ASSOCIATION BETWEEN PSYCHOSTIMULANT AGENTS AND SUICIDE RELATED EVENTS IN
INDIVIDUALS WITH ATTENTION-DEFICIT HYPERACTIVITY DISORDER
by
CHRISTINE LEIGH BAGLEY Pharm.D., University of Georgia, 2010
A dissertation submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment
of the requirements for degree of Doctor of Philosophy Clinical Science Program
ii
This dissertation for the Doctor of Philosophy by Christine Leigh Bagley
has been approved for the Clinical Science Program
by
Lisa Cicutto, Chair Robert Valuck, Advisor
Heather Anderson Glen Schumock Christian Hopfer
iii Bagley, Leigh Christine (PhD, Clinical Science Program)
Association between Psychostimulant Agents and Suicide Related Events in Individuals with Attention-Deficit Hyperactivity Disorder (ADHD)
Dissertation directed by Professor Robert Valuck ABSTRACT
Attention-Deficit Hyperactivity Disorder (ADHD) is a common condition among the pediatric population with escalating use of psychostimulant agents to treat ADHD. No previous studies have directly aimed to evaluate the potential association between psychostimulant medications and suicide related events. Over the past decade there has been an increasing awareness for the potential of prescription medications to increase the risk of suicide related events, including agents such as antidepressants, leukotriene modifying agents, atomoxetine, and antiepileptics. Given the lack of data to support or refute the association between psychostimulants and suicide related events, additional, more definitive studies are warranted. Therefore, it was the objective of this dissertation to determine if a potential association exists between psychostimulants and suicide related events utilizing three different methodologies. This series of analyses were the first well-conducted observational studies investigating this association. We well-conducted an ecologic study utilizing Food and Drug Administration (FDA) adverse event data, a time series analysis describing overall Attention-Deficit Hyperactivity Disorder (ADHD) diagnosis and treatment characteristics as well as suicide attempt trends using aggregated medical claims data, and a nested case-control study utilizing person level medical claims data. Each aim provided additional insight on the potential association of suicide related events and psychostimulants. Our first analysis using FDA data indicated that psychostimulants had no statistically significant association with suicide related events with the exception of lisdexamfetamine. Our second study showed that the percentage of suicide attempts and psychostimulant coverage declined over time among our ADHD cohort. Our
iv
third analysis, a case control study in an ADHD cohort of children and adolescents showed that psychostimulants are not associated with an increased risk of suicide attempt. Based on our findings across all analyses and accounting for the limitations of each, we conclude that
psychostimulant treatment is not associated with an increased risk of suicide related events. In fact, while not statistically significant our findings suggest a potentially protective role in this association thus further supporting the safety of these agents which play a beneficial in the treatment of ADHD.
The form and content of this abstract are approved. I recommend its publication. Approved: Robert Valuck
v
vi
ACKNOWLEDGEMENTS
COMIRB Protocol 12-1033 Initial Application deemed not human subject research and exempt from review. COMIRB response presented in Appendix A.
Dissertation supported by NIH/NCATS Colorado CTSI Grant Number UL1 TR001082. Contents are the authors sole respo si ilit a d do ot ecessarily represent official NIH views.
vii TABLE OF CONTENTS
CHAPTER PAGE
I. INTRODUCTION ... 1
1.1 Attention-Deficit Hyperactivity Disorder (ADHD) ... 1
1.2 ADHD Treatment ... 3
1.3 Psychostimulant Warnings And Drug Label Revisions ... 6
1.4 Study Purpose ... 7 1.5 Specific Aims ... 9 1.5.1 Study Hypothesis ... 10 1.5.2 Specific Aim 1 ... 10 1.5.3 Specific Aim 2 ... 10 1.5.4 Specific Aim 3 ... 10
II. REVIEW OF THE LITERATURE ... 13
2.1 Suicide ... 13
2.2 Defining Suicide Related Event Nomenclature ... 14
2.3 Psychostimulant Mechanistic Plausibility... 15
2.4 Evidence of Psychostimulants-Suicide Association ... 16
2.5 Implications of this Study ... 19
III. METHODS ... 20
3.1 Study Overview... 20
3.2 Specific Aim 1 ... 20
3.2.1 Data Source for Suicide Related Events ... 20
3.2.2 Data Sources for National Psychostimulant Prescriptions ... 22
viii
TABLE OF CONTENTS (Continued)
CHAPTER PAGE
3.2.4 Statistical Analysis ... 25
3.3 Specific Aim 2 ... 26
3.3.1 Data Source ... 26
3.3.2 Annual ADHD Incident and Prevalent Rate Measures ... 32
3.3.3 Monthly ADHD Related Measures ... 32
3.3.4 Statistical Analysis ... 33
3.4 Specific Aim 3 ... 33
3.4.1 Data Source ... 36
3.4.2 Study Cohort ... 36
3.4.3 Study Population ... 39
3.4.4 Definition of Psychostimulant Exposure ... 40
3.4.5 Covariates ... 41
3.4.6 Statistical Analysis ... 41
3.4.7 Sensitivity Analysis ... 42
IV. RESULTS ... 44
4.1 Specific Aim 1 ... 44
4.1.1 Spontaneous Reports, Prescription Frequencies, and Aggregate Suicide Related Event Reporting Rates ... 44
4.1.2 Empirical Bayes (EB) Rate Multiplier Estimates ... 53
4.2 Specific Aim 2 ... 56
4.2.1 Annual ADHD Incidence And Prevalence Rates ... 56
4.2.2 Monthly ADHD Related Measures ... 63
ix
TABLE OF CONTENTS (Continued)
CHAPTER PAGE
4.3.1 Study Population ... 71
4.3.2 Conditional Logistic Regression Analysis ... 75
4.3.3 Stratified Analyses ... 78
V. DISCUSSION/CONCLUSION ... 79
5.1 Study Overview... 79
5.2 Specific Aim 1 Discussion ... 79
5.2.1 Specific Aim 1 Limitations ... 81
5.2.2 Specific Aim 1 Conclusions ... 82
5.3 Specific Aim 2 Discussion ... 83
5.3.1 Specific Aim 2 Limitations ... 87
5.3.2 Specific Aim 2 Conclusions ... 88
5.4 Specific Aim 3 Discussion ... 88
5.4.1 Specific Aim 3 Limitations ... 91
5.4.2 Specific Aim 3 Conclusions ... 92
5.5 Overall Conclusions ... 92
REFERENCES ... 95
x LIST OF TABLES
TABLE PAGE
Table 1. Psychostimulant agents approved by the Food and Drug Administration
(FDA) to treat ADHD ... 4 Table 2. Non-stimulant agents approved by the FDA ... 5 Table 3. Number of SREs reported, prescriptions (RX), and SRE reporting rate for
psychostimulant agents by year in the US. ... 45 Table 4. Number of SREs reported by drug and type ... 46 Table 5. Empirical bayes rate multiplier estimates and confidence intervals by drug:
psychostimulant agents and selective serotonin reuptake inhibitors model ... 53 Table 6. Empirical bayes rate multiplier estimates and confidence intervals by drug:
psychostimulant agents and alpha-adrenergic agonists model ... 55 Table 7. Overall time series results of ADHD related measures for April 2001 to
December 2011 among an ADHD cohort ... 64 Table 8. ADHD case and control demographic and clinical characteristics ... 73 Table 9. Exposure to psychostimulant agents and risk of suicide attempt ... 76 Table 10. Alternative psychostimulant exposure definitions and risk of suicide
attempt ... 77 Table 11. Exposure to psychostimulant agents and risk of suicide attempt – stratified
xi
LIST OF FIGURES
FIGURE PAGE
Figure 1. Conceptual model of factors influencing youth suicide ... 3
Figure 2. Potential causal pathway ... 9
Figure 3. Triangulation approach ... 11
Figure 4. Average rate of US suicide (from 1999-2010) by age ... 13
Figure 5. Average annual rate of suicide ... 14
Figure 6. Flow Diagram of Adverse Drug Event Reporting Sequence ... 21
Figure 7. NAMCS and NHAMCS survey design (adapted from Budnitz, et al.) ... 24
Figure 8. Defining rolling cohort of new ADHD diagnoses ... 31
Figure 9. Aim 3 study design ... 38
Figure 10. Psychostimulant alternative exposure group definitions ... 43
