prevalence be limited; further improvements in these treatments
and incidence were attenuated. for chronic conditions might change the risk of devel-
• Education and vascular factors had a small effect, but the trends of decreasing oping dementia in later life and might, therefore, have
prevalence and incidence remained significant. further effects on dementia incidence and prevalence in
Framingham Heart Study, USA40 the future46,65. Other lifestyle factors, such as changes in
• Adjusted for factors to test whether the observed decrease in dementia incidence diet and physical activity, have been suggested as reasons
was attenuated. for a declining incidence of dementia, but primary evi-
• No significant effect of any investigated factor — changes were <10% dence to confirm such hypotheses is currently lacking,
and these patterns are changing with each generation.
Health and Retirement Study, USA30
Some of the prevalence and incidence studies
• Adjusted for factors to test whether the observed decrease in dementia prevalence revealed sex differences in the changes (TABLE 3), and
was attenuated.
these differences could provide further insight into pos-
• Education, cardiovascular factors and BMI attenuated the change by up to 12%, but sible reasons. Life expectancy at the age of 60 years is a
the decrease in prevalence remained significant.
good marker of the overall health status of elderly people
Rotterdam Study, Netherlands64 (based on UN data), and despite the fact that women
• Calculated population attributable risk (PAR; the proportion of dementia cases that in western countries have consistently had a longer life
could be prevented if the risk factor was removed) for each risk factor in the two expectancy at the age of 60 years, the life expectancy of
cohorts studied. men has increased to a greater degree over the past two
• The PAR was reduced by smoking and cholesterol levels. decades. Reductions in smoking and improvements
• The PAR was unaffected by education and cardiovascular disease. in the prevention and treatment of cardiovascular dis-
• The PAR was increased by diabetes mellitus and hypertension. ease might have had a larger effect on the health and
life expectancy of men than those of women, and such
major changes in risk factors might be important in the
generations. Indeed, in the Rotterdam study, improve- observed sex difference in dementia occurrence.
ment of brain health — as indicated by larger brain Although the reasons for stable or decreasing preva-
volume, less brain atrophy and less cerebral small ves- lence and incidence of dementia remain unclear, a sin-
sel disease — was reported in the most recent cohort36. gle risk factor is unlikely to be responsible. Changes in
Several possible explanations for these findings have Western societies and improvements in living conditions
been suggested5,63, but only four of the studies30,37,40,64 since the two World Wars have led to improvements in
identified the key factors associated with the trend of overall health and in cognitive development and reserve
decreasing dementia incidence (BOX 2), and these factors throughout the lifecourse66. Population-level invest-
vary between American, French and Dutch cohorts. ments in infrastructure, education, health services and
Educational level explained some of the decline in social welfare have contributed to substantial improve-
incidence: up to 6% in the FHS and nearly 10% in the ments in multiple dimensions of physical, mental and
Bordeaux incidence study. In the HRS, education along cognitive health from early life and might have conse-
with other socioeconomic factors explained 10% of the quently reduced the risk of dementia in later life. For
decrease in prevalence. However, in the Rotterdam study, example, education level has been related to an increase
the percentage of preventable dementia cases related to in cognitive reserve67. A higher level of education has
low education remained similar over two decades, sug- been associated with better cognitive performance, but
gesting that, despite improvements in education in the not with lower rates of cognitive decline68. The latest gen-
more recent cohort, the level of education still contrib- erations to reach old age have had more years of statu-
uted substantially to dementia occurrence in the more tory education than previous generations, which might
recent cohort64. be associated with greater cognitive reserve, which, in
The change in the proportion of cases that were pre- turn, might partly explain a delayed onset of dementia.
ventable by cessation of smoking partly explained the Such effects on incidence can only be observed over
changes in incidence in the Rotterdam study, but smok- decades. Nevertheless, the observations we have indi-
ing did not explain the decline in incidence in the FHS cate that factors that are related to social disadvantage
and Bordeaux study. and health inequality might have an important role in
In these four studies, the changing prevalence of sev- cognitive health over the lifecourse.
