The Prevalence and Incidence of Dementia Due to Alzheimer’s Disease: a Systematic Review and Meta-Analysis

Abstract Background Updated information on the epidemiology of dementia due to Alzheimer’s disease (AD) is needed to ensure that adequate resources are available to meet current and future healthcare needs. We conducted a systematic review and meta-analysis of the incidence and prevalence of AD. Methods The MEDLINE and EMBASE databases were searched from 1985 to 2012, as well as the reference lists of selected articles. Included articles had to provide an original population-based estimate for the incidence and/or prevalence of AD. Two individuals independently performed abstract and full-text reviews, data extraction and quality assessments. Random-effects models were employed to generate pooled estimates stratified by age, sex, diagnostic criteria, location (i.e., continent) and time (i.e., when the study was done). Results Of 16,066 abstracts screened, 707 articles were selected for full-text review. A total of 119 studies met the inclusion criteria. In community settings, the overall point prevalence of dementia due to AD among individuals 60+ was 40.2 per 1000 persons (CI95%: 29.1-55.6), and pooled annual period prevalence was 30.4 per 1000 persons (CI95%: 15.6-59.1). In community settings, the overall pooled annual incidence proportion of dementia due to AD among individuals 60+ was 34.1 per 1000 persons (CI95%: 16.4-70.9), and the incidence rate was 15.8 per 1000 person-years (CI95%: 12.9-19.4). Estimates varied significantly with age, diagnostic criteria used and location (i.e., continent). Conclusions The burden of AD dementia is substantial. Significant gaps in our understanding of its epidemiology were identified, even in a high-income country such as Canada. Future studies should assess the impact of using such newer clinical diagnostic criteria for AD dementia such as those of the National Institute on Aging–Alzheimer’s Association and/or incorporate validated biomarkers to confirm the presence of Alzheimer pathology to produce more precise estimates of the global burden of AD. RÉSUMÉ Prévalence et incidence de la démence due à la maladie d’Alzheimer : revue systématique et méta-analyse. Contexte: Nous avons besoin d’informations sur l’épidémiologie de la démence due à la maladie d’Alzheimer (MA) afin de nous assurer que des ressources adéquates sont disponibles pour satisfaire les besoins actuels et futurs de la population en soins de santé. Nous avons effectué une revue systématique et une méta-analyse de l’incidence et de la prévalence de la MA. Méthodologie: Nous avons effectué une recherche dans les bases de données MEDLINE et EMBASE de 1985 à 2012 ainsi que dans la liste de références d’articles retenus. Les articles retenus devaient fournir des estimations de l’incidence et/ou de la prévalence populationnelle de la MA. Deux évaluateurs ont revu indépendamment les résumés et le texte intégral ainsi que l’extraction des données des publications et en ont évalué la qualité. Nous avons utilisé des modèles à effets aléatoires pour générer des estimations regroupées stratifiées par âge, sexe, critères diagnostiques, lieu (continent) et temps (moment où l’étude a été réalisée). Résultats: Parmi les 16 066 résumés examinés, 707 articles ont été retenus pour une revue du texte intégral. En tout, 119 études rencontraient les critères d’inclusion. Dans la communauté, la prévalence ponctuelle globale de la démence due à la MA chez les individus de 60 ans et plus était de 40,2 par 1 000 (IC à 95%: 29,1 à 55,6) et la prévalence annuelle pour les données regroupées était de 30,4 par 1 000 (IC à 95%: 15,6 à 59,1). Dans la communauté, la proportion d’incidence annuelle globale regroupée de la démence due à la MA chez les individus de 60 ans et plus était de 34,1 par 1 000 (IC à 95%: 16,4 à 70,9) et le taux d’incidence était de 15,8 par 1 000 personnes-années (IC à 95%: 12,9 à 19,4). Les estimations variaient significativement selon l’âge, les critères diagnostiques utilisés et le lieu (continent). Conclusions: Le fardeau de la démence dû à la MA est considérable. Nous avons identifié des lacunes importantes dans notre compréhension de son épidémiologie, même dans un pays à revenu élevé comme le Canada. Des études ultérieures devraient évaluer l’impact de l’utilisation de critères diagnostiques plus récents pour identifier la démence due à la MA tels ceux du National Institute on Aging-Alzheimer’s Association et/ou incorporer des biomarqueurs validés pour confirmer la présence de la pathologie de la MA et fournir des estimations plus précises de son fardeau global.


