Global, regional and national population attributable fraction of potentially modifiable risk factors for dementia in the adult population: an analysis using global health databases

Abstract

Introduction

Approximately 45% of dementia cases are potentially modifiable; however, their population attributable fraction (PAF) may vary according to sociodeterminants and geographical regions. Thus, this descriptive, ecologic study aimed to estimate overall and individual global, regional and national PAFs due to 14 modifiable risk factors for dementia in adults.

Methods

Age-standardised prevalence, prevalence proportion and relative risk of risk factors for dementia were extracted for each nation from the Global Burden of Disease Study, WHO Observatory, World Bank and Lancet commission for dementia. Communalities by income group and weighted PAFs were calculated. Sociodemographic index (SDI), gross national income (GNI) and Health Access and Quality Index (HAQI) were used for correlation analysis with PAF.

Results

Approximately 34.03% (95% uncertainty interval (UI) 30.42% to 37.37%) of dementia cases were potentially preventable worldwide. South Asia accumulated the largest amounts of weighted overall PAFs (42.16% (95% UI 39.76% to 42.58%)) while Western Europe had the lowest amount (23.25% (95% UI 22.76% to 23.81%)). Globally, vision loss was the most important risk factor (4.04% (95% UI 2.88% to 5.09%), followed by low education (3.61% (95% UI 2.18% to 5.33%)) and inactivity (2.41% (95% UI 1.71% to 3.07%)). We identified regional and interregional patterns for alcohol, diabetes, hearing loss, inactivity, obesity, smoking and vision loss. Overall, PAF was negatively correlated with GNI (r=−0.39, p<0.0001), SDI (r=−0.39, p<0.0001) and HAQI (r=−0.37, p<0.0001). All individual risk factors except diabetes, smoking and inactivity were correlated with SDI and HAQI. Sensitivity analysis using a fixed communality evidenced changes in risk factors ranking within regions and distribution of PAFs.

Conclusions

One-third of dementia cases were potentially preventable worldwide, being vision loss, low education and inactivity their main contributors. Since estimates varied according to socioeconomic and regional-specific factors, tailored whole-population approaches should be prioritised.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Based on previous reports, approximately 45% of dementia cases are potentially preventable by controlling 14 risk factors associated with dementia, which may vary according to sociodeterminants and geographical regions.

WHAT THIS STUDY ADDS

• This study introduced a layer of variability, according to nations’ income group communality, to better understand the complexity behind the distribution of the population attributable fraction of dementia risk factors worldwide.

• Efforts to avoid or correct vision loss, increase education attainment and increase physical activity in populations should be encouraged across the world to reduce the global burden of dementia.

• However, our results suggest the importance of implementing tailored local and regional whole-population approaches to reduce the risk factor of dementia.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study identified priority risk factors for dementia at national, regional and global levels.

Introduction

Dementia is a major global health challenge, affecting over 55 million individuals worldwide and expected to triple by 2050 due to population ageing.¹ This progressive brain disorder not only diminishes individual quality of life but also imposes significant economic and social burden on families and healthcare systems. The urgency to identify and mitigate modifiable risk factors for dementia has gained momentum, with growing evidence suggesting that a considerable proportion of dementia cases may be preventable through targeted interventions.^{2 3}

The landmark Lancet Commission paper on dementia prevention, intervention and care highlighted 14 potentially modifiable risk factors, including hypertension, diabetes, obesity, smoking, physical inactivity, air pollution and low education.² Collectively, these factors were estimated to account for approximately 45% of dementia cases globally. However, while these estimates offer critical insights, they assume a constant distribution of dementia risk factors worldwide, neglecting the existence of regional and national variations. Previous studies have demonstrated significant geographic and socioeconomic disparities in the distribution and impact of modifiable risk factors for dementia.4,7 For instance, studies in Argentina⁸ and Brazil⁹ highlighted that lower socioeconomic status is associated with higher population attributable fraction (PAFs), particularly for risk factors such as hypertension, obesity and low education. Likewise, ethnic and racial differences have also been identified, with certain populations exhibiting higher dementia burden due to variations in risk factor prevalence and access to healthcare, emphasising the need for targeted prevention efforts.48,10

