The potential for dementia prevention in Japan: a population attributable fraction calculation for 14 modifiable risk factors and estimates of the impact of risk factor reductions

Summary

Background

As dementia prevalence increases globally, preventive strategies targeting modifiable risk factors have become increasingly important. In Japan, with its super-ageing society, dementia is the leading cause of increased disability-adjusted life years among older adults. This study quantified the contribution of 14 potentially modifiable risk factors for dementia in older adults using Japan-specific prevalence data.

Methods

We calculated population attributable fractions (PAFs) and potential impact fractions (PIFs) using recent publicly available prevalence data from national surveys and cohort studies in Japan, and relative risks and communality weights from the 2024 Lancet Commission report on dementia. We then modelled how 10% and 20% reductions in each risk factor would affect national dementia prevalence.

Findings

The weighted combined PAF for all 14 risk factors was 38.9%, indicating that nearly 4 in 10 dementia cases in Japan might be preventable. Hearing loss (6.7%), physical inactivity (6.0%), and high LDL cholesterol (4.5%) were the largest contributors. Reducing all risk factors by 10% could prevent ∼208,000 dementia cases; reducing them by 20% could prevent ∼407,000 cases.

Interpretation

Dementia preventive efforts in Japan should prioritise hearing care, physical activity, and metabolic health. Japan-specific data confirmed that hearing loss is a leading contributor to dementia, underscoring the urgency to increase public awareness and access to hearing interventions.

Funding

The Royal Danish Embassy in Japan, Danish Ministry of Foreign Affairs, Danish Ministry of Health, and Japan Agency for Medical Research and Development funded this study.

Research in context.

Evidence before this study

The 2017, 2020, and 2024 versions of The Lancet Commission report on dementia proposed that 35–45% of dementia cases worldwide could be reduced by targeting potentially modifiable key risk factors. These estimates were based on pooled global data and expert consensus. Country-specific prevalence or exposure data were not considered. Although the pooled data approach produced valuable insights about dementia risk at a global level, it did not account for possible country-specific differences in demographic structure, health systems, and risk factor prevalence. In Japan, the population is ageing rapidly. With a comparatively high dementia prevalence, Japan lacks comprehensive, nationally representative estimates of the extent modifiable risk factors contribute to dementia incidence. A local, community-based Japanese study from 2019 estimated the contribution of seven modifiable risk factors to incident dementia, but no nationally representative Japanese study has been published. Without country-specific estimates, it is difficult for policymakers in Japan to prioritise prevention strategies or effectively allocate resources. Thus, there is a critical need for dementia risk factor data that reflect the true epidemiological situation in Japan, and more generally, in other countries.

Added value of this study

The present study is the first to quantify the contribution of 14 modifiable dementia risk factors in Japan using current, population-specific prevalence data and standardised population attributable fractions (PAF) methodology. By integrating Japan-specific epidemiological data with relative risk estimates and communality adjustments from the 2024 Lancet Commission report on dementia, we provide robust national estimates of the proportion of potentially preventable dementia cases. We also model the impact of proportional (10–20%) reductions in these risk factors on dementia prevalence, highlighting the top contributors in the Japanese context, notably hearing loss and physical inactivity.

Implications of all the available evidence

Our findings bolster global evidence that substantial dementia prevention may be feasible through targeted public health interventions. In Japan, hearing care, promotion of physical activity, and management of metabolic risk factors should be prioritised by implementing national dementia prevention strategies. Given the enactment of Japan's Basic Act on Dementia in 2024, our present data offer timely and actionable insights to guide implementation. Future research should evaluate intervention effectiveness and monitor progress in reducing dementia risk factors and the national dementia burden.

Introduction

The prevalence of dementia is increasing worldwide. According to the Global Burden of Disease study (GBD) 2021,¹ Alzheimer's disease and other dementias accounted for 1.26% of global disability-adjusted life years (DALYs) across all ages, and 4.76% among those aged 70 years and older. Consequently, the costs associated with dementia have been estimated at 1.3 trillion USD in 2019.² These costs are projected to rise to 1.7 trillion USD by 2030, and when the increasing expenses of caregiving are taken into account, they are expected to reach as high as 2.8 trillion USD.³

Japan has the highest average life expectancy in the world and has become a super-ageing society. In 2010, more than 21% of its population was 65 years or older. This proportion rose to 29.3% in 2024 and is projected to exceed one-third of the population in Japan by 2045.⁴ As ageing is the greatest risk factor for developing dementia,⁵ Japan is vulnerable. In Japan, Alzheimer's disease and other dementias represent the second leading cause of DALYs across all ages and is the leading cause for those over 70. This underscores that dementia is a critical social and public health issue.¹ Behera et al. reported that the disease burden of Alzheimer's disease and other dementias in Japan have increased steadily from 1990 to 2021 and is projected to continue to rise over the next decades.⁶ According to the Ministry of Health, Labour and Welfare, the prevalence rate of dementia in Japan is approx. 12.3% among individuals aged 65 years or older, and the prevalence rate of mild cognitive impairment (MCI) is approx. 15.5%.⁷ Extrapolating this to Japan's overall population (as of in 2022) produced the following estimates: 4.43 million people (95% CI: 4.18–4.68 million) with dementia and 5.58 million (95% CI: 3.82–7.35 million) with MCI. By 2050, these numbers are expected to rise to 5.87 million (15.1%) and 6.31 million (16.2%), respectively.⁷

Although pharmaceutical developments, such as monoclonal antibodies aimed at removing protofibrillar forms of amyloid-beta from the brain,^8,9 show promise for reducing clinical dementia, their limited effectiveness and high costs have resulted in inconsistent approval and limited adoption in clinical practice.¹⁰ At present, there is a growing consensus among medical professionals and researchers that reducing dementia risk and delaying its onset through prevention is considered the most effective strategy.¹¹

In 2014, Norton et al. identified seven potentially modifiable risk factors for developing dementia.¹² Building on this, Livingston and colleagues published updates as part of The Lancet Commission on dementia in 2017,¹³ 2020,¹⁴ and 2024.¹⁵ These reports and their recommendations have influenced policy planning worldwide, but as they are based on global data, they do not necessarily represent the situation in Japan.

