Published on 16 September 2026
ADHD medication in children and adolescents: who stops, and when?
In brief
A Norwegian national cohort of 8961 children and adolescents aged 5 to 17 years, diagnosed with ADHD in specialist services between 1 January 2010 and 31 December 2012, was followed through the prescription dispensing registers until 31 December 2021, that is, for at least nine years per person. 74.3% of these young people, 6661 in all, received at least one dispensing, after a median delay of 106 days (interquartile range 17 to 231 days), and 12% waited more than a year. Discontinuation comes sooner when treatment starts in adolescence: the difference in restricted mean survival time at one year is 31.9 fewer days in 15 to 17 year olds compared with 5 to 11 year olds (95% CI 23.1 to 40.8 fewer days; p<0.001). Girls start treatment more often than boys in every age band, with the widest gap between 15 and 17 years (72.8% against 54.2%); across all age bands combined, they stay on treatment 13.2 fewer days over the first year (95% CI 6.5 to 19.8 fewer days; p<0.001). The major limitation lies in the nature of the data: this is an association observed on pharmacy dispensings, not measured adherence, and the register says nothing about the reasons for stopping.
The context
The question does not arise at the moment of prescribing, it arises three months later. A child or adolescent diagnosed with ADHD, for whom medication has been chosen, will not necessarily stay on that treatment for long, and the clinician often notices the discontinuation after the fact, at a renewal appointment that never took place.
Long-term follow-up data are scarce in this field: controlled trials last a few weeks to a few months, whereas the treatment decision commits the patient for years. National health registers offer a partial answer, by observing what happens outside any protocol, across an entire country. They answer a descriptive question, the real-world timeline of dispensings, and not a question of efficacy.
The study at a glance
| Item | Content |
|---|---|
| Population | Children and adolescents aged 5 to 17 years, Norway, incident diagnosis of ADHD (ICD-10 F90) made in specialist child and adolescent mental health services between 2010 and 2012. N=8961, of whom 6661 were treated. |
| DetailPopulation identified from the national patient register and the national prescription dispensing register, with national coverage rather than a sample. Incident status rests on the absence of a specialist F90 diagnosis and the absence of any ADHD medication dispensing during the year preceding inclusion. | |
| Intervention | ADHD medications: psychostimulants (methylphenidate, amphetamine, dexamfetamine, lisdexamfetamine) and non-stimulants (atomoxetine, guanfacine). |
| DetailExposure is defined by pharmacy dispensing. Neither the prescription alone nor actual intake is observed. In practice, the first dispensing is a stimulant in 98% of cases, methylphenidate alone in 97%, atomoxetine in 2.4%. | |
| Comparator | No control arm. Comparisons between pre-specified subgroups. |
| DetailThree age bands at initiation (5 to 11 years, 12 to 14 years, 15 to 17 years), sex, class of the first medication, number of psychiatric comorbidities and parental education level. | |
| Outcomes | Initiation rate, time to first discontinuation, and longitudinal adherence measured by the PDC over nine years. |
| DetailDiscontinuation is defined by a gap of at least 90 days without estimated medication coverage. The PDC (proportion of days covered) is the share of days covered by a dispensing, calculated over successive 90-day intervals and capped at 1.0; no adherence threshold is applied, and the value is analysed as a continuous variable. Time to discontinuation is summarised by the restricted mean survival time (RMST) at one year, and by that measure alone. | |
| Design | National retrospective register-based cohort, inclusions from 2010 to 2012, follow-up until 31 December 2021. CEBM level of evidence 2b. |
| DetailNo randomisation, no allocation of treatment by the investigators. The design supports associations, it does not establish cause and effect. | |
Quality control
| Domain | Judgement |
|---|---|
| Sample size and representativeness | Strength |
| Finding8961 young people diagnosed across an entire country, 6661 of them treated and followed for discontinuation, drawn from registers designed to be exhaustive. The volunteer bias inherent in recruited cohorts is ruled out. | |
| Length of observation | Strength |
| FindingInclusions from 2010 to 2012 and follow-up to the end of 2021, that is, at least nine years guaranteed for each person, whereas long-term data are scarce in paediatric ADHD. | |
| Measurement of adherence | Indirect measure |
| FindingThe PDC is built on packs dispensed, not on tablets taken. A dispensing that is not consumed counts as coverage: real adherence is probably lower than the adherence measured here. The authors themselves point out this limitation. | |
| Reasons for stopping | Not collected |
