Introduction
Attention-deficit hyperactivity disorder (ADHD) is a highly heritable common neurodevelopmental disorder with a worldwide prevalence of about 5% among school-aged children (Polanczyk, de Lima, Horta, Biederman, & Rohde, Reference Polanczyk, de Lima, Horta, Biederman and Rohde2007). Although the onset of ADHD is typically in childhood, ADHD often persists in adulthood. It has been shown that at least 15% of children with ADHD meet diagnostic criteria for ADHD also in adulthood (Faraone, Biederman, & Mick, Reference Faraone, Biederman and Mick2006). The average prevalence of ADHD in adults has been found to be 3–4% (Fayyad et al., Reference Fayyad, Graaf, Kessler, Alonso, Angermeyer, Demyttenaere and Jin2007). However, ADHD in adults is often underdiagnosed and/or untreated, which may lead to negative psychosocial and health consequences (Ginsberg, Quintero, Anand, Casillas, & Upadhyaya, Reference Ginsberg, Quintero, Anand, Casillas and Upadhyaya2014). Evidence from previous studies has shown substantial comorbidity between ADHD and both psychiatric and somatic medical conditions, e.g. hypertension, migraine, obesity, and type 2 diabetes, in adulthood (Brikell, Burton, Mota, & Martin, Reference Brikell, Burton, Mota and Martin2021; Chen et al., Reference Chen, Hartman, Haavik, Harro, Klungsøyr, Hegvik and Larsson2018; Kittel-Schneider et al., Reference Kittel-Schneider, Arteaga-Henriquez, Vasquez, Asherson, Banaschewski, Brikell and Reif2022). Yet, the comorbidities in undiagnosed or subclinical ADHD populations are less clear.
Although shared genetic factors have been implicated in explaining the link between ADHD and psychiatric outcomes (Brikell et al., Reference Brikell, Burton, Mota and Martin2021), little is known about the role of shared genetic mechanisms in ADHD and medical conditions. ADHD is highly heritable with an average heritability of 76% in twin studies (Faraone & Larsson, Reference Faraone and Larsson2019; Faraone et al., Reference Faraone, Perlis, Doyle, Smoller, Goralnick, Holmgren and Sklar2005) and the heritability attributed to single nucleotide polymorphisms (SNPs) identified in large genome-wide association studies (GWAS), e.g. SNP-heritability of 22% (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019). The genetic correlations based on the results of ADHD GWAS showed significant genetic overlap between ADHD and several mental health (e.g. depression, autism, subjective well-being, and neuroticism), as well as physical health traits and conditions (e.g. body mass index (BMI), obesity, HDL cholesterol, and type 2 diabetes) (Demontis et al., Reference Demontis, Walters, Athanasiadis, Walters, Therrien, Nielsen and Børglum2023, Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019). Another approach to investigate shared genetics across traits and disorders is by using polygenic risk scores (PRSs), which capture an individuals weighted sums of risk alleles as detected with a large-scale, independent GWAS (Lewis & Vassos, Reference Lewis and Vassos2020). Currently, PRS for ADHD (PRSADHD) based on the GWAS explains up to 5.5% of variance in ADHD and individuals in the top 10% of this PRS show five times higher odds of ADHD compared to individuals in the lowest PRS decile (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019). Although many individuals with high PRSADHD will not have ADHD diagnosis, PRSADHD has still great potential to identify individuals with high ADHD genetic liability in populations with low ADHD prevalence, potentially expressing a higher degree of ADHD-like traits (e.g. impulsivity, inattention).
Despite the strong advantage of exploiting electronic health records (EHR) in medical genetics research in biobanks (Abul-Husn & Kenny, Reference Abul-Husn and Kenny2019) there are currently few available reports on the associations between PRSADHD and EHR-based medical diagnoses. Such studies would help to generate hypothesis about potential underlying mechanisms, e.g. shared genetic etiology, confounding, or causal effects seen in studies of ADHD and somatic health comorbidities. A phenome-wide association study (PheWAS) is a hypothesis-free approach that can be used to test associations between a single predictor (e.g. genetic variants, PRS, and diagnosis) and a wide range of phenotypes, such as the EHR data (Pendergrass et al., Reference Pendergrass, Brown-Gentry, Dudek, Torstenson, Ambite, Avery and Ritchie2011). A PheWAS based on UK Biobank questionnaire and diagnosis data showed that PRSADHD was associated with several health-related traits (e.g. physical abuse, younger age at first sexual intercourse, smoking behavior, obesity, higher BMI, and several blood measures) (Leppert et al., Reference Leppert, Millard, Riglin, Davey Smith, Thapar, Tilling and Stergiakouli2020). Another EHR-based PheWAS in 10 000 Penn Medicine Biobank Cohort participants using more than 1800 phecodes over 6,6 years on average also found PRSADHD associations with medical conditions (e.g. tobacco use disorder, chronic airway obstruction, and type 2 diabetes) (Kember et al., Reference Kember, Merikangas, Verma, Verma, Judy, Abecasis and Bućan2021).
