555973
research-article2014
CPJXXX10.1177/0009922814555973Clinical PediatricsWietecha et al
Article
Differential Response Profiles in Children and Adolescents With Attention-Deficit/Hyperactivity Disorder: Treatment With Atomoxetine
Clinical Pediatrics 2015, Vol. 54(2) 164–173 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/0009922814555973 cpj.sagepub.com
Linda A. Wietecha, BSN, MS1, Shufang Wang, PhD1, Keith E. Saylor, PhD2, Kathleen A. Day, RN, MSN1, Sheng Hu Wu3, and Doug Kelsey, MD1
Abstract Atomoxetine has been shown to be safe and effective in the treatment of attention-deficit/hyperactivity disorder (ADHD). The purpose of this post hoc analysis was to examine response trajectories of pediatric patients treated with atomoxetine. Data were pooled from 7 atomoxetine double-blind, placebo-controlled clinical trials conducted in pediatric patients between November 1998 and June 2004. Growth mixture modeling was applied to the investigator-rated ADHD rating scale (ADHDRS-Inv) and Clinical Global Impressions-ADHD-Severity (CGIADHD-S) scores in the randomized acute phase (6-9 weeks) to explore whether there were groups of patients who differed in their response to atomoxetine. Classification and regression tree analyses were performed to identify predictors that can help categorize subjects to different response profiles. Patients (N = 925) were mostly male (73%) and of the combined subtype (74%). Most patients had a response pattern characterized by gradual, modest improvement, while a smaller group experienced early, robust improvement. Keywords atomoxetine, ADHD, growth mixture modeling (GMM) analysis, classification and regression tree (CART) analysis
Introduction Stimulant medications are well established for the management of attention-deficit/hyperactivity disorder (ADHD). In a 2010 meta-analysis of double-blind placebo-controlled studies, the response rate for the stimulant dexamphetamine was 88%, and the response rate for the stimulant methylphenidate was 79%.1 A comparative study of the stimulants methylphenidate and amphetamine concluded that 87% of children with ADHD would respond well to 1 of the 2 stimulant classes.2 Additionally, the American Academy of Pediatrics, an international consensus statement,3 and the Texas Children’s Medication Project4 have recommended stimulants as the first line of treatment for ADHD, particularly in patients without comorbidities. However, atomoxetine, a selective noradrenergic reuptake inhibitor, may be considered as the first medication for ADHD in individuals with an active substance abuse problem, comorbid anxiety, or tics. Atomoxetine is also preferred for patients experiencing severe side effects to stimulants, such as mood lability or tics.5
Atomoxetinewas first approved on November 26, 2002, for use in children and adolescents 6 to 17 years of age and adults as a nonstimulant treatment for ADHD. Atomoxetine has been shown to be effective (and have a favorable safety profile) in the treatment of ADHD in children and adolescents.6-8 However, even though the mean change from baseline and improvement relative to placebo were statistically significant in these clinical trials, there were also many patients who did not respond to treatment with atomoxetine. For example, Spencer et al7 reported the results of 2 studies of atomoxetine in patients with ADHD. They found, by using a criterion of ≥25% reduction in the ADHD Rating Scale-IV-Parent Version: Investigator-Administered and Scored (ADHDRS) total 1
Eli Lilly and Company, Indianapolis, IN, USA NeuroScience, Inc, Herndon, VA, USA 3 Eli Lilly Asia, Shanghai, People’s Republic of China 2
Corresponding Author: Linda Wietecha, US Medical Neuroscience, Eli Lilly and Company, Lilly Corporate Center, DC 4135, Indianapolis, IN 26285, USA. Email: [email protected]
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
165
Wietecha et al Table 1. Patients.