Figure 11. SRE reporting rate for individual psychostimulant agents by year in the US ... 48
Figure 12. SRE reporting rate for each drug class by year in the US ... 50
Figure 13. Aggregated SRE reporting rate by psychostimulant, SSRI, and alpha- adrenergic drug from 1998 – 2010 in the US ... 52
Figure 14. Empirical bayes rate multiplier estimates and confidence intervals by drug: psychostimulant agents and selective serotonin reuptake inhibitors model ... 54
Figure 15. Empirical bayes rate multiplier estimates and confidence intervals by drug: psychostimulant agents and alpha-adrenergic agonists model ... 55
Figure 16. Annual ADHD incidence and prevalence rates, 2001-2011 ... 56
Figure 17. Annual ADHD incidence rates by gender, 2001-2011 ... 57
Figure 18. Annual ADHD prevalence rates by gender, 2001-2011 ... 58
Figure 19. Annual ADHD incidence rates by age group, 2001-2011 ... 59
xii
LIST OF FIGURES (Continued)
FIGURE PAGE
Figure 21. Annual ADHD incidence rates by age group and gender in adolescents
and young adults, 2001-2011 ... 61 Figure 22. Annual ADHD incidence rates by age group and gender in adults and
seniors, 2001-2011 ... 61 Figure 23. Annual ADHD prevalence rates by age group and gender in adolescents
and young adults, 2001-2011 ... 62 Figure 24. Annual ADHD prevalence rates by age group and gender in adults and
seniors, 2001-2011 ... 62 Figure 25. Percent of new ADHD diagnoses by provider type (all ages), 2001-2011 ... 65 Figure 26. Percent of new ADHD diagnoses by provider type (children and
adolescents), 2001-2011 ... 66 Figure 27. Percent of filled psychostimulant prescriptions by type within 30 days of
index diagnosis (all ages), 2001-2011 ... 67 Figure 28. Percent of filled psychostimulant prescriptions by type within 30 days of
index diagnosis (children and adolescents), 2001-2011 ... 67 Figure 29. Percent of psychostimulant use within 30 days of index diagnosis by
prescriber type (all ages), 2001-2011 ... 68 Figure 30. Percent of psychostimulant use within 30 days of index diagnosis by
prescriber type (children and adolescents), 2001-2011 ... 69 Figure 31. Monthly percent psychostimulant medication possession and suicide
attempts in ADHD prevalent cohort (all ages), 2001-2011 ... 70 Figure 32. Monthly percent psychostimulant medication possession and suicide
attempts in ADHD prevalent cohort (children and adolescents), 2001-2011 ... 70 Figure 33. Aim three study population ... 72
xiii
LIST OF ABBREVIATIONS ADHD Attention-Deficit Hyperactivity Disorder ADR Adverse Drug Reactions
ARIMA Autoregressive Integrated Moving Average
AE Adverse Event
CD Continuous Dose
CI Confidence Interval
CPT-4 Current Procedural Terminology
DSM-5 Diagnostic and Statistical Manual of Mental Diseases Fifth Edition DSM-IV Diagnostic and Statistical Manual of Mental Diseases Fourth Edition
EB Empirical Bayes
ED Emergency Department
ER Extended Release
FAERS FDA Adverse Events Reporting System FDA Food and Drug Administration
HCPCS Healthcare Common Procedure Coding System
HIV/AIDS Human Immunodeficiency Virus/Acquired Immune Disease
ICD-9-CM International Classification of Diseases Ninth Revision Clinical Modification
IR Immediate Release
LA Long Acting
MedDRA Medical Dictionary for Regulatory Activities MPR Medication Possession Ratio
NAMCS National Ambulatory Medical Care Survey NCHS National Center for Health Statistics NDC National Drug Codes
NHAMCS National Hospital Ambulatory Medical Care Survey
OR Odds Ratio
RCT Randomized Clinical Trial
RX Prescription
SD Standard Deviation
SR Sustained Release SRE Suicide Related Event
SSRI Selective Serotonin Reuptake Inhibitor
US United States
1 CHAPTER I
INTRODUCTION
1.1 Attention-Deficit Hyperactivity Disorder (ADHD)
Attention-deficit hyperactivity disorder (ADHD) is a common developmental disorder that can afflict all age groups including children, adolescents, and adults but must have a childhood onset before age 7.[1, 2] The Diagnostic and Statistical Manual of Mental Diseases Fourth Edition (DSM-IV) outlined criteria for diagnosis of ADHD requiring six or more symptoms of inattention and/or hyperactivity/impulsivity for at least six months to a point that is inappropriate and/or disruptive for the developmental level.[3] Based on the DSM-IV ADHD criteria, three types of ADHD were defined: predominantly inattentive type, predominantly hyperactive-impulsive type, and a combined type. In May 2013, the Diagnostic and Statistical Manual of Mental Diseases Fifth Edition (DSM-5) replaced the DSM-IV ADHD diagnostic criteria, but the fundamental principle for the diagnosis of ADHD remained the same.[3-5] The DSM-5 continues to utilize the same 18 symptoms requiring at least six from either or both symptom domains (inattention group of criteria and/or the
hyperactivity/impulsivity criteria), while older adolescents and adults (over age 17 years) must present with five.[4]
ADHD is of unknown etiology and is a descriptive condition without diagnostic biologic markers making diagnosis and treatment reliant on symptom identification including inattention, over-activity, and impulsivity.[1, 3] These hallmark symptoms are often present in children as early as four years of age.[1] There have been varying estimates of the prevalence of ADHD, but most recently it has been estimated that 5 percent of children worldwide suffer from ADHD.[1, 6, 7] In the United States, the percentage of children aged 4 to 17 years with ADHD increased from 7.8% during 2003 to 9.5% in 2007.[8] Further, boys and children ages 5 to 12 years of age are diagnosed with ADHD twice as frequently as girls and adolescents aged 13 to 18 years of age.[7] The severity level of
2
ADHD symptoms decreases with age; however, the majority of children with ADHD continue to show impairment throughout adolescence and into adulthood.[7] Prevalence estimates of ADHD among adults worldwide is approximately 2.5 percent.[9] The prevalence estimates for all ages is expected to increase following the DSM-5 update, specifically in adolescents and adults.[5]
ADHD is associated with significant morbidity and financial burden. Children and
adolescents diagnosed with ADHD often experience problems with academic achievement, social relationships, life skills, and risk of injury.[1, 10, 11] ADHD is associated with multiple comorbidities including oppositional defiant disorder, conduct disorder, learning disabilities.Substance abuse is also more common in AHDH but it is unclear if it is ADHD or the primary comorbidities associated with ADHD. [12, 13] Children with ADHD also have higher annual medical costs than children without ADHD. [14] The economic burden associated with ADHD in children and adolescents, including direct and indirect costs, are estimated to be approximately 42.5 billion (in 2005 dollars) per year. [14-21]
A positive relationship between ADHD and suicide has been described.[22-31] It is thought that ADHD symptoms occur more frequently in suicidal populations and may be a reason for more completed suicide.[23, 26, 27, 32] Some studies have previously suggested that the symptoms and/or the comorbid mental health disorders of ADHD may be reasons for an increased risk of suicide in this disease.[28, 32] However, the exact mechanism is still unclear as to why individuals with ADHD are more likely to commit suicide.[26] Investigators have indicated that additional studies are needed to properly understand this risk specifically in younger populations without comorbidities as this relationship is possible medicated by the comorbid conditions. [22, 24, 26] Moreover, given that the comorbidities associated with ADHD are also associated with an increased risk of suicide, it is likely there are multiple factors attributing to the increased risk of suicide within ADHD. A previously published and notable conceptual model is shown in Figure 1 highlighting common factors that may play a role in children and adolescent suicide.[33, 34]
