ADHD medication use and risk of self-harm: a multinational self-controlled case series study Authors: Adrienne Y L Chan, Kebede Beyene, Sohee Park, Chin-Yao Shen, Andrew S C Yuen, Kirstie H T W Wong, Shiu Lun Au Yeung, Esther W Chan, David Coghill, Celine S L Chui, Sharon Cook, Stephen Crystal, Patrick Ip, Tatia M C Lee, Wallis C Y Lau, Xue Li, Terry Y S Lum, Hao Luo, Nirmala Rao, Andrew Tomlin, Li Wei, Yun K Wing, Amy H Y Chan, Edward Chia-Cheng Lai, Ju-Young Shin, Kenneth K C Man, Ian C K Wong Published: 2026 ABSTRACT AbstractBackgroundConflicting evidence from observational studies and randomised trials has created uncertainty about the association between attention-deficit/hyperactivity disorder (ADHD) medication use and self-harm.ObjectiveTo assess the risk of incident self-harm associated with use of ADHD medication across multiple healthcare systems.MethodsWe conducted a multinational self-controlled case series study using a common data modelling approach to assess the risk of self-harm among individuals treated with ADHD medication across population-based databases from Hong Kong, New Zealand, South Korea, Taiwan and the United Kingdom. We included individuals who initiated ADHD medication and had a record of self-harm between 2001–2020. Incidence rate ratios (IRRs) for self-harm were calculated using conditional Poisson regression, comparing three risk periods of ADHD medication exposure (ie, the 90 days before medication initiation, the first 90 days of medication use and the subsequent treatment period) to non-exposure periods. Adjusted IRRs and 95% CIs were estimated for various risk periods. Estimates from each site were pooled using a random-effects model.FindingsA total of 461 024 individuals with at least one prescription for ADHD medication were included, of whom 6847 had an incident self-harm event during the observation period. The risk was highest during the 90 days before medication initiation (pooled adjusted IRR: 2.58; 95% CI 2.12 to 3.13), followed by the first 90 days of medication use (pooled adjusted IRR: 1.78; 95% CI 1.23 to 2.57), and lowest in the subsequent treatment period (pooled adjusted IRR: 1.27; 95% CI 1.01 to 1.60). Results were consistent across included databases.ConclusionsWhile the use of ADHD medication was associated with self-harm, the highest risk occurred during the unexposed period immediately before starting treatment. The consistent findings across various health systems do not support ADHD medications as the main driver of self-harm. Instead, the increased risk observed prior to treatment may be related to untreated ADHD features.Clinical implicationsAlthough our findings provide reassurance that ADHD medications do not appear to be the main driver of the increased self-harm risk, the persistently elevated risk highlights the need for ongoing clinical monitoring and counselling before and during treatment.Keywords: Attention Deficit and Disruptive Behavior Disorders, Attention-Deficit/Hyperactivity Disorder, Mental Health WHAT IS ALREADY KNOWN ON THIS TOPIC Observational studies examining attention-deficit/hyperactivity disorder (ADHD) medication use and self-harm have reported inconsistent findings, with some suggesting an increased risk shortly after treatment initiation and others a reduced risk during treatment, while randomised trials are underpowered to detect rare outcomes. Observational studies examining attention-deficit/hyperactivity disorder (ADHD) medication use and self-harm have reported inconsistent findings, with some suggesting an increased risk shortly after treatment initiation and others a reduced risk during treatment, while randomised trials are underpowered to detect rare outcomes. Most evidence comes from single-country cohort or case–control studies, with few using self-controlled case series designs or adequately addressing time-varying confounding. Most evidence comes from single-country cohort or case–control studies, with few using self-controlled case series designs or adequately addressing time-varying confounding. Uncertainty remains regarding the temporal association between ADHD medication use and self-harm, and regulatory concerns persist, including a US Food and Drug Administration black box warning for atomoxetine. Uncertainty remains regarding the temporal association between ADHD medication use and self-harm, and regulatory concerns persist, including a US Food and Drug Administration black box warning for atomoxetine. WHAT THIS STUDY ADDS Using a self-controlled case series design in individuals from five population-based databases across the Asia-Pacific region and the United Kingdom, this study shows that the risk of self-harm is highest before treatment initiation and was lower, but still modestly elevated, during subsequent treatment periods. Using a self-controlled case series design in individuals from five population-based databases across the Asia-Pacific region and the United Kingdom, this study shows that the risk of self-harm is highest before treatment initiation and was lower, but still modestly elevated, during subsequent treatment periods. These findings do not support ADHD medication use as the main driver of the observed increase in self-harm risk, but they highlight the importance of timely diagnosis and treatment rather than delaying intervention because of safety concerns. These findings do not support ADHD medication use as the main driver of the observed increase in self-harm risk, but they highlight the importance of timely diagnosis and treatment rather than delaying intervention because of safety concerns. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY The findings support timely diagnosis and treatment of ADHD while emphasising the need for ongoing monitoring of self-harm risk before and during treatment, rather than delaying treatment because of safety concerns. The findings support timely diagnosis and treatment of ADHD while emphasising the need for ongoing monitoring of self-harm risk before and during treatment, rather than delaying treatment because of safety concerns. Introduction Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder. Although ADHD features may attenuate over the lifespan, adult ADHD is far more prevalent than once believed, affecting an estimated 6% of US adults.1 Worldwide, the prevalence of ADHD is estimated at around 7.2% in the child and youth population and 2.5% in adults.2 3 ADHD is typically characterised by a persistent pattern of inattention, hyperactivity and impulsivity.4 It can have a long-lasting and debilitating impact on various aspects of an individual’s life as well as on those around them.2 During the child’s formative years, features of ADHD can result in challenges with learning and social functioning.4 As the child matures, these difficulties may be further exacerbated, potentially contributing to aggressive and antisocial behaviours, academic failure, criminal behaviours and an increased risk of suicidal behaviour.4 There are currently effective pharmacologic treatments available for managing the core challenges of ADHD, including reducing distractibility, improving sustained attention and minimising impulsive behaviours.5 ADHD medications can be categorised into two main classes: stimulants (such as methylphenidate and amphetamine-type stimulants) and non-stimulants (such as atomoxetine, clonidine and guanfacine).6 Recent meta-analyses recommend methylphenidate for children and adolescents and amphetamine-type stimulants for adults as first-line treatments for ADHD.7 Although studies have raised concerns about a potential positive association between ADHD treatment and self-harm events, more recent findings suggested that stimulant medication use was associated with a reduced risk of self-harm in patients with ADHD.8–10 In view of the current rise in ADHD medication use, there is an urgent need to clarify the relationship between ADHD medication use and the risk of self-harm events.11 12 However, the rarity of self-harm events complicates this effort. Randomised controlled trials are often based on relatively small sample sizes, short durations of follow-up and strict inclusion and exclusion criteria, which constrains the ability to explore drug safety, especially for rare events. While observational studies can provide large sample sizes with adequate numbers of rare events and long-term follow-up, insufficient adjustment for confounding or geographically specific or setting-specific idiosyncrasies might result in biased conclusions or limited generalisability. Therefore, we aimed to assess the association between ADHD medication use and the risk of self-harm, using a self-controlled design with multinational healthcare data to minimise confounding and ensure sufficient statistical power. In this study, self-harm refers to intentional self-poisoning or self-injury, irrespective of the apparent purpose of the act.13 This broad definition was adopted because suicidal intent is often unobservable or inconsistently documented in routine healthcare data.14 Therefore, recorded self-harm may include acts with suicidal intent, acts without suicidal intent and acts where intent is unclear. Methods Data sources In this study, we used patient-level electronic health or claims data derived from the respective national or regional administrative databases, clinical databases or registry databases. These databases include Hong Kong’s Clinical Data Analysis and Reporting System (CDARS), New Zealand’s (Ministry of Health national databases), South Korea’s Health Insurance Review & Assessment Service (HIRA), Taiwan’s National Health Insurance Database (NHID) and the United Kingdom’s (UK) Clinical Practice Research Datalink (CPRD) linked to Hospital Episode Statistics (HES) and Office for National Statistics (ONS). A full list of participating sites and databases is provided in online supplemental eTable 1. We applied a distributed network approach with a study-specific common data model (CDM) that standardises the data structures in different databases.15 16 Briefly, the coordination centre distributed the common analysis package, generating aggregated results based on the CDM. Analyses were conducted locally. Each participating site then shared only the aggregated results with the coordinator.15 This approach allowed us to ensure the consistency of analysis among sites. The mapping codes for diagnosis and prescription are presented in online supplemental eTable 2. SAS and R codes adopted in this study are available on GitHub repository at https://github.com/adrienneylc/ADHD-Rx-Self-Harm. Study design This is a multinational population-based self-controlled case series study (SCCS) based on healthcare databases from five geographies (Hong Kong, New Zealand, South Korea, Taiwan and the UK). In the SCCS design, the incidence rate ratio (IRR) was estimated based on within-person comparisons where participants served as their own control rather than between-person comparisons.17 IRRs were calculated by comparing the incidence rate of self-harm during prespecified risk windows with the incidence rate during unexposed baseline periods. A major advantage of this design, unlike traditional cohort or case–control studies, is that it controls for potential effects of measured and unmeasured time-invariant confounders that vary between individuals (ie, genetic factors, disease severity and socioeconomic factors).17 It has previously been employed to examine self-harm and other outcomes in multiple studies.10 18 19 We also adjusted for time-varying factors, including age, season and comedications, which are known to influence the prescribing of ADHD treatment.10 Study population and case identification The study population comprised all individuals aged six or older who received at least one prescription for ADHD medication and had at least one self-harm event during the study period from 1 January 2001 to 31 December 2020, a period for which data were available from the included databases. We began follow-up at 6 years of age, as ADHD medication is generally not recommended for younger children.4 Observation periods began on 1 January 2001, the sixth birthday of the patient or registration start date (whichever was later) and ended on 31 December 2020, date of death or date of registration end (whichever was earlier). Individuals who had a self-harm event before the study period and those with missing age or sex information were also excluded. Exposure The study design schema for defining ADHD medication exposure is illustrated in online supplemental eFigure 1. All formulations (standard and extended release) of ADHD medications (methylphenidate, dexmethylphenidate, amphetamine, dexamphetamine, metamphetamine, lisdexamfetamine, atomoxetine, guanfacine and clonidine) and all strengths were included in the analysis. These medications were identified based on prescription/dispensing records of products listed in the WHO Anatomical Therapeutic Chemical (ATC) Classification System codes whenever possible or equivalent codes of other drug coding systems used at the study site. A list of the medication codes used is available in online supplemental eTable 2. In the main analysis, individuals with a prescription/ dispensing record for any ADHD medication during their observation period were considered to be exposed. The exposure period was defined based on prescription date and prescribed days in the database. Periods within the observation period other than exposure periods were classified as non-exposure periods. We first conducted an overall exposure analysis, where periods of ADHD medication exposure were compared with person-time without recorded ADHD medication use (unexposed period). We then conducted analysis by separating different risk windows, where person-time was divided into four mutually exclusive categories with reference to a previous study10: absence of ADHD medications (baseline period, including patient-time before and after medication exposure), 90 days before the first ADHD medication exposure (pre-exposure period), first 90 days of ADHD medication use and subsequent ADHD medication use. Individuals could contribute to consecutive exposure risk periods while they were continuously exposed to the drug. However, since individuals may take drug holidays, they may experience treatment gaps throughout the observation periods. Therefore, subsequent exposure after the first exposure was not treated as ‘new’ exposure but was instead classified as subsequent ADHD medication use. All remaining time before and after exposure was considered as baseline periods (online supplemental eFigure 1). In subgroup analyses, the exposures were different types of ADHD medication, namely, stimulants (amphetamine-type stimulants, methylphenidate) and non-stimulants (atomoxetine, clonidine, guanfacine), using the same exposure period definition as outlined in the main analysis. Outcome The primary outcome of our study was the first-ever recorded incident event of self-harm during the individual observation period, identified using diagnosis codes available at each participating site and irrespective of the apparent purpose of the act (online supplemental eTable 2, materials 2&3).10 18 The outcome should, therefore, be interpreted as clinically recorded self-harm captured in healthcare data, rather than completed suicide. Restricting the primary analysis to the first event ensured that each individual contributed only one outcome event, thereby avoiding potential dependence between recurrent self-harm events in the primary SCCS analysis.17 Covariate We included the following time-varying covariates in the model: age (years; by 1-year age band), seasons (in 3-month intervals) and use of comedications (antipsychotics—ATC code: N05A, antidepressants—ATC code: N06A, anxiolytics—ATC code: N05B, lithium—ATC code: N05AN01 and antiseizure medications—ATC code: N03A). Statistical analysis We determined the duration of exposure during the study period and calculated the incidence rates for self-harm in each of the predefined risk periods. The risk of self-harm was evaluated using conditional Poisson regression in which incidence rates in risk periods of interest were compared with baseline periods, adjusting for time-varying covariates. We estimated the adjusted IRRs and their 95% CIs for exposure overall and for each predetermined exposure risk period. A significance level of 5% was used in all statistical analyses. IRRs for each risk window at each site were pooled using a random-effects model.20 Subgroup analyses were conducted according to different types of ADHD medication (stimulants and non-stimulants). In this analysis, patients who switched from the study drug to a different type of ADHD medication were censored at the switching date. Subgroup analyses by sex were also conducted. Further analyses investigating the association between ADHD medication and the risk of recurrent self-harm were conducted (online supplemental eTable 3). A subgroup of patients with at least two self-harm events, where the incident and second events were recorded during the individual’s observation period, were included. The follow-up period began 30 days after the incident self-harm, and the IRRs of a subsequent event were evaluated with respect to the risk window as defined above, using the same definition and analysis approach in the main analysis. To assess the robustness of our study results against potential violation of the SCCS assumptions, time-varying confounding and biases, we conducted seven sensitivity analyses. One of them excluded individuals with a record of death during the study period to meet an important assumption of the SCCS study design, ‘study participants must not be censored within the study period’. The list of sensitivity analyses and their respective rationales are available in online supplemental eTable 3. To examine the temporal pattern of self-harm risk without prespecifying exposure windows, we conducted post hoc spline-based SCCS analyses for available sites (Hong Kong, New Zealand and the United Kingdom). In this approach, the relative incidence of self-harm was estimated continuously across the observation period using spline functions, allowing the temporal pattern of risk to emerge directly from the data rather than relying on predefined risk windows.10 21 Microsoft Excel, R, and Statistical Analysis System (SAS) (SAS, USA) were used for data management and analysis. Ethics and inclusion statement The study used healthcare data obtained from Hong Kong, New Zealand, Taiwan, South Korea and the UK. These data encompassed various sources such as electronic health records, registers and insurance records. Each participating site followed the relevant local ethics and regulatory frameworks for study approval, the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Reference number: UW12-136) in Hong Kong; the New Zealand Health and Disability Ethics Committee in New Zealand (considered out of scope for review given that no patient-level data will be shared); institutional review board of Sungkyunkwan University (Reference number: SKKU2020-04-003) in South Korea; the National Cheng Kung University Human Research Ethics Committee (Reference number: HREC (Exempt) No. 110–677) in Taiwan; and the Independent Scientific Advisory Committee in the UK (Reference number: 20_000116). Patient and public involvement No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. Patient and public involvement activities will be conducted through ADDISS, the National Attention Deficit Disorder Information and Support Service, to disseminate the results of the research to study participants or the relevant patient community. Patient’s detail has been removed from these case descriptions to ensure anonymity. Participant consent is not required for studies using purely observational and anonymised data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.22 Results We identified a total of 461 024 individuals (Hong Kong, n=44 271; New Zealand, n=108 809; Taiwan, n=179 646; South Korea, n=404, where only data from patients with both ADHD medication and outcome were collected; and the UK, n=127 894) who received at least one prescription of ADHD medication between 1 January 2001 and 31 December 2020. Among these individuals, we included 6847 individuals who had an ADHD medication prescription and a self-harm event, respectively, during their individual observation period in the analysis (online supplemental eFigure 2). The baseline characteristics by databases are available in table 1. Table 1Baseline characteristics of included individualsHong Kong(N%)New Zealand(N%)South KoreaN (%)TaiwanN (%)United KingdomN (%)Number of patients treated with ADHD medications44 271108 809404*179 646127 894Number (%) of patients with at least one self-harm event in patients treated with ADHD medications45248412742221058Among patients with at least one self-harm event who are treated with ADHD medicationNumber (%) of male275 (60.84)1988 (41.07)107 (39.05)106 (47.75)721 (68.15)Number (%) of patients exposed to methylphenidate419 (92.70)2335 (48.23)264 (96.35)126 (56.76)811 (76.65)Number (%) of patients exposed to amphetamine-type stimulantsNA222 (4.59)NANANANumber (%) of patients exposed to atomoxetine121 (26.77)424 (8.76)41 (14.96)16 (7.21)365 (34.50)Number (%) of patients exposed to other non-stimulants36 (7.96)2624 (54.20)6 (2.19)109 (49.10)62 (5.86)Age (years) at baseline (mean; SD)9.29 (9.29)24.05 (14.89)21.19 (16.75)24.90 (16.55)10.34 (7.03)Number (%) with comorbidity* (ICD-9-CM or equivalent) Acute reaction to stress (308)98 (21.68)85 (1.76)18 (6.57)31 (13.96)73 (6.90) Adjustment disorder (309)101 (22.35)569 (11.75)51 (18.61)72 (32.43)14 (1.32) Anxiety disorder (293.84, 300)98 (21.68)1129 (23.32)237 (86.50)182 (81.98)412 (38.94) Depressive disorder (296.2, 296.3, 311)79 (17.48)1790 (36.98)259 (94.53)177 (79.73)401 (37.90) Disturbance of conduct not elsewhere classified (312)54 (11.95)470 (9.71)73 (26.64)28 (12.61)85 (8.03) Disturbance of emotion (313)94 (20.80)327 (6.75)112 (40.88)99 (44.59)116 (10.96) Intellectual disabilities (317-319)40 (8.85)86 (1.78)21 (7.66)22 (9.91)139 (13.14) Specific delays in development (315)35 (7.74)127 (2.62)19 (6.93)21 (9.46)0 (0)Number (%) with exposure to medication† (ATC code): Lithium (N05AN01)23 (5.09)238 (4.92)99 (36.13)66 (29.73)9 (0.85) Antipsychotics (N05A)271 (59.96)2213 (45.71)259 (94.53)208 (93.69)445 (42.06) Antidepressants (N06A)232 (51.33)3265 (67.44)259 (94.53)208 (93.69)685 (64.74) Antiseizure medications (N03A)139 (30.75)2044 (42.22)182 (66.42)193 (86.94)188 (17.77) Anxiolytics (N05B)205 (45.35)1144 (23.63)267 (97.45)218 (98.20)300 (28.36)*Only data on patients with both exposure and case were included in the data extraction in South Korea.ADHD, attention-deficit/hyperactivity disorder; ATC, Anatomical Therapeutic Chemical; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; NA, not applicable. Baseline characteristics of included individuals Only data on patients with both exposure and case were included in the data extraction in South Korea. ADHD, attention-deficit/hyperactivity disorder; ATC, Anatomical Therapeutic Chemical; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; NA, not applicable. The pooled IRR for overall exposure to any ADHD medication and non-exposure periods was 1.20 (95% CI 0.96 to 1.49). When broken down by prespecified risk periods, ADHD medication exposure was most strongly associated with self-harm during the 90 days before initiation of ADHD medication (pooled IRR: 2.58; 95% CI 2.12 to 3.13), followed by the first 90 days of ADHD medication use (pooled IRR: 1.78; 95% CI 1.23 to 2.57) and lowest during the subsequent ADHD medication use period (pooled IRR: 1.27; 95% CI 1.01 to 1.60; figure 1; table 2). Crude incidence rates were highest during the 90 days before ADHD medication initiation (11 to 29 events per 100 person-years across sites), compared with 5–7 events per 100 person-years during unexposed periods and generally declined during the first 90 days (0 to 23 events per 100 person-years) and subsequent treatment periods (8 to 19 events per 100 person-years; table 2). Site-specific estimates for each database are presented in table 2 and the corresponding forest plots in figure 1, with heterogeneity statistics, including I² and τ², reported in the forest plots. Results are consistent across included databases. Results from post hoc spline-based analyses were also consistent with the primary findings. In all three sites with available data, the relative incidence of self-harm began to increase at least 100 days before ADHD medication initiation and declined thereafter, with no period after initiation exceeding the risk observed before initiation (online supplemental eFigure 14). This indicates that the prespecified 90-day risk windows adequately captured the underlying temporal pattern and did not artificially drive the results. Meta-analyses of adjusted IRRs for self-harm associated with ADHD medication use. ADHD, attention-deficit/hyperactivity disorder; IRR, incidence rate ratio. Table 2Results of main analysis by data sourcesRisk windowNumber of eventsPatient-yearsCrude incidence (per 100 person-years)Adjusted IRR (95% CI)P valueHong KongUnexposed2915583.045 Reference NA 90 days before ADHD medication initiation23114.64204.10 (2.61 to 6.46)<0.001First 90 days of ADHD medication use19105.16183.21 (1.95 to 5.28)<0.001Subsequent ADHD medication use1191306.1791.24 (0.93 to 1.67)0.148New ZealandUnexposed396555 105.567ReferenceNA90 days before ADHD medication initiation2271166.24192.40 (2.10 to 2.75)<0.001First 90 days of ADHD medication use120964.24121.54 (1.28 to 1.85)<0.001Subsequent ADHD medication use5296730.3481.06 (0.94 to 1.19)0.349South KoreaUnexposed2133304.306ReferenceNA90 days before ADHD medication initiation2069.39292.67 (1.61 to 4.44)<0.001First 90 days of ADHD medication use1148.29231.35 (0.70 to 2.62)0.370Subsequent ADHD medication use30155.73192.22 (1.30 to 3.79)0.003TaiwanUnexposed1893099.976ReferenceNA90 days before ADHD medication initiation1558.21262.66 (1.49 to 4.74)0.001First 90 days of ADHD medication use040.860NANASubsequent ADHD medication use18195.3490.87 (0.46 to 1.66)0.682United KingdomUnexposed93517 910.285ReferenceNA90 days before ADHD medication initiation28260.31112.07 (1.41 to 3.03)<0.001First 90 days of ADHD medication use11108.66101.47 (0.80 to 2.69)0.219Subsequent ADHD medication use84844.06101.45 (1.08 to 1.94)0.014*Results with counts <5 were masked to protect patient confidentiality..ADHD, attention-deficit/hyperactivity disorder; IRR, incidence rate ratio; NA, not applicable. Results of main analysis by data sources Results with counts <5 were masked to protect patient confidentiality. .ADHD, attention-deficit/hyperactivity disorder; IRR, incidence rate ratio; NA, not applicable. Similar patterns of association were observed when the analyses were stratified by stimulants only (online supplemental eFigure 3, eTable 4) and non-stimulants only (online supplemental eFigure 4, eTable 4). In the stimulants only analysis, the pooled IRR for overall exposure to stimulants and the non-exposure period was 1.21 (95% CI 1.01 to 1.44). When broken down by prespecified risk periods, stimulant exposure was most strongly associated with self-harm during the 90 days before initiation of stimulants (pooled IRR: 2.74; 95% CI 2.23 to 3.36), followed by the first 90 days of stimulant use (pooled IRR: 1.92; 95% CI 1.58 to 2.34) and was lowest during the subsequent stimulant use period (pooled IRR: 1.23; 95% CI 0.99 to 1.52). In the non-stimulants only analysis, the pooled IRR for overall exposure to non-stimulant use and non-exposure period was 1.10 (95% CI 0.80 to 1.51). When broken down by prespecified risk periods, non-stimulant exposure was most strongly associated with self-harm during the 90 days before initiation of non-stimulants (pooled IRR: 2.87; 95% CI 1.95 to 4.22), followed by the first 90 days of non-stimulant use (pooled IRR: 1.66; 95% CI 0.79 to 3.49) and was lowest during the subsequent non-stimulant use period (pooled IRR: 1.13; 95% CI 0.87 to 1.47). Results from the subgroup analyses by sex (online supplemental eFigures 5–6, eTable 4) and sensitivity analyses (online supplemental eFigures 7–13, eTables 5,6) showed consistent patterns regarding the association between ADHD medication use and the risk of self-harm. Despite an elevated risk of self-harm during periods of ADHD medication use compared with periods of non-use, the risk of self-harm was highest during the 90 days before initiation of ADHD medication when broken down by prespecified risk periods. Discussion Principal findings Our study addresses an important clinical need to understand the relationship between ADHD medication use and self-harm events, particularly during a period of increasing ADHD diagnoses and consequent use of ADHD medication across the world.11 12 We employed a multinational, population-based within-individual design, which allowed us to mitigate confounding factors while ensuring a sufficient sample size to detect the risk of self-harm among individuals using ADHD medications. With a multinational sample of over 460 000 individuals, we identified 6847 individuals who had both an ADHD medication record and a self-harm event during the observation period. The results suggest that the risk of self-harm is highest in the 90 days before the initiation of ADHD medication. This elevated risk decreases after ADHD treatment begins, during the first 90 days of medication use and further during the subsequent treatment period. This trend held across all included health systems and for both stimulant and non-stimulant medications. Furthermore, the consistent results observed across all sensitivity analyses in our study underscore the robustness of our results against potential biases. The findings indicate that the elevated self-harm risk immediately surrounding treatment initiation is most likely driven by worsening underlying ADHD features rather than by the medications themselves. The progressive decline in incidence rate ratios from 90 days before treatment start to subsequent treatment periods demonstrates that, once the underlying features are appropriately managed (whether with stimulants or non-stimulants), the excess risk of self-harm is substantially attenuated. Thus, our results reinforce that ADHD treatment is not causally linked to self-harm and instead may help mitigate the risk associated with untreated features. Comparison with other studies In our previous study to evaluate whether methylphenidate increases risk of self-harm in Hong Kong, we demonstrated that the risk of self-harm increased prior to starting methylphenidate and remained high immediately after commencement of it.10 However, results from a single database study may not be generalisable. A Swedish register-based study found that ADHD medication use was linked to higher self-harm rates in between-individual comparisons, but showed an 11% reduction in self-harm rates compared with periods off ADHD medication in the within-individual analysis, suggesting confounding by underlying conditions.23 More recently, a Swedish nationwide SCCS study reported that ADHD medication use was associated with lower risks of self-harm and other real-world outcomes.24 Zhang et al, using the Swedish register data and a target trial emulation framework, similarly reported that ADHD drug treatment was associated with reduced rates of suicidal behaviours.25 In addition, Taipale et al reported in a Swedish nationwide cohort study of adolescents and adults with ADHD that ADHD medication use was associated with lower suicidal behaviour and fewer psychiatric and non-psychiatric hospitalisations.26 Similarly, a within-individual design study using US data found a protective effect on self-harm behaviours in patients with stimulant use but not non-stimulant use.8 Consistent with these findings, Taiwan was the only participating healthcare system in our study to demonstrate a statistically significant reduction in the overall risk of self-harm during ADHD medication exposure. However, this protective association was not consistently observed across the remaining healthcare systems, our pooled multinational results support the more conservative conclusion that ADHD medication use was not associated with an increased risk of self-harm. Finally, a systematic review and meta-analysis investigated the association between methylphenidate or atomoxetine use in patients with ADHD and the risk of self-harm, respectively, and found no increase in self-harm behaviour.27 However, the meta-analysis was subjected to heterogeneity arising from differences in methodology. Our study, using a common data protocol and common programming approach across five healthcare systems, minimises this source of heterogeneity. Clinical implications and future research This study contributes valuable evidence to inform both clinical practice and public health policy especially after the recent rejection of proposed inclusion of methylphenidate for ADHD in children and adolescents in the WHO’s Model List of Essential Medicines, which reflected concerns about the pressing need for robust evidence on the safety and effectiveness of ADHD medications.28 Future research is needed to integrate data on ADHD medication use into multinational investigations and on other long-term outcomes associated with ADHD medications, especially in diverse populations, to further clarify their safety profiles. Moreover, additional studies are needed to explore the impact of ADHD medication on other high-risk populations, such as individuals with co-occurring mental health conditions. Our study highlights the importance of continued patient monitoring and the need for adherence to clinical guidelines when prescribing ADHD medications. As ADHD medication use continues to rise, policymakers must ensure that robust pharmacovigilance systems are in place to monitor potential adverse effects, including self-harm events. The elevated risk of self-harm observed in the 90 days before ADHD medication initiation suggests that the period leading up to treatment may be characterised by worsening features, emerging psychiatric comorbidity, psychosocial stressors or increased treatment-seeking. This finding highlights the importance of timely clinical assessment for self-harm around the time of ADHD medication initiation. Although our findings do not support a causal increase in self-harm risk attributable to ADHD medication, the first 90 days after treatment initiation remain an important period for clinical monitoring. During this early phase, features may not yet be fully controlled, treatment may still be undergoing dose titration, and patients may continue to experience the clinical difficulties that prompted treatment initiation. The reduction in self-harm risk after ADHD medication initiation compared with the 90-day preinitiation period should also not be interpreted as reflecting the pharmacological effect of ADHD medication alone. Treatment initiation is typically accompanied by clinical assessment, treatment titration, counselling, management of coexisting psychiatric symptoms and closer monitoring, all of which may contribute to attenuation of risk after initiation. The risk of self-harm during the subsequent treatment period, although attenuated relative to the preinitiation period, remained modestly elevated compared with unexposed time (pooled IRR 1.27, 95% CI 1.01 to 1.60). This residual elevation may, at least in part, reflect continued clinical surveillance and detection, as individuals receiving ADHD medication remain under ongoing clinical management and monitoring, rather than a direct pharmacological effect of the medication. Irrespective of the underlying explanation, this finding supports continued vigilance for self-harm among individuals receiving ADHD medication, particularly where psychiatric comorbidity is present. Clinicians should, therefore, ensure close follow-up during the initial phase of treatment, with appropriate counselling for patients and families to monitor changes in mood, behaviour, self-harm thoughts and treatment tolerability. Strengths and limitations The large multinational population included in our study delivers greater statistical power, broader population representation and enhanced validity, thereby enabling more generalisable conclusions. However, most of the databases were built for administrative or reimbursement purposes rather than research purposes, representing a variety of data sources, healthcare settings and coding practices with their respective features and limitations; hence, our study may have some limitations. First, there may be inaccuracy of coding and incompleteness of records. The outcome measures are limited to patients with diagnoses of self-harm events, and we could miss outcomes not diagnosed or coded. Coding practices, healthcare access and the healthcare settings captured by each database may also vary across countries/regions, which could contribute to differences in absolute event rates between databases. However, since the SCCS design compares periods within the same individual, stable differences in coding practice or healthcare access are not likely to bias the within-person estimates. Second, we defined the study outcome as a primary or secondary diagnosis for self-harm events. Though this is likely to increase the power of our study by capturing all possible cases of self-harm events, some minor events may have been captured. Third, some