Article Text

Stimulant treatment and injury among children with attention deficit hyperactivity disorder: an application of the self-controlled case series study design
1. Sudha R Raman1,
2. Stephen W Marshall1,2,
3. Kevin Haynes3,
5. Albert Jackson Naftel4,
6. Til Stürmer1
1. 1 Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA
2. 2 Injury Prevention Research Center, University of North Carolina at Chapel Hill, Chapel Hill, Chapel Hill, North Carolina, USA
3. 3 Department of Biostatistics and Epidemiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, USA
4. 4 Department of Psychiatry, School of Medicine, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA
1. Correspondence to Dr Sudha R Raman, c/o Til Stürmer, Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, McGavran-Greenberg Hall, CB #7435, Chapel Hill, NC 27599, USA; ramans{at}email.unc.edu

## Abstract

Objective The aim of the present work was to assess the short-term effects of stimulant medication use on risk of injury among children diagnosed as having attention deficit hyperactivity disorder (ADHD).

Methods The study group for this self-controlled case series study was children aged 1–18 years old diagnosed as having ADHD who experienced an incident medically-attended injury event and received at least one prescription for stimulant medication between 1993 and 2008 (n=328), identified from The Health Improvement Network primary care database from the UK. Conditional Poisson regression was used to estimate incident rate ratios (IRR) and 95% CIs for injury comparing periods of time exposed to stimulant medication to unexposed periods.

Results Among children with ADHD prescribed stimulant medication, the rate of medically-attended injury was decreased during periods of stimulant medication use as compared to unexposed periods (IRR 0.68, 95% CI 0.50 to 0.91). There was evidence of a protective association among males and among children aged 10–14 years. This effect did not change over time on treatment.

Conclusions Stimulant medication use may decrease the risk of injury among children treated for ADHD, although unmeasured time varying confounding may be an alternative explanation. Injury risk may be considered during the decision-making process with regard to medication continuation among children with ADHD.

View Full Text

## Introduction

Attention deficit hyperactivity disorder (ADHD) in childhood is characterised by inattention, hyperactivity and impulsivity. Children with ADHD experience high rates of injuries, including fractures, head injuries and burns.1–3 While stimulant medications for the treatment of ADHD such as methylphenidate (MPH) have been shown to reduce core symptoms and improve school performance and family interactions,4 it is unclear whether treatment also affects injury risk.

Stimulant medication use is hypothesised to decrease injury risk by reducing ADHD symptoms such as inattention or impulsivity. Conversely, periods of medication non-use may increase risk for injury.5 ,6 Clinical trials have reported too few injuries to evaluate these hypotheses.7 MPH use has been found to improve behaviours that may influence injury rates. For example, MPH improved driving performance among adolescent boys with ADHD when using driving simulators and community driving8–10 and MPH improved pedestrian road crossing behaviour in a virtual reality setting.11 Observational study results evaluating injury rates showed that use of long-acting MPH was associated with a lower odds of injury compared with use of standard formulation MPH.12 In addition, injury rates were slightly lower among adherent stimulant users compared with children who were less adherent.13 Although the latter two studies suggest that overall exposure to stimulant medication use may be protective for injuries, residual confounding may have resulted from uncontrolled between-group factors associated with stimulant use and injury events, such as parental supervision and health care seeking behaviour. Better control of individual-level confounders can be obtained using a study design in which a child's untreated time is used as unexposed time (rather than using untreated children as the unexposed group).

The current study assesses the association between stimulant medication use and risk of injury, using a self-controlled case series method.14 The self-controlled case series design takes advantage of the intermittent periods of medication use that are often observed among children with ADHD to make within-individual comparisons of injury risk between time periods on and off stimulant medications. The unit of analysis is an individual person; therefore the analysis cannot be confounded by individual characteristics that are stable over time (eg, sex, genetics). We hypothesise that injury risk among children with ADHD is decreased during stimulant treatment periods compared with untreated periods.