INTRODUCTION
2][3] Up to 70% of the dementias occurring in older adults are attributed in whole or in part to AD. 4 Though described more than a century ago, 5 treatment options remain limited.Available pharmacotherapies provide modest symptomatic benefits 6 of debatable cost-effectiveness. 7pdated information on the epidemiology of dementia due to AD is needed if we are to ensure that adequate resources are mobilized to deal with the needs of those with this condition and their families.Such studies can also inform prevention strategies and approaches to management.Systematic reviews on the epidemiology of dementia generally do not deal with specific causes such as AD, but rather provide estimates of overall dementia. 8,9The last systematic review of the global incidence of dementia specifically due to AD was published in 2008. 10While the age-specific incidence rate of AD dementia doubles approximately every 5.5 years in older populations 11 and several studies have produced estimates stratified by sex and geographic region, 10,[12][13][14] an unexplored issue is the heterogeneity produced by differing diagnostic criteria and study setting.In a majority of the more recent reviews, either a systematic methodology was not utilized 10,12,15 or it was uncertain whether one was. 14,16,17n this report, we present an updated systematic review and meta-analysis of population-based studies of the incidence and prevalence of dementia due to AD.We also examine the extent and causes of heterogeneity in these estimates.

METHODS
This is one in a series of systematic reviews on the prevalence and incidence of priority neurological conditions as part of the National Population Health Study of Neurological Conditions. 18

Search Strategy
The systematic review and meta-analysis were conducted according to a predetermined protocol based on the PRISMA statement for systematic reviews and meta-analyses. 19The search strategy (Appendix A) was developed by the study authors (who have expertise in dementia and/or disease epidemiology) in consultation with a research librarian with systematic review expertise.The primary search was conducted in the MEDLINE and EMBASE databases in February of 2011 and updated in July of 2012.References were exported and managed using EndNote X5. 20 International studies published before the year 2000 and Canadian studies published prior to 1985 were excluded because of the availability of prior meta-analyses summarizing earlier work.The earlier date for Canadian studies was to ensure that all relevant national work was included, as this was part of a nationally funded study examining the burden of neurological conditions in Canada.The review was restricted to articles published in English or French.The reference lists of included articles were manually searched for additional articles.

Study Selection
Two reviewers independently examined the titles and abstracts of all retrieved references in order to identify papers likely reporting original population-based data on the prevalence and/or incidence of AD dementia.Two reviewers also independently examined the fulltext papers identified during the first phase.To be included in the systematic review, reviewed papers had to: (1) report original research; (2) be population-based; and (3) provide an incidence and/ or prevalence estimate of dementia due to AD. Disagreements about the inclusion of articles were resolved by consensus or involvement of a third author if necessary.

Data Extraction and Study Quality
Two reviewers extracted data from included articles using a standard data collection form.Any disagreement was resolved by consensus.When multiple articles reported data on the same study population, the most accurate and comprehensive data as determined by the reviewers were used.In cases where the studies reported on different data collection years or subgroups (e.g., by sex and/or age), all data were included.The demographic data recorded included age, sex, setting (community-only, both community and institution, institution-only) and study location (i.e., Africa, Asia, Australia, Europe, North America, South America).The approach to ascertain cases was noted, as were sources of data and definitions/diagnostic criteria used.Incidence and prevalence estimates of AD dementia from each study were recorded, along with any stratification by age, sex or year of data collection.The quality of the included studies was evaluated using an assessment tool 21,22 (Appendix B), with each study given a quality score that ranged from 0 to 8 (higher scores indicating a higher-quality assessment).

