Educational attainment reduces the risk of suicide attempt among individuals with and without psychiatric disorders independent of cognition: a bidirectional and multivariable Mendelian randomization study with more than 815,000 participants

Suicide is a leading cause of death with approximately 800,000 deaths per year worldwide. In the United States, the age-adjusted suicide rate increased 33% from 1999 to 2017, with more than 120 deaths by suicide each day. Beyond mortality, suicidal behavior imposes substantial morbidity and economic burden, with an estimated 2.1 million discharges for self-inflicted injuries in 2013 and $508 billion in combined medical expenditures and lost productivity. Identifying causal risk factors is essential for developing effective prevention strategies. Observational studies suggest that lower educational attainment (EA) is associated with higher risk of suicidal behavior and may contribute to socioeconomic disparities in mortality. However, EA is tightly correlated with cognitive performance (CP), which is also linked to suicidal behavior, complicating causal inference. Observational designs are vulnerable to residual confounding, reverse causation, and the influence of psychiatric symptoms and disorders, as well as behavioral factors such as smoking and alcohol use. Randomized trials to assess the effects of EA or CP on suicide risk are unethical or infeasible. Mendelian randomization (MR) leverages randomly allocated genetic variants as instrumental variables to strengthen causal inference. Suicidal behavior has a polygenic, partly heritable basis that is partly distinct from psychiatric disorders, and EA and CP are also heritable and polygenic. Using large GWAS datasets, this study applies bidirectional and multivariable MR to disentangle the causal relationships among EA, CP, and suicide attempt risk, while additionally accounting for socioeconomic, psychiatric, and behavioral factors.

Prior observational literature consistently links low educational attainment with increased suicidal behavior across the life course and indicates strong associations between cognitive performance, EA, and suicidality. However, these associations are potentially confounded by correlated socioeconomic status indicators, cultural factors, psychiatric disorders and symptoms, and behaviors such as smoking and alcohol use. The genetic architecture of suicidal behavior is polygenic and, in part, distinct from that of psychiatric disorders, with genome-wide significant loci identified for suicide attempts in both general and psychiatric cohorts. Mendelian randomization studies have begun to probe causal relationships between modifiable risk factors (e.g., smoking) and suicide. EA and CP themselves are polygenic and heritable, with substantial genetic correlations between CP and years of schooling. Despite these findings, the causal directions and independent effects of EA and CP on suicide attempt risk had remained unclear prior to this study.

Study design: A bidirectional and multivariable Mendelian randomization (MR) framework was used to assess causal relationships between educational attainment (EA), cognitive performance (CP), and suicide attempt risk, and to test potential reverse causation from suicide attempts to EA or CP. Single-variable MR (SMR) and multivariable MR (MVMR) models were applied, with multiple sensitivity analyses to address pleiotropy. Additional MVMR models simultaneously incorporated socioeconomic, psychiatric, and behavioral factors as potential confounders/mediators. Data sources: Publicly available summary statistics were obtained from large GWAS consortia and pipelines: Social Science Genetic Association Consortium (SSGAC), iPSYCH, and the MRC-IEU UK Biobank GWAS Pipeline. Participants were predominantly white individuals of European ancestry; total combined sample exceeded 815,000. Exposures and outcome: - EA: SSGAC GWAS meta-analysis (N ≤ 766,345; predominantly European ancestry; UK Biobank included), measured as years-of-education equivalent mapped across cohorts; genetic associations reported in SD units. - CP: Combined SSGAC meta-analysis of COGENT (N=135,298) and UK Biobank (N=222,543), total N ≈ 257,841; CP defined via principal components of neuropsychological tests (COGENT) and a standardized fluid intelligence score (UK Biobank). SNP heritability ≈ 0.209; genetic correlation with education rG ≈ 0.662. - Outcome: Hospital-based records of suicide attempts (N ≤ 50,264). Instrument selection and clumping: - For EA and CP, SNPs reaching genome-wide significance (p < 5×10⁻⁸) were selected and pruned for independence at LD r² ≤ 0.001. - Additional instruments were extracted for potential confounders/mediators: household income (UK Biobank data-field 738; N=397,751), ever smoker (UK Biobank data-field 2016-00; N=461,006), alcohol units consumed per week (SSGAC, N=112,117), self-reported visits to a GP or psychiatrist for nerves, anxiety, tension, or depression (UK Biobank; N≈460k), psychotic experiences, and psychiatric disorders: major depressive disorder (MDD), schizophrenia (SCZ), and bipolar disorder (BD), with genome-wide significant SNPs pruned at LD r² ≤ 0.001. Statistical analysis: - SMR analyses estimated total effects of EA and CP on suicide attempt risk using inverse-variance weighted (IVW) MR, with complementary sensitivity methods (MR Egger, weighted median, weighted mode, MR-PRESSO) to assess robustness to horizontal pleiotropy and outliers. - Bidirectional MR tested effects of suicide attempt liability on EA and CP. - MVMR estimated direct effects of EA and CP jointly (e.g., effect of EA holding CP constant), using IVW MVMR, MR Egger MVMR, and MR-PRESSO extensions. - Extended MVMR models included EA, CP, and additional variables (household income, smoking, alcohol consumption, psychiatric symptoms/visits, psychotic experiences, MDD, SCZ, BD) simultaneously to account for potential confounding/mediation. - Harmonization procedures aligned effect alleles across datasets; standard errors and 95% CIs were computed; genome-wide significant thresholds and rigorous QC from source GWAS were applied. Ethical considerations: All contributing studies had institutional review board approvals and informed consent; only publicly available summary data were used.