Acknowledging such variations is crucial for informing region-specific prevention strategies, particularly in low-income (LIC) and lower-middle income countries (LMICs) where the prevalence of risk factors and healthcare infrastructure differs markedly from high-income countries (HICs).^{5 7 11} Therefore, in this study, we aimed to address these gaps by calculating global, regional and national-level PAFs for potentially modifiable risk factors for dementia across 204 countries, 23 regions and four World Bank (WB) income groups. Specifically, we seek to: (1) Quantify the overall and risk factor-specific PAFs for dementia at the global, regional and national levels; (2) Examine geographic and income-related disparities in PAFs, highlighting the key drivers of variation and (3) Explore the relationship between PAFs and multidimensional indicators of health and socioeconomic development, including sociodemographic index (SDI), Health Access and Quality Index (HAQI) and gross national income (GNI). For this purpose, we leveraged the most recent prevalence data from the Global Burden of Disease (GBD) 2021 Study¹² and other global health datasets from the WHO and WB providing the first comprehensive global analysis of regional and national PAFs for dementia.

Methods

Study design

This descriptive, ecological study used national-level prevalence data from 12 out of the 14 potentially modifiable risk factors for dementia reported by the Lancet standing commission for dementia² to estimate their respective PAFs for 27 geographical regions, four WB income levels and 204 countries. Data from the remaining two risk factors (low-density lipoprotein (LDL) cholesterol and social isolation) were obtained from the Lancet standing commission for dementia² to estimate overall and subgroup PAF calculations only. Strengthening the Reporting of Observational Studies in Epidemiology checklist can be found as online supplemental material 1.

Locations definition

We subdivided the world into 27 geographical regions and 204 countries based on the GBD study 2021.¹² America was composed of two macro regions: High income North America and Latin America & the Caribbean. Subsequently, this last one was subclassified in five subregions: Caribbean, Central, Andean, Tropical and Southern Latin America. Europe was subdivided into three regions: Western, Central and Eastern Europe. Africa was divided into two macroregions: Northern Africa & Middle East and Sub-Saharan Africa, which was subdivided into four subregions: Central, Eastern, Western- and Southern-Sub-Saharan Africa. Asia was composed of five subregions: Central, East, South, Southeast Asia and High-income Asia-Pacific. By contrast, Oceania included only Pacific Ocean Island nations, while Australia and New Zealand were under the subregion of Australasia. Finally, we included the four WB income groups: LIC, LMIC, upper-middle income countries (UMIC) and HIC.¹³ Classification of nations by region could be found in online supplemental tables 1 and 2.

Risk factor definition and data sources

We used the GBD Study 2021¹² to obtain age-standardised prevalence rates per 100 000 people from five potentially modifiable risk factors for dementia (diabetes mellitus, hearing loss, vision loss, major depression and traumatic brain injury (TBI)). The GBD study 2021 extracted raw data from multiple primary sources including death certificates, scientific studies, governmental and non-governmental reports among others. Afterwards, prevalence modelling was performed using the DisMod-MR 2.1 tool, a Bayesian modelling software that incorporates data on several disease variables and their epidemiological relationships. Estimates were modelled for every country, even for those without primary data. According to the GBD study 2021, diabetes mellitus definition (including types I and II) was based on fasting plasma glucose concentration (7 mmol/L or greater) or use of either insulin or diabetes medications.¹² Hearing loss was defined as the quietest sound a person can perceive in their better ear according to pure-tone average of different audiometric thresholds.¹² Likewise, vision loss was defined considering the vision in the better-seeing eye mapped on the Snellen scale.¹² By contrast, major depression diagnosis ascertainment was based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV: 296.21–24, 296.31–34) criteria and the tenth International Classification Criteria (ICD-10: F32.0–9, F33.0–9).¹² Likewise, TBI was based on the ICD-10 (F07.2, F07.8, F07.81, F07.89, F07.9, S06, S07, T90.2 and T90.5).¹² Prevalence data for education attainment came from the UNESCO Institute for Statistics displayed in the WB portal.¹⁴ We defined low education attainment as the proportion of people (≥25 years) in a country with less than primary school education. For this purpose, we extracted nation-level data of the proportion of people (≥25 years) with at least primary school completed.¹⁴ This was computed as 1 minus the proportion of adults who had completed at least primary school. Likewise, age-standardised proportion of smoking per country was obtained from the WB portal.¹⁵ Estimates were modelled using population-based surveys and were defined as percentage of population over 15 years old currently using any tobacco product.¹⁵ By contrast, due to the lack of data for air pollution prevalence at nation level, we followed the definition from the Lancet standing commission for dementia² and used percentage of urban population as a proxy. Data was obtained from the WB portal, displaying data from World Urbanisation prospects from United Nations.¹⁶