## Methods

### Study population

This study was conducted using data from The Health Improvement Network (THIN) from 1 January 1993 to 30 June 2008. Contributing UK general practices (GP) provide THIN with longitudinal electronic data at the individual level about GP visits, the date and details of prescriptions and referrals for care received in secondary care facilities.15 The 2008 THIN dataset includes information about 2.3 million patients who were actively registered with 386 GPs. The THIN dataset is obtained from approximately 4% of the UK population and has an age, gender and geographic distribution that is similar to the overall population of the UK.15

Cases were selected from patients who were registered with a THIN practice, had a diagnosis of ADHD, experienced a first-ever incident injury event and who had been prescribed stimulant medication at least once between the ages of 1 and 18 years during the study period.

When a new patient registers with a THIN practice, diagnoses and events that are part of the patient's history may be recorded using the date that the information was entered into the computer, not the date of the event. Therefore we used only events and diagnoses that occur more than 12 months after registration as these are most likely to be dated correctly.16

Information in THIN is coded using the Read clinical classification system. The Read codes used to define ADHD included any code that referred specifically to ADHD, attention deficit disorder or hyperkinetic disorder, plus the two most common codes in THIN that refer to ADHD symptoms (see online only appendix).

### Injury outcome

Each child contributed a single, first-ever medically-attended injury event to the analysis. A medically-attended injury was defined as a record that described bodily harm with an ‘injury and poisoning’ Read code at any location of service (GP, emergency room or hospital). Injury and poisoning Read codes are classified into 27 injury types. All injury types were included except for ‘late effects of injury or poisoning events’ and ‘medical/surgical procedures causing complications’.

### Exposure to medication

MPH and dexamphetamine were the only types of stimulant medications approved for ADHD treatment in the UK during the study period.17 Different formulations (eg, standard formulation and long-acting MPH) and all strengths (eg, 20, 30, 40 mg) for MPH and dexamphetamine were used to determine exposure status.

Assessment of stimulant medication treatment periods required the determination of the treatment length for each prescription (ie, the no. of days supplied). The days supplied value was calculated from the prescribed daily dose and quantity whenever possible (55% of all prescriptions). For the remaining 45% of prescriptions where either dose or quantity prescribed was missing, the number of days supplied was set to 30 days. This value was chosen because 82% of the calculated days supplied values were between 28 and 32 days and a 30-day supply is the recommended prescription length for controlled drugs (such as stimulant medication) in the UK.18

Observation time started on the day of initial ADHD diagnosis and continued until the earliest of day the child turned 19 years old, died, transferred out of the practice or the last date of data collection from the practice. All available observation time was categorised as either treated or untreated. A treatment period started on the day a stimulant was prescribed and extended until the end of the days supplied value plus a 30-day grace period unless another stimulant was prescribed within this period. Any observation time that was not classified as treated was classified as untreated. Each child could have multiple treated and untreated periods. To assess for differences in the effect of medication according to the time since the start of a treatment period, observation time during the first 30 days of each continuous treatment period was classified separately from observation time occurring after the 30th day of treatment (figure 1).

Figure 1

Self-controlled case series study design for a single individual prescribed medication. The length of each segment of time varies between individuals. Injury could occur in either treated or untreated periods. Rate ratios are pooled estimates comparing the rate of injury events during treated periods divided by the rate of events during untreated periods. Unexposed person-time can occur prior to the first medication prescription or after a medication course. ADHD, attention deficit hyperactivity disorder.

Since the length of the effect of both types of ADHD medication is less than a day (about 3–6 h for standard MPH and about 8 h for long-acting MPH),17 ,19 the grace period in this study was established not primarily as a biological washout period for the medication but to account for misclassification of days that a child may be on medication. Children with ADHD may not use medication every day (eg, only on school days but not weekends) and thus a 30-day supply may last longer than 30 days. We assumed that this irregular use would be similar for the standard formulation of MPH as well as the long-acting MPH and thus used the same grace period for both types of medication.

### Application of the self-controlled case series method

As stated above, the self-controlled case series design was used in order to control for individual-level time-independent factors that might confound the estimation of the effect of stimulant treatment on injuries, such as parental supervision and care-seeking behaviour. There are three relevant methodological considerations when applying the self-controlled case series design. First, while the design controls for confounding by factors that are stable over time, time-varying risk factors for the outcome within an individual may lead to within-person confounding. Therefore, two strong risk factors for injury, age (in 2-year categories) and season (winter, spring, summer, autumn) were controlled for in the analyses. Second, events must be independent within a person, an assumption that does not hold with injuries as the occurrence of an injury may alter the risk of a subsequent injury.20 Thus this analysis includes only the first injury event during the observation period.21 Finally, an assumption of the self-controlled case series is that the treatment at any time point is not affected by the occurrence of the event at a prior time point. In this study, there was concern that a medically-attended injury may precipitate the initial prescription of medication among children with ADHD. To address this hypothesis, we assessed the distribution of injuries in the 6 months prior to initial medication prescription and found no indication that injury events precipitate initial medication. Therefore all person-time between diagnosis and initial medication prescription was included in the main analysis. To further assess whether injuries that occurred prior to the first treatment may have affected the main estimate, we also repeated the analysis excluding untreated time prior to medication initiation.

### Analysis

The primary analysis assessed the impact of stimulant medication on the occurrence of first incident injury by comparing the rate of events during all periods of treatment with stimulant medication to the rate during all untreated time periods. Self-controlled incident rate ratios (IRR) and their 95% CIs were estimated using conditional Poisson regression. All data were analysed using SAS V.9.2 (SAS Institute Inc, Cary, North Carolina, USA).

Stratified analyses were conducted to assess the association between stimulant medication and injury events by gender, age group (1–5, 6–10, 11–14, 15–18 years) and type of medication.

### Sensitivity analyses

1. Medically-attended injury: we performed several sensitivity analyses using different definitions for injury events. Types of injuries that were hypothesised to be more consistently medically-attended (upper limb, lower limb and skull fractures, head injuries) were examined separately. Injuries were also stratified by the day of the week (weekend or weekday).

2. Definition of ADHD: the broadening of the diagnostic criteria for ADHD and the associated increase in stimulant treatment for ADHD in recent years may modify the association under study. Children treated with stimulants before and after the year 2000 were therefore analysed separately.

3. Assumptions of the study design: the results of self-controlled case series may be biased by incorrect specification of the treatment period and by the inclusion of events that alter the distribution of post-event treatments. Sensitivity analyses were undertaken to determine if the results were influenced by (i) varying the length of the grace period used to determine treatment period length (eg, 15, 45, 60, 90 days), (ii) excluding untreated person-time prior to medication initiation or, (iii) excluding fatal injuries that obviously precluded post-event treatment.

The University of North Carolina Public Heath and Nursing Institutional Ethics Review Board approved the study protocol.

## Results

From 4234 eligible children in THIN data, 328 children had an incident injury event after initial ADHD diagnosis and at least 1 prescription for stimulant medication. This study group was predominantly male (86.9%, n=285) and the mean age at ADHD diagnosis was 9.7 years (SD 3.0). The mean person-years of observation per child was 5.9 years (SD 2.7). About 22% (n=78) of children were prescribed medication on the day of the diagnosis (the start of the observation period), and the majority initiated medication within 1 year of diagnosis (73%, n=239).

The initial type of medication prescribed was MPH for 98.5% (n=323); 76.8% (n=252) and 21.7% (n=71) were prescribed standard formulation MPH and long-acting MPH, respectively (table 1). Using a 30-day grace period to define exposure, 23.8% had a single treatment period, 46.0% had 2–5 treatment periods, 20.1% had 6–10 treatment periods, and 9.1% had more than 10 treatment periods.

Table 1

Description of study population: children with ADHD who have at least one prescription for stimulant medication and an incident injury event (n=328)

Common types of injuries included upper limb fractures (15.6%), head injuries (14.0%), sprains and strains (11.3%) and superficial injuries (11.0%). Fractures occurred in 24.4% (n=80) of the study group (table 2).

Table 2

Types of injury events as classified by Read code subchapters

The unadjusted incident rate ratio (IRR) for an injury event comparing treatment periods to all untreated periods was 0.70 (95% CI 0.53 to 0.94). Adjustment for age (in 2-year categories) and season did not change the estimate substantially (IRR 0.68, 95% CI 0.50 to 0.91).

Compared to untreated periods, the rate of injury was lower during the first 30 days of any stimulant (IRR 0.62, 95% CI 0.38 to 1.02) and after the first 30 days of any stimulant treatment course (IRR 0.69, 95% CI 0.51 to 0.94).

In subgroup analyses, there was stronger evidence of a protective association between medication and injury among boys (IRR 0.63, 95% CI 0.46 to 0.87) than among girls (IRR 1.01, 95% CI 0.46 to 2.21). Age group estimates were less precise than the main estimate; only among children aged 10–14 years did the estimate provide evidence of a protective association (table 3).

Table 3

Self-controlled case series analyses for the association between stimulant medication treatment and first medically-attended injury event, stratified by case characteristics

Effect estimates stratified by injury type, day of injury and period of time (pre/post 2000) were less precise but similar to the main estimate, except for the estimate for head injuries (IRR 1.30, 95% CI 0.63 to 2.67; table 4). Excluding five children who were initially treated with dexamphetamine did not change the IRR (IRR 0.71, 95% CI 0.53 to 0.95).

Table 4

Results of sensitivity analyses

Excluding untreated time prior to medication initiation (and thus the cases with an injury event during this time) yielded a slightly higher estimate (IRR 0.83, 95% CI 0.55 to 1.20). The results did not change substantially when the length of the grace period was varied or when the one fatal injury event was excluded (table 4).

## Discussion

In this study, stimulant medication was associated with a decreased risk of injury in children treated for ADHD. This protective effect was clearly present among boys and children 10–14 years of age. The first injury rate was decreased during treatment periods within the first 30 days of a treatment course and in the remainder of the treatment course, as compared to untreated periods. The estimates appeared to be minimally influenced by changes in outcome and exposure specification relevant to the assumptions of the self-controlled case series design.

The results support the hypothesis that treating children with ADHD with medication can decrease injury risk. We were not able to conclude whether medication had an effect in girls due to the low number of female injuries and resulting loss in precision for the analysis of girls. A gender difference in the effect of medication is plausible as injury experiences and ADHD diagnosis also differ by gender. The increased rates of all types of injuries among boys as compared with girls across all ages has been attributed to a complex interaction of behavioural, environmental and social factors.22 ,23 Information about type and severity of ADHD symptoms was not available in this study. However it is possible, if the distribution of symptoms and/or types of injuries varies by gender, that the association between stimulant medication and injury risk may also be different between genders.

The protective effect of medication in the initial 30 days and in the later portion of each treatment course indicates that the effects of stimulant medication are constant over time on treatment. This is consistent with research that suggests that medication has direct effects and indirect effects, such as the facilitation of improved function at home and school over time.24 Additionally, these results can support recent treatment guidelines that discourage breaks in stimulant medication use (specifically breaks in treatment greater than 1 month) and highlight decreased injury risk as a potential additional benefit of medication.25

Our study examined whether stimulant medication had an acute effect on injury risk among children with ADHD treated with stimulants. Two previous observational studies focused on related dimensions of the association between stimulant medication use and injury risk. Comparing children by stimulant medication type, long-acting MPH was associated with fewer outpatient visits or hospitalisations for injuries as compared with the standard formulation of MPH.12 In a second study, Marcus et al assessed injury risk among stimulant-treated children over a single treatment period of at least 4 months. Children who were more adherent to medication were less likely to experience an injury than children who were less adherent (injuries/person-year: 0.227; 95% CI 0.214 to 0.240 vs 0.230; 95% CI 0.204 to 0.254; adjusted HR 0.89, p=0.07).13 Both studies compared different groups of children and thus attempted to control for relevant child-level confounders. However the use of administrative data limited the measurement of confounders that may have been relevant to medication use/prescription and injury such as family functioning (eg, level of supervision), healthcare-seeking behaviour or ADHD severity.

This study was able to build upon the previous literature to examine injury rates for children with ADHD during periods on and off medication and assess injury risk over different periods of treatment. The key strength of the self-controlled case series design in this context is that the results are perfectly controlled by design for time-invariant differences between individuals.

Our results suggest that medication use, and the presumed resulting decrease in ADHD symptoms, is associated with decreased injury risk. However, confounding could nevertheless occur in the self-controlled case series design if an uncontrolled factor was associated with the timing or duration of treatment periods and risk of injury (such as a change in school environment, parental supervision, non-pharmacological treatments or a change in the underlying severity of ADHD symptoms). For example, the unexpected finding of higher rates of head injuries during stimulant treatment could reflect confounding by an unmeasured time varying factor, such as an increase in the reporting of head injuries during periods of stimulant treatment. A time varying factor would need to have a large effect on injury risk among many children within the study group to substantially affect our findings. Future research should examine the distribution and effect of these factors to determine their impact on this association.

We recognise there are a number of limitations to the current study. The first limitation concerns the diagnosis of ADHD. According to national UK guidelines, an ADHD diagnosis should be made by a specialist (usually a paediatrician or psychiatrist) when a child meets the diagnostic criteria for ADHD according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) or the International Classification of Diseases, 10th edition (ICD-10) (hyperkinetic disorder, which is viewed as severe ADHD), exhibits impairment (based on interview or direct observation) and has pervasive symptoms, (ie, symptoms occur in two or more settings17). Apart from these general guidelines we cannot know what specific criteria doctors used when assigning ADHD codes to the children in our study. However we would not expect that some lack of specificity of ADHD diagnosis would have a major impact on our results.

Injury events have not been validated in THIN data. As such, some medically attended injury events may not have been recorded in THIN data (eg, injuries that were attended to in after-hours clinics but not reported to the child's GP). This limitation could have reduced sample size; however, in order to bias the study results, these injured children would have to differ systematically from the study participants in how medication treatment was associated with the timing of injuries.

Exposure misclassification may have occurred due of the nature of the THIN primary care dataset. Following diagnosis and initial medication prescription by specialists, continued prescribing and monitoring of ongoing pharmacological treatment may be performed by general practitioners, under ‘shared care’ arrangements.17 Data about specialist prescribing is not routinely included in THIN, and this may introduce bias. The exclusion of children who are prescribed medication exclusively by specialists would decrease sample size and potentially limit the generalisability of our results to all children with ADHD if the excluded children differ systematically in how medication is associated with injury events. Misclassification of person time as unexposed when in fact the child has received ADHD medication from a specialist may also bias the effect estimate, however, this bias is likely towards the null. No empirical data exists about the distribution of ADHD prescribing between consultants/specialists and GPs. However a substantial number of ADHD medication prescriptions do originate from primary care and have been used to obtain estimates of the incidence and prevalence of drug-treated ADHD that have been comparable to other European countries.29 It is also reasonable to suggest that GPs may prescribe the majority of ongoing ADHD prescriptions, because of the guidelines on shared care and because child mental health specialist services are not as accessible in all regions.30 ,31 Hence, we expect that the misclassification of exposure due to specialist prescriptions would not have affected the overall results to a great extent.

This study was able to take advantage of the detailed prescription information to characterise treatment periods. However, prescription records are only a proxy for prescription dispensing and actual consumption and do not necessarily indicate exposure to stimulant medication on a given day during the observation period.

Finally, a 30-day grace period, as used in similar ADHD medication pattern research,32–34 was chosen so that children who take medication only during the school week would not be misclassified as having discontinued treatment too early. Sensitivity analyses indicated that these results were not sensitive to the definition of the grace period.

## Conclusions

This study supports the clinical hypothesis that current stimulant medication use may reduce risk of injury among children with ADHD treated with stimulants. These findings are important for care and treatment of children with ADHD. However, it is important to note that this analysis cannot answer the question whether injury risk can be reduced by stimulant medication initiation in children with ADHD not previously treated. Up to 15% of children in the UK experience a medically-attended injury each year35 and children with ADHD experience a disproportional number of these injuries.2 Continued stimulant medication treatment may be able to prevent injuries and the resulting impairments in this high-risk group. Consideration of the potential benefit of continuous stimulant medication on injury risk may thus aid clinicians and families in the decision-making process about stimulant medication use.