eral chronic diseases that are associated with dementia
— such as stroke, heart diseases, hypertension and dia- Beyond Western countries
betes mellitus — was reported but differed across stud- Primary evidence that provides information about the
ies. These changes explain only a limited proportion prevalence and incidence of dementia outside Western
(<10%) of the reduction in incidence and prevalence of Europe and the USA is lacking. Systematic reviews and
Table 3 | Studies in which sex differences were reported for trends in the prevalence and incidence of dementia
Study Prevalence trends Incidence trends Life expectancy at age 60 years‡ (years)
Men Women Men Women Men Women
1990 2000 2012 1990 2000 2012
CFAS, UK (2008 vs 1991) ↓* ↓* ↓* ↔ 18 20 22 22 23 25
Zaragoza, Spain (1994 vs 1987) ↓* ↔ N/A N/A 19 21 22 24 25 27
Nordanstig, Sweden (2001 vs 1995) ↓* ↔ N/A N/A 19 21 23 23 24 25
Bordeaux, France (1999 vs 1988) N/A N/A ↔ ↓* 20 20 23 25 26 27
FHS, USA (2005 vs 1985) N/A N/A ↓ ↓* 19 20 21 23 23 24
↓ Decrease. ↔ Stable. *Trend was statistically significant. ‡Based on WHO data. CFAS, Cognitive Function and Ageing Study; FHS, Framingham Heart Study;
N/A, not applicable.
meta-analyses can be used to synthesize evidence on the early life of the study cohorts. The long-term effects
dementia prevalence and incidence in low and mid- of these societal factors can lead to variation in popu-
dle income countries, but such analyses of trends are lation health, cognitive reserve and dementia occur-
unlikely to be robust if variations in methodologies and rence across generations in countries. Future trends in
population characteristics are not taken into account5. dementia occurrence outside western countries are less
Of the primary, population-based studies discussed predictable because the interplay between health during
in the previous sections, two were conducted outside the lifecourse, protective factors and risk factors varies
Western Europe and the USA: one in Japan and one in considerably between social contexts.
Nigeria, and these reported different trends. The Hisayama study includes longitudinal data on
The Hisayama study reported an increasing preva- cardiovascular disease in middle-aged cohorts, which might
lence of dementia between 1985 and 2005, and findings also provide some insight into potential reasons for the
from a subsample of autopsy specimens suggested an observed changes in dementia prevalence in Japan. Since
even higher prevalence in 2012 (REFS 31,69). However, the mid‑1980s, the prevalence of hypertension, stroke and
the analysis of the autopsy subsample did not take into smoking has declined, and the prevalence of diabetes mel-
account age and potential selection bias. Another study litus, hypercholesterolaemia and obesity has increased72.
conducted in the Diasen-Cho area of Japan, which The increasing prevalence of dementia might, therefore,
was excluded from the discussion in previous sections be related to changes in lifestyle, such as an increasingly
because the screening approaches used did not ensure western diet, physical inactivity and increasing obe-
representation of the whole population, also identified sity, metabolic syndromes and diabetes mellitus31,73. An
an increase in prevalence across three time points (1980, analysis published in 1995 showed that these factors
1990 and 2000)70. By contrast, the incidence of dementia were not associated with an increased risk of all-type
in the Nigerian cohort of the IIDP was stable39. dementia after a 7‑year follow‑up, and a subtype analysis
A unifying explanation for these different results found only one significant association, between diabetes
is difficult to formulate because these countries have mellitus and vascular dementia, the prevalence of which
different economic, political, social and cultural back- appeared to be stable31,74. However, a more recent analysis
grounds and rates of change over the past few dec- after a 15‑year follow‑up showed that diabetes mellitus
ades. Some insight might be gained by considering the was related to an increased risk of all-type dementia and
changes in life expectancy at birth over the past century Alzheimer disease, but not with vascular dementia75.
(FIG. 3). Changes in life expectancy are associated with Although these findings indicate a long-term effect of
the effects of societal factors and might therefore also diabetes mellitus on dementia in old age and a possible
indicate different determinants of cognitive health across time-lagged effect on increasing prevalence, increasing
generations71. Different life events, health statuses and recognition of mixed dementia will make interpretation
disease profiles experienced by different generations, of these changes in the occurrence of dementia subtypes
and trends in the prevalence and incidence of dementia over time even more challenging. Until deeper pheno-
might reflect complex interactions between these fac- typing, both during life and after death, is conducted and
tors. The life expectancy profiles of Japan and Nigeria is consistent across time, the detail of neurobiological
over the past century differ dramatically from those changes that underlie the risks and clinical manifestations
of the western countries. In Japan, the life expectancy of dementia will be unknown.
