Aspirin Use and Risk of Alzheimer’s Disease: A 2-Sample Mendelian Randomization Study

Abstract

Background:

Observational studies have shown inconsistent findings of the relationships between aspirin use and the risk of Alzheimer’s disease (AD).

Objective:

Since residual confounding and reverse causality were challenging issues inherent in observational studies, we conducted a 2-sample Mendelian randomization analysis (MR) to investigate whether aspirin use was causally associated with the risk of AD.

Methods:

We conducted 2-sample MR analyses utilizing summary genetic association statistics to estimate the potential causal relationship between aspirin use and AD. Single-nucleotide variants associated with aspirin use in a genome-wide association study (GWAS) of UK Biobank were considered as genetic proxies for aspirin use. The GWAS summary-level data of AD were derived from a meta-analysis of GWAS data from the International Genomics of Alzheimer’s Project (IGAP) stage I.

Results:

Univariable MR analysis based on these two large GWAS data sources showed that genetically proxied aspirin use was associated with a decreased risk of AD (Odds Ratio (OR): 0.87; 95%CI: 0.77–0.99). In multivariate MR analyses, the causal estimates remained significant after adjusting for chronic pain, inflammation, heart failure (OR = 0.88, 95%CI = 0.78–0.98), or stroke (OR = 0.87, 95%CI = 0.77–0.99), but was attenuated when adjusting for coronary heart disease, blood pressure, and blood lipids.

Conclusion:

Findings from this MR analysis suggest a genetic protective effect of aspirin use on AD, possibly influenced by coronary heart disease, blood pressure, and lipid levels.

Keywords

Alzheimer’s disease aspirin blood pressure cardiovascular disease inflammation lipids pain 2-sample Mendelian randomization analysis

INTRODUCTION

With an estimated 6.2 million Americans over 65 years of age suffering from AD in 2021, the Alzheimer’s Association has reported $355 billion in total health care spending for people with dementia. It is estimated that Alzheimer’s disease (AD) will affect 13.8 million Americans in 2060, making it a major health challenge in the US [1]. In June 2021, the US Food and Drug Administration (FDA) approved aducanumab, a monoclonal antibody, as the first-ever-disease-modifying drug for AD [2]. Although aducanumab could reduce amyloid plaques [3], what remains controversial is whether the amyloid clearance protects patients from cognitive and functional decline [4]. In this scenario, targeting modifiable risk factors is a promising strategy to prevent AD. Growing evidence has indicated that neuroinflammation and blood clots play important roles in the progression of AD [5, 6]. Aspirin, also known as acetylsalicylic acid, is a commonly used medication for anti-inflammatory and antiplatelet therapies [7].

Aspirin use may be associated with decreased risk of AD because of its anti-inflammatory and cardioprotective properties [8] (http://www.alzdiscovery.org). However, observational study and clinical trial have yielded inconsistent and often contradictory findings. Etminan et al. [9] conducted a meta-analysis of 3 case-control and 5 cohort studies and reported that aspirin had no significant association with the risk of developing AD. Nilsson et al. [10] conducted a cross-sectional analysis which indicated that high-dose aspirin could significantly reduce AD risk for people aged≥80 years. Szekely et al. [11] pooled six prospective studies to report a reduced risk of AD in aspirin users. Subsequent large prospective studies have reported either no effect [12 –15], increased risk [16, 17], or reduced risk [18, 19] of aspirin on AD. The ASPREE (ASPirin in Reducing Events in the Elderly) clinical trial [20] indicated low-dose aspirin was not associated with decreased risk of developing AD in healthy elderly individuals. A recent observational study found that long-term low-dose aspirin use can slow the development of AD in patients with coronary heart disease (CHD) but not in other individuals [19]. Another retrospective cohort study presented that aspirin use may present AD in patients with ischemic stroke [21]. These inconsistent findings demonstrated the inherent limitations of observational studies, the highly heterogenous nature of AD, and the possibility that aspirin use is associated with decreased risk of AD in specific populations which are not included in existing clinical trials.

In this study, we assess potential causal associations between aspirin use and AD through Mendelian randomization (MR) analysis based on genome-wide association study (GWAS) summary statistics. MR analysis utilizes single-nucleotide variants (SNVs) as unconfounded genetic proxies for exposure and infers their causal effects on outcomes, which can minimize the bias encountered in observational epidemiologic studies [22 –24]. MR analysis has been used to assess the relationships between aspirin use and risk of lung cancer [25] and between opioid medication and risk of major depressive disorder, anxiety, and stress-related disorder, and cardiovascular disease (CVD) [26, 27]. We conducted a two-sample univariable MR (UVMR) study utilizing summary-level data of 175,936 human subjects from large-scale GWAS studies that included self-reported aspirin use in the UK Biobank [28], as well as AD genetic associations from the International Genomics of Alzheimer’s Project [29]. We conducted multivariable (MVMR) analyses to investigate whether the causal relationship between aspirin use and AD is influenced by pain, inflammation, CVD, or other risk factors. To the best of our knowledge, this is the first large-scale MR analysis investigating potential causal relationships between aspirin use and AD based on genetics.

METHODS

All analyses used publicly available GWAS summary-level data, so ethical approval from an institutional review board was not necessary. Figure 1 presented the conceptual MR design based on three core assumptions: 1) instrumental SNVs have strong associations with the exposure variable; 2) instrumental SNVs do not associate with any confounders between exposure and outcome; 3) instrumental SNVs do not have a direct influence on outcome [22 –24]. As shown in Fig. 1, the path signed with a black cross is blocked by removing horizontally pleiotropic SNVs, which ensures the validity of the instrumental variable assumptions [30] to estimate total effect without confounding bias. The paths signed with red crosses are blocked by MVMR, which estimates direct effect, so that we could conduct mediation analysis by jointly considering direct and total effects [31]. The relationship (distortion or total effect) estimated with UVMR tends to be reduced if MVMR adjusts for confounders or mediators. In this study, removal of horizontally pleiotropic SNVs eliminates any confounding bias in UVMR; therefore, UVMR and MVMR can be combined to conduct mediation analysis [31, 32].

Fig. 1

Overview of the MR analysis for aspirin use and risk of AD.

GWAS data sources

The details of GWAS data sources are listed in Table 1. All data used were publicly available and obtained from European ancestry population. We retrieved the genetic associations of aspirin (acetylsalicylic acid, Anatomical Therapeutic Chemical (ATC) code: N02BA01) use (ASA) from a case-control GWAS performed among United Kingdom Biobank (UKB) study participants [28], which comprised of 61,583 cases and 50,427 controls. A case group included participants who take ASA or medications of the same ATC levels (N02BA01), while a control group included participants who take neither ASA nor medications of the same first two ATC levels [28]. We also used summary statistics of AD from a GWAS meta-analysis of the International Genomics of Alzheimer’s Project (IGAP) [29] stage I. The GWAS was based on a European ancestry meta-analysis of 21,982 cases and 41,944 controls.

