Alzheimer’s disease and related dementias in South and Southeast Asia: a systematic review and meta-analysis of epidemiology and associated factors

Abstract

Objectives

Alzheimer’s disease and related dementias (AD/ADRDs) are rising global public health challenges, disproportionately affecting ageing populations. Although higher prevalence estimates have been reported in high-income countries (HICs), the most rapid increase in dementia burden is projected to occur in low- and middle-income countries (LMICs), particularly in South and Southeast Asia. Genetic predisposition and non-modifiable risk factors interplay with vascular, lifestyle and environmental factors, shaping future disease burden. However, evidence from South and Southeast Asia remains fragmented, limiting targeted policy responses. This review systematically synthesises available evidence on prevalence and associated factors to inform prevention and care strategies.

Design

Systematic review and meta-analysis of AD/ADRDs prevalence and associated factors related studies from South and Southeast Asia.

Data sources

Five electronic databases including PubMed, Scopus, CINAHL, EMBASE, APA PsycNet and hand searching of reference lists up to 1 May 2025.

Eligibility criteria for selecting studies

The designed study included observational studies, specifically cross-sectional, casecontrol and cohort designs that reported the epidemiological characteristics and associated factors of AD/ADRDs.

Data extraction and synthesis

Study screening, data extraction and methodological assessments were conducted by two independent reviewers. Risk of bias was assessed using the modified Newcastle-Ottawa Scale (mNOS). A restricted maximum likelihood model was used to synthesise results, complemented by synthesis of associated risk factors. Between-study heterogeneity was assessed using I² statistic.

Results

A total of 31 studies were included, of which 13 were analysed to estimate pooled prevalence (N=56 340). The methodological quality was good (mean mNOS=7.1, SD 0.99). The pooled age of onset of AD was 70.7 (SD 5.5) years, with one-third (46.3%) of the men surviving for 5 years. The overall estimated prevalence of dementia was 3.0% (95% CI 2.0% to 5.0%, I²=98.7%), AD 2.0% (95% CI 1.0% to 4.0%, I²=99.0%) and vascular dementia 1.0% (95% CI 0% to 1.0%, I²=93.7%). Subgroup analysis showed that AD prevalence was 3.0% in LMICs (95% CI 1.0% to 6.0%) compared with 2.0% in HICs (95% CI 1.0% to 4.0%), and increased to 8.0% after 2010 (95% CI 5.0% to 14.0%) compared with 1.0% before 2000 (95% CI 0% to 2.0%). Additionally, results suggested that older adults had higher odds of AD (OR: 3.98, 95% CI 1.77% to 8.99%), while higher literacy, adequate sleep, periodontal therapy and cognitive stimulation were identified as protective factors. No clear evidence of an association between female gender, family history and AD risk was observed in the pooled analysis.

Conclusions

AD affects a substantial and growing proportion of populations in South and Southeast Asia, with marked heterogeneity across settings. Evidence of an upward trend over the time in LMICs poses significant challenges in the coming years. The rising prevalence, associated factors particularly in resource-limited contexts, underscores the need for early-screening in clinical practices and region-specific epidemiological research.

PROSPERO registration number

CRD 420251047105.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Comprehensive regional synthesis of Alzheimer’s disease and related dementias epidemiology in South and Southeast Asia using subgroup meta-analysis.

• Conducted following rigorous methodology, and reported according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA-2020) and meta-analysis guidelines.

• Clinical diagnosis without biomarker confirmation, may introduce misclassification bias.

• Limited data from several countries may affect regional prevalence estimates.

• Risk factors were synthesised through meta-analysis, but causality cannot be inferred.

Introduction

Alzheimer’s disease and related dementias (AD/ADRDs) constitute a rapidly expanding global public health epidemic marked by progressive neurodegeneration, cognitive decline and functional impairments.¹ Collectively, AD/ADRDs affect about 57 million people globally, including 5.5 million Americans in the USA, and this number is projected to reach 8.5 million within the next 5 years.² These disease burdens rank among the leading causes of death and disability, posing an enormous challenge to healthcare systems.³ The global prevalence of all-cause dementia is estimated at 6.97% among individuals aged ≥50 years,⁴ with the highest observed in the USA (9.7% ≥70 years), followed by Europe (4.4% ≥65 years), China and West Pacific regions (4% ≥60 years) and sub-Saharan Africa (1.6%).^{5 6}

While dementia is an umbrella clinical syndrome characterised by impaired cognition and daily functioning, AD/ADRDs refer to specific four disorders, including AD, vascular dementia (VaD), dementia with Lewy bodies (LBD) and frontotemporal dementia (FTD).⁷ Among AD/ADRDs, AD accounts for nearly 60%–70% of all dementia cases and is the fifth leading cause of mortality globally.⁸ AD not only impairs cognition and quality of life but also creates substantial economic and emotional strain, with annual per-capita costs ranging from US$468 to US$171 284. This wide variation reflects differences in healthcare systems, health insurance, informal caregiving burden, disease severity, institutional versus community-based care, economic disparities between countries and healthcare infrastructures.⁹ Although the reported prevalence of dementia is lower in East Asia (0.89%) and South Asia (0.44%)¹⁰ compared with the global prevalence, this likely reflects under-reporting and limited prior screening initiatives.¹¹ The burden is intensifying in low- and middle-income countries (LMICs) particularly in South and Southeast Asia where rapid population ageing, limited diagnostic infrastructure and lack of long-term care amplify disease impact.^{10 12} While high-income countries (HICs) are expected to observe a twofold increase in cases by 2040, LMICs like India, China and other Asian and Pacific nations are projected to experience more than a three-fold rise.¹³ The proportion of people aged ≥60 years is expected to double from 12% in 2015 to 22% by 2050,¹⁴ with South and Southeast Asia witnessing the fastest growth. According to Global Burden of Disease 2019, dementia prevalence in these regions has risen by over 150% since 1990, with AD as the predominant subtype.¹⁰ Consequently, LMICs are projected to bear 37% of the global disability burden (South Asia 20% and Southeast Asia 17%)^{15 16} (dataset). These gaps underscore the need for region-specific evidence to inform evidence-based national strategies and design culturally appropriate prevention, diagnostic and care interventions.