| FindingAn adverse effect, remission of symptoms, a shared decision on a trial discontinuation, a house move or loss to follow-up all leave the same trace in the register: no dispensing. These situations cannot be told apart in these data. | |
| Analytical method | Appropriate |
| FindingThree complementary approaches: Kaplan-Meier curves with the log-rank test, RMST at one year with confidence intervals obtained by bootstrap over 1000 resamples, and a linear mixed-effects model for the PDC trajectory over nine years. The RMST does not require the proportional hazards assumption, unlike the hazard ratio, and it is expressed in days, a unit that can be interpreted in the consulting room. The summary table in the supplementary material confirms this set-up and specifies that the mixed-effects model includes age, sex, medication class, psychiatric comorbidities and parental education; the survival comparisons, for their part, are crude comparisons between strata, and no sensitivity analysis is reported. A reservation on internal consistency: the note to supplementary figure 2 lines up five log-rank p values for four strata cited, and one of them, p=0.064, does not match the value given in the main text for age. The age difference retained here is that of the RMST, whose confidence interval excludes zero. | |
| Transferability | Limited |
| FindingThe Norwegian system combines universal access with diagnosis made in specialist child and adolescent mental health services. Its conditions for prescribing, dispensing and renewal are specific to that setting and cannot be assumed to apply in another country. | |
The findings
| Outcome | Reported value |
|---|---|
| Starting treatment | 6661 of the 8961 young people diagnosed, that is 74.3%, receive at least one dispensing. |
| ReadingRoughly a quarter of the cohort never appears in the dispensing register. The reasons for this absence, a choice not to treat, refusal, referral to non-drug care, are not documented. | |
| Time to first dispensing | Median delay of 106 days (interquartile range 17 to 231); 55% within 90 days; 12% beyond one year. |
| ReadingSlightly under half of the treated young people start more than three months after the diagnosis used by the authors as the starting point. This delay describes a care pathway, it does not measure reluctance. | |
| Effect of age at treatment start | Compared with 5 to 11 year olds (RMST at one year of 290.3 days): 17.6 fewer days in 12 to 14 year olds (95% CI 9.9 to 25.3) and 31.9 fewer days in 15 to 17 year olds (95% CI 23.1 to 40.8), p<0.001. |
| ReadingThis is the clearest result of the study, and it is graded: the later the initiation, the shorter the time on treatment in the first year. It is consistent with what is observed for other long-term treatments in adolescence, which does not, however, prove it. | |
| Difference between girls and boys | Girls start treatment more often in every age band (72.8% against 54.2% in 15 to 17 year olds) and, across all ages combined, stay on treatment 13.2 fewer days over the first year (95% CI 6.5 to 19.8; p<0.001). |
| ReadingThe two figures do not cover the same scope: the initiation gap is described age band by age band, the duration gap is an overall comparison between girls and boys. The observation is counter-intuitive only if one assumes that starting treatment more often translates into better adherence later: these data show that the two do not go together, without saying why. | |
| Discontinuation trajectories | Among the 6661 treated young people: 37% stop within the first year, 19% between one and two years, 16% at three years or more, 29% continue until the end of follow-up. |
| ReadingFewer than a third of the treated young people are still covered at the end of follow-up. A reminder on method: restarting after a discontinuation is not taken into account by the survival analysis, which treats the first discontinuation as a terminal event, and this may overestimate permanent cessation. | |
| Parental education level | No significant association with discontinuation (log-rank p=0.26; RMST not significant). Slower decline in PDC with medium or high parental education (0.7 and 0.6 percentage points per 90-day interval; p<0.001). |
| ReadingThe social gradient appears in medication coverage over time, not in the time to first discontinuation. Writing that a high level of education goes with longer continuation would go beyond what these analyses show. No mechanism is tested here. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Measurement of the outcome | Moderate risk |
| FindingDiscontinuation is defined by a gap of at least 90 days without estimated coverage, that is, by the absence of a dispensing, which is an administrative event. It is the same thing neither as a decision to stop nor as non-adherence. The authors discuss this limitation explicitly. | |
| Residual confounding | Moderate risk |
| FindingSeverity of the disorder, tolerability, concurrent non-drug care: all determinants of stopping that do not appear, or do not appear fully, in administrative registers. The authors acknowledge this, and also point out that psychiatric comorbidity is captured only as a simple count of F-chapter diagnoses. The differences between subgroups remain crude, unadjusted comparisons, not effects. | |