was lower than that in western countries in the first Population ageing and an increasing burden of non-
half of the 20th century and increased dramatically in communicable diseases are important challenges world-
the 1960s. Life expectancy in Nigeria has increased by wide, but the effect of chronic diseases on dementia
30 years over the past century, but is still 20 years lower trends might vary between contexts, with time-lags of
than the other countries. These dramatic changes are uncertain length. Forecasts of dementia burden must take
generally related to war and historical events, with a con- into account these different contexts of health profiles,
sequence of extremely deprived living conditions, inter- deprivation and social environments rather than only
ruption of education and lack of health and social care in focusing on the potential effect of individual risk factors.
100
World War I Great Depression World War II African American Civil End of Cold
(1914–1918) (1929–1937) (1939–1945) Rights Movement (1960s) War (1991)
90
Famines in Continental Europe
70
Life expectancy (age)
60
50
40
30
Meiji restoration Nuclear bombs Japanese economic Japanese asset
(1868–1912) in Japan (1945) miracle (1960s) bubble collapse (1991)
20
End of British HIV/AIDS
Nigeria (1960) (1990s)
10 Unification of Niger areas
under British Colony Nigeria Civil War Nigeria democratization
(1914) (1967–1970) (1999)
0
1900 1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015
Year
Spain Sweden Netherlands United France Japan Nigeria United United States:
Kingdom States black
Figure 3 | Life expectancy at birth in all countries included in population-based studies of dementia incidence and
Nature
prevalence. Data obtained from from Gapminder, the National Center for Health Statistics, USA 84
andReviews Neurology
Wu, Y.‑T.| et al.5.
studies and analytical epidemiological approaches on health and social policies in relation to dementia
(neuroscientific epidemiological approaches) is vitally prevention and risk reduction.
important and provides an opportunity to integrate No single factor has been identified that fully explains
understanding of brain health, neurobiology and neuro the observed changes in dementia prevalence and inci-
pathology within the general population to support dence, but reductions in absolute inequalities — including
better prevention, care and cure of dementia. improvements in living conditions, better access to educa-
tion and improved healthcare systems — are likely to have
Conclusions influenced multiple risk and protective factors throughout
In the past few years, descriptive epidemiological stud- an individual’s lifecourse that are related to physical, men-
ies, in which the effects of changing diagnostic criteria tal and cognitive health, and thereby reduced the risk of
and study methods were minimized, have strengthened dementia in later life. This conclusion sends an important
the evidence that dementia — age for age — is declining message to all in society about the need for long-term
in some countries and that the number of people with action to address factors that determine both healthy
dementia can remain stable despite population age- and unhealthy ageing, and to make further efforts to
ing28,30. Substantial reductions in the risk of dementia in reduce inequalities within and between nations, with the
whole populations can balance the growing numbers of expectation of health with age, including a lower risk of
older people. Given that the global population is ageing, dementia. Only an integrated approach that incorporates
identifying the factors that contribute to reductions in lifecourse health and brings together many disciplines
the prevalence and incidence of dementia in particu- underpinned by neuroscience and population-based
lar countries and regions should become a major pri- epidemiological studies can provide the robust evidence
ority, as the findings will have important implications required to understand the observed changes.
1. Prince, M. et al. World Alzheimer Report 2015. The 15. Roehr, S., Pabst, A., Luck, T. & Riedel-Heller, S.G. 31. Sekita, A. et al. Trends in prevalence of Alzheimer’s
global impact of dementia: an analysis of prevalence, Secular trends in the incidence of dementia in high- disease and vascular dementia in a Japanese
incidence, cost and trends. Alzheimer’s Disease income countries: a protocol of a systematic review and community: the Hisayama Study. Acta Psychiatr.
International https://www.alz.co.uk/research/ a planned meta-analysis. BMJ Open 7, e013630 (2017). Scand. 122, 319–325 (2010).
