IEA International Journal of Epidemiology, 2023, 386–402 https://doi.org/10.1093/ije/dyac128 Advance Access Publication Date: 12 June 2022 International Epidemiological Association Original article ADHD and Autism Spectrum Disorder Mendelian randomization analysis for attention deficit/hyperactivity disorder: studying a broad range of exposures and outcomes Marıa Soler Artigas ,1,2,3,4* Cristina Sánchez-Mora,1,2,3,4 Paula Rovira,1,2,5,6 Laura Vilar-Ribó,1,2,3 Josep Antoni Ramos-Quiroga1,2,3,7 and Marta Ribase s1,2,3,4 1 Psychiatric Genetics Unit, Group of Psychiatry, Mental Health and Addiction, Vall d’Hebron Research Institute (VHIR), Universitat Autònoma de Barcelona, Barcelona, Spain, 2Department of Mental Health, Hospital Universitari Vall d’Hebron, Barcelona, Spain, 3Biomedical Network Research Centre on Mental Health (CIBERSAM), Instituto de Salud Carlos III, Madrid, Spain, 4Department of Genetics, Microbiology, and Statistics, Faculty of Biology, Universitat de Barcelona, Barcelona, Spain, 5 Vicerectorat de Recerca, postdoctoral researcher Margarita Salas, Universitat de Barcelona, Barcelona, Spain, 6Departament of Psychiatry, Faculty of Medicine, Universidad de Granada, Spain and 7Department of Psychiatry and Forensic Medicine, Universitat Autònoma de Barcelona, Barcelona, Spain *Corresponding author. Vall d’Hebron Research Institute (VHIR), Passeig Vall d’Hebron 119-129 08035 Barcelona, Spain. E-mail: maria.soler@vhir.org Received 31 March 2021; Editorial decision 13 May 2022; Accepted 26 May 2022 Abstract Background: Attention deficit/hyperactivity disorder (ADHD) is a highly prevalent neuro- developmental disorder caused by a combination of genetic and environmental factors and is often thought as an entry point into a negative life trajectory, including risk for comorbid disorders, poor educational achievement or low income. In the present study, we aimed to clarify the causal relationship between ADHD and a comprehensive range of related traits. Methods: We used genome-wide association study (GWAS) summary statistics for ADHD (n ¼ 53 293) and 124 traits related to anthropometry, cognitive function and intelli- gence, early life exposures, education and employment, lifestyle and environment, lon- gevity, neurological, and psychiatric and mental health or personality and psychosocial factors available in the MR-Base database (16 067 n 766 345). To investigate their causal relationship with ADHD, we used two-sample Mendelian randomization (MR) with a range of sensitivity analyses, and validated MR findings using causal analysis using summary effect estimates (CAUSE), aiming to avoid potential false-positive results. Results: Our findings strengthen previous evidence of a causal effect of ADHD liability on smoking and major depression, and are consistent with a causal effect on odds of de- creased average total household income [odds ratio (OR) ¼ 0.966, 95% credible interval (CrI) ¼ (0.954, 0.979)] and increased lifetime number of sexual partners [OR ¼ 1.023, 95% C The Author(s) 2022. Published by Oxford University Press on behalf of the International Epidemiological Association. V 386 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unre- stricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. International Journal of Epidemiology, 2023, Vol. 52, No. 2 387 CrI ¼ (1.013, 1.033)]. We also found evidence for a causal effect on ADHD for liability of arm predicted mass and weight [OR ¼ 1.452, 95% CrI ¼ (1.307, 1.614) and OR ¼ 1.430, 95% CrI ¼ (1.326, 1.539), respectively] and time spent watching television [OR ¼ 1.862, 95% CrI ¼ (1.545, 2.246)], and evidence for a bidirectional effect for age of first sexual in- tercourse [beta ¼ 0.058, 95% CrI ¼ (0.072, 0.044) and OR ¼ 0.413, 95% CrI ¼ (0.372, 0.457), respectively], odds of decreased age completed full-time education [OR ¼ 0.972, 95% CrI ¼ (0.962, 0.981) and OR ¼ 0.435, 95% CrI ¼ (0.356, 0.533), respectively] and years of schooling [beta ¼ -0.036, 95% CrI ¼ (0.048, 0.024) and OR ¼ 0.458, 95% CrI ¼ (0.411, 0.511), respectively]. Conclusions: Our results may contribute to explain part of the widespread co-occurring traits and comorbid disorders across the lifespan of individuals with ADHD and may open new opportunities for developing preventive strategies for ADHD and for negative ADHD trajectories. Key words: ADHD, Mendelian randomization, causal analysis using summary effect estimates Key Messages • Our results are consistent with a causal effect of attention deficit/hyperactivity disorder (ADHD) genetic liability decreasing average total household income and increasing lifetime number of sexual partners. • We detect a positive effect of the liability of anthropometric traits (arm predicted mass and weight) and of time spent watching television on ADHD. • We show evidence for a bidirectional negative effect between liability of ADHD and of education outcomes (years of schooling and age completed full-time education), age of first sexual intercourse and past tobacco smoking for non- heavy smokers. Introduction as response inhibition, vigilance, working memory or plan- Attention deficit/hyperactivity disorder (ADHD) is a neu- ning, personality profiles with low effortful control and rodevelopmental disorder with a prevalence of around high neuroticism,8–11 or emotion dysregulation problems 5.3% in childhood and 2.8% in adulthood.1,2 The aetiol- such as irritability or temper outbursts.12 In addition, up to ogy of ADHD involves a combination of genetic and envi- 70–80% of ADHD patients suffer from comorbid disor- ronmental factors, with an estimated heritability of 70– ders across their lifespan.13 These include other psychiatric 80%.3,4 Potential environmental risk factors include pre- conditions, such as major depressive, oppositional defiant, and perinatal risk factors (maternal smoking or alcohol bipolar or substance use disorders, but also somatic dis- consumption, low birthweight, prematurity), exposure to eases such as obesity, sleep disorders or migraine.7 The environmental toxins, unfavourable psychosocial condi- presence of comorbidities in ADHD worsens symptom tions (severe early childhood deprivation, maternal hostil- progression, disorder course and outcome, and also ity) or low socioeconomic status.5,6 increases mortality rates.8,14 In this context ADHD can be ADHD is characterized by a persistent pattern of inat- thought of as an entry point into a negative life trajectory tentive, hyperactive and impulsive behaviour; however, its with higher risks for poor educational achievement, unem- clinical presentation is heterogeneous, with a wide spec- ployment or criminality, among others.7 trum of severity and symptoms that often overlap with Most studies undertaken to date have reported associa- other conditions.7 There are a number of traits that are not tion between ADHD and comorbid traits, but inferring part of the core diagnostic criteria for ADHD, which can causality can be more challenging due to the potential ef- nevertheless influence severity, persistence and treatment fect of confounding factors or reverse causality. Different decisions. For instance, individuals with ADHD often have strategies have been developed to overcome these inference a poor cognitive performance in executive functions, such problems, and the causality for ADHD and a number of 388 International Journal of Epidemiology, 2023, Vol. 52, No. 2 traits have been tested using: (i) longitudinal analyses, some metabolites’ levels (n ¼ 132), to procedural metrics or bio- of them undertaken in twins; (ii) Mendelian randomization logical samples (n ¼ 76) and to clinical traits (n ¼ 600), ex- (MR), which uses genetic variants as proxies for an expo- cept those related to neurological, psychiatric and mental sure (instrumental variables) to test for a causal effect on an health; and selected traits related to anthropometry, cogni- outcome15; and (iii) the latent causal variable (LCV) model, tive function and intelligence, early life exposures, educa- based on a latent variable that mediates the genetic correla- tion and employment, lifestyle and environment, longevity, tion between two traits and quantifies the degree of causal- neurological, psychiatric and mental health or personality ity between them.16 When using only one approach, and psychosocial factors. We then removed duplicated and longitudinal analyses have reported a causal role for low unordered categorical traits (n ¼ 150, Supplementary family income in early childhood on ADHD6 and of ADHD Table S1). Finally, in order to reduce further the number of on lower educational achievement.17 In addition, MR stud- traits analysed to those with causal relationships most ies have reported an effect of the liability of low birthweight likely to be identified by MR methods, we removed traits increasing the risk for ADHD,18 an effect of higher intelli- with a heritability Z score 4 and an ADHD genetic corre- gence lowering the risk for ADHD19 and an effect of the ge- lation Z score 2, obtained using LD score regression,35,36 netic liability to ADHD increasing the risk for asthma20 and ending up with a total of 124 traits included in the analysis coronary artery disease, as well as a positive bidirectional ef- (Supplementary Tables S1 and S2, available as fect for childhood obesity.21 When more than one approach Supplementary data at IJE online). The MR-Base summary was used, consistent results were found for an effect for statistics used for these 124 traits were obtained from dif- ADHD liability on major depression22 and inconsistent ferent sources: European Bioinformatics Institute (EBI) results, which require further investigation, were identified database,37 Neale Lab and Integrative Epidemiology Unit for body mass index (BMI),23–25 phone use,26 smoking, can- (IEU) analyses of UK Biobank phenotypes [http://www.nea nabis and alcohol use.27–30 lelab.is/uk-biobank]38 and manually collected and curated Overall, the evidence from causal inference analyses un- data from different consortia for MR-Base.19,39–47 dertaken for ADHD is hard to interpret, in some cases due to the limited number of strategies applied19,20 or because inconsistent findings are found when using different Mendelian randomization approaches.29,30 In the present study we aim to clarify the Main analysis relationship between ADHD and a comprehensive range of We ran two-sample MR in both directions to assess the po- related traits, using genome-wide association studies tential causal relationship between ADHD and every se- (GWAS) datasets available,31,32 following current guide- lected trait, using the multiplicative random effects inverse lines on two-sample MR analyses33 and a range of sensitiv- variance weighted (IVW) method as the main analysis.48 ity analyses. In addition, we validated MR findings using MR analyses only included independent single nucleotide causal analysis using summary effect estimates (CAUSE), a polymorphisms (SNPs) (r2 <0.001 or distance >10 000 kb) recently developed method to account for horizontal plei- with P <5.00e-08 in the exposure, and variants meeting otropy that acts through a common shared heritable factor, the threshold for both the exposure and the outcome were and avoids potential false-positive results.34 removed. For exposure variants not found in the outcome, GWAS proxies were used instead (r2 0.8, obtained using Methods 1000 Genomes European sample). ADHD liability genetic instruments are provided in Supplementary Table S3 GWAS dataset selection (available as Supplementary data at IJE online). False-dis- We selected GWAS summary statistics available in the covery rate (FDR) across all tests considered was used to MR-Base database31,32 in May 2020 (n ¼ 31 772) which correct for multiple testing. fulfilled the following inclusion criteria: (i) sample size [N effective ¼ 4  N cases  N controls/(N cases þ N con- MR sensitivity analyses trols) for binary traits] >5000; (ii) results available in more For IVW results with FDR corrected P <0.05, we under- than 450 000 genetic variants; (iii) European ancestry; (iv) took sensitivity analyses to assess the robustness of the non sex-specific; and (v) more than three independent findings under weaker assumptions, given that the genetic genome-wide significant signals (P <5.00e-08). This re- variants used as instruments must meet three assumptions duced the number of traits to 1259 (Supplementary Table for IVW results to be valid: (i) robust association with the S1, available as Supplementary data at IJE online). At this exposure; (ii) no horizontal pleiotropy, or association with point we removed traits related to diet (n ¼ 52), to the outcome through a pathway independent of the International Journal of Epidemiology, 2023, Vol. 52, No. 2 389 exposure; and (iii) independence of confounders that influ- (CAUSE).34 Uncorrelated horizontal pleiotropy occurs ence the exposure and the outcome. when the effects on the exposure and the outcome are We used weighted median and weighed mode methods, uncorrelated and it can be accounted for by methods such which only require a subset of variants to be valid instru- as MR-Egger or MR-PRESSO. Correlated horizontal plei- ments and are robust to outliers.49,50 Under equal weights, otropy takes place when the effects on the exposure and the weighted median requires at least half of the variants to the outcome act through a shared heritable factor, and is be valid instruments, and the weighted mode requires more harder to account for by current MR methods. CAUSE can variants to estimate the true causal effect than any other deal with both kinds of horizontal pleiotropy, avoiding quantity.49,50 We used MR-PRESSO, which performs a therefore potential false-positive findings. This method test to detect horizontal pleiotropy (MR-PRESSO global uses expected log pointwise posterior density (ELPD), a test), and if detected, it removes horizontal pleiotropic out- Bayesian approach, to compare two nested models: a shar- liers and then performs the IVW method using the remain- ing model where the causal effect (c) is zero but allows for ing instruments.51 We also applied MR-Egger, which is horizontal pleiotropic effects; and a causal model where c affected by outlying data points but allows all genetic var- is a free parameter.34 Independent variants (r2 <0.01; dis- iants to have pleiotropic effects, assuming that these effects tance >250 kb) associated to the exposure with P <1.00e- are independent of the variant-exposure associations.52,53 03 were included, and the cause v1.2.0 R package was MR-Egger also implements a pleiotropy test, however, used for these analyses.34 A Bonferroni correction for the when the assumption of no measurement error in the SNP- number of tests undertaken was used to correct for multi- exposure effect estimates (NOME assumption) is violated ple testing at this stage (P <1.21e-03). (I2GX <0.9). MR-Egger causal estimates are biased towards the null, and the pleiotropy test type I error can be in- flated.54 For this reason, when 0.6 10 indicate strong-enough instruments.55 In or- in both directions (Figure 1; Supplementary Table S4, der to avoid results due to reverse causation, we also ap- available as Supplementary data at IJE online). After sensi- plied Steiger filtering, removing instruments from the tivity analyses there was evidence of a positive effect of the analysis if they did not explain substantially more of the genetic liability of eight traits (arm, leg, whole body and variance in the exposure trait than in the outcome and un- trunk fat-free mass, arm and trunk predicted mass, whole dertaking the IVW method on the remaining set of instru- body water mass and weight) on ADHD (Table 1), show- ments.56 Variance explained for binary traits was ing moderate heterogeneity (40.128%0.9 or 0.610 for continuous exposures. In the IVW analysis, a negative effect of ADHD genetic lia- bility on mean time to correctly identify matches, a mea- Causal analysis using summary effect estimates sure of raw processing and reaction speed, was detected as For those traits that met the MR sensitivity criteria, we well as a negative effect of intelligence when ADHD was also ran causal analysis using summary effect estimates the outcome, and for cognitive performance and fluid 390 International Journal of Epidemiology, 2023, Vol. 52, No. 2 Figure 1 Number of traits included in each stage of the analysis for each direction and category. Total number of traits analysed (‘All’) are presented as well as number of traits meeting the significance criteria in the MR main analysis (‘MR main’), in the MR sensitivity analyses (‘MR sens’) and in CAUSE (‘CAUSE’). ADHD, attention deficit/hyperactivity disorder; CAUSE, causal analysis using summary effect estimates; MR, Mendelian randomization intelligence in both directions (IVW FDR P <0.05) remained significant [from OR ¼ 1.037, 95% confidence (Table 1; Supplementary Table S4; Figure 1). Of them, interval (CI) ¼ (1.023, 1.052) to OR ¼ 1.03, 95% the effect of ADHD as exposure on cognitive performance CI ¼ (1.018, 1.043), I2 ¼ 36.476%, Supplementary Table survived the sensitivity analyses, but showed high S5c, Supplementary Figure S1am]. The effect of ADHD on heterogeneity (I2 ¼ 84.231%, Supplementary Table S5b, maternal smoking around birth, however, did not survive Supplementary Figure S1a–h). When outliers were re- multiple comparisons correction with CAUSE (DELPD moved in the MR-PRESSO analysis, the magnitude of the P ¼ 2.19e-03, Table 2). causal effect estimate increased slightly (Supplementary Table S5b,) and the heterogeneity went down to an I2 of Education and employment 68.844%. The CAUSE analysis did not show evidence for the effect of ADHD liability on cognitive performance The IVW analysis detected results with FDR P <0.05 in (DELPD P ¼ 0.069, Table 2). both directions for all traits analysed, with a negative effect for age completed full-time education and years of school- ing and a positive effect for job involving heavy manual, Early life events physical work or mainly walking or standing (Figure 1, Out of all early life exposure traits, a positive effect be- Table 1; Supplementary Table S4). Only age completed tween maternal smoking around birth and ADHD was full-time education and years of schooling met the sensitiv- identified in both directions in the IVW analysis; but only ity analysis criteria in both directions. However, when when ADHD was considered as exposure were the sensitiv- ADHD was the outcome, the Steiger analysis provided evi- ity criteria met (Figure 1; Supplementary Table S4). There dence for reverse causation,with reduced effect sizes after was moderately high heterogeneity for this analysis (I2 ¼ filtering, which suggests that at least part of the observed 63.896%), and although removing one outlier in the MR- effect was due to ADHD liability causing the educational PRESSO analysis reduced the heterogeneity and the causal outcomes (Supplementary Table S5d). The heterogeneity effect estimate, measured as odds of maternal smoking per for these analyses was moderate (I2 <50%) except for the unit increase in log odds ratio [log(OR)] of ADHD, it effect of ADHD liability on years of schooling, which was International Journal of Epidemiology, 2023, Vol. 52, No. 2 Table 1 Mendelian randomization results Traita ADHD ! Trait Trait ! ADHD Number Effect 95% CI FDR P Meets Number OR 95% CI FDR P Meets of SNPs size sensitivity? of SNPs sensitivity? Anthropometric measures Arm fat-free mass (left) 9 0.026 (0.005, 0.046) 3.09E-02 No 479 1.671 (1.449, 1.927) 2.00E-11 Yes Arm predicted mass (left) 9 0.025 (0.004, 0.046) 4.38E-02 No 471 1.665 (1.438, 1.929) 1.11E-10 Yes Whole body fat-free mass 9 0.016 (0.006, 0.038) 2.31E-01 – 509 1.431 (1.251, 1.637) 1.13E-06 Yes Whole body water mass 9 0.016 (0.006, 0.038) 2.25E-01 – 518 1.456 (1.27, 1.669) 5.24E-07 Yes Leg fat-free mass (left) 9 0.012 (0.016, 0.04) 4.88E-01 – 460 1.461 (1.262, 1.69) 2.12E-06 Yes Trunk fat-free mass 9 0.017 (0.003, 0.037) 1.47E-01 – 520 1.397 (1.218, 1.602) 8.25E-06 Yes Trunk predicted mass 9 0.017 (0.003, 0.037) 1.58E-01 – 518 1.406 (1.225, 1.613) 6.09E-06 Yes Weight 9 0.045 (0.002, 0.088) 7.67E-02 – 458 1.662 (1.485, 1.859) 1.54E-17 Yes Cognitive function and intelligence Cognitive performance 9 0.072 (0.131, 0.013) 3.55E-02 Yes 134 0.600 (0.499, 0.722) 5.24E-07 No Early life events Maternal smoking around birthb 9 1.037 (1.023, 1.052) 1.89E-06 Yes 15 143.936 (25.34, 817.581) 1.70E-07 No Education and employment Age completed full-time educationb 9 0.930 (0.909, 0.951) 2.52E-09 Yes 36 0.167 (0.102, 0.275) 2.00E-11 Yes Years of schooling 9 0.098 (0.138, 0.059) 5.99E-06 Yes 297 0.298 (0.251, 0.354) 1.20E-41 Yes Lifestyle and environment Alcohol intake frequencyb 9 1.108 (1.049, 1.171) 8.51E-04 Yes 92 1.399 (1.168, 1.676) 8.51E-04 No Alcohol intake versus 10 years previouslyb 9 1.042 (1.026, 1.058) 1.13E-06 Yes 12 8.389 (2.293, 30.696) 3.62E-03 Yes Alcohol usually taken with mealsb 9 0.959 (0.942, 0.976) 2.08E-05 Yes 33 0.085 (0.033, 0.221) 2.19E-06 No Average weekly champagne plus white wine intakeb 9 0.955 (0.936, 0.975) 4.74E-05 Yes 4 0.148 (0.041, 0.528) 7.92E-03 No Frequency of stair-climbing in past 4 weeksb 9 0.954 (0.918, 0.991) 3.21E-02 Yes 17 0.383 (0.204, 0.717) 7.00E-03 No Nitrogen dioxide air pollution 2010 9 0.032 (0.016, 0.049) 5.37E-04 Yes 8 0.610 (0.159, 2.339) 5.43E-01 – Particulate matter air pollution (pm2.5) 2010 9 0.047 (0.029, 0.064) 1.55E-06 Yes 7 1.989 (0.38, 10.401) 5.01E-01 – Age first had sexual intercourse 9 0.130 (-0.18, -0.08) 1.74E-06 Yes 184 0.223 (0.187, 0.266) 8.34E-61 Yes Lifetime number of sexual partnersb 9 1.063 (1.029, 1.099) 8.61E-04 Yes 60 2.958 (1.76, 4.971) 1.63E-04 No Current tobacco smokingb 9 1.041 (1.028, 1.055) 4.94E-09 Yes 32 12.305 (3.945, 38.38) 6.54E-05 No Ever smokedb 9 1.032 (1.022, 1.043) 3.64E-09 Yes 73 15.251 (7.528, 30.898) 7.20E-13 No (Continued) 391 392 Table 1 Continued Traita ADHD ! Trait Trait ! ADHD Number Effect 95% CI FDR P Meets Number OR 95% CI FDR P Meets of SNPs size sensitivity? of SNPs sensitivity? Pack/years adult smoking as proportion of lifespan 9 0.075 (0.039, 0.111) 1.73E-04 Yes 13 1.335 (0.971, 1.835) 1.28E-01 – exposed to smoking Past tobacco smokingb 9 0.905 (0.88, 0.931) 4.14E-11 Yes 87 0.381 (0.296, 0.491) 1.02E-12 Yes Average total household income before taxb 9 0.909 (0.885, 0.933) 2.65E-11 Yes 44 0.329 (0.237, 0.455) 2.37E-10 No Number of full sistersb 9 1.029 (1.016, 1.043) 8.78E-05 Yes 5 9.746 (0.4, 237.152) 2.38E-01 – Townsend deprivation index at recruitment 9 0.075 (0.055, 0.095) 1.35E-12 Yes 17 5.400 (1.952, 14.941) 3.25E-03 Yes Length of mobile phone useb 9 1.013 (0.975, 1.052) 5.87E-01 – 29 1.983 (1.314, 2.993) 3.14E-03 Yes Time spent watching television (TV)b 9 1.030 (0.997, 1.065) 1.34E-01 – 107 3.495 (2.521, 4.847) 1.00E-12 Yes Longevity Parental longevity (combined parental attained age, 9 0.050 (0.029, 0.071) 1.07E-05 Yes 10 0.972 (0.462, 2.041) 9.62E-01 – Martingale residuals)c Neurological, psychiatric and mental health Major depressive disorderb 10 1.200 (1.122, 1.283) 7.12E-07 Yes 5 1.739 (1.048, 2.886) 6.25E-02 – Personality and psychosocial factors Able to confideb 9 0.919 (0.881, 0.959) 3.51E-04 No 13 0.635 (0.46, 0.878) 1.39E-02 Yes International Journal of Epidemiology, 2023, Vol. 52, No. 2 Frequency of tiredness/lethargy in past 2 weeksb 9 1.029 (1.004, 1.055) 5.16E-02 – 36 5.220 (2.809, 9.698) 1.13E-06 Yes Frequency of unenthusiasm/disinterest in past 2 weeksb 9 1.034 (1.021, 1.047) 1.13E-06 yes 11 8.077 (1.753, 37.221) 1.69E-02 No For all the traits with inverse variance weighted FDR P <0.05 which also met the sensitivity criteria in at least one direction, the number of SNPs included, the inverse variance weighted results and information on whether the MR sensitivity criteria were met are presented for both directions. MR, Mendelian randomization; ADHD, attention deficit/hyperactivity disorder; SNP, single nucleotide polymorphism; CI, 95% confidence interval; FDR, false-discovery rate; P, P-value. a The unit reported for all continuous traits is SD, except for years of schooling, which was provided in years and cognitive performance, which was provided using a cognitive score.44 b The causal effect size provided for the comparison with ADHD as the exposure for these traits is odds raio (OR) per ADHD log(OR) unit increase, since they were all analysed as categorical ordered, except for maternal smoking around birth, ever smoked, alcohol usually taken with meals and major depressive disorder, which were analysed as binary. c For parental longevity (combined parental attained age, Martingale residuals) a positive effect size indicates decreased attained age.47 International Journal of Epidemiology, 2023, Vol. 52, No. 2 393 Figure 2 Diagram of the relationships found with consistent evidence across methods. Traits are coloured by category (education and employment in green; anthropometric measures in blue; neurological, psychiatric and mental health-related traits in red; and lifestyle and environment in purple) and CAUSE effect sizes are presented. Details on effect size units can be found in Table 2. ADHD, attention deficit/hyperactivity disorder; CAUSE, causal analysis using summary effect estimates; TV, television considerable (I2 ¼ 87.572%, Supplementary Table S5d). analyses, we found evidence for an effect of ADHD as ex- After excluding outliers, MR-PRESSO results remained sig- posure on traits related to alcohol use (increased alcohol nificant (Supplementary Table S5d, Supplementary Figure intake frequency, alcohol intake versus 10 years previously, S1au) and the heterogeneity decreased, although it remained decreased alcohol usually taken with meals and average substantial (I2 ¼ 67.755%). CAUSE confirmed these results, weekly champagne plus white wine intake), physical exer- providing strong evidence for a causal effect per unit increase cise (decreased frequency of stair-climbing in past in the log(OR) of ADHD on odds of decreased age completed 4 weeks), air pollution (increased nitrogen dioxide and par- full-time education and years of schooling [OR ¼ 0.972, ticulate matter), sexual behaviour (decreased age first had 95% CrI ¼ (0.962, 0.981), DELPD P ¼ 2.85e-04 and beta ¼ - sexual intercourse and increased lifetime number of sexual 0.036, 95% CrI ¼ (-0.048, -0.024), DELPD P ¼ 8.04e-05, re- partners), smoking (increased current tobacco smoking, spectively] (Figure 2, Table 2). Evidence in the opposite direc- ever smoking and pack/years adult smoking, and decreased tion was also found, with causal effect per unit increase in past tobacco smoking) and sociodemographic information odds of increased age completed full-time education and (decreased average total household income and increased years of schooling on ADHD odds [OR ¼ 0.435, 95% Townsend deprivation index and number of sisters) CrI ¼ (0.356, 0.533), DELPD P ¼ 2.51e-06 and OR ¼ 0.458, (Table 1). Alcohol intake versus 10 years previously, age of 95% CrI ¼ (0.411, 0.511), DELPD P ¼ 1.08e-010, respec- first sexual intercourse, past tobacco smoking and tively] (Figure 2, Table 2). Townsend deprivation index also had an effect when ADHD was used as outcome, as did length of mobile phone use and time spent watching television (Table 1). Lifestyle and environment When ADHD was considered as outcome, all analyses In the IVW analysis 12 lifestyle and environment traits had that met the sensitivity criteria also showed evidence of re- FDR P <0.05 when ADHD was the exposure, three when verse causation, with smaller effect sizes after Steiger filter- ADHD was the outcome and 18 in both directions ing particularly for Townsend deprivation index, (Figure 1; Supplementary Table S4). After the sensitivity suggesting that at least some the observed effect was due to 394 Table 2 Causal analysis using summary effect estimates results Traita ADHD ! Trait Trait ! ADHD c 95% CrI DELPD (SE) DELPD P c 95% CrI DELPD (SE) DELPD P Anthropometric measures Arm fat-free mass (left) – – – – 1.441 (1.287,1.606) 4.444 (1.555) 2.13E-03 Arm predicted mass (left) – – – – 1.452 (1.307,1.614) 4.57 (1.396) 5.32E-04 Whole body fat-free mass – – – – 1.312 (1.175,1.464) 3.275 (1.405) 9.87E-03 Whole body water mass – – – – 1.312 (1.176,1.464) 3.14 (1.346) 9.83E-03 Leg fat-free mass (left) – – – – 1.371 (1.234,1.523) 3.79 (1.4) 3.39E-03 Trunk fat-free mass – – – – 1.232 (1.1,1.381) 2.51 (1.44) 4.06E-02 Trunk predicted mass – – – – 1.244 (1.108,1.398) 2.505 (1.406) 3.74E-02 Weight – – – – 1.430 (1.326,1.539) 5.71 (1.202) 1.01E-06 Cognitive function and intelligence Cognitive performance 0.023 (0.04,0.006) 2.681 (1.81) 6.92E-02 – – – – Early life events Maternal smoking around birthb 1.012 (1.007,1.017) 4.886 (1.715) 2.19E-03 – – – – Education and employment Age completed full time educationb 0.972 (0.962,0.981) 5.655 (1.641) 2.85E-04 0.435 (0.356,0.533) 6.159 (1.349) 2.51E-06 International Journal of Epidemiology, 2023, Vol. 52, No. 2 Years of schooling 0.036 (0.048,-0.024) 6.5 (1.723) 8.04E-05 0.458 (0.411,0.511) 7.184 (1.131) 1.08E-10 Lifestyle and environment Alcohol intake frequencyb 1.033 (1.017,1.049) 5.103 (2.079) 7.04E-03 – – – – Alcohol intake versus 10 years previouslyb 1.010 (1.003,1.016) 2.66 (1.813) 7.12E-02 2.099 (1.294,3.384) 3.324 (1.814) 3.34E-02 Alcohol usually taken with mealsb 0.986 (0.98,0.993) 4.543 (1.8) 5.80E-03 – – – – Average weekly champagne plus white wine intakeb 0.985 (0.977,0.994) 3.654 (1.905) 2.75E-02 – – – – Frequency of stair climbing in last 4 weeksb 0.991 (0.979,1.002) 0.55 (1.288) 3.35E-01 – – – – Nitrogen dioxide air pollution 2010 0.003 (-0.005,0.012) 0.653 (0.47) 9.18E-01 – – – – Particulate matter air pollution (pm2.5) 2010 0.008 (-0.001,0.016) 0.936 (1.484) 2.64E-01 – – – – Age first had sexual intercourse 0.058 (-0.072,0.044) 6.939 (1.562) 4.47E-06 0.413 (0.372,0.457) 7.797 (1.105) 8.70E-13 Lifetime number of sexual partnersb 1.023 (1.013,1.033) 5.869 (1.911) 1.06E-03 – – – – Current tobacco smokingb 1.015 (1.01,1.021) 5.488 (1.703) 6.35E-04 – – – – Ever smokedb 1.011 (1.006,1.016) 5.368 (1.974) 3.27E-03 – – – – (Continued) International Journal of Epidemiology, 2023, Vol. 52, No. 2 Table 2 Continued Traita ADHD ! Trait Trait ! ADHD c 95% CrI DELPD (SE) DELPD P c 95% CrI DELPD (SE) DELPD P Pack years adult smoking as proportion of life 0.034 (0.019,0.047) 4.232 (1.751) 7.81E-03 – – – – span exposed to smoking Past tobacco smokingb 0.963 (0.947,0.978) 5.747 (1.829) 8.37E-04 0.588 (0.523,0.664) 5.685 (1.399) 2.41E-05 Average total household income before taxb 0.966 (0.954,0.979) 5.777 (1.809) 7.01E-04 – – – – Number of full sistersb 1.007 (0.999,1.013) 1.207 (1.546) 2.17E-01 – – – – Townsend deprivation index at recruitment 0.024 (0.015,0.033) 4.996 (1.701) 1.65E-03 1.870 (1.38,2.522) 3.676 (1.546) 8.72E-03 Length of mobile phone useb – – – – 1.281 (1.084,1.516) 2.874 (1.805) 5.56E-02 Time spent watching television (TV)b – – – – 1.862 (1.545,2.246) 4.491 (1.475) 1.16E-03 Longevity Parental longevity (combined parental attained age, 0.017 (0.007,0.026) 3.663 (1.873) 2.52E-02 – – – – Martingale residuals)c Neurological, psychiatric and mental health Major depressive disorderb 1.110 (1.073,1.148) 5.828 (1.666) 2.34e-04 – – – – Personality and psychosocial factors Able to confideb – – – – 0.864 (0.739,1.012) 0.836 (1.416) 2.77E-01 Frequency of tiredness/lethargy in past 2 weeksb – – – – 1.772 (1.388,2.25) 3.717 (1.523) 7.34E-03 Frequency of unenthusiasm/disinterest in past 2 weeksb 1.013 (1.007,1.019) 5.665 (1.982) 2.13E-03 – – – – For all analyses that met the MR sensitivity criteria, CAUSE results are presented, including an estimate of the causal effect size (c) with 95% credible intervals and the difference between the ELPD in the causal and in the sharing models (DELPD) with its standard error and one-sided P-value. P-values meeting a Bonferroni corrected threshold are highlighted in bold (0.05/41¼1.21E-03). MR, Mendelian randomization; CAUSE, causal analysis using summary effect estimates; ADHD, attention deficit/hyperactivity disorder; Crl, credible interval; ELPD, expected log pointwise posterior density; SE, standard error; P, P-value. a The unit reported for all continuous traits is SD, except for years of schooling, which was provided in years and cognitive performance, which was provided using a cognitive score.44 b The causal effect size (c) provided for these traits is odds ratio (OR) per ADHD log(OR) unit increase when ADHD is the exposure, since they were all analysed as categorical ordered, except for maternal smoking around birth, ever smoked, alcohol usually taken with meals and major depressive disorder, which were analysed as binary. c For parental longevity (combined parental attained age, Martingale residuals) a positive effect size indicates decreased attained age.47 395 396 International Journal of Epidemiology, 2023, Vol. 52, No. 2 ADHD liability causing the lifestyle and environmental Neurological, psychiatric and mental outcomes (Supplementary Table S5e). Also, high heteroge- health-related traits neity (I2 >70%) was detected for alcohol intake frequency, We found evidence of an increase in the odds of major de- age of first sexual intercourse and lifetime number of sex- pression per unit increase in the log(OR) of ADHD in the ual partners when ADHD was the exposure and for IVW analysis [OR ¼ 1.200, 95% CI=(1.122, 1.283), Townsend deprivation index in the other direction Table 1], which survived the sensitivity analyses and also (Supplementary Figure S1s, af, ah, au, Supplementary met the Bonferroni correction for multiple testing in Table S5e). After removing outliers there was still evidence CAUSE [OR ¼ 1.110, 95% CrI=(1.073, 1.148), DELPD of a causal effect and the heterogeneity decreased, although P ¼ 2.34e-04] (Figure 2, Table 2). An alternative and more it remained around 60% for age of first sexual intercourse, broad definition of depression, however, did not reach sta- lifetime number of sexual partners and Townsend depriva- tistical significance at any stage, and neither did the other tion index (Supplementary Table S5e). analyses undertaken (Figure 1; Supplementary Table S4). After multiple comparisons correction, CAUSE analyses provided evidence with ADHD as the exposure for lower odds of increased average total household income Personality and psychosocial factors [OR ¼ 0.966, 95% CrI ¼ (0.954, 0.979)], higher odds of a In the IVW analysis, we found evidence of an effect for increased lifetime number of sexual partners [OR ¼ 1.023, ADHD as exposure on one trait, with ADHD as outcome 95% CrI ¼ (1.013, 1.033)] and a negative effect in both for nine traits and in both directions for three traits directions for age first had sexual intercourse (Supplementary Table S4; Figure 1). The effect of the ge- [beta ¼ 0.058, 95% CrI ¼ (0.072, 0.044) and netic liability of ADHD on frequency of unenthusiasm/dis- OR ¼ 0.413, 95% CrI ¼ (0.372, 0.457) for ADHD as ex- interest in past 2 weeks met the sensibility analysis criteria, posure and outcome, respectively] (Figure 2, Table 2). but was only suggestive after multiple comparison correc- Evidence was also found for a positive effect of ADHD lia- tion in the CAUSE analysis (DELPD P ¼ 2.13e-03, Table 2) bility on current tobacco smoking [OR ¼ 1.015, 95% and evidence of reverse causation was found in the oppo- CrI ¼ (1.01, 1.021)], and a negative effect for past tobacco site direction (Supplementary Table S5h). The effect of the smoking in both directions [OR ¼ 0.963, 95% genetic liability of frequency of tiredness/lethargy in past CrI ¼ (0.947, 0.978), OR ¼ 0.588, 95% CrI ¼ (0.523, 2 weeks and of being able to confide on ADHD remained 0.664) for ADHD as exposure and outcome, respectively] after the sensitivity analyses, but only suggestive evidence (Figure 2, Table 2). In addition, time spent watching televi- was found for being able to confide in the CAUSE analyses sion seemed to increase significantly the odds of ADHD (DELPD P ¼ 7.34e-03, Table 2). [OR ¼ 1.862, 95% CrI ¼ (1.545, 2.246)] (Figure 2, Table 2). Suggestive evidence (DELPD P <0.05) was shown Overall, the direction of the causal effect size estimates in the CAUSE analysis for a causal effect of the genetic lia- was consistent between IVW MR and CAUSE for all analy- bility of ADHD on the remaining smoking-related traits ses undertaken and, in general, CAUSE effect size estimates and on all alcohol-related traits, except for alcohol intake were smaller than those estimated by IVW MR (Figure 3, versus 10 years previously, which showed an effect in the Tables 1 and 2). opposite direction, and for Townsend deprivation index, which showed an effect in both directions (Table 2). Discussion In the present study we assessed for the first time the causal Longevity relationships between ADHD and a broad range of ADHD- In the IVW analysis, there was no evidence of a causal effect related traits, applying complementary approaches. Through when ADHD was considered as outcome, but evidence for this comprehensive strategy we found consistent evidence an effect of ADHD as exposure was found in all longevity- across methods for a causal effect of the genetic liability of related traits, decreasing maternal, paternal and combined anthropometric measures and time spent watching television age of death or attained age and decreasing the odds of both on ADHD, for the genetic liability of ADHD on average parents being in the top 10% of survival (Figure 1, Table 1; household income and major depressive disorder, and a bidi- Supplementary Table S4). However, only effects on com- rectional relationship with educational achievement, smoking bined parental attained age met the sensitivity analysis crite- and sexual behaviour. ria and showed suggestive evidence of a causal relationship in Our findings give support to the relationship between the CAUSE analysis, although it did not meet multiple testing ADHD and risk-taking behaviours and to existing evidence correction (DELPD P ¼ 0.025, Table 2). indicating that ADHD is an entry point into a harmful life International Journal of Epidemiology, 2023, Vol. 52, No. 2 397 Figure 3 Causal effect estimates obtained by MR, with 95% CI, vs those obtained by CAUSE, with 95% CrI, in both directions. Effect sizes are provided as logOR for binary and categorical ordered traits. Different categories have been plotted in different colours, and traits with results meeting the Bonferroni corrected threshold in CAUSE are labelled. ADHD, attention deficit/hyperactivity disorder; CAUSE, causal analysis using summary effect estimates; CI, confidence interval; CrI, credible interval; MDD, major depressive disorder; MR, Mendelian randomization; TV, television; OR, odds ratio trajectory, where ADHD individuals are more likely to en- on alcohol dependence, they failed to find a causal connec- gage in behaviours that put them at risk for negative out- tion between ADHD and alcohol amount or alcohol use dis- comes, including smoking, problematic substance use or order.29 Differences between the traits considered, sample unsafe sexual behaviour.58,59 We confirm previous findings sizes or the way of measuring alcohol consumption may ac- for a causal effect of the genetic liability of ADHD on smok- count for inconsistent findings among studies. With respect ing behaviour.29 In particular, we found a positive effect of to sexual behaviour, we show, for the first time, evidence ADHD liability on frequency of current smoking and a neg- for a causal role of the genetic liability to ADHD on lower ative effect on past tobacco smoking, which indicates lower age at first sexual intercourse and on increased lifetime num- frequency of smoking in the past for individuals who were ber of sexual partners. In line with all these results, and with not heavy smokers when asked. These results may suggest additional evidence from observational literature, we also that those individuals who were heavy smokers in the past found suggestive evidence for a causal effect of ADHD ge- carried on being heavy smokers when asked, and support netic liability on decreased longevity.14 that ADHD genetic liability may increase frequency of Our results also support a positive effect of time spent smoking and make smokers less likely to give up, which watching television on ADHD, which goes in line with a agrees with a reported negative effect of ADHD liability in reported association between ADHD and television usage smoking cessation.29 A recent MR study, however, did not and with evidence from longitudinal studies reporting an ef- find evidence for a causal relationship from liability for fect of increased screen time worsening a child’s develop- ADHD to nicotine dependence, although their sample size ment and increasing risk for autism spectrum disorder.61–63 was more limited and these results may just reflect lack of In addition, our findings strengthen previous evidence statistical power.60 We also found suggestive evidence for linking ADHD with academic, employment and financial an effect of the genetic liability of ADHD increasing smok- problems.17,64 In fact we report, for the first time, consis- ing initiation and pack/years of smoking that support previ- tent evidence for a negative causal effect of ADHD liability ous findings.29,30 Also, despite not surpassing the strict on years of schooling, age completed full-time education Bonferroni correction applied, suggestive evidence of a and average total household income before tax, and sug- causal effect of the genetic liability of ADHD increasing al- gestive evidence of a positive effect on Townsend depriva- cohol intake frequency and making individuals less likely to tion index at recruitment. In addition, evidence was found have their alcohol intake with meals, was found, which sug- in the other direction for the educational traits and sugges- gests increased risk for prejudicial use of alcohol. Although tive evidence for Townsend deprivation index. Given that previous studies reported a weak effect of liability to ADHD ADHD is a neurodevelopmental disorder, the latest 398 International Journal of Epidemiology, 2023, Vol. 52, No. 2 relationships are temporally implausible and results may some spurious findings due to instrument mis- reflect previous studies showing that children in families specification. with lower parental education, family income or socioeco- ii. Aiming to avoid false-positive results, we designed a nomic status are at higher risk for ADHD or ADHD symp- strict analysis pipeline. We undertook a comprehen- toms.6,65,66 This is consistent with dynastic effects, when sive set of sensitivity analyses, including the weighted genetic variants in parents may affect the next generation mode, recently reported to maintain the correct type I indirectly through their effect on the environment rather error rate in a diverse set of scenarios but also to be than through the inherited DNA, affecting our results.67 too conservative, particularly for large sample sizes.70 For instance, it might be that parent’s socioeconomic status In addition, we applied a strict multiple testing correc- could influence parenting skills, social development or tion, despite the presence of correlated traits. stress levels and these, in turn, may impact on children’s iii. Despite undertaking this range of different analyses, mental health.66 each one under a different set of assumptions, and Along these lines, we also found evidence for liability of selecting only results that were consistent across meth- lower age at first sexual intercourse and of lower rate of ods, we still identified some temporally implausible past tobacco smoking in non-heavy smokers increasing the relationships. These associations could be explained odds of ADHD. Evidence from the literature has linked re- statistically because the instruments were used as lated traits such as young parental age or maternal smok- measures of the liability for a trait, not necessarily its ing with increased risk of ADHD in children,5,68,69 observed manifestation,70 although they may also in- suggesting that these chronological implausible results may dicate the invalidity of the genetic instruments. As dis- also be driven by dynastic effects. cussed above, some of them, such the effect of We also show evidence consistent with a positive effect socioeconomic status or smoking behaviour liability of anthropometric measures on ADHD, a finding which is on ADHD, are likely driven by dynastic effects. In ad- likely related to the effect of BMI liability on ADHD, dition, for those traits with evidence of bidirectional reported in a previous MR study.23 Later studies with lon- causality, we cannot rule out a scenario where most of gitudinal and family designs, however, pointed to this rela- the heritable variation of both exposure and outcome tionship being largely explained by a variety of is mediated by the same unobserved process, as ac- psychosocial factors and shared genetic and environmental knowledged by CAUSE authors.34 confounders, also including a role for parental education iv. Some of the scenarios where MR sensitivity analyses dynastic effects.24,25 have been carried out may not have been optimal for Our findings also confirm the effect of ADHD liability their performance: for instance, the limited number of on major depression and the lack thereof when using a genetic instruments available for ADHD (particularly broader definition,22 but no evidence supporting previous relevant for MR Egger) or the difference in sample size between the exposure and the outcome in some results of a causal relationship of ADHD and birthweight comparisons (relevant for Steiger filtering). It may be or intelligence was found. These discrepancies may be that under more suitable circumstances, MR sensitiv- explained by methodological differences between studies, ity analyses would be more efficient in detecting false- including: (i) the selection of genetic instruments and addi- positive results. tional covariates taken into account by authors considering v. In addition, genetic variants could be associated with birthweight18; or (ii) differences in sensitivity analyses un- more than one trait, which would make it difficult to dertaken when the protective effect of intelligence on ascertain which one is the true causal exposure. This ADHD was described.19 is particularly relevant when analysing correlated The results of the present study should be interpreted in traits, as it is the case in this study. Further sensitivity the light of several limitations, as follows. analyses, which were out of the scope of this study, ex- i. Given that we aimed to give an overview of potential cluding variants associated with other traits or under- causal relationships between ADHD and a consider- taking mediation analysis would contribute to deepen able number of related traits, using publicly available our understanding and provide more robust evidence summary statistics datasets, it was not feasible to tai- for the causal relationships identified. For instance, lor the analytical strategy separately for each trait or multivariable MR analyses have been used to detect to carefully curate each phenotype. This may have an effect of educational attainment mediating the rela- prevented us from identifying additional evidence for tionship between BMI and ADHD.25 causal relationships, as may be the case for birth- vi. We must be cautious when comparing effect sizes be- weight mentioned above,18 but also may have led to tween analyses with ADHD as exposure and outcome, International Journal of Epidemiology, 2023, Vol. 52, No. 2 399 since they are often presented in different scales, and depressive disorder, of the genetic liability of anthropomet- there are a number of assumptions that need to hold ric measures and time spent watching television on for reliable interpretation of causal effects for binary ADHD, and of a bidirectional relationship with educa- exposures.71 Also, due to differences in sample size, tional achievement, smoking and sexual behaviour. the power was often different between analyses, which Additional analyses with complementary study designs to in turn makes it difficult to establish a prevailing direc- avoid the effect of potential biases will be required to fol- tion of causality for traits with a bidirectional effect. low up these findings. However, our results may still con- vii. There are some methodological issues that should also tribute to explain part of the widespread co-occurring be considered. Given the large resources of GWAS traits and comorbid disorders across the lifespan of indi- summary statistics currently available and the flour- viduals with ADHD, and may open new opportunities for ishment of MR-related methods being developed, developing preventive strategies for ADHD and for nega- there is a huge potential for MR analyses to shed light tive ADHD trajectories. into the causal relationships between many complex traits. We should, however, also bear in mind their limitations when designing studies and interpreting Ethics approval results. In order to avoid the effect of horizontal plei- Ethics approval was not required for this study, since no new data otropy, which is relevant given that emerging results were generated and only summary statistics were analysed. from GWAS point to a large number of genetic var- iants being associated with multiple traits,51,71 we fol- Data availability lowed up MR findings with CAUSE.34 Overall we The data underlying this article are available in the MR-Base data- observed consistency in direction of effects between base (doi: 10.1101/2020.08.10.244293 and doi : 10.7554/ both methods, with smaller effect sizes estimated by eLife.34408.001). CAUSE than by MR and narrower intervals in gen- eral. The difference in effect sizes may be due to CAUSE also accounting for horizontal pleiotropy in Supplementary data its model. Although, CAUSE authors’ also acknowl- Supplementary data are available at IJE online. edge that in the event of causality and no correlated pleiotropy, their causal estimates tend to shrink to- wards zero in comparison with other methods, partly Author contributions due to prior distribution being centred at zero.34 M.R. and M.S.A. designed the study and the analytical strategy and Under these circumstances they also report lower wrote the manuscript. M.S.A. conducted the analysis. C.S-M., J.A.R-Q., L.V-R., M.R., M.S.A. and P.R. interpreted the findings mean squared error for CAUSE compared with MR if and reviewed critically the manuscript. causal effects are small, and there is also low power on the exposure, which seems to be the case when we consider ADHD as exposure.34 Overall, despite the Funding use of a range of complementary approaches in this This work was supported by European Union H2020 Programme study and of the evidence provided for causal relation- (H2020/2014–2020) under grant agreements no. 667302 (CoCA), ships supported by the literature and by alternative no. 728018 (Eat2beNICE), no. 848228 (DISCOvERIE) and no. study designs, such as the effect of ADHD on depres- 2020604 (TIMESPAN); by the Instituto de Salud Carlos III (PI18/ 01788, PI19/00721, P19/01224, PI20/00041, FI18/00285 to L.V.R., sion,22 it seems that some of our results may still have CD15/00199 to C.S.M, CP09/00119 and CPII15/00023 to M.R.); been affected by biases such as dynastic effects. This by the Pla estratègic de recerca i innovació en salut (PERIS), highlights the caution that must still be exerted when Generalitat de Cata-lunya (METAL-Cat; SLT006/17/287); by the interpreting MR findings and the need for other stud- Agència de Gestiód’Ajuts Universitaris i de Recerca AGAUR, ies with alternative designs, such as those in families,72 Generalitat de Catalunya (2017SGR1461, 2016FI_B00899 to P.R.); to triangulate their findings and confirm MR results. Ministry of Science, Innovation and Universities (IJC2018-035346-I to M.S.A); by the European Regional Development Fund (ERDF); and by ‘la Marató de TV3’ (092330/31) and by the ECNP Network Conclusion ‘ADHD across the Lifespan’ [https://www.ecnp.eu/researchinnova tion/ECNP-networks/List-ECNP-Networks/]. This paper reflects Our results are consistent with a causal effect of the genetic only the authors’ views, and the European Union is not liable for liability of ADHD on average household income and major any use that may be made of the information contained therein. 400 International Journal of Epidemiology, 2023, Vol. 52, No. 2 Conflict of interest 14. Sun S, Kuja-Halkola R, Faraone SV et al. Association of psychi- atric comorbidity with the risk of premature death among chil- J.A.R.Q. was on the speakers’ bureau and/or acted as consultant for dren and adults with attention-deficit/hyperactivity disorder. Janssen-Cilag, Novartis, Shire, Takeda, Bial, Shionogi, Sincrolab, JAMA Psychiatry 2019;76:1141–49. Novartis, BMS, Medice, Rubiand Raffo in the past 3 years. He also received travel awards (air tickets þ hotel) for taking part in psychi- 15. Davey Smith G, Hemani G. Mendelian randomization: genetic atric meetings from Janssen-Cilag, RubiShire, Takeda, Shionogi, anchors for causal inference in epidemiological studies. Hum Bial and Medice. The Department of Psychiatry chaired by him re- Mol Genet 2014;23:R89–98. ceived unrestricted educational and research support from the fol- 16. O’Connor LJ, Price AL. Distinguishing genetic correlation from lowing companies in the past 3 years: Janssen- Cilag, Shire, Oryzon, causation across 52 diseases and complex traits. Nat Genet Roche, Psious and Rubió. 2018;50:1728–34. 17. de Zeeuw EL, van Beijsterveldt CEM, Ehli EA, de Geus EJC, References Boomsma DI. Attention deficit hyperactivity disorder symptoms and low educational achievement: evidence supporting a causal 1. Polanczyk G, de Lima MS, Horta BL, Biederman J, Rohde LA. hypothesis. Behav Genet 2017;47:278–89. The worldwide prevalence of ADHD: a systematic review and 18. Orri M, Pingault J-B, Turecki G. Contribution of birth weight to metaregression analysis. Am J Psychiatry 2007;164:942–48. mental health, cognitive and socioeconomic outcomes: two- 2. Fayyad J, Sampson NA, Hwang I et al.; WHO World Mental sample Mendelian randomisation. Br J Psychiatry 2021;219: Health Survey Collaborators. The descriptive epidemiology of 507–514. DSM-IV Adult ADHD in the World Health Organization World 19. Savage JE, Jansen PR, Stringer S et al. Genome-wide association Mental Health Surveys. Atten Defic Hyperact Disord 2017;9: meta-analysis in 269,867 individuals identifies new genetic and 47–65. functional links to intelligence. Nat Genet 2018;50:912–19. 3. Faraone SV, Perlis RH, Doyle AE et al. Molecular genetics of 20. Zhu Z, Zhu X, Liu C-L et al. Shared genetics of asthma and men- attention-deficit/hyperactivity disorder. Biol Psychiatry 2005; tal health disorders: a large-scale genome-wide cross-trait analy- 57:1313–23. sis. Eur Respir J 2019;54:1901507. 4. Larsson H, Chang Z, D’Onofrio BM, Lichtenstein P. The herita- 21. Leppert B, Riglin L, Wootton RE et al. The effect of ADHD on bility of clinically diagnosed attention deficit hyperactivity disor- physical health outcomes: a two-sample Mendelian randomiza- der across the lifespan. Psychol Med 2014; 44:2223–29. tion study. Am J Epidemiol. 2020;190:1047–1055. 5. Faraone SV, Asherson P, Banaschewski T. Attention-deficit/hy- 22. Riglin L, Leppert B, Dardani C et al. ADHD and depression: in- peractivity disorder. Nat Rev Dis Prim 2015;1:15020. vestigating a causal explanation. Psychol Med 2020;51: 6. Larsson H, Sariaslan A, Långström N, D’Onofrio B, 1890–1897. Lichtenstein P. Family income in early childhood and subsequent 23. Martins-Silva T, Vaz JDS, Hutz MH et al. Assessing causality in attention deficit/hyperactivity disorder: a quasi-experimental the association between attention-deficit/hyperactivity disorder study. J Child Psychol Psychiatry 2014;55:428–35. and obesity: a Mendelian randomization study. Int J Obes 2019; 7. Franke B, Michelini G, Asherson P et al. Live fast, die young? A 43:2500–08. review on the developmental trajectories of ADHD across the 24. Donnchadha SÓ, Bramham J, Greene C. Rethinking the associa- lifespan. Eur Neuropsychopharmacol 2018;28:1059–88. tion between overweight/obesity and ADHD in children: a longi- 8. Kooij JJS, Bijlenga D, Salerno L et al. Updated European tudinal and psychosocial perspective. Ir J Psychol Med 2020; 75: Consensus Statement on diagnosis and treatment of adult 1–14. ADHD. Eur Psychiatry 2019;56:14–34. 25. Liu C-Y, Schoeler T, Davies NM et al. Are there causal relation- 9. Martel MM, Nigg JT, Lucas RE. Trait mechanisms in youth ships between attention-deficit/hyperactivity disorder and body with and without attention-deficit/hyperactivity disorder. J Res mass index? Evidence from multiple genetically informed Pers 2008;42:895–913. designs. Int J Epidemiol 2021;50:496–509 10. De Pauw SSW, Mervielde I. The role of temperament and per- 26. Byun Y-H, Ha M, Kwon H-J et al. Mobile phone use, blood lead sonality in problem behaviors of children with ADHD. J levels, and attention deficit hyperactivity symptoms in children: Abnorm Child Psychol 2011;39:277–91. a longitudinal study. PLoS One 2013;8:e59742. 11. Marx I, Hacker T, Yu X, Cortese S, Sonuga-Barke E. ADHD 27. Soler Artigas M, Sánchez-Mora C, Rovira P et al.; ADHD Group and the choice of small immediate over larger delayed rewards: a of the Psychiatric Genomics Consortium, International Cannabis comparative meta-analysis of performance on simple choice- Consortium. Attention-deficit/hyperactivity disorder and life- delay and temporal discounting paradigms. J Atten Disord 2021; time cannabis use: genetic overlap and causality. Mol Psychiatry 25:171–87. 2020;25:2493–503. 12. Shaw P, Stringaris A, Nigg J, Leibenluft E. Emotion dysregula- 28. Vilar-Ribó L, Sánchez-Mora C, Rovira P et al. Genetic overlap tion in attention deficit hyperactivity disorder. Am J Psychiatry and causality between substance use disorder and attention- 2014;171:276–93. deficit and hyperactivity disorder. Am J Med Genet B 13. Kessler RC, Adler LA, Berglund P et al. The effects of temporally Neuropsychiatr Genet 2020;186:140–150. secondary co-morbid mental disorders on the associations of 29. Treur JL, Demontis D, Davey Smith G et al. Investigating causal- DSM-IV ADHD with adverse outcomes in the US National ity between liability to ADHD and substance use, and liability to Comorbidity Survey Replication Adolescent Supplement (NCS- substance use and ADHD risk, using Mendelian randomization. A). Psychol Med 2014;44:1779–92. Addict Biol 2021; 26:e12849. International Journal of Epidemiology, 2023, Vol. 52, No. 2 401 30. Jang SK, Saunders G, Liu M, Jiang Y, Liu DJ, Vrieze S; 23andMe 44. Lee JJ, Wedow R, Okbay A et al.; Social Science Genetic Research Team. Genetic correlation, pleiotropy, and causal asso- Association Consortium. Gene discovery and polygenic predic- ciations between substance use and psychiatric disorder. Psychol tion from a genome-wide association study of educational attain- Med 2022;52:968–78. ment in 1.1 million individuals. Nat Genet 2018;50:1112–21. 