About 70% of AD risk is attributed to genetic susceptibility, while age, female sex and family history further increase the vulnerability.¹⁷ Modifiable risk factors, notably vascular diseases, low education, comorbidities, poor diet, psychiatric illness and environmental exposures^{16 18} contribute roughly ~35% of the global population attributable fraction. In contrast, higher education, cognitive engagement, physical activity and social networks may reduce risk by up to 45%. The 2024 Lancet Commission estimated that addressing modifiable risks could potentially delay nearly half of the global AD burden.^{19 20} Despite the rapidly increasing burden of AD/ADRDs in LMICs, evidence from South and Southeast Asia remains fragmented and under-represented in global estimates. Current global estimates are largely country-specific, methodologically heterogeneous, predominantly derived from HICs, while LMICs suffer from limited surveillance systems and scarcity of epidemiological data.²¹ Systematic reporting of existing evidence is crucial to address the future disease burden and seek public health importance. This study addresses this critical gap from South Asia and Southeast Asian regions by providing pooled regional estimates of AD/ADRDs prevalence and associated risk factors and by generating region-specific evidence from under-represented LMIC settings.

Methods

Study design

This review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA-2020),²² Meta-Analysis of Observational Studies in Epidemiology²³ guidelines to ensure methodological rigour and was prospectively registered in PROSPERO²⁴ (CRD 420251047105). The review was conducted in accordance with the previously published protocol. Detailed methodology of the study was described in the pre-published protocol.²¹

Research objective and research question

Our primary objective was to synthesise epidemiological data from South and Southeast Asia to estimate pooled prevalence of AD/ADRDs, including subgroup analysis by study settings, time frame and income level. The secondary objective was to synthesise the associated risk (modifiable and non-modifiable) and protective factors using standard meta-analysis and explore their association with AD. Additionally, where quantitative synthesis was not feasible for meta-analysis, risk factors were described narratively as described in the published protocol.²¹

Based on the PECOT: population (P), exposure (E), comparison (C), outcome (O) and time frame (T) framework, this review will answer the following research question: ‘What is the epidemiology of AD/ADRDs and its associated risk factors among South and Southeast Asian population?’²³

Search strategy, outcome definition and selection criteria

The literature search was conducted in PubMed (via NLM), Scopus (via Elsevier), CINAHL (via EBSCOhost), EMBASE (via Elsevier) and APA PsycNet (via APA). In addition, to ensure the comprehensiveness and quality, citation chaining and grey literature searching were performed to identify additional eligible studies. A comprehensive search strategy was developed with input of a librarian, using Medical Subject Headings (MeSH) terms, truncated keywords and database filters, as described in prior protocol.²³ Preliminary search strings are summarised in online supplemental table 1A, with complete strategies in online supplemental file 1.

Eligible studies included empirical, population-based observational designs reporting quantitative AD/ADRDs estimates in South Asian (Afghanistan, Bangladesh, Bhutan, India, Sri Lanka, Maldives, Nepal, Pakistan) and/or Southeast Asian (Brunei, Cambodia, Indonesia, Laos, Malaysia, Myanmar, Philippines, Singapore, Thailand, Timor-Leste, Vietnam) populations, using validated diagnostic tools were presented in online supplemental table 1B. Studies lacking specific AD/ADRDs data, using non-validated criteria, focusing other populations, or with inaccessible full texts, or having multiple publications from the same project will be excluded. Full criteria were detailed in the published protocol.²³ To ensure comprehensive and unbiased study inclusion, Rayyan (V.1.4.3) and EndNote (V.25.0) were used. The web-based application (Rayyan) facilitated independent blinded screening by three independent reviewers, thus improved screening efficiency, transparency and minimised selection bias during study inclusion. In addition, reference management software (EndNote) was used for systematic reference management and duplicate article removal. A total of 467 articles were identified from five databases (n=423) and hand searches (n=44). After removing 94 duplicates, two reviewers (AT and TS) screened titles and abstracts, excluding 284 articles. Inter-rater reliability was moderate (Cohen’ k=0.41; and percentage agreement=87%), with discrepancies resolved by discussion and consensus with third reviewer (SS). The selection process is depicted in the PRISMA flow diagram.²⁴

For longitudinal studies reporting prevalence at multiple time points, only the baseline reported prevalence estimate (a single effect estimate) was extracted to avoid duplication of participants within the meta-analysis. When studies reported subgroup-specific prevalence estimates, pooled estimates for the overall study population were preferentially extracted where available. In addition, risk factor outcomes were defined as the association between predefined exposures (eg, age, sex, education, family history and lifestyle factors) and AD. Effect estimates (OR with 95% CI) were extracted from each study.