WorldAlzheimerReport2015.pdf (2015). 16. Erkinjuntti, T., Østbye, T., Steenhuis, R. & Hachinski, V. 32. American Psychiatric Association. Diagnostic and
2. Department of Health. G8 dementia summit The effect of different diagnostic criteria on the Statistical Manual of Mental Disorders 3rd edn
declaration. GOV.UK https://www.gov.uk/ prevalence of dementia. N. Engl. J. Med. 337, revised (American Psychiatric Association, 1987).
government/publications/g8-dementia-summit- 1667–1674 (1997). 33. Folstein, M., Folstein, S. & McHugh, P. R. A practical
agreements/g8-dementia-summit-declaration 17. Grimmer, T. et al. Trends of patient referral to a memory method for grading the cognitive state of patients for
(2013). clinic and towards earlier diagnosis from 1985–2009. the clinician. J. Psychiatr. Res. 12, 189–198 (1975).
3. World Health Organization. First WHO ministerial Int. Psychogeriatr. 27, 1939–1944 (2015). 34. Lawton, M. P. & Brody, E. M. Assessment of older
conference on global action against dementia. 18. Kukull, W. A. et al. Interrater reliability of Alzheimer’s people: self-maintaining and instrumental activities
World Health Organization http://www.who.int/ disease diagnosis. Neurology 40, 257–260 (1990). of daily living. Gerontologist 9, 179–186 (1969).
mental_health/neurology/dementia/ministerial_ 19. Pérès, K. et al. Trends in the prevalence of dementia 35. Crimmins, E. M., Kim, J. K., Langa, K. M. & Weir, D. R.
conference_2015_report/en/ (2015). in French farmers from two epidemiological cohorts. Assessment of cognition using surveys and
4. Brayne, C. & Davis, D. Making Alzheimer’s and J. Am. Geriatr. Soc. 65, 415–420 (2017). neuropsychological assessment: the Health and
dementia research fit for populations. Lancet 380, 20. Hebert, L. et al. Change in risk of Alzheimer disease Retirement Study and the Aging, Demographics, and
1441–1443 (2012). over time. Neurology 75, 786–789 (2010). Memory Study. J. Gerontol. B Psychol. Sci. Soc. Sci.
5. Wu, Y.‑T. et al. Dementia in Western Europe: 21. Rocca, W. A. et al. Trends in the incidence and 66, i162–i171 (2011).
epidemiological evidence and implications for policy prevalence of Alzheimer’s disease, dementia, and 36. Schrijvers, E. M. C. et al. Is dementia incidence
making. Lancet Neurol. 15, 116–124 (2016). cognitive impairment in the United States. Alzheimers declining? Trends in dementia incidence since 1990 in
6. Brayne, C., Stephan, B. C. M. & Matthews, F. E. Dement. 7, 80–93 (2011). the Rotterdam Study. Neurology 78, 1456–1463
European perspective on population studies of 22. Manton, K., Gu, X. & Ukraintseva, S. Declining (2012).
dementia. Alzheimers Dement. 7, 3–9 (2011). prevalence of dementia in the U.S. elderly population. 37. Grasset, L. et al. Trends in dementia incidence:
7. Hofman, A. et al. The prevalence of dementia in Adv. Gerontol. 16, 30–37 (2005). evolution over a 10‑year period in France. Alzheimers
Europe: a collaborative study of 1980–1990 findings. 23. Langa, K. M. et al. Trends in the prevalence and Dement. 12, 272–280 (2016).
Int. J. Epidemiol. 20, 736–748 (1991). mortality of cognitive impairment in the United States: 38. Matthews, F. E. et al. A two decade dementia incidence
8. Fratiglioni, L. et al. Incidence of dementia and major is there evidence of a compression of cognitive comparison from the Cognitive Function and Ageing
subtypes in Europe: a collaborative study of morbidity? Alzheimers Dement. 4, 134–144 (2008). Studies I and II. Nat. Commun. 7, 11398 (2016).
population-based cohorts. Neurologic Diseases in the 24. Wiberg, P., Waern, M., Billstedt, E., Östling, S. & 39. Gao, S. et al. Dementia incidence declined in African-
Elderly Research Group. Neurology 54, S10–S15 Skoog, I. Secular trends in the prevalence of dementia Americans but not in Yoruba. Alzheimers Dement. 12,
(2000). and depression in Swedish septuagenarians 244–251 (2016).