Study
Week Postbaseline in Which Data Were Collected (Visit Is Indicated)
Patients, n (%); Total = 925
0
1
2
3
4
206 (22.3) 81 (8.8) 65 (7.0) 63 (6.8) 208 (22.5) 126 (13.6) 176 (19.0)
V3 V2 V3 V3 V2 V2 V2
V4 V3 V4 V4
V5 V4 V5 V5 V3
V6
V7 V5 V7 V7 V4 V4
B4Z-MC-LYAC B4Z-MC-LYAT B4Z-MC-HFBD B4Z-MC-HFBK B4Z-MC-LYBI B4Z-MC-LYBG B4Z-US-LYCC
V3 V3
V6 V6
V4
5
6
V8 V8
V8 V6 V9 V9 V5
7
8
9
Reference
Michelson et al (2001)8 Michelson et al (2002)22 V11 V12 Spencer et al (2002)7 V11 V12 Spencer et al (2002)7 Newcorn et al (2008)24 V5 Kelsey et al (2004)13 Block et al (2009)12 V9
V10 V10
V5
Abbreviation: V, visit.
score, that response rates were 64.1% for study 1 and 58.7% for study 2. There are some indications that subpopulations vary in their response to atomoxetine.9 Clinical experience and exploratory research findings suggest that there is a subgroup of responders who are much improved with atomoxetine (≥40% decrease on ADHDRS total score), some of whom show preferential response to atomoxetine over stimulants.9,10 Newcorn et al9 found that the distribution of clinical responses to atomoxetine was bimodal, with most subjects being either responders who were much improved (47% of the sample) or nonresponders (40% of the sample). The reasons for this heterogeneity of response are unclear. It is possible that baseline symptom severity or patient characteristics play a role. For example, it is known that the rate of clearance of atomoxetine in cytochrome P450 2D6 poor metabolizers is approximately one tenth that of extensive metabolizers11; therefore, poor metabolizers are exposed to greater concentrations of the drug and run a greater risk of adverse reactions. Predictors of response could provide useful information that would lead to a more appropriate treatment for the patient and minimize the time spent exploring a treatment that is unlikely to be successful for that individual. Because the time course for response to atomoxetine can be up to 6 weeks,7,8,12,13 we conducted this post hoc analysis to explore potential differential response profiles to atomoxetine among pediatric patients. The purpose of this post hoc analysis was to (a) examine response trajectories of pediatric patients treated with atomoxetine to determine if there are identifiable groups of patients with distinctly different patterns of response to atomoxetine over time and (b) determine if there is a clearly definable robust response to atomoxetine, to attempt to identify those patients with a high probability of having a good response to atomoxetine before treatment begins or very early in treatment.
Methods Studies Data were pooled from 7 atomoxetine double-blind, placebo-controlled, randomized clinical trials conducted in pediatric patients (ages 6-18 years) between November 1998 and June 2004 (Table 1).7,8,12,13 Only patients treated with atomoxetine were included in this analysis (regardless of how well they responded to treatment). However, patients who were primarily of the hyperactive/impulsive subtype were excluded (because of small sample size). In addition to meeting the clinical criteria for ADHD diagnosis, all patients were required to meet a symptom severity threshold: a score of at least 1.5 standard deviations (SDs) above age and gender norms on the ADHDRS-Inv, for the total score or either of the inattention or hyperactivity/impulsivity subscales. These studies examined the effect of atomoxetine on symptom reduction as measured by the ADHD Rating Scale (ADHD-RS) total score as the primary outcome measure. The study designs, inclusion and exclusion criteria, and assessments in each of the 7 trials presented here were similar and have been published previously. Importantly, subjects taking any other psychotropic medication were excluded as were subjects who had a documented history of bipolar I or bipolar II disorder or any history of psychosis. Other studies that used a relapse-prevention design or recruited subjects with a specific comorbid disorder (ie, major depressive disorder, anxiety disorders, or enuresis) were not included in the present analysis either because the study methodology was different or because adding smaller and potentially heterogeneous samples might have limited our ability to detect predictors of response. However, subjects with comorbid disorders (except bipolar I or II disorder or any psychotic disorder) were permitted to enroll in the 7 trials included in this analysis, provided that their primary disorder was ADHD and enrolling in an
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
166
Clinical Pediatrