A risk ratio of 1.34 is not a 34% probability of an adverse event. It compares the event risk in a treatment group with the risk in its comparator. A 2025 meta-analysis of prescription stimulants reported that ratio for overall adverse events, but interpreting it requires the comparison risk, drug classes, follow-up and uncertainty.
The review included lisdexamfetamine and other amphetamines alongside methylphenidate. Its combined estimate should not be relabeled an amphetamine-only risk, an illicit-methamphetamine risk or an individual forecast.
What did the review include?
Oliva and colleagues published the meta-analysis in JAMA Network Open on May 9, 2025. It included 93 placebo-controlled randomized clinical trials and reported 16,382 participants overall. The included trials addressed ADHD and several other clinical conditions, as well as healthy-volunteer research.
ADHD predominated: the authors identified 77 trials, or 82.7%, involving ADHD alone. Trials differed in stimulant, formulation, dose, population and duration. Follow-up extended up to 52 weeks, but many studies were much shorter. The overall population count does not mean every participant contributed to every adverse-event or vital-sign analysis.
This was a synthesis of clinical trial reports. It did not administer a common methamphetamine dose or collect one standardized set of toxicology specimens. Prescription amphetamine formulations, lisdexamfetamine and methylphenidate should retain their individual names rather than being collapsed into methamphetamine.
What did the principal estimate mean?
The pooled risk ratio for overall adverse events was 1.34, with a 90% confidence interval of 1.27–1.41. This corresponds to an estimated 34% relative increase compared with placebo, not an increase of 34 percentage points.
For illustration only, if a comparison population had a 20% event risk, multiplying by 1.34 would give 26.8%, an absolute increase of 6.8 percentage points. If the comparison risk were 50%, the same multiplication would give 67%, an increase of 17 percentage points. Those hypothetical values explain the arithmetic; they are not the review’s observed pooled event rates.
A risk ratio alone therefore cannot supply a universal number needed to harm or a fixed absolute probability for a patient. The baseline risk and the relevance of the study population matter.
Why must the confidence level be preserved?
The paper reports 90% intervals and specifies a two-sided significance threshold of p < 0.10. Those are the study’s stated choices. Its intervals should not be quoted as 95% confidence intervals.
Confidence intervals describe statistical uncertainty around an estimated effect under the analytical assumptions. They are not a probability range for one person’s outcome. They also do not incorporate every concern about differences between trials or the transfer of a pooled finding to a different population.
Were lisdexamfetamine and other amphetamines separately considered?
Yes. The authors report that the overall-adverse-event association remained statistically significant when methylphenidate, lisdexamfetamine and other amphetamines were analyzed separately. That finding supports attention to the amphetamine-related evidence, but does not make the combined 1.34 ratio the estimate for each subgroup.
Likewise, separate significant comparisons with placebo do not establish that the stimulant classes have identical risks or that one is safer than another. Such comparisons require the relevant direct or appropriately modeled between-group analysis.
The review reported substantial variation in its overall adverse-event analysis, with I² = 67%. That statistic describes heterogeneity among study estimates. It is not the percentage of participants harmed and does not mean 67% of adverse events were caused by medication.
What were “adverse events” in this evidence?
The primary outcome concerned developing any reported adverse event. It did not mean that every event was severe, that every participant stopped medication, or that every reported symptom had an independently established drug cause.
The review separately analyzed decreased appetite, headache, insomnia, dry mouth, nausea, irritability and anxiety. It also examined vital signs. Combining these categories into an undifferentiated claim of “serious harm” would misstate the endpoint.
Adverse-event occurrence, severity, clinical importance and discontinuation are distinct questions. A trial can report more nonserious symptoms while reporting few serious events, and neither observation erases the other.
How should its cardiovascular findings be bounded?
The authors found higher pooled diastolic blood pressure and heart rate in stimulant groups, while the systolic-pressure comparison was nonsignificant. They judged the pooled vital-sign changes not clinically significant. That judgment concerns the measured average differences under the reviewed protocols.
It does not establish that no individual experienced a clinically important change, nor that rare cardiovascular outcomes are excluded over many years. The review itself distinguishes its controlled trial evidence, with follow-up no longer than about one year, from longer observational and pharmacovigilance evidence.
Those evidence types have different strengths and limitations. A spontaneous report cannot automatically establish drug causation; a short randomized trial cannot automatically resolve rare long-term risks.
Did the review establish an absence of misuse or psychosis risk?
No. The authors describe misuse potential and the emergence of psychosis as understudied and call for more standardized assessment. The absence of directly identified addiction behaviors in the included trials is not proof that misuse cannot occur outside those controlled settings.
Drug liking, dependence, diversion, psychiatric symptoms and an adverse-event count are different outcomes. An observed result on one cannot stand in for a validated estimate of every other outcome.
What is the defensible conclusion?
This historical meta-analysis found a higher reported adverse-event risk with prescription stimulants than with placebo across its included trials. The numerical estimate is relative, pooled across classes and accompanied by 90% uncertainty intervals and substantial heterogeneity.
For an individual forensic question, it supplies neither an amphetamine concentration threshold nor proof that a particular symptom, crash or psychiatric episode was caused by the drug. Preserve the formulation, population, endpoint and observation period before transferring a group estimate to a case.
Scientific source
Oliva HNP, Prudente TP, Mayerson TF, et al. Safety of Stimulants Across Patient Populations: A Meta-Analysis. JAMA Network Open. 2025;8(5):e259492. doi:10.1001/jamanetworkopen.2025.9492. Full-text meta-analysis. Historical synthesis of placebo-controlled human trials across stimulant classes.
Leave a Reply