9. EU Joint Programme – Neurodegenerative Disease 1976–2006. Psychol. Med. 43, 2627–2634 (2013). 40. Satizabal, C. L. et al. Incidence of dementia over
Research. 21st century EURODEM. Redefining 25. Wimo, A. et al. Cohort effects in the prevalence and three decades in the Framingham Heart Study.
dementia for epidemiology. JPND research http:// survival of people with dementia in a rural area in N. Engl. J. Med. 374, 523–532 (2016).
www.neurodegenerationresearch.eu/wp-content/ Northern Sweden. J. Alzheimers Dis. 50, 387–396 41. World Health Organization. The ICD‑10 classification
uploads/2015/02/21st-Century-EURODEM.pdf (2016). of mental and behavioural disorders: clinical
(2014). 26. Qiu, C., von Strauss, E., Bäckman, L., Winblad, B. descriptions and diagnostic guidelines. World Health
10. Bell, J. F. et al. Existing data sets to support studies of & Fratiglioni, L. Twenty-year changes in dementia Organization http://www.who.int/classifications/icd/en/
dementia or significant cognitive impairment and occurrence suggest decreasing incidence in central bluebook.pdf (1992).
comorbid chronic conditions. Alzheimers Dement. 11, Stockholm, Sweden. Neurology 80, 1888–1894 42. American Psychiatric Association. Diagnostic and
622–638 (2015). (2013). Statistical Manual of Mental Disorders 4th edn
11. Alzheimer Association. 2015 Alzheimer’s disease facts 27. Lobo, A. et al. Prevalence of dementia in a southern (American Psychiatric Association, 2000).
and figures. Alzheimers Dement. 11, 332–384 (2015). European population in two different time periods: 43. Doblhammer, G., Fink, A. & Fritze, T. Short-term
12. World Health Organisation & Alzheimer’s Disease the ZARADEMP Project. Acta Psychiatr. Scand. 116, trends in dementia prevalence in Germany between
International. Dementia: a public health priority. 299–307 (2007). the years 2007 and 2009. Alzheimers Dement. 11,
World Health Organization http://www.who.int/ 28. Matthews, F. E. et al. A two-decade comparison of 291–299 (2015).
mental_health/publications/dementia_report_2012/en/ prevalence of dementia in individuals aged 65 years 44. Doblhammer, G., Fink, A., Zylla, S. & Willekens, F.
(2012). and older from three geographical areas of England: Compression or expansion of dementia in Germany?
13. Sachdev, P. S. Is the incidence of dementia declining? results of the Cognitive Function and Ageing Study I An observational study of short-term trends in
A report for Alzheimer’s Australia. Alzheimer’s and II. Lancet 382, 1405–1412 (2013). incidence and death rates of dementia between
Australia https://www.fightdementia.org.au/sites/ 29. Hall, K. S. et al. Prevalence rates for dementia and 2006/07 and 2009/10 based on German health
default/files/Paper_39_Is_the_incidence_of_dementia_ Alzheimer’s disease in African Americans: 1992 versus insurance data. Alzheimers Res. Ther. 7, 66 (2015).
declining.pdf (2014). 200. Alzheimers Dement. 5, 227–233 (2009). 45. Rocca, W. A., Cha, R. H., Waring, S. C. & Kokmen, E.
14. Prince, M. et al. Recent global trends in the prevalence 30. Langa, K. et al. A comparison of the prevalence of Incidence of dementia and Alzheimer’s disease:
and incidence of dementia, and survival with dementia in the United States in 2000 and 2012. a reanalysis of data from Rochester, Minnesota,
dementia. Alzheimers Res. Ther. 8, 23 (2016). JAMA Intern. Med. 177, 51–58 (2017). 1975–1984. Am. J. Epidemiol. 148, 51–62 (1998).