Table 1

GWAS Data Sources and Description from Open GWAS (https://gwas.mrcieu.ac.uk) and GWAS Catalog (http://www.ebi.ac.uk/gwas/)

Phenotype	Consortium	First Author	Discovery Sample Description	Ancestry	Download
ASA	N/A	Wu et al. [28]	61,583 cases, 50,427 controls	European	https://www.ebi.ac.uk/gwas/studies/GCST007937
AD	IGAP	Kunkle et al. [29]	21,982 cases, 41,944 controls	European	https://www.ebi.ac.uk/gwas/studies/GCST007511
Chronic Pain (>3 months)
Back	MRC-IEU	Elsworth et al. [36]	80,588 cases, 36,816 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-8463/
Knee	MRC-IEU	Elsworth et al. [36]	76,910 cases, 20,979 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-8906/
Hip	MRC-IEU	Elsworth et al. [36]	40,152 cases, 11,364 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-133/
Neck/shoulder	MRC-IEU	Elsworth et al. [36]	72,887 cases, 32,509 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-16118/
Headache	MRC-IEU	Elsworth et al. [36]	41,719 cases, 49,550 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-13092/
Abdominal/stomach	MRC-IEU	Elsworth et al. [36]	21,711 cases, 17,200 controls	European	https://gwas.mrcieu.ac.uk/datasets/ukb-b-19097/
Inflammation:
CRP levels	the within family consortium	Howe et al. [37]	61,623 individuals	European	https://gwas.mrcieu.ac.uk/datasets/ieu-b-4764/
CVD:
CHD	CARDIoGRAM	Schunkert et al. [38]	22,233 cases, 64,762 controls	European	https://www.ebi.ac.uk/gwas/studies/GCST000998
HF	HERMES	Shah et al. [39]	47,309 cases, 930,014 controls	European	https://www.ebi.ac.uk/gwas/studies/GCST009541
Stroke	MEGASTROKE	Malik et al. [40]	40,585 cases, 406,111 controls	European	https://www.ebi.ac.uk/gwas/studies/GCST006906
Risk factors
SBP	ICBP	Evangelou et al. [41]	757,601 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST006624
DBP	ICBP	Evangelou et al. [41]	757,601 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST006630
LDL cholesterol	GLGC	Willer et al. [42]	94,595 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST002222
HDL cholesterol	GLGC	Willer et al. [42]	94,595 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST002223
Total Cholesterol	GLGC	Willer et al. [42]	94,595 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST002221
Triglycerides	GLGC	Willer et al. [42]	94,595 individuals	European	https://www.ebi.ac.uk/gwas/studies/GCST002216

AD, Alzheimer’s disease; ASA, acetylsalicylic acid; CHD, coronary heart disease; CRP, C-reactive protein; CVD, cardiovascular disease; DBP, diastolic blood pressure; GLGC, Global Lipids Genetics Consortium; HDL, high-density lipoprotein; HF, heart failure; ICBP, International Consortium for Blood Pressure; IGAP, International Consortium for Blood Pressure; LDL, low-density lipoprotein; MRC-IEU, MRC Integrative Epidemiology Unit; SBP, systolic blood pressure.

In general, pain, inflammation, blood pressure, and lipid levels could be considered as risk factors for neurocognitive disorders [33 –35]. To test effect of chronic pain, we used summary-level data from MRC-IEU GWASs for multisite pain (back, knee, hip, neck/shoulder, headache, abdominal/stomach) lasting longer than 3 months [36]. For C-reactive protein (CRP), a sensitive blood biomarker of inflammation, we used summary statistics of serum CRP levels from 61,623 participants included in the within family GWAS consortium [37]. For CVD, we obtained summary statistics for CHD, heart failure (HF), and stroke from several GWAS data resources. The CHD GWAS data were obtained from the Coronary Artery Disease Genome wide Replication and Meta-analysis (CARDIoGRAM) plus the Coronary Artery Disease (C4D) Genetics consortium (CARDIoGRAMplusC4D). To ensure that all participants were of European descent, we used summary data of CARDIoGRAM meta-analysis from 22 GWAS studies [38], which can minimize the potential bias due to population stratification. The summary statistics for HF were retrieved from the Heart Failure Molecular Epidemiology for Therapeutic targets (HERMES) consortium [39]. This GWAS meta-analysis pooled 17 population cohorts and 9 case-control studies. We used summary statistics for stroke (ischemic stroke, intracerebral hemorrhage, and stroke of unknown type) from the MEGASTROKE consortium [40]. The systolic blood pressure and diastolic blood pressure (DBP) traits were defined using a GWAS from the International Consortium for Blood Pressure [41]. These data were from a meta-analysis of 757,601 European participants. Blood lipids including high-density lipoprotein cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, and total cholesterol were defined using a GWAS from the Global Lipids Genetics Consortium [42]. Overall, the ASA GWAS may have used overlapping samples with secondary exposure but did not overlap with the sample used to estimate the SNV-AD (outcome) associations.

Univariable Mendelian randomization

We identified a set of genetic variants associated with ASA (p < 5*10^–6). Then, the PLINK clumping algorithm was employed to conduct linkage disequilibrium clumping based on the 1000 Genomes Project, to obtain independent SNVs. In this study, relaxed thresholds (r ² = 0.1 and window size = 100 kb) were selected, which has been used in previous MR research when few significant SNVs are extracted [43, 44]. A genetic variant was considered sufficient if the corresponding instrument strength quantified with F-statistic was larger than 10. SNV-exposure and SNV-outcome data were harmonized, and palindromic SNVs [45] were removed (Supplementary Table 1). Prior to the analysis, SNVs exhibiting reverse causation were identified and removed with Steiger filtering [46] based on the harmonized data. The MR-PRESSO (MR-pleiotropy residual sum and outlier) global test and outlier test were used for evaluating the presence of horizontal pleiotropy and for outlier removal respectively [47, 48]. The remaining SNVs were utilized as genetic instruments to assess the causal effect of ASA on the risk of developing AD.

