Introduction

Attention-deficit hyperactivity disorder (ADHD) is a common neurodevelopmental condition affecting 5–7% of children, with prevalence declining to 1% in older adults [1]. Stimulant medications are first-line pharmacological treatment and found effective in reducing core symptoms in randomized controlled trial [2]. Observational studies also report positive effects on real-world outcomes, including injury rates, educational attainment, and substance misuse [3]. However, 20–40% of individuals have suboptimal response to stimulants, and while generally well tolerated, side effects such as headaches, appetite loss, sleep disturbances, and elevated blood pressure or heart rate are common, particularly during initiation [1, 2, 4]. Adverse effects and poor response are along with dosing inconvenience, stigma, and patient attitudes common reported reason for medication discontinuation in ADHD [5, 6]. A recent multi-national study found that 35–61% of individuals with ADHD discontinued medication within one year of starting, with higher rates among adolescents and adults compared to children [7]. U.S. studies further show that more than half of children [8] and adults [9] with ADHD do not refill their first prescription, suggesting that many disengage during the titration phase—a crucial period for identifying optimal treatment [1]. Early discontinuation can represent a major challenge, resulting in potential symptom exacerbation, functional impairment, and elevated risk of adverse outcomes [1,2,3]. Understanding why some individuals discontinue treatment early is therefore essential to improve long-term management of ADHD.

It is increasingly recognized that variability in treatment response and tolerability to psychotropic medication is influenced by genetics. Yet large-scale studies aimed at discovering genetic factors affecting pharmacotherapeutic outcomes in psychiatry in general, and ADHD in particular, are sparse [10]. Identifying genetic markers associated with stimulant discontinuation could offer new insights into the biological mechanisms underlying treatment response, tolerability, as well as behaviors related to medication persistence. This may inform more tailored treatment approaches, including risk stratification at medication initiation, prediction of tolerability profiles, or alternative treatment strategies for those at higher risk of discontinuation. To date, pharmacogenomic ADHD research has been dominated by candidate gene studies focused on methylphenidate (MPH). Replicated associations are reported for variants in DAT, DRD4, ADRA2A, COMT, and NET genes, though individual effect sizes are modest [11, 12]. Only three small genome-wide association studies (GWAS; N < 200) have examined MPH response [13, 14] or adverse effects [15], with no genome-wide significant loci identified. We recently reported suggestive evidence that common genetic variants contribute to stimulant discontinuation [16] defined by prescriptions from 7427 individuals with ADHD [17, 18]. SNP-heritability (h2SNP) was estimated to be 0·13 (SE 0·08) and higher polygenic scores (PGSs) for bipolar disorder and schizophrenia were associated with increased rates of discontinuation. However, broader genetic liabilities, including traits related to cognition, physical comorbidity (e.g., obesity, blood pressure, cardiometabolic risk), and side-effects [2,3,4, 19] may also influence discontinuation. Additionally, individuals with ADHD show an elevated burden of protein-truncating variants (PTVs) in highly constrained genes [20, 21] and rare variants are proposed to influence medication metabolism, response and tolerability [22]. Yet, no studies have investigated the role of rare variants in genes potentially relevant for stimulant pharmacodynamics. Finally, no studies have examined genetic contributions to age-related differences in discontinuation, despite known age-related variation in discontinuation rates, comorbidity profiles, and ADHD h2SNP [5, 7, 21]. To address these knowledge gaps, we aimed to comprehensively characterize the contributions of common and rare genetic variants to stimulant discontinuation in individuals with ADHD. We specifically focus on early medication discontinuation, based on the hypothesis that individuals who experience strong side effects or lack of therapeutic response are more likely to discontinue treatment early than individuals for whom the medication is working well, but might discontinue later due to less biological factors. Consequently, our design may increase the likelihood of capturing the genetic and biological components of treatment discontinuation.

