Hostname: page-component-5db58dd55d-h5th4 Total loading time: 0 Render date: 2026-06-01T07:41:27.943Z Has data issue: false hasContentIssue false
  • English
  • Français

Educational attainment and mental health outcomes: A within-sibship Mendelian randomization study

Published online by Cambridge University Press:  26 September 2025

María Fernanda Vinueza Veloz Open the ORCID record for María Fernanda Vinueza Veloz [Opens in a new window] ,
Laxmi Bhatta ,
Paul R. Jones ,
Martin Tesli ,
George Davey Smith ,
Neil M. Davies ,
Ben M. Brumpton  and
Øyvind E. Næss
María Fernanda Vinueza Veloz*
Affiliation:
Department of Community Medicine and Global Health, Institute of Health and Society, Faculty of Medicine, University of Oslo, Oslo, Norway
Laxmi Bhatta
Affiliation:
K.G. Jebsen Center for Genetic Epidemiology, Department of Public Health and Nursing, Norwegian University of Science and Technology – NTNU, Trondheim, Norway FIU-PH, Division of Mental Health Care, St Olavs Hospital, Trondheim, Norway
Paul R. Jones
Affiliation:
Department of Community Medicine and Global Health, Institute of Health and Society, Faculty of Medicine, University of Oslo, Oslo, Norway
Martin Tesli
Affiliation:
Department of Mental Disorders, Norwegian Institute of Public Health, Oslo, Norway Centre for Research and Education in Forensic Psychiatry, Department of Mental Health and Addiction, Oslo University Hospital, Oslo, Norway Department of Adult Psychiatry, Diakonhjemmet Hospital, Oslo, Norway Department of Psychiatry, Østfold Hospital, Grålum, Norway
George Davey Smith
Affiliation:
MRC Integrative Epidemiology Unit (IEU), University of Bristol, Bristol, UK
Neil M. Davies
Affiliation:
K.G. Jebsen Center for Genetic Epidemiology, Department of Public Health and Nursing, Norwegian University of Science and Technology – NTNU, Trondheim, Norway Division of Psychiatry, University College London, London, UK Department of Statistical Sciences, University College London, London, UK
Ben M. Brumpton
Affiliation:
K.G. Jebsen Center for Genetic Epidemiology, Department of Public Health and Nursing, Norwegian University of Science and Technology – NTNU, Trondheim, Norway HUNT Research Center, Department of Public and Nursing, Norwegian University of Science and Technology – NTNU, Trondheim, Norway Clinic of Medicine, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Øyvind E. Næss
Affiliation:
Department of Community Medicine and Global Health, Institute of Health and Society, Faculty of Medicine, University of Oslo, Oslo, Norway Department Chronic diseases, Norwegian Institute of Public Health, Oslo, Norway
*
Corresponding author: María Fernanda Vinueza Veloz; Email: m.f.v.veloz@medisin.uio.no
Rights & Permissions [Opens in a new window]

Abstract

Background

Observational studies indicate that higher educational attainment (EA) is associated with a lower risk of many mental health conditions (MHC). We assessed to what extent this association is influenced by genetic nurture and demographic factors (i.e., assortative mating and population structure).

Methods

We conducted a within-sibship Mendelian randomization (MR) study. The sample consisted of 61 880 siblings (27 507 sibships) from the Trøndelag Health Study-HUNT (Norway) and UK Biobank (United Kingdom). MHC outcomes included symptom scores for anxiety, depression, and neuroticism, measured using the Hospital Anxiety and Depression Scale, the 7-item Generalized Anxiety Disorder Scale, the 9-item Patient Health Questionnaire, and the Eysenck Personality Questionnaire, along with self-reported psychotropic medication use.

Results

One standard deviation (SD) increase in liability to EA was associated with lower anxiety (−0.20 SD [95% CI: −0.26, −0.14]), depression (−0.11 SD [−0.43, −0.22]), and neuroticism scores (−0.30 SD [−0.53, −0.06]), as well as lower odds of psychotropic medication use (OR: 0.60 [0.52, 0.69]). Within-sibship MR estimates remained consistent with population-based estimates: anxiety (−0.17 SD [−0.33, −0.00]); depression (−0.18 SD [−1.26, 0.89]); neuroticism (−0.29 SD [−0.43, −0.15]); psychotropic medication use (OR, 0.52 [0.34, 0.82]).

Conclusions

Higher EA or genetic liability to education reduces symptoms of anxiety, neuroticism, and psychotropic medication use. These mental health benefits do not seem to be explained by EA-linked genetic nurture or demographic factors. Regarding depression, results were less conclusive due to imprecise estimates, though beneficial effects of genetic liability to higher EA are possible and warrant further investigation.

Keywords

anxietydepressioneducational attainmentMendelian randomizationneuroticismpsychotropic medication

Information

Type
Original Article
Information
Psychological Medicine , Volume 55 , 2025 , e282
Check for updates
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press

Introduction

Mental health conditions (MHC), which are defined as mental disorders, psychosocial disabilities, and mental states associated with significant distress, impaired functioning, or risk of self-harm, represent a major global health burden (World Health Organization, 2022). Common MHC such as anxiety and mood disorders (e.g., depression) are among the leading causes of years lived with disability across all age groups (Abbafati et al., Reference Abbafati, Abbas, Abbasi-Kangevari, Abd-Allah, Abdelalim, Abdollahi, Abdollahpour, Abegaz, Abolhassani, Aboyans, Abreu, Abrigo, Abualhasan, Abu-Raddad, Abushouk, Adabi, Adekanmbi, Adeoye, Adetokunboh and Amini2020). Furthermore, individuals with MHC face mortality rates at least twice as high as those of the general population or individuals without these conditions (Walker et al., Reference Walker, McGee and Druss2015).

Socioeconomic factors, including those related to education, are strongly implicated in the risk of developing MHC (Kivimäki et al., Reference Kivimäki, Batty, Pentti, Shipley, Sipilä, Nyberg, Suominen, Oksanen, Stenholm, Virtanen, Marmot, Singh-Manoux, Brunner, Lindbohm, Ferrie and Vahtera2020). Higher educational attainment (EA) is associated with a reduced risk of developing MHC such as anxiety, depression, personality disorders, and substance abuse (Bjelland et al., Reference Bjelland, Krokstad, Mykletun, Dahl, Tell and Tambs2008; Cohen et al., Reference Cohen, Nussbaum, Weintraub, Nichols and Yen2020; Erickson et al., Reference Erickson, El-Gabalawy, Palitsky, Patten, Mackenzie, Stein and Sareen2016). This inverse relationship may be attributed to greater access to mental and physical resources in adulthood, which help educated individuals cope with adversity and mitigate the development of MHC (Niemeyer et al., Reference Niemeyer, Bieda, Michalak, Schneider and Margraf2019).

Mendelian randomization (MR) is a method that uses genetic variants or polygenic scores (PGS) as instrumental variables to investigate relationships between exposures and outcomes. MR relies on the premise that there is a causal pathway from an individual’s genotype to the individual’s phenotype (e.g., a genotype–phenotype association). MR is robust to exposure-outcome confounding and reverse causation because it leverages the principle that genetic variants are randomly assigned at conception, and do not change throughout life (Davies et al., Reference Davies, Holmes and Smith2018; Evans & Davey Smith, Reference Evans and Davey Smith2015; Smith & Ebrahim, Reference Smith and Ebrahim2003).

Valid causal inference using MR requires that three core instrumental variable assumptions hold: (i) The genetic variants must be associated to the exposure of interest and this relationship is required to be reasonably strong (relevance). (ii) There must be no uncontrolled common causes of the genetic variants and the outcome (independence). (iii) The genetic variants must influence the outcome solely through the exposure of interest, which implies that the effect of the genetic variants on the outcome is fully mediated by the exposure (exclusion) (Walker et al., Reference Walker, Sanderson, Levin, Damraurer, Feeney and Davies2024).

Two recent population-based MR studies support the role of EA in the etiology of common MHC, including anxiety and depression (Jones et al., Reference Jones, Wootton, Gill, Carter, Gunnell, Munafò and Sallis2021; Yuan et al., Reference Yuan, Xiong, Michaëlsson, Michaëlsson and Larsson2021). These estimates may however conflate the effects of EA with genetic nurture – a phenomenon where parental genotypes influence offspring outcomes through environmental pathways, independent of the child’s own genetic makeup (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020; Kong et al., Reference Kong, Thorleifsson, Frigge, Vilhjalmsson, Young, Thorgeirsson, Benonisdottir, Oddsson, Halldorsson, Masson, Gudbjartsson, Helgason, Bjornsdottir, Thorsteinsdottir and Stefansson2018; Morris et al., Reference Morris, Davies, Hemani and Smith2020). For instance, parents with a genetic predisposition to higher EA may provide resources (e.g., cognitive stimulation, financial stability), which promote their children’s educational success and mental well-being.

Additionally, population-based MR estimates are likely biased by demographic factors (e.g., assortative mating and population structure) (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020). Bias related to assortative mating (i.e., non-random matching between reproductive partners) can arise when assortment leads to a genetic correlation between parents, which induces spurious genotype–phenotype associations in the offspring (Border et al., Reference Border, Athanasiadis, Buil, Schork, Cai, Young, Werge, Flint, Kendler, Sankararaman, Dahl and Zaitlen2022; Hartwig et al., Reference Hartwig, Davies and Davey Smith2018). Bias related to population structure (i.e., the presence of systematic differences in allele frequencies between subpopulations) can occur when ancestry is correlated with both the genotype (e.g., EA genotype) and the phenotype (e.g., MHC) (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020).

The objective of the present study is to evaluate the extent to which the associations between EA and MHC, specifically symptoms of anxiety, depression, and neuroticism, along with psychotropic medication use, are accounted for by genetic nurture and demographic factors. To address this, we applied a within-sibship MR design, which accounts for genetic nurture and reduces bias from assortative mating and population structure.

Methodology

Study design and data sources

The present is the report of a within-sibship MR study. MR assumptions and how they have been addressed are summarized in eTable 1. One-sample and two-sample MR methods were applied using individual-level data, and summary statistics from genome wide association studies (GWAS). Individual-level data came from the Trøndelag Health Study (HUNT) and UK Biobank (UKB) (Bycroft et al., Reference Bycroft, Freeman, Petkova, Band, Elliott, Sharp, Motyer, Vukcevic, Delaneau, O’Connell, Cortes, Welsh, Young, Effingham, McVean, Leslie, Allen, Donnelly and Marchini2018; Olav et al., Reference Olav, Svold, Langhammer, Rehn, Kjelvik, Grøntvedt, Sørgjerd, Fenstad, Heggland, Holmen, Stuifbergen, Aalberg Vikjord, Brumpton, Vard, Skjellegrind, Thingstad, Sund, Selbaek, Mork and Krokstad2022). A brief description of contributing GWAS can be found in eTable 2.

Setting and participants

HUNT is a population-based cohort study that is held in the Trøndelag County in Norway and started in 1984 (Olav et al., Reference Olav, Svold, Langhammer, Rehn, Kjelvik, Grøntvedt, Sørgjerd, Fenstad, Heggland, Holmen, Stuifbergen, Aalberg Vikjord, Brumpton, Vard, Skjellegrind, Thingstad, Sund, Selbaek, Mork and Krokstad2022). We used data from all participants of the second (HUNT2) and third HUNT wave (HUNT3) who had been genotyped. From those, we selected all individuals who were > 30 years of age when they participated in the survey, and had at least one sibling. The final sample from HUNT2 and HUNT3 included 26 770 (10 428 sibships) and 16 718 siblings (7010 sibships), respectively (eTable 3).

The UKB is a prospective cohort study that began in 2006. UKB is following nearly 500 000, 40- to 69-year-old participants from across the UK, who volunteered to be part of the study and provided consent for follow-up through linkage to their health records (Bycroft et al., Reference Bycroft, Freeman, Petkova, Band, Elliott, Sharp, Motyer, Vukcevic, Delaneau, O’Connell, Cortes, Welsh, Young, Effingham, McVean, Leslie, Allen, Donnelly and Marchini2018). We included all participants who had been genotyped. After restricting the sample to sibships with two or more individuals, our analysis sample included 35 118 participants from 17 079 sibships (eTable 3).

Genetic variants

For HUNT participants, we used a weighted PGS as an instrumental variable for EA (PGS-edu). The calculation of the PGS-edu was based on the genetic variants reported as significantly associated with years of education at the genome-wide level (p < 5 × 10−8) in a recent GWAS (Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati, Sidorenko, Kweon, Goldman, Gjorgjieva, Jiang, Hicks, Tian, Hinds, Ahlskog, Magnusson, Oskarsson, Hayward, Campbell and Young2022). From the 3952 independent genetic variants reported by Okbay et al. (pairwise r2 = 0.1, no physical distance cut-off), we included those that were well imputed in the target population (eTable 4) (Choi et al., Reference Choi, Mak and O’Reilly2020). For further information, see eMethods.

Exposure and outcomes

EA was the exposure of interest and was assessed through the question: For HUNT participants: ‘What is your highest level of education?’ For UKB participants ‘What qualifications do you have?’ Number of years of education was assigned for each of the answers based on The International Standard Classification of Education (ISCED) mapping 1997 (eTable 5). Symptoms of anxiety, depression and neuroticism, as well as the use of psychotropic medication, were the outcomes of interest. In HUNT, symptoms of anxiety and depression were assessed by the Hospital Anxiety and Depression Scale (HADS), and in UKB by the 7-item Generalized Anxiety Disorder Scale (GAD-7) and 9-item Patient Health Questionnaire (PHQ-9) (Kroenke et al., Reference Kroenke, Spitzer and Williams2001; Spitzer et al., Reference Spitzer, Kroenke, Williams and Löwe2006; Zigmond & Snaith, Reference Zigmond and Snaith1983). In both HUNT and UKB, neuroticism was measured using the Eysenck Personality Questionnaire and use of psychotropic medication was self-reported (Eysenck & Tambs, Reference Eysenck and Tambs1990). A detailed description of how the exposure and outcomes were processed can be found in eMethods.

Ethics approval and informed consent

The study protocol was approved by the Regional Committees for Medical Research Ethics South East (REK 2017/2479) and Mid-Norway (REK 2015/1197). All participants signed informed consent for participation and the use of data in research. UK Biobank obtained ethics approval from the North West Multi-centre Research Ethics Committee and obtained informed consent from all study participants.

Statistical analysis

Before running the analyses using individual-level data, we standardized all numerical variables so that they had a mean of 0 and a standard deviation (SD) of 1. EA, as well as symptoms of anxiety, depression, and neuroticism, were analyzed as continuous, while use of psychotropic medication was analyzed as categorical (yes/no). We estimated the association of EA and the outcomes using one-sample MR (two-stage least squares regression), and ordinary least squares (OLS) or logistic regression for comparison.

All models were adjusted by sex and age. However, when the PGS-edu was included as a predictor, the model was also adjusted by the first 10 principal components of ancestry (PCA) to account for population structure and genotyping batch. In all cases, we assumed that standard errors were correlated within sibships and therefore clustered standard errors were computed using the “vcov = cluster” command. OLS, logistic, and one-sample MR analyses were performed using the ‘feols’ (for continuous) and ‘felgm’ (for categorical) functions of the ‘fixest’ package in R (Bergé, Reference Bergé2018; R Core Team, 2021).

Any difference between families due to genetic nurture, assortative mating, and population structure was accounted for by using family fixed effects (anxiety, depression, and neuroticism) or the sibling difference method (psychotropic medication use) (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020). All analyses were performed using the “fixest” package in R (see eMethods for a detailed description) (Bergé, Reference Bergé2018).

Individual-level data for the two cohorts were analyzed separately using the same model specifications and R packages. Then, results were meta-analyzed using the ‘rma’ function from the ‘metafor’ R package (Viechtbauer, Reference Viechtbauer2010). We applied fixed effect models, except when heterogeneity between HUNT and UK Biobank estimates was detected, that is Cochran’s Q Chi2 < 0.05 and I2 > 50% (eTable 6).

Two-sample MR analyses were performed using summary statistics from GWAS described in eTable 2 and the R package ‘TwoSampleMR’ (Hemani et al., Reference Hemani, Zheng, Elsworth, Wade, Haberland, Baird, Laurin, Burgess, Bowden, Langdon, Tan, Yarmolinsky, Shihab, Timpson, Evans, Relton, Martin, Davey Smith, Gaunt and Haycock2018). The inverse-variance estimator weighted (IVW) estimator and its 95% confidence intervals (CI) are reported in the present work. Pleiotropy robust estimators, including weighted median, weighted mode, and MR-Egger, were used to investigate pleiotropy and reported in Supplementary Tables. The directionality of the effect was evaluated using the Steiger test of directionality. The MR-Egger intercept test was performed to assess pleiotropy. We only had access to summary statistics from within-sibship GWAS meta-analysis for depressive symptoms and neuroticism to conduct within-sibship two-sample MR (eTable 2) (Howe et al., Reference Howe, Nivard, Morris, Hansen, Rasheed, Cho, Chittoor, Lind, Palviainen, van der Zee, Cheesman, Mangino, Wang, Li, Klaric, Ratliff, Bielak, Nygaard, Reynolds and Davies2021).

Handling of missing information

We imputed missing data for various questions of the HADS score, education, and psychotropic medication use in HUNT (see eTable 7 for details on missing information). We applied multivariate imputation to each HUNT survey, using the conditional specification implemented by the MICE algorithm (van Buuren & Groothuis-Oudshoorn, Reference van Buuren and Groothuis-Oudshoorn2011). For each imputed data set, we then calculated the corresponding score. For further information, see eMethods.

Results

Descriptive statistics

In both cohorts, participation rates were higher among females than among males. HUNT participants were younger and had slightly lower mean education years than UKB participants (13 versus 14 years). HUNT participants more frequently reported symptoms of anxiety than UKB participants (8% versus 3%). About 5% of HUNT and UKB participants reported symptoms of depression (Table 1). Psychotropic medication use was slightly higher among HUNT than UKB participants (9% versus 8%). Differences between females and males can be seen in Table 1. For each cohort, the total number of siblings varied depending on the outcome (eTable 2).

Table 1.

General characteristics of HUNT and UKB samples

Note: In HUNT symptoms of anxiety and depression were assessed by the Hospital Anxiety and Depression Scale (HADS), and in UK Biobank by the 7-item Generalized Anxiety Disorder Scale (GAD-7) and 9-item Patient Health Questionnaire (PHQ-9), respectively. Neuroticism was assessed using a six-item and a 12-item Eysenck Personality Questionnaire in HUNT and UK Biobank, respectively (see eMethods).

Abbreviations and symbology: n, number; %, percentage; IQR, interquartile range; SD, standard deviation; Psych. med., psychotropic medication usage; a, assessed in 4863 participants (n female = 2959 (60.85%), n male = 1904 (39.15%)); b, assessed in 16 718 participants (n female = 8965 (53.62%), n male = 7753 (46.38%)); c, assessed in 23 852 participants (n female = 13 664 (57.29%), n male = 10 188 (42.71%)).

The PGS-edu was associated with years of education, conditional from age, sex, first 10 PCA, and batch in both cohorts. In HUNT, one SD increase in the PGS-edu was associated with a 0.19 SD (~0.55 years) increase in years of education (95%CI, 0.18: 0.20, p = 4.40×10−187, F-test stat. = 162.99, r2 = 0.03). In UK Biobank, each SD increase of the PGS-edu was associated with a 0.24 SD (~1.22 years) increase in years of education (95%CI, 0.23: 0.24, p = 2.20×10−16, F-test stat. = 103.90, r2 = 0.06). This association was attenuated after including a family fixed effect (HUNT: 0.13, 95%CI: 0.11–0.15, p = 3.47×10−44, F-test stat. = 1349.62, r2 = 0.64; UK Biobank: 0.13, 95%CI, 0.11: 0.14, p = 3.38×10−43, F-test stat. = 2362.40, r2 = 0.33). The associations between the PGS-edu and the outcomes are depicted in eTable 8.

Main analyses

The direction of the regression and MR estimates was consistent across both cohorts and all analyses. However, there were some differences in association strength between HUNT and UKB (Table 2). Differences were more pronounced for the depression and neuroticism MR estimates. There was little evidence of weak instrument bias, as all F-stats were higher than 10 (eTable 9).

Table 2.

Association between educational attainment and mental health outcomes

Note: HUNT and UK Biobank data were analyzed separately and results then meta-analyzed (see Methodology).

Abbreviations and symbology: B, coefficient; SE, standard error; LCI, low 95% confidence interval; UCI, upper 95% confidence interval; p, p value; n, number; OLS, ordinary least squares regression; LOG, logistic regression; EA, educational attainment; FE, within-sibship adjustment; PGS-edu, educational attainment polygenic score; 1SMR, one-sample Mendelian randomization; Psych. med., psychotropic medication usage; *, coefficient and confidence intervals were exponentiated and hence odd ratios are presented.

The results of the population-based MR analyses indicate that a genetic liability to higher EA has protective effects on anxiety, neuroticism, and psychotropic medication use (Figures 1, 3, and 4). A one SD increase in genetic liability to EA was associated with reduced anxiety (−0.20 SD [−0.26, −0.14]), neuroticism scores (−0.30 SD [−0.53, −0.06]), and lowered the odds of psychotropic medication use (odds ratio (OR): 0.60 [0.52, 0.69]). For depression, MR estimates included the null (−0.11 SD [−0.43, 0.22]) (Figure 2). The one-sample MR estimates were larger than the phenotypic regression estimates, except for depression, but were consistent in direction (Table 2).

Figure 1.

Educational attainment and symptoms of anxiety. Standard deviation (SD) changes in the anxiety score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’.

Figure 2.

Educational attainment and symptoms of depression. Standard deviation (SD) changes in the depression score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’; 2SMR, ‘two-sample Mendelian randomization’.

Figure 3.

Educational attainment and neuroticism. Standard deviation (SD) changes in the neuroticism score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’; 2SMR, ‘two-sample Mendelian randomization’.

Figure 4.

Educational attainment and use of psychotropic medication. Log odds changes in psychotropic medication use and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for logistic regression (LOG) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; LOG EA, ‘logistic regression model with educational attainment as exposure’; LOG PGS-edu, ‘logistic regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’.

The within-sibship MR estimates were generally consistent with the population-based MR estimates (Table 2 and Figures 1–4). Although some changes were observed, within-sibship estimates overlapped with population-based MR estimates (Figures 1–4). Specifically, the effect of the genetic liability to higher EA on anxiety changed from −20 SD to −0.17 SD [−0.33, −0.00], on depression from −11 SD to −18 SD [−1.26, 0.89], and on neuroticism from −0.30 SD to −0.29 SD [−0.43, −0.15]. The OR for psychotropic medication use decreased from 0.60 to 0.52 [0.34, 0.82].

Two-sample MR estimates for depression and neuroticism were consistent in direction with those from one-sample MR (eTable9). A one SD increase in genetic liability to EA reduced depression and neuroticism scores by (−0.22 [−0.25, −0.18]) and (−0.21 [−0.25, −0.17]), respectively. Also, the weighted median, weighted mode, and MR-Egger estimates were consistent with IVW estimates. In both cases, the directionality test indicated that the causal direction is likely correct and that horizontal pleiotropy is unlikely (eTable 10). Within-sibship two-sample MR estimates were consistent with population-based estimates but were less precise, as the depression estimate including the null hypothesis.

Discussion

Our phenotypic and population-based one-sample MR analyses indicated that a genetic liability to higher EA reduces symptoms of anxiety and neuroticism and lowers psychotropic medication use. Complementary population-based two-sample MR analyses further indicated that a genetic liability to higher EA reduces depression symptoms. When accounting for genetic nurture and demographic factors, effect estimates changed slightly and became less precise but remained consistent with the population-based estimates. This implies that these associations are unlikely to be attributable to genetic nurture linked to EA or demographic factors. Overall, these findings suggest that higher EA (its genetic liability or some other closely related traits) reduces the risk of developing these MHC in adulthood.

In line with our results, previous research found little evidence for genetic nurture linked to parental EA influencing children’s depression, anxiety, and attention-deficit hyperactivity symptoms (Hughes et al., Reference Hughes, Torvik, van Bergen, Hannigan, Corfield, Andreassen, Ystrom, Ask, Smith, Davies and Havdahl2024). Hughes et al. reported that children’s PGS-Edu were negatively associated with these traits, independent of genetic nurture, implying that liability to higher EA may protect against MHC. In contrast, another study using the same cohort identified broad parental genetic nurture effects on depression (i.e., not specifically related to EA), which were partially mediated via parental anxiety and depression symptoms (Cheesman et al., Reference Cheesman, Eilertsen, Ahmadzadeh, Gjerde, Hannigan, Havdahl, Young, Eley, Njølstad, Magnus, Andreassen, Ystrom and McAdams2020). These results suggest that while genetic nurture linked to EA likely does not contribute to MHC risk, genetic nurture tied to other parental traits (e.g., mental health) could still play a role.

While the magnitude of change in effect estimates varied across the examined MHC, it was generally modest. Moreover, all within-sibship MR effect estimates displayed wide confidence intervals that overlapped with the population-based MR estimates. The degree of imprecision was most pronounced for depression, for which within-sibship MR estimates also overlapped with the null hypothesis. Given this uncertainty, we cannot exclude a potential effect of EA on adult depression symptoms. In contrast, a recent study using a similar approach reported complete attenuation of the effects of EA on depression, anxiety, and neuroticism (van de Weijer et al., Reference van de Weijer, Demange, Pelt, Bartels and Nivard2024). However, like our findings with respect to depression, their imprecise confidence intervals preclude definitive conclusions about the influence of EA on depression.

Importantly, in MR studies on EA, genetic instruments serve as proxies for liability to EA, rather than direct measures of educational duration. Consequently, MR estimates are unlikely to reflect the pure effect of an additional year of schooling (Howe et al., Reference Howe, Tudball, Davey Smith and Davies2022). That is, the genetic liability to higher EA likely operates through both measured EA (e.g., postgraduate qualifications) and other closely related traits. For instance, the PGS-edu likely influences mental health not only by increasing the likelihood of education achievement, but also by shaping unmeasured characteristics like personality or cognitive ability (Krapohl et al., Reference Krapohl, Rimfeld, Shakeshaft, Trzaskowski, McMillan, Pingault, Asbury, Harlaar, Kovas, Dale and Plomin2014). Furthermore, the phenotypic expression of the genetic liability to higher EA depends on social, historical, and cultural contexts, implying that its impact on the risk of developing MHC may differ across populations (Border et al., Reference Border, Athanasiadis, Buil, Schork, Cai, Young, Werge, Flint, Kendler, Sankararaman, Dahl and Zaitlen2022; Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati, Sidorenko, Kweon, Goldman, Gjorgjieva, Jiang, Hicks, Tian, Hinds, Ahlskog, Magnusson, Oskarsson, Hayward, Campbell and Young2022; Rutherford et al., Reference Rutherford, Muthukrishna, Tropf, Mills, Zietsch, Verweij and Visscher2025).

Strengths and limitations

A major strength of our work is that we applied robust methods to evaluate the impact of genetic nurture and demographic factors on the association between EA and the studied MHC (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020; Howe et al., Reference Howe, Nivard, Morris, Hansen, Rasheed, Cho, Chittoor, Lind, Palviainen, van der Zee, Cheesman, Mangino, Wang, Li, Klaric, Ratliff, Bielak, Nygaard, Reynolds and Davies2021). As genetic variants are randomly assigned at conception, MR studies are potentially less susceptible to bias from confounding or reverse causation. Within-family MR addresses key limitations of population-based MR, such as violations of the independence assumption (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020). Importantly, our results were replicated in two cohorts and across different analyses: phenotypical and MR estimates were consistent in both cohorts. Nevertheless, our approach has limitations that must be considered when interpreting our findings.