Data Synthesis and Analysis
The significance of age, sex, diagnostic criteria, location (i.e., continent) and time (i.e., when the study was conducted) on incidence and prevalence estimates was assessed using metaregression.Age was examined using the youngest-aged person in the study as a continuous factor of potential heterogeneity (note that few studies provided data on mean or median age).Sex, diagnostic criteria and geographic location were treated as categorical variables.Changes over time were examined using the study start, midpoint and end-years as potential sources of heterogeneity.All pooled estimates were restricted to studies reporting on older individuals (i.e., aged 60 + , 65 + , 70 + ) to mitigate the potential confounding effects of age.All period estimates were converted to annual estimates (e.g., period prevalence represents the annual period prevalence).Studies were also stratified by the location of participants (i.e., communityonly, community and institution, institution-only) to minimize confounding by disease severity.Studies were included in the meta-analysis if they reported the estimate with 95% confidence intervals (CI 95% ), the number of AD cases along with overall sample size, or the information with which to calculate an estimate.Additionally, subgroup meta-analysis was only performed if more than one study was available for each subgroup (e.g., a region could have been omitted from this analysis if only one study was available in a region; however, if more than one study was included in the other regions, these data were then analyzed).
To compare study quality characteristics across groups (i.e., continent), ANOVA testing was utilized to determine differences.To assess for significant between-study heterogeneity, the Cochrane Q statistic was calculated and I 2 was employed to quantify the magnitude.All the pooled estimates and 95% confidence intervals were calculated using random-effects models.Publication bias was investigated visually using funnel plots and statistically using Begg's, 23 Egger's 24 and the trim-andfill tests.The trim-and-fill method identifies funnel plot asymmetry by imputing the effect estimates of potentially missing studies and assessing the influence of these studies on the pooled estimate.For all tests, a value of p less than 0.05 was deemed to be significant.All statistical analyses were carried out in R version 2.14. 25 The meta package was used to produce the pooled estimates, forest plots and publication bias assessments. 26he metafor package was used to conduct the meta-regression using restricted maximum-likelihood estimation. 27

Identification and Description of Studies
The search strategy yielded a total of 16,066 citations, including duplicates (8743 from MEDLINE, 7323 from EMBASE) (Figure 1).After screening titles and abstracts, 707 articles were selected for full-text review.Of them, 547 were excluded (230 international studies were published before 2000, 164 did not report incidence or prevalence of dementia, 114 were not population-based, 39 did not report original data).The updating of the search and hand searching the references led to an additional 4 and 12 articles, respectively.Among the 176 eligible papers meeting the inclusion criteria, 57 were excluded, as they did not report on the incidence or prevalence of AD dementia.A total of 119 papers reported on AD dementia.
The characteristics of the 119 included studies are shown in Tables 1-3.Seventy-five reported on prevalence, 43 on incidence and 1 on both.Forty-four studies provided data from Europe,

Prevalence of AD
31,33,35,36,38,39,42,45,46,48,57,[60][61][62][63]65,67,70 In community-only settings, the point prevalence among those 60 + years of age was 40.19 (CI 95% : 29.06-55.59)per 1000 (Figure 2). Point prevalence estimates in the community ranged fr a low of 15.51 per 1000 in one study from India 60 to a high of 204.13 per 1000 in a study from Israel.70 The pooled point prevalence in those 60 + in combined community and institution studies was 26.57(CI 95% : 11.83-59.69)per 1000.In community and institution studies, point prevalence estimates ranged from a low of 12.34 per 1000 in a study from the United States 35 to a high of 51.00 per 1000 in a study from Canada.28 The pooled point prevalence of AD among those 60+ in institution-only settings was 226.97 (CI 95% : 88. 23-583.87).
0,102 In community settings, the pooled annual period prevalence among those aged 60 + was 30.4 (CI 95% : 15.6-59.1)per 1000 (Figure 3).In combined community and institution settings, the pooled annual period prevalence was 44.0 (CI 95% : 19.9-97.1)per 1000.A single study from an institution in the United States reported an annual period prevalence of 101.0 (CI 95% : 89.4-114.1)per 1000. 93Annual period prevalence estimates for any setting ranged from 1.1 per 1000 in a community study from India 77 to 123.0 in a community study from the United States. 91

