Introduction
Several compounds are now recognized as effective treatments for the major symptoms of attention-deficit/hyperactivity disorder (ADHD) in adulthood. The most effective of these include methylphenidate and dextroamphetamine (or mixed dextro- and levoamphetamine); however, the use of other agents, such as bupropion and desipramine, has also received some support. In addition to these, atomoxetine, a highly selective noradrenergic reuptake inhibitor with little affinity for other neurotransmitter systems [
1], has been shown to be well tolerated and effective in reducing the symptoms of ADHD in adulthood. In fact, the benefits of atomoxetine for adults with ADHD have now been demonstrated in three studies of adult patients [
2,
3], with each report establishing the superiority of atomoxetine over placebo in reducing inattentive, hyperactive, and impulsive symptoms of the illness [
3]. As a result of its demonstrated efficacy and low occurrence of clinically meaningful side effects [
4], atomoxetine recently became the first non-stimulant medication approved for use in the United States for the treatment of ADHD in adults.
Thus, the ability of atomoxetine to reduce symptoms of ADHD among adults has been sufficiently established; however, several key questions about its clinical utility remain unresolved. For example, although the initial studies of the efficacy of atomoxetine provided useful information for clinicians treating adults with ADHD, such as the average magnitude of the decrease in ADHD symptoms associated with drug treatment and the reliability of this effect, the standard methods of data presentation in these reports do not provide information about the full range of effects of this compound. To further characterize the clinical performance of atomoxetine, we completed a drug-placebo response curve analysis of the data initially reported by Michelson
et al. [
3] This method, described by Faraone
et al. [
5], is a generalization of receiver operating characteristic (ROC) analysis [
6], which has been widely applied to assessing the accuracy of diagnostic tests [
7‐
9]. The goal of this method is to identify additional characteristics of drug-placebo differences that have already been shown to be statistically significant, including: 1) the size of the effect using different response criteria; 2) the nature of individual responses; and 3) the portion of the drug's effect that is due to symptom improvement, the prevention of symptom worsening, or both.
Discussion
The results of two large randomized, double-blind, placebo-controlled trials of atomoxetine for the treatment of ADHD in adults were initially reported by Michelson
et al. [
3], who documented the superiority of this compound relative to placebo in reducing total ADHD symptoms, as well as inattentive and hyperactive symptoms of the illness. Due to the ample size and rigorous design of these studies, as well as the strong statistical significance of their results, the efficacy of atomoxetine has been firmly established. However, the simple knowledge that, on average, atomoxetine is efficacious does not tell clinicians much about its full range of effect.
Drug-placebo response curves provide an easily interpretable format for further evaluating clinically informative characteristics of a compound with proven efficacy. Because atomoxetine has demonstrable efficacy, drug-placebo response curve analysis of its performance against placebo was warranted. Collectively, the drug-placebo response curves presented here for each of the different reporters and the various dependent measures paint a consistent picture of the benefits of atomoxetine. First, it is clear that atomoxetine is superior to placebo in reducing total ADHD symptoms as well as individual symptom clusters, such as inattention and hyperactivity. For each of these measures, the drug-placebo response curve was always situated above the line of no effect, indicating that subjects were more likely to respond to atomoxetine than to placebo over the entire range of possible criteria of responsiveness. In addition, it is clear that atomoxetine targeted the core features of ADHD rather than only one of its most conspicuous features of inattention and hyperactivity, as AUCs across total, inattention, and hyperactivity change scores were quite similar. Second, responsiveness to atomoxetine was reliably assessed by clinicians, investigators, and patients, as the AUCs for the various dependent measures varied little (0.58–0.61) across reporters. Third, atomoxetine not only reduced the symptoms of ADHD, but prevented the worsening of these symptoms as well, a finding that has been seen for drug-placebo response curve analyses of other medications [
5,
10,
11]. In contrast however, these prior drug-placebo response curve analyses have also revealed stronger effects of other medications on clinician-rated ADHD symptomatology, as evidenced by AUCs of 0.86 for Adderall [
5,
10,
11], 0.89 for methylphenidate [
5,
10,
11], and 0.93 for desipramine [
5,
10,
11], as compared to the AUC of approximately 0.60 presently observed for atomoxetine.