| Interpretation of the sex difference | Untested hypotheses |
| FindingNothing in this dataset allows a choice between the possible explanations: differences in tolerability, symptom profile, age at diagnosis, family expectations or attitudes to treatment. The authors put these forward as hypotheses. What has not been analysed must not be read as a negative result. | |
| External validity | Transfer with caution |
| FindingThe proportions and delays measured in Norway are not benchmarks for other health systems. The authors add that their cohort was diagnosed between 2010 and 2012 and that diagnostic practice has changed since then. The profile of the subgroups that stop early is more plausibly generalisable than the figures themselves, and that remains a hypothesis. | |
| Power and completeness | Strength |
| FindingThe national sample size and the length of follow-up give the subgroup analyses a precision that is unusual in this field. The comparison between medication classes is the exception: only 2.4% of the young people started on a non-stimulant, and the authors themselves describe this comparison as underpowered. | |
Level of evidence
Confidence is high in the description: at this scale and over this duration, the timeline of dispensings is described with a precision that few studies achieve, and the direction of the differences between age bands is hardly open to question. Confidence is lower in what those differences mean. A dispensing register describes dispensing behaviour, not a clinical decision and not medication intake.
On the difference between girls and boys, the statistical observation and its explanation do not sit at the same level of evidence. The first is documented; the second is open, and any causal interpretation, biological or social, goes beyond what these data can support. CEBM level 2b: a cohort, therefore an association.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“That teenagers stop sooner, I see every day, but at least now I have an order of magnitude. What interests me is the finding on girls: they are treated more often and they stop earlier. Mind you, we are talking about prescriptions not collected from the pharmacy, not about girls anyone actually asked. I don’t know whether they stop because it isn’t working, because they are better, or because nobody chased up the follow-up. And this is Norway; dispensing does not work the same way everywhere. In practice, what it mainly tells me is to bring my sixteen-year-old patients back sooner, not to prescribe them something else.”
What this means in practice: this study identifies moments and profiles where follow-up slips. It does not say which treatment to choose, nor whether a discontinuation is bad news.
What you can do with this
- Anticipate the timeline rather than discover it. In these data, 37% of treated young people stop during the first year, and the time on treatment in that first year falls steadily as age at initiation rises. Planning an early review during the first year, and an even closer one when treatment starts in adolescence, costs little and is justified without invoking a risk that is not demonstrated here.
- Do not infer adherence from initial acceptance. A family that readily accepts treatment is not, for that reason alone, a family that will keep it going. This is the most transferable message of this work, and it holds beyond the question of sex.
- Document the reason when a discontinuation occurs. What these registers cannot distinguish, a patient’s notes can: adverse effect, improvement judged sufficient, a trial discontinuation agreed together, or a break in follow-up. Recording which one it was changes what to do at the next appointment.
- Answer a parent who asks how long this lasts. One can say honestly that many children and adolescents stop early, that this does not mean failure, and that the useful duration is revisited at each stage of development rather than decided at the outset.
- Check the local framework before transferring these figures. The rules for prescribing, dispensing and renewing stimulant medication are set country by country. Those that applied to the Norwegian cohort are not necessarily those that apply where you practise, and they shape the real rhythm of renewals: check them in their current version before relying on them with a patient or a family.
Frequently asked questions
Does this study show that adolescent girls tolerate ADHD medication less well?
No. It shows that, in this cohort, girls aged 15 to 17 start treatment more often than boys of the same age (72.8% against 54.2%) and that, across all age bands combined, girls stay on treatment on average 13.2 fewer days during the first year. Tolerability is not measured in this work, and no analysis compares adverse effects between girls and boys. An explanation based on tolerability remains a hypothesis.
Is ADHD medication discontinuation the same as poor adherence?