For UVMR analysis, we conducted the inverse variance weighted (IVW) method that combines the Wald ratio to determine the causal effect of ASA on AD risk. The IVW method is an efficient and consistent estimate of causal effect if all selected genetic instruments are valid [49]. Moreover, we performed the iterative Mendelian randomization and pleiotropy (IMRP) analysis [50], which could simultaneously search for horizontal pleiotropic variants and estimate the causal effect. We conducted sensitivity analysis to examine any violation of MR assumptions. Cochran’s Q statistic [51] was used to evaluate heterogeneity between single genetic variants that may indicate the presence of invalid instruments. Heterogeneity was considered present if the Cochran Q-test p-value was less than 0.05. Excessive heterogeneity indicates the violation of assumptions for selecting valid SNVs, which would induce horizontal pleiotropy [52]. We also inspected horizontal pleiotropy with the MR-Egger intercept, which can be regarded as the average pleiotropic effect of genetic variants [53]. The MR Steiger directionality test was used to examine whether the direction in our hypothesis is valid [46]. We implemented all UVMR analyses using the TwoSampleMR package in R.

Multivariable Mendelian randomization

MVMR makes causal inference of each exposure on the outcome conditional on other exposures [54]. UVMR and MVMR estimate total and direct effects, respectively, so that combining these two MR analyses can detect if a mediator is present between an exposure and an outcome [31]. For MVMR analysis, we identified a set of genetic variants that were related to either ASA, chronic pain, CRP levels, CVD, blood pressure, or blood lipids at a significance threshold of p < 5*10^–6 or 5*10^–8 (Supplementary Table 2). The PLINK clumping algorithm was executed to extract independent SNVs. Appropriate proxy SNV was selected at a threshold of LD r2 of 0.8, when an instrumental SNV-trait association is available in either primary exposure or secondary exposure. Moreover, overall SNVs were ensured to be independent with an additional clumping step (LD r ² < 0.1, window size > 100 kb). Palindromic SNVs and SNVs exhibiting reverse causation that were identified with Steiger filtering were eliminated. The extended version of the IVW model (MV IVW) [55] in MVMR analysis was utilized as the primary method to estimate causal effect when adjusting for secondary exposure. The generalized version of Cochran’s statistical test was used to inspect potential heterogeneity of genetic variants. MV MR-Egger intercept was used to inspect horizontal pleiotropy [56]. All MVMR analyses were carried out using the MVMR and TwoSampleMR packages in R.

RESULTS

In the present study, 59 SNVs were associated with ASA at a threshold of p < 5*10^–6, among which one (rs77507211) was absent in the AD GWAS summary statistics. No appropriate proxy SNVs at the threshold of LD r ² > 0.8 were identified. After Steiger filtering and removal of palindromic SNVs, 40 SNVs were available for analysis. F-statistics of selected SNVs ranged from 20.9 to 60.7, indicating there are no weak instruments in the MR analysis. MR-PRESSO did not detect any potential pleiotropic outliers.

Aspirin use is causally associated with decreased risk for AD based on UVMR analysis

UVMR analysis revealed that ASA is associated with a lower risk of AD (OR = 0.87; 95%CI = 0.78–0.97, 40 SNVs, Table 2). Cochran’s Q-test (p > 0.05) uncovered no heterogeneity in the IVW model (Table 3). The MR-Egger intercept (p > 0.05) revealed no directional pleiotropy in our analysis. Among 40 SNVs for aspirin use, 3 SNVs (rs1510226, rs1831733, rs583104) are associated with HF, 2 SNVs (rs28752929, rs635634) are associated with DBP, and 6 SNVs (rs117733303, rs1510226, rs4299376, rs4927207, rs583104, rs635634) are associated with LDL cholesterol. After removing potential pleiotropic SNVs related to LDL cholesterol, the association between aspirin use and AD risk remained significant (OR = 0.87, 95%CI = 0.77–0.99; Fig. 2, Tables 2 and 3). While there is a yet-to-confirm relationship between AD and HF (or DBP), our study showed a causal estimate of aspirin use on the decreased risk of AD (removal of HF SNVs: OR = 0.83, 95%CI = 0.74–0.93; removal of DBP SNVs: OR = 0.87, 95%CI = 0.78–0.98; removal of all potentially pleiotropic SNVs: OR = 0.85, 95%CI = 0.75–0.96, Tables 2 and 3). Moreover, leave-one-out analysis suggested the estimate effect of IVW was independent of any single SNV. Consistent results were obtained using a more stringent threshold of LD r ² < 0.01 (Supplementary Table 3). The MR Steiger directionality test showed that the assumption that aspirin use causes AD is valid (Table 3). IMRP provides evidence of no horizonal pleiotropy and significantly protective effect of ASA on risk of AD (Tables 2 and 3). In the frailty analysis (Supplementary Methods 1), the IVW method estimated the mean effect induced by frailty alone (OR = 0.998, 95%CI = 0.92–1.08, Supplementary Figure 1), indicating that frailty effects negligibly influence the true effect estimate (OR = 0.87, 95%CI = 0.77–0.99).

Table 2

The causal estimate of ASA on AD from IVW and IMRP model

Exposure	Methods	#SNVs	OR	95%CI	p
ASA¹	IVW	40	0.87	0.78–0.97	0.012
	IMRP	39	0.85	0.76–0.95	0.003
ASA²	IVW	37	0.83	0.74–0.93	0.001
	IMRP	37	0.83	0.74–0.93	0.001
ASA³	IVW	38	0.87	0.78–0.98	0.02
	IMRP	37	0.85	0.76–0.95	0.006
ASA⁴	IVW	34	0.87	0.77–0.99	0.03
	IMRP	33	0.84	0.75–0.96	0.008
ASA⁵	IVW	32	0.85	0.75–0.96	0.01
	IMRP	32	0.85	0.75–0.96	0.01

ASA¹, All ASAs SNVs; ASA², ASAs SNVs that are not associated with HF; ASA³, ASAs SNVs that are not associated with DBP; ASA⁴, ASAs SNVs that are not associated with LDL cholesterol; ASA⁵, ASAs SNVs that are not associated with HF, DBP, or LDL cholesterol. AD, Alzheimer’s disease; ASA, acetylsalicylic acid; CI, confidence interval; DBP, diastolic blood pressure; HF, heart failure; IMRP, iterative Mendelian randomization and pleiotropy; IVW, inverse variance weighted; LDL, low-density lipoprotein; OR, odds ratio; SNVs, single-nucleotide variants.