Subjects and methods

Study population and genomic data

iPSYCH is a population-based case-cohort identified from all singletons born in Denmark 1 May 1981 until 31 December 2008, linked to national registries using the personal identification number assigned to all Danish residents [17, 18]. iPSYCH includes 50,615 randomly selected population-based controls and 79,336 individuals with at least one diagnosis of ADHD (F90.0), schizophrenia spectrum disorders (SCZ;F20-F29), affective disorders (F30–F39), or autism (ASD;F84.0,F84.1,F84.5,F84.8,F84.9) in the Danish Psychiatric Central Research Register (DPCRR) between 1 January 1994 and 31 December 2015 [23]. The DPCCR contains data on all in-patient admissions since 1969 and outpatient since 1995. Diagnoses are recorded according the International Classification of Diseases, 10th revision, Diagnostic Criteria for Research from 1994 [23]. We selected all iPSYCH individuals diagnosed with ADHD and with >1 dispensation of a stimulant ADHD medication (methylphenidate [N06BA04], dexamfetamine [N06BA02], lisdexamfetamine [N06BA12]) between 1 January 1996 and 31 December 2018 in the Danish National Prescription Register (DNPR) [24]. DNPR includes all redeemed prescriptions in Denmark since 1995 classified by Anatomical Therapeutic Chemical (ATC) codes. Information on sex assigned at birth, date of birth, emigration and death were obtained from the Danish Civil Registration System and the Cause of Death Register. Individuals who died, emigrated, or with missing information during follow-up were excluded.

DNA from all individuals in iPSYCH was extracted from dried bloodspots obtained from the Danish Neonatal Screening Biobank. iPSYCH individuals were genotyped in two waves (iPSYCH2012 and iPSYCH2015i), using Illumina´s PsychChip and Global Screening array, and subsequently merged. Detailed descriptions of the sample, DNA extraction [17, 18], imputation, and quality control (QC) [19] are provided elsewhere. For current analyses, best guess genotypes from the combined, imputed and QC´ed iPSYCH sample (nSNPs = 8785,478; nsample = 134,628) were extracted. We excluded variants with imputation info score <0.8, minor allele frequencies (MAF) < 0.01, missing call rates >0.02, and with Hardy-Weinberg equilibrium exact test p-value < 1×10−8, as well as samples with missing call rates >0.05, leaving 7,124,623 SNPs from 22 autosome chromosomes. Filtering was done using the bigsnpr R package [25] and PLINK1.9 [26]. After phenotypic and genomic filtering, 18,362 individuals were retained for analyses (Fig. 1).

Fig. 1: Filtering pipeline of individuals in the iPSYCH cohort diagnosed with ADHD and initiating stimulant medication treatment.
Fig. 1: Filtering pipeline of individuals in the iPSYCH cohort diagnosed with ADHD and initiating stimulant medication treatment.
Full size image

Individuals diagnosed with ADHD, affective disorders, schizophrenia spectrum disorder, and/or autism from the iPSYCH cohort was the initial sample before filtering. Individuals who died, emigrated, has missing information in the civil registration system or only redeemed one stimulant prescription during follow-up were excluded from analyses. After the filtering steps, 18,362 individuals diagnosed with ADHD and > 1 stimulant medication dispensation were retained for analyses.

Stimulant discontinuation

DNPR does not contain information about prescription length or cause of discontinuation. Therefore and in line with prior research [7, 16], we defined stimulant discontinuation by a gap ≥180 days between dispensations. The ≥180 days cut-off was selected to capture clinically meaningful medication gaps rather than shorter breaks. We focused on discontinuation within one year of initiation (i.e. first stimulant dispensation in the DNPR) as this may be more likely to reflect discontinuation due to poor response or tolerability, which may in turn be more influenced by genetics, compared to discontinuation after longer-term treatment. Individuals with a gap ≥180 days between consecutive stimulant dispensations were defined as discontinuers if the last dispensation followed by the gap occurred within the first year of initiation (Figure S1).