Nation-level high alcohol consumption was extracted from the global health observatory of the WHO and was defined as age-standardised proportion of population over 15 years with at least one heavy episodic drinking (minimum 60 g or more of pure alcohol) in the past 30 days. Estimates were calculated based on population-based surveys.¹⁷ Obesity prevalence was defined as age-standardised proportion of body mass index higher than or equal to 30 kg/m² among adults (≥18 years).¹⁸ Nation-level data were obtained from the Global Health Observatory of the WHO.¹⁸ Obesity prevalence estimates were calculated based on population-based surveys. Hypertension prevalence was defined as age-standardised proportion of people with systolic blood pressure ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg or taking medication for hypertension among adults between 30 and 79 years.¹⁹ Like obesity, we extracted nation-level data from the Global Health Observatory of the WHO which based their estimates calculation on population-based survey and surveillance systems.¹⁹ Physical inactivity data were obtained from Strain et al under the definition of age-standardised proportion of people (≥18 years) not doing 75 min of vigorous-intensity activity, 150 min of moderate-intensity activity or an equivalent combination per week.²⁰ Strain et al modelled physical inactivity estimates for 2022 using population-based surveys. Conversely, nation-level data for social isolation and high LDL cholesterol was not available; therefore, we used the prevalence reported by Livingston et al² only to estimate the overall PAF and risk factors subcategories. LDL cholesterol prevalence was obtained from the Norwegian Hunt Study²¹ and was defined as the proportion of adults (+45 years) with more than 3 mmol/L of LDL in plasma. By contrast, the prevalence for social isolation was obtained from Livingston et al.² These authors used the pooled prevalence of social isolation reported by Teo et al, which was a systematic review and meta-analysis on 41 studies targeting community dwelling older adults (+60 years).²² Risk factors definitions, categories, measurement methods and limitations are shown in online supplemental table 3.

Additionally, we extracted nation-level data from three Multidimensional Health and Economic Indicators: (1) GNI,²³ (2) HAQI²⁴ and (3) SDI.²⁵ GNI is measured in US dollars and was obtained from the WB 2021.²³ It is defined as the sum value added by all resident producers plus any product taxes, excluding subsidies. The HAQI is a scale from 0 to 100 that measures the healthcare access and quality based on amenable mortality or the mortality that could be avoided in the presence of high-quality healthcare.²⁴ It was obtained from the GBD Study 2019. Finally, the SDI was obtained from the GBD Study 2021. It is a scale from 0 to 100 that measures the overall development of a nation or region based on income per capita, average years of education and total fertility rates.²⁵

Statistical analysis

We grouped risk factors under five categories: vascular and metabolic (diabetes, hypertension, obesity and LDL cholesterol), behavioural (alcohol consumption, smoking and physical inactivity), sensorial impairment (hearing and visual), social determinants (low education), environmental (air pollution) and others (traumatic brain injury, depression and social isolation).

Age-standardised rates were transformed into prevalence proportion using the formula:

$${\displaystyle {\displaystyle Prevalence\left(\mathrm{\%}\right)=\frac{Agestandardizedprevalencerate}{100,000}\times 100}}$$

We calculated communalities to control for share variability among risk factors estimating each factor’s unique risk. Because the relationship among risk factors may vary according to the context of the region or nation, we created four datasets of risk factors prevalence according to nation’s income group. Thus, we estimated four different sets of communalities for each risk factor. Afterwards, we scaled and standardised the prevalence data so that each variable has a mean of 0 and an SD of 1. Then we removed missing data on risk factors prevalence and performed principal component analysis (PCA). From the PCA, we kept components with ≥1 eigenvalues and calculated communalities for each variable by adding the squared scaled of the ‘loadings’ (scaled to their respective eigenvalue) across retained principal components.