In this study, we analysed the contribution of 14 potentially modifiable risk factors for dementia, as identified in the 2024 Lancet Commission report: less education, hearing loss, high LDL cholesterol, depression, traumatic brain injury (TBI), physical inactivity, smoking, diabetes, hypertension, obesity, excessive alcohol consumption, social isolation, air pollution exposure, and untreated visual loss¹⁵ using risk factor prevalence data for Japan.

Methods

The 14 risk factors were selected by the Lancet Commission based on a comprehensive literature review focussing on high quality, consistent dose-respondent, validly measured evidence. Only biologically plausible risk factors associated with dementia and backed up by evidence were included. The commission considered several other potentially modifiable risk factors (e.g., insufficient sleep, unhealthy diet, infections, other mental health conditions), but they concluded that those possible risk factors lacked sufficient evidence for them to be included in the 2024 Lancet Commission report.¹⁵

Data sources and risk factor definitions

In the preparatory phase of this study, we considered multiple publicly available data sources of risk factor prevalence in Japan based on their representativeness, sample size, recency, and transparency of documentation. Whenever possible, we chose data sources that used risk factor definitions consistent with those used in the 2024 Lancet report.¹⁵

Six of the 14 risk factor prevalence estimates for dementia were based on data from the 2019 National Health and Nutrition Survey (NHNS) Japan¹⁶: high LDL cholesterol, physical inactivity, diabetes, hypertension, obesity, and excessive alcohol consumption. The estimated prevalence of smoking was based on data from the 2022 NHNS.¹⁷ The NHNS has been conducted annually by the Ministry of Health, Labour, and Welfare since 1946 to document the health status, nutritional intake, and lifestyle habits of people in Japan, and to obtain basic data to aid comprehensive health promotion in Japan.¹⁸ The survey uses random sampling of households across Japan to ensure coverage of different regions and demographics. From its inception, the NHNS has played a major role in gathering information about the Japanese people and contribute to national policy using evidence-based results. The 2019 NHNS consisted of data from the following: 5074 participants who underwent a standard physical examination, 2431 of which also underwent a standard blood panel; 5709 participants who completed a questionnaire on lifestyle habits; and 5865 participants who completed a dietary survey. For the 2019 NHNS, data were presented in age epochs of 10 years. For our analyses, we restricted the age range to the epochs of 40–69 years, which is roughly equivalent to the definition of midlife (45–64 years) published in the 2024 Lancet Commission report.¹⁵

Risk factor prevalence

Definitions of the 14 risk factors are presented in Table 1.

Table 1.

Definitions of risk factors for dementia.

Risk factor	Definition	Source
Less education	Below upper secondary school	Education at a Glance 2024
Hearing loss	Hearing level ≥26 dB (WHO Grade 1–4)	Wasano et al., 2022
High LDL cholesterol	LDL cholesterol ≥130 mg/dL	2019 NHNS
Depression	Any mood disorder according to the WHO Composite Diagnostic Interview version 3.0	World Mental Health Japan 2nd Survey
Traumatic brain injury	Diagnosis of head injury in neurotrauma data bank	Kameyama et al., 2008
Physical inactivity	Self-report: not exercising more than 30 min at least twice a week for more than a year	2019 NHNS
Smoking	Self-reported daily smoking	2022 NHNS
Diabetes	HbA1c ≥ 6.5% and/or diabetes medication	2019 NHNS
Hypertension	Systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg, or antihypertensive medication	2019 NHNS
Obesity	BMI ≥30 measured at physical examination (height and weight)	2019 NHNS
Excessive alcohol consumption	Self-report: >40 g alcohol daily (men); >20 g alcohol daily (women)	2019 NHNS
Social isolation	Self-reported living alone	National Survey on the Actual Conditions of Solitude and Isolation
Air pollution	Exposure to PM2.5 concentration ≥15 μg/m3 according to satellite-derived measurement	Atmospheric Composition Analysis Group 2022
Untreated visual loss	US criteria for visual impairment (best-corrected visual acuity of the better-seeing eye <20/40 but >20/200), and blindness (≤20/200 in the better-seeing eye)	Yamada et al., 2010

NHNS = National Health and Nutrition Survey, WHO = World Health Organization, LDL = low density lipoprotein, HbA1c = haemoglobin A1c, BMI = body mass index, PM_2.5 = particulate matter <2.5 μm.

The risk factors can be grouped according to a life course model¹⁵ comprising three stages: (i) early life (less formal education); (ii) midlife (hearing loss, high LDL cholesterol, depression, traumatic brain injury [TBI], physical inactivity, smoking, diabetes, hypertension, obesity, excessive alcohol consumption); and (iii) late life (social isolation, air pollution exposure, and untreated visual loss). However, the impact of each risk factor is not necessarily limited to a specific life stage or age range. Due to the nature of our data sources, the estimates on the prevalence of TBI and exposure to air pollution could not be restricted to a specific age range in our study.

Education

Population-based results for attained educational level of 25- to 64-year-old Japanese citizens were derived from the Education at a Glance 2024 report published by the Organisation for Economic Co-operation and Development (OECD).¹⁹ According to OECD, the report is the definitive guide to the state of education around the world. In Japan, compulsory education consists of six years of elementary school plus three years of lower secondary school. Consequently, we defined ‘less formal education’ as below the level of upper secondary school, which is equivalent to ≤9 years of formal education.

Hearing loss

Data on the prevalence of hearing loss were extracted from a retrospective cross-sectional study that reported audiometric data on 23,860 participants in Japan in the age range of 10–99 years.²⁰ The participants had sought medical attention for various reasons at the Department of Otolaryngology at the National Hospital Organization Tokyo Medical Center between April 2000 and March 2020. We adopted the World Health Organization (WHO) grade 1–4 (hearing level ≥26 dB) definition for hearing loss.²¹ To make our prevalence estimate comparable to the estimate presented in the 2024 Lancet Commission report,¹⁵ we restricted the age range to ≥55 years. Thus, we used data from 14,567 participants.

High LDL cholesterol

Data on the prevalence of high LDL cholesterol were drawn from the 2019 NHNS.¹⁶ High LDL cholesterol was defined as LDL levels ≥130 mg/dl, as reported in the 2024 Lancet Commission report.¹⁵ We used data from participants in the age range of 40–69 years (n = 1255).

Depression

Data on the prevalence of depression were derived from the World Mental Health Japan 2nd Survey. This is a representative household survey of 20- to 75-year-old residents in Japan; it was conducted from 2013 to 2015 (n = 2450).²² The participants were selected from a multi-stage, clustered, area probability sample of households based on a government-sponsored resident registry. The data were collected from the Kanto region in 2013, in the East and Chubu (central) regions of Japan in 2014, and in the West region of Japan in 2015. The Kanto region comprises Tokyo (the capital) and surrounding areas plus seven prefectures. The data we used from the survey came from the category ‘any mood disorder’. According to the WHO Composite Diagnostic Interview version 3.0, ‘any mood disorder’ includes major depression, bipolar I and II disorders, and dysthymia.²³

Traumatic brain injury

As we were unable find valid data on the prevalence of TBI in Japan, we used data on the incidence of TBI in Miyagi Prefecture, a prefecture north of Tokyo, Japan. Data were from patients admitted to one of the 17 neurosurgical institutes in the prefecture and registered in the Miyagi Neurotrauma Data Bank from 1995 to 2006 (n = 10,373).²⁴ TBI was defined as a hospital diagnosis of either diffuse brain injury, contusion, acute subdural haematoma, or acute epidural haematoma. The estimated cumulative risk of TBI during the average life expectancy (81.08 years) in the year 2000 was used as a crude estimate of the TBI prevalence rate. The age range was the full population age range.

Physical inactivity

Data on the prevalence of self-reported physical inactivity were drawn from the 2019 NHNS.¹⁶ Physical inactivity was defined as failing to meet the survey criterion of exercising at least 30 min per session, at least twice a week, for at least one year. We used exercise prevalence data from participants in the age range of 40–69 years (n = 1437).

Smoking

Data on the prevalence of smoking were drawn from the 2022 NHNS.¹⁷ We adopted the WHO definition of daily smoking (smoke any tobacco product at least once a day).²⁵ In the 2022 NHNS, data are in age epochs of 5 years, but we restricted the age range to 45–64 years (n = 31,814).

Diabetes

Data on the prevalence of diabetes were extracted from the 2019 NHNS.¹⁶ Diabetes was defined as haemoglobin A1c ≥ 6.5% and/or prescribed use of anti-diabetic medication. We used data from participants in the age range of 40–69 years (n = 1253).

Hypertension

Data on the prevalence of hypertension were drawn from the 2019 NHNS.¹⁶ Hypertension was defined as systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg, or prescribed use of antihypertensive medication. We used data from participants in the age range of 40–69 years (n = 1322).

Obesity

Data on the prevalence of obesity were calculated from the 2019 NHNS.¹⁶ Obesity was defined as a body mass index of ≥30 according to participants’ body weight and height recorded at physical examination. We used data from participants in the age range of 40–69 years (n = 2261).

Excessive alcohol consumption

Data on the prevalence of alcohol consumption were drawn from the 2019 NHNS.¹⁶ Excessive alcohol consumption was defined as follows: for men, consuming ≥40 g pure alcohol daily (≥280 g = 35 alcohol units per week); for women, consuming ≥20 g pure alcohol daily (≥140 g = 17.5 alcohol units per week). For example, this is equivalent to the following: for men, consuming ≥360 mL of sake every day, ≥360 mL 5–6 times a week, ≥540 mL 3–4 times a week, ≥900 mL once or twice a week, or ≥900 mL 1–3 times a month; for women, consuming ≥180 mL of sake every day, ≥180 mL 5–6 times a week, ≥180 mL 3–4 times a week, ≥540 mL once or twice a week, or ≥900 mL 1–3 times a month. We used data from participants in the age range of 40–69 years (n = 2960).

Social isolation

Data on the prevalence of social isolation were derived from the National Survey on the Actual Conditions of Loneliness and Isolation 2023 conducted by the Cabinet Office, Government of Japan (n = 11,141).²⁶ A total of 20,000 individuals aged 16 or more nationwide were selected using a random sampling method based on residence. Social isolation was defined as living alone, according to an individual's self-reported living status. In the 2023 survey, data were in age epochs of 10 years. We restricted the age range to ≥60 years (n = 5155).

