Education and stroke: evidence from epidemiology and Mendelian randomization study

Stroke is the second leading cause of death and disability worldwide. Identifying modifiable risk factors is crucial to reduce morbidity, mortality, and economic burden. Observational evidence suggests that lower educational attainment is associated with higher stroke risk, but such associations may be confounded and cannot establish causality, and randomized trials of education are not feasible. Mendelian randomization (MR) uses genetic variants as instruments to infer causal effects of modifiable exposures on outcomes, minimizing confounding and reverse causation. Prior MR work indicated education may protect against ischemic stroke independent of cognition, but the prospective association over long follow-up and causal relationships, especially for hemorrhagic stroke, remain unclear. This study aims to define the nature and magnitude of the association between education and incident stroke in the ARIC cohort and to test for causal effects of education on total, ischemic, and hemorrhagic stroke using two-sample MR.

Epidemiological studies have repeatedly linked low education with increased stroke risk across settings and sexes. Meta-analytic evidence reports that having fewer than 11 years of education versus 11 or more is associated with about one-third higher stroke risk. Large cohorts (e.g., >250,000 participants, mean follow-up 4.7 years) also found higher stroke risk in those with lower education. MR studies have suggested that educational attainment protects against coronary heart disease and stroke, with mediation through body mass index, systolic blood pressure, and smoking. However, heterogeneity by stroke subtype is plausible given distinct pathophysiology, and prior evidence on hemorrhagic stroke is limited. The present study extends this literature with long-term prospective data from ARIC and two-sample MR analyses by stroke subtype.

Design and population: The ARIC Study is a population-based prospective cohort from four U.S. communities, enrolling 15,792 adults aged 45–64 years at visit 1 (1987–1989), with subsequent visits in 1990–1992, 1993–1995, 1996–1998, and 2011–2013, plus ongoing telephone follow-up and hospital surveillance. For this analysis, exclusions were missing stroke follow-up (n=279), prior stroke (n=266), prior CHD or heart failure (n=1062), missing education (n=15), and missing other covariates (n=2661), yielding 11,509 participants. Exposure: Education was self-reported at visit 1 as highest grade completed and categorized as basic (less than high school), intermediate (high school degree or vocational school), and advanced (attending or completed college/professional school). Outcomes: Incident stroke (total, ischemic, hemorrhagic) from visit 1 through 2014 was ascertained via annual/semiannual interviews, clinic visits, and community hospital surveillance. Stroke was defined as rapid onset focal neurological deficit lasting >24 hours or until death without a non-stroke cause. Hospitalization events were identified using ICD-9 codes 430–438, cerebrovascular keywords, or imaging evidence. Subtyping followed established ARIC procedures. Covariates: Baseline (visit 1) covariates included age, sex, race, smoking, drinking, family income; measured BMI; seated blood pressure (random-zero sphygmomanometer); physical activity (Baecke questionnaire); diabetes (based on glucose thresholds, medication use, or self-report); lipids (TC, HDL-c, triglycerides; LDL-c via Friedewald); glucose; platelet count; creatinine (modified kinetic Jaffe). Medication use (statin, aspirin, anticoagulants) was recorded. Statistical analysis (cohort): Baseline characteristics were compared across education levels using ANOVA, χ2, or Kruskal–Wallis tests as appropriate. Kaplan–Meier curves and log-rank tests compared cumulative incidence by education. Cox proportional hazards models estimated hazard ratios (HRs) for incident total stroke and by subtype, with follow-up from visit 1 to event, loss, death, or 12/31/2014. Model 1 adjusted for age, race, sex; Model 2 additionally for diabetes, creatinine, HDL-c, LDL-c, triglycerides, glucose, platelet count, statin, aspirin, anticoagulants; Model 3 further for income. Prespecified subgroup analyses (sex, age, race, smoking, drinking, BMI, creatinine) assessed heterogeneity; interactions were tested via likelihood ratio tests, considering P>0.1 as similar across subgroups. Mendelian randomization: Two-sample MR was conducted using MR-Base with summary GWAS data. Instruments were SNPs associated with education at genome-wide significance (P<5×10−8) from SSGAC (European ancestry, n=293,723). Outcomes: total stroke (MRC-IEU UKB-b:6813, European, n=463,010), ischemic stroke (ISGC, mixed ancestry, n=29,633), hemorrhagic stroke (MRC-IEU UKB-b:4538, mixed ancestry, n=463,010). For each SNP, effect allele, frequency, beta, SE, and P-value were extracted, ensuring quality (heterogeneity P<0.001, HWE P<0.001, imputation info/r2>0.90). Harmonization aligned effect alleles; palindromic SNPs were resolved using allele frequencies. Proxy SNPs were not used. Primary causal estimates used inverse-variance weighted (IVW) meta-analysis; sensitivity analyses included weighted median, weighted mode, and MR-Egger regression to assess pleiotropy, plus leave-one-out analyses and funnel plots. Instrument strength was summarized by F-statistic and variance explained; ORs were scaled per 1 SD (3.6 years) higher education. Bonferroni thresholds for SNP-level multiple testing were P<0.0009 (58 SNPs) or P<0.0007 (70 SNPs); overall MR analyses by outcome considered P<0.05 statistically significant. Ethics approval was obtained; procedures followed the Declaration of Helsinki; participants provided informed consent.

