Unveiling the link between physical activity levels and dementia risk: Insights from the UK Biobank study

Abstract

Background:

There is limited information on the mixture effect and weights of light physical activity (LPA), moderate physical activity (MPA), and vigorous physical activity (VPA) on dementia risk.

Methods:

A prospective cohort study was conducted based on the UK Biobank dataset. We included participants aged at least 45 years old without dementia at baseline between 2006-2010. The weighted quantile sum regression was used to explore the mixture effect and weights of three types of physical activity on dementia risk.

Results:

This study includes 354,123 participants, with a mean baseline age of 58.0-year-old and 52.4 % of female participants. During a median follow-up time of 12.5 years, 5,136 cases of dementia were observed. The mixture effect of LPA, MPA, and VPA on dementia was statistically significant (β: −0.0924, 95 % Confidence Interval (CI): (−0.1402, −0.0446), P < 0.001), with VPA (weight: 0.7922) contributing most to a lower dementia risk, followed by MPA (0.1939). For Alzheimer’s disease, MPA contributed the most (0.8555); for vascular dementia, VPA contributed the most (0.6271).

Conclusion:

For Alzheimer’s disease, MPA was identified as the most influential factor, while VPA stood out as the most impactful for vascular dementia.

1. Introduction

Dementia affects more than 50 million individuals globally, with Alzheimer’s disease being the most prevalent subtype (60-70 %), followed by vascular dementia (25 %) (Patterson, 2018). With one new case diagnosed every 3.2 seconds, it is projected that the number of people living with dementia will reach 152 million in 2050 (Patterson, 2018; Prince et al., 2015). There is no medication available for the treatment of dementia, and six drugs approved by the U.S. Food and Drug Administration (FDA) only manage symptoms and have severe side effects such as stroke and death (Gaugler et al., 2022). However, The Lancet Commission reported that the number of dementia could be potentially reduced by 40 % by addressing twelve modifiable risk factors (Livingston et al., 2020). Therefore, targeting these modifiable risk factors is an effective strategy for dementia prevention and management.

Among the twelve modifiable risk factors, physical activity is considered one of the most important modifiable risk factors for dementia. Strong evidence from meta-analyses (Groot et al., 2016; Hersi et al., 2017), longitudinal studies (Najar et al., 2019; Zhu et al., 2022), and genetic studies (Zhu et al., 2022) generally supports the causal relationship between physical activity and dementia, despite a few exceptions reporting null associations (Sabia et al., 2017; Young et al., 2015). The reasons for the null association could be due to under-diagnosed dementia cases (Sabia et al., 2017) and the dose-response relationship between physical activity and incident dementia (Young et al., 2015). A recent systematic review (Erickson et al., 2022) has shown that physical activity could be conducive to cognitive function through cellular and molecular changes such as angiogenesis and neurogenesis, structural and functional brain changes such as volume and thickness and elasticity, and behavioral and socioemotional changes such as strengthening and reducing fatigue.

Previous studies (Groot et al., 2016; Hu et al., 2022; Shao et al., 2022) have primarily concentrated on the overall effect of physical activity on cognitive function or the effects of individual types of physical activities on cognition through subgroup analyses. Several studies (Hu et al., 2022; Shao et al., 2022) have focused on the effect of different types (light, moderate, and vigorous) of physical activity on incident dementia. However, as multiple types of physical activity are correlated and might interact with each other (i.e., persons doing moderate physical activity tend to have high levels of other types of physical activity), there are two research gaps. First, the mixture effect of three types of physical activity has not been examined. Second, the relative contribution of each type of physical activity to dementia risk remains unknown.

Due to high correlations between individual types of physical activity and the presence of coexistence, the traditional Cox proportional hazards model may underestimate the true effects of each physical activity on demntia and be biased or underpowered in estimating these effects. Recent advancements in statistical analysis, such as the weighted quantile sum (WQS) regression, have become increasingly popular in studies of chemical mixtures. For example, the WQS method effectively transfer correlated components, categorized as ordinal variables, into quantiles. These quantiles are then combined into an index for analyzing the mixture’s collective effect (Czarnota et al., 2015). By constraining the weights to sum to 1 (with values ranging from 0 to 1), the WQS regression simplifies the data structure and mitigates the challenges posed by collinearity among correlated variables (Czarnota et al., 2015). These advanced models provide the tools necessary for investigating the joint effects of different types of physical activity on the risk of incident dementia. This study aims to estimate the mixture effect of three types of physical activity and their relative contribution to incident dementia using WQS regression.