Two reviewers (AT and TS) independently extracted data, cross-checked by a third (SS) using a standardised Excel form. Extracted data included study details, patient demographics, diagnostic methods and epidemiological outcomes (prevalence, incidence, survivability, mortality). Studies reporting overlapping populations were identified,^{25 26} and the overlap was considered when extracting and synthesising the data to avoid treating the same population as independent. For pooled prevalence estimation, only one study reporting prevalence estimation²⁵ was considered and analysed from studies with overlapping populations to avoid double-counting of participants. Associated factors were qualitatively (meta-analysis) synthesised and grouped into five categories (a) demographic and genetic, (b) socioeconomic and occupational, (c) vascular and psychological, (d) lifestyle and nutritional and (e) environmental and other factors to enhance interpretability of heterogeneous correlates. Moreover, ORs with 95% CIs of risk factors were also recorded where available. Inter-rater reliability was assessed to ensure data accuracy and transparency. RoB was independently assessed by two reviewers following the modified NOS²⁷ for prevalence studies, covering selection (five stars); comparability (two stars) and exposure/outcome (three stars). For comparability, studies when adjusted for key confounders including age, sex and at least one major sociodemographic or clinical variable (eg, education or comorbidity) were considered as two stars. Studies were categorised as unsatisfactory (0–4), satisfactory (5–6), good (7–8) and very good (9–10). Detailed quality (assessment results) of individual studies is provided in online supplemental table 3, with conflicts resolved through discussion.

Patients and public involvement

This study did not directly involve patients, caregivers or any public entity.

Statistical analysis and data visualisation

A random-effects meta-analysis model with restricted maximum likelihood (REML) estimation was adopted in R (V.4.3.2, meta packages) to estimate only the pooled prevalence of all-cause dementia (ADRDs) and AD²⁸ to account for both within-study sampling error and between-study heterogeneity. The Freeman-Tukey double arcsine transformation was applied to stabilise the variance of prevalence estimates before pooling. Between-study heterogeneity was quantified using the I² statistic (percentage of variability) and τ² (between-study level variance), and potential sources of heterogeneity were further explored through subgroup and sensitivity analyses. Descriptive analysis summarised study characteristics and patient demographics and diagnostic tools; where ‘n’ denotes the number of studies, and ‘N’ denotes the total number of participants. Subgroup analyses were performed across dementia subtypes (all-cause dementia, AD), time period (before 2000, 2000–2010, after 2010), economic settings (HICs, LMICs), region (rural, urban, mixed) and gender using the REML model with 95% CIs and visualised via forest plots. Heterogeneity was evaluated using the I² statistic classified as low (≤25%), moderate (>25% to <75%) or high (≥75%) with significance at p<0.10. Funnel plots were used for visual inspection to explore potential publication bias and small study effects. Associated risk factors with quantitative estimates were meta-analysed where feasible, otherwise synthesised narratively. Risk factors were further synthesised and visualised as heatmaps across five predefined categories. Where ≥2 studies reported ORs with 95% CIs (age, gender, education, family history and depression), pooled estimates were derived using REML with inverse variance weighting, log ORs were analysed and back-transformed for interpretation. Univariate meta-regression analysis was conducted to explore potential sources of heterogeneity using study-level covariates, including age (years), sex (female), sample size (per 1000) and quality assessment (modified Newcastle-Ottawa Scale, mNOS) scores. Sensitivity analysis employed the Jackknife (leave-one-out) approach using ‘metainf ()’, to assess study influence on pooled estimates.

Results

Characteristics of the included studies and study participants

A total of 31 studies representing 30 independent study populations from South and Southeast Asia met the eligibility criteria, summarised in online supplemental table 1B. Two studies from India reported data from an overlapping population, with one conducting a prevalence survey²⁵ and the other conducting a 2-year follow-up incidence survey.²⁶ Of these, 21 were from South Asia including 16 from India^{5 25 26 29–35} and Pakistan,^36–41 followed by Bangladesh,^{42 43} Nepal^{44 45} and Sri Lanka ⁴⁶(figure 1). In Southeast Asia, most originated from Singapore^{11 47 48} and Indonesia.^{49 50} Study designs were mainly cross-sectional (n=22), with five case–control and four were longitudinal. 10 studies were conducted in LMICs and three in HICs. The study periods spanned three decades, before 2000 (1991 to 1999), between 2000 and 2009, and after 2010 (2010 to 2024), with 12 were population-based studies (N=55 520), while four studies (N=4435) were dementia or memory clinic-based.^{11 49 51 52} Together, these studies (n=13) represented 56 340 participants, including 861 had dementia and 671 AD cases. Sample sizes ranged from 45 to 24 488. Among the AD cases age of onset ranged from 40 to 100 (based on the age category of majority included studies ≥60 years threshold was defined as older adulthood or aged group rather than the universal definition, while 40–59 years was reported as middle-aged adults) with a pooled mean age of 70.7 (SD 5.5) years and 59.88% (N=291) were female. Eight studies were conducted among urban populations, three rural, and two included both settings. Educational attainment was low in the majority of the studies, and familial history ranged from 16.0% to 31.7%. Across population-based 12 studies, prevalence estimates were meta-analysed from samples of adults (≥40 years), while four studies in memory clinics (as outpatient basis hospital services) were analysed separately as complementary clinical evidence rather than representative population-based estimates. Substantial heterogeneity was observed across studies (I² ≥50%) indicating that the study results varied considerably across included studies, and thus a random-effects model was applied for pooled analysis. Eight studies (n=8) reported on comorbidities, most commonly type-II diabetes mellitus, hypertension, smoking, stroke, ischaemic heart disease and obesity reflecting a strong vascular-metabolic profile among regional cohorts.