Relative risks (RRs) were obtained from Livingston et al for PAF calculation²; therefore, they were considered constant among countries. Overall and specific weighted PAF (wPAFs) were calculated for each risk factor following Livingston et al² using the following formulas:

$${\displaystyle {\displaystyle wPAF=Weight\times \frac{Prevalence\times \left(RR-1\right)}{1+Prevalence\times \left(RR-1\right)}}}$$

PAF weight was calculated according to the risk factor communality² using the formula:

$${\displaystyle {\displaystyle Weight=1-communality}}$$

Overall wPAF and risk factor categories were calculated using the formula:

$$\begin{gathered} {\displaystyle {\displaystyle \begin{array}{l}OverallPAF\end{array}=\begin{array}{l}1-\left[\left(1-wPAF1\right)\left(1-wPAF2\right) \\ \left(1-wPAF3\right)\left(1-wPAF4\right)...\right]\end{array}}} \end{gathered}$$

For global and regional PAFs respectively, we calculated population-weighted means to adjust the contribution of small countries.

$${\displaystyle {\displaystyle RegionalPAF=\frac{\mathrm{\Sigma }\left(PAF\times Population\right)}{\mathrm{\Sigma }\left(Population\right)}}}$$

Tables with prevalence by risk factor and country can be found in online supplemental table 4, while communalities and weights by income group and RRs are displayed in online supplemental table 5–9. Estimates were expressed in mean and 95% uncertainty intervals (UI) which were estimated by bootstrap resampling method using 10 000 resamples. Additionally, we performed Spearman correlations to test whether a relationship between risk factors PAF and the three Multidimensional Health and Economic Indicators exists. All the analysis, data arrangement and maps were performed using R V.4.4.2

Sensitivity analysis

We conducted a sensitivity analysis using the communalities reported in the Lancet standing commission for dementia² to allow a better comparison of our results with previous studies using fixed communalities and enable the identification of constant risk factor PAF distributions across the world.

Patient and public involvement

No patients or members of the public were involved in the planning or conduction of any step of this study.

Results

Globally, 34.03% of dementia cases were potentially preventable, though substantial regional differences were observed (figure 1). Among regions, South Asia had the largest amount of potentially preventable cases of dementia (42.16% (95% UI 39.76% to 42.58%)) while Western Europe had the lowest amount (23.25% (95% UI 22.76% to 23.81)) (online supplemental table 10).

Figure 1. Worldwide distribution of the proportion of potentially preventable dementia cases, expressed as overall population attributable fraction (PAF).

Iraq accumulated the largest amount of potentially preventable cases of dementia (44.27% (95% UI 35.59% to 52.87%)) and North Korea the lowest (20.42% (95% UI 16.44% to 24.36%); online supplemental table 11). Considering economic regions, a U-shape was observed with high- and low-income regions demonstrating lower PAFs (24.51% (95% UI 23.68% to 25.63%) and 24.6% (23.60% to 25.42%), respectively), while upper-middle and lower-middle income regions accumulated larger PAFs (34.48% (32.80% to 38.37%) and 39.13% (35.29% to 41.51%)).

Vision loss was the most important risk factor globally with a PAF of 4.04% (95% UI 2.88% to 5.09%), followed by low education (3.61 (95% UI 2.18% to 5.33%)) and inactivity (2.41% (95% UI 1.71% to 3.07%), figure 2). Nevertheless, the distribution of PAFs due to modifiable risk factors for dementia worldwide varies due to regional and income influences. For instance, the global distribution of PAFs due to low education varies according to income, as LICs accumulated larger proportions of PAFs, contrasting with high income regions (online supplemental figure 1). By consequence, Oceania was the region with the largest amount of PAF due to low education (7.32% (95% UI 2.23% to 8.72%)), followed by South Asia (7.21% (95% UI 6.92% to 8.51%)). Across all nations, Tanzania located in Eastern Sub-Sahara had the largest breach of education with a PAF of 13.93% (95% UI 10.72% to 16.64%), while countries such as the UK and Austria located in Western Europe (region with the least amount of PAF due to low education) had a PAF of 0%.

Figure 2. Ranking of potentially modifiable risk factors of dementia based on their population attributable fraction by region. HIC, high-income country; LIC, low-income country; LMIC, low-income and middle-income country; TBI, traumatic brain injury; UMIC, upper-middle-income country.