Incidence of AD
Fourteen studies reported on the incidence proportion of AD, [104][105][106][107][108][109][110][111][112][113][114][115][116][117] with six included in the meta-analysis of 60 + studies. 107,108,111,113,114,116In community settings, the pooled annual incidence proportion among those aged 60 + was 34.1 (CI 95% : 16.4-70.9)per 1000 (Figure 4).A single U.S. study reported on the annual incidence proportion in combined community and institution settings with an estimate of 27.2 (CI 95% : 22.2-33.3)per 1000. 105There were no studies from an institution-only setting.Annual incidence proportion estimates for any setting ranged from 11.5 per 1000 in a community study from Nigeria 108 to 97.8 per 1000 in a community study from the United States. 111140,[143][144][145] In community-only settings, the pooled incidence rate of AD among those 60 + was 15.8 (CI 95% : 12.9-19.4)per 1000 person-years (Figure 5).A single Italian study reported on the incidence rate in combined community and institution settings with an estimate of 7.0 (CI 95% : 5.5-8.9)per 1000 person-years. 120There were no studies from an institution-only setting.The lowest estimate for any setting was 7.0 (CI 95% : 4.8-10.3)per 1000 person-years in a community study from the Netherlands 145 and the aforementioned study from Italy, and the highest 30.0 (CI 95% : 25.4-35.5)per 1000 person-years in a community study from the United States. 143

Sources of Heterogeneity
The effect of important potential sources of heterogeneity (i.e., age, sex, diagnostic criteria, location [continent], time [when the study was done]) on incidence and prevalence estimates in those aged 60 + was assessed using univariate metaregressions.

Age
Increasing age was associated with increasing point prevalence, period prevalence, incidence rate and incidence proportion estimates (p < 0.001).

Sex
Though the differences did not reach statistical significance (p values ranged from 0.102 to 0.582), estimates of incidence and prevalence by sex of the subjects were higher in females than in males, in the 22 studies that reported on this.

Diagnostic Criteria for AD
Within community settings, DSM-IV criteria 147 (n = 2) produced a statistically significant (p = 0.044) higher estimate for AD dementia point prevalence (91.7 [CI 95% : 19.0-442.8]per 1000) than those based on NINCDS-ADRDA criteria for probable AD 148  No statistically significant differences between the aforementioned criteria were seen for period prevalence in the community (p = 0.065), though the association was in the same direction as seen in the pooled point prevalence.All incidence studies used NINCDS-ADRDA criteria for probable AD.

Time
There was no effect of the time of study initiation, midpoint or conclusion on point prevalence, period prevalence, incidence rate or incidence proportion estimates.

Publication Bias
For the period prevalence, point prevalence, incidence rate and incidence proportion of AD dementia, significant funnel plot asymmetry was not found using Begg's or Egger's test (p > 0.05).Upon visual inspection, the funnel plots appeared symmetrical.

Study Quality
The median study quality score for studies reporting on the incidence or prevalence of AD dementia was 6/8 (range 3-8) (Table 4).Study quality did not vary by continent based on the results of ANOVA analyses.