Figure 1. PRISMA flow chart showing the study screening process throughout the review. PRISMA flow chart representing study identification, screening and final inclusion from PubMed, Scopus, CINAHL, EMBASE and APA PsycNet using institutional access (icddr,b). A two-stage screening process was applied before final study inclusion. *Two included studies from India^{25 26} represented an overlapping study population; only the prevalence study²⁵ was included in the pooled prevalence analysis to avoid double-counting participants, considering 30 independent study populations from South and Southeast Asia (SA/SEA). ¹Rayyan, a web-based tool used to screen and manage studies during a systematic review. AD, Alzheimer’s disease; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Heterogeneity in the screening practice and case identification of AD/ADRDs

Majority of the studies (n=14) used multiple screening tools to identify probable or possible cases (online supplemental table 1B and online supplemental table 2) following standard diagnostic criteria like National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA). However, other criteria like National Institute on Aging-Alzheimer’s Association (NIA-AA) were also used across the studies. To categorise the patients, Mini–Mental State Examination (MMSE) was a commonly adopted tool to assess cognitive impairment among AD/ADRDs patients across included studies, though the cut-off scores were slightly different across studies^{30 38 53–55} (online supplemental file 2). In addition to MMSE, Clinical Dementia Rating (CDR) was used to classify dementia. Six studies incorporated clinical history, neuroimaging (CT scan and MRI)¹¹ and biochemical tests to exclude reversible causes like hypothyroidism or vitamin B₁₂ deficiency for AD/ADRDs classification.^{11 29 32 38 46 56}

Pooled prevalence of AD/ADRDs

Across 12 population-based studies (N=55 520), the pooled prevalence of dementia was 3.0% (95% CI 2.0% to 5.0%, I²=98.7%) (figure 2A). 13 studies (N=56 340) estimated the pooled prevalence of AD was 2.0% (95% CI 1.0% to 4.0%, I²=99.0%). The 95% CI indicates the range within which the true pooled prevalence is likely to lie with 95% certainty, while high heterogeneity (I² ≥75%) indicates considerable variability in prevalence estimates across studies rather than a single consistent estimate (figure 2B).

Figure 2. Forest plots of pooled prevalence estimates dementia and AD. Forest plots showing the pooled prevalence of (A) dementia and (B) AD in South and Southeast Asia. Pooled estimates were derived using a REML model, with data from South Asia (n=8 for dementia; n=9 for AD) and Southeast Asia (n=4 for both outcomes). REML, Random-effects meta-analysis.

Subgroup analysis

By country and outcome: Regional estimates were comparable between South Asia (2.0%) and Southeast Asia (2.0%). Country-specific prevalence of AD across the regions was depicted in figure 3. Studies from memory clinic settings (n=3 studies, N=4348 participants) reported a substantially higher pooled prevalence of dementia 68% (95% CI 8.0% to 98%, I²=98.6%)^{11 49 51} (online supplemental figure 1A) and 29% (95% CI 9.0% to 62%, I²=98.5%) for AD (n=4, N=4435)^{11 49 51 52} (online supplemental figure 1B). However, the wide CI indicates considerable uncertainty around the point estimate, suggesting that the true prevalence may differ substantially from the point estimates. The small number of studies (n=3 for dementia, and n=4 for AD) and substantial variability between them might have contributed to the uncertainty and high heterogeneity (I² ≥75%) of the pooled estimates. 11 population-based studies reported the prevalence of non-AD dementia, with VaD being the most common, ranging from 0.06% to 2.86%,^{45 46} followed by Multi-Infarct Dementia at 1.43%,⁴⁷ mixed dementia at 0.03%–0.05%^{29 46} and LBD at 0.14%⁴⁶ (online supplemental table 2). The pooled prevalence of VaD across 10 studies (n=10, N=50 045) was 1.0% (95% CI 0% to 1.0%, I²=93.7%), indicating a lower point estimate compared with AD, although substantial variability was observed between studies (online supplemental figure 1C). In contrast, studies from memory clinic settings reported higher and more variable prevalence rates of VaD (8.3%–31.2%), mixed dementia (4.4%–31.5%), FTD 6.16%, Parkinson’s Disease Dementia 0.19% and autoimmune dementia 0.89% (online supplemental table 2).

Figure 3. Prevalence of AD across South and Southeast Asian countries based on extracted data. Data were synthesised from included studies, covering South Asia (n=9) and Southeast Asia (n=11). Each country is depicted in a distinct colour by the authors (no copyright issue). Reported prevalence estimates were available from India, Pakistan, Sri Lanka, Nepal, Singapore, Thailand and the Philippines published included studies. For countries with multiple studies, mean prevalence values were calculated.

By region, sex, study period and income settings

Subgroup analysis showed a pooled prevalence of AD was 2.0% in both rural (n=3, N=8639) and urban areas (n=8, N=44 947) and 5.0% in mixed urban–rural (n=2, N=2754) settings (95% CI 1.0% to 35%, I²=97.4%) (figure 4A). By study period, pooled prevalence was 1.0% (n=4, N=31 897) before 2000 (95% CI 0% to 2.0%), 2.0% (n=5, N=21 624) between 2000 and 2010 (95% CI 1.0% to 3.0%) and rose sharply to 8.0% (n=4, N=2819) after 2010 (95% CI 5.0% to 14%) (figure 4B). Gender-specific prevalence was similar 1.0% in both male and female participants (n=9, N=52 568) (figure 4C). By income settings, studies from LMICs (n=10, N=39 682) showed higher pooled prevalence of 3.0% (95% CI 1.0% to 6.0%, I²=99.1%) compared with 1.0% (n=3, N=16 658) in HICs (95% CI 1.0% to 2.0%, I²=99%) (figure 4D). Across the subgroups (A and B) considerable high heterogeneity was observed, indicating substantial variability in prevalence estimates between studies within each subgroup, meaning the results were not consistent across studies. The subgroup analysis did not fully explain the observed heterogeneity in main estimate, as substantial variability persisted within most subgroups.