When grouped by categories, the vascular and metabolic risk factors for dementia (diabetes mellitus, hypertension, LDL and obesity) accounted for the largest proportion of PAF with a global PAF of 10.98% (95% UI 10.39% to 11.73%, online supplemental figure 2). Among them, HTA accounted for the largest amount of PAF globally (1.60% (95% UI 1.42% to 1.87%)) with Tropical Latin America accumulating the largest amount of PAF regionally (2.77% (95% UI 2.76% to 3.33%)) and Australasia the least amount of PAFs (1.09% (95% UI 1.08% to 1.14%)), virtually tied with Western Europe and high-income Asia Pacific (online supplemental figure 3). Although there was no trend in the ranking among income groups, there was a PAF peak in UMIC (2.04% (95% UI 1.77% to 2.55%)). After HTA,

Diabetes mellitus accounted for 1.27% (95% UI 0.98% to 1.64%) global PAF with Oceania accounting for the largest regional PAFs (3.38% (95% UI 2.21% to 3.66)%) and Eastern Sub-Saharan Africa the lowest (0.42% (95% UI 0.28% to 0.63%), online supplemental figure 4). Not far from diabetes, obesity accounted for 1.20% (95% UI 0.83% to 1.72%) of the global PAF with High income North America as the region with the largest fraction (2.86% (95% UI 2.05% to 2.96%)). However, it is important to address the existence of a regional cluster of obesity in Oceania, with Tokelau accounting for the largest PAF at nation-level (5.80% (95% UI 2.27% to 8.76%)) followed by Samoa (5.29% (95% UI 2.18% to 8.00%)) and American Samoa (5.03% (95% UI 2.04% to 7.54%); online supplemental figure 5). This could be related to the high diabetes PAF values in Oceania. Additionally, we identified other possible regional and interregional influences. For example, despite a potential income influence on obesity as HICs tend to have higher PAFs, high-income Asia Pacific accumulated one of the lowest regional PAFs (0.48% (95% UI 0.43% to 0.83%)), suggesting the existence of sociocultural factors influencing obesity prevalence. This suggests the existence of different risk factors prioritisation within regions in spite of their values worldwide.

Regarding behavioural risk factors, global PAF (5.86% (95% UI 4.62% to 6.84%), online supplemental figure 6), inactivity was ranked third among all risk factors globally (2.41% (95% UI 1.71% to 3.07%)). Tropical Latin America shows the largest amount of regional PAFs (4.26% (95% UI 3.87% to 4.27%)). Nevertheless, Cuba, located in the Caribbean, presented the largest amount at nation-level (6.14% (95% UI 0.2% to 11.23)). Inactivity seems to have an income influence with UMIC and LMIC accumulating larger PAF amounts and HIC and LIC regions lower (online supplemental figure 7).

Furthermore, smoking (ranked fourth globally) accounted for 2.41% (95% UI 1.71% to 3.07%) of global PAF with Oceania presenting the largest regional PAF (4.94% (95% UI 2.37% to 5.42%), followed by Southeast Asia (4.15% (95% UI 2.92% to 5.21%)). LMIC accumulated the largest amount of PAF (3.44% (95% UI 2.18% to 4.41%)) and PAF tended to decrease with higher income. However, LIC presented the lowest amount of PAF (0.77% (95% UI 0.59% to 0.96%)). Despite this, regions composed in their majority by HIC ranked smoking among the five most relevant risk factors, and regions composed by UMIC among the least important. This pattern may be underscoring potential regional and intraregional influences (online supplemental figure 8). For heavy alcohol consumption, Andean Latin America figured as the region with the largest PAF contribution (3.00% (95% UI 2.01% to 3.55%)) and North Africa and the Middle East with the lowest (0.23% (95% UI 0.18% to 0.28%)). Cultural factors influence the distribution of alcohol PAF among regions (eg, low PAF among countries with Muslim tradition, online supplemental figure 9). In addition, it is important to highlight that heavy alcohol consumption was ranked in third position in the HIC group and in first and second position in Central and Western Europe, respectively.