31. Hemani G, Zheng J, Elsworth B et al. The MR-Base platform 45. Clifton EAD, Perry JRB, Imamura F et al. Genome-wide associa- supports systematic causal inference across the human phenome. tion study for risk taking propensity indicates shared pathways Elife 2018;7. https://elifesciences.org/articles/34408 (15 January with body mass index. Commun Biol 2018;1:36. 2022, date last accessed). 46. Demontis D, Walters RK, Martin J et al.; 23andMe Research 32. Elsworth B, Lyon M, Alexander T et al. The MRC IEU Team. Discovery of the first genome-wide significant risk loci for at- OpenGWAS data infrastructure. bioRxiv 2020. doi: tention deficit/hyperactivity disorder. Nat Genet 2019;51:63–75. 101101/20200810244293. http://biorxiv.org/content/early/2020/ 47. Pilling LC, Kuo C-L, Sicinski K et al. Human longevity: 25 ge- 08/10/2020.08.10.244293 (15 January 2022, date last accessed). netic loci associated in 389,166 UK biobank participants. Aging 33. Burgess S, Davey Smith G, Davies NM et al. Guidelines for per- (Albany NY) 2017;9:2504–20. forming Mendelian randomization investigations. Wellcome 48. Pierce BL, Burgess S. Efficient design for Mendelian randomiza- Open Res 2019;4:186.Available from: https://wellcomeopenre tion studies: subsample and 2-sample instrumental variable esti- search.org/articles/4-186/v2 (1 January 2022, date last accessed). mators. Am J Epidemiol 2013;178:1177–84. 34. Morrison J, Knoblauch N, Marcus JH, Stephens M, He X. 49. Hartwig FP, Davey Smith G, Bowden J. Robust inference in sum- Mendelian randomization accounting for correlated and uncor- mary data Mendelian randomization via the zero modal pleiot- related pleiotropic effects using genome-wide summary statistics. ropy assumption. Int J Epidemiol 2017;46:1985–98. 50. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent esti- Nat Genet 2020;52:740–47. 35. Bulik-Sullivan BK, Loh P-R, Finucane HK et al.; Schizophrenia mation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol 2016;40: Working Group of the Psychiatric Genomics Consortium. LD 304–14. Score regression distinguishes confounding from polygenicity in 51. Verbanck M, Chen C-Y, Neale B, Do R. Detection of widespread genome-wide association studies. Nat Genet 2015;47:291–95. Mar horizontal pleiotropy in causal relationships inferred from 36. Bulik-Sullivan B, Finucane HK, Anttila V et al.; Genetic Mendelian randomization between complex traits and diseases. Consortium for Anorexia Nervosa of the Wellcome Trust Case Nat Genet 2018;50:693–98. Control Consortium 3. An atlas of genetic correlations across 52. Bowden J, Davey Smith G, Burgess S. Mendelian randomization human diseases and traits. Nat Genet 2015;47:1236–41. with invalid instruments: effect estimation and bias detection 37. Buniello A, MacArthur JAL, Cerezo M et al. The NHGRI-EBI through Egger regression. Int J Epidemiol 2015;44:512–25. GWAS Catalog of published genome-wide association studies, 53. Burgess S, Thompson SG. Interpreting findings from Mendelian targeted arrays and summary statistics 2019. Nucleic Acids Res randomization using the MR-Egger method. Eur J Epidemiol 2019;47:D1005–12. 2017;32:377–89. 38. Elsworth BL, Mitchell R, Raistrick CA, Paternoster L, Hemani G, 54. Bowden J, Greco MF, Del Minelli C et al. Assessing the suitabil- Gaunt T. MRC IEU UK Biobank GWAS pipeline version 1. MRC ity of summary data for two-sample Mendelian randomization Integrative Epidemiology Unit, University of Bristol, 2017. https:// analyses using MR-Egger regression: the role of the I2 statistic. doi.org/10.5523/bris.2fahpksont1zi26xosyamqo8rr (13 February Int J Epidemiol 2016;45:1961–74. 2021, date last accessed). 55. Staiger D, Stock JH. Instrumental variables regression with weak 39. Berndt SI, Gustafsson S, Mägi R et al. Genome-wide meta- instruments. Econometrica 1997;65:557. analysis identifies 11 new loci for anthropometric traits and pro- 56. Hemani G, Tilling K, Davey Smith G. Orienting the causal rela- vides insights into genetic architecture. Nat Genet 2013;45: tionship between imprecisely measured traits using GWAS sum- 501–12. mary data. PLoS Genet 2017;13:e1007081. 40. Lambert JC, Ibrahim-Verbaas CA, Harold D et al.; European 57. Lee SH, Goddard ME, Wray NR, Visscher PM. A better coeffi- Alzheimer’s Disease Initiative (EADI). Meta-analysis of 74,046 cient of determination for genetic profile analysis. Genet individuals identifies 11 new susceptibility loci for Alzheimer’s Epidemiol 2012;36:214–24. disease. Nat Genet 2013;45:1452–58. 58. Charach A, Yeung E, Climans T, Lillie E. Childhood attention- 41. Nagel M, Watanabe K, Stringer S, Posthuma D, van der Sluis S. deficit/hyperactivity disorder and future substance use disorders: Item-level analyses reveal genetic heterogeneity in neuroticism. comparative meta-analyses. J Am Acad Child Adolesc Nat Commun 2018;9:905. Psychiatry 2011;50:9–21. 42. Howard DM, Adams MJ, Shirali M et al.; 23andMe Research 59. Schoenfelder EN, Kollins SH. Topical review: ADHD and Team. Genome-wide association study of depression phenotypes health-risk behaviors: toward prevention and health promotion. in UK Biobank identifies variants in excitatory synaptic path- J Pediatr Psychol 2016;41:735–40. ways. Nat Commun 2018;9:1470. 60. Vink JM, Treur JL, Pasman JA, Schellekens A. Investigating ge- 43. Wray NR, Ripke S, Mattheisen M et al.; Major Depressive netic correlation and causality between nicotine dependence and Disorder Working Group of the Psychiatric Genomics ADHD in a broader psychiatric context. Am J Med Genet B Consortium. Genome-wide association analyses identify 44 risk Neuropsychiatr Genet 2021;186:423–29. variants and refine the genetic architecture of major depression. 61. Lingineni RK, Biswas S, Ahmad N, Jackson BE, Bae S, Singh KP. Nat Genet 2018;50:668–81. Factors associated with attention deficit/hyperactivity disorder 402 International Journal of Epidemiology, 2023, Vol. 52, No. 2 among US children: results from a national survey. BMC Pediatr 67. Davies NM, Howe LJ, Brumpton B, Havdahl A, Evans DM, 2012;12:50. Davey Smith G. Within family Mendelian randomization stud- 62. Kushima M, Kojima R, Shinohara R et al.; Japan Environment ies. Hum Mol Genet 2019;28:R170–79. and Children’s Study Group. Association between screen time 68. Min X, Li C, Yan Y. Parental age and the risk of ADHD in off- exposure in children at 1 year of age and autism spectrum disor- spring: a systematic review and meta-analysis. Int J Environ Res der at 3 years of age. JAMA Pediatr 2022;176:384–91. Public Health 2021; 18:4939. 63. Madigan S, Browne D, Racine N, Mori C, Tough S. Association 69. Hvolgaard Mikkelsen S, Olsen J, Bech BH, Obel C. Parental age between screen time and children’s performance on a develop- and attention-deficit/hyperactivity disorder (ADHD). Int J mental screening test. JAMA Pediatr 2019;173:244–50. Epidemiol 2017;46:409–20. 64. Barkley RA. Major life activity and health outcomes associated 70. Hemani G, Bowden J, Haycock P et al. Automating Mendelian with attention-deficit/hyperactivity disorder. J Clin Psychiatry randomization through machine learning to construct a putative 2002;63(Suppl 1):10–15. causal map of the human phenome. bioRxiv 2017. doi: 65. Russell AE, Ford T, Williams R, Russell G. The association be- 101101/173682. https://www.biorxiv.org/content/early/2017/ tween socioeconomic disadvantage and attention Deficit/ 08/23/173682 (15 January 2022, date last accessed). Hyperactivity Disorder (ADHD): a systematic review. Child 71. Burgess S, Labrecque JA. Mendelian randomization with a bi- Psychiatry Hum Dev 2016;47:440–58. nary exposure variable: interpretation and presentation of causal 66. Torvik FA, Eilertsen EM, McAdams TA et al. Mechanisms link- estimates. Eur J Epidemiol 2018;33:947–52. ing parental educational attainment with child ADHD, depres- 72. Brumpton B, Sanderson E, Heilbron K et al.; 23andMe Research sion, and academic problems: a study of extended families in Team. Avoiding dynastic, assortative mating, and population The Norwegian Mother, Father and Child Cohort Study. J Child stratification biases in Mendelian randomization through Psychol Psychiatry 2020;61:1009–18. within-family analyses. Nat Commun 2020;11:3519.