Air pollution

We used data from the 2022 Atmospheric Composition Analysis Group (ACAG) to estimate the prevalence of exposure to air pollution. These are satellite-derived data on fine particulate matter for various regions, including Japan.²⁷ The ACAG website (sites.wustl.edu/acag.) defines values as follows: ‘Regional instruments and their respective retrievals were combined with simulation based upon their relative uncertainties as determined using ground-based sun photometer observations to produce geophysical estimates that explain most of the variance in ground-based particulate matter with a diameter of ≤2.5 μm (PM2.5) measurements’. According to WHO air quality guidelines, the average daily exposure to PM2.5 should not exceed 15 μg/m³ for more than 3–4 days per year.²⁸ Consequently, we defined exposure to air pollution as exposure to concentrations of ≥15 μg/m³ of PM 2.5. The age range was the full population age range.

Untreated visual loss

Data on the prevalence of untreated visual loss were derived from a comprehensive review synthesising results from 13 epidemiological studies of visual impairment in Japan, and from three official Japanese databases managed by the Ministry of Health, Labour and Welfare, and by the Ministry of Internal Affairs and Communications.²⁹ The epidemiological studies were conducted between 1984 and 2004, with the majority being conducted in the 1990s. These studies covered most of the major regions of Japan. Untreated visual loss was defined as visual impairment (best-corrected visual acuity of the better-seeing eye <20/40 but better than 20/200), and blindness (≤20/200 in the better-seeing eye). Since 9 of the 13 studies were based on participants aged ≥40 years, and one study was based on participants aged ≥50 years, we restricted the age range to ≥40 years (n = 23,071).

Statistical analysis

Relative risk and communality

The 2024 Lancet Commission report presents relative risk (RR) estimates for each of the 14 dementia risk factors. These are based on a comprehensive literature review and meta-analyses of observational studies.¹⁵ We used these RRs in our analysis, as we reasoned that they represent the best estimates available. The population attributable fraction (PAF) of a risk factor is an estimate of the proportion of dementia cases that may be attributed to this factor. Individual PAFs can be adjusted to account for shared variance (communality) and combined to obtain an overall estimate of prevention potential. As our data were derived from several different sources, we could not calculate communalities for the present study. As an alternative, we used the communality values of the 2024 Lancet Commission report, which are based on data from 37,000 participants in a Norwegian population-based health study, the Trøndelag Health (HUNT) Study.³⁰

Population attributable fractions

The unweighted PAF for each risk factor was calculated as in³¹:

$${\text{PAF}}_{\mathrm{i}}={\mathrm{P}}_{\mathrm{i}}\times \left({\text{RR}}_{\mathrm{i}}-1\right)/\left[1+{\mathrm{P}}_{\mathrm{i}}\times \left({\text{RR}}_{\mathrm{i}}-1\right)\right]$$

where P_i is the estimated prevalence rate of each risk factor, and RR_i is the relative risk of dementia associated with exposure to each risk factor.

We calculated weights for each risk factor using the following formula:

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

The combined weighted PAF for all 14 risk factors was calculated using the formula reported by Norton et al.¹²:

$$\text{Combined}\text{weighted}\text{PAF}=1-\left[\left(1-\text{Weight}1\times \text{PAF}1\right)\times \left(1-\text{Weight}2\times \text{PAF}2\right)\times \left(1-\text{Weight}3\times \text{PAF}3\right)\cdots \right]$$

Separately, we also calculated the weighted PAF for each risk factor using the formula from the 2017 Lancet Commission report¹³:

$$\text{Individual}\text{weighted}\text{PAF}=\left[\text{Unweighted}\text{PAF}/\sum \left(\text{unweighted}\text{PAF}\right)\right]\times \text{combined}\text{weighted}\text{PAF}$$

The individual weighted PAFs indicate the maximum proportions of dementia cases that could hypothetically be prevented by eliminating each of the individual risk factors. The combined weighted PAF indicates the proportion of dementia cases that could be prevented by hypothetically eliminating all risk factors.

Potential impact fractions

The potential impact fraction (PIF) is an estimate of the proportion of dementia cases that may be prevented by a partial (e.g., 10% or 20%) reduction of the risk factor in question. Individual PIFs can be combined to obtain an overall estimate of the impact of proportional reductions of several risk factors. The PIF for each risk factor was calculated as in³²:

$$\left({\text{PIF}}_{\mathrm{i}}\right)=\left(\left[{\mathrm{P}}_{\mathrm{i}}-{\mathrm{P}}_{\mathrm{i}}^{\prime }\right]\times \left[{\text{RR}}_{\mathrm{i}}-1\right]\right)/\left({\mathrm{P}}_{\mathrm{i}}\times \left[{\text{RR}}_{\mathrm{i}}-1\right]+1\right)$$

where P'_i is the counterfactual prevalence rate of a risk factor following a proportional (e.g., 10%) reduction.

The combined weighted PIF for all 14 factors was calculated using the formula reported by Lee et al.³³ They took communalities (the 14 weights from the PAF calculation) of the risk factors into account:

$$\text{Combined}\text{weighted}\text{PIF}=1-\left[\left(1-\text{Weight}1\times \text{PIF}1\right)\times \left(1-\text{Weight}2\times \text{PIF}2\right)\times \left(1-\text{Weight}3\times \text{PIF}3\right)\cdots \right]$$

We estimated the number of dementia cases that could be prevented in Japan following a 10% or 20% reduction in each risk factor. We did this by multiplying each PIF by the estimated prevalence of dementia in Japan in 2022, which was approximately 4,432,000 cases.⁷ Confidence intervals (CI) for the unweighted PIFs and for the potential number of dementia cases prevented for each risk factor were calculated by first entering the value of the lower limit of the CI of the RR for each risk factor into the equations and then repeating the calculation by entering the higher limit of the CI of the RR for each risk factor.