Table 3

Sensitivity analyses with different methods

	Heterogeneity	Pleiotropy			Direction
Outcome	Q-statistic (P)	MR Egger intercept (P)	MR-PRESSO p	IMRP p	Steiger test
ASA¹	37.8 (0.53)	0.003 (0.69)	0.546	0.838	True
ASA²	28.5 (0.81)	0.003 (0.73)	0.838	0.85	True
ASA³	37.5 (0.44)	0.004 (0.64)	0.479	0.777	True
ASA⁴	33.9 (0.43)	0.008 (0.39)	0.45	0.782	True
ASA⁵	26.6 (0.69)	0.006 (0.53)	0.708	0.746	True

Fig. 2

Analysis of the causal relationship between ASA and risk of AD.

The protective effects of aspirin use on AD is partially influenced by coronary heart disease, blood pressure and lipid levels based on MVMR analysis

In MVMR, we examined whether the association between ASA and decreased risk of AD was influenced by chronic pain, inflammation, CVD, blood pressure, and blood lipids. Because there is a yet-to-confirm relationship between AD and HF (or DBP), we retained ASA SNVs associated with HF or DBP to conduct the MVMR analysis. When adjusting for the pain- and inflammation-related traits, the genetic causality of ASA on AD remained statistically significant (adjusting for chronic back pain: OR = 0.87, 95%CI = 0.77–0.99; chronic headache: OR = 0.88, 95%CI = 0.78–0.999; CRP levels: OR = 0.84, 95%CI = 0.73–0.96) (Fig. 3, Supplementary Table 4). Potential heterogeneity (p > 0.05 in Cochran’s statistical test) was not detected (Supplementary Table 4). MVMR Egger intercepts suggest no horizontal pleiotropy. The causal estimates between ASA and AD remained statistically significant when adjusting for HF (OR = 0.88, 95%CI = 0.78–0.98) and stroke (OR = 0.87, 95%CI = 0.77–0.99), while the relationship between ASA and AD risk reduction was attenuated when adjusting for CHD (OR = 1.0, 95%CI = 0.85–1.17), blood pressure (OR = 0.95, 95%CI = 0.87–1.04) and blood lipids (OR = 0.89, 95%CI = 0.76–1.03). These results suggest that the protective association between aspirin and AD is partially influenced by CHD, blood pressure, and blood lipids. Again, MV Egger intercepts suggest no horizontal pleiotropy.

Fig. 3

MVMR results of ASA and risk of AD.

DISCUSSION

Our UVMR study suggests a causally protective effect of ASA on the risk of developing AD. These results are consistent with some previous findings [18]. Moreover, MVMR analysis indicated that ASA’s effects on AD might be influenced by CHD, blood pressure, and blood lipids.

This two sample MR analysis have several strengths. MR analysis minimizes bias caused by reverse causality and residual confounding, which are common in observational studies. In addition, all SNV effect estimates in this study were obtained from GWAS studies restricted to populations of European descent, therefore, bias due to population stratification was minimized [49, 57]. We performed rigorous sensitivity analyses to assess any violation of MR assumptions. We identified eight SNVs of rs1510226, rs1831733, rs583104, rs28752929, rs635634, rs117733303, rs4299376, and rs4927207 that were associated with heart failure, diastolic blood pressure, or LDL cholesterol. Growing evidence suggests that patients with CVD are at increased risk of AD and that there is a strong causal association between CVD, cognitive decline, and AD [6]. After removal of eight CVD-associated SNVs, UVMR still showed a significant causal relationship between ASA and AD, suggesting that protective effects of aspirin on AD are not biased. Finally, combined UVMR and MVMR analyses revealed potential mediators underlying the observed causal association between ASA and AD. MVMR results suggest that the effect of ASA on AD risk reduction is independent of pain and inflammation. MVMR analysis further indicated that the causal protective association of ASA on AD might be influenced by CHD, blood pressure, and blood lipid levels. Previous studies have reported that CHD, blood pressure, and blood lipids are associated with a risk of AD [6 , 59]. Clinical trials have also demonstrated the strong cardioprotective properties of aspirin; currently, aspirin is recommended to decrease the risk of adverse cardiovascular events and all-cause mortality in patients with higher cardiovascular risk and lower risk of bleeding [60].

Although the exact causes of AD remain largely unknown, the “amyloid hypothesis” suggests that impaired clearance of amyloid-beta protein plaques is implicated in AD pathogenesis [61]. Thus, activating or boosting the cellular pathways in disposing of waste in the brain could be a promising strategy to slow or delay the progression of AD. Transcription factor EB (TFEB) regulates the expression of the brain’s debris-clearing machinery. A previous study revealed that aspirin use upregulated TFEB, stimulated the production of lysosomes, and decreased amyloid plaque pathology in AD model mice [62].

Epidemiological and clinical trial studies showed inconsistent and often contradictory relationships between aspirin use and risk of AD. Observational studies susceptible to residual confounding and bias. In addition, the protective effect of aspirin use on AD may be limited to specific subgroups of individuals. For example, a previous observational study has suggested aspirin is associated with reduced cognitive decline in patients with stroke or CHD [19, 63]. ASPREE trial that focused on healthy elderly participants [20] may have not detected protective effect of aspirin use on reducing the risk of AD in specific high risk populations. Our study, by leveraging summary genetics from large-scale GWAS studies, provides evidence that aspirin use may be causally associated with decreased risk for AD, partially influenced by CHD, blood pressure, and lipid levels. It would probably mean that aspirin can provide a more significant protective effect in patients at higher risk for CHD, abnormal blood pressure or lipid levels.

This study has several limitations. First, potential dose effects of aspirin use on the risk of AD is not considered in our two-sample MR analysis, however it is important because any benefit needs to be balanced against the risks of excessive bleeding. Moreover, the adverse effects of aspirin change with age, leading to recommendations for cardiovascular prophylaxis that are both risk and age dependent [60, 64]. Future work in stratified populations (e.g., according to ASA dose, age, comorbidities, etc.) is warranted to further investigate the heterogeneity of the effect of ASA on AD in specific sub-populations [65], which could be a basis of clinical studies. Second, to compensate for the lack of SNVs, we selected SNVs related to aspirin use and chronic pain risk at a relaxed threshold (p < 5*10^–6), which is less than the conventional genome-wide significance threshold (p < 5*10^–8). However, selecting SNVs with smaller effect sizes for ASA may increase the potential of weak instrument bias, although we used F-statistics to validate the instrument strength. A relaxed threshold was also selected to perform LD clumping, ensuring instrument independence. Although no horizontal pleiotropy was detected by sensitivity analyses, it remains a possibility that existing pleiotropy caused bias. Third, the ASA GWAS summary statistics from individual GWAS studies may have limited power to detect genetic associations, in which the effect size of a significant SNV-exposure association tends to be overestimated, leading to the winner’s curse phenomenon [66]. Thus, the causal estimate in our study might be biased with the data-driven selection of instrumental variants [67]. Fourth, MR analysis only estimated effect sizes corresponding to a long-term difference. Work is needed to examine the effect size from a short-term intervention [68]. Fifth, the reliability of the employed MR investigation depends on the validity of the instrumental SNVs. In this study, we used several strategies to assess the validity of instruments and remove invalid ones. Future work is necessary to examine how other MR designs may impact the findings in our study, including identifying instrumental SNVs in the genes of the targeted protein to mimic the mechanism of action or performing robustly MR estimates in the presence of invalid instruments. Finally, this study was based on summary statistics of GWAS studies. While there was no sample overlap between exposure and outcome GWAS in UVMR, primary exposure may have overlapping samples with secondary exposure that had potential in distorting the estimated causal relationship of ASA with AD. Therefore, future work is warranted to perform individual-level data MVMR analysis for preventing the sample overlap problem in the two-sample summary data setting.