46. Sposato, L. A. et al. Declining incidence of stroke and 62. Zeng, Y., Feng, Q., Hesketh, T., Christensen, K. & 81. Rorsman, B., Hagnell, O. & Lanke, J. Prevalence and
dementia: coincidence or prevention opportunity? Vaupel, J. W. Survival, disabilities in activities of daily incidence of senile and multi-infarct dementia in the
JAMA Neurol. 72, 1529–1531 (2015). living, and physical and cognitive functioning among the Lundby Study: a comparison between the time periods
47. Kosteniuk, J. G. et al. Simultaneous temporal trends oldest-old in China: a cohort study. Lancet 389, 1947–1957 and 1957–1972. Neuropsychobiology
in dementia incidence and prevalence, 2005‑2013: 1619–1629 (2017). 15, 122–129 (1986).
a population-based retrospective cohort study in 63. Larson, E. B., Yaffe, K. & Langa, K. M. New insights 82. Mathillas, J., Lövheim, H. & Gustafson, Y. Increasing
Saskatchewan, Canada. Int. Psychogeriatr. 29, into the dementia epidemic. N. Engl. J. Med. 369, prevalence of dementia among very old people.
1643–1658 (2016). 2275–2277 (2013). Age Ageing 40, 243–249 (2011).
48. Abdulrahman, G. O. Alzheimer’s disease: current 64. de Bruijn, R. F. et al. The potential for prevention of 83. Li, S. et al. Is the dementia rate increasing in Beijing?
trends in Wales. Oman Med. J. 29, 280–284 dementia across two decades: the prospective, Prevalence and incidence of dementia 10 years later
(2014). population-based Rotterdam Study. BMC Med. 13, in an urban elderly population. Acta Psychiatr. Scand.
49. Bertrand, M., Tzourio, C. & Alperovitch, A. Trends in 132 (2015). 115, 73–79 (2007).
recognition and treatment of dementia in France 65. Hachinski, V. Stroke and potentially preventable 84. Arias, E., Heron, M. & Xu, J. United States life tables,
analysis of the 2004 to 2010 database of the National dementias proclamation. Updated World Stroke Day 2012. Centers for Disease Control and Prevention
Health Insurance plan. Alzheimer Dis. Assoc. Disord. proclamation. Stroke 46, 3039–3040 (2015). https://www.cdc.gov/nchs/data/nvsr/nvsr65/
27, 213–217 (2013). 66. Skoog, I. Dementia: dementia incidence — the times, nvsr65_08.pdf (2016).
50. Ukraintseva, S., Sloan, F., Arbeev, K. & Yashin, A. they are a‑changing. Nat. Rev. Neurol. 12, 316–318
Increasing rates of dementia at time of declining (2016). Acknowledgment
mortality from stroke. Stroke 37, 1155–1159 67. Stern, Y. Cognitive reserve. Neuropsychologia 47, We would like to thank Ms Lesile Grasset (Bordeaux study),
(2006). 2015–2028 (2009). Dr Sujuan Gao (Indianapolis–Idaban Dementia Project) and
51. Chien, I. C. et al. Treated prevalence and incidence of 68. Blazer, D. G. et al. Cognitive Aging: Progress in Professor Anders Wimo (Nordanstig study) for providing
dementia among National Health Insurance enrollees Understanding and Opportunities for Action (National age-specific and sex-specific prevalence and incidence
in Taiwan, 1996–2003. J. Geriatr. Psychiatry Neurol. Academies Press, 2015). estimates.
21, 142–148 (2008). 69. Honda, H. et al. Trends in autopsy-verified dementia
52. Menec, V. H., Lix, L. & MacWilliam, L. Trends in the prevalence over 29 years of the Hisayama study. Author contributions
health status of older Manitobans, 1985 to 1999. Neuropathology 36, 383–387 (2016). All authors researched data for the article and reviewed
Can. J. Aging 24 (Suppl. 1), 5–14 (2005). 70. Wakutani, Y. et al. Longitudinal changes in the and/or edited the manuscript before submission. Y.-T.W. and
53. Dodge, H. H. et al. Trends in the prevalence of prevalence of dementia in a Japanese rural area. C.B. wrote the article and made substantial contributions to
dementia in Japan. Int. J. Alzheimers Dis. 2012, Psychogeriatrics 7, 150–154 (2007). discussion of the content.
956354 (2012). 71. Jones, D. S. & Greene, J. A. Is dementia in decline?