- Single-variable MR (SMR): Both EA and CP appeared protective against suicide attempt risk. For EA, the odds ratio (OR) per SD increase in EA was reported with a 95% CI of 0.412–0.666 and P = 1.70 × 10⁻⁷ (abstract lists an OR value that appears discordant with the CI; direction indicates reduced risk). For CP, the OR per SD increase in CP was 0.714 (95% CI: 0.572–0.885; P = 0.020). - Bidirectional MR: No evidence that genetic liability to suicide attempt causally affects EA or CP. - Multivariable MR (EA and CP jointly): EA showed a strong independent protective effect on suicide attempt risk (OR 0.342; 95% CI: 0.206–0.568; P = 1.61 × 10⁻⁴) when adjusting for CP, while CP showed no independent effect (OR 1.182; 95% CI: 0.842–1.659; P = 0.333). - Extended MVMR adjusting simultaneously for behavioral and psychiatric factors (tobacco smoking, alcohol consumption, and self-reported nerves/tension/anxiety/depression): The protective effect of EA persisted with little attenuation (OR 0.541; 95% CI: 0.421–0.696; P = 3.33 × 10⁻¹⁰). - Robustness: Results were consistent across complementary MR methods (IVW, MR Egger, weighted median/mode, MR-PRESSO), supporting robustness to pleiotropy and strengthening causal inference. - Population: Findings pertain to white individuals of European ancestry drawn from large-scale GWAS (>815,000 combined across datasets).

The study set out to determine whether educational attainment (EA) and/or cognitive performance (CP) have causal effects on suicide attempt risk and to disentangle their potentially correlated influences. The MR results indicate that higher EA confers a causal reduction in suicide attempt risk, independent of CP. CP alone did not exhibit an independent protective effect once EA was accounted for. Bidirectional analyses argue against reverse causality from suicide attempt liability to reductions in EA or CP. The persistence of EA’s protective effect after simultaneously adjusting for behavioral (smoking, alcohol use) and psychiatric (self-reported anxiety/depression, and major psychiatric disorders) factors suggests that EA itself may be a key upstream determinant of suicide attempt risk rather than merely a proxy for these correlates. Consistency across multiple MR approaches that handle pleiotropy differently supports the robustness of the causal interpretation. These findings highlight EA as a potentially impactful lever for suicide prevention strategies and public health policy, reinforcing the importance of educational opportunities and attainment as part of broader efforts to reduce suicidal behavior. Nonetheless, applicability beyond white individuals of European ancestry remains to be tested.

Using large-scale bidirectional and multivariable Mendelian randomization, the study provides evidence that educational attainment causally reduces the risk of suicide attempts, while cognitive performance does not have an independent effect when EA is considered. The protective association of EA remains robust even after accounting for psychiatric disorders and behavioral mediators. These findings have potential implications for policy and interventions that enhance educational opportunities as part of suicide prevention strategies. Future research should evaluate generalizability across diverse ancestral populations and further elucidate pathways linking EA to reduced suicidality, including sociocultural and economic mechanisms.

- Generalizability: Analyses were limited to white individuals of European ancestry; results may not generalize to other populations. - Measurement heterogeneity: EA was harmonized across cohorts via imputed years-of-education equivalents, which may introduce phenotype heterogeneity. - MR assumptions: Although multiple sensitivity analyses were used, MR relies on assumptions (e.g., no horizontal pleiotropy affecting the outcome except via the exposure) that cannot be fully verified. - Summary-level data: Use of GWAS summary statistics precluded stratified analyses and exploration of non-linear effects or effect modification. - Outcome definition: Suicide attempt cases were based on hospital records (primary or secondary diagnoses), which may miss non-hospitalized attempts and introduce misclassification.