CONCLUSIONS

This study leveraged 2-sample MR analyses to demonstrate a genetic causality between ASA and decreased risk of AD. We further showed that this causal relationship might be influenced by blood pressure, blood lipid levels, and CHD. Findings from this study provide evidence supporting the benefit of aspirin use in reducing the risk of AD, which may help identify patient populations to curb AD by prevention and treatment strategies.

Footnotes

ACKNOWLEDGMENTS

The authors have no acknowledgments to report.

FUNDING

This work has been supported by the NIH National Institute of Aging (grants nos. AG057557, AG061388, AG062272, AG076649) and the Clinical and Translational Science Collaborative (CTSC) of Cleveland (TR002548).

CONFLICT OF INTEREST

The authors have no conflict of interest to report.

DATA AVAILABILITY

The data supporting the findings of this study are available within the article.

The supplementary material is available in the electronic version of this article: .

References

Alzheimer’s Association (2021) 2021 Alzheimer’s disease facts and figures. Alzheimers Dement 17, 327–406.

Torre

, Gonzalez-Lima

(2021) The FDA approves Aducanumab for Alzheimer’s disease, raising important scientific questions. J Alzheimers Dis 82, 881–882.

Sevigny

, Chiao

, Bussière

, Weinreb

, Williams

, Maier

, Dunstan

, Salloway

, Chen

, Ling

, O’Gorman

, Qian

, Arastu

, Li

, Chollate

, Brennan

, Quintero-Monzon

, Scannevin

, Arnold

, Engber

, Rhodes

, Ferrero

, Hang

, Mikulskis

, Grimm

, Hock

, Nitsch

, Sandrock

(2016) The antibody aducanumab reduces Aβ plaques in Alzheimer’s disease. Nature 537, 50–56.

Walsh

, Merrick

, Milne

, Brayne

(2021) Aducanumab for Alzheimer’s disease? BMJ 374, n1682.

Leng

, Edison

(2021) Neuroinflammation and microglial activation in Alzheimer disease: Where do we go from here? Nat Rev Neurol 17, 157–172.

Stampfer

(2006) Cardiovascular disease and Alzheimer’s disease: Common links. J Intern Med 260, 211–223.

DiNicolantonio

, O’Keefe

, Lavie

(2012) Benefits and risks of aspirin use. JAMA 308, 1088–1089.

Alzheimer’s Drug Discovery Foundation (2015) Aspirin (low-dose) and NSAID. https://www.alzdiscovery.org/uploads/cognitive_vitality_media/Aspirin-low-dose-and-NSAID-Cognitive-Vitality-For-Researchers.pdf. Last updated on March 19, 2015, Accessed on July 1, 2022.

Etminan

, Gill

, Samii

(2003) Effect of non-steroidal anti-inflammatory drugs on risk of Alzheimer’s disease: Systematic review and meta-analysis of observational studies. BMJ 327, 128.

10.

Nilsson

, Johansson

, Takkinen

, Berg

, Zarit

, McClearn

, Melander

(2003) Does aspirin protect against Alzheimer’s dementia? A study in a Swedish population-based sample aged≥80 years. Eur J Clin Pharmacol 59, 313–319.

11.

Szekely

, Green

, Breitner

, Østbye

, Beiser

, Corrada

, Dodge

, Ganguli

, Kawas

, Kuller

, Psaty

, Resnick

, Wolf

, Zonderman

, Welsh-Bohmer

, Zandi

(2008) No advantage of Aβ₄₂-lowering NSAIDs for prevention of Alzheimer dementia in six pooled cohort studies. Neurol 70, 2291–2298.

12.

Kelley

, McClure

, Unverzagt

, Kissela

, Kleindorfer

, Howard

, Wadley

(2015) Regular aspirin use does not reduce risk of cognitive decline. J Am Geriatr Soc 63, 390.

13.

Grodstein

, Skarupski

, Bienias

, Wilson

, Bennett

, Evans

(2008) Anti-inflammatory agents and cognitive decline in a bi-racial population. Neuroepidemiology 30, 45–50.

14.

Arvanitakis

, Grodstein

, Bienias

, Schneider

, Wilson

, Kelly

, Evans

, Bennett

(2008) Relation of NSAIDs to incident AD, change in cognitive function, and AD pathology. Neurology 70, 2219–2225.

15.

Szekely

, Breitner

, Fitzpatrick

, Rea

, Psaty

, Kuller

, Zandi

(2008) NSAID use and dementia risk in the Cardiovascular Health Study*: Role of APOE and NSAID type. Neurology 70, 17–24.

16.

Waldstein

, Wendell

, Seliger

, Ferrucci

, Metter

, Zonderman

(2010) Nonsteroidal anti-inflammatory drugs, aspirin, and cognitive function in the Baltimore longitudinal study of aging. J Am Geriatr Soc 58, 38–43.

17.

Mandas

, Mereu

, Catte

, Saba

, Serchisu

, Costaggiu

, Peiretti

, Caminiti

, Vinci

, Casu

, Piludu

, Fossarello

, Manconi

, Dessi

(2014) Cognitive impairment and age-related vision disorders: Their possible relationship and the evaluation of the use of aspirin and statins in a 65 years-and-over Sardinian population. Front Aging Neurosci 6, 309.

18.

Côté

, Carmichael

, Verreault

, Lindsay

, Lefebvre

, Laurin

(2012) Nonsteroidal anti-inflammatory drug use and the risk of cognitive impairment and Alzheimer’s disease. Alzheimers Dement 8, 219–226.

19.