3
Figure 1. Conceptual model of factors influencing youth suicide.[33, 34] 1.2 ADHD Treatment
As with most mental health disorders, the goal of treatment for ADHD is to decrease symptoms, enhance functionality, and improve the i di idual s well-being.[13] Treatment of ADHD includes non-pharmacologic interventions, pharmacologic treatment, or a combination of the two.[1, 35]
Non-pharmacologic treatment often includes psychosocial support, counseling, behavioral training, and cognitive behavioral therapy.[1, 35] Pharmacologic treatment includes psychostimulant medications, such as methylphenidate, dextroamphetamine, and mixed amphetamine salts in extended release (ER) and immediate release (IR) formulations (Table 1).[1, 6]
Both pharmacologic treatment and non-pharmacologic treatment are utilized in the management of ADHD symptoms however the National Institute of Mental Health Multimodal treatment study demonstrated that pharmacotherapy is substantially more effective than
behavioral therapy.[1, 36] First line pharmacologic therapy for ADHD includes psychostimulants. These agents have been shown effective in improving the core symptoms of ADHD and may also improve aggression, oppositionality, internalizing symptoms, and peer relations.[37-40]
4
Table 1. Psychostimulant agents approved by the Food and Drug Administration (FDA) to treat ADHD.
Generic Name Brand Name(s) (FDA Approval Date)
FDA Approved Age(s) FDA Approved Indication(s) Methylphenidate Aptensio XR® (4/17/2015) Adults and children 6 years
and older ADHD Concerta® (XR) (08/01/2000) Daytrana® (04/06/2006) Metadate ER® (06/01/1988) Metadate CD® (04/03/2001) Quillivant XR® (09/27/2012) Methylin® (12/19/2002) Narcolepsy (except Aptensio XR®, Concerta®, Metadate CD®, Ritalin LA®, Quillivant XR® and Daytrana®) Methylin ER® (05/09/2000) Ritalin® (12/05/1955) Ritalin LA® (06/05/2002) Ritalin SR® (03/30/1982)
Dexmethylphenidate Focalin® (11/13/2001) Adults and children 6 years and older
ADHD Focalin XR® (05/26/2005)
Methamphetamine Desoxyn® (12/31/1943) Adults and children 6 years and older
ADHD Adults and children 12
years and older
Obesity Mixed Amphetamine
Salts: Amphetamine/ Dextroamphetamine
Adderall® (01/19/1960) Adults and children 3 years and older
ADHD Adults and children 6 years
and older
Narcolepsy (except XR formulation) Adderall XR® (10/11/2001) Adults and children 6 years
and older
ADHD Dextroamphetamine Dextrostat® (IR)*
(05/29/1975)
Adults and children 3 years and older
ADHD Adults and children 6 years
and older
Narcolepsy Dexedrine® (ER) (02/26/1976) Adults and children 6 years
and older
ADHD Narcolepsy Lisdexamfetamine Vyvanse® (02/23/2007) Adults and children 6 years
and older
ADHD Amphetamine Dyanavel XR (10/21/2015) Children 6 years and older ADHD CD = continuous dose; ER = extended release; LA = long acting; IR = immediate release; SR = sustained release; XR = extended release
5
The use of other psychotropic medications in children and adolescents with ADHD is becoming more common.[41, 42] These other pharmacologic treatments for ADHD include alpha-adrenergic agonist agents (e.g., clonidine and guanfacine) as well as norepinephrine reuptake inhibitor (e.g., atomoxetine) as shown in Table 2. Other less common pharmacologic treatment options include antidepressants, such as bupropion, and antipsychotics; however, these are not the mainstay of therapy and are typically reserved for individuals who respond poorly to other
pharmacologic agents as well as those with comorbid mental health diagnoses requiring additional therapy. Generally, psychostimulant agents are considered first-line in the pharmacologic treatment of ADHD.[1, 6, 35, 43]
Table 2. Non-stimulant agents approved by the FDA.
Generic Name Brand Name(s)
(FDA approval Date)
FDA Approved Age(s) FDA Approved Indication Clonidine Kapvay® (ER) (09/29/2009) Children and adolescents 6
to 17 years of age
ADHD Catapres® (IR) (9/03/1974) Adults and children 12
years and older
Hypertension Guanfacine Intuniv® (ER) (09/02/2009) Children and adolescents 6
to 17 years of age
ADHD Tenex® (IR) (10/27/1986) Adults and children 12
years and older
Hypertension Atomoxetine Strattera® (11/26/2002) Children and adolescents 6
to 18 years of age
ADHD
ER = extended release; IR = immediate release
Due to the increasing prevalence of ADHD as well as the continuation of ADHD into
adulthood, psychostimulant medication utilization is on the rise.[44-46] Utilization will likely continue to increase given the recent 2012 update of the United States (US) pediatric ADHD guidelines that included an expansion of the age range recommended for treatment of ADHD from children 6 to 12 to 4 through 18 years of age.[1] Psychostimulant utilization varies considerably by gender, age, race/ethnicity, and geographical area, with a rate threefold higher in boys than in girls, and higher among ages 6 to 12 years, whites and individuals in the northeastern region of the US.[43-45, 47]
6
1.3 Psychostimulant Warnings And Drug Label Revisions
Psychostimulant agents have been on the market for many years; however, only recently FDA post-marketing safety concerns have been raised. The FDA has identified a possible safety concern regarding methylphenidate agents and psychiatric adverse events. [48] Post-marketing adverse event reports received by the FDA regarding methylphenidate products included psychiatric events such as visual hallucinations, suicidal ideation, psychotic behavior, as well as aggression or violent behavior. This issue was evaluated at the March 22, 2006 Pediatric Advisory Committee meeting.[49] The committee concluded from re-examination of clinical trial data that there was a statistically significant increase in psychosis or mania among patients taking methylphenidate products compared with patients taking placebo, and that some hallucinations appeared to be drug-related. The committee recommended labeling changes to include quantitative information on rates from clinical trial data and a description of possible visual or tactile hallucinations. Further, it was determined that clinical trial data suggested an increased frequency of aggression events relative to placebo for some drugs (e.g., Daytrana, Ritalin LA, and Strattera) but not others; thus, the
committee suggested labeling changes to highlight that aggression can be a feature of ADHD and information should be provided to inform patients about the risks of an increase in aggression. Last, the committee concluded that suicide related events defined as suicidal behavior (i.e., suicidal actions, attempts and/or completions) and suicidal ideation (i.e., passive and active) for approved ADHD products were not significantly increased in randomized clinical trial (RCT) analysis other than for atomoxetine; thus, no labeling changes were recommended by the committee to address suicide related events at that time.[49]
The FDA has conducted additional small-scale analyses utilizing the FDA Adverse Event Reporting System (FAERS).[48-50] However, these analyses have been on subsets of psychostimulant
7
agents, over small periods of time, and have used numerator-based methods (i.e., a variant of the empirical Bayes screening) that do not account for the population at risk.[51]
Currently psychostimulant medication package inserts warn that the drugs can exacerbate underlying depression, treatment-emergent psychotic or manic symptoms, as well as aggressive behavior or hostility; however, they do not warn against suicide related events.