Sensitivity analysis

For the sensitivity analysis, we used Japan-specific hazard ratios (HRs) for dementia for seven of the 14 risk factors. HRs for six risk factors (less education, physical inactivity, smoking, diabetes, hypertension, obesity) were extracted from the Ohsaki Cohort 2006 Study,³⁴ and an HR for dementia associated with hearing loss was derived from a Japanese cohort study.³⁵ The Ohsaki Cohort Study was conducted in the city of Ohsaki between 2006 and 2012. The participants were 65 years or older at baseline; 8563 people completed the 5.7 years of follow-up.³⁴ The study on hearing impairment was a community-based prospective cohort study from the Japan Gerontological Evaluation Study, which compared normal hearing with hearing loss and dementia. The baseline survey was conducted from 2010 to 2012 in 13 municipalities. The participants were 65 years or older at baseline; 53,549 people completed the six years of follow-up.³⁵ Prior to the calculation of PAFs and PIFs, we converted HRs to RR estimates using the following formula reported by Livingston et al.^15,36:

$$\text{RR}=\left(1-{\mathrm{e}}^{\text{HR}\times \text{in}\left(1-\mathrm{r}\right)}\right)/\mathrm{r}$$

Where r is the rate of dementia for the reference (‘non-exposed’) group.

Calculations were performed using Microsoft Excel 2016®.

Role of the funding source

The funding source had no role in the study design, data analysis, interpretation, or writing of this report.

Ethical considerations

This study was conducted using a database containing only publicly available anonymised information. Thus, the present study was not subject to ethical review, because the Ethical Guidelines for Medical and Biological Research Involving Human Subjects stipulate that ethical review is not required for these kinds of data.³⁷

Results

Population attributable fraction (PAF)

The combined PAF of all 14 dementia risk factors, weighted for communality, was 38.9%. This result indicates that nearly 4 out of 10 prevalent dementia cases in Japan could be preventable if, hypothetically, all risk factors were eliminated (Table 2; Fig. 1).

Table 2.

Prevalence rates and PAFs for 14 potentially modifiable dementia risk factors for the Japanese population.

	RR for dementia (95% CI)	Risk factor prevalence	Communality, %	Unweighted PAF	Weighted PAF
Early life
Less education	1.6 (1.3–2.0)	7.0%	0.608	4.0%	1.5%
Midlife
Hearing loss	1.4 (1.0–1.9)	57.8%	0.609	17.6%	6.7%
High LDL cholesterol	1.3 (1.3–1.4)	43.2%	0.469	12.1%	4.5%
Depression	2.2 (1.7–3.0)	6.0%	0.452	6.9%	2.6%
Traumatic brain injury	1.7 (1.4–1.9)	3.1%	0.423	2.2%	0.8%
Physical inactivity	1.2 (1.2–1.3)	75.9%	0.567	16.0%	6.0%
Smoking	1.3 (1.2–1.4)	20.5%	0.650	5.8%	2.2%
Diabetes	1.7 (1.6–1.8)	11.7%	0.493	7.9%	3.0%
Hypertension	1.2 (1.1–1.4)	41.8%	0.595	7.7%	2.9%
Obesity	1.3 (1.0–1.7)	6.0%	0.622	1.8%	0.7%
Excessive alcohol consumption	1.2 (1.0–1.5)	16.6%	0.772	3.4%	1.3%
Late life
Social isolation	1.6 (1.3–1.8)	17.9%	0.408	9.3%	3.5%
Air pollution	1.1 (1.1–1.1)	77.5%	0.341	6.5%	2.5%
Untreated visual loss	1.5 (1.4–1.6)	3.7%	0.553	1.7%	0.6%
Overall PAF for all risk factors					38.9%

RR = relative risk, CI = confidence interval, PAF = population attributable fraction.

Fig. 1.

Population attributable fraction (PAF) of potentially modifiable risk factors for dementia in Japan.

The three risk factors with the highest weighted PAF values (all >4%) were hearing loss (prevalence rate: 57.8%; weighted PAF: 6.7%); physical inactivity (prevalence rate: 75.9%; weighted PAF: 6.0%); and high LDL cholesterol (prevalence rate: 43.2%; weighted PAF: 4.5%). This indicates that 17.3% of the total prevention potential was attributed to these three risk factors.

Six risk factors (social isolation, diabetes, hypertension, depression, air pollution, and smoking) had intermediate weighted PAF values in the range of 2.2–3.5%. The sum of weighted PAFs for these risk factors was 16.7%. This indicates that a substantial proportion of the total prevention potential was attributed to these factors. Five risk factors (less formal education, excessive alcohol consumption, TBI, obesity, and untreated visual loss) had the lowest weighted PAF values (≤1.5%). The sum of weighted PAF values for these five risk factors was 4.9%, indicating that a relatively small proportion of the potentially preventable dementia cases was attributed to these factors.

Potential impact fraction (PIF)

A 10% reduction across all 14 risk factors was associated with a 4.7% (95% CI: 2.7 to 6.6) lower prevalence of dementia with time. This reduction is equivalent to 208,185 (95% CI: 119,659 to 292,092) fewer cases of dementia (Table 3).

Table 3.

Potential impact fractions (PIFs) and estimated reductions in Japanese dementia prevalence with proportional risk factor reductions.