- Cohort and events: Among 11,509 ARIC participants (median follow-up 25.3 years), 915 incident strokes occurred (8.0%). Baseline lower education was associated with higher prevalence of Black race, current smoking, higher BMI, SBP, and DBP, among others. - Total stroke (Model 3, adjusted for demographics, risk factors, and income): Compared with basic education, intermediate education HR 0.88 (95% CI 0.74–1.04; P=0.135) and advanced education HR 0.75 (0.62–0.91), indicating 25% lower rate with advanced education. - Ischemic stroke (Model 3): Intermediate education HR 0.82 (0.69–0.98; P=0.032); advanced education HR 0.73 (0.60–0.90; P=0.003). - Hemorrhagic stroke (Model 3): Intermediate education HR 1.23 (0.73–2.06; P=0.439); advanced education HR 0.98 (0.54–1.80; P=0.955); no significant associations. - Subgroups: Associations were broadly consistent across sex, race, smoking, drinking, and creatinine strata (all P-interactions >0.1). Stronger protective associations were observed for BMI 24–28 kg/m2 (intermediate HR 0.68 [0.52–0.90]; advanced HR 0.57 [0.42–0.78]; P-interaction=0.042) and for age <60 years (intermediate HR 0.80 [0.66–0.98]; advanced HR 0.67 [0.54–0.84]; P-interaction=0.016). - MR instrument: 70 education-associated SNPs explained 1.04% of variance; mean F-statistic 68, indicating strong instruments. - MR total stroke: No causal effect detected (IVW OR 0.999, 95% CI 0.998–1.000; P=0.166). No heterogeneity (P=0.616 MR-Egger; P=0.621 IVW) or horizontal pleiotropy (MR-Egger intercept P=0.368). Sensitivity analyses were concordant; leave-one-out showed no influential SNPs. - MR hemorrhagic stroke: No causal effect (IVW OR 1.000; P=0.811); no heterogeneity or pleiotropy; sensitivity analyses consistent. - MR ischemic stroke: Evidence for a protective causal effect of higher education (IVW OR 0.764, 95% CI 0.585–0.998; P=0.048). No heterogeneity (P=0.917 MR-Egger; P=0.923 IVW) and no pleiotropy (MR-Egger intercept P=0.751). Weighted median/mode/Egger were directionally similar but not statistically significant.

The study addressed whether educational attainment is associated with and causally related to stroke risk. In a well-characterized, long-term cohort (ARIC), higher education was associated with lower incidence of total and ischemic stroke after extensive adjustment, but not with hemorrhagic stroke. Two-sample MR indicated that genetically proxied higher education likely reduces ischemic stroke risk, while showing no evidence for a causal effect on total or hemorrhagic stroke. These findings support a causal pathway from education to ischemic stroke specifically, consistent with distinct etiologies of ischemic versus hemorrhagic stroke. Mechanistically, education is linked to healthier behaviors, safer occupations, better healthcare access, and favorable cardiovascular risk profiles; mediation by BMI, blood pressure, and smoking is plausible and supported by prior MR mediation analyses. The lack of effect for hemorrhagic stroke aligns with divergent relationships between lipids and stroke subtypes and potential opposite effects of intensive LDL-C lowering. Subgroup analyses suggest stronger protective associations among individuals with BMI 24–28 kg/m2 and those younger than 60 years, indicating potential effect modification by adiposity and age. Overall, the results reinforce education as a social determinant with meaningful impact on ischemic stroke risk and provide causal evidence that complements observational associations.

Higher educational attainment is associated with lower incidence of total and ischemic stroke in the ARIC cohort, with no association for hemorrhagic stroke. MR analyses suggest a protective causal effect of education on ischemic stroke, but not on total or hemorrhagic stroke. These findings emphasize the public health importance of educational policies and interventions as upstream strategies for ischemic stroke prevention. Future research should use larger and more diverse GWAS datasets, investigate mediation pathways (e.g., blood pressure, BMI, smoking), and evaluate potential effect modifiers to inform targeted prevention.

- Residual confounding: Important biomarkers (e.g., inflammatory markers such as BNP, hsCRP) were not measured at ARIC visit 1 and could not be adjusted for. - Outcome ascertainment: ARIC stroke surveillance may miss transient ischemic attacks and some events, potentially underestimating incidence. - MR data sources: MR analyses relied on available MR-Base datasets (SSGAC for education; MRC-IEU/ISGC for stroke), which may limit power and scope; larger GWAS are needed for confirmation. - Ancestry/generalizability: SSGAC and MRC-IEU datasets are primarily European; findings may not generalize to other ethnic/racial groups. - Some MR sensitivity methods lacked statistical significance, and the ischemic stroke MR finding, while significant by IVW, was near the threshold and should be interpreted cautiously.