### What is already known on this subject

• Children with attention deficit hyperactivity disorder experience high rates of injuries. Clinicians hypothesise that stimulant medication use can decrease injury risk by reducing attention deficit hyperactivity disorder (ADHD) symptoms; however there are few examples of a clear association between injuries experienced in the community and medication use.

• For children with ADHD who are treated with medication, periods of stimulant medication use were associated with lower injury rates as compared with untreated periods.

• Injury risk may be considered in the decision-making process about the continued use of stimulant medication for ADHD in treated children.

• This analysis, however, cannot answer the question whether injury risk can be reduced by stimulant medication initiation in children with ADHD not previously treated.

### A hit and run kills newlyweds

Expectant parents in a cab en route to see a doctor in Brooklyn were hit by a BMW whose driver and passenger fled. The report states that ‘It was not clear if one or both of the drivers was at fault’ but the driver of the BMW will be charged with hit and run. The wife, 24 weeks pregnant, was rushing to seek medical attention because she could no longer feel the baby. Editor's comment: This was sent to me by Lou Lombardo who runs the Care for Crash Victims website and who suggests that ‘National Highway Traffic Safety Administration (NHTSA) should start counting the number of orphans that result from fatal crashes.’ I certainly agree. http://www.nytimes.com/2013/03/04/nyregion/expectant-couple-killed-on-way-to-hospital.html

### Banking firms targeted in US gun control debate

In a column in PostMedia news at the end of January, William Marsden reported that the Toronto Dominion (TD) Bank, one of Canada's largest, was being asked to stop lending money to Smith and Wesson (SW), the gun manufacturer that makes the AR-15 assault weapon. The request came from the mayor of Chicago, not just because SW make guns, but because they, not surprisingly, do not support gun controls. Apparently TD gives SW a $60 million revolving line of credit. The mayor sent a similar letter to the Bank of America which has given a$35 million line of credit to Sturm, Ruger and Co. another gun manufacturer. Editor's comment: Sadly, I think this approach is far more likely to succeed (if the banks agree) than public opinion alone.

## Acknowledgments

CSD Medical Research in London, UK kindly provided the THIN data.

View Abstract

• ## Supplementary Data

This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Files in this Data Supplement:

## Footnotes

• Contributors SRR drafted, designed and implemented this study under the supervision of SWM and TS. SRR contributed to the conception and design of this study, the conduct of the statistical analysis and its interpretation, the construction of the manuscript and drafting its revisions, and critical review of the article. SWM and TS contributed to the conception and design of this study, the conduct and interpretation of the statistical analysis and revisions of the article for important intellectual content. KH, BNG, AJN contributed to the design of this study, the statistical analysis and interpretation of the results, critical review of the article for important intellectual content and essential edits of the initial and revised manuscript.

• Funding This work was supported by a University of North Carolina—GSK Center of Excellence in Pharmacoepidemiology and Public Health Doctoral Fellowship. KH received an NIH Clinical and Translational Science Award (grant #UL1-RR024134) to work on this manuscript. TS receives investigator-initiated research funding and support as Principal Investigator (R01 AG023178) from the National Institute on Aging at the National Institutes of Health. He also receives research funding as Principal Investigator of the UNC-DEcIDE center from the Agency for Healthcare Research and Quality. TS does not accept personal compensation of any kind from any pharmaceutical company, though he receives salary support from the Center for Pharmacoepidemiology (GlaxoSmithKline) and from unrestricted research grants from pharmaceutical companies (Merck, Sanofi) to UNC. The above-mentioned funding sources did not have any input into the manuscript under consideration.

• Competing interests None.

• Ethics approval University of North Carolina Public Heath and Nursing Institutional Ethics Review Board.

• Provenance and peer review Not commissioned; externally peer reviewed.

## Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.