Statistical analyses

h2 SNP and GWAS

To evaluate the variance in stimulant discontinuation explained by the included SNPs, h2SNP was estimated using genome-based restricted maximum likelihood (GREML) implemented in GCTA [27]. We report h2SNP on the observed scale due to the variability prevalence estimates of discontinuation in the ADHD population. To explore age-related genetic differences, we split the sample by age at initiation using a genetically informed cut-off, moving the cut-off one year at a time. At each cut-off, h2SNP and genetic correlation (rg) between “child” and “adolescent/adult” groups were estimated to identify the cut-off with the largest genetic difference (lowest rg) between groups, while maintaining adequate sample sizes. GWAS was then performed using logistic regression in PLINK1.9 for all individuals, the child group, and the adolescent/adult group. Analyses were corrected for sex assigned at birth, genotyping wave, age at initiation, and the first 10 ancestral principal components (PCs). See Supplement for details.

Polygenic score analyses

We selected 36 polygenic scores (PGSs) for phenotypes relevant to ADHD and ADHD medication outcomes from a catalog of 943 PGSs generated in the iPSYCH cohort using LDpred2 [28] and GWAS summary statistics without iPSYCH data, which has been described in details [29] (see also Supplement). PGSs for ADHD, major depressive disorder (MDD), SCZ, bipolar [BP] and autism were generated using meta-PRS described in Ref. 28, which combine internal individual-level data and external GWAS summary statistics for polygenic prediction, and has proven more precise when large individual-level data are available. PGSs represented domains including psychiatric disorders, education and cognition, cardiometabolic traits, and potential adverse effects of ADHD medication or self-medication (Table S1). Associations with stimulant discontinuation were estimated using logistic regression in the full sample and by age group, adjusting for age at first dispensation, sex, genotyping batch, and the first 10 principal components. Results are reported as odds ratios (ORs) with 95% confidence intervals (CI) per standard deviation increase in PGS. The Benjamini–Hochberg False discovery rate (FDR) correction was applied in the full sample and each age-group, adjusting for 36 tests. For PGSs with associations surviving FDR correction, we estimated the odds of discontinuation across quintiles relative to the lowest. Variance explained by the PGS was estimated as the difference in Pseudo-R² between a model with covariates only and covariates plus PGS. We tested for replication of associated PGSs in an independent sample of 829 individuals with ADHD who had initiated stimulant treatment from the Child and Adolescent Twin Study in Sweden (CATSS) [30]. We used generalized estimating equations with a logit link and cluster-robust standard errors to account for twin relatedness. Analyses were conducted in R and adjusted for the same covariates as the main analyses. We report nominally significant associations (p < 0.05) and FDR-corrected p-values. Details in Supplement.

Rare variant analyses

We investigated the role of protein-truncating variants (PTVs) in discontinuation using exome sequencing data from a subsample of 5999 individuals in iPSYCH (2280 discontinuers and 3719 continuers). Although PTVs with strong deleterious effects are typically rare due to negative selection, stimulant discontinuation is unlikely to have been under such evolutionary pressure. Therefore, we included both rare and low-frequency PTVs (minor allele frequency ≤0.01). We analyzed three gene sets (Tables S2–S4), selected for their hypothesized involvement in stimulant and/or dopamine response, as stimulant medications act by increasing extracellular dopamine levels [2]. These were: (1) pLI genes, including 2991 loss-of-function-intolerant genes (pLI > 0.9) based on gnomAD v2.1.1; (2) dopamine response genes, including 82 genes involved in cellular or organismal responses to dopamine, identified in MSigDB; and (3) stimulant medication genes, including 44 genes identified via DGIdb v5.0 using the term “methylphenidate+hydrochloride”. We focused on genes interacting with methylphenidate, as around 80% of individuals with ADHD in Denmark initiate treatment with methylphenidate [7]. The remaining patients are treated primarily with non-stimulants and initiating treatment with other stimulants is very rare [7]. As a negative control, we also evaluated the burden of synonymous variants in each gene set, as these are not expected to affect protein function or medication response. Details are in Supplement.