Figure 4. Forest plots of subgroup analysis of AD prevalence. (A) Subgroup by demographic region, using a REML model: rural (n=3), urban (n=8) and both (n=2) studies from South and Southeast Asia. (B) Subgroup by time frame: studies conducted before 2000 (n=4), between 2000 and 2010 (n=5) and after 2010 (n=4). (C) Subgroup by gender (male, female), based on data from nine studies. (D) Subgroup by country income setting: LMICs included South Asia (n=9; India (5), Sri Lanka, Nepal, Pakistan (2)) and Southeast Asia (n=1; Philippines); HICs included Southeast Asia (n=3; Singapore (2), Thailand). For this analysis, low- and lower-middle-income countries were grouped as LMICs, while upper-middle-income and high-income countries were grouped as HICs. AD, Alzheimer’s disease; HICs, high-income countries; LMICs, low- and middle-income countries; REML, Random-effects meta-analysis.

Meta-regression analysis

Univariate meta-regression analysis was conducted for age, sex, sample size and quality assessment scores (mNOS). None of the covariates were found significant effects for predicting dementia and AD (p>0.05) outcomes, suggesting that none of them had influenced the pooled prevalence estimates or account for heterogeneity across studies (table 1).

Table 1. Univariate meta-regression for dementia and AD.

Variables	β (95% CI)	SE	P value
Dementia
Age (years)	0.123 (−0.181 to 0.427)	0.155	0.428
Sex (female)	−0.021 (−0.181 to 0.427)	0.091	0.821
Sample size (per 1000)	−0.094 (−0.370 to 0.182)	0.141	0.503
Quality assessment score (mNOS)	−1.025 (−4.039 to 1.988)	1.538	0.505
Alzheimer’s Disease (AD)
Age (years)	0.047 (−0.129 to 0.223)	0.090	0.598
Sex (female)	0.014 (−0.089 to 0.118)	0.053	0.789
Sample size (per 1000)	−0.130 (−0.290 to −0.030)	0.082	0.110
Quality assessment score (mNOS)	−0.185 (−1.934 to 1.565)	0.893	0.836

N.B: Age is expressed in years; sex represents the percentage of female participants; study quality was assessed using the mNOS and sample size is scaled per 1000 participants across the included studies.

AD, Alzheimer’s disease; mNOS, modified Newcastle-Ottawa Scale.

Incidence, severity and survivability of AD/ADRDs

Two studies from India^{26 32} (N=2698 and N=1066) reported AD incidence rate of 1.7 per 1000 person-years (95% CI 0.8 to 3.2), which increased to 11.7 (95% CI 10.9 to 12.4) person-years among individuals aged ≥55 years, providing additional epidemiological evidence on the incidence of AD across the included studies. In summary, AD occurred more frequently in older adults (≥55 years), with a much higher incidence compared with the overall population. The overall incidence of all-cause dementia between 2001 and 2011 was 12.05 per 1000 person-years.³² Four studies (n=4) assessed AD severity,^{33 36 43 45} identifying moderate AD being the most prevalent (31.03%–60%). In Singapore,¹¹ a tertiary memory clinic study (N=2890) found young-onset dementia (YOD) accounted for 15.8%, with young-and-late onset AD representing 6.9% and 26.4%, respectively. Two studies investigated survivality.^{25 57} One reported (N=5126) a median survival of 2.7 years from onset of disease for individuals with CDR score ≥1, while another found 5-year survival rates of 46.3% in males, 34.3% in females and overall (40.8%), indicating variation in survival outcomes across studies (online supplemental table 2).

Associated factors of AD

A synthesis (standard meta-analysis) of associated factors (both modifiable and non-modifiable), divided into five categories, is presented in online supplemental table 4).

1. Demographic and genetic: Majority of the studies (n=15) identified advanced age or older adults (≥60 years), ethnicity, familial history and genetic (APOE ε4)/familial predisposition as key non-modifiable risk factors for AD in this region (figure 5 and online supplemental table 4). Meta-analysis of six studies (n=6) showed a higher AD risk among older adults (aged ≥60 years) compared with middle-aged (40–59 years) individuals (OR: 3.98, I²=89.4%, p<0.001) (figure 6A). No significant association between sex (female) and AD risk (n=6) was observed in the pooled analysis (OR: 1.09, 95% CI 0.68 to 1.74) compared with male in six included studies, although moderate heterogeneity was present across studies (I²=58.3%, p=0.035) (figure 6B). Two studies (N=2054) assessed familial history of AD and reported a possible association with increased risk; however, the pooled estimate was highly imprecise (OR: 2.28, 95% CI 0.11 to 45.13), with substantial heterogeneity (I²=85.3%, p=0.009), indicating considerable uncertainty around the direction and magnitude of the effect (figure 6C).

2. Socioeconomic and occupational: Half of the studies (n=11) reported modifiable risk factors for AD, including early life illiteracy, low level of education or some education (≤4 years), low socioeconomic status and occupational factors including stressful profession (figure 5 and online supplemental table 4). A meta-analysis of five studies (N=23 896) indicated no significant association between low education (≤4 years) and AD (OR: 0.78, 95% CI 0.48 to 1.26), with moderate heterogeneity (I²=65.6%, p=0.02), indicating variability in effect estimates across studies (figure 6D).