2. Methods

2.1. Study design

This study is presesented according to the Strengthening the Reporting of Observational studies in Epidemiology guidelines (Von Elm et al., 2007) (Supplementary Table 1). A retrospective cohort study was established based on the UK Biobank and included more than 500,000 participants aged 40 to 70 at baseline (Sudlow et al., 2015) (Application ID: 61915). The participants were assessed between 2006 and 2010 in 22 centers throughout the UK (Sudlow et al., 2015). Participants provided information on physical activity and other potentially health-related aspects through questionnaires, interviews, accelerometers, and biological samples (Said et al., 2018; Sudlow et al., 2015). The participants have also agreed to have their future health, including disease events, monitored. All participants provided written informed consents (Said et al., 2018). The UK Biobank study has been approved by the North West Multi-center Research Ethics Committee (Said et al., 2018).

In this study, we included 502,486 participants at baseline. Participants were excluded if they were 1) less than 45 years old; 2) with the self-reported prevalent cognitive impairment or Alzheimer’s disease or dementia at baseline; 3) diagnosed with dementia; and 4) missing data on physical activity and covariates. Finally, a total of 345,123 participants were included in the final analyses. The flow chart of subject selection is shown in Fig. 1.

Fig. 1.

The flow chart of study participants.

NOTE: ¹ MET, metabolic equivalent.

2.2. Physical activity assessment

In this study, the metabolic equivalent (MET) (Ainsworth et al., 2011) $\frac{\text{energy expended per kilogram of body weight per hour}}{\text{energy expended per kilogram of body weight per hour when sitting}}$) was calculated to lated to represent LPA, MPA, and VPA. First, physical activities were assessed based on several questions that were adapted from the international physical activity questionnaire (IPAQ) short form (Craig et al., 2003). Questions are listed in Supplementary Table 2 and are available to view on the UK Biobank Web site (http://biobank.ctsu.ox.ac.uk/crystal/label.cgi?id=100054). Based on the questions, the frequencies and durations of the three types of physical activity were collected and achieved. For example, regarding MPA, the questions were “In a typical week, on how many days did you do 10 min or more of moderate physical activities like carrying light loads, cycling at a normal pace?” and “How many minutes did you usually spend doing moderate activities on a typical day?” The frequency and duration of walking were captured. Then, each level was assigned the following MET hour values: 3.3 for the LPA, 4.0 for the MPA, and 8.0 for the VPA (Ainsworth et al., 2011; Craig et al., 2003). In line with IPAQ guidelines (Craig et al., 2003), physical activity with less than 10 minutes per day for walking, moderate, or vigorous activity was recoded to 0. The total METs were calculated by summing METs from each activity level. The unit of MET was minutes per week.

2.3. Dementia assessment

The incident dementia cases within the UK Biobank were identified through linkage to the hospital in-patient admission records (hospital Episode Statistics for England, Scottish Morbidity Record data for Scotland, and the Patient Episode Database for Wales) and death registries (National Health Service Digital for England and Wales and the Information and Statistics Division for Scotland) (Lourida et al., 2019). Participants with dementia and its subtypes were identified using ICD-9 and ICD-10 codes for dementia (ICD-9: 290.2, 290.3, 291.2, 294.1, 331.1, 331.2, 331.5; ICD-10: A81.0, F02, F02.0, F02.1, F02.2, F02.3, F02.4, F02.8, F03, F05.1, F10.6, G31.0, G31.1, G31.8), Alzheimer’s disease (ICD-9: 331.0; ICD-10: F00, F00.0, F00.1, F00.2, F00.9, G30, G30.0, G30.1, G30.8, G30.9), and vascular dementia (ICD-9: 290.4; ICD-10: F01, F01.0, F01.1, F01.2, F01.3, F01.8, F01.9, I67.3). The codes used in the UK Biobank study to identify dementia cases are shown in Supplementary Table 3. The follow-up began from the date attending the assessment center to the earliest date of dementia diagnosis, the date of loss to follow-up, the date of death, and the censoring (September 2021), whichever occurred first.