1.4 Study Purpose
Attention-deficit hyperactivity disorder is a common condition among the pediatric population with over 5 million children and adolescents reported to have the disease in 2007.[52] Psychostimulant medication utilization in children and adolescents has doubled over the previous decade.[53] Large cohort studies have been conducted to investigate the cardiovascular risks of psychostimulant agents;[54-56] however, no previous studies have evaluated the potential association between psychostimulant agents and suicide related events including suicidal ideation and behavior. Suicide is the tenth leading cause of death among the pediatric population and rates continue to rise.[57] Given the increased use of psychostimulants with little knowledge of the affect they can have on suicide risk, additional information is needed.
The FDAs de isio not to add a warning for psychostimulants and suicide related events was based on clinical trial data showing no association between suicide related events and
psychostimulant agents.[49] However, suicide related events are considered rare adverse events (AE), and often clinical trials have multiple limitations in detecting rare AEs, including the likelihood of excluding those patients at highest risk of the AE (e.g., suicidal ideation and/or behavior), small sample sizes, short follow-up time, and ascertainment biases that are associated with a focus on spontaneously reported AE.[51, 58] Thus, additional studies with alternate study designs are warranted to investigate the potential association between psychostimulant agents and suicide related events.
8
Over the past decade there has been an increasing awareness for the potential of prescription medications to increase the risk of suicide related events, including agents such as antidepressants (most notably selective serotonin reuptake inhibitors), leukotriene modifying agents, isotretinoin, atomoxetine, and antiepileptics.[59-61] These agents have received black-box warnings for suicide related events based on evidence from meta-analyses of clinical trial data, observational studies, and early signal detection utilizing FAERS. Often these warnings have been controversial and some would argue that they were made with limited information and
prematurely. However, the FDA concluded a potential risk may occur with these agents indicating the biologic plausibility that pharmacologic treatment could increase the risk for suicide related events.
The effect ps hosti ula ts a ha e o a i di idual s sui ide risk is ot k o . Given the biologic plausibility that pharmacologic treatment could increase the risk of suicide related events and the lack of observational studies investigating this association in ADHD, additional information is needed on the relationship between psychostimulant medications and suicide related events (Figure 2).
9 Figure 2. Potential causal pathway.
1.5 Specific Aims
This set of studies aimed to determine if an association exists between psychostimulant agents and suicide related events. The most frequent indication for psychostimulant use is treatment of ADHD; therefore, the focus of this study will be on the relationship between psychostimulant medications and suicide related events, in an ADHD population. Although
premarketing clinical trials are required for all new drugs before they are available to the public, rare AEs are often not detected until data are available in a larger population.[51, 58] This study utilized a triangulation and progression of evidence approach using three different methodologies that frequently comprise the mainstay of drug safety research (Figure 3). The objective in utilizing the different methodologies was to provide further granularity with each subsequent approach as well as to provide a better estimation of the association between psychostimulant medications and suicide related events.
10 1.5.1 Study Hypothesis
Three different studies/approaches were conducted to test the following hypothesis: psychostimulant medication use is associated with an increased risk of suicide related events. The null hypothesis was that there is no association between psychostimulants and risk of suicide related events.
1.5.2 Specific Aim 1
Evaluate the ecologic association between psychostimulant agents and spontaneously reported suicidal ideation and behavior using FDA Adverse Event Reporting System (FAERS) data among all ages: an ecologic study.
1.5.3 Specific Aim 2
Describe the trends in ADHD diagnosis, treatment characteristics, and suicide attempt over time in all ages using aggregated, person-level medical claims data: a time-series analysis.
1.5.4 Specific Aim 3
Evaluate the association between dispensed psychostimulant agents and suicide attempt in children and adolescents with ADHD using medical claims data: a retrospective, observational, nested case-control study.
Figure 3. Triangulation approach.
12 1.6 Study Summary
In the first aim we evaluated whether a relationship exists between suicidal ideation and behavior and psychostimulants utilizing spontaneously reported FAERS data. The FDA database, FAERS, utilizes five different categories for describing suicidal ideation and behavior including completed suicide, suicide attempt, suicidal behavior, suicidal ideation, and depression suicidal. A random-effects Poisson regression analysis developed by Gibbons et al. was utilized to model the association between psychostimulant agents and suicidal ideation and behavior.[51, 62] This method uses a denominator approach that includes an estimate of the population at risk for the particular AE. The FAERS data is the most common source used by the FDA in identifying AE signals in order to determine if medication warnings or labeling changes need to be made. Although this is the same data being utilized, most often the FDA uses numerator only methodological analyses.[51] In this study we used a proxy denominator in order to understand the population at risk. In the second aim we described trends in ADHD diagnosis and treatment characteristics as well as suicide attempt rates among all ages using aggregated, person-level medical claims data. The third aim was a patient-level nested case-control study using medical claims data with suicide attempts as the outcome of interest. Medical claims data often provide a large enough sample to evaluate rare AEs and a case-control study is more generalizable compared to RCTs because individuals at greater risk do not need to be excluded from the study design.[51, 63, 64]
13 CHAPTER II
REVIEW OF THE LITERATURE
2.1 Suicide
Suicide is a significant global public health issue. In 2009, suicide was ranked as the 10th leading cause of death among persons ages 10 years and older, accounting for 36,891 deaths.[57, 65] Suicide is the second leading cause of death among 25 to 34 year olds and the third leading cause of death among 15 to 24 year olds.[57] Among young adults ages 15 to 24 years old, there are
approximately 100 to 200 attempts for every completed suicide. [66]
Males are at nearly 4 times greater risk for completed suicide compared to females; however, attempt rates are similar between genders.[57, 66, 67] In the US, suicide completion rates also differ between age groups with a higher rate in middle age groups (45-49 and 50-54 age groups) and the elderly ears as shown in Figure 4. [57] US suicide rates have historically remained stable over time, while rates over the previous decade indicate a slight and gradual increase, as shown in Figure 5.[57]
Figure 4. Average rate of US suicide (from 1999-2010) by age. 0 10 20 30 40 50 60 05-09 10-14 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85+ R ate p e r 100, 000 p o p u lati o n Age
Suicide Rate by Age
14 Figure 5. Average annual rate of suicide.
While suicide has multiple risk factors, in the US the most important is psychiatric disorders. It has been estimated that over 90% of all suicides in the US are associated with mental illness.[66] Most commonly, suicide occurs in individuals with affective disorders, substance abuse, and schizophrenia.[24, 66, 68-71] Studies have also reported a higher suicide rate in individuals with depression,anxiety and panic disorders, personality disorders, ADHD, eating disorders, body dysmorphic disorder, and bipolar disorder.[22, 26, 66, 68-80] An increased risk of suicide has also been reported among individuals with non-mental health diseases including dementia, malignant neoplasm/cancer, human immunodeficiency virus/acquired immune disease (HIV/AIDS), epilepsy, irritable bowel disease, migraine headaches, multiple sclerosis, renal disease, chronic pain, and asthma.[66, 73, 74, 81-85]
2.2 Defining Suicide Related Event Nomenclature
Suicidal behavior, thoughts, and action were previously referred to as suicidality. However, suicidality is a broad term and often does not appropriately indicate the occurrence. Recently the FDA has urged more specific terminology for the suicide related events. Suicidality could be further defined as any of the following: suicide completion, suicide attempt, suicidal behavior, and suicidal ideation. Suicidal ideation can further be considered passive, which is also sometimes referred to as
0 2 4 6 8 10 12 14 16 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 R ate p e r 100, 000 p o p u lati o n Year
Suicide Rate by Year
15
depression suicidal, or active. However, active suicidal ideation is further classified based on if a method, intent, and/or plan has been formulated. Suicidal behavior can include completed suicide, suicide attempt, interrupted attempt, aborted attempt, and preparatory actions toward imminent suicidal behavior. Overall, the most well-known and accepted terminology includes completed suicide, attempted suicide, suicidal behavior (including interrupted attempt, aborted attempt, and preparatory actions), and suicidal ideation (passive and active).[86]
2.3 Psychostimulant Mechanistic Plausibility
There are multiple psychostimulant medications on the market (as shown in Table 1), but all of the psychostimulant agents act via similar mechanism of action and could broadly be
differentiated into two categories: amphetamines and methylphenidate products. There are also combinations of agents, pro-drugs that require conversion to the active formulation, and
formulations with longer half-lives resulting in a longer duration of action. All of these have similar sites and mechanistic effects within the respective category.
Amphetamines are non-catecholamine sympathomimetic agents that stimulate the release of norepinephrine and other biologic amines from central adrenergic receptors. At higher dosages, they cause release of dopamine from the mesocorticolimbic system and the nigrostriatal dopamine systems. It is thought that the release of dopamine is responsible for the reinforcing properties of amphetamine. At still higher doses, amphetamine stimulates the release of serotonin. Finally, amphetamine may act as a direct agonist on central serotonin receptors. Thus, amphetamine is both a direct and an indirect stimulant. Amphetamines may also inhibit monoamine oxidase, but this is a minor action. However, there is no conclusive evidence for the mechanism(s) of action of
amphetamines on the mental and behavioral characteristics of ADHD. Current research suggests that the modulation of serotonergic pathways by the amphetamines may contribute to the calming effects in the treatment of this disorder.[87]
16
Conversely, methylphenidate exerts many of its effects through dopamine uptake blockade of central adrenergic neurons, in contrast to the amphetamines that increase catecholamine release as a primary mechanism. Specifically, methylphenidate blocks dopamine transport of carrier
proteins. As a result, sympathomimetic activity in the central nervous system is increased. There is some evidence that the alteration of dopamine transport systems by methylphenidate may indirectly augment the action of serotonin, but further pharmacologic research is needed to understand these processes. The main sites of activity appear to be the brain stem arousal system and the cerebral cortex, including the subcortical structures of the thalamus.[87]
It has been suggested that biological changes in the brain are associated with completed and attempted suicide. Abnormal functioning of the hypothalamic-pituitary-adrenal axis, a major component of adaptation to stress, has been shown to be a predictor of future completed suicides, but not consistently for suicide attempts. Also, it has been shown that serotonergic function is reduced and noradrenergic function is altered in the central nervous systems of both suicide attempters and completers.[66]
Based on the mechanisms of the psychostimulants and their potential effect, there may be some relationship between the pharmacologic mechanisms and the evidence for altered
serotonergic and noradrenergic function and suicidal ideation and behavior. This is only a theoretical process with no definitive studies to indicate a pathway in suicide related events or direct role in ADHD.
2.4 Evidence of Psychostimulants-Suicide Association
Currently there are very few studies evaluating the relationship between psychostimulants and suicide. Case reports have been published documenting methylphenidate poisonings with high dose consumption of the prescription agent with suicidal intent.[88-90] Studies with more
17
Premarketing clinical trials were required for the FDA approval of psychostimulant agents, but clinical trials often do not capture rare AEs such as suicide.[51] Many of these trials were short in duration and were conducted before suicidal ideation and behavior became a heightened area of focus in clinical trial research. Future FDA guidance for measuring suicidal ideation and behavior in clinical trials will be focused on actively querying patients about the occurrence of suicidal thinking and behavior, the current standard of practice is to rely on patients to report such occurrences spontaneously, followed by retrospective classification of events into appropriate categories.[86]
A clinical trial conducted by Hoare et al.[92] investigating methylphenidate ER in children and adolescents with ADHD detected two cases of suicide related events. In both cases, the event was classified as depression-suicidal thoughts. Although this is an infrequent result, this is an instance where a clinical trial captured the rare AE of suicidal ideation in conjunction with psychostimulant agents.
A meta-analysis of atomoxetine and methylphenidate comparator clinical trials examined suicide related events associated with ADHD treatment.[93] The meta-analysis included seven comparator trials among which 5 suicide related events were detected (none of which were suicide attempts or completions). Of the five suicide related events, three were atomoxetine related (3/559) and two methylphenidate related (2/465). The Mantel-Haenszel risk ratio was 0.52 (95% CI; 0.06, 4.54), concluding no risk difference between these agents.
In addition to case reports, meta-analyses and clinical trials, a previous retrospective cohort study investigated psychostimulants and atomoxetine to determine if an association exists between these and sudden death.[56] From a cohort with 18,637 patient-years, seven patients died, none of which were deemed to be cases of sudden cardiac related death. However, an increased risk of suicide was seen in the ADHD cohort in comparison with the general population, although the investigators could not conclude whether the medications contributed to this.[56]
18
A study conducted by Tobaiqy et al.[94] of adverse drug reactions (ADR) associated with ADHD medications in children found that parents reported common side effects such as decreased appetite as well as rare ADRs including suicidal ideation. However, no conclusion could be made about a potential association between ADHD medications and suicide related events.
Marshall et al.[95] conducted a prospective cohort study of 1800 drug users looking
specifically at illicit intravenous methamphetamine abuse and found an increased risk of attempted suicide. Subjects utilizing illicit methamphetamine were at an 80% increased risk of attempting suicide over the seven year period in comparison to those who injected other illegal substances, including heroin or cocaine. Subjects who injected more often were also at greatest risk. These findings suggest that intravenous methamphetamine drug abusers are at an increased risk for suicidal related events. However, it is important to note that suicide is more common in individuals with substance abuse.[75-77] This study does provide some question in regards to a potential biologic mechanism with psychostimulants, but cannot be generalizable to those with clinically indicated psychostimulant use as our study aims to investigate given this study was specific to illicit drug users only.
A recent study by Ruggiero et al.[96] aimed to increase the number of ADRs recorded in the Italian spontaneous reporting database and collect more thorough data on a subset of individuals using a systematic adverse event checklist for both atomoxetine and methylphenidate in the pediatric setting. Most of the ADRs reported were common and defined as not serious events. There were several uncommon ADRs identified by the intensive monitoring program including hepatomegaly, suicidal ideation, weight gain, and drug interactions. This study reinforced the fact that under reporting of ADRs is an issue. This study also supports the safety of these medications while acknowledging rare ADR can occur with atomoxetine and methylphenidate.