3. Vascular and psychological: Vascular and metabolic comorbidities in mid and late-life (n=11 studies) including type II diabetes, obesity, dyslipidaemia, hypertension, cardiovascular diseases (MI, IHD) and cerebrovascular events (stroke) were substantially associated with AD (figure 5). Psychological factors reported in two (n=2) studies including mid-life and late-life depression and poor mental health also emerged as potential triggers for late-life.^{34 36 42} Meta-analysis of two studies (N=14 287) from Indonesia⁵⁰ and the Philippines⁵⁸ showed, depression was associated with lower odds of AD (OR: 0.85, 95% CI 0.46 to 1.57, p=0.88), although no statistical significance was evident from the findings (figure 6E).

4. Lifestyle and nutritional: Lifestyle factors such as midlife tobacco and alcohol consumption identified as modifiable risk factors for AD in several studies (n=7).^35–37 Five studies reported nutritional factors including unhealthy dietary patterns, malnutrition, low bone mineral density (BMD) and vitamin B₁₂ deficiency were also linked to the disease progression^{34 40 44 50 52} (figure 5).

5. Environmental and others: Environmental factors such as exposure to poor, non-purified water exposure, pollutants, toxins, metals, along with rural dwelling and urbanisation were linked to AD risk in later-life reported in two studies.^{35 39} Modifiable factors (n=5) including poor sleep quality, sleep disturbances and oral health problems like chronic periodontitis, tooth loss, early tooth extraction, gingivitis and poor oral hygiene were also associated with AD development in late life^{33 34 41 50 57} (figure 5).

Figure 5. Risk factors (modifiable and non-modifiable) of AD in South and Southeast Asia. The left panel (Lt) categorises reported risk factors into five broad domains: (A) demographic (age, gender, ethnicity) and genetic (family history); (B) socioeconomic (education, socioeconomic status) and occupational (stressful profession); (C) vascular (diabetes, obesity, dyslipidaemia, hypertension, stroke, cardiovascular disease) and psychological (depression, anxiety); (D) lifestyle (tobacco, alcohol) and nutritional (diet, malnutrition, vitamin B₁₂ deficiency, low bone mineral density) and (E) environmental (residence, water exposure, pollutants) and other factors (sleep quality and disorders, oral hygiene, periodontitis, gingivitis). The right panel (Rt) showed a Sankey diagram mapping modifiable (green) and non-modifiable (red) risk factors across early, mid and late life. AD, Alzheimer’s disease.

Figure 6. Forest plots of associations between AD and selected risk factors. (A) Association between AD and age (≥60 years). (B) Association between AD and female gender. (C) Association between AD and family history of dementia. (D) Association between AD and low education (≤4 years). (E) Association between AD and depression. AD, Alzheimer’s disease.

Protective factors of AD

Four studies (n=4) identified potential protective factors for AD,^{25 42 50 57} including adequate sleep, periodontal therapy, high literacy and cognitive interventions such as training, rehabilitation and cognitive stimulation therapy (CST), all of which showed modest yet significant cognitive benefits. Interestingly, a case–control study from Indonesia reported that a history of stroke appeared to be a protective factor for AD incidence compared with other dementias (online supplemental table 3, online supplemental file 2).⁵⁰

Quality of the included studies

Based on the mNOS online supplemental table 3, 2 studies rated very good (9 points), 21 were good (7–8 points) and 8 were satisfactory (5–6 points). The overall mean mNOS score was 7.06 (SD 0.99) across the included studies, indicating overall good methodological quality of included evidence. Two studies^{44 50} included controls, although the recruitment methods were unclear (Selection), while two studies^{25 47} lack subgroup analysis (Comparability). Case confirmation methods varied, four studies^{11 38 46 56} used radiological assessments, while others relied on cognitive tools or clinical subgrouping.^{5 30 31 48} Diagnostic cut-off values were undefined in nine studies.^{35 39 40 42 44 49–51 57} Additionally, two studies^{29 47} provided insufficient statistical details, omitting p values and subgroup stratification methods (Outcome) (online supplemental file 2).

Publication bias and sensitivity analysis

Funnel plots for dementia (n=12) and AD (n=13) appeared symmetrical visually, suggesting minimal publication bias and no strong evidence of small-study effect (online supplemental figure 2AB). Sensitivity analysis showed that omitting any single individual study did not substantially alter pooled estimates, with all individual studies remaining highly significant (p<0.001), indicating that the estimates were robust and not driven by any individual study (online supplemental figure 2CD).

Discussion

This systematic review and meta-analysis revealed that dementia affected about 3.0% of the population in South and Southeast Asia, with AD and VaD comprising most cases. AD was the predominant subtype, contributing over half of all dementia diagnoses. Compared with global estimates, pooled prevalence in these regions were slightly lower, with substantial heterogeneity across studies. In addition, vascular, lifestyle, psychological, environmental and sociodemographic factors particularly advancing age, low educational years and vascular comorbidities (eg, hypertension, stroke) were identified as individual risk factors for AD across the population.