We aimed to take the hypothesis-free PheWAS approach in a large population-based biobank (Estonian Biobank; EstBB) to identify the associations between ADHD genetic liability and lifetime history of medical conditions based on ICD-10 diagnoses from EHRs in individuals without a history of ADHD diagnosis (PRSADHD analysis). Considering the sex differences in the prevalence of ADHD (Biederman, Faraone, Monuteaux, Bober, & Cadogen, Reference Biederman, Faraone, Monuteaux, Bober and Cadogen2004; Gaub & Carlson, Reference Gaub and Carlson1997), we also investigated potential sex differences for the associations between PRSADHD and medical conditions. Additionally, we compared the direction and strength of the identified ICD-10 associations that passed multiple corrections to the associations between these medical conditions and EHR-based ADHD diagnosis (ADHD diagnosis analysis) as well as to the high ADHD risk based on an adult ADHD brief screening instrument (questionnaire-based ADHD analysis). Although existing evidence shows substantial comorbidity and genetic correlation between ADHD and depression, as well as depression and other medical conditions (Carney & Freedland, Reference Carney and Freedland2017; Demontis et al., Reference Demontis, Walters, Athanasiadis, Walters, Therrien, Nielsen and Børglum2023, Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019; Kan et al., Reference Kan, Pedersen, Christensen, Bornstein, Licinio, MacCabe and Rijsdijk2016; Katzman, Bilkey, Chokka, Fallu, & Klassen, Reference Katzman, Bilkey, Chokka, Fallu and Klassen2017; Levey et al., Reference Levey, Stein, Wendt, Pathak, Zhou, Aslan and Gelernter2021; Rajan et al., Reference Rajan, McKee, Rangarajan, Bangdiwala, Rosengren and Gupta2020), it is unknown to what degree depression may mediate ADHD-medical condition link in adults. Therefore, our secondary aim was to explore the mediation effect of depression in the identified PRS – medical condition associations.
Methods and materials
Study population
The Estonian Biobank (EstBB) is a large data-rich population-based biobank (Leitsalu et al., Reference Leitsalu, Haller, Esko, Tammesoo, Alavere, Snieder and Metspalu2015), covering approximately 20% of the adult population in Estonia (N = ~210 000; 66% females; mean birth year 1971). All EstBB participants have signed an informed consent form and provided blood samples for genotyping. The EHR data is regularly retrieved by linking to the national health databases and registries, such as the National Health Insurance Funds (NHIF) database, cause of death register, and hospital records. Estonia has universal health care, covering more than 95% of the population. The research project has obtained approval from the Estonian Council on Bioethics and Human Research. More details about the recruitment of participants can be found in Supplementary information.
Phenotype data
Medical conditions via ICD-10 diagnoses
Medical diagnosis data was drawn from the EHRs based on the NHIF́s database covering the period from 2004 to 2020. NHIF́s database is a nationwide system integrating data from all healthcare providers in Estonia. EHR data included 2004 ICD-10 codes, of which 489 were excluded as these were not medical diagnoses or were in a very low prevalence (online Supplementary Table S1). In total, 1515 phenotypes from all ICD-10 disease categories (online Supplementary Table S2) were included in the analyses. ADHD diagnosis was defined by ICD-10 diagnosis code F90 (including diagnoses F90.0, F90.1, F90.8, and F90.9).
High ADHD risk based on adult ADHD self-report scale
As a sensitivity analysis, we included self-reported ADHD symptoms score, measured with the modified Estonian version of Adult ADHD Self-Report Scale (ASRS v1.1) part A (Kessler et al., Reference Kessler, Adler, Ames, Demler, Faraone, Hiripi and Walters2005). ASRS part A is a brief six-item screening questionnaire and items are assessed on Likert scale ranging from 0 to 4: 0 = never; 1 = rarely; 2 = sometimes; 3 = often; 4 = very often. A score cutoff of 14 out of 24 represents individuals with higher risk for ADHD (Kessler et al., Reference Kessler, Adler, Gruber, Sarawate, Spencer and Van Brunt2007). ASRS data in EstBB is available for 86 245 participants (Ojalo et al., Reference Ojalo, Haan, Kõiv, Kariis, Krebs, Uusberg and Lehto2024).
ADHD polygenic risk scores
To compute the PRS, the PRS-CS-auto algorithm was first used to apply a Bayesian regression framework to infer posterior effect sizes of SNPs by using the summary results from a large Psychiatric Genomics Consortium ADHD GWAS (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019) and an external linkage disequilibrium reference panel (Ge, Chen, Ni, Feng, & Smoller, Reference Ge, Chen, Ni, Feng and Smoller2019). Per-individual risks were further calculated with plink 2.0 (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015). The PRSADHD was standardized for the PheWAS analysis and categorized into deciles for subsequent analyses to test differences in high v. low/middle genetic liability groups. Prior to the main analysis, we tested the association between PRSADHD and ICD-10 ADHD diagnosis history. The variance explained by the PRSADHD was computed by subtracting the Nagelkerke pseudo R2 from the full (PRS, birth year, recruitment year, sex, and 10 genetic principal components (PCs)) and the reduced (birth year, recruitment year, sex, and 10 PCs) logistic regression models. Details about the genotyping and imputation procedures in the EstBB can be found in Supplementary information.
Statistical analysis
Main analysis
PheWAS was implemented to test for the associations between PRSADHD and 1515 ICD-10 diagnosis codes in the undiagnosed subsample where all individuals with a history of ADHD diagnosis (F90*) were removed (N = 464). Individuals with ADHD diagnosis were excluded because compared to the PRSADHD analysis sample, they were significantly younger and had likely received treatment or other support to manage their ADHD-related risks. Analyses were performed as described in our pre-registered protocol (Haan, Krebs, Võsa, & Lehto, Reference Haan, Krebs, Võsa and Lehto2021). Analyses were conducted using logistic regression and adjusted for sex (except in sex-stratified analyses), birth- and recruitment year to adjust for birth cohort effects and different EstBB recruitment strategies used across two decades, and 10 PCs to adjust for genetic ancestry (except in secondary analyses). Relatedness was handled by excluding related individuals separately in the PRSADHD and ADHD diagnosis analysis samples (PI-HAT > 0.2) which remained 111 261 individuals in the PRSADHD analysis sample and 111 601 individuals in the ADHD diagnosis sample. Additionally, for comparison we ran the main analyses in the full cohort without excluding related individuals (N = 196 935). We corrected for multiple testing by applying Bonferroni correction (p-value threshold 3.3 × 10−5). PheWAS and mediation analyses were run in R (version 3.6.0). PheWAS results were visualized using the PheWAS library https://github.com/PheWAS/PheWAS. Other analyses were performed using Stata v.17.
PRSADHD deciles
We further examined the associations that survived the multiple testing correction in the PheWAS by using PRSADHD deciles (ranging from 10th to 90th percentiles) to explore the difference in the odds of being diagnosed with a specific medical condition in high v. low (10th v. 1st PRS decile) and in high v. middle (10th v. 5th PRS decile) ADHD genetic liability groups. While the difference in the top and bottom genetic liability groups reflects the differences between the two extreme ends of ADHD genetic liability, the top v. middle comparison reflects the differences between the highest and the population average ADHD genetic liability groups. Additionally, we explored sex differences in the top and bottom PRS decile analysis.
Secondary analyses
Comparison of effects across PRSADHD, ADHD diagnosis, and questionnaire-based ADHD analyses
To shed light on the different patterns in medical condition risk in biobank participants with high ADHD genetic liability but without an ADHD diagnosis, EHR-based ADHD diagnosis and clinically relevant ADHD symptoms, we conducted two sets of additional logistic regression analyses only on these medical outcomes that passed the multiple testing correction in the PRSADHD PheWAS analysis. First, we used the EHR-based ADHD diagnosis as the predictor and second, we used the questionnaire-based high ADHD risk as the predictor.
Causal mediation analysis with depression diagnosis
Considering the well-established link between depression, ADHD and more severe somatic health conditions, we ran causal mediation analysis for the identified phenotypes and included lifetime depressive episode diagnosis as a mediator. Lifetime depressive episode diagnosis was defined based on the ICD-10 diagnosis codes F32* and F33*. Causal mediation analysis was run using the R package ́mediatioń and 1000 simulations was used for calculating estimates (Tingley, Yamamoto, Hirose, Keele, & Imai, Reference Tingley, Yamamoto, Hirose, Keele and Imai2014). Causal mediation analysis provides estimates for the average causal mediation effect, average direct effect, and total effect. Here we were interested in quantifying the proportion of effects on the medical conditions that were mediated by depression.
Results
A descriptive overview of the study samples is shown in Table 1. The main analysis study sample (PRSADHD) size was 111 261 individuals, after exclusion of related individuals (N = 85 674), ADHD cases (N = 464), and individuals with missing EHR records (N = 2887). In total 65% of the study sample was female and the mean birth year was 1970. In the ADHD diagnosis analysis sample, the sample size was 111 601, of which 464 were diagnosed ADHD cases (0.5%). In the questionnaire-based ADHD sample, 3556 participants screened positive for high risk for ADHD (8%). Cross-group comparisons indicate that although the female–male ratio was similar in all three samples, individuals in the ADHD diagnosis analysis sample were considerably younger (mean birth year 1990) compared to the two other study samples (mean birth year 1970 for PRSADHD analysis sample and 1979 for questionnaire-based ADHD analysis sample).
Table 1. Descriptive overview of study samples
Note: *Positive screening; **based on the complete set subset (N ADHD PRS = 76 927; N ADHD diag = 211/77 064; N ADHD ASRS = 2662/35 388), education was categorized as primary (1–4 grades, 5–9 grades, or no education; secondary (10–12 grades and vocational education); higher (university and research degree); ***data available for 211 participants in the ADHD diagnosis analysis sample and for 2662 participants who screened positive for ADHD in the questionnaire-based ADHD analysis sample; PRS, polygenic risk score; SD, standard deviation.
PheWAS of PRSADHD
First, we tested the association of the PRSADHD with ADHD diagnosis (OR 1.34, CI 1.22–1.47) and questionnaire-based ADHD (OR 1.06, CI 1.02–1.09). Results indicate that 1 s.d. increase in PRSADHD corresponds to 34% increase in the odds of ADHD diagnosis and 6% increase in the odds of higher risk for self-reported questionnaire-based ADHD. PRSADHD explained 0.6% variance in EHR-based ADHD diagnosis and 0.04% in questionnaire-based high-risk ADHD.
Our PheWAS showed evidence of association between PRSADHD and for 80 ICD-10 diagnoses after correction for multiple testing (Figure 1, Table 2, Table S3). Overall, the top five medical conditions with the strongest evidence for associations were chronic obstructive pulmonary disease (COPD) (OR 1.15, CI 1.11–1.18), obesity (OR 1.13, CI 1.11–1.15), type 2 diabetes (OR 1.11, CI 1.09–1.14), dorsalgia (OR 1.08, CI 1.07–1.10), and polyarthrosis (OR 1.09, CI 1.07–1.12).
Figure 1. Associations between PRSADHD and ICD-10 codes after adjustment for multiple testing.
Note: The X axis indicates groups of ICD10 main codes colored respectively and Y axis −log10 of the p values. Each triangle in the plot represents one ICD-10 main code and the direction of the triangle represents direction of effect. Red line − Bonferroni-corrected significance level (3.3 x 10−5). Phenotypes passed the Bonferroni correction from the lowest to highest p-values: E66, ‘obesity'; M54, ‘dorsalgia'; M15, ‘polyarthrosis'; J44, ‘chronic obstructive pulmonary diseases'; E11, ‘non-insulin dependent diabetes'; O04, ‘medical abortion'; F32, ‘depressive episode'; R51, ‘headache'; M77, ‘other enthesopathies'; F10, ‘mental and behavioral disorders due to use of alcohol'; M51, ‘other intervertebral disc disorders'; J20, ‘acute bronchitis'; F41, ‘other anxiety disorders'; G56, ‘mononeuropathies of upper limb'; M79, ‘other soft tisuse disorders'; M75, ‘shoulder lesion'; D22, ‘melanocytic naevi'; G47, ‘sleep disorders'; K29, ‘gastritis and duodenitis'; I11, ‘hypertensive heart disease'; J45, ‘asthma'; M25, ‘other joint disorders'; I10, ‘primary hypertension'; J03, ‘acute tonsillitis'; M50, ‘cervical disc disorders'; M16, ‘coxarthrosis'; G44, ‘other headache syndromes'; K21, ‘gastro-oesophageal reflux disease'; M19, ‘other arthrosis'; M70, ‘soft tissue disorders related to use, overuse and pressure'; J04, ‘acute laryngitis and tracheitis'; K04, ‘diseases of pulp and periapical tissues'; I50, ‘heart failure'; G43, ‘migraine'; I21, ‘acute myocardial infarction'; B86, ‘scabies'; M10, ‘gout'; A63, ‘other predominantly sexually transmitted diseases'; L82, ‘seborrhoeic keratosis'; R10, ‘abdonminal and pelvic pain'; K05, ‘gingivitis and periodontal diseases'; F43, ‘reaction to severe stress', M47, ‘spondylosis'; N30, ‘cystitis'; M48, ‘other spondylopathies'; F33, ‘recurrent depressive disorder'; N71, ‘inflammatory disease of uterus'; I48, ‘atrial inflammation of vagina and vulva'; M65, ‘synovitis and tenosynovitis'; R11, ‘nausea and vomiting'; F07, ‘personality and behavioral disorders due to brain damage'; F06, ‘other mental disorders due to brain damage'; A56, ‘other sexually transmitted chlamydial diseases'; K80, ‘cholelithiasis'; R73, ‘elevated blood glucose level'; N92, ‘excessive, frequent and irregulaar menstruation'; K86, ‘other diseases of pancreas'; K20, ‘oesophagitis'; O20, ‘haemorrhage in early pregnancy'; M13, ‘other arthritis'; L02, ‘cutaneous abscess, furuncle and carbuncle'; M96, ‘postprocedural musculoskeletal disorders'; K25, ‘gastric ulcer'; O99, ‘other maternal diseases'; Z95, ‘presence of cardiac and vascular implants and grafts'; K76, ‘other diseases of liver'; T51, ‘toxic effect of alcohol'; N70, ‘salpingitis and oophoritis'; M17, ‘gonarthrosis'; O26, ‘maternal care for other conditions related to pregnancy'; O23, ‘infections of genitourinary tract in pregnancy'; O47, ‘false labor'; K26, ‘duodenal ulcer'; O00, ‘ectopic pregnancy'; M06, ‘other rheumatoid arthritis'; I70, ‘atherosclerosis'; H36, ‘retinal disorders in diseases classified elsewhere'; B07, ‘viral warts'; I25, ‘chronic ischaemic heart disease'.
Table 2. Associations between PRSADHD and ICD-10 codes after adjustment for multiple testing
Note: n, number of cases; CI, confidence interval.
Sex-stratified analyses showed that for females, the top five associated medical conditions were obesity (OR 1.14, CI 1.11–1.16), polyarthrosis (OR 1.10, CI 1.08–1.13), medical abortion (OR 1.10, CI 1.08–1.13), dorsalgia (OR 1.09, CI 1.07–1.09), and headache (OR 1.08, CI 1.06–1.10). For males, the top five associated medical conditions were COPD (OR 1.17, CI 1.12–1.23), mental and behavioral disorders due to alcohol use (OR 1.11, CI 1.07–1.15), obesity (OR 1.11, CI 1.07–1.15), type 2 diabetes (OR 1.10, CI 1.06–1.14), and dorsalgia (OR 1.07, CI 1.04–1.09). Results are shown in online Supplementary Table S4 and Figures S1-S2.