Nguyen

TNM

, Chen

, Trares

, Stocker

, Holleczek

, Beyreuther

, Brenner

, Schoettker

(2021) Long-term low-dose acetylsalicylic use shows protective potential for the development of both vascular dementia and Alzheimer’s disease in patients with coronary heart disease but not in other individuals from the general population: Results from two large cohort studies. Alzheimers Res Ther 14, 75.

20.

Ryan

, Storey

, Murray

, Woods

, Wolfe

, Reid

, Nelson

, Chong

, Williamson

, Ward

, Lockery

, Orchard

, Trevaks

, Kirpach

, Newman

, Ernst

, McNeil

, Shah

(2020) Randomized placebo-controlled trial of the effects of aspirin on dementia and cognitive decline, Neurology 95, e320–e331.

21.

Gorenflo

, Davis

, Kendall

, Olaker

, Kaelber

, Xu

(2022) Association of aspirin use with reduced risk of developing Alzheimer’s disease in elderly ischemic stroke patients: A retrospective cohort study. J Alzheimers Dis 91, 1–8.

22.

Smith

, Ebrahim

(2003) ‘Mendelian randomization’: Can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 32, 1–22.

23.

Smith

, Hemani

(2014) Mendelian randomization: Genetic anchors for causal inference in epidemiological studies, Hum Mol Genet 23, R89–R98.

24.

Sheehan

, Didelez

, Burton

, Tobin

(2008) Mendelian randomisation and causal inference in observational epidemiology, PLoS Med 5, e177.

25.

Jiang

, Su

, Li

, Wang

, Wen

, Liang

, He

, Liang

(2021) Association between the use of aspirin and risk of lung cancer: Results from pooled cohorts and Mendelian randomization analyses. J Cancer Res Clin Oncol 147, 139–151.

26.

Rosoff

, Smith

, Lohoff

(2021) Prescription opioid use and risk for major depressive disorder and anxiety and stress-related disorders: A multivariable mendelian randomization analysis. JAMA Psychiatry 78, 151–160.

27.

Cai

, He

, Wang

, Rong

, Chen

, Shen

, Li

, Peng

(2022) Genetic liability for prescription opioid use and risk of cardiovascular diseases: A multivariable Mendelian randomization study. Addict 117, 1382–1391.

28.

, Byrne

, Zheng

, Kemper

, Yengo

, Mallett

, Yang

, Visscher

, Wray

(2019) Genome-wide association study of medication-use and associated disease in the UK Biobank. Nat Commun 10, 1891.

29.

Kunkle

, Grenier-Boley

, Sims

, Bis

, Damotte

, Naj

, Boland

, Vronskaya

, Van Der Lee

, Amlie-Wolf

, et al (2019) Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet 51, 414–430.

30.

Burgess

, Smith

, Davies

, Dudbridge

, Gill

, Glymour

, Hartwig

, Holmes

, Minelli

, Relton

, Theodoratou

(2020) Guidelines for performing Mendelian randomization investigations. Wellcome Open Res 4, 186.

31.

Sanderson

(2021) Multivariable Mendelian randomization and mediation, Cold Spring Harb Perspect Med 11, a038984.

32.

Rijnhart

, Lamp

, Valente

, MacKinnon

, Twisk

, Heymans

(2021) Mediation analysis methods used in observational research: A scoping review and recommendations. BMC Med Res Methodol 21, 1–17.

33.

Higgins

, Martin

, Baker

, Vasterling

, Risbrough

(2018) The relationship between chronic pain and neurocognitive function: A systematic review. Clin J Pain 34, 262.

34.

Albu

, Padureanu

, Boldeanu

, Bumbea

, Enescu

, Albulescu

, Silosi

, Enescu

(2018) Vascular neurocognitive disorders and the vascular risk factors. J Mind Medl Sci 5, 7–15.

35.

Salis

, Loddo

, Zanda

, Peralta

, Serchisu

, Mandas

(2022) Comprehensive Geriatric Assessment: Application and correlations in a real-life cross-sectional study. Front Med 9, 984046.

36.

Elsworth

, Lyon

, Alexander

, Liu

, Matthews

, Hallett

, Bates

, Palmer

, Haberland

, Smith

, Zheng

, Haycock

, Gaunt

, Hemani

The MRC IEU OpenGWAS data infrastructure. BioRxiv. doi: https://doi.org/10.1101/2020.08.10.244293.

37.

Howe

, Nivard

, Morris

, Hansen

, Rasheed

, Cho

, Chittoor

, Lind

, Palviainen

, Zee

, Cheesman

, Mangino

, Wang

, Li

, Klaric

, Ratliff

, Bielak

, Nygaard

, Giannelis

, Willoughby

, Reynolds

, Balbona

, Andreassen

, Ask

, Baras

, Bauer

, Boomsma

, Campbell

, Chen

, Christofidou

, Corfield

, Dahm

, Dokuru

, Evans

, Geus

EJC

, Giddaluru

, Gordon

, Harden

, Hill

, Hughes

, Kerr

, Kim

, Kweon

, Latvala

, Lawlor

, Li

, Lin

, Magnus

, Magnusson

PKE

, Mallard

, Martikainen

, Mills

, Njølstad

, Overton

, Pedersen

, Porteous

, Reid

, Silventoinen

, Southey

, Stoltenberg

, Tucker-Drob

, Wright

, Social Science Genetic Association Consortium, Within Family Consortium, Hewitt

, Keller

, Stallings

, Lee

, Christensen

, Kardia

SLR

, Peyser

, Smith

, Wilson

, Hopper

, Hägg

, Spector

, Pingault

, Plomin

, Havdahl

, Bartels

, Martin

, Oskarsson

, Justice

, Millwood

, Hveem

, Naess

, Willer

, Asvold

, Koellinger

, Kaprio

, Medland

, Walters

, Benjamin

, Turley

, Evans

, Smith

, Hayward

, Brumpton

, Hemani

, Davies

(2022) Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. Nat Genet 54, 581–592.

38.