No, and the distinction is central here. The study observes the absence of any pharmacy dispensing for at least 90 days. A treatment stopped by mutual agreement, an improvement that makes continuation unnecessary in the family’s eyes, and silent drop-out all produce the same signal. The register cannot separate them.
Is a median of 106 days from ADHD diagnosis to first medication abnormally long?
This delay describes the time elapsed between diagnosis and first dispensing in a given health system, with its waiting lists and its initiation pathways. It is comparable neither with a benchmark from another health system nor with a quality standard. Fifty-five per cent of treated young people start within 90 days, twelve per cent more than a year later.
Do these Norwegian figures apply to young patients in other health systems?
The proportions and durations, no: access to care, the ways treatment is started and the dispensing rules differ. What is more reasonably transferable is the shape of the phenomenon, that is, the existence of subgroups that stop earlier. Even that needs confirmation on local data.
Should follow-up be stepped up for girls in particular?
This study tests no intervention, so it cannot answer that. It shows where to look. Closer follow-up in adolescence, whatever the sex, is a reasonable clinical decision, but its benefit remains to be demonstrated.
Annotated bibliography
Source study. Garcia-Argibay M, Hofstad T, Bjelland I, Cortese S, Mykletun A. ADHD medication discontinuation and non-adherence: a Norwegian population-based register study. BMJ Mental Health, 2026; 29(1): e302649. Norwegian national cohort of 8961 children and adolescents aged 5 to 17 years, care and dispensing registers, incident diagnoses from 2010 to 2012, follow-up until 31 December 2021. Contribution: a country-wide description of the timeline of initiation and discontinuation of ADHD medication, with subgroup analyses that few samples allow. Limitations: indirect measurement of adherence through dispensings, reasons for stopping not collected, unadjusted comparisons between subgroups, transferability restricted to comparable health systems. Funding: Norwegian Centre of Expertise for Neurodevelopmental Disorders and Hypersomnias (NEVSOM), with no role in the conduct of the study or in the decision to publish.
DOI 10.1136/bmjment-2026-302649
PMID 42331561
Brikell I, Yao H, Li L, et al. ADHD medication discontinuation and persistence across the lifespan: a retrospective observational study using population-based databases. Lancet Psychiatry, 2024; 11: 16-26. DOI 10.1016/S2215-0366(23)00332-2. Contribution: this is the multinational study with which the authors compare their results, with a discontinuation peak around 18 to 19 years that extends the trend observed here. Limitation, as the authors of the source study put it: sex differences were not formally tested in that work.
Zhang L, Zhu N, Sjölander A, et al. ADHD drug treatment and risk of suicidal behaviours, substance misuse, accidental injuries, transport accidents, and criminality: emulation of target trials. BMJ, 2025; 390: e083658. DOI 10.1136/bmj-2024-083658. Contribution: a reference cited by the source study in support of the idea that continued treatment is associated with fewer serious adverse outcomes. Limitation: these are trial emulations on observational data, and this Norwegian study measures none of those outcomes.
Cortese S. Pharmacologic treatment of attention deficit-hyperactivity disorder. N Engl J Med, 2020; 383: 1050-1056. DOI 10.1056/NEJMra1917069. Contribution: a reference review of drug treatment for ADHD, useful for placing the therapeutic framework that this cohort observes without evaluating it. Limitation: a narrative review, predating the data presented here.
The identifiers of these context references were taken from the reference list of the source publication. Their content was not analysed as part of this verification: they are cited for what the source publication uses them to say.
What was consulted. References verified on 12 August 2026 against the published version and its supplementary material. That material is limited to a summary table of the five analyses (objective, outcome, time horizon, covariates) and the legends of three figures: population selection flowchart, cumulative incidence curves of discontinuation, adherence trajectories. The figures themselves are images without readable values, so none of their values could be checked; the values cited here all come from the published text and its tables. Reported funding: Norwegian Centre of Expertise for Neurodevelopmental Disorders and Hypersomnias (NEVSOM). Competing interests, as declared in the publication: one author (SC) declares reimbursement of travel and accommodation expenses for educational activities on ADHD and honoraria from Medice; the other authors declare no conflict of interest. This article underwent an independent double reading.