19
Based on the current scientific literature, it is evident there is a lack of data confirming or refuting the association between psychostimulants and suicide related events. While there is some evidence, it is severely limited in scope and a definitive association cannot be determined.
2.5 Implications of this Study
No previous studies have directly aimed to evaluate the potential association between psychostimulant medications and suicide related events as the primary outcome in therapeutic users. Over the past decade there has been an increasing awareness of the potential for prescription medications to increase the risk of suicide related events, including antidepressants, leukotriene modifying agents, atomoxetine, and antiepileptics.[59-61] Investigators have noted that there is no difference in risk for suicide related events between atomoxetine and methylphenidate; however, the warnings issued by the FDA are inconsistent between these agents.[93, 97] Given the lack of data to support or refute the association between psychostimulants and suicide related events,
additional, more definitive studies are warranted. Therefore, it was the objective of this dissertation to determine if a potential association exists between psychostimulants and suicide related events.
20 CHAPTER III
METHODS
3.1 Study Overview
To examine the association between psychostimulant use and suicide, this investigation utilized a triangulation and progression of evidence approach - with three different studies each using different methodologies that comprise the mainstay of drug safety research. The objective in utilizing the different methodologies was to provide further granularity with each subsequent analysis. There are a variety of approaches for studying AEs, ranging from analyses of spontaneous reports to ecological studies to medical claims database studies; each has its strengths and
weaknesses. We aimed to better understand the relationship being studied by combining these methods while keeping in mind the limitations of each.
3.2 Specific Aim 1
For this aim we conducted an ecologic study of the association between psychostimulants and spontaneously reported suicidal ideation and behavior with the objective of identifying and comparing the rate of reported suicide related events based on adverse event reports and a denominator of prescribed prescriptions.[62, 98]
3.2.1 Data Source for Suicide Related Events
To obtain the numerator for our analyses (counts of AEs of interest), suicide-related events reported to the FDA Adverse Event Reporting System (FAERS) between January 1, 1998 through December 31, 2010 (1998 being the earliest data available and 2010 the most recent with
corresponding prescription data available) were identified. A FAERS report submission can be made by any of the following: patient, caregiver, manufacturer, healthcare provider, or hospital and the case submission is assigned an individual safety report identification number as well as a case number (Figure 6). In instances where follow-up reports are submitted, these have the same case
21
number as the initial report but a separate individual safety report number. Each report lists the patie t s rea tio s e.g. sui ide atte pt, sui ide o pletio , out o e s e.g. hospitalized, death , primary suspected drug(s), secondary suspected drug(s), concomitant drug(s), as well as patient demographic and clinical characteristics. The demographic and clinical characteristics are not
o siste tl reported i FAE‘S so e restri ted a al ses to patie t s rea tio a d the pri ar suspected drug(s). We only included cases where the drug of interest was listed as the primary suspected drug to minimize potential reporting bias given all case reports submitted to the FAERS database are self-reported without any validation process to confirm the event and true causative agent.
Figure 6. Flow Diagram of Adverse Drug Event Reporting Sequence. 3.2.1.1 FAERS case selection
Reports were included if the generic or brand name for methylphenidate, dexmethylphenidate, dextroamphetamine, dextroamphetamine and amphetamine, or lisdexamfetamine was listed as the primary suspected drug, and the report listed one of the
following Medical Dictionary for Regulatory Activities (MedDRA)[99] reaction terms utilized by FAERS: completed suicide, suicide attempt, suicidal behaviour, suicidal ideation, and depression suicidal. Methamphetamine cases were excluded from this analysis because this drug is infrequently prescribed for the treatment of ADHD and we are unable to differentiate whether the FAERS cases were associated with prescription methamphetamine versus illegal methamphetamine
22
Due to the nature of this passive reporting system, multiple reports can be submitted for one event and data can be incomplete. To account for this limitation of the data, we attempted to eliminate duplicate reports through the case selection process if it was evident based on the submission information. If more than one report existed, we used the most complete and recent follow-up report of the same event. Given that multiple reactions could be submitted for a single event, for reports with more than one suicide-related outcome, we used the higher order term as follows: completed suicide > suicide attempt > suicide behavior > suicidal ideation > depression suicidal.
3.2.2 Data Sources for National Psychostimulant Prescriptions
To obtain the denominator for our analyses (i.e. population at risk), we used estimates of the national prescription rates of psychostimulant agents (i.e., methylphenidate,
dexmethylphenidate, dextroamphetamine, dextroamphetamine and amphetamine, or
lisdexamfetamine) utilizing data from the US National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) for 1998 through 2010 (most recent data available). US NAMCS and US NHAMCS are national, cross-sectional, annual surveys with a national probability sample design, conducted by the US National Center for Health Statistics (NCHS).
The NAMCS collects data each year from approximately 3000 randomly selected office-based physicians for a randomly selected 1- eek period o of their patie ts isits, resulti g i data for approximately 25,000 patient visits each year (Figure 7). The randomly selected physicians are asked to fill out a standardized form based on medical records for each patient visit including information on patient demographics and visit information such as chief complaint,
diagnostic/screening tests, rendered diagnoses, and medications prescribed or provided during the visit.
23
The NHAMCS collects data each year from approximately 600 nationally representative emergency and outpatient department hospitals for a randomly selected 4-week period for
approximately 70,000 patient visits a year (Figure 7). The NHAMCS collects information on the visit, patient, and hospital. The patient visit data form that is completed for each patient visit is based upon the medical record and includes the following information: patient demographics, visit information including chief complaint, diagnostic/screening tests, rendered diagnoses, and
medications prescribed or provided during the visit, as well as hospital information including type of clinic, hospital ownership, use of electronic medical records, and medical school affiliation.
Both surveys report data on medications prescribed and/or provided during the visit. The NAMCS and NHAMCS surveys coded drug data as (1) directly written on the survey and (2) to reflect the drug s ge eri o po e ts up u til . I , this lassifi atio s ste for drugs as replaced by a system that coded drugs according to the generic components and therapeutic classes with a Multum drug identification code. Due to the change in methods across survey years of interest, we utilized coding methodology developed by NCHS that matched medication codes from previous years of NAMCS and NHAMCS data with codes from the current ambulatory care drug database, and then dropped the old drug characteristics associated with that medication code, while replacing them with the new Multum characteristics. Each NAMCS/NHAMCS record contained a single weight known as the patient visit weight that is used for both visits and drug/procedures. We utilized this patient visit weight to estimate the national prescribing rates for the drugs of interest. Because the NAMCS and NHAMCS surveys use a multistage stratified probability sampling design and included the patient visit weight that can be utilized for drugs, the derived NAMCS/NHAMCS drug prescribing estimates for the denominator in the analysis are generalizable to the US.
24
Figure 7. NAMCS and NHAMCS survey design (adapted from Budnitz, et al.)[100] 3.2.3 Drug Comparators
Because there are many limitations with the FAERS data, we modeled the psychostimulant AEs with a positive and negative comparator group. The comparator groups provide a point of reference to better interpret the meaning of the AEs rate for psychostimulants and help account for potential reporting bias, potential changes in reporting over time and the inability to validate the reaction or causative agent reported. We examined reports of suicidal related events from the FAERS database involving selective serotonin reuptake inhibitor (SSRIs) (i.e., citalopram,
escitalopram, fluoxetine, fluvoxamine, paroxetine, and sertraline) and alpha-adrenergic agents (i.e., clonidine and guanfacine, immediate and extended release formulations). SSRIs served as the positive control since this class of antidepressants has previously received FDA warnings regarding an increased risk of suicide related events and this is a frequently reported drug class with the aforementioned reactions of interest in FAERS.[98, 101] The alpha-adrenergic agents served as the negative control since these agents have not been previously shown to be associated with suicide
25
related events. The alpha-adrenergic agents can also be used for the treatment of ADHD, so any underlying risk for suicide related events in the ADHD population is expected to be similar. For both control groups, the case selection was collected from the FAERS data utilizing the same case
selection process as described previously for the psychostimulants. The population at risk (denominator) for these comparator drugs was obtained using the annual, national prescribing estimates from NAMCS and NHAMCS.