Schunkert

, König

, Kathiresan

, Reilly

, Assimes

, Holm

, Preuss

, Stewart

, Barbalic

, Gieger

, Absher

, Aherrahrou

, Allayee

, Altshuler

, Anand

, Andersen

, Anderson

, Ardissino

, Ball

, Balmforth

, Barnes

, Becker

, Berger

, Bis

, Boekholdt

, Boerwinkle

, Braund

, Brown

, Burnett

, Buysschaert

, Cardiogenics , Carlquist

, Chen

, Cichon

, Codd

, Davies

, Dedoussis

, Dehghan

, Demissie

, Devaney

, Diemert

, Do

, Doering

, Eifert

, Mokhtari

NEE

, Ellis

, Elosua

, Engert

, Epstein

, Faire

, Fischer

, Folsom

, Freyer

, Gigante

, Girelli

, Gretarsdottir

, Gudnason

, Gulcher

, Halperin

, Hammond

, Hazen

, Hofman

, Horne

, Illig

, Iribarren

, Jones

, Jukema

, Kaiser

, Kaplan

, Kastelein

JJP

, Khaw

, Knowles

, Kolovou

, Kong

, Laaksonen

, Lambrechts

, Leander

, Lettre

, Li

, Lieb

, Loley

, Lotery

, Mannucci

, Maouche

, Martinelli

, McKeown

, Meisinger

, Meitinger

, Melander

, Merlini

, Mooser

, Morgan

, Mühleisen

, Muhlestein

, Münzel

, Musunuru

, Nahrstaedt

, Nelson

, Nöthen

, Olivieri

, Patel

, Patterson

, Peters

, Peyvandi

, Qu

, Quyyumi

, Rader

, Rallidis

, Rice

, Rosendaal

, Rubin

, Salomaa

, Sampietro

, Sandhu

, Schadt

, Schäfer

, Schillert

, Schreiber

, Schrezenmeir

, Schwartz

, Siscovick

, Sivananthan

, Sivapalaratnam

, Smith

, Snoep

, Soranzo

, Spertus

, Stark

, Stirrups

, Stoll

, Tang

WHW

, Tennstedt

, Thorgeirsson

, Thorleifsson

, Tomaszewski

, Uitterlinden

, Rij

, Voight

, Wareham

, Wells

, Wichmann

, Wild

, Willenborg

, Witteman

JCM

, Wright

, Ye

, Zeller

, Ziegler

, Cambien

, Goodall

, Cupples

, Quertermous

, März

, Hengstenberg

, Blankenberg

, Ouwehand

, Hall

, Deloukas

, Thompson

, Stefansson

, Roberts

, Thorsteinsdottir

, O’Donnell

, McPherson

, Erdmann

, the CARDIoGRAM Consortium (2011) Large-scale association analysis identifies 13 new susceptibility loci for coronary artery disease. Nat Genet 43, 333–338.

39.

Shah

, Henry

, Roselli

, Lin

, Sveinbjörnsson

, Fatemifar

, Hedman

, Wilk

, Morley

, Chaffin

, Helgadottir

, Verweij

, Dehghan

, Almgren

, Andersson

, Aragam

, Ärnlöv

, Backman

, Biggs

, Bloom

, Brandimarto

, Brown

, Buckbinder

, Carey

, Chasman

, Chen

, Chung

, Chutkow

, Cook

, Delgado

, Denaxas

, Doney

, Dörr

, Dudley

, Dunn

, Engström

, Esko

, Felix

, Finan

, Ford

, Ghanbari

, Ghasemi

, Giedraitis

, Giulianini

, Gottdiener

, Gross

, Guðbjartsson

, Gutmann

, Haggerty

, Harst

, Hyde

, Ingelsson

, Jukema

, Kavousi

, Khaw

, Kleber

, Køber

, Koekemoer

, Langenberg

, Lind

, Lindgren

, London

, Lotta

, Lovering

, Luan

, Magnusson

, Mahajan

, Margulies

, März

, Melander

, Mordi

, Morgan

, Morris

, Morrison

, Nagle

, Nelson

, Niessner

, Niiranen

, O’Donoghue

, Owens

, Palmer

CNA

, Parry

, Perola

, Portilla-Fernandez

, Psaty

, Regeneron Genetics Center, Rice

, Ridker

, Romaine

SPR

, Rotter

, Salo

, Salomaa

, Setten

, Shalaby

, Smelser

, Smith

, Stender

, Stott

, Svensson

, Tammesoo

, Taylor

, Teder-Laving

, Teumer

, Thorgeirsson

, Thorsteinsdottir

, Torp-Pedersen

, Trompet

, Tyl

, Uitterlinden

, Veluchamy

, Völker

, Voors

, Wang

, Wareham

, Waterworth

, Weeke

, Weiss

, Wiggins

, Xing

, Yerges-Armstrong

, Yu

, Zannad

, Zhao

, Hemingway

, Samani

, McMurray

JJV

, Yang

, Visscher

, Newton-Cheh

, Malarstig

, Holm

, Lubitz

, Sattar

, Holmes

, Cappola

, Asselbergs

, Hingorani

, Kuchenbaecker

, Ellinor

, Lang

, Stefansson

, Smith

, Vasan

, Swerdlow

, Lumbers

(2020) Genome-wide association and Mendelian randomisation analysis provide insights into the pathogenesis of heart failure. Nat Commun 11, 163.

40.

Malik

, Chauhan

, Traylor

, Sargurupremraj

, Okada

, Mishra

, Rutten-Jacobs

, Giese

, Van Der Laan

, Gretarsdottir

, Anderson

, Chong

, Adams

, Ago

, Almgren

, Amouyel

, Ay

, Bartz

, Benavente

, Bevan

, Boncoraglio

, Brown

, Butterworth

, Carrera

, Carty

, Chasman

, Chen

, Cole

, Correa

, Cotlarciuc

, Cruchaga

, Danesh

, Bakker

PIW

, DeStefano

, Hoed

, Duan

, Engelter

, Falcone

, Gottesman

, Grewal

, Gudnason

, Gustafsson

, Haessler

, Harris

, Hassan

, Havulinna

, Heckbert

, Holliday

, Howard

, Hsu

, Hyacinth

, Ikram

, Ingelsson

, Irvin

, Jian

, Jiménez-Conde

, Johnson

, Jukema

, Kanai

, Keene

, Kissela

, Kleindorfer

, Kooperberg

, Kubo

, Lange

, Langefeld

, Langenberg

, Launer

, Lee

, Lemmens

, Leys

, Lewis

, Lin

, Lindgren

, Lorentzen

, Magnusson

, Maguire

, Manichaikul

, McArdle

, Meschia

, Mitchell

, Mosley

, Nalls

, Ninomiya

, O’Donnell

, Psaty

, Pulit

, Rannikmäe

, Reiner

, Rexrode

, Rice

, Rich

, Ridker

, Rost

, Rothwell

, Rotter

, Rundek

, Sacco

, Sakaue

, Sale

, Salomaa

, Sapkota

, Schmidt

, Schminke

, Sharma

, Slowik

, Sudlow

CLM

, Tanislav

, Tatlisumak

, Taylor

, Thijs

VNS

, Thorleifsson

, Thorsteinsdottir

, Tiedt

, Trompet

, Tzourio

, Duijn

, Walters

, Wareham

, Wassertheil-Smoller

, Wilson

, Wiggins

, Yang

, Yusuf

, AFGen Consortium, Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium, International Genomics of Blood Pressure (iGEN-BP) Consortium, INVENT Consortium, STARNET, Bis