The results in the full cohort (relatives not excluded) were largely similar, but additionally 52 associations passed correction for multiple testing potentially due to larger sample size and non-independence between observations (online Supplementary Table S5).
PRSADHD deciles
The PRSADHD decile analysis of the 80 significantly associated medical conditions showed that individuals in the top PRSADHD decile had 70% higher risk for COPD (OR 1.70; CI 1.48–1.95); 59% higher risk for obesity (OR 1.59; CI 1.47–1.73), 51% higher risk for toxic effects of alcohol (OR 1.51; CI 1.16–1.97), 45% higher risk for type 2 diabetes (OR 1.45; CI 1.30–1.61), and 34% higher risk for polyarthrosis (OR 1.34, CI 1.23–1.47), compared to individuals in the lowest PRSADHD decile. The results using top v. middle PRSADHD decile were mostly in line with the results from top v. bottom PRSADHD decile analysis, but were non-significant for 23 medical conditions, e.g. other sexually transmitted diseases, heart failures, atherosclerosis, acute bronchitis, other arthritis, and pregnancy related conditions (Fig. 2; online Supplementary Tables S6-S7). The results comparing the high PRSADHD group separately in females and males are shown in Fig. 3 and online Supplementary Tables S8-S9. Eleven phenotypes were only available in females (such as pregnancy-related conditions). The majority of the associations between females and males were in the same direction. Of the phenotypes that were available in both genders, 15 associations were significant only in females (including, cystitis, cholelithiasis, gastro-oesophageal reflux disease (GORD), migraine, gastric ulcer, other arthrosis, other joint disorders, other spondylopathies, abdominal and pelvic pain, and other sexually transmitted disease (STD) diagnoses). In contrast, associations with personality and behavioral disorders due to brain damage/dysfunction, toxic effects of alcohol, and cardiac/vascular implants and grafts were only significantly associated in males.
Figure 2. Comparison of associations between top vs bottom and top vs medium PRSADHD risk on ICD-10 codes after adjustment for multiple testing.
Note: PRS, ‘polygenic risk score'; 95% CI, ‘95% confidence intervals'.
Figure 3. Comparison of associations between high PRSADHD risk and ICD-10 codes in females and males.
Note: PRS, ‘polygenic risk score'; 95% CI, ‘95% confidence intervals'; ICD-10 codes less than 10 cases were excluded from analyses.
Secondary analyses
Comparison of effects across PRSADHD, ADHD diagnosis, and questionnaire-based ADHD analyses
In the ADHD diagnosis analysis, 18 medical conditions had insufficient case numbers (<10 cases) to be included in these analyses (online Supplementary Table S10). We observed significant associations between ADHD diagnosis and 31 medical conditions, of which 6 medical conditions showed protective associations (e.g. other STDs, melanocytic nevus, excessive and irregular menstruation, medical abortion, and pregnancy-related conditions) (Fig. 3). Overall, 40% of the PRSADHD associations were also observed in the ADHD diagnosis analysis. We observed the strongest evidence for associations with psychiatric disorders, e.g. behavioral disorders due to brain disease (OR 6.42, CI 4.08–10.11), reaction to severe stress (OR 4.70, CI 3.83–5.76), recurrent depressive disorders (OR 3.97, CI 3.08–5.11), but also with asthma (OR 2.11, CI 1.64–2.70) and acute bronchitis (OR 1.72, CI 1.43–2.06).
In the questionnaire-based ADHD analysis, the direction of the effects was largely similar as in the PRS decile analysis and 78% of the PRSADHD associations were observed in the questionnaire-based ADHD analysis. We observed the strongest evidence for associations with recurrent depressive disorders (OR 3.78, CI 3.44–4.16), anxiety disorders (OR 2.31, CI 2.14–2.50), polyarthrosis (OR 1.94, CI 1.70–2.20), sleep disorders (OR 1.94, CI 1.77–2.13), and asthma (OR 1.64, CI 1.48–1.81). These results are shown in Fig. 4 and online Supplementary Tables S10-S11.
Figure 4. Comparison of associations between ADHD diagnosis and questionnaire-based ADHD on ICD-10 codes after adjustment for multiple testing.
Note: PRS, ‘polygenic risk score'; 95% CI, ‘95% confidence intervals'; ICD-10 codes less than 10 cases were excluded from analyses.
Causal mediation analysis
Our findings from causal mediation analysis showed that depression was a significant mediator for all 78 medical conditions (two depression diagnoses were excluded from the analysis), but the proportion of the effect between PRSADHD and medical conditions that was mediated by depression for majority of outcomes was generally small, ranging from 2 to 16%. Only for mental disorders, e.g. sleep disorders, anxiety disorders, reaction to severe stress, and mental disorders due to brain damage/dysfunction, the mediation effect accounted for more than 20% of the observed associations between PRSADHD and respective ICD-10 diagnosis (online Supplementary Table S12).