3.2.4 Statistical Analysis
A random-effects Poisson regression analysis developed by Gibbons et al. was utilized to model the association between psychostimulant agents (i.e., methylphenidate,
dexmethylphenidate, dextroamphetamine, dextroamphetamine and amphetamine, or
lisdexamfetamine) and suicide related events.[62] This approach has previously been validated by Gibbons et al. and is suitable for this analysis because (1) we are looking at rare, count data; (2) this approach provides a method to compare the rate of an AE simultaneously for all individual drugs within a class of drugs (e.g. psychostimulants) with the overall rate of AE of the class as a whole and/or between alternate drug comparator classes; (3) this method utilizes a denominator approach by including a proxy for the population at risk for a particular AE which minimizes the bias produced by restricting the analysis to only those patients who report the AE. [62]
For this analysis, we first determined the rate of suicide related events per one-hundred thousand prescriptions by year (1998-2010) for each medication class (psychostimulants, SSRIs, alpha-adrenergic agents) and each individual drug. We then determined the Empirical Bayes (EB) rate multiplier estimates (exp(EB)) for each drug utilizing SuperMix version 1.0.[103] The AE rate multipliers describe the difference in suicide related event reporting rates for the drug or class of interest from the overall rate for the comparative drugs/class (i.e., SSRIs and alpha-adrenergic agents)in the analysis.
26
A rate multiplier of 1.0 indicates that the drug/class has the same rate as the average for the other drugs in the analysis (i.e., SSRIs in the first analysis, and alpha adrenergic agents in the second analysis), a multiplier of 2.0 is twice the average rate, and a multiplier of 0.5 is one-half the average rate. Confidence (posterior) intervals that do not include 1.0 provide evidence for either significant protective (below 1.0) or harmful associations (above 1.0) of the drug/class and the AE.
Psychostimulant and SSRIs were modeled and psychostimulants and alpha-adrenergic agonists were modeled separately and statistical inference was made based on the large sample confidence (posterior) intervals for the rate multipliers. All statistical analyses were performed with SAS version 9.3 [102] and SuperMix version 1.0.[103]
3.3 Specific Aim 2
For this aim we first estimated the annual incidence and prevalence rates of ADHD. For the second part of this aim, we examined and described five ADHD related measures of interest for ADHD diagnosis and treatment characteristics as well as psychostimulant possession and suicide attempts among a monthly cohort of incident and prevalent ADHD subjects using time series analysis.
3.3.1 Data Source
The data for aim two was obtained from the PharMetrics Legacy Health Plans Claims Data from IMS Health Incorporated. The PharMetrics Legacy Health Plans Claims Data is a commercial, longitudinal database of integrated adjudicated medical and pharmacy claims from over 90
managed health care plans, representing more than 60 million unique, anonymous covered lives at a given point in the US and is a nationally representative managed care population. The PharMetrics Legacy Health Plans Claims Data includes: patient year of birth, gender, and geographic region (West, Midwest, South, or Northeast); diagnosis codes [international classification of diseases, ninth revision, clinical modification (ICD 9-CM codes)], procedure codes [Current Procedural Terminology
27
(CPT-4) and Healthcare Common Procedure Coding System (HCPCS)]; drug codes [National Drug Codes (NDC)], days supplied, and quantity dispensed for medications dispensed; payer type; provider specialty; service date; service cost; and a monthly health plan enrollment flag. A random 10% sample of all patients from the PharMetrics Legacy Health Plans Claims Data (i.e., PharMetrics Legacy 10% random sample) from January 2001 through December 2011 and the corresponding annual population files of the PharMetrics Legacy Health Plan Claims Data was utilized.
3.3.1.1 Study population for annual rates of incident and prevalent ADHD diagnosis - numerator From the PharMetrics Legacy 10% random sample, we created an annual rolling cohort of incident and prevalent ADHD cases. These incident and prevalent ADHD diagnosis counts were used as the numerator to determine annual ADHD diagnosis rates. Subjects had to meet all three criteria below to enter the incident ADHD cohort while subjects only had to meet one criterion for an index ADHD diagnosis to be included in the prevalent ADHD cohort. After the first year a subject met the incident case definition, the subject was considered a prevalent case.
1. Subject had one of the following definitions for index ADHD diagnosis:
a. Subject had two or more diagnosis codes for ADHD (ICD 9-CM codes 314.0-314.01) at any time in the dataset (first diagnosis code was the index ADHD diagnosis); or b. Subject had one diagnosis code for ADHD and at least one ADHD-related dispensed
prescription [psychostimulant (methylphenidate, dexmethylphenidate,
dextroamphetamine, dextroamphetamine and amphetamine, lisdexamfetamine, or methamphetamine) alpha-adrenergic agent or atomoxetine] da s AFTE‘ the ADHD diagnosis code (diagnosis code was the index diagnosis); or
28
c. Subject had one diagnosis code for ADHD and at least one ADHD-related dispensed pres riptio da s BEFO‘E the ADHD diag osis ode first dispe sed ADHD-related dispensed prescription was the index diagnosis).
2. Subject had at least 3 months (90 days) of continuous benefit enrollment prior to the date of the index ADHD diagnosis (based on one of the above definitions).
3. Subject had a period of at least 3 months (90 days) prior to the index ADHD diagnosis (based on one of the above definitions) where no previous diagnosis for ADHD or ADHD-related medication appeared in the claims history (90 day look back period).
We required a minimum 3 months of continuous eligibility for incident subjects to allow for identification of their incident status with no prior ADHD related diagnoses or prescriptions.
Whereas, we only required one month of enrollment for prevalent subjects since we did not have to account for a look back period to ensure there was no prior ADHD related claims.
3.3.1.2 Study population for annual rates of incident and prevalent ADHD diagnosis - denominator
Next we utilized the PharMetrics Legacy Health Plan general population file to determine the population counts which we utilized as our denominator to estimate the annual ADHD incidence and prevalence rates. For the incident case denominator, subjects were required to have at least 3 months (90 days) of continuous benefit enrollment at any point during the year. For the prevalent case denominator, subjects were required to have at least 1 month of benefit enrollment at any point during the year. Because incident subjects had to have a minimum 3 months of continuous eligibility and prevalent subjects only had to have 1 month of continuous eligibility, we required the denominator populations to meet the same criteria.
29
3.3.1.3 Study population/analytic file for ADHD related measures
Again we utilized the PharMetrics Legacy 10% random sample to create a monthly rolling cohort of new ADHD diagnoses. Subjects remained in the cohort until the end of the period or disenrolled (i.e., became ineligible). Subjects had to meet all three criteria below to enter the incident ADHD cohort. A subject counted as an incident case during the first month in which they met the below criteria and afterwards subjects were counted as a prevalent case in each subsequent month until they were no longer eligible (Figure 8).