2.4. Covariates

Covariates included socio-demographic factors (age, sex, ethnicity, education, Townsend deprivation index), lifestyle factors (smoking status and alcohol status), diseases (stroke, hypertension, diabetes, and depression), and the genetic factor (polygenic risk score, PRS). Age was a continuous variable. Sex was classified as female and male. Ethnicity was categorized as White, Mixed, Asian or Chinese, Black, and others. Education was classified as higher, upper secondary, lower secondary, vocational, and others. Townsend deprivation index was categorized into three levels (quantiles 1, 2 to 4, and 5), combining information on social class, employment, car availability, and housing) (Townsend, 1987). Smoking status was classified as never, previous, and current. Smoking status was categorized as never, previous, and current. Diseases including stroke, hypertension, diabetes, and depression were categorized as yes and no.

The genetic factor of PRS was considered an effect modifier instead of the confounder to avoid the effect modification bias (VanderWeele and Robins, 2007). An effect modifier is a cause of the outcome and is not associated with the exposure (VanderWeele and Robins, 2007). The PRS for dementia was calculated as below briefly. Detailed information on PRS can be accessed in the UK Biobank (Said et al., 2018). In a brief, the genotype data were generated using a custom Axiom genotyping array assaying about 825,927 genetic variants, followed by genome-wide imputation (Said et al., 2018). Phenotype information on dementia was assessed using self-reported dementia/Alzheimer/cognitive impairment. Genetic quality control included single nucleotide polymorphisms (SNPs) with a minor allele frequency greater than 0.05, and SNPs were tested for Hardy-Weinberg equilibrium. The principal component-based ancestry centering step was applied to approximately centre the score distributions on zero across all ancestries. Per-individual PRS values were calculated as the genome-wide sum of the per-variant posterior effect size multiplied by allele dosage. The total number of participants with PRS was 486,133. PRS was centered and standardized with numbers ranging from −0.72731 to 6.20547 (Leonenko et al., 2021). The PRS was categorized into three groups: low (lowest quantile), intermediate (quantiles 2 to 4), and high (highest quintile). The field ID used in analyses is shown in Supplementary Table 4. The directed graph was drawn to explore the confounders, effect modifiers, and the minimal sufficient adjustment set. (Supplementary Figure 1).

2.5. Statistical analyses

In the descriptive analysis, continuous variables were presented as the means with standard deviation (SD) if they were normally distributed. Otherwise, a median with an interquartile range (IQR) was used. The categorical variables were presented as numbers with percentages. Spearman correlation coefficients were calculated between three types of physical activity.

The Cox proportional hazards model was used to explore the effects of each type of physical activity (YES versus No) on incident dementia (Marrie et al., 2009). We further analyzed the dose-response relationships between three types of physical activity and dementia. The nonlinearity test was performed using the analysis of variance (ANOVA). If there was a nonlinear relationship (P for nonlinearity < 0.05) between physical activity and dementia, the restricted cubic spline (RCS) (Marrie et al., 2009) with four knots was performed (Harrell, 2001). The confounders (age, sex, ethnicity, education, Townsend deprivation index, smoking status, alcohol status, stroke, hypertension, depression, and diabetes) were considered in the model. The results were represented as hazards ratio (HR) with 95 % confidence intervals (CIs).

The weighted quantile sum (WQS) was utilized to assess the contributions of three types of physical activity and identify the one that has the most significant association with the risk of developing dementia (Czarnota et al., 2015). Specifically, the WQS method estimates the joint effects of three types of physical activities on dementia by assigning weights to reflect their relative importance, calculatings quantile scores for each type of physical activity, and then combining them into a single index for each individual. The WQS allows for examining collective impacts of three types of physical activities on dementia, taking into account varying significance and distribution of each activity type. The weighted scores are constrained to add up to 1 with each score ranging between 0 and 1, thereby reducing dimensionality and addressing issues associated with collinearity (Czarnota et al., 2015). The model can be described as: $g(\mu )={\beta }_0+{\beta }_1({\sum }_{i=1}^c{w}_i{q}_i)+z^{\prime }\phi$. In this analysis, β₀ is the model intercept. β₁ is the coefficient summarizing the overall effect to the weighted index $({\sum }_{i=1}^c{w}_i{q}_i)$ for the three types of physical activity. The w_i represents the weight for the ith physical activity type q_i. Furthermore, z denotes a vector of covariates determined prior to the estimation of the weights, φ are the coefficients for the covariates in z. The g(μ) is the logit link that relates the incident dementia (μ), to the variables in the right-hand side of equation (Czarnota et al., 2015). The weights are derived through a bootstrap and constrained to sum to one and to be bounded between zero and one. For each bootstrap sample (B=100), a sample dataset is created by sampling with replacement from the training dataset, and the model’s parameters are estimated through an optimization algorithm. The analysis was performed using the R 4.2.2 Version with the packages included “survival”, “rms”, and “gWQS”.

