Published on 15 September 2026
ADHD medication: at what dose does the benefit stop rising?
In brief
Guidelines give dose ranges, not curves. This network meta-analysis rebuilds the relationship between dose and effect from 113 double-blind randomised trials and more than 25,000 patients, separately in children and in adults. In children and adolescents, median efficacy rises up to 45 mg/day for methylphenidate, up to 25 mg/day for amphetamines and up to 4 mg/day for guanfacine, with no demonstrated benefit beyond. In adults, amphetamines plateau above approximately 50 mg/day, whereas methylphenidate keeps rising with no visible plateau, possibly for want of data. The authors set two opposite errors against each other: settling for an insufficient response without titrating, and pushing the dose beyond licensed limits for no reason.
The context
The question is the most ordinary one in an ADHD clinic. The patient has partly improved, should the dose go up? Until now the answer rested on licensed dose ranges and on habit. The meta-analyses available compared drugs against one another, at fixed or mean doses, without modelling the relationship between dose and effect.
Two opposite errors therefore coexist in practice. Underdosing leaves patients at doses that never reached their useful range. Mechanical escalation prolongs titration well past the point where the average effect stops improving. Without a dose-effect curve, nothing tells the two situations apart.
The method
What a dose-effect meta-analysis changes
A conventional network meta-analysis compares treatments. Here dose is handled as a continuous variable, in a hierarchical Bayesian model with restricted cubic splines. The shape of the relationship is therefore not imposed in advance: it can rise then flatten, rise without end, or fall back.
The protocol was pre-registered on OSF, the trials come from the MED-ADHD database updated in February 2025, with no language restriction, and unpublished data were included. Risk of bias was assessed with the RoB 2 tool. People with lived experience of ADHD took part in the design and in the interpretation. Transitivity, the assumption every indirect comparison depends on, was examined through the distribution of the main effect modifiers, with no anomaly detected.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Patients aged 5 years and older meeting standardised ADHD criteria. Children and adolescents: 68 trials, 14,138 participants, 70.6% male. Adults: 45 trials, 11,016 participants, 54.0% male. Excluded: genetic syndromes, treatment-resistant populations, withdrawal-phase designs | |
| Intervention | |
| Oral monotherapy for ADHD, stimulants and non-stimulants, analysed as a function of daily dose | |
| Comparator | |
| Placebo and indirect comparisons within the network | |
| Outcomes | |
| Efficacy on validated clinical scales, primary outcome. Tolerability measured by discontinuation due to adverse events, secondary outcome | |
| Design | |
| Systematic review and dose-effect network meta-analysis of 113 double-blind randomised trials, hierarchical Bayesian model, protocol pre-registered · CEBM 1a |
Of 15,096 references screened, 164 reports were retained in the systematic review and 113 trials in the dose-effect analysis. Ethnicity of participants is reported inconsistently from one trial to the next.
Quality control
| Criterion | Status |
|---|---|
| Registration | Sound |
| FindingProtocol pre-registered on OSF | |
| Completeness of the search | Sound |
| FindingMED-ADHD database, no language restriction, unpublished data included | |
| Robustness | Sound |
| FindingMultiple sensitivity analyses confirming the findings, no intransitivity detected | |
| Precision at high doses | Reservation |
| FindingWide credible intervals in the upper part of the curves in children and adolescents, a reservation the authors flag themselves: this is exactly where the decision is made | |
| Patient involvement | Sound |
| FindingPeople with lived experience involved in the design and in the interpretation | |
| Funding and declared interests | Reservation |
| FindingUK public funding. Several authors declare links with industry, one of them employed by a pharmaceutical company and holding shares in it at the time of writing, with no declared role of that company in the work | |
| Published correction | Reservation |
| FindingA correction to this article was published in August 2026. Check it before quoting any precise figure | |
The findings
| Drug and age group | Efficacy |
|---|---|
| Methylphenidate, children and adolescents | Rises up to 45 mg/day, no demonstrated benefit above |
| TolerabilityNo clearly dose-dependent risk of discontinuation | |
| Amphetamines, children and adolescents | Rises up to 25 mg/day |
| TolerabilityDiscontinuations for adverse events rise above 25 mg/day | |
| Guanfacine, children and adolescents | Rises up to 4 mg/day, no demonstrated benefit above |
| TolerabilityNot reported in the published abstract | |
| Amphetamines, adults | Plateau above approximately 50 mg/day |
| TolerabilityDiscontinuations rise above 50 mg/day | |
| Methylphenidate, adults | Keeps rising with no visible plateau, possibly for want of data at high doses |
| TolerabilityDiscontinuations rise above 50 mg/day | |
| Atomoxetine and modafinil | No dose-effect relationship shown, atomoxetine having been analysed only in fixed-dose trials |
| TolerabilityNot reported in the published abstract | |
Efficacy and tolerability do not follow the same curve
This is the second contribution of the work, and probably the most useful in the consulting room. One reading point comes before the figures: amphetamine doses are reproduced as published, with no salt and no formulation specified, and the amphetamine products licensed for ADHD differ from one country to another, so these values do not transpose directly to a given product. For amphetamines, the probability of discontinuation for adverse events rises with dose above 25 mg/day in children and above 50 mg/day in adults. For methylphenidate, that rise appears in adults above 50 mg/day, with no clearly dose-dependent risk in children. Where the two curves cross, past the plateau, you pay without buying anything. The rule is not universal: in children on methylphenidate no clearly dose-dependent risk of discontinuation appears, and that is the only one of the four cases where tolerability provides no counterweight.
