Effect of ADHD medication on risk of injuries: a preference-based instrumental variable analysis

ADHD is associated with an increased risk of injury. Causal evidence for effects of pharmacological treatment on injuries is scarce. We estimated effects of ADHD medication on injuries using variation in provider preference as an instrumental variable (IV). Using Norwegian registry data, we followed 8051 patients who were diagnosed with ADHD aged 5 to 18 between 2009 and 2011 and recorded their ADHD medication and injuries treated in emergency rooms and emergency wards up to 4 years after diagnosis. Persons with ADHD had an increased risk of injuries compared to the general population (RR 1.35; 95% CI: 1.30–1.39), with higher risk in females (RR 1.47; 95% CI: 1.38–1.56) than males (RR 1.23; 95% CI: 1.18–1.28). The between-clinics variation in provider preference for ADHD medication was large and had a considerable impact on patients’ treatment status. There was no causal evidence for protective effects of pharmacological treatment on injuries overall for young individuals with ADHD characterized by milder or atypical symptoms. However, there was an apparent effect of pharmacological treatment over time on the risk of injuries treated at emergency wards in this patient group. Supplementary Information The online version contains supplementary material available at 10.1007/s00787-023-02294-6.


Covariates
Table S2.Data sources for covariates.

Instrumental variable analysis
Provider preference need to meet the following requirements to be considered a valid IV. [1][2][3] First, provider preference must predict treatment (relevance).This is tested with the F-statistic of the IV in first stage regressions.Second, provider preference can only impact injuries by its effect on treatment (exclusion).This was assessed by reduced form analyses in the general population.Third, provider preference must be as good as random for patients (independence), which we account for by including covariates for patient mix.Fourth, provider preference can only impact patients' treatment either positively or negatively (monotonicity), which is examined by analyses of the relationship between provider preference and medication.Fifth, there should be no interference nor treatment variation (stable unit treatment value assumption) which we assess through analyses of medication type.

Figure S1 .
Figure S1.Risk ratios for specific types of injuries at emergency room in persons with ADHD vs general population by 4 years follow-up.x-axisdiffers for poison due to large estimates (but also low frequency of events).

Figure S3 .
Figure S3.Coefficient plot for first stage results with F-statistics for the IV.

Figure S4 .
Figure S4.Coefficient plot examining balance of covariates for the IV with the joint F-test.

Figure S5 .
Figure S5.Reduced form.Associations between provider preference for ADHD medication and injuries in the general population sample.Coefficient plots with 95% confidence intervals based on linear probability models.

Figure S6 .
Figure S6.Variation between clinics in ADHD medication among patients diagnosed with ADHD for the first to fourth year following ADHD diagnosis.Provider preference for ADHD medication at clinic level as mean defined daily dosages for ADHD medication by years after ADHD diagnosis among patients on x-axis.Residuals from first stage regressions of treatment on IV plotted against values of IV with local polynomial regression line and residual values on right side y-axis.

Figure S6 .
Figure S6.Probit results for association between ADHD medication and injuries.

Figure S7 .
Figure S7.Linear probability model results for association between ADHD medication and injuries in patients aged below and above 12 years.

Figure S8 .
Figure S8.Linear probability model results for associations between ADHD medication and injury by type of medication.

Figure S9 .
Figure S9.Linear probability model results for the association between ADHD medication and types of injuries.

Figure S10 .
Figure S10.IV Probit results for the effect of ADHD medication on the probability of injury.

Figure S11 .
Figure S11.2SLS results for the effect of ADHD medication on the probability of injury excluding patients who had one or more prescription prior to diagnosis.Panel A exclude patients with prescriptions prior to waitlist end date (sample n=6942).Panel B exclude patients with prescriptions prior to diagnosis (sample n=6,528).

Figure S12 .
Figure S12.2SLS results for the effect of ADHD medication on the probability of injury in patients aged below and above 12 years.

Figure S13 .
Figure S13.2SLS results for the effect of ADHD medication on the probability of injury by type of medication.

Figure S14 .
Figure S14.2SLS results for the effect of ADHD medication on the probability of injury for types of injuries.