Self-Harm Incidence Peaked Before ADHD Medication Initiation

Key Takeaways

• Among people prescribed ADHD medication who had recorded self-harm in healthcare databases in Hong Kong, New Zealand, South Korea, Taiwan, and the UK, relative incidence peaked before treatment initiation.
• Relative incidence remained elevated during early and later medication use compared with non-exposure time, but was lower than before initiation.
• Investigators observed a similar temporal pattern across the databases and did not regard medication as the primary driver of the excess incidence.

Recorded self-harm near the start of attention-deficit/hyperactivity disorder (ADHD) treatment raises a difficult question: did the circumstances leading to a prescription precede the event, or might medication exposure be involved? Regulatory concern extends to the United States, where atomoxetine carries a Food and Drug Administration boxed warning. For people beginning treatment, the timing of events around the first prescription is central to interpreting that uncertainty.

Investigators conducted a multinational, population-based self-controlled case series using a common data model across healthcare systems in Hong Kong, New Zealand, South Korea, Taiwan, and the UK from 2001–2020. They identified 461,024 people with an ADHD medication prescription; 6,847 with an incident self-harm event contributed to the self-controlled analysis.

Eligible people were aged at least 6 years. Exposure was recorded through stimulant or non-stimulant prescriptions or dispensing. The primary event was first clinically recorded intentional self-poisoning or self-injury, regardless of apparent suicidal intent, not completed suicide. Within-person non-exposure time, including baseline time outside the designated preinitiation window, was the comparator for the 90 days before first exposure, the first 90 days on medication, and subsequent use. These were not randomized head-to-head comparisons between windows. Conditional Poisson models adjusted for age, season, and psychotropic comedication; local estimates were pooled using a random-effects model.

Before medication initiation, recorded self-harm reached its highest relative incidence against within-person non-exposure time. The pooled adjusted incidence rate ratio (IRR) was 2.58 (95% confidence interval [CI] 2.12–3.13).

During initial medication use, the pooled adjusted IRR was 1.78 (95% CI 1.23–2.57) against the same non-exposure comparator. The estimate was lower than before treatment, but the windows were not randomized comparisons. Across medication exposure overall, incidence was not clearly different from non-exposure time.

During subsequent medication use, the pooled adjusted IRR was 1.27 (95% CI 1.01–1.60) versus non-exposure time. Relative incidence remained elevated but was lower than before treatment. Investigators observed a similar temporal pattern across the included databases.

Recorded prescriptions or dispensing may not reflect actual medication use because of nonadherence or treatment gaps. Uncoded self-harm can escape ascertainment, while differences in healthcare access or diagnostic coding across locations can affect observed absolute rates. Within-person comparisons account for stable individual differences but not unmeasured changes in ADHD features, comorbidity, stress, treatment-seeking, or concurrent care around initiation. Neither the decline after the pretreatment peak nor elevated incidence during use establishes medication-caused protection or harm; care surrounding initiation could also contribute to the observed timing.

The authors concluded that the pretreatment peak does not support ADHD medication as the main driver of excess recorded self-harm incidence. They raised untreated ADHD features as a possible explanation rather than an established cause. Given the elevated incidence around treatment initiation, they emphasized monitoring and counseling both before and during medication use.

Clinician Questions

How were ADHD medication gaps classified when examining self-harm after treatment started?

Treatment gaps or drug holidays followed by another prescription did not restart the initiation window. Subsequent exposure after the first exposure was classified as later use; time outside exposure and the designated preinitiation window counted toward the non-exposure baseline. Prescription records did not confirm adherence.

Were stimulant and non-stimulant ADHD medications examined separately for self-harm?

Investigators assessed stimulant and non-stimulant ADHD medication groups separately and reported broadly similar timing patterns for recorded self-harm. In those drug-class analyses, individuals switching from the class being assessed were censored at the switch date. The subgroup patterns do not establish identical effects or causation.