, Pastinen

, Ruusalepp

, Schadt

, Koplev

, Björkegren

JLM

, Codoni

, Civelek

, Smith

, Trégouët

, Christophersen

, Roselli

, Lubitz

, Ellinor

, Tai

, Kooner

, Kato

, He

, Harst

, Elliott

, Chambers

, Takeuchi

, Johnson

, BioBank Japan Cooperative Hospital Group, COMPASS Consortium, EPIC-CVD Consortium, EPIC-InterAct Consortium, International Stroke Genetics Consortium (ISGC),METASTROKE Consortium, Neurology Working Group of the CHARGE Consortium, NINDS Stroke Genetics Network (SiGN), UK Young Lacunar DNA Study, MEGASTROKE Consortium, Sanghera

, Melander

, Jern

, Strbian

, Fernandez-Cadenas

, Longstreth

, Rolfs

, Hata

, Woo

, Rosand

, Pare

, Hopewell

, Saleheen

, Stefansson

, Worrall

, Kittner

, Seshadri

, Fornage

, Markus

, Howson

JMM

, Kamatani

, Debette

, Dichgans

(2018) Multiancestry genome-wide association study of 520,000 subjects identifies 32 loci associated with stroke and stroke subtypes. Nat Genet 50, 524–537.

41.

Evangelou

, Warren

, Mosen-Ansorena

, Mifsud

, Pazoki

, Gao

, Ntritsos

, Dimou

, Cabrera

, Karaman

, et al. (2018) Genetic analysis of over 1 million people identifies 535 new loci associated with blood pressure traits. Nat Genet 50, 1412–1425.

42.

Global Lipids Genetics Consortium. (2013) Discovery and refinement of loci associated with lipid levels. Nat Genet 45, 1274.

43.

Vaucher

, Keating

, Lasserre

, Gan

, Lyall

, Ward

, Smith

, Pell

, Sattar

, Paré

, Holmes

(2018) Cannabis use and risk of schizophrenia: A Mendelian randomization study. Mol Psychiatry 23, 1287–1292.

44.

Papiol

, Schmitt

, Maurus

, Rossner

, Schulze

, Falkai

(2021) Association between physical activity and schizophrenia: Results of a 2-sample Mendelian randomization analysis. JAMA Psychiatry 78, 441–444.

45.

Burgess

, Davies

, Thompson

(2016) Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol 40, 597–608.

46.

Hemani

, Tilling

, Smith

(2017) Orienting the causal relationship between imprecisely measured traits using GWAS summary data, PLoS Genet 13, e1007081.

47.

Verbanck

, Chen

C-y

, Neale

, Do

(2018) Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet 50, 693–698.

48.

Zhu

(2020) Mendelian randomization and pleiotropy analysis. Quant Biol 10, 1–11.

49.

Hemani

, Bowden

, Smith

(2018) Evaluating the potential role of pleiotropy in Mendelian randomization studies, Hum Mol Genet 27, R195–R208.

50.

Zhu

, Li

, Xu

, Wang

(2021) An iterative approach to detect pleiotropy and perform Mendelian Randomization analysis using GWAS summary statistics. Bioinform 37, 1390–1400.

51.

Greco

M FD

, Minelli

, Sheehan

, Thompson

(2015) Detecting pleiotropy in Mendelian randomisation studies with summary data and a continuous outcome. Stat Med 34, 2926–2940.

52.

Bowden

, Greco

M FD

, Minelli

, Zhao

, Lawlor

, Sheehan

, Thompson

, Smith

(2019) Improving the accuracy of two-sample summary-data Mendelian randomization: Moving beyond the NOME assumption. Int J Epidemiol 48, 728–742.

53.

Burgess

, Thompson

(2017) Interpreting findings from Mendelian randomization using the MR-Egger method. Eur J Epidemiol 32, 377–389.

54.

Sanderson

, Smith

, Windmeijer

, Bowden

(2019) An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. Int J Epidemiol 48, 713–727.

55.

Burgess

, Thompson

(2015) Multivariable Mendelian randomization: The use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol 181, 251–260.

56.

Rees

, Wood

, Burgess

(2017) Extending the MR-Egger method for multivariable Mendelian randomization to correct for both measured and unmeasured pleiotropy. Stat Med 36, 4705–4718.

57.

Bowden

, Smith

, Haycock

, Burgess

(2016) Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol 40, 304–314.

58.

DeCarli

(2021) The link between blood pressure and Alzheimer’s disease. Lancet Neurol 20, 878–879.

59.

Iwagami

, Qizilbash

, Gregson

, Douglas

, Johnson

, Pearce

, Evans

, Pocock

(2021) Blood cholesterol and risk of dementia in more than 1· 8 million people over two decades: A retrospective cohort study, Lancet Healthy Longev 2, e498–e506.

60.

Mogul

, Leppien

, Laughlin

, Spinler

(2021) Aspirin for primary prevention of cardiovascular disease: A review of recent literature and updated guideline recommendations. Expert Opin Pharmacother 22, 83–91.

61.

Hardy

, Selkoe

(2002) The amyloid hypothesis of Alzheimer’s disease: Progress and problems on the road to therapeutics. Science 297, 353–356.

62.

Chandra

, Jana

, Pahan

(2018) Aspirin induces lysosomal biogenesis and attenuates amyloid plaque pathology in a mouse model of Alzheimer’s disease via PPARα. J Neurosci 38, 6682–6699.

63.

Douiri

, McKevitt

, Emmett

, Rudd

, Wolfe

CDA

(2013) Long-term effects of secondary prevention on cognitive function in stroke patients. Circulation 128, 1341–1348.

64.

Patrono

, Baigent

(2019) Role of aspirin in primary prevention of cardiovascular disease. Nat Rev Cardiol 16, 675–686.

65.

Coscia

, Gill

, Benítez

, Pérez

, Malats

, Burgess

(2022) Avoiding collider bias in Mendelian randomization when performing stratified analyses. Eur J Epidemiol 37, 1–12.

66.

Dudbridge

, Newcombe

(2019) Replication and metaanalysis of genome-wide association studies. Handbook of Statistical Genomics: Two Volume Set, pp. 631–650.

67.

Burgess

, Butterworth

, Thompson

(2013) Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol 37, 658–665.

68.

Davies

, Holmes

, Smith

(2018) Reading Mendelian randomisation studies: A guide, glossary, and checklist for clinicians, BMJ 362, k601.