In conclusion, we have extended the statistical results of Michelson
et al. [
3] by using drug-placebo response curves to describe the clinical significance of the efficacy of atomoxetine in the treatment of ADHD among adults. Our method of data presentation provides readers and clinicians with a means of understanding the nature of the effects of this drug, and the degree to which they are clinically relevant. Rather than collapsing individual responses into means or single rates of response, the drug-placebo response curve illustrates clinically meaningful details that often are lost in a standard analysis, such as the ability of atomoxetine to improve outcome and prevent worsening throughout the full range of outcome scores. The present drug-placebo response analysis provided strong support for the efficacy of atomoxetine relative to placebo for reducing inattention, hyperactivity, and total ADHD symptoms assessed by a variety of reporters, and for preventing the worsening of these symptoms. The finding that atomoxetine is efficacious through the full range of outcome further emphasizes the clinical value of treating ADHD adults with this medication.
Methods
Subjects
Two identical randomized, double-blind, placebo-controlled studies were conducted concurrently at 17 (Study I) and 14 (Study II) outpatient sites in North America. Each site's institutional review board evaluated and approved the study protocol, and written informed consent was obtained from each patient. Adults who met DSM-IV criteria for ADHD as assessed by clinical interview and confirmed by the Conners' Adult ADHD Diagnostic Interview for DSM-IV were recruited from clinics and by advertisement. Patients were required to have at least moderate symptom severity, and the diagnosis had to be corroborated by a second reporter for either current symptoms (by a significant other) or childhood symptoms (by a parent or older sibling). Patients who met diagnostic criteria for any other Axis-I disorder using the Structured Clinical Interview for DSM-IV were excluded, as were patients with serious medical illness or habitual substance abuse.
Atomoxetine and Placebo Administration
Following an initial one-week medication washout and evaluation period, patients entered a two-week placebo lead-in phase. Patients who maintained the initial severity criteria required for study entry were randomized to receive atomoxetine or placebo for a 10-week period. Atomoxetine was administered in evenly divided doses in the morning and late afternoon/early evening beginning at a total daily dose of 60 mg. Patients with residual symptoms received higher doses of up to 90 mg/day after two weeks and 120 mg/day after four weeks. If patients developed problems tolerating this regimen, the dose could be decreased to the last tolerated dose or an increase in dosage could be omitted. Across both studies, 270 subjects received atomoxetine, while 263 subjects received placebo. Of these, 197 completed acute treatment with atomoxetine, while 211 placebo-treated subjects completed the trial, a difference that was not significant.
Outcome Measures
The outcome measures examined in this study were derived from the CAARS and the CGI. A clinician completed the CGI before and after the treatment regimen, while both the subject and an investigator completed the CAARS before and after treatment. The three groups of primary dependent measures of this study included: 1.) clinician-rated CGI ADHD Severity change scores and endpoint scores; 2.) investigator-rated inattention, hyperactivity/impulsivity, total symptoms, and ADHD index scores on the CAARS; and 3.) self-rated inattention, hyperactivity/impulsivity, total symptoms, and ADHD index scores on the CAARS.
Drug-Placebo Response Curve Analysis
The rationale and methodology for drug-placebo response curve analysis methods are described in detail by Faraone
et al. [
5] The goal of response curve analysis is not to demonstrate statistically significant group differences; rather, this method provides an alternative means of displaying differences that have already been demonstrated to be statistically significant. Thus, it does not replace a standard statistical analysis, but augments that analysis by showing the clinical significance of drug effects. For the present study, the use of drug-placebo response curve analysis is warranted, as the statistically significant effects of atomoxetine on reducing symptoms of ADHD in adults have been documented previously.
The drug-placebo response curve is constructed in the following six steps: 1.) Choose an outcome variable, for example the change in CAARS Inattention score from baseline to the end of the study; 2.) At each observed score, calculate separately for the drug and placebo groups the proportion of subjects having that score or a better score. For CAARS change scores, therapeutic change is indicated by negative numbers,
i.e., a decrease in the symptom score; 3.) For each observed score, plot these proportions for the drug group on the vertical axis against the proportions computed for the placebo group on the horizontal axis; 4.) Connect the plotted points and label those that correspond to the best response, the 25
th percentile of response, the median response, the 75
th percentile of response and the worst response; 5.) If the outcome variable is a change score, also label the point corresponding to no change; 6.) Plot the line of no effect, which is the diagonal line from the [0, 0] point to the [
1,
1] point. Each point along a curve represents an observed outcome score on that measure, and the points on each plot are then connected by line segments. The line of no effect comprises all points for which the proportion of subjects who respond to drug is the same as the proportion who respond to placebo.
The drug-placebo response curve is a graphical method of describing results from a clinical trial, not a statistical test. It is most sensibly used to describe an effect that has been demonstrated with appropriate statistical tools. Nevertheless, the drug-placebo response curve's roots in (ROC) analysis motivate the computation of one statistic, the AUC, which is computed through integration. The area under the drug-placebo response curve ranges from 0.5 (when the drug effect equals the placebo effect) to 1.0 (when the drug is completely effective and the placebo has no effect). The AUC is a useful index of clinical significance because it equals the probability that a randomly selected member of the drug group will have a better result than a randomly selected member of the placebo group [
12,
13],
i.e., the probability that drug will outperform placebo.