Discussion
In this study, we performed a PheWAS between PRSADHD and EHR-based diagnoses across all ICD-10 categories, using nationwide, population-representative clinical data with up to 17 years follow-up. This approach enabled the investigation of the associations between genetic liability to ADHD and medical conditions in a population without a history of ADHD diagnosis. Overall, our results indicate robust associations between ADHD genetic liability and 80 medical conditions, with obesity, dorsalgia, polyarthrosis, COPD, and type 2 diabetes among the top associated medical conditions.
These results are in line with previous findings on the associations between ADHD diagnosis and comorbid medical conditions (Du Rietz et al., Reference Du Rietz, Brikell, Butwicka, Leone, Chang, Cortese and Larsson2021; Momen et al., Reference Momen, Plana-Ripoll, Agerbo, Benros, Børglum, Christensen and McGrath2020). Furthermore, similarly to our findings, other studies using PRSADHD (Kember et al., Reference Kember, Merikangas, Verma, Verma, Judy, Abecasis and Bućan2021; Leppert et al., Reference Leppert, Millard, Riglin, Davey Smith, Thapar, Tilling and Stergiakouli2020) also showed associations between PRSADHD, mental and physical health problems, indicating a shared genetic background between ADHD and medical conditions. Another large-scale genetic study, based on UK Biobank data, found that genetic liability to multiple complex traits in adulthood, such as type 2 diabetes, obesity, peripheral artery disease, and polyarthritis, increases the risk for ADHD in childhood, which further shows the role of underlying genetic effects between physical health and ADHD across development (García-Marín et al., Reference García-Marín, Campos, Cuéllar-Partida, Medland, Kollins and Rentería2021a).
Several of the observed associations between PRSADHD and medical conditions in non-diagnosed individuals are consistent with the few previous genetically informative studies on diagnosed ADHD and medical comorbidities, such as the nervous system (Du Rietz et al., Reference Du Rietz, Brikell, Butwicka, Leone, Chang, Cortese and Larsson2021), respiratory (Du Rietz et al., Reference Du Rietz, Brikell, Butwicka, Leone, Chang, Cortese and Larsson2021; García-Marín et al., Reference García-Marín, Campos, Kho, Martin, Cuéllar-Partida and Rentería2021b), musculoskeletal (Du Rietz et al., Reference Du Rietz, Brikell, Butwicka, Leone, Chang, Cortese and Larsson2021), and cardiometabolic diseases (Du Rietz et al., Reference Du Rietz, Brikell, Butwicka, Leone, Chang, Cortese and Larsson2021; Garcia-Argibay et al., Reference Garcia-Argibay, du Rietz, Lu, Martin, Haan, Lehto and Brikell2022). Several studies have demonstrated that some of the observed associations between medical conditions and ADHD diagnosis were mediated by lifestyle factors, such as tobacco use and alcohol misuse (Faraone et al., Reference Faraone, Banaschewski, Coghill, Zheng, Biederman, Bellgrove and Wang2021; Garcia-Argibay et al., Reference Garcia-Argibay, du Rietz, Lu, Martin, Haan, Lehto and Brikell2022), which may reflect the impulsive behavior characteristic to ADHD. Although we focused only on undiagnosed individuals in this study, the PRSADHD likely captures the genetic predisposition to ADHD-related traits, such as impulsivity and disorganization, which are strongly associated with risk behavior and unhealthy lifestyle (Solmi et al., Reference Solmi, Civardi, Corti, Anil, Demurtas, Lange and Carvalho2021). Therefore, future studies should explore whether health-related behavior mediates the associations between ADHD genetic predisposition and medical conditions in different populations where detailed information on such behavior is available through questionnaires (lifestyle) or electronic health records (e.g. missing medical appointments and accident-related injuries). However, given the strong comorbidity between ADHD and depression, we utilized the diagnosis information to test mediation by depression diagnosis. As expected, we found that depression was a significant mediator for all observed medical conditions, but the effect from depression alone was rather small. Additionally, it is possible that adults with undiagnosed or subclinical ADHD may be more likely diagnosed with depression, either because of potentially expressed symptoms overlapping with depression (e.g. concentration problems, inattention) or being in an increased depression risk due to their unmanaged ADHD.
A recent Mendelian randomization study showed a bidirectional causal effect between ADHD and obesity-related traits (e.g. waist-hip-ratio (WHR) and BMI-adjusted WHR) and reported genetic overlap between BMI and ADHD (Karhunen et al., Reference Karhunen, Bond, Zuber, Hurtig, Moilanen, Järvelin and Rodriguez2021). It has been also shown that obesity increases the risk for a chronic inflammatory state (Andersen, Murphy, & Fernandez, Reference Andersen, Murphy and Fernandez2016), which can lead to several diseases, such as hypertension, type 2 diabetes, cardiovascular disease, and musculoskeletal diseases (Furman et al., Reference Furman, Campisi, Verdin, Carrera-Bastos, Targ, Franceschi and Slavich2019; Nikiphorou & Fragoulis, Reference Nikiphorou and Fragoulis2018). There is also some evidence that inflammation plays an important role in the etiology of ADHD (Dunn, Nigg, & Sullivan, Reference Dunn, Nigg and Sullivan2019). It has been reported that individuals with ADHD had an increased inflammatory marker interleukin 6 (IL-6) which has been linked with an increased risk of diabetes (Donfrancesco et al., Reference Donfrancesco, Nativio, Di Benedetto, Villa, Andriola, Melegari and Di Trani2020; Wang et al., Reference Wang, Bao, Liu, Ouyang, Wang, Rong and Liu2013). Similarly, the role of obesity and inflammation markers are well described in the etiology of cholelithiasis and GORD (El-Serag, Reference El-Serag2008; Littlefield & Lenahan, Reference Littlefield and Lenahan2019). Therefore, it is plausible that obesity can act as a mediator on the pathway between ADHD and several medical conditions. In fact, the study by Garcia-Argibay et al. found that BMI-mediated associations between PRSADHD and medical conditions like hypertension, type 2 diabetes, migraine, and sleep disorders (Garcia-Argibay et al., Reference Garcia-Argibay, du Rietz, Lu, Martin, Haan, Lehto and Brikell2022).