Limited precision hinders definitive conclusions about whether the effect of EA on depression is explained by genetic nurture and demographic factors. This limitation likely stems from four methodological constraints. First, cohort heterogeneity between the HUNT and UKB, including instrument differences (e.g., HADS versus PHQ-9) and population characteristics (e.g., age, socioeconomic contexts), increases heterogeneity in effect estimates. Second, self-reported depression symptoms increase the likelihood of measurement error, which reduces statistical power and attenuates effect sizes. Third, while MR estimates within cohorts are robust to individual-level confounding, between-cohort heterogeneity in pooled estimates may reflect either true population differences or methodological constraints. With only two cohorts, we cannot distinguish these scenarios. Fourth, given the episodic nature of depression, which is strongly influenced by short-lived environmental effects, single time-point assessments may attenuate the observed associations between EA and depression (Kendler & Gardner, Reference Kendler and Gardner2017).

While previous evidence supports the use of self-reported medication as an alternative or supplementary phenotype for anxiety and depression in genetic studies, an important implementation limitation warrants consideration (Skelton et al., Reference Skelton, Rayner, Purves, Coleman, Gaspar, Glanville, Hunjan, Hübel, Breen and Eley2021). Factors such as socioeconomic barriers to healthcare access, stigma, and regional variations in prescribing practices may introduce measurement error. This can manifest as both under-ascertainment of true medication need among individuals with MHC and misclassification of case status, thereby potentially attenuating effect estimates toward the null. Despite this limitation, we observed consistent protective effects associated with a genetic liability to higher EA across both psychotropic medication use and symptom severity measures. This convergence across multiple proxies strengthens the hypothesis that genetic liability to higher EA reduces the risk of developing these MHC in adulthood.

While within-sibship MR designs are robust to genetic nurture and demographic factors, biases may persist due to unmeasured family-level environmental influences (Brumpton et al., Reference Brumpton, Sanderson, Heilbron, Hartwig, Harrison, Vie, Cho, Howe, Hughes, Boomsma, Havdahl, Hopper, Neale, Nivard, Pedersen, Reynolds, Tucker-Drob, Grotzinger, Howe and Davies2020; Howe et al., Reference Howe, Nivard, Morris, Hansen, Rasheed, Cho, Chittoor, Ahlskog, Lind, Palviainen, van der Zee, Cheesman, Mangino, Wang, Li, Klaric, Ratliff, Bielak, Nygaard and Davies2022). Indirect genetic effects between siblings, such as the influence of sibling genotypes on the shared environment, could explain our results (Howe et al., Reference Howe, Evans, Hemani, Davey Smith and Davies2022). For example, an older sibling’s academic success raises parental expectations for younger siblings, independent of the younger siblings’ genetics. Unmeasured family-level environmental factors, such as differential parental treatment (e.g., unequal resource allocation between siblings in response to inherited variants), could further confound results (Sjölander et al., Reference Sjölander, Frisell and Öberg2022). While sibling influences (including birth order dynamics) could potentially distort within-sibship MR estimates, their magnitude is expected to be smaller than major confounding pathways addressed by our design (Demange et al., Reference Demange, Hottenga, Abdellaoui, Eilertsen, Malanchini, Domingue, Armstrong-Carter, de Zeeuw, Rimfeld, Boomsma, van Bergen, Breen, Nivard and Cheesman2022).

Other limitations include possible weak instrument bias. In the general population, the PGS-edu’s explanatory power (R2) was relatively small (~3%), which is expected to attenuate further in within-sibship analyses due to reduced within-family variation (Sjölander et al., Reference Sjölander, Frisell and Öberg2022). Because instrument strength depends on both R2 and sample size, the instrument’s effective strength is often diminished in within-sibship analyses. Although our F-statistics exceeded conventional thresholds for instrument strength (F > 10, see eTable 9), the low R2 reflects a weak genetic signal, increasing susceptibility to weak instrument bias. Assuming that our PGS-edu is a valid instrument, effect estimates would be biased toward the null, potentially masking a true protective effect of EA. This limitation is particularly critical for interpreting the effect of EA on depression symptoms compared to other outcomes, as it could obscure a true protective effect of a genetic liability to higher EA on depression.

Selection bias is a recognized limitation of the UK Biobank, as participants are non-randomly sampled and over-represent healthier, wealthier, and more educated individuals (Tyrrell et al., Reference Tyrrell, Zheng, Beaumont, Hinton, Richardson, Wood, Davey Smith, Frayling and Tilling2021). While simulations suggest selection bias may have less impact on MR than pleiotropy or population stratification, it may still differentially influence our analyses (Gkatzionis & Burgess, Reference Gkatzionis and Burgess2019). In within-sibship, MR, non-random participation within families, such as the systematic enrolment of siblings with higher education or better health, may induce selection or collider bias (Sjölander et al., Reference Sjölander, Frisell and Öberg2022). Additionally, as MR analyses rely on available GWAS data, our study is constrained by the lack of ancestral and geographic diversity, which limits generalizability. These limitations highlight the need for replication in more representative cohorts.

Conclusions

Our study suggests that higher EA (or genetic liability to education) may help reduce anxiety, neuroticism, and psychotropic medication use. These mental health benefits do not seem to be explained by EA-linked genetic nurture or demographic factors (e.g., assortative mating, population structure). Regarding depression, results were inconclusive due to imprecise estimates, though beneficial effects of genetic liability to higher EA are possible and warrant further investigation. Additionally, future research should investigate how education (or other closely related phenotypes, such as cognitive skills and income) impacts mental health across diverse populations. Broader areas of research might include: (1) Examining mediating pathways (e.g., mental health literacy and improved living conditions); (2) Identifying critical developmental windows; (3) Understanding gene–environment interplay; and (4) Developing and testing interventions for high-risk groups.

Supplementary material

The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725101736.

Information on the author’s access to data

MFVV and LB had full access to all of the study’s data and are responsible for its integrity and the accuracy of the data analysis.

Data availability statement

Researchers associated with Norwegian research institutes can apply for the use of HUNT material: data and samples – given approval by a Regional Committee for Medical and Health Research Ethics. Researchers from other countries can also apply in cooperation with a Norwegian principal investigator. Information regarding data access can be found at https://www.ntnu.edu/hunt/data. UK Biobank individual-level participant data are available via enquiry to . All GWAS summary statistics used in the present manuscript are publicly available and can be downloaded from https://gwas.mrcieu.ac.uk/ and https://thessgac.com/papers/.

Acknowledgments

The Trøndelag Health Study (HUNT) is a collaboration between HUNT Research Center (Faculty of Medicine and Health Sciences, NTNU, Norwegian University of Science and Technology), Trøndelag County Council, Central Norway Regional Health Authority, and the Norwegian Institute of Public Health. The genotyping in HUNT was financed by the National Institutes of Health, University of Michigan, the Research Council of Norway, the Liaison Committee for Education, Research and Innovation in Central Norway, and the Joint Research Committee between St Olavs Hospital and the Faculty of Medicine and Health Sciences, NTNU. We also wish to express our gratitude to Dr. David Carslake for his valuable comments. MFVV, PRJ, and ØEN are funded by The Research Council of Norway, Project Inequalities in non-communicable diseases: Indirect selection or social causation (INDI-INEQ #287347). BMB is funded by the Research Council of Norway (FRIMEDBIO #287347). Grants from the Liaison Committee for Education, Research and Innovation in Central Norway and the Joint Research Committee between St Olavs Hospital and the Faculty of Medicine and Health Sciences, NTNU, support LB and BMB. GDS works at the Medical Research Council (MRC) Integrative Epidemiology Unit at the University of Bristol, which is supported by MRC (MC_UU_00011/1). NMD is supported by The Research Council of Norway (#295989), the National Institute of Mental Health (MH130448), and the UCL Division of Psychiatry.