In summary, the placebo-response curve provides four pieces of clinically relevant data not typically available from traditional statistical analyses of outcomes data. First, the effect size of a drug on an outcome measure can be determined as the distance between the curve and the line of no effect at any given cut-point. Second, the ratio of drug responders to placebo responders across the range of outcomes can be determined as the area under the curve. Third, the likelihood of a drug to elicit a specific outcome (e.g., a clinically meaningful cut-point) can be determined as the proportion of drug-responders to placebo-responders at any given cut-point. Fourth, the ability of a drug to improve functioning vs. prevent worsening of functioning can be determined as the proportion of drug-responders to placebo-responders at the outcome score representing no change.
Competing interests
Stephen V. Faraone, PhD.- Stephen Faraone receives research funding from Lilly, McNeil and Shire.
Joseph Biederman, MD.- Joseph Biederman receives research support from the following sources: Shire Laboratories, Inc and Eli Lilly & Company, Pfizer Pharmaceutical, Cephalon Pharmaceutical,, Janssen Pharaceutical, Neurosearch. Pharmaceuticals, Stanley Medical Institute, Lilly Foundation, Prechter Foundation, NIMH, NICHD and NIDA
Dr. Joseph Biederman is a speaker for the following speaker's bureaus: Eli Lilly & Company, Pfizer Pharmaceutical, Novartis Pharmaceutical, Wyeth Ayerst, Shire Laboratories Inc, McNeil Pharmaceutical, and Cephalon Pharmaceutical
Dr. Joseph Biederman is on the advisory board for the following pharmaceutical companies: Eli Lilly & Company, CellTech, Shire Laboratories Inc, Novartis Pharmaceutical, Noven Pharmaceutical, McNeil Pharmaceuticals, Janssen, Johnson & Johnson, Pfizer, and Cephalon Pharmaceuticals
Thomas Spencer, MD- Dr. Thomas Spencer receives research support from the following sources: Shire Laboratories, Inc and Eli Lilly & Company, Glaxo-Smith Kline, Pfizer Pharmaceutical, McNeil Pharmaceutical, Novartis Pharmaceutical, and NIMH
Dr. Thomas Spencer is a speaker for the following speaker's bureaus: Glaxo-Smith Kline, Eli Lilly & Company, Novartis Pharmaceutical, Wyeth Ayerst, Shire Laboratories Inc, McNeil Pharmaceutical
Dr. Thomas Spencer is on the advisory board for the following pharmaceutical companies: Shire Laboratories, Inc and Eli Lilly & Company, Glaxo-Smith Kline, Pfizer Pharmaceutical, McNeil Pharmaceutical, and Novartis Pharmaceutical
David Michelson, MD- David Michelson is a Lilly employee
Lenard Adler, MD- Lenard Adler receives grant and Research Support, is a Consultant or on Advisory Boards: Abbott Laboratories, Bristol-Myers Squibb, Eli Lilly and Co., McNeil/Johnson & Johnson, Merck & Co., Inc., Neurosearch, Novartis Pharmaceuticals Corp., Pfizer Labs, Cortex Pharmaceuticals, Cephalon and Shire Pharmaceuticals
Fred Reimherr, MD- Fred Reimherr has been part of Lilly advisory board.
Stephen J Glatt, PhD- Stephen Glatt has no conflicts of interest to declare.
Authors' contributions
Stephen V. Faraone, PhD- Steve Faraone contributed to the analysis and interpretation of the data, the drafting and revision of the manuscript.
Joseph Biederman, MD- Joseph Biederman contributed to the analysis and interpretation of the data, the drafting and revision of the manuscript.
Thomas Spencer, MD- Thomas Spencer contributed to the conception and design of the study, the acquisition of data, interpretation of data and drafting/reviewing the manuscript.
Lenard Adler, MD- Lenard Adler contributed to the conception and design of the study, the acquisition of data, interpretation of data and drafting/reviewing the manuscript.
David Michelson, MD- David Michelson contributed to the study conception design, the data acquisition as well as critical reviewing of the manuscript.
Fred Reimherr, MD- Fred Reimherr contributed to the study conception design, the data acquisition as well as critical reviewing of the manuscript.
Stephen J Glatt, PhD- Stephen Glatt contributed to the analysis and interpretation of the data, and the drafting and revision of the manuscript.