Sensorial loss accounted for 5.97% (95% UI 4.59% to 7.09%) global PAF (online supplemental figure 10). Individually, hearing loss accounted for 1.93% (95% UI 1.65% to 2.19%) globally, with East Asia accounting for the largest amount of regional PAF (2.64% (95% UI 1.96% to 2.68%)) and Western Europe the lowest values (0.94% (95% UI 0.90% to 0.98%)). At national level, China accumulated the largest PAF due to hearing loss with 2.68% (95% UI 1.44% to 3.88%). There was no clear trend according to income groups; however, it was ranked second and third in LIC and UMIC, respectively. Moreover, regional clusters seem to exist with larger PAFs in Tropical Latin America, Eastern and Southern Sub-Saharan Africa and East and Southeast Asia with respect to other regions of the world (online supplemental figure 11). Vision loss was the most important risk factor globally and Southern Sub-Sahara was the region with the largest amount of PAFs accounting for 6.95% (95% UI 4.36% to 8.00%), with South Africa being the country with the highest PAF (8.00% (95% UI 5.11% to 10.65%)). By contrast, high income North America was the region with the lowest amount of PAF due to visionl loss (0.66% (95% UI 0.59% to 0.67%), online supplemental figure 12). Like for low education and inactivity, UMIC and LMIC accumulated the largest amount of PAF due to vision loss by income group.

Furthermore, TBI presented a global PAF of 0.28% (95% UI 0.23–0.31) being Eastern Europe the region with the largest amount of PAF due to TBI (0.34% (95% UI 0.31% to 0.42%), online supplemental figure 13). Ukraine accumulated the largest amount of PAF at nation-level (0.43% (95% UI 0.02% to 1.06%)). Overall, there was a low PAF due to TBI around the world, with even lower PAFs in Central and Eastern Sub-Saharan Africa. In the case of depression, it accounted for a global PAF of 1.20% (95% UI 0.89% to 1.52%) with Southern Sub-Saharan Africa presenting the largest values (1.83 (95% UI 1.52 to 2.44) without a specific regional pattern and a similar income influence as for low-education and visual loss (online supplemental figure 14). Likewise, air pollution with a global PAF of 1.64% (95% UI 1.31% to 1.96%) had a high accumulation of PAF in Tropical Latin America (2.73% (95% UI 2.05% to 2.75%)) at regional level, despite Argentina being the country with the largest PAF at nation level (2.87% (95% UI 0.00% to 7.14%), online supplemental figure 15). Although air pollution was ranked first in HIC, UMIC accumulated more PAF than the other three income groups (2.22% (95% UI 1.95% to 2.57%)).

Correlation analysis

There was a significant negative relationship between overall PAF and the three Multidimensional Health and Economic Indicators (income, SDI and HAQI). While almost all individual risk factors for dementia showed a significant correlation with the three Multidimensional Health and Economic Indicators, diabetes, smoking and inactivity showed no correlation with HAQI and SDI, while diabetes, inactivity and TBI with income (table 1). Large amounts of PAF due to obesity, TBI (but income) and air pollution were correlated with higher values in the three Multidimensional Health and Economic Indicators. By contrast, small amounts of PAF due to sensorial impairments, HTA, low education and depression were correlated with higher values of Multidimensional Health and Economic Indicators. Low smoking PAF only correlated with higher values of income.

Table 1. Correlations between population attributable fraction of 12 potentially modifiable risk factors of dementia and three Multidimensional Health and Economic Indicators (income, SDI and HAQI).

Risk factor	Income		SDI		HAQI
	rho	P value	rho	P value	rho	P value
Vascular and metabolic
Diabetes mellitus	−0.08	0.304	0.06	0.376	−0.12	0.077
Hypertension	−0.31	<0.0001	−0.28	<0.0001	−0.30	<0.0001
Obesity	0.38	<0.0001	0.35	<0.0001	0.27	<0.0001
Behavioural
Alcohol	0.46	<0.0001	0.39	<0.0001	0.37	<0.0001
Smoking	−0.18	0.013	−0.06	0.362	−0.05	0.45
Inactivity	0.04	0.634	0.06	0.431	0.07	0.33
Sensorial impairment
Hearing impairment	−0.50	<0.001	−0.51	<0.0001	−0.51	<0.0001
Visual impairment	−0.51	<0.0001	−0.53	<0.0001	−0.52	<0.0001
Social determinant
Low education	−0.62	<0.0001	−0.65	<0.0001	−0.60	<0.0001
Environmental
Air pollution	0.64	<0.0001	0.62	<0.0001	0.58	<0.0001
Others
Depression	−0.57	<0.0001	−0.53	<0.0001	−0.47	<0.0001
TBI	0.05	0.519	0.16	0.025	0.18	0.009
Overall	−0.39	<0.0001	−0.39	<0.0001	−0.37	<0.0001

Bold values indicate statistical significance.