Risk factor	Unadjusted PIF percentage at 10% reduction (95% CI)	Fewer cases at 10% reduction, No. (95% CI)	Unadjusted PIF percentage at 20% reduction (95% CI)	Fewer cases at 20% reduction, No. (95% CI)
Early Life
Less education	0.4 (0.2–0.7)	17,578 (7922 to 29,265)	0.8 (0.4–1.3)	35,156 (15,844 to 58,530)
Midlife
Hearing loss	1.8 (0.0–3.3)	78,039 (256–148,227)	3.5 (0.0–6.7)	156,078 (512–296,455)
High LDL cholesterol	1.2 (1.0–1.4)	53,816 (43,191 to 62,484)	2.4 (1.9–2.8)	107,632 (86,382 to 124,968)
Depression	0.7 (0.4–1.1)	30,729 (17,506 to 46,781)	1.4 (0.8–2.1)	61,458 (35,012 to 93,561)
Traumatic brain injury	0.2 (0.1–0.3)	9646 (5411 to 11,992)	0.4 (0.2–0.5)	19,291 (10,822 to 23,984)
Physical inactivity	1.6 (1.3–1.8)	70,700 (55,871 to 79,971)	3.2 (2.5–3.6)	141,401 (111,743 to 159,943)
Smoking	0.6 (0.4–0.8)	25,642 (15,750 to 37,382)	1.2 (0.7–1.7)	51,284 (31,500 to 74,764)
Diabetes	0.8 (0.7–0.9)	34,957 (31,398 to 38,889)	1.6 (1.4–1.8)	69,915 (62,796 to 77,778)
Hypertension	0.8 (0.2–1.3)	34,155 (10,831 to 56,506)	1.5 (0.5–2.5)	68,310 (21,662 to 113,011)
Obesity	0.2 (0.0–0.4)	8053 (529–17,290)	0.4 (0.0–0.8)	16,107 (1057 to 34,580)
Excessive alcohol consumption	0.3 (0.0–0.7)	14,947 (735–31,477)	0.7 (0.0–1.4)	29,894 (1471 to 62,954)
Late life
Social isolation	0.9 (0.5–1.3)	41,075 (24,037 to 58,585)	1.9 (1.1–2.6)	82,149 (48,073 to 117,169)
Air pollution	0.7 (0.5–0.8)	28,898 (22,806 to 34,815)	1.3 (1.0–1.6)	57,795 (45,613 to 69,630)
Untreated visual loss	0.2 (0.1–0.2)	7535 (5795 to 9574)	0.3 (0.3–0.4)	15,071 (11,589 to 19,149)
Combined factors	4.7 (2.7 to 6.6)	208,185 (119,659 to 292,092)	9.2 (5.3 to 12.8)	407,547 (236,510 to 566,821)

PIF = potential impact fraction, CI = confidence interval.

A 20% reduction across all risk factors was associated with 9.2% (95% CI: 5.3 to 12.8) lower prevalence of dementia with time. This level of reduction is equivalent to 407,547 (95% CI: 236,510 to 566,821) fewer cases of dementia. The relative impact of reducing the prevalence of individual risk factors followed the same rank ordering as the PAF values. For instance, a 10% reduction in the prevalence of hearing loss would have a relatively large impact on dementia prevalence, equivalent to 78,039 cases (not weighted for communality). On the other hand, a 10% reduction in the prevalence of untreated visual loss would have a relatively minor impact, equivalent to 7535 cases.

Sensitivity analysis

To examine the effect of using Japan-specific risk estimates for dementia for selected risk factors, we performed a sensitivity analysis. We converted the HRs to RR estimates and repeated the calculations for unweighted and weighted PAFs for dementia associated with the following seven risk factors: less education (HR 1.77; 95% CI: 1.49–2.09); hearing loss (HR 1.28; 95% CI: 1.19–1.38); physical inactivity (HR 1.63; 95% CI: 1.38–1.92); smoking (HR 1.06; 95% CI: 0.84–1.35); diabetes (HR 1.49; 95% CI: 1.20–1.85); hypertension (HR 1.34; 95% CI: 1.13–1.59); and obesity (HR 1.29; 95% CI: 0.82–2.04) (Supplementary Table S1). This procedure resulted in higher unweighted and weighted PAFs for physical inactivity, hypertension, and less education, but lower PAFs for hearing loss, smoking, diabetes, and obesity. The overall weighted PAF for all risk factors increased from 38.9% to 41.8%. Using the Japan-specific RR estimates to calculate PIFs, we found that a 10% risk factor reduction would result in approximately 226,211 fewer cases of dementia (95% CI: 137,240 to 316,521) as compared to 208,185 fewer cases in the main analysis. A 20% risk factor reduction would result in approximately 442,327 fewer cases of dementia (95% CI: 270,934 to 612,898) as compared to 407,547 fewer cases in the main analysis (Supplementary Table S2).

Discussion

Japan has one of the most rapidly ageing populations in the world and the estimated prevalence of dementia in Japan was approximately 4,432,000 cases in 2022.⁷ Our primary results indicate that 38.9% of dementia cases in Japan could be potentially preventable. The leading modifiable risk factor in the present study was hearing loss, accounting for 6.7% of dementia cases in Japan. Given that age-related hearing loss typically occurs in middle and older ages, it is critical that individuals have wide access to facilities capable of accurately diagnosing hearing loss and providing effective interventions. Although hearing aid use may help prevent cognitive decline in individuals with hearing loss,³⁸ in Japan, only 15% of those aware of their hearing impairment use hearing aids, which is significantly lower than in other high-income countries. This disparity emphasises the urgency of encouraging older adults with hearing loss to consider using hearing aids.³⁹

The second largest contributor to dementia in Japan was physical inactivity with a weighted PAF of 6.0%. Similar PAFs have been reported in other high-income countries,^33,40, 41, 42, 43 likely because they share some characteristics with Japan such as sedentary work conditions and a well-developed transportation infrastructure. Other major contributors to dementia included high LDL cholesterol, diabetes, and hypertension, which are to some extent lifestyle-related health conditions. Mental and social issues such as depression and social isolation also showed considerable contributions to dementia. Together, these findings highlight the importance of policy interventions to promote increased physical activity, healthy lifestyles, expand access to healthcare where individuals can obtain appropriate diagnoses, and treatment and expand social support services targeting mental and social issues in Japan. Generally, it must be considered whether an individual-based approach,^38,44 a population-based approach,^45,46 or a combination of both, would be more effective for each specific factor.