Sensitivity analyses

We conducted two sensitivity analyses. First, we re-estimated h2SNP further adjusting for co-occurring iPSYCH-ascertained disorders (SCZ, ASD and affective disorders). Second, we re-estimated h2SNP, and performed GWAS and PGS analyses after excluding 3257 individuals who discontinued within three months of treatment initiation, as this may reflect limited treatment engagement [8, 9] which is potentially less influenced by genetics.

Results

Among 18,362 individuals with ADHD who started stimulant treatment (31·0% female), the median age at initiation was 12 years (IQR 8–18). Within one year, 7102 (39·0%) had discontinued. Discontinuers had a higher median age at initiation than continuers (15 years [IQR 10–20] vs 10 years [IQR 8–15]) and more often at least one co-occuring iPSYCH diagnosis (32·0% of discontinuers vs 28·0% of continuers [Table 1]).

Table 1 Description of study sample.
Full size table

SNP-heritability and GWAS

In the full sample, the h2SNP of discontinuation was 0.06 (95% CI 0.02–0.11) and significantly different from zero (p < 0·001), indicating a small common variant contribution. Table S5 shows the h2SNP, rg and associated test statistics from age-stratified analyses. The lowest rg, between children and adolescents/adults was observed when splitting at age 16 (rg = 0·23, 95%CI 0·37–0·83), which was significantly different from 1 (p < 0·012). The h2SNP of discontinuation was 0·08 (95% CI 0·02–0·14) in children <16 years and 0·14 (95% CI 0·02–0·27) in adolescents/adults ≥16 years. Henceforth, the full sample is referred to as DISCALL (n = 18,362; 7102 discontinuers), the child group as DISCbelow16 (n = 12,229; 3598 discontinuers), and the adolescents/adult group as DISCabove16 (n = 6133; 3504 discontinuers). Adjusting for co-occurring iPSYCH ascertained disorders did not affect h2SNP estimates (Table S6) and were therefore not included as covariates in subsequent analyses. No genome wide significant loci were identified for DISCALL or DISCabove16 (Figure S2 and S4). One locus on chromosome 11 reached genome wide significance in DISCbelow16 (index variant rs73438427, OR 0·65, p = 1·54 × 10⁻⁸; Figure S3), located in the protein-coding gene SLC5A12 (Figure S5). Table S7 presents the top 10 SNPs associated with DISCbelow16 and their corresponding associations in DISCALL and DISCabove16. Genomic inflation factors (λ) were <1·02 across GWASs and visual inspection of QQ plots (Figures S6–S8) showed no strong evidence of residual confounding.

Polygenic score analyses

Table S8-S10 show all PGS associations. Ten of the 36 selected PGSs were statistically significantly associated with stimulant discontinuation in at least one of three groups (DISCALL, DISCbelow16 or DISCabove16) after FDR correction (Table 2). These included PGSs for five psychiatric (psychopathology [P]-factor, SCZ, MDD, BP obssesive compulsive disorder [OCD]), two cognitive (educational attainment [EA], intelligence quotient [IQ]), and three phenotypes related to adverse effects or self-medication (cannabis use [CU], body mass index [BMI], and migraine). No cardio-metabolic PGSs were significantly associated with discontinuation. In DISCALL, the strongest associations were with the P-factor-PGS, derived from a factor analysis of GWAS of SCZ and MDD, (OR 1·11, 95% CI 1·08–1·14) and SCZ-PGS (OR 1·10, 95% CI 1·06–1·13). In DISCbelow16 the strongesG associations were with the P-factor-PGS (OR 1·12, 95% CI 1·08–1·17) and the BMI-PGS (OR 0·90, 95% CI 0·86–0·94). In DISCabove16, the strongest associations were with SCZ-PGS (OR 1·09, 95% CI 1·03–1·14) and IQ-PGS (OR 0·91, 95% CI 0·87–0·96). Notably, the EA-PGS showed opposite directions of association in DISCbelow16 (OR 1·06, 95%CI 1·02–1·11) and DISCabove16 (OR 0·93, 95%CI 0·88–0·97). Each PGS explained from 0.01–0·23% of the variance in discontinuation (Tables S8–S10). ORs by PGS quintiles for the 10 associated PGSs are shown in Fig. 2 and Table S11.

Fig. 2: Odds Ratios (ORs) of stimulant discontinuation by polygenic score quintiles.
Fig. 2: Odds Ratios (ORs) of stimulant discontinuation by polygenic score quintiles.
Full size image

Odds ratio (OR) and 95% confidence intervals (CI) for stimulant discontinuation among individuals with ADHD who initiated stimulant treatment, by their polygenic score (PGS) quintiles (on the x-axis) for the following phenotypes: psychopathology (P) factor, schizophrenia, major depressive disorder, bipolar disorder, obsessive compulsive disorder, body mass index, IQ, educational attainment, migraine, and cannabis use. First quintile is used as reference. Results are shown for DISCALL (in orange), DISCbelow16 (purple), and DISCabove16 (green) and vertical bars indicate 95% confidence intervals.

Table 2 Polygenic scores (PGS) associated with stimulant discontinuation in individuals with ADHD.
Full size table

Testing for replication in an independent sample (n = 829; 39% discontinuers; Table S12), only the BMI- and IQ-PGSs were nominally associated with discontinuation in the full sample, showing consistent direction of effects as in iPSYCH. In DISCbelow16 (n = 592, 33% discontinuers), MDD-PGS and SCZ-PGS were nominally associated with increased odds of discontinuation, while in DISCabove16 (n = 237, 56% discontinuers) BMI-PGS again showed a nominal negative association. No associations survived FDR correction. PGSs for the P-factor, OCD, and cannabis use disorders could not be generated due to data linkage constraints.

Rare variant analyses

Neither PTVs in the pLI gene set (OR 1·01, 95% CI 0·98–1·05) nor in the stimulant medication gene set (OR 0·96, 95% CI 0·75–1·23) were associated with discontinuation (Table S13). PTVs in dopamine response genes showed a nominally significant (p < 0.05) association in DISCALL (OR 0·81, 95% CI 0·66–0·99), indicating a lower PTV burden in discontinuers compared to continuers. A similar, though non-significant, effect was observed in DISCbelow16 (OR 0·89, 95% CI 0·69–1·15), while a somewhat stronger and nominally significant association was seen in DISCabove16 (OR 0·64, 95% CI 0·45–0·91; p < 0·02) (Fig. 3).

Fig. 3: Load of protein truncating variants (PTVs) and synonymous variants in the dopamine response gene set in individuals who discontinued stimulants compared to individuals who continued.
Fig. 3: Load of protein truncating variants (PTVs) and synonymous variants in the dopamine response gene set in individuals who discontinued stimulants compared to individuals who continued.
Full size image

Boxplot of odds ratios (OR) and 95% confidence intervals (CI) from logistic regression of the effect of protein truncating variants (PTVs) in the “dopamine response” gene set on stimulant discontinuation. ORs are shown as a solid vertical line and upper and lower 95%CI as the upper and lower boundaries of the boxes for DISCALL, DISCbelow16 and DISCabove16. Results are shown for PTVs in red, and for comparison the load of synonymous variants in green.

Sensitivity analyses

In sensitivity analyses excluding individuals who discontinued stimulants within the first three months, 15,105 were retained of whom 25% discontinued within 1 year (n discontinuers=3845 [eDISCall]). The age split yielding the lowest rg between children and adolescents/adults was 14 years (rg 0·12, 95% CI –0·47–0·71). h2SNP was 0·07 (95% CI 0·01–0·12) for eDISCALL, 0·09 (0·01–0·17) for eDISCbelow14, and 0·18 (0·05–0·31) for eDISCabove14 (table S14). No genome-wide significant loci were identified (Figure S9-S14). PGSs for the P-factor, SCZ and BMI (figures S15–17; table S15) were significantly associated with discontinuation in the eDISC samples, with consistent directions of effect as in the main analyses.