Author contribution

Conception and design: MFVV, LB, BMB, and ØEN. Analysis: MFVV and LB. Interpretation of data for the manuscript: MFVV, LB, PRJ, MT, GDS, NMD, BMB, and ØEN. Drafting of the manuscript: MFVV and LB. Critical review of the manuscript for important intellectual content: MFVV, LB, PRJ, MT, GDS, NMD, BMB, and ØEN. Final approval of the version to be published: MFVV, LB, PRJ, MT, GDS, NMD, BMB, and ØEN. Agreement to be accountable for all aspects of the work: MFVV, LB, PRJ, MT, GDS, NMD, BMB, and ØEN.

Funding statement

The funders had no role in the design, analysis, interpretation of the data, drafting of the manuscript, review, or final approval of the version to be published, nor in the decision to submit the manuscript for publication.

Competing interests

ØEN reports receiving funding from the Norwegian University of Science and Technology for a review assignment. GDS reports a Scientific Advisory Board Membership for Relation Therapeutics and Insitro. MFVV, LB, PRJ, MT, NMD, and BMB declare no conflicts of interest to disclose.

Ethical standard

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Footnotes

M.F.V.V., and L.B. authors contributed equally to this work and share first authorship. M.F.V.V., B.M.B., and Ø.N. authors contributed equally to this work and share last authorship.