DISCUSSION
A substantial societal burden from AD dementia was demonstrated in our systematic review and meta-analyses.In community settings, the point prevalence of AD dementia among those 60 + was 40.2 per 1000, while its incidence proportion was 34.1 per 1000 and incidence rate was 15.8 per 1000 person-years.Despite the large number of studies included in our meta-analysis, the resulting estimates lacked precision at times due to significant statistical heterogeneity.Our finding that the period prevalence of AD dementia in community settings (30.4 per 1000 persons) was lower than the point prevalence (40.2 per 1000 persons) was unexpected and should be interpreted with caution.You would typically expect the opposite finding (i.e., a higher pooled estimate from the period prevalence studies).This was likely due to the significant heterogeneity (i.e., >99% for period prevalence studies) that existed between these two pools of studies, leading to wide confidence intervals.In addition, there were several outliers, particularly in the period prevalence estimates, which ranged from a low of 1.1 in India 77 to a high of 123.0 per 1000 persons in a U.S. study. 91

THE CANADIAN JOURNAL OF NEUROLOGICAL SCIENCES
Our exploration of the sources of this heterogeneity led to several interesting findings.There was an insufficient number of population-based institution studies to do meta-analyses for this setting, but our descriptive analysis indicated that incidence rate for AD dementia is higher in the community while prevalence is greater in institutions.This is not surprising.There are few at-risk individuals within institutions, and the high mortality rate from other causes in the small at-risk institutional group likely means that they will likely die from another cause before they have time to develop AD.The inclusion of an institutionalized sample as well as region-specific variation in the availability of facility-based care and/or likelihood for institutionalization can have a substantial impact on the estimated prevalence of AD dementia in a community. 149Institutionalization typically occurs as a result of the functional impairments, behavioural challenges and associated burden on family caregivers that arise as the disease progresses and largely explain the high prevalence in this setting.Information on the incidence and prevalence of AD stratified by setting is particularly relevant for planning resource allocation.We identified a significant gap when it comes to the population-based epidemiology of AD dementia in institutional and residential settings.Future studies are required to understand the true burden of AD dementia in long-term and supportive care facilities.The use of standardized assessments based on data abstracted from interRAI instruments to provide estimates for the prevalence of dementia in these settings holds promise, but it is unclear whether they could be utilized for estimates of dementia arising specifically from AD. 150 All estimates of incidence and prevalence were higher for females compared to males, though the differences were not statistically different.In economically developed nations, about two-thirds of individuals diagnosed with AD dementia are women. 151This is primarily due to the fact that women on average live longer than men, and increasing age is the most important non-genetic risk factor for AD dementia.Incidence studies suggest an age-dependent relationship between sex and likelihood of developing AD dementia.One of the studies we included noted differences in incidence rates by sex after 90 years of age. 142Other reports indicate that the incidence of AD dementia increases with age in both sexes until 85-90 years of age, after which it plateaus for men but continues to increase among women. 152,153A prior meta-analysis reported slightly longer doubling times with increasing age for AD dementia in men compared to women, 10 while another study reported that women tend to have a higher incidence at very advanced ages. 12These noted differences between the sexes could be due to methodological issues, the differential impact of historical environmental risk factors, or true biological differences in disease susceptibility between the sexes. 154Interestingly, recent data suggest that, relative to women, men who survive to older ages may exhibit a lower risk for developing AD because of a healthier cardiovascular risk profile. 155ifficulties examining the effect of the diagnostic criteria utilized to diagnose AD cases were encountered due to the ubiquitous use of NINCDS-ADRDA criteria.However, for point prevalence in community-only settings, DSM-IV criteria were found to produce significantly higher estimates than studies utilizing NINCDS-ADRDA criteria for probable AD (and possibly ICD-10 criteria 156 ); a trend in the same direction was also shown in community period prevalence estimates.The choice and operationalization of diagnostic criteria can have a large effect on estimated incidence and/or prevalence. 157,158In one study, the use of DSM-III criteria led to 29.1% of subjects receiving a dementia diagnosis compared to 13.7% when DSM-IV criteria were employed. 157Newer diagnostic criteria for AD decouple AD from the presence of a dementia and no longer require the presence of a memory impairment, the impact of which on epidemiological estimates of the incidence and prevalence of AD is yet unknown. 159In the future, a diagnosis of preclinical AD may be made on the basis of biomarkers, though clinical criteria will be required to diagnose symptomatic (i.e., mild cognitive impairment or dementia) AD. 160 We suspect that the National Institute on Aging-Alzheimer's Association clinical diagnostic criteria for dementia due to AD 161 will be used in future incidence and prevalence studies of AD.Studies are needed to assess the potential impact of using these newer diagnostic approaches compared to the criteria that have been used to date on the estimated incidence and prevalence of AD and trends over time.
Estimates of AD dementia incidence and prevalence tended to be higher in North America as compared to Asia, but these differences were not statistically different except for estimates of period prevalence in community settings.Geographical differences in epidemiological estimates of AD could be due to a variety of factors other than true differences in age-specific disease risk, such as differing screening methods and thresholds for diagnosis, age distribution of the assessed population, duration of survival after the onset of AD dementia, overall life expectancy and competing risks. 17Nonetheless, the possibility of true regional differences in AD incidence and prevalence has important implications.It is unlikely that the observed findings can be fully explained by differences in life expectancy.While North America has a relatively high life expectancy, estimates of life expectancy are similar or even higher in several Asian countries (e.g., Japan). 162Similar findings (i.e., lower estimates in Asia) have been reported for Huntington's disease and Parkinson's disease, where it is felt that differences in the distribution, life expectancy and degree of stigmatization associated with a diagnosis of the condition may contribute to variations in disease reporting. 163,164he methodology utilized for this systematic review and metaanalysis closely followed established guidelines.We feel we were able to identify most eligible studies as multiple sources of study ascertainment were employed.We found no evidence for publication bias.We did, however, find a good deal of statistical heterogeneity.
In order to accurately plan for future needs, there remains an ongoing requirement to provide accurate estimates of the incidence and prevalence of AD.Relying on older data may lead to either over-or underestimating the resources required if incidence and prevalence rates are changing over time.Though, using meta-regression analysis, we did not find that time had an effect on the incidence or prevalence of AD, this does not preclude the possibility of true changes in age-standardized incidence and/or prevalence rates for dementia from AD occurring either now or in the near future due to changes in the presence of risk factors at a population level. 165There could well be rising and/or falling rates of AD within specific nations or regions that could be obscured by looking at international changes.For dementia overall, as an example, there is a suggestion that   rates are falling in high-rate areas (often high-income countries) and might be rising in low-and middle-income countries, where premature mortality is decreasing. 166,167This underscores the need for future studies on the epidemiology of this important condition.