Discussion

We found evidence of a small contribution of common genetic variants to stimulant discontinuation, with h²snp estimates indicating slightly higher heritability in adolescents and adults than in children. The age cut-off between children and adults at 16 years was determined by genetics analyses, but grouping of adolescents together with young adults are in line with findings from a large epidemiological study that identified differences in treatment discontinuation in children (7–11 years) compared to adolescents (12–17) and young adults (18 – 25 years), with lower discontinuation in the childhood group compared to the other two groups [7]. The GWAS identified one significantly associated locus in children below 16 at initiation, located in the gene SLC5A12. PGSs for mainly adult-onset psychiatric disorders, cognitive, side-effect and body weight related phenotypes were associated with discontinuation, with the strongest signals seen for the P-factor, schizophrenia, educational attainment and BMI PGSs. Several of these associations showed age-specific patterns. Rare variant analyses showed no associations with protein-truncating variants (PTVs) in broad constraint- or medication-related gene sets, but nominal evidence of reduced PTV burden in dopamine response genes among discontinuers. Together, these findings suggest that both common and rare genetic variants can influence stimulant discontinuation, albeit with small effects, and that genetic contributions may be partially distinct across age groups.

We identified one genome-wide significant locus for stimulant discontinuation in children, located on chromosome 11 within SLC5A12, a gene encoding a sodium-coupled monocarboxylate transporter involved in lactate and pyruvate transport. The gene is primarily expressed in peripheral tissues with limited expression in the brain and has not been previously been implicated in ADHD, neurodevelopment, or drug metabolism. Thus, the relevance of this finding for stimulant pharmacokinetics remains unclear and would need to be replicated in independent cohorts. This is the largest GWAS of stimulant-treatment related outcomes to date, yet we did not identify genome-wide significant associations beyond this single locus. This underscores the challenge to detect common variants with small to moderate effects influencing multifactorial traits like medication discontinuation, even in relatively large pharmacogenomic samples. Replication and meta-analysis with other cohorts is needed to increase power and validate findings, yet this remains challenging due to the limited availability of pharmacogenomic ADHD samples. Further, our findings of modest SNP-heritability and the lack of strong genome-wide significant signals suggest that biological influences likely only explain a small part of stimulant discontinuation. Genetic associations may be more readily detectable for measures of stimulant response and tolerability, as found in recent pharmacogenomics studies of symptom change following antidepressant treatment [31].

Among the 36 polygenic scores examined, psychiatric PGSs showed the clearest pattern of association with stimulant discontinuation, particularly for the general psychopathology (P-factor) and schizophrenia. These associations were largely consistent across age, suggesting that higher genetic burden for mood and psychotic disorder increases the likelihood of discontinuation. Effects of psychiatric PGSs were somewhat stronger in children, yet the overall pattern supports a link between broad psychiatric vulnerability and reduced treatment persistence across ages. This aligns with clinical observations and some prior research, suggesting that psychiatric comorbidities can complicate ADHD treatment [2, 5]. Our findings highlight the potential value of genetic predisposition as an early indicator of treatment challenges, providing insights that precede, and may be independent of, the later emergence of psychiatric disorders. Interestingly, the ADHD-PGS was not associated with discontinuation, suggesting distinct genetic influences on ADHD diagnosis and stimulant persistence. This is also partly supported by inconsistent findings, where some studies have linked more severe ADHD symptoms to better treatment response and others the opposite patterns [32].