One point calls for nuance rather than a shortcut. In adults, methylphenidate shows no plateau. The authors do not conclude that the dose should be raised indefinitely: they put forward as a possible explanation that data at high doses are too sparse for the curve to settle. No visible plateau does not mean no plateau.
Critical appraisal
| Domain | Finding |
|---|---|
| Nature of the data | |
| Aggregate data, not individual participant data. The plateau is a group average, not a rule that applies to a given patient, and any responders at high dose cannot be identified | |
| Upper end of the curves | |
| Wide credible intervals, flagged by the authors themselves. The zone where the clinical decision is hardest is also the least well estimated | |
| Population | |
| Treatment-resistant patients are excluded by design: the results do not apply to them | |
| Oral monotherapy only | |
| Combinations and non-oral formulations fall outside the scope | |
| Robustness | |
| Multiple sensitivity analyses, transitivity examined, no anomaly found | |
Level of evidence
PEB appraisal: high confidence in the shape of the curves, moderate confidence in the threshold values. The level of evidence is the highest available for a question of comparative efficacy, the protocol is pre-registered and the conduct exemplary. What limits precision is not the method, it is the raw material: aggregate data and few trials at high doses. The figures of 45 mg/day and 25 mg/day are titration landmarks, not regulatory thresholds.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Curves at last, instead of ranges. What I take from it is that we often stop too early and sometimes push for nothing. But the intervals are wide exactly where I need precision, so I will use it as a landmark and look at the patient, not at the table. ”
What this means in practice: the plateau gives a titration target and a reasonable stopping point, it does not replace the individual assessment of efficacy and tolerability.
What you can do with this on Monday morning.
- Faced with a partial response, consider carrying titration on towards the plateau of the drug rather than stopping short on principle. The authors name growing concerns about subtherapeutic prescribing as one of the reasons for the work, and present their results as support for shared decision making, not as a titration protocol.
- Once the plateau is reached, change the question rather than the dose: adherence, diagnosis, comorbidity, school or working context.
- Watch tolerability as a signal in its own right, particularly on amphetamines, where discontinuations rise with dose while efficacy does not follow.
- Explain the plateau to the patient or to the parents. It is an argument that can be understood in one sentence and it defuses the routine request for an increase.
The doses reported here describe what the trials studied, and a median from the literature is not a prescribing recommendation. Who may initiate a stimulant, on what prescription and for how long, is set nationally and changes over time. Refer to the summary of product characteristics in force where you practise.
Frequently asked questions
Does the plateau mean that no patient benefits from a higher dose?
No. It describes the median effect in a group. Some patients may respond above it, but aggregate data cannot identify them. Going past the plateau then becomes an individual decision to be justified and monitored, not a routine step.
Why does methylphenidate show no plateau in adults?
The authors offer a possible explanation, the scarcity of trials at high doses in adults, which would prevent the curve from settling. They present it as a hypothesis. It is in no way an invitation to raise the dose without limit.
Do these curves apply to long-term treatment?
They describe what double-blind randomised trials measured, and those are short by nature. The dose-effect relationship over several years, with growth, tolerance and adaptation, is not estimated here.
What should be made of atomoxetine?
No dose-effect relationship could be shown in the fixed-dose trials. The study neither supports nor discourages escalation in its case, and that is worth telling the patient.
Is this an argument against ADHD medication?
No, rather the opposite. The work describes curves that rise before they flatten, which gives a numerical landmark for not stopping too low. It does not address the efficacy of stimulants against placebo, which belongs to other work.
Annotated bibliography
Nourredine M, Jurek L, Hamza T, Cipriani A, Subtil F, Parlatini V, Farhat LC, Veronesi GF, Efthimiou O, Salanti G, Cortese S (2026). Pharmacological interventions for ADHD: a systematic review and dose-effect network meta-analysis. The Lancet Psychiatry, 13(6), 485-495. DOI 10.1016/S2215-0366(26)00091-X · PMID 42134365. Source study analysed here, open access. Funded by the National Institute for Health and Care Research. A correction has been published, PMID 42285112.
Cortese S, Adamo N, Del Giovane C, et al. (2018). Comparative efficacy and tolerability of medications for attention-deficit hyperactivity disorder in children, adolescents, and adults: a systematic review and network meta-analysis. The Lancet Psychiatry, 5(9), 727-738. DOI 10.1016/S2215-0366(18)30269-4 · PMID 30097390. The reference network meta-analysis on the choice of drug, which the present work completes on the question of dose.