References

Abbafati, C., Abbas, K. M., Abbasi-Kangevari, M., Abd-Allah, F., Abdelalim, A., Abdollahi, M., Abdollahpour, I., Abegaz, K. H., Abolhassani, H., Aboyans, V., Abreu, L. G., Abrigo, M. R. M., Abualhasan, A., Abu-Raddad, L. J., Abushouk, A. I., Adabi, M., Adekanmbi, V., Adeoye, A. M., Adetokunboh, O. O., … Amini, S. (2020). Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: A systematic analysis for the global burden of disease study 2019. Lancet, 396(10258), 1222. https://doi.org/10.1016/S0140-6736(20)30925-9.Google Scholar
Bergé, L. (2018). Efficient estimation of maximum likelihood models with multiple fixed-effects: The R package FENmlm. DEM Discussion Paper Series, p. 13. https://github.com/lrberge/fixest/blob/master/_DOCS/FENmlm_paper.pdf.Google Scholar
Bjelland, I., Krokstad, S., Mykletun, A., Dahl, A. A., Tell, G. S., & Tambs, K. (2008). Does a higher educational level protect against anxiety and depression? The HUNT study. Social Science and Medicine, 66(6), 1334–1345. https://doi.org/10.1016/j.socscimed.2007.12.019.Google ScholarPubMed
Border, R., Athanasiadis, G., Buil, A., Schork, A. J., Cai, N., Young, A. I., Werge, T., Flint, J., Kendler, K. S., Sankararaman, S., Dahl, A. W., & Zaitlen, N. A. (2022). Cross-trait assortative mating is widespread and inflates genetic correlation estimates. Science, 378(6621), 754–761. https://doi.org/10.1126/science.abo2059.CrossRefGoogle ScholarPubMed
Brumpton, B., Sanderson, E., Heilbron, K., Hartwig, F. P., Harrison, S., Vie, G. Å., Cho, Y., Howe, L. D., Hughes, A., Boomsma, D. I., Havdahl, A., Hopper, J., Neale, M., Nivard, M. G., Pedersen, N. L., Reynolds, C. A., Tucker-Drob, E. M., Grotzinger, A., Howe, L., & Davies, N. M. (2020). Avoiding dynastic, assortative mating, and population stratification biases in Mendelian randomization through within-family analyses. Nature Communications, 11(1), 1–13. https://doi.org/10.1038/s41467-020-17117-4.Google ScholarPubMed
Bycroft, C., Freeman, C., Petkova, D., Band, G., Elliott, L. T., Sharp, K., Motyer, A., Vukcevic, D., Delaneau, O., O’Connell, J., Cortes, A., Welsh, S., Young, A., Effingham, M., McVean, G., Leslie, S., Allen, N., Donnelly, P., & Marchini, J. (2018). The UK biobank resource with deep phenotyping and genomic data. Nature, 562(7726), 203–209. https://doi.org/10.1038/s41586-018-0579-z.CrossRefGoogle ScholarPubMed
Cheesman, R., Eilertsen, E. M., Ahmadzadeh, Y. I., Gjerde, L. C., Hannigan, L. J., Havdahl, A., Young, A. I., Eley, T. C., Njølstad, P. R., Magnus, P., Andreassen, O. A., Ystrom, E., & McAdams, T. A. (2020). How important are parents in the development of child anxiety and depression? A genomic analysis of parent-offspring trios in the Norwegian mother father and child cohort study (MoBa). BMC Medicine, 18, 284. https://doi.org/10.1186/s12916-020-01760-1.CrossRefGoogle ScholarPubMed
Choi, S. W., Mak, T. S. H., & O’Reilly, P. F. (2020). A guide to performing polygenic risk score analyses. Nature Protocols, 15(9), 2759–2772. https://doi.org/10.1038/s41596-020-0353-1.CrossRefGoogle ScholarPubMed
Cohen, A. K., Nussbaum, J., Weintraub, M. L. R., Nichols, C. R., & Yen, I. H. (2020). Association of adult depression with educational attainment, aspirations, and expectations. Preventing Chronic Disease, 17, 200098. https://doi.org/10.5888/pcd17.200098CrossRefGoogle ScholarPubMed
Davies, N. M., Holmes, M. V., & Smith, G. D. (2018). Reading Mendelian randomisation studies: A guide, glossary, and checklist for clinicians. BMJ, 362, k601. https://doi.org/10.1136/bmj.k601.CrossRefGoogle ScholarPubMed
Demange, P. A., Hottenga, J. J., Abdellaoui, A., Eilertsen, E. M., Malanchini, M., Domingue, B. W., Armstrong-Carter, E., de Zeeuw, E. L., Rimfeld, K., Boomsma, D. I., van Bergen, E., Breen, G., Nivard, M. G., & Cheesman, R. (2022). Estimating effects of parents’ cognitive and non-cognitive skills on offspring education using polygenic scores. Nature Communications, 13, 4801. https://doi.org/10.1038/s41467-022-32003-x.CrossRefGoogle ScholarPubMed
Erickson, J., El-Gabalawy, R., Palitsky, D., Patten, S., Mackenzie, C. S., Stein, M. B., & Sareen, J. (2016). Educational attainment as a protective factor for psychiatric disorders: Findings from a nationally representative longitudinal study. Depression and Anxiety, 33(11), 1013–1022. https://doi.org/10.1002/da.22515.CrossRefGoogle ScholarPubMed
Evans, D. M., & Davey Smith, G. (2015). Mendelian randomization: New applications in the coming age of hypothesis-free causality. Annual Review of Genomics and Human Genetics, 16, 327–350. https://doi.org/10.1146/ANNUREV-GENOM-090314-050016.CrossRefGoogle ScholarPubMed
Eysenck, S. B. G., & Tambs, K. (1990). Cross-cultural comparison of personality: Norway and England. Scandinavian Journal of Psychology, 31(3), 191–197. https://doi.org/10.1111/j.1467-9450.1990.tb00830.x.CrossRefGoogle Scholar
Gkatzionis, A., & Burgess, S. (2019). Contextualizing selection bias in Mendelian randomization: How bad is it likely to be? International Journal of Epidemiology, 48(3), 691–701. https://doi.org/10.1093/IJE/DYY202.CrossRefGoogle ScholarPubMed
Hartwig, F. P., Davies, N. M., & Davey Smith, G. (2018). Bias in Mendelian randomization due to assortative mating. Genetic Epidemiology, 42(7), 608–620. https://doi.org/10.1002/gepi.22138.CrossRefGoogle ScholarPubMed
Hemani, G., Zheng, J., Elsworth, B., Wade, K. H., Haberland, V., Baird, D., Laurin, C., Burgess, S., Bowden, J., Langdon, R., Tan, V. Y., Yarmolinsky, J., Shihab, H. A., Timpson, N. J., Evans, D. M., Relton, C., Martin, R. M., Davey Smith, G., Gaunt, T. R., & Haycock, P. C. (2018). The MR-base platform supports systematic causal inference across the human phenome. eLife, 7, e34408. https://doi.org/10.7554/ELIFE.34408.CrossRefGoogle ScholarPubMed
Howe, L. J., Evans, D. M., Hemani, G., Davey Smith, G., & Davies, N. M. (2022). Evaluating indirect genetic effects of siblings using singletons. PLoS Genetics, 18(7), e1010247. https://doi.org/10.1371/journal.pgen.1010247.CrossRefGoogle ScholarPubMed
Howe, L. J., Nivard, M. G., Morris, T. T., Hansen, A. F., Rasheed, H., Cho, Y., Chittoor, G., Ahlskog, R., Lind, P. A., Palviainen, T., van der Zee, M. D., Cheesman, R., Mangino, M., Wang, Y., Li, S., Klaric, L., Ratliff, S. M., Bielak, L. F., Nygaard, M., … Davies, N. M. (2022). Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. Nature Genetics, 54(5), 581–592. https://doi.org/10.1038/s41588-022-01062-7.CrossRefGoogle ScholarPubMed
Howe, L. J., Nivard, M. G., Morris, T. T., Hansen, A. F., Rasheed, H., Cho, Y., Chittoor, G., Lind, P. A., Palviainen, T., van der Zee, M. D., Cheesman, R., Mangino, M., Wang, Y., Li, S., Klaric, L., Ratliff, S. M., Bielak, L. F., Nygaard, M., Reynolds, C. A., … Davies, N. M. (2021). Within-sibship GWAS improve estimates of direct genetic effects. bioRxiv, 2021.03.05.433935. https://doi.org/10.1101/2021.03.05.433935CrossRefGoogle Scholar
Howe, L. J., Tudball, M., Davey Smith, G., & Davies, N. M. (2022). Interpreting Mendelian-randomization estimates of the effects of categorical exposures such as disease status and educational attainment. International Journal of Epidemiology, 51(3), 948–957. https://doi.org/10.1093/ije/dyab208.CrossRefGoogle ScholarPubMed
Hughes, A. M., Torvik, F. A., van Bergen, E., Hannigan, L. J., Corfield, E. C., Andreassen, O. A., Ystrom, E., Ask, H., Smith, G. D., Davies, N. M., & Havdahl, A. (2024). Parental education and children’s depression, anxiety, and ADHD traits, a within-family study in MoBa. npj Science of Learning, 9(1), 1–9. https://doi.org/10.1038/s41539-024-00260-8.CrossRefGoogle ScholarPubMed
Jones, D. P., Wootton, R. E., Gill, D., Carter, A. R., Gunnell, D., Munafò, M. R., & Sallis, H. M. (2021). Mental health as a mediator of the association between educational inequality and cardiovascular disease: A mendelian randomization study. Journal of the American Heart Association, 10(17), 19340. https://doi.org/10.1161/JAHA.120.019340Google ScholarPubMed
Kendler, K. S., & Gardner, C. O. (2017). Genetic and environmental influences on last-year major depression in adulthood: A highly heritable stable liability but strong environmental effects on 1-year prevalence. Psychological Medicine, 47(10), 1816–1824. https://doi.org/10.1017/S0033291717000277.Google ScholarPubMed
Kivimäki, M., Batty, G. D., Pentti, J., Shipley, M. J., Sipilä, P. N., Nyberg, S. T., Suominen, S. B., Oksanen, T., Stenholm, S., Virtanen, M., Marmot, M. G., Singh-Manoux, A., Brunner, E. J., Lindbohm, J. V., Ferrie, J. E., & Vahtera, J. (2020). Association between socioeconomic status and the development of mental and physical health conditions in adulthood: A multi-cohort study. The Lancet Public Health, 5(3), e140–e149. https://doi.org/10.1016/S2468-2667(19)30248-8.CrossRefGoogle ScholarPubMed
Kong, A., Thorleifsson, G., Frigge, M. L., Vilhjalmsson, B. J., Young, A. I., Thorgeirsson, T. E., Benonisdottir, S., Oddsson, A., Halldorsson, B. V., Masson, G., Gudbjartsson, D. F., Helgason, A., Bjornsdottir, G., Thorsteinsdottir, U., & Stefansson, K. (2018). The nature of nurture: Effects of parental genotypes. Science , 359(6374), 424–428. https://doi.org/10.1126/science.aan6877CrossRefGoogle Scholar
Krapohl, E., Rimfeld, K., Shakeshaft, N. G., Trzaskowski, M., McMillan, A., Pingault, J.-B., Asbury, K., Harlaar, N., Kovas, Y., Dale, P. S., & Plomin, R. (2014). The high heritability of educational achievement reflects many genetically influenced traits, not just intelligence. Proceedings of the National Academy of Sciences, 111(42), 15273–15278. https://doi.org/10.1073/pnas.1408777111.CrossRefGoogle Scholar
Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2001). The PHQ-9. Journal of General Internal Medicine, 16(9), 606–613. https://doi.org/10.1046/j.1525-1497.2001.016009606.x.CrossRefGoogle ScholarPubMed
Morris, T. T., Davies, N. M., Hemani, G., & Smith, G. D. (2020). Population phenomena inflate genetic associations of complex social traits. Science Advances, 6(16), eaay0328. https://doi.org/10.1126/SCIADV.AAY0328/SUPPL_FILE/AAY0328_SM.PDF.CrossRefGoogle ScholarPubMed
Niemeyer, H., Bieda, A., Michalak, J., Schneider, S., & Margraf, J. (2019). Education and mental health: Do psychosocial resources matter? SSM – Population Health, 7, 100393. https://doi.org/10.1016/J.SSMPH.2019.100392.CrossRefGoogle ScholarPubMed
Okbay, A., Wu, Y., Wang, N., Jayashankar, H., Bennett, M., Nehzati, S. M., Sidorenko, J., Kweon, H., Goldman, G., Gjorgjieva, T., Jiang, Y., Hicks, B., Tian, C., Hinds, D. A., Ahlskog, R., Magnusson, P. K. E., Oskarsson, S., Hayward, C., Campbell, A., … Young, A. I. (2022). Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nature Genetics, 54(4), 437–449. https://doi.org/10.1038/S41588-022-01016-Z.CrossRefGoogle ScholarPubMed
Olav, B., Svold, A., Langhammer, A., Rehn, T. A., Kjelvik, G., Grøntvedt, V., Sørgjerd, E. P., Fenstad, J. S., Heggland, J., Holmen, O., Stuifbergen, M. C., Aalberg Vikjord, S. A., Brumpton, B. M., Vard, H., Skjellegrind, K., Thingstad, P., Sund, E. R., Selbaek, G., Mork, P. J., … Krokstad, S. (2022). Cohort profile update: The HUNT study, Norway. International Journal of Epidemiology, 2022, 1–12. https://doi.org/10.1093/IJE/DYAC095.Google Scholar
R Core Team. (2021). R: A language and environment for statistical computing [computer software]. R Foundation for Statistical Computing. https://www.r-project.org/Google Scholar
Rutherford, A., Muthukrishna, M., Tropf, F. C., Mills, M. C., Zietsch, B. P., Verweij, K. J. H., & Visscher, P. M. (2025). Socio-economic status is a social construct with heritable components and genetic consequences. Nature Human Behaviour 9(5), 864–876. https://doi.org/10.1038/s41562-025-02150-4.Google Scholar
Sjölander, A., Frisell, T., & Öberg, S. (2022). Sibling comparison studies. Annual Review of Statistics and Its Application, 9(9), 71–94. https://doi.org/10.1146/annurev-statistics-040120-024521.CrossRefGoogle Scholar
Skelton, M., Rayner, C., Purves, K. L., Coleman, J. R. I., Gaspar, H. A., Glanville, K. P., Hunjan, A. K., Hübel, C., Breen, G., & Eley, T. C. (2021). Self-reported medication use as an alternate phenotyping method for anxiety and depression in the UK biobank. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 186(6), 389–398. https://doi.org/10.1002/ajmg.b.32878.CrossRefGoogle Scholar
Smith, G. D., & Ebrahim, S. (2003). Mendelian randomization”: Can genetic epidemiology contribute to understanding environmental determinants of disease? International Journal of Epidemiology, 32(1), 1–22. https://doi.org/10.1093/ije/dyg070.CrossRefGoogle ScholarPubMed
Spitzer, R. L., Kroenke, K., Williams, J. B. W., & Löwe, B. (2006). A brief measure for assessing generalized anxiety disorder: The GAD-7. Archives of Internal Medicine, 166(10), 1092–1097. https://doi.org/10.1001/archinte.166.10.1092.CrossRefGoogle ScholarPubMed
Tyrrell, J., Zheng, J., Beaumont, R., Hinton, K., Richardson, T. G., Wood, A. R., Davey Smith, G., Frayling, T. M., & Tilling, K. (2021). Genetic predictors of participation in optional components of UK biobank. Nature Communications, 12(1), 886. https://doi.org/10.1038/S41467-021-21073-Y.CrossRefGoogle ScholarPubMed
van Buuren, S., & Groothuis-Oudshoorn, K. (2011). Mice: Multivariate imputation by chained equations in R. Journal of Statistical Software, 45(3), 1–67. https://doi.org/10.18637/JSS.V045.I03.Google Scholar
van de Weijer, M. P., Demange, P. A., Pelt, D. H. M., Bartels, M., & Nivard, M. G. (2024). Disentangling potential causal effects of educational duration on well-being, and mental and physical health outcomes. Psychological Medicine, 54(7), 1403–1418. https://doi.org/10.1017/S003329172300329X.CrossRefGoogle ScholarPubMed
Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36, 1–48. https://doi.org/10.18637/jss.v036.i03.CrossRefGoogle Scholar
Walker, E. R., McGee, R. E., & Druss, B. G. (2015). Mortality in mental disorders and global disease burden implications: A systematic review and meta-analysis. JAMA Psychiatry, 72(4), 334. https://doi.org/10.1001/JAMAPSYCHIATRY.2014.2502.CrossRefGoogle ScholarPubMed
Walker, V., Sanderson, E., Levin, M. G., Damraurer, S. M., Feeney, T., & Davies, N. M. (2024). Reading and conducting instrumental variable studies: Guide, glossary, and checklist. BMJ (Clinical Research Ed.), 387, e078093. https://doi.org/10.1136/bmj-2023-078093.Google ScholarPubMed
World Health Organization. (2022). World mental health report: Transforming mental health for all. In World Mental Health Report. World Health Organization. https://www.who.int/publications/i/item/9789240049338Google Scholar
Yuan, S., Xiong, Y., Michaëlsson, M., Michaëlsson, K., & Larsson, S. C. (2021). Genetically predicted education attainment in relation to somatic and mental health. Scientific Reports, 11(1), 4296. https://doi.org/10.1038/s41598-021-83801-0.CrossRefGoogle ScholarPubMed
Zigmond, A. S., & Snaith, R. P. (1983). The hospital anxiety and depression scale. Acta Psychiatrica Scandinavica, 67(6), 361–370. https://doi.org/10.1111/J.1600-0447.1983.TB09716.X.CrossRefGoogle ScholarPubMed
Figure 0