2.6. Subgroup analyses

We conducted subgroup analyses on subtypes of dementia, including Alzheimer’s disease and vascular dementia. In addition, to explore whether the associations between physical activity and dementia might be modified by disease susceptibility, we performed stratified analyses by the PRS for dementia.

3. Results

3.1. Population characteristics

Among the 354,123 participants, the mean (SD) age at recruitment was 58.0 (6.85) years old, and 52.4 % were female. During a median (interquartile range) follow-up time of 12.5 years (11.80 - 13.26 years), a total of 5,136 cases of dementia was observed, including 2,090 cases of Alzheimer’s disease and 1,153 cases of vascular dementia. A total of 68,975 (19.5 %) participants had low PRS, 207,542 (58.6 %) had intermediate PRS, and 69,031 (19.5 %) had high PRS. The number of missing data on PRS was 8,575 (2.4 %). The baseline characteristics of participants is shown in Table 1.

Table 1.

The baseline characteristics of study participants (N = 354,123).

	All-cause dementia
	Total (N=354,123)	No (N=348,987)	Yes (N=5,136)
Age (years)
Mean (SD)	58.0 (6.85)	57.9 (6.84)	64.3 (4.53)
Sex
Female	185,466 (52.4 %)	183,228 (52.5 %)	2,238 (43.6 %)
Male	168,657 (47.6 %)	165,759 (47.5 %)	2,898 (56.4 %)
Ethnicity
White	338,291 (95.5 %)	333,358 (95.5 %)	4,933 (96.0 %)
Mixed	1,780 (0.5 %)	1,758 (0.5 %)	22 (0.4 %)
Asian or Chinese	6,786 (1.9 %)	6,711 (1.9 %)	75 (1.5 %)
Black	4,600 (1.3 %)	4,523 (1.3 %)	77 (1.5 %)
Others	2,666 (0.8 %)	2,637 (0.8 %)	29 (0.6 %)
Education
Higher	174,335 (49.2 %)	172,449 (49.4 %)	1,886 (36.7 %)
Upper secondary	26,336 (7.4 %)	26,010 (7.5 %)	326 (6.3 %)
Lower secondary	74,204 (21.0 %)	73,320 (21.0%)	884 (17.2 %)
Vocational	21,689 (6.1 %)	21,334 (6.1 %)	355 (6.9 %)
Others	57,559 (16.3 %)	55,874 (16.0%)	1,685 (32.8 %)
Townsend deprivation index
1 (least deprived)	72,790 (20.6 %)	71,870 (20.6 %)	920 (17.9 %)
2-4	214,500 (60.6 %)	211,536 (60.6 %)	2,964 (57.7 %)
5 (most deprived)	66,833 (18.9 %)	65,581 (18.8%)	1,252 (24.4 %)
Smoking status
Never	191,652 (54.1 %)	189317 (54.2 %)	2,335 (45.5 %)
Previous	128,039 (36.2 %)	125786 (36.0 %)	2,253 (43.9 %)
Current	34,432 (9.7 %)	33884 (9.7 %)	548 (10.7 %)
Alcohol status
Never	13,980 (3.9 %)	13655 (3.9 %)	325 (6.3 %)
Previous	12,393 (3.5 %)	12038 (3.4 %)	355 (6.9 %)
Current	327,750 (92.6 %)	323294 (92.6 %)	4,456 (86.8 %)
Stroke
No	349,362 (98.7 %)	344499 (98.7 %)	4,863 (94.7 %)
Yes	4,761 (1.3 %)	4488 (1.3 %)	273 (5.3 %)
Hypertension
No	257,533 (72.7 %)	254572 (72.9 %)	2,961 (57.7 %)
Yes	96,590 (27.3 %)	94415 (27.1 %)	2,175 (42.3 %)
Depression
No	348,743 (98.5 %)	343,681 (98.5 %)	5,062 (98.6 %)
Yes	5,380 (1.5 %)	5,306 (1.5 %)	74 (1.4 %)
Diabetes
No	338,332 (95.5 %)	333,802 (95.6 %)	4,530 (88.2 %)
Yes	15,791 (4.5 %)	15,185 (4.4 %)	606 (11.8 %)
PRS1
Low	68,975 (19.5 %)	68,444 (19.6%)	531 (10.3 %)
Intermediate	207,542 (58.6 %)	205,150 (58.8 %)	2,392 (46.6 %)
High	69,031 (19.5 %)	66,963 (19.2%)	2,068 (40.3 %)
Missing	8,575 (2.4 %)	8,430 (2.4 %)	145 (2.8 %)
LPA2
Median [Min, Max]	693 [0, 4,160]	693 [0, 4,160]	693 [0, 4,160]
MPA3
Median [Min, Max]	480 [0, 5,040]	480 [0, 5,040]	480 [0, 5,040]
VPA4
Median [Min, Max]	160 [0, 10,100]	160 [0, 10,100]	0 [0, 10,100]

Note:

¹PRS, polygenic risk scores;

²LPA, light physical activity;

³MPA, moderate physical activity;

⁴VPA, vigorous physical activity.

3.2. Correlations among three types of physical activity

In addition to three types of physical activity, the total physical activity was included to examine its associations with three types of physical activity. The results of Spearman analyses showed that the correlation coefficients (ρ) between the three types of physical activity ranged from 0.22 to 0.82. The correlation between the MPA and the total was strong (ρ = 0.82, P < 0.001). The correlation between the LPA and the VPA was relatively weak (ρ = 0.22, P < 0.001). The correlations between the MPA, and LPA (ρ = 0.44, P < 0.001) and VPA (ρ = 0.47, P < 0.001) were moderate (Supplementary Figure 2).

3.3. The effect of three types of physical activity on dementia

The results of the Cox proportional hazard regression analysis showed that the LPA (YES/NO) (HR 0.61, 95 %CI 0.55 to 0.68, P < 0.001), the MPA (HR 0.61, 95 %CI 0.55 to 0.68, P < 0.001), and the VPA (HR 0.80, 95 %CI 0.76 to 0.85, P < 0.001) were significantly associated with a lower risk of all-cause dementia. (Supplementary Table 5). The results of RCS showed that there were nonlinear relationships between three types of physical activity and all-cause dementia. Individuals with an LPA with 690 (about 209 minutes of light physical activity per week) (Fig 2A&D), or MPA with 633 (about 203 minutes of any moderate physical activity per week) (Fig. 2B&D), or VPA of 1621 (about 158 minutes of any vigorous physical activity per week) had the lowest risk of developing all-cause dementia (Fig. 2C&D).

Fig. 2.

Dose-response relationships between three types of physical activity and all-cause dementia.

NOTE: ¹ CI, confidence interval; ² LPA, light physical activity; ³ MPA, moderate physical activity; ⁴ VPA, vigorous physical activity.

3.4. The mixture effect and weights in associations between physical activities and dementia

For all-cause dementia, in the crude model, the results of the WQS showed that the mixture effect of three types of physical activity on all-cause dementia was statistically significant (β −0.1876, 95 %CI −0.2324 to −0.1429, P < 0.001). In the fully covariate-adjusted model, this association was still statistically significant (β −0.0924, 95 %CI −0.1402 to −0.0446, P < 0.001) (Supplementary Table 6). The estimated weights of three types of physical activity showed that the VPA (0.7922) contributed most to reduce the risk of all-cause dementia, followed by the MPA (0.1939) (Fig. 3).

Fig. 3.

Weights of three types of physical acitivity contributing to reducing the risk of all-cause dementia.

NOTE: ¹ LPA, light physical activity; ² MPA, moderate physical activity; ³ VPA, vigorous physical activity.

3.5. Subgroup analyses

3.5.1. Subtypes of the dementia

For Alzheimer’s disease, the results of the Cox proportional hazard regression analysis showed that the MPA was significantly associated with a lower risk of Alzheimer’s disease (HR 0.86, 95 %CI 0.77 to 0.96, P = 0.009). The LPA and VPA were not significantly associated with a lower risk of Alzheimer’s disease (P > 0.050) (Supplementary Table 7). The results of RCS showed that there were non-linear relationships between the three types of physical activity and Alzheimer’s disease (Fig. 4). For vascular dementia, the results of the Cox proportional hazard regression analysis showed that three types of physical activity (HR ranging from 0.52 to 0.79) were significantly associated with a lower risk of vascular dementia (Supplementary Table 7). The results of RCS showed that there were non-linear relationships between the three types of physical activity and vascular dementia (Fig. 4).