54. Kim, K. W. et al. A nationwide survey on the Historical trends and future trajectories. N. Engl. Competing interests statement
prevalence of dementia and mild cognitive impairment J. Med. 374, 507–509 (2016). The authors declare no competing interests.
in South Korea. J. Alzheimers Dis. 23, 281–291 72. Kiyohara, Y. Epidemiology of dementia: the Hisayama
(2011). study [Japanese]. Nihon Rinsho 72, 601–606 (2014). Publisher’s note
55. Yu, R. et al. Trends in prevalence and mortality of 73. Kishimoto, H. et al. The long-term association between Springer Nature remains neutral with regard to jurisdictional
dementia in elderly Hong Kong population: physical activity and risk of dementia in the community: claims in published maps and institutional affiliations.
projections, disease burden, and implications for the Hisayama study. Eur. J. Epidemiol. 31, 267–274
long-term care. Int. J. Alzheimers Dis. 2012, 406852 (2016). Review criteria
(2012). 74. Yoshitake, T. et al. Incidence and risk factors of vascular We conducted a literature search in PubMed using search
56. Fuh, J. & Wang, S. Dementia in Taiwan: past, present, dementia and Alzheimer’s disease in a defined elderly terms (“dementia” OR “Alzheimer”) AND (“prevalence” OR
and future. Acta Neurol. Taiwan 17, 153–161 Japanese population: the Hisayama study. Neurology “incidence”) AND (“trend*” OR “change*”) to identify relevant
(2008). 45, 1161–1168 (1995). publications up to February 2017. We selected potential
57. Zhang, Y. et al. Prevalence of dementia and major 75. Ohara, T. et al. Glucose tolerance status and risk of population-based studies which investigated change in preva
dementia subtypes in the Chinese populations: a meta- dementia in the community: the Hisayama study. lence or incidence of dementia using similar methods across
analysis of dementia prevalence surveys, 1980–2010. Neurology 77, 1126–1134 (2011). time. We also cross-checked the reference lists of previous
J. Clin. Neurosci. 19, 1333–1337 (2012). 76. Mueller, S. G. et al. Ways toward an early diagnosis in reports on dementia trends. We excluded studies in which
58. Chan, K. Y. et al. Epidemiology of Alzheimer’s disease Alzheimer’s disease: the Alzheimer’s Disease medical records or healthcare administrative databases were
and other forms of dementia in China, 1990–2010: Neuroimaging Initiative (ADNI). Alzheimers Dement. 1, used, systematic reviews, meta-analyses and those focusing
a systematic review and analysis. Lancet 381, 55–66 (2005). only on Alzheimer disease and cognitive impairment. We also
2016–2023 (2013). 77. Savva, G. M. et al. Age, neuropathology, and dementia. excluded five studies owing to methodological limita-
59. Wu, Y.‑T., Brayne, C. & Matthews, F. E. Prevalence of N. Engl. J. Med. 360, 2302–2309 (2009). tions20–22,24,25. In total, we included 14 full-text studies from
dementia in East Asia: a synthetic review of time 78. Snowdon, D. A. Healthy aging and dementia: findings Sweden, the Netherlands, Spain, France, the UK, the USA,
trends. Int. J. Geriatr. Psychiatry 30, 793–801 from the Nun Study. Ann. Intern. Med. 139, 450–454 Japan and Nigeria, and all were published in English.
(2015). (2003).
60. Wu, Y.‑T. et al. Period, birth cohort and prevalence of 79. Flicker, L. & Holdsworth, K. Aboriginal and Torres Strait
dementia in mainland China, Hong Kong and Taiwan: islander people and dementia: a review of the research. FURTHER INFORMATION
a meta-analysis. Int. J. Geriatr. Psychiatry 29, Alzheimer’s Australia https://www.fightdementia.org.au/ UN data: http://data.un.org/
1212–1220 (2014). files/NATIONAL/documents/Alzheimers-Australia- Gapminder: http://www.gapminder.org/data/
61. Gao, M. et al. The time trends of cognitive impairment Numbered-Publication-41.pdf (2014).
incidence among older Chinese people in the 80. Breeze, E., Hart, N. J., Aarsland, D., Moody, C. & SUPPLEMENTARY INFORMATION
community: based on the CLHLS cohorts from 1998 to Brayne, C. Harnessing the power of cohort studies for See online article: S1 (table) | S2 (table) | S3 (table)
2014. Age Ageing https://www.dx.doi.org/10.1093/ dementia research. J. Public Mental Health 14, 8–17 ALL LINKS ARE ACTIVE IN THE ONLINE PDF
ageing/afx038 (2017). (2015).
CORRIGENDUM