1. Subject had one of the following definitions for index ADHD diagnosis:
a. Subject had two or more diagnosis codes for ADHD at any time in the dataset (first diagnosis code was the index ADHD diagnosis); or
b. Subject had one diagnosis code for ADHD and at least one ADHD-related dispensed prescription 90 days AFTER the ADHD diagnosis code (diagnosis code was the index diagnosis); or
c. Subject had one diagnosis code for ADHD and at least one ADHD-related dispensed prescription 90 days BEFORE the ADHD diagnosis code (first dispensed ADHD-related dispensed prescription was the index diagnosis).
2. Subject had 3 months (90 days) of continuous benefit enrollment prior to the date of the index ADHD diagnosis (based on one of the above definitions) and 1 month (30 days) after. 3. Subject had a period of 3 months (90 days) prior to the index ADHD diagnosis (based on one
of the above definitions) where no previous diagnosis for ADHD or ADHD-related medication appeared in the claims history (90 day look back period).
Based on the cohort of new ADHD diagnoses as defined above, we created monthly aggregated measures from the person-level data. As new episodes of ADHD accrued within the cohort over
30
time, measures of interest were aggregated into successive monthly values for incident and prevalent cases.
Figure 8. Defining rolling cohort of new ADHD diagnoses.
32
3.3.2 Annual ADHD Incident and Prevalent Rate Measures
We utilized the counts of incident and prevalent ADHD diagnosis as the numerator from the annual rolling cohort of incident and prevalent ADHD cases and the corresponding population counts as our denominator from the annual PharMetrics Legacy population files. Incident cases that were in the cohort for multiple years were considered a prevalent case after the first year. We evaluated incident and prevalent diagnosis rates by age at index ADHD diagnosis (5-11 years, 12-18 years, 19-24 years, 25-49 years, and 50 -74 years) and gender groups using the respective
numerator and denominator counts. 3.3.3 Monthly ADHD Related Measures
The measures to evaluate psychostimulant treatment characteristics were calculated for each month in the 11-year time period based on the previously defined monthly rolling cohort of new ADHD diagnoses. For the following three measures, we calculated monthly proportions for two age groups (1) all ages and (2) 5 to 18 years of age only based on the incident cases from the ADHD cohort.
1. Type of provider who diagnosed ADHD: We calculated the monthly proportion of
index ADHD diagnoses by each of the following provider types: pediatrician,
non-pediatrician primary care physician (e.g., primary care, family medicine), psychiatrist, and other mental health provider (e.g., psychologist, social worker, therapist).
2. Psychostimulant drug dispensing for ADHD: We calculated the monthly proportion of new ADHD diagnoses for which psychostimulants were filled within 30 days before and after the index diagnosis date.
3. Type of provider prescribing index psychostimulant prescription: We calculated
33
before and after the index diagnosis date that were written by the following
prescriber types: pediatrician, non-pediatrician primary care physician (e.g., primary
care, family medicine), psychiatrist, and other mental health provider (e.g., psychologist, social worker, therapist).
For the following two measures, we calculated monthly proportions for two age groups (1) all ages and (2) 5 to 18 years of age only based on the prevalent cases from the monthly rolling cohort of new ADHD diagnoses.
4. Suicide attempt in individuals diagnosed with ADHD: We calculated the proportion of subjects within each month that had a suicide attempt (ICD 9-CM codes
E950-E959). Duplicate claims for one event were excluded to avoid double counting one
suicide attempt; however, recurrent suicide attempts were included.
5. Psychostimulant medication coverage/possession: We calculated the proportion of subjects within each month that had a psychostimulant prescription fill that
overlapped with that month. 3.3.4 Statistical Analysis
First we examined annual rates of ADHD using PharMetrics Legacy Health Plans Claims Data. We utilized the counts of incident and prevalent ADHD diagnoses as the numerator from the annual rolling cohort of incident and prevalent ADHD cases and the corresponding population counts as our denominator from the PharMetrics Legacy annual population files. We then estimated the observed annual diagnosis rates (incident and prevalent) over the 11-year period (2001 to 2011) by age and gender groups as defined above. We then plotted these data points to estimate annual incidence and prevalence diagnosis trends.
The second set of analyses used time series methods to examine and describe the five ADHD related measures of interest for ADHD diagnosis and treatment characteristics as well as
34
psychostimulant possession and suicide attempts among the monthly cohort of incident and prevalent ADHD subjects over the 11-year period. We used time series methods as these are appropriate when the data has sequenced observations ordered in time. This method also assumes that errors are correlated thus time series methods assist to identify patterns in the correlated data which can be adjusted for during the modeling. For this analysis we were not evaluating an
interruption over the time period since there was not a direct psychostimulant intervention (i.e., no FDA issued warnings for suicidality) nor were we attempting to forecast trends. Because of this, we focused on the descriptive and explanation phases of time series analysis to identify patterns (i.e., correlation(s) and/or seasonality) in the time series data for each ADHD related measure using an autoregressive integrated moving average (ARIMA) (p, d, q) time series model as well as estimated the overall trend.
We utilized the Box-Jenkins approach for the time series model building process which takes a step wise approach to first identify the model and underlying correlations (i.e., seasonality), estimate model parameters including autoregressive and/or moving average components and then validate the selected model to ensure all of the patterns have been addressed. [115] First we plotted the monthly measures by age groups to examine patterns over the 11-year period among each ADHD related measure of interest. Next we tested for stationarity, the degree to which there is homogeneity for the mean and variance over the observation period, using autocorrelation and partial autocorrelation functions, correlograms (residuals versus time) and the Durbin-Watson test statistic which tested for serial correlations. Next, the autocorrelation, partial autocorrelation, and inverse autocorrelation functions were evaluated to estimate the autoregressive and moving average components (p, q) while only including significant parameters to ensure model parameter appropriateness. To validate the models, we examined the residuals (differences between actual and predicted values) for each model to confirm they were a series of random errors.
35
First-order autocorrelation was detected indicating correlation between the prior month and subsequent month values (i.e., one lag period) for each ADHD related measure. We also detected an additive seasonal pattern repeating at month 12 (i.e., lag 12) for psychostimulant drug dispensing, suicide attempts and psychostimulant medication possession measures. We adjusted for the autocorrelation within each model as well as the additive seasonal pattern repeating at lag 12 for psychostimulant drug dispensing, suicide attempts and psychostimulant medication possession measures. Models were implemented using an ARIMA (p ,d, q) time series model in STATA® version 11.2.[116] The autoregressive component, p, represents the lingering effects of preceding values, the integrated component, d, represents trend(s) in the data, and the moving average component, q, represents the lingering effect of preceding random component(s) of the time series. For the diagnosing provider type and prescribing provider type measures, we adjusted for the first-order autocorrelation with an autoregressive parameter (1, 0, 0) which resulted in the most appropriate model. For the psychostimulant drug dispensing, suicide attempts and psychostimulant medication possession measures, we adjusted for the first-order autocorrelation with an autoregressive parameter and adjusted for the additive seasonal pattern with an autoregressive and moving average parameter as well as seasonal differencing at lag 12 (2, 1, 1). Because this was an additive model it would be described as (2, 1, 1) with differencing at lag 12; however, if the seasonal model was multiplicative a common notation is ARIMA (p,d,q)(P,D,Q)s, where s is the seasonal cycle. 3.4 Specific Aim 3
For this aim we conducted a retrospective, nested case-control study of the association between psychostimulants and suicide attempt in children and adolescents with ADHD. Cases and controls were identified from an ADHD cohort of children and adolescents using a large national administrative claims database as described above in aim two.