GNI, gross national income; HAQI, Healthcare Access and Quality index; SDI, sociodemographic index; TBI, traumatic brain injury.

Sensitivity analysis

Fixed communality, based on the Lancet standing commission for dementia, isolated the distribution of dementia risk factors according to prevalence only (online supplemental figure 16). Thus, global overall PAF for dementia risk factors was approximately 5% higher with some differences in regional distributions and income trends (online supplemental figure 17). Low education was the most important contributor to global overall PAF instead of vision loss. Regionally, fixed communalities reduced the relevancy of alcohol in HIC including western and central Europe, and inactivity and smoking globally, and increased relevancy of air pollution (online supplemental figures 18–21). It confirmed a high concentration of diabetes mellitus and obesity in Oceania and showed a relatively high prevalence of HTA in Central Europe, alcohol and inactivity in high-income Asia Pacific, and air pollution in Southern Latin America.

Discussion

Overall, one-third of dementia cases around the world were potentially preventable with substantial regional and economic variations in the PAFs for dementia risk factors. The highest overall regional PAF was observed in South Asia, and the lowest in Western Europe. A clear economic gradient was evident, with higher PAFs in LICs compared with HICs, reflecting disparities in the prevalence and impact of modifiable risk factors. Globally, vision loss, low education and inactivity emerged as the top contributors to dementia risk, but the relative importance of individual risk factors varied across regions. Our findings emphasise the need for region-specific strategies that consider local risk factor profiles and health system capacities to mitigate the growing dementia burden.

A key contribution of this study is the identification of regional patterns in PAFs, which demonstrate that the modifiable risk factors for dementia are not uniformly distributed. For instance, vascular and metabolic factors, such as diabetes and obesity, were leading contributors in HICs, while sensorial impairments and social determinants, such as low education, predominated in LMICs. These findings align with prior studies from Latin America, Africa and Asia, which have also reported that LMICs face a disproportionate burden of modifiable risk factors due to limited healthcare access and lower educational attainment.5,811

Despite that, it is important to acknowledge that, compared with previous studies on single nations, our results differ in risk factor distributions and PAF values, possibly due to the source of data, introduction of communalities according to income group and methods to estimate communalities. For instance, in most of those cases, data came from cross-sectional national representative surveys^{6 9 26 27} which contain greater granularity and allow for capturing more precisely the correlation structure among dementia risk factors according to the setting. By contrast, our data came from global health datasets, which may contain limitations in their estimation methods due to missing data in some countries or assumptions in their models that may overestimate or underestimate real values. However, they allow direct comparisons across countries and regions, giving support in the identification of regional trends and construction of policy development, especially in less advantaged nations that lack local reliable data. In addition, due to the nature of our data, communalities calculations differed from those performed by the Lancet standing commission for dementia,² which used individual level data from the Trondelag Health Study (HUNT) —a population-based study in Norway—reflecting shared variance of risk factors across individuals. Nevertheless, while the correlation structure of risk factors is more precise, an HIC such as Norway may not possess the same correlation structure as an LIC. Hence the importance of our approach.

Additionally, the influence of race and ethnicity on PAFs has been noted in studies from the USA, where Black and Hispanic populations exhibit higher PAFs compared with White and Asian populations, driven by disparities in risk factor prevalence and healthcare access.⁴ Likewise, research in New Zealand found substantial differences in PAFs across ethnic groups, with Māori and Pacific peoples experiencing a higher burden of modifiable risk factors compared with European and Asian populations.¹⁰ Nevertheless, due to the characteristics of our data, it was not possible to assess potential influences either by ethnicity, race or sex. Therefore, future studies should aim to disentangle further possible different distribution in these populations.

The findings of this study provide critical insights for policymakers and healthcare providers. Reducing air pollution, improving access to education and addressing vascular, metabolic and sensorial-related risk factors should be prioritised, particularly in UMICs and LMICs where these factors contribute significantly to dementia risk. In fact, considering the modifiable plausibility of dementia risk factors, its influence by socioeconomic factors and the regional distribution of dementia risk factors, policy-makers and public health professionals should switch from at-risk-individual-based to whole-population-based approaches to achieve a substantial impact in dementia prevention.²⁸ This involves the introduction of policies that act on the structure rather than on the individual, favouring the development of pro-brain health environments. However, these new public health strategies should account for regional and subregional disparities. Consequently, the development of region-specific guidelines for dementia prevention, informed by local data and health system capacities, and the adoption of methods able to incorporate and model complex systems such as system thinking approaches could facilitate the understanding of the problems embedded in the population and enhance the development and success of population-tailored strategies.²⁸

One of the most striking findings of our study is the strong correlation between overall PAFs and multidimensional health and economic indicators. Regions with lower SDI and HAQI values exhibited higher overall PAFs, highlighting the role of socioeconomic development in shaping the distribution and impact of dementia risk factors. Interestingly, sensory impairment (hearing and vision loss) demonstrated inverse correlations with these indicators in high-income settings, likely reflecting better prevention and management of these conditions through access to healthcare services.²⁹ However, diabetes, inactivity and smoking showed no significant associations with these indicators, which may be attributed to differences in regional management strategies or cultural influences. Future research should explore these dynamics further to better understand the role of healthcare systems and regional practices in modulating these risks.

In addition, this is the first study to comprehensively estimate national, regional and global PAFs for dementia using the most updated prevalence data from diverse sources such as the GBD Study, WB and WHO databases. By applying consistent methodologies, this study provides a detailed and actionable understanding of the global distribution of modifiable dementia risk factors. Furthermore, allowing communalities to vary according to countries’ income groups introduced an extra layer of complexity in the distribution of dementia risk factors worldwide, approaching a more realistic scenario. Sensitivity analysis with a unique communality identified robust distributions in risk factors and those more prone to vary according to the setting. Besides, the introduction of multidimensional health and economic indicators enables a nuanced analysis of how socioeconomic and health system factors interact with dementia risk.

Despite that, this study has some limitations. First, relative risks were assumed to be constant across regions with a potential risk of low transportability. Likewise, even though we allowed some variations on communalities according to nations’ income group, risk factors’ correlation structure may vary even more among and within countries of the same income group due to cultural or geographic factors. Therefore, full local variations in the strength of associations between risk factors and dementia may not be fully captured. Second, missing prevalence data for certain countries necessitated the use of proxy estimates from the literature, which may introduce inaccuracies. For instance, social isolation and high LDL cholesterol lacked nation-level prevalence data, potentially underestimating the overall PAF. In addition, some risk factors, such as air pollution, were measured using proxies (eg, percentage of the urban population exposed to pollution), which may not reflect true exposure levels. Moreover, our prevalence data came from multiple sources with heterogeneous methods. Age ranges were not perfectly aligned across risk factors, and some sources reported crude rather than age-standardised estimates. Finally, data versions differed across sources, creating temporal mismatches upward or downward in our estimates. However, in spite of these limitations—mostly due to the difficulty of assembling global health datasets—our study, like the Lancet commission for dementia prevention, aims to serve as an initial guide to dementia prevention by mapping the global distribution of dementia risk factors. Future studies should incorporate region-specific relative risks and prevalence estimates to enhance the precision of PAF calculations.

Conclusions and public health implications

This study provides a comprehensive global analysis of the modifiable risk factors for dementia, highlighting significant regional and economic disparities. Our findings underscore the need for targeted whole-population public health interventions to address key risk factors such as vision loss, low education, inactivity, smoking, air pollution and hypertension, particularly in LMICs where the burden is highest. The observed correlations between PAFs and socioeconomic indicators and allowing risk factors contribution to vary according to their income context emphasise the critical role of addressing structural inequities in healthcare and education to reduce dementia prevalence. Future research should focus on refining region-specific risk factor estimates, integrating emerging risk factors and exploring longitudinal relationships between socioeconomic development and dementia risk. By addressing these gaps, global efforts can better prioritise prevention strategies, ultimately reducing the burden of dementia worldwide.

The funder of the study had no role in the study design, data extraction, analysis, interpretation or writing and submission of the report. We plan to spread our results to the general public, healthcare professionals, scientists and policy-makers through our monthly newsletter, social media platforms and conventional media.

Data availability statement

Data are available on reasonable request.