The sex-stratified analysis generally showed similar associations in males and females. Several conditions that were statistically significant only in one gender (e.g. diseases of the musculoskeletal and digestive system, personality and behavioral disorders due to brain damage/dysfunction, toxic effects of alcohol, and cardiac/vascular implants and grafts), we observed similar effect sizes in both genders, potentially indicating insufficient case numbers in the sex-stratified analyses. However, we also observed some sex differences. The association with sexually transmitted chlamydial diseases was more strongly associated with females. The literature supports the link between ADHD and risky sexual behavior. Several studies have shown that ADHD is associated with higher risk of sexually transmitted diseases and more and younger age at first pregnancies (Hechtman et al., Reference Hechtman, Swanson, Sibley, Stehli, Owens, Mitchell and Stern2016; Hosain, Berenson, Tennen, Bauer, & Wu, Reference Hosain, Berenson, Tennen, Bauer and Wu2012). The study by Leppert et al. (Reference Leppert, Millard, Riglin, Davey Smith, Thapar, Tilling and Stergiakouli2020) also demonstrated the association between PRSADHD and younger age at first sexual intercourse in the UK Biobank. Although the prevalence of chlamydia is similar in females and males, the disease is more often asymptomatic in males. Considering that women are more frequently screened in a routine gynecological care, it is possible that the stronger association observed in females can be explained by higher likelihood of receiving a diagnosis (Dielissen, Teunissen, & Lagro-Janssen, Reference Dielissen, Teunissen and Lagro-Janssen2013). Other associations observed more strongly in females were with cystitis, migraine, cholelithiasis, abdominal and pelvic pain, and GORD diagnoses. Cystitis, migraine, abdominal and pelvic pain, and cholelithiasis have been reported to be more common in women, which has been suggested to be explained by female hormones and inflammation processes (Allais et al., Reference Allais, Chiarle, Sinigaglia, Airola, Schiapparelli and Benedetto2020; Curran, Reference Curran2015; Littlefield & Lenahan, Reference Littlefield and Lenahan2019; Patnaik et al., Reference Patnaik, Laganà, Vitale, Butticè, Noventa, Gizzo and Dandolu2017). However, although GORD symptoms are equally prevalent in males and females (El-Serag, Reference El-Serag2008), it has been suggested that estrogen may mediate the association in females (Nilsson, Johnsen, Ye, Hveem, & Lagergren, Reference Nilsson, Johnsen, Ye, Hveem and Lagergren2003).
Although the majority of the associations observed in the PRSADHD analysis sample were in the same direction as in the ADHD diagnosis and questionnaire-based ADHD analysis samples, we observed opposite effects for some conditions in the ADHD diagnosis sample (such as conditions only expressed in females, e.g. pregnancy-related conditions). It is possible that the protective associations observed with these conditions could be affected by treatment effects, potentially reducing the likelihood of psychosocial or environmental risk factors (Fuller-Thomson & Lewis, Reference Fuller-Thomson and Lewis2015), and altering reproductive behavior (Østergaard, Dalsgaard, Faraone, Munk-Olsen, & Laursen, Reference Østergaard, Dalsgaard, Faraone, Munk-Olsen and Laursen2017; Skoglund et al., Reference Skoglund, Kopp Kallner, Skalkidou, Wikström, Lundin, Hesselman and Sundström Poromaa2019). Additionally, it is possible that the diagnosed ADHD sample in EstBB is healthier, more educated, and practicing healthy lifestyle compared to the average ADHD patient.
Our results suggest that ADHD-related medical conditions are also present in individuals without ADHD diagnosis but who have high genetic liability for ADHD. Evidence suggests that ADHD in adults is likely underdiagnosed and undertreated across countries (Fayyad et al., Reference Fayyad, Sampson, Hwang, Adamowski, Aguilar-Gaxiola and Al-Hamzawi2017). For example, ADHD lifetime prevalence in EstBB is 0.5% and yearly prevalence of ADHD was 0.8% in 2015–2020 in Estonia overall. This further highlights the importance of improving screening and management of ADHD in adult populations, in order to reduce the risk of severe physical comorbidities and potentially shorter life expectancy in individuals with subclinical or currently undiagnosed ADHD. This is also supported by our findings of the ADHD screening questionnaire, which revealed that 8,7% of the participants have high ADHD risk. However, for broader applicability, future studies should replicate these findings in other cohorts with varying rates of ADHD prevalence among adults and investigate the impact of high ADHD risk on overall mortality and survival.