While dementia prevalence in Latin America (8.5%) and the UK (7.2%) exceeds regional estimates, Southeast Asia and sub-Saharan Africa reported similar or lower rates (2.0%–4.0%).^59–61 AD prevalence (~2.0%) of the current study aligns with findings from Iran (2.3%),⁶² China Coast (3.1%)⁶³ and Nigeria (2%)⁶⁴ but remains lower than the USA (9.7%),⁶⁵ Japan (7.1%),⁶⁶ Europe (5.05%)⁶⁷ and China (5%).⁶⁸ The comparative low prevalence may be attributed to multifactorial causes. In many LMIC settings, under-reporting and limited access to cognitive screening and specialist diagnostic services may lead to substantial underdiagnosis. Differences in population age structure, with a relatively smaller proportion of older adults, may further contribute to lower observed prevalence. In addition, socioeconomic disparities, educational attainment and healthcare access influence case detection, while genetic, lifestyle, ethnic and cultural factors may modify disease expression and help-seeking behaviours. Finally, the scarcity of large, population-based epidemiological studies and inconsistent diagnostic criteria across regions likely reflect in underestimation of the true disease burden.^{6 23}

Variations likely reflect socioeconomic disparities, vascular risk exposure, genetic susceptibility and healthcare access.⁶⁹ Earlier Asian studies reported higher VaD prevalence, but recent evidence showed an AD shift, likely reflecting improved vascular risk management (0.96% to 1.09%) and hypertension control rates (26% to 47%).^{3 69} This progress, reflecting the flagship priority given to non-communicable disease prevention and primary care strategies. AD prevalence rose from 1.0% before 2000 to 8.0% after 2010, reflecting higher study quality in newer studies, better diagnostic facilities, addition of biomarker with clinical and neuroimaging techniques and rapid socioeconomic changes,⁶⁷ consistent with global trends showing a 160.84% rise in AD incidence between 1990 and 2019.^{10 68} However, unlike western nations¹⁰ where early detection has reduced rates, survival outcomes remain poorer in these regions with median survival 2.7 years²⁵ and 5-year survival 46% in males⁵⁷ compared with 7.2 years for typical AD and 11.3 years for YOD cases globally.⁷⁰

Subgroup analysis revealed no difference in AD prevalence between urban and rural populations, aligning with findings from China (2010-2020)⁶⁸ and Germany.⁷¹ However, several reviews reported higher prevalence in rural areas,^{3 63 71 72} whereas studies from South Asia^{26 30 39} found consistent disparity, except for one study from Vietnam by Tran et al⁵² that observed a high prevalence in urban regions. These rural-urban disparities likely reflect variations in modifiable risk factors, healthcare access and prevention strategies. In rural settings, ORs ranging from 0.89 to 1.79,⁷² suggesting that improved chronic disease management, healthcare awareness and expanded social insurance had helped reduce the gap.^{30 68} Conversely, in urban areas, greater exposure to air pollution, psychological stress and reduced green spaces may elevate AD risk. In addition, despite performing subgroup analyses to explore potential sources of heterogeneity; however, substantial variability persisted across most subgroups in our study, suggesting that other unmeasured factors including differences in study design, population characteristics (demographics) and diagnostic criteria may not fully explain the observed heterogeneity, which had also been evident in few other global studies.^{4 19 33} Furthermore, our pooled estimate suggested a tendency towards increased AD risk among individuals with a family history, although the association was not statistically significant and the estimate was highly imprecise. This finding contracts with the strong association reported by Shaji et al in a substantially larger population, as well as other previous studies reporting increased AD risk among individuals with a first-degree relative with a history of dementia.^{30 50 58} The discrepancy may partly reflect the small number of studies contributing to the pooled analysis and substantial heterogeneity between studies. Moreover, previous evidence suggests that familial risk of dementia may not be specific to AD but may extend to other neurological disorders, highlighting the complexity of interpreting family history as a risk factor.^{64 73} Nevertheless, the direction of our pooled estimate suggested a potential familial contribution to AD risk, warranting further investigation in larger, well-designed longitudinal studies.

We found that the pooled prevalence of AD in LMICs was nearly double that observed in HICs. With population ageing, AD/ADRDs are projected to rise disproportionately in LMICs, which by 2050, will account for 65% of the global value of a statistical life years burden (up from 18% in 2019), largely driven by upper-middle-income countries (92% of LMIC share and 60% global).⁷⁴ By 2050, nearly 80% of all AD cases are expected to occur in LMICs, particularly across South and Southeast Asia,²³ where up to 90% remain underdiagnosed according to the World Alzheimer Report 2021.¹² This underdiagnosis reflects persistent disparities in diagnostic capacity, inadequate care infrastructure, overburdened systems, lack of referral system and insufficient early screening, and poor biomarker availability.⁷³ Despite growing burden, most quantitative estimates still originated from HICs, with only ~10% derived from LMICs.⁷⁵ Within Southeast Asia, research is concentrated in two HICs (Singapore and Brunei), while studies from other LMICs remain scarce.²³ Addressing this evidence gap, requires large-scale, region-specific, cutting-edge epidemiological and translational research in LMICs, particularly among Asian populations. Although global meta-analysis has often reported higher AD risk among females, our review found no comparable prevalence between sexes in South and Southeast Asia. A meta-analysis of 998 187 participants from 43 countries similarly reported no sex difference in AD incidence,⁷⁶ consistent with our findings, and other population-based studies from Stockholm⁷⁷ and Rotterdam,⁷⁸ both reporting a 1:1 male-to-female ratio. In contrast, a multiregional systematic review across Asia, South America, Europe/North America and Africa identified a higher prevalence among women,⁴ possibly due to genetic susceptibility (apolipoprotein E-4 allele, telomere shortening), hormonal influence, pregnancy-related complications and psychiatric comorbidities. Country-level differences in life expectancy, education and gender disparities may further explain this variability.