of the subgroup analyses may have limited sample size; careful interpretation of these results is warranted. Fourth, given that this is an observational study, non-compliance with the prescribed drug could influence our results. If patients do not take their drug after prescription, our exposure measure will be misclassified, which could affect our results. Similarly, we did not assume that participants received continuous treatment on ADHD medication initiation because clinicians may offer drug holidays to ADHD patients during school holidays.10 29 Nonetheless, we conducted sensitivity analyses to assess the robustness of our study results against various drug non-adherence scenarios. As some ADHD medication may be indicated for other non-ADHD conditions, such as atomoxetine and clonidine for hypertension, and methylphenidate for fatigue, we excluded patients with a cancer diagnosis in our main analysis and further conducted a sensitivity analysis by including only patients with an ADHD diagnosis. Fifth, baseline psychiatric comorbidity and psychotropic comedication use varied across the included databases. In particular, patients from South Korea and Taiwan had higher recorded levels of psychiatric comorbidity and psychotropic comedication use than those from other countries/regions. These differences may reflect variation in healthcare systems, coding practices, treatment pathways, prescribing practices and the clinical complexity of patients who received ADHD medication and experienced self-harm. They may also partly explain the heterogeneity observed in some country-specific and subgroup analyses. However, because the SCCS design is based on within-person comparisons, fixed differences in baseline clinical severity and between-country case mix are inherently controlled for within each database. In addition, analyses were conducted separately within each country/region before pooling, individual country-specific/region-specific estimates were presented, and time-varying comedication use was also adjusted for in the models. Finally, similar to other observational studies, we cannot rule out the effect of residual confounding from unmeasured time-varying factors beyond those adjusted for in this study. These factors, such as worsening ADHD features, emerging psychiatric comorbidity, psychosocial stressors or treatment-seeking behaviour, are most likely to change around the time of ADHD medication initiation. We, therefore, conducted risk-window stratified analyses to separately evaluate the preinitiation, early treatment, and subsequent treatment periods, allowing more appropriate interpretation of the temporal pattern of self-harm risk in the presence of potential residual time-varying confounding. Conclusion The temporal pattern of risk, with the highest self-harm risk observed before ADHD medication initiation and lower risk thereafter, does not support the hypothesis that ADHD medication initiation is causally associated with a higher risk of self-harm, with similar results observed across five different healthcare systems. Instead, these findings underscore the importance of timely and appropriate treatment for ADHD to mitigate the risk of adverse outcomes such as self-harm. While our study provides reassuring evidence that ADHD medications, particularly stimulants, do not increase the risk of self-harm, ongoing vigilance in clinical practice and public health monitoring remains essential. Supplementary material Footnotes Funding: This work was supported by the Collaborative Research Fund of Hong Kong Research Grant Council 510 (C7009-19G and C7101-23W). The availability of data from Taiwan was supported in part by grants from the National Science and Technology Council of Taiwan (NSTC 115-2628-B-006-002-; 115-2320-B-006-014-MY3). Provenance and peer review: Not commissioned; externally peer-reviewed. Patient consent for publication: Not applicable. Data availability free text: The Hong Kong data used in this study will not be accessible to external parties, as the data custodians have not given permission due to concerns regarding patient privacy protection. Requests can be submitted to hacpaaedr@ha.org.hk (the Central Panel on Administrative Assessment of External Data Requests, the Hospital Authority, Hong Kong). It should be noted that the processing time for such requests may vary as the data provided will be customised for the specific purpose of each project. The data that support the findings of this study are available from the Health Insurance Review and Assessment Service of South Korea but restrictions apply to the availability of these data due to domestic laws and regulations that prohibit the distribution or release of individual’s data to the public and thus are not publicly available. Data are however available from the authors upon reasonable request and with permission of the Health Insurance Review and Assessment Service of South Korea. Due to data privacy laws, individual-level data from New Zealand cannot be made publicly available. The National Health Research Insurance Database of Taiwan can only be accessed at the Health and Welfare Data Center due to data privacy concerns. Data availability statement Data are available upon reasonable request. References Associated Data Supplementary Materials Data Availability Statement Data are available upon reasonable request.