Table 1. General characteristics of HUNT and UKB samples

Figure 1

Table 2. Association between educational attainment and mental health outcomes

Figure 2

Figure 1. Educational attainment and symptoms of anxiety. Standard deviation (SD) changes in the anxiety score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’.

Figure 3

Figure 2. Educational attainment and symptoms of depression. Standard deviation (SD) changes in the depression score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’; 2SMR, ‘two-sample Mendelian randomization’.

Figure 4

Figure 3. Educational attainment and neuroticism. Standard deviation (SD) changes in the neuroticism score and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for ordinary least squares regression (OLS) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; OLS EA, ‘ordinary least squares regression model with educational attainment as exposure’; OLS PGS-edu, ‘ordinary least squares regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’; 2SMR, ‘two-sample Mendelian randomization’.

Figure 5

Figure 4. Educational attainment and use of psychotropic medication. Log odds changes in psychotropic medication use and its 95% confidence interval per SD increase in years of education are shown. Estimated associations are displayed for logistic regression (LOG) and Mendelian randomization models. Note: SD, ‘standard deviation unit’; LOG EA, ‘logistic regression model with educational attainment as exposure’; LOG PGS-edu, ‘logistic regression model with the educational attainment polygenic score as exposure’; 1SMR, ‘one-sample Mendelian randomization’.

Supplementary material: File

Vinueza Veloz et al. supplementary material

Vinueza Veloz et al. supplementary material
Download Vinueza Veloz et al. supplementary material(File)
File 183.2 KB