Table 1 :
Studies Reporting on the Prevalence of Alzheimer's Disease

Table 1 .
(Continued) ://doi.org/10.1017/cjn.2016.36Published online by Cambridge University Press 36 from Asia, 32 from North America, 5 from South America, 2 from Africa and 2 from Australia (2 studies reported data from more than one continent).Nine studies reported on those aged 60 + , 68 on those 65 + and 19 on those 70 + . https

Table 2 :
Studies Reporting on the Incidence Rate of Alzheimer's Disease ://doi.org/10.1017/cjn.2016.36Published online by Cambridge University Press https

Table 3 :
Studies Reporting on the Incidence Proportion of Alzheimer's Disease

Table 4 :
Quality assessment scores of Alzheimer's Disease incidence and prevalence studies

Table 4 .
(Continued) ://doi.org/10.1017/cjn.2016.36Published online by Cambridge University Press https Kirsten Fiest, Jodie Roberts, Colleen Maxwell, Eric Smith, Alexandra Frolkis, Adrienne Cohen, Andrew Kirk, Dawn Pearson, Tamara Pringsheim, and Andres Venegas-Torres have nothing to disclose.David Hogan holds the Brenda Strafford Foundation Chair in Geriatric Medicine, though receives no salary support from this.Nathalie Jetté has the following disclosures: Public Health Agency of Canada, Principal Investigator, research support; Canada Research Chair in Neurological Health Services Research, Researcher, research support; Alberta Innovates Health Solutions Population Health Investigator, Researcher, research support.