Similarly, no association was observed for alcohol use disorder PGS, despite prior evidence linking substance misuse to treatment complications and drop-out [2, 5, 16]. Polygenic load for cognitive traits showed diverging effects by age. In adolescents/adults, higher IQ- and educational attainment (EA)-PGSs were associated with reduced odds of discontinuation, consistent with studies linking higher education to better treatment adherence [6]. In contrast, in children, higher EA-PGS was associated with increased odds of discontinuation, potentially reflecting less severe ADHD symptoms or greater access to alternative strategies for managing ADHD in families with higher EA. BMI and migraine were associated with reduced discontinuation in children. Higher genetic propensity for higher BMI may offset stimulant-related appetite suppression, which is a common cause of discontinuation for children with a standard-range BMI [2]. The Migraine-PGS was also inversely associated with discontinuation in children, potentially due to a hypothesized beneficial effect of stimulants on dopamine-linked migraine symptoms [33]. Cardiometabolic PGSs showed no association with discontinuation, consistent with emerging evidence that stimulants does not confer significant cardiovascular risk when considering short-term treatment [34]. While our PGS findings highlight biologically plausible mechanisms and potential age-specific effects underlying stimulant discontinuation, PGSs explained only a small proportion of variance (0.003–0.23%), limiting clinical utility of current PGSs.

Rare variant analyses suggested that individuals who discontinued stimulants had a lower burden of protein-truncating variants (PTVs) in dopamine response genes, particularly among adolescents and adults. This may reflect less disruption of dopamine pathways and potentially less severe ADHD, indicating a reduced biological need for stimulant treatment. In contrast, individuals with a higher PTV burden may have more impaired dopamine signaling and therefore benefit more from stimulants, leading to continued use. While this points to a possible link between rare variant load and treatment efficacy or necessity, the lack of association of the ADHD-PGS, which is linked to ADHD symptom severity [35], complicates this interpretation. Given the modest effect sizes and nominal significance, findings should be interpreted cautiously but highlight need for further research into how rare genetic variation may shape treatment response in ADHD.

In summary, we observed age-related differences in the genetic underpinnings of discontinuation. Among adolescents and adults 16 or older, stronger genetic associations in common and rare variant signals could suggest that discontinuation in this group more often reflects individual biological responses, such as perceived efficacy or side-effects. In contrast, discontinuation among children may be more influenced by external factors, including clinicians and parental preferences. These age-related patterns, though modest, point to developmental and contextual influences on how genetic liability shapes treatment discontinuation in ADHD.

This study has several limitations. First, the sample was underpowered to conduct sex-stratified or subgroup analyses based on specific medication patterns. Second, discontinuation was defined using dispensation data, without information on adherence or reasons for stopping treatment. Genetic influences may differ by discontinuation cause (e.g., adverse effects vs poor response), but we could not disentangle these. Furthermore, our sensitivity analysis did not suggest that inclusion of individuals who discontinued within three months had a strong impact on the results. Nevertheless, it should be noted that we did not have information on the cause of discontinuation and therefore cannot separate discontinuation due to side effects or poor response from low treatment engagement, which in particular could affect those with early discontinuation, where the former likely has a larger biological component. Third, registry coverage for ADHD diagnoses were available from 1994 and prescription data from 1996, resulting in possible misclassification of treatment initiation. However, ADHD diagnosis and medication use were rare in Denmark prior to mid-2000s, suggesting this is unlikely to substantially impact findings. Finally, we only investigated the genetic architecture underlying early stimulant discontinuation which might differ from the genetic underpinnings of longe-term treatment patters, and the sample was limited to individuals of European ancestry and findings may therefore not generalize to other populations.

We found that common and rare genetic variants modestly contribute to stimulant discontinuation in ADHD, without identifying strong effects from individual genes. Our findings tentatively support genetic contributions to age-related differences in discontinuation and suggest that genetic liability for broad psychiatry vulnerability, BMI and cognition, but not ADHD itself, impacts treatment discontinuation. Future large-scale studies are needed to clarify the relevance of our findings for stimulant discontinuation, response and tolerability and in developing personalized ADHD treatment strategies.