Strengths and limitations
The major strength of this study is inclusion of medical conditions across all ICD-10 diagnoses codes based on EHR with 17 years of follow-up, as well as combining genetic data with EHR and questionnaire data. EHR data is, as compared to self-reported data, not affected by recall bias and has better validity for many diagnoses as information is collected prospectively. Furthermore, we used data from a population-based biobank with a large sample size and high coverage of the whole population, thereby improving our statistical power to detect associations of even small effects.
However, this study also has some limitations. First, although EHR is a comprehensive data source, it only includes individuals with more severe symptoms and/or diseases who have received medical treatment. This may lead to misclassification bias, particularly in psychiatric categories, given the low EHR-based ADHD prevalence in EstBB. It is also possible that some individuals who received their ADHD diagnosis in childhood could have been left out of the study as EHR covers the period from 2004 to 2020. However, ADHD is often underdiagnosed in middle-age and older adults because ADHD in adulthood has been recognized fairly recently (Franke et al., Reference Franke, Michelini, Asherson, Banaschewski, Bilbow, Buitelaar and Reif2018). Availability of data from 2004 can also affect our mediation analysis results as we cannot be certain that depression was not present prior to any of the medical conditions identified in the PRSADHD analysis. Second, population-based cohort studies may be affected by selection bias as healthier and better educated people are more likely to participate in health studies (Larsson, Reference Larsson2021). The ‘healthy volunteer’ selection bias has been previously described in the UK Biobank (Fry et al., Reference Fry, Littlejohns, Sudlow, Doherty, Adamska, Sprosen and Allen2017) and may also exist in the EstBB. Third, given the small number of ADHD cases in our sample, we could not test the effect of ADHD medications on the results. It is well-reported that ADHD drug treatment can help to reduce various negative health outcomes and decrease the risk of risky behavior (Faraone et al., Reference Faraone, Banaschewski, Coghill, Zheng, Biederman, Bellgrove and Wang2021). It is possible that the protective and statistically non-significant associations observed with some phenotypes in the ADHD diagnosis analysis could be affected by treatment effects. However, it is also plausible that individuals with ADHD diagnosis in the EstBB differ substantially from the average individual with ADHD diagnosis due to selection bias. Fourth, since the ADHD diagnosis analytical sample had higher mean birth year, the observed findings may have been affected by underestimation of associations as some diseases typically emerge in older age.
Conclusion
Our study showed that genetic liability for ADHD is associated with increased risk for various medical conditions in individuals without ADHD diagnosis history. The results largely mirror the known associations between diagnosed ADHD and physical disease comorbidities of which many may be linked to adverse health behaviors This knowledge can have implications for prevention and health policy as better detection and timely management of ADHD symptoms may help to reduce the risks for poor health choices and adverse medical consequences across the life span. Although PheWAS design does not allow us to conclude which specific mechanisms underlying behind the observed associations, it provides insights for future studies to further investigate potential underlying pathways.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724000606.
Acknowledgements
We thank all participants and staff of the Estonian biobank for their contribution to this research. The analytical work of EstBB was carried out in part in the High Performance Computing Center of the University of Tartu. We also thank Liisi Panov from the National Institute for Health Development for providing country-wide ADHD prevalence estimates and Silva Kasela for helping with causal mediation analysis.
Estonian Biobank Research Team: Andres Metspalu, Lili Milani, Tõnu Esko, Reedik Mägi, Mari Nelis and Georgi Hudjashov. This work was written at writing retreats and writing days organized by the Institute of Genomics, University of Tartu.
Funding statement
This research in the Estonian Biobank was supported by the European Union through the European Regional Development Fund (Project No. 2014-2020.4.01.15-0012 GENTRANSMED and 2014-2020.4.01.16-0125), and the Estonian Research Council's grant No. PSG615. This study was also funded by EU H2020 grant 692145, Estonian Research Council Grant IUT20-60, IUT24-6. This research is also part of the TIMESPAN project that has received funding from the European Union's Horizon 2020 research and innovation program under grant agreement No 965381. This reflects only the author`s view, and the European Commission is not responsible for any use that may be made of the information it contains. Henrik Larsson acknowledges financial support from the Swedish Research Council (2018-02599) and the Swedish Brain Foundation (FO2021-0115). Isabell Brikell acknowledges financial support from the Swedish Brain Foundation.
This work has been presented as a poster presentation (‘Phenome-wide association study between ADHD polygenic risk and ICD-10 diagnosis codes in the Estonian Biobank’) at the World Congress of Psychiatric Genetics in 2021. Preprint of this manuscript is available in medRxiv (https://www.medrxiv.org/content/10.1101/2022.11.28.22282824v1).
Competing interests
Henrik Larsson reports receiving grants from Shire Pharmaceuticals; personal fees from and serving as a speaker for Medice, Shire/Takeda Pharmaceuticals, and Evolan Pharma AB; and sponsorship for a conference on attention-deficit/hyperactivity disorder from Shire/Takeda Pharmaceuticals and Evolan Pharma AB, all outside the submitted work.
All other authors report no competing interests, including authors from the Estonian Biobank Research Team.
Open access