Low educational attainment, identified as a major risk factor in our review, may increase vulnerability to AD/ADRDs by reducing cognitive reserve, whereas higher education supports cortical integrity and language processing. Education has been shown to lower AD risk by up to 36% and delay onset by ~24%.^{8 68} Among non-modifiable factors age, sex and APOE-ε4 genotype remain dominant predictors. Evidence from Singapore reported higher age-adjusted prevalence among Indians compared with Malays and Chinese, suggesting genetics and lifestyle contributors.^{3 8} Protective factors including social engagement, healthy relationships, cognitive activity and bilingualism enhance cognitive resilience (reducing ~30% to 45% future risk), while widowed individuals particularly APOE ε4 carriers, face elevated AD risk.⁸ Furthermore, prior history of vascular events (stroke) does not imply stroke prevents AD biologically, rather, it reflects diagnostic and pathological distinctions which shift AD towards VaD through inflammatory responses.⁵⁰ Depression also exhibits a bidirectional link with AD via inflammatory and hypothalamic-pituitary-adrenal axis dysregulation, the HUNT study reported anxiety‐and depressive symptoms decades before AD/ADRDs onset, underscoring the need to address midlife mental distress.⁷⁹ Nutritional and metabolic factors including malnutrition, dyslipidaemia, low vitamin B₁₂ or folate, anaemia and reduced BMD are linked with impaired cognition, and reduced albumin levels,^{3 44} while elevated homocysteine and altered lipid metabolism markers may serve as early predictors for AD. Multimorbidity, nearly threefold higher in AD patients and affecting ~40% in South and Southeast Asia¹⁰ highlighting the complex interplay of biological and systematic determinants. Integrative bioinformatics approaches combining comorbidity and metabolic pathway data may help elucidate underlying mechanisms and identify novel therapeutic targets.

Strengths and limitations of the review

The review has several limitations. Although studies from 19 countries were included, no eligible data were available from four South Asian and Southeast Asian nations, and participants number were small relative to the region’s 2.79 billion population. Only one large multistaged study (Mumbai, India) and four post-2010 were identified. Large population-based data from least-developed countries were lacking, despite the region comprising ~33.8% of the global population and exhibiting substantial demographic, environmental and healthcare variations that may introduce bias.^{10 80} Substantial heterogeneity across the studies could not be explained despite subgroup and meta-regression analysis, and the heterogeneity remained high across subgroup analyses. Furthermore, wide CI due to small number of studies and substantial variability across studies might impact the confidence in the point estimate, which needs to be interpreted cautiously. In addition, variability in diagnostic criteria and cut-off values likely contributed to the observed heterogeneity and may have influenced the reported prevalence estimates. Inclusion of probable or possible AD/ADRDs based on clinical diagnostic criteria, without biomarker confirmation, may introduce ascertainment bias due to misdiagnosis. Furthermore, inclusion was restricted to English language publications, potentially introducing language bias; all grey literature sources were not systematically captured, small-study effects, and most included studies were cross-sectional, which may limit the generalisability of the findings. In addition, moderate study quality could also have led to over- or underestimation. Furthermore, insufficient quantitative data prevented reliable analysis of individual risk factors. Despite these constraints, this review adhered to standard reporting guidelines and comprehensive search methods, providing the most up-to-date and robust estimates of AD/ADRDs prevalence and subgroup variations in South and Southeast Asia.

Potential implications and future directions

This review highlighted a regional overview of AD/ADRDs in South and Southeast Asia, emphasising the need for standardised diagnostic criteria, clinicians’ training, early detection of cognitive impairment in clinical settings and large-scale research to investigate the true disease burden. As 40%-80% of dementia cases remain undiagnosed in primary settings, early identification and screening of high-risk groups are crucial. Delaying Alzheimer’s onset by just 5 years could reduce prevalence and caregiving needs by ~36%.^{10 23 68} Until, public awareness remains limited regarding the disease course and associated correlates while, only half of the population had adequate knowledge and 11% were aware of the risk factors.²³ Strengthening community awareness, promoting dementia-friendly environments and integrating lifestyle interventions into public health initiatives are essential. Given the disease’s progressive burden, caregiver training and psychological support should be prioritised. To serve the disease burden, annual per capita costs ranged from US$468 (mild) to over US$171 283 (severe), contributing to the projected global cost of US$2 trillion by 2030.⁹ Large-scale longitudinal studies are warranted to clarify country-specific risk stratification. Global funders (eg, like Alzheimer’s Society, WHO, NINDS) should support affordable, sensitive diagnostics facility for low-resource settings. Community-based and digital interventions including bibliotherapy, CST could offer promising avenues to improve well-being and caregiver outcomes particularly in LMICs.^{35 43}

Conclusions

We observed an obvious epidemiological transition of AD/ADRDs prevalence across South and Southeast Asia. Our meta-analysis showed that about 2 in every 100 people is living with the risk of AD and the number is rising with population ageing. Higher prevalence was observed among older adults and women where no major geographic differences. The growing burden in LMICs poses serious economic and caregiving challenges which urgently urge for early-screening for cognitive impairments in routine clinical practices. However, addressing modifiable risk factors from early to midlife, increasing community awareness, early screening and targeted strategies could help to reduce the disease burden and slow the progression. Consistent diagnostic criteria, early detection in clinical settings and future large-scale epidemiological and bioinformatics-based translational research are needed to better understand the demographic diversity in neurodegenerative diseases.

Data availability statement

Data are available in a public, open access repository. Data are available on reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.