In the main analysis, the Japanese risk factor prevalence rates were combined with RR estimates for dementia from the 2024 Lancet report to calculate PAFs and PIFs. These risk estimates came from meta-analyses based on data from numerous cohort studies conducted mainly in Western countries. As the resulting RRs may not be fully applicable to Japan, we performed a sensitivity analysis combining Japanese data on risk factor prevalence with Japanese risk estimates for dementia for seven risk factors (we were unable to identify Japanese risk estimates for all risk factors). This resulted in a minor increase in the overall weighted PAF for all risk factors. Our sensitivity analysis also affected the rank order of risk factors, as determined by PAF magnitude; physical inactivity now had the largest prevention potential with hearing loss in second place. Applying Japanese risk estimates to the PIF calculations resulted in slightly higher estimated reductions in dementia prevalence. This effect was mainly driven by a substantially higher PIFs for physical inactivity.

A key feature of the present study was the use of Japan-specific data on risk factor prevalence to estimate the contribution of modifiable risk factors for dementia. Comparison of Japanese versus global data on risk factor prevalence indicates that the prevalence rates of TBI (3.1% vs. 12.1%), untreated visual loss (3.7% vs. 12.7%), and less formal education (7.0% vs. 23.2%) may be markedly lower in Japan than globally.¹⁵ These differences probably reflect fundamental socio-economic differences between Japan and the rest of the world. Conversely, the prevalence rates of lifestyle-related risk factors such as physical inactivity, hypertension, diabetes, and excessive alcohol consumption seem to be higher in Japan than globally.

Combined PAFs of risk factors for dementia have been calculated for many high- and middle-income countries using nationally representative data,⁴⁷ including the USA³³; New Zealand⁴⁰; Australia⁴¹; Denmark⁴²; Canada⁴⁸; Italy⁴³; Brazil^49,50; Chile⁵¹; China; India, and six Latin America countries combined.⁵² A direct comparison of the results of our present study with results from other nations is problematic for several reasons. First, the included risk factors differ across the studies. Several studies did not include prevalence rate estimates for TBI, air pollution, or excessive alcohol consumption, and studies published prior to the 2024 Lancet Commission report did not include high LDL cholesterol or untreated visual loss as risk factors for dementia in their analyses. Only the 2024 Lancet Commission report and the 2025 study from Brazil⁵⁰ included all 14 risk factors. Second, the specific definitions of risk factors varied across the studies, as did the age ranges of the exposed groups. Despite these reservations, it may be noted that the weighted combined PAF for Japan (38.9%) is substantially lower than the global estimate (45.3%), and also lower than estimates for other high-income countries such as Australia (40.6%), the US (41.0%), New Zealand (47.7%), Canada (49.2%), and Italy (39.6%). The Japanese estimate, however, is higher than the weighted combined PAF for Denmark (35.2%), based on 12 risk factors.

Comparison of PIFs is further complicated by the fact that the hypothesised percentage reductions in risk factor prevalence differ across studies. A Danish study reported a weighted combined PIF of 4.1% at a 10% reduction in the prevalence of 12 risk factors, and a PIF of 8.1% at a 20% reduction in risk factor prevalence.⁴² These estimates are broadly in line with the PIF estimates in the present study (4.7% and 9.2%). For further comparison, an American study reported a weighted combined PIF of 7.3% at a 15% reduction in the prevalence of 12 risk factors,³³ and a recent study from New Zealand reported a PIF of 9.0% at a 15% reduction in the prevalence of 12 risk factors, and a PIF of 14.6 at a 25% reduction in risk factor prevalence.⁵³ PIF estimates are, per definition, cross-sectional and do not inform us about the pace at which the reductions in dementia prevalence may happen. The New Zealand study also presented modelled projections over a 30-year period indicating that a 15% reduction in risk factor prevalence may result in a 4.3% reduction in the prevalence of dementia after three decades, whereas a 25% risk factor reduction may produce a 7.0% reduction in dementia prevalence.⁵³ These modelled reductions over 30 years are substantially lower than the PIF estimates, which may be perceived as theoretical upper limits of prevention potential.

In the present study, we examined the 14 modifiable risk factors reported in the 2024 Lancet Commission report, but other factors such as sleep deprivation⁵⁴ and infectious diseases, including SARS-CoV-2,⁵⁵ have also been considered as potential risk factors for dementia which may be included in future analyses. Furthermore, genetic predispositions, such as the ε4-allele of the ApoE gene, which currently represents the largest known genetic risk factor,⁵⁶ may in the future become modifiable through biomedical advances.⁵⁷ If so, the overall proportion of potentially preventable dementia could increase substantially.

The outcome of dementia prevention countermeasures prioritised based on the present study may influence dementia prevention strategies in other countries soon to become super-aged societies and the study may work as a model for similar studies in other ageing societies. In June 2023, Japan enacted The Basic Act on Dementia to Promote an Inclusive Society, which became law in 2024⁵⁸ prompting the government to establish The Basic Plan for Dementia Policy Promotion.⁵⁹ The Guideline of Measures for Ageing Society was also approved, laying the groundwork for future dementia prevention policies in Japan.⁶⁰ At present, although the overall policy direction has been established, discussions are still ongoing regarding the specific measures to be undertaken.

Limitations

Our study has several limitations and potential sources of bias. The risk factor prevalence data for Japan were derived from various sources. We extracted prevalence rates from the NHNS for half of the risk factors,^16,17 but for the other half of the risk factors, prevalence rates were derived from other surveys, epidemiological studies, and even satellite-derived data (air pollution).^{19,20,22,24,26,27,29}

The use of various data sources hinders the delineation of a well-defined study period. The majority of the data were collected in 2019 and 2022 in the context of the NHNS,^16,17 but the data on TBI prevalence were collected between 1995 and 2006,²⁴ and the data on the prevalence of untreated vision loss were mainly collected in the 1990s.²⁹ This sampling period variability may limit the robustness of the results. As regional data on risk factor prevalence have not been systematically reported for Japan, we were unable to perform analyses at the prefecture or regional level. However, we hope that future prevalence surveys conducted by local governments will make such analyses possible.

Ideally, the HRs used to calculate PAF estimates should be derived from epidemiological studies that include roughly the same age ranges used for PAFs calculations. But the HRs used in the primary analysis in our study mainly came from meta-analyses presented in the 2024 Lancet report,¹⁵ regardless of eventual incongruities of age ranges. It is possible that the RR estimates for the risk factors presented in the 2024 Lancet Commission report,¹⁵ and applied in our PAF calculations, have limited validity in a Japanese context.⁴⁷ We addressed this issue by performing a sensitivity analysis involving Japan-specific risk estimates of dementia associated with seven risk factors, but we were not able to identify Japan-specific risk estimates for all risk factors.

The use of Japan-specific risk estimates in the sensitivity analysis has its own possible limitations. The Ohsaki Cohort 2006 Study, from which we extracted six of the seven HRs, is a local community-based study of people ≥65 years of age. This means that the risk estimates may not be fully applicable to the population of Japan.³⁴

Ideally, the communality values used as weights in the PAF and PIF calculations should be based on a national data set including all risk factors, but as our data were derived from different sources, we could not produce Japan-specific communalities. Instead, we used communality values from the 2024 Lancet Commission report based on Norwegian data, which may have limited relevance for Japan.

There are further possible limitations concerning the definitions of specific risk factors. We generally aimed at defining risk factors that were compatible with the definitions used in the 2024 Lancet report,¹⁵ but due to the way that the Japanese data were collected and presented it was necessary to adjust definitions and age ranges for some risk factors. The definition of depression in our study is relatively broad, as it includes major depression, bipolar disorder, and dysthymia.²² Consequently, the prevalence of depression may be overestimated in our study. Although we chose a relatively broad definition, the resulting prevalence rate for depression in Japan is lower than the global prevalence rate (7.2%) reported in the 2024 Lancet report.¹⁵

It was challenging to estimate the prevalence rate of TBI in Japan, as we could only retrieve data on incidence.²⁴ Our crude TBI prevalence rate estimate of 3.1% is markedly lower than the global prevalence rate estimate of 12.1% presented in the 2024 Lancet Commission report, but relatively close to the population- and registry-based estimate of 4.7% for Denmark, a country that is comparable to Japan regarding high traffic safety and low rates of violent crime.⁴² We suspect that the true prevalence of TBI in Japan may be higher than our estimate here, which is based only on patients admitted to neurosurgical institutes and thus likely misses the full range of milder TBIs and concussions.

The definition of physical inactivity in our study makes lower demands on the amount of physical activity compared to the definition in the 2024 Lancet report.^15,16 Consequently, the prevalence of physical inactivity may be underestimated in our study. The definition of high alcohol consumption in our study is higher for men (≥35 alcohol units per week), but lower for women (≥17.5 units)¹⁶ compared to the definition in the 2024 Lancet report (≥21 units regardless of sex).¹⁵ Consequently, the prevalence of high alcohol consumption may be underestimated for men in our study and overestimated for women. This is not ideal, but as we used publicly available data on alcohol consumption, we were not able to apply alternative definitions. The definition of social isolation in our study was based on living status,²⁶ which may be considered a proxy variable for social isolation. Despite this possible limitation, we chose the definition to make our results comparable to the results of the 2024 Lancet report.¹⁵ The data on air pollution exposure used in our study are for the entire population of Japan and does not enable subgroup analyses according to differential risk of exposure associated with, for example, living in an urban versus rural setting.

In conclusion, dementia preventive efforts in Japan should prioritise better hearing care, physical activity, and metabolic health in middle-aged and older adults. In addition, local governments should also consider promoting healthy lifestyles and expanding access to enhanced healthcare for older individuals as soon as possible.

Contributors

KW and KJ conceptualised the study, conducted all analyses, wrote the manuscript, reviewed the manuscript, and revised the final manuscript.

Data sharing statement

As the study uses anonymised, publicly available data, it does not contain individual participant data. The results reported in this article (text, tables, figures, and Supplement Tables) will be shared. The study protocol and statistical analysis plan will be available. The data will become available beginning 3 months and ending 5 years following article publication. The data will be shared with researchers who provide a methodologically sound proposal. Proposals should be directed to wasano@tokai.ac.jp. To gain access, data requestors will need to sign a data access agreement. Requests for clarification of specific issues related to the current publication will be considered by the steering committee as long as provision of such data does not interfere with future publications by the research team.

Editor note

The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations.

Declaration of interest

KW reports personal fees for consulting from AMGEN Inc. and KYORIN Pharmaceutical Co., Ltd., outside the submitted work.