Fig. 4.

Subgroup analysis on the subtype of dementia (A) and polygenic risk scores (B) in dose-response associations between three types of physical activity and dementia.

NOTE: ¹ CI, confidence intervals; ² LPA, light physical activity; ³ MPA, moderate physical activity; ⁴ VPA, vigorous physical activity.

In the crude model, the results showed that the WQS index was significantly associated with a lower risk of Alzheimer’s disease (β −0.1177, 95 %CI −0.1859 to −0.0494, P < 0.001) and vascular dementia (β −0.1902, 95 %CI −0.2890 to −0.0915, P < 0.001). In the fully covariate-adjusted model, these associations were not statistically significant (P > 0.050) (Supplementary Table 8). The results of estimated weights of three types of physical activity showed that the MPA (0.8555) contributed most to Alzheimer’s disease, and the VPA (0.6271) contributed most to vascular dementia (Fig. 5).

Fig. 5.

Subgroup analysis on the subtype of dementia (A) and polygenic risk scores (B) regarding weights in associations between three types of physical activity and dementia.

NOTE: ¹ LPA, light physical activity; ² MPA, moderate physical activity; ³ VPA, vigorous physical activity.

3.5.2. Polygenic risk scores

The results of the Cox proportional hazard regression analysis showed that the LPA (HR ranging from 0.57 to 0.60), MPA (HR ranging from 0.60 to 0.78), and VPA (HR ranging from 0.57 to 0.85) were significantly associated with a lower risk of dementia in different polygenic risk groups (Supplementary Table 9). The results of the RCS showed that there were non-linear relationships between the three types of physical activity and dementia in different polygenic risk groups (Fig. 4).

In the crude model, the results of the WQS showed that the mixture effect of three types of physical activity was significantly associated with dementia in low (β −0.2901, 95 %CI −0.4493 to −0.1309, P < 0.001), intermediate (β −0.2377, 95 %CI −0.3037 to −0.1718, P < 0.001), and high polygenic genetic risk groups (β −0.1863, 95 %CI −0.2624 to −0.1102, P < 0.001). In the fully covariate-adjusted model, these associations were still statistically significant in low (β −0.1310, 95 %CI −0.2575 to −0.0044, P = 0.042) and intermediate polygenic risk groups (β −0.1562, 95 %CI −0.226 to −0.0864, P < 0.001), but not the high polygenic risk group (β −0.0339, 95 %CI −0.0885 to 0.0206, P = 0.2223) (Supplementary Table 10). The results of estimated weights of three types of physical activity showed that the VPA contributed most in the low (0.7025) and intermediate (0.8615) polygenic risk groups while the MPA contributed most in the high polygenic risk group (0.9156) (Fig. 5).

4. Discussion

This study identified the mixture effect and weights in associations between LPA, MPA, and VPA and dementia. Given genetic risk, physical activity patterns that are more adherent to MPA and VPA were associated with a reduced risk of dementia. Our study also found that the specific type of physical activity that was more effective in reducing the risk of dementia varied based on the subtype of dementia. For Alzheimer’s disease, MPA was found to be more effective, while for vascular dementia, VPA was found to be more powerful. These findings suggest that a combination of both MPA and VPA, tailored to individual risk factors, could be an effective strategy for reducing the risk of dementia.

We found that the LPA, MPA, and VPA could significantly reduce the risk of dementia while considering the genetic risk and other confounders. This evidence was in line with previous studies (Paula et al., 2022; Sizhen et al., 2022). One recent meta-analysis (Paula et al., 2022) taking account of the effects of reverse causation showed that, even in longer follow-ups, physical activity was associated with lower incidences of all-cause dementia and Alzheimer’s disease. Another meta-analysis (Sizhen et al., 2022) with 2,154,818 participants showed that even leisure activities, such as social activities, could reduce the risk of dementia. Through the analysis of dose-response relationships between LPA, MPA, and VPA and dementia, the best dose for reducing the risk of dementia were 209 minutes per week for LPA, 158 minutes per week for MPA, and 203 minutes/weekly for VPA. This result was similar with the guideline published by the World Health Organization (WHO) (WHO, 2020). WHO has suggested that adults and older adults should do at least 150 to 300 minutes of MPA, or at least 75-150 minutes of VPA, or an equivalent combination of MPA and VPA throughout the week for substantial health benefits (WHO, 2020).

The significant mixture effect of LPA, MPA, and VPA on dementia suggested that the combination of three types of physical activity was conducive to reduce dementia risk. We also found that the VPA and MPA conferred more cognitive benefits than LPA. This results also supported by the PRS subgroup analysis with VPA and MPA contributing most in the low, intermdaite, and high polygenic risk group. Previous studies rarely studied relative contribution of each physical activity contributing to dementia, but several studies compared effects of different types of physical activity on dementia. A previous meta-analysis with 32 longitudinal studies (Guure et al., 2017) using both Bayesian parametric and non-parametric models concluded that high physical activity was more protective against Alzheimer’s disease (HR: 0.62, 95 % CI 0.49, 0.75) than vascular dementia (HR: 0.92, 95 % CI 0.62, 1.30). However, the measurement of physical activity focused on the frequencies of physical activity and ignored the role of intensity and detailed duration of physical activity. One recent umbrella review (López-Ortiz et al., 2023) exploring the effect of physical activity intervention on Alzheimer’s disease found that Alzheimer’s disease risk reduction was higher among those performing VPA (38 %) than MPA (29 %). Current evidence indicated the the MPA and VPA might be more effective in reducing dementia risk compared with LPA.

The strengths of associations between three types of physical activity and subtypes of dementia risk differed. The MPA played a more important role than VPA in preventing Alzheimer’s disease. The VPA was more effective than MPA in preventing vascular dementia. One prospective study (Hansson et al., 2019) with 21 years follow-up duration explored the effect of VPA on Alzheimer’s disease and vascular dementia, and results showed that VPA was effective to reduce the risk of vascular dementia but not Alzheimer’s disease. Generally, both MPA and VPA could improve the level of neurotrophins, the structural integrity of the hippocampus, the clearance of amyloid, and cardiovascular health (Erickson et al., 2022). Specifically, the physiological pathways and mechanisms that mediate cognitive benefits might differ in MPV and VPA (Erickson et al., 2022). Previous evidence (Koščak Tivadar, 2017) showed that MPA has a positive impact on general congtive ability, working memory and attention, verbal memory and attention, which was prominent in Alzheimer’s disease. Meanwhile, VPA requires more attention to PA and less to cognitive processes, particularly in terms of reducing reactions, selective attention and flexibility to tasks (Koščak Tivadar, 2017). Moreover, compared with MPA, VPA could produce more cardiovascular fitness that is more powerful to protect against vascular dementia than Alzheimer’s disease (Koščak Tivadar, 2017). Therefore, in general, the combination of MPA and VPA could be performed to reduce the risk of dementia. Specifically, for Alzheimer’s disease, more MPA could be implemented, while for vascular dementia, more VPA could be implemented. However, we found that the proportion of VPA was the smallest in the total physical activity. The awareness of doing VPA is needed to be improved.

4.1. Clinical and policy Implications

For clinical environment, our findings underscore the significance of physical activity in mitigating the risk of dementia, considering an individual’s genetic predisposition. Specifically, our study reveals that MPA can be especially advantageous in reducing the risk of Alzheimer’s disease, while engaging in VPA has the greatest impact on lowering the risk of vascular dementia. These results emphasize the potential of customized physical activity interventions that target specific dementia types, providing valuable insights for preventive strategies and personalized care approaches.

For policy makers, we suggested that the MPA and VPA should be prompted to the public regarding dementia management. Specifically, the VPA shoud be emphasized to prevent vascular dementia. The proportion of participants doing VPA was lower (58.2 %), compared with that of MPA (83.7 %) and LPA (95.9 %). The proportion of participants doing VPA was even higher than other populations. For example, the number was 27.0 % (13.0 % from leisure and 14.0 % from work) in National Health and Nutrition Examination Survey (https://wwwn.cdc.gov/Nchs/Nhanes/2011-2012/PAQ_G.htm) in 2011. The number was 34.1 % in China Health and Retirement Longitudinal Study (https://charls.charlsdata.com/pages/data/111/en.html) in 2011. The differences in the proportion might due to ethic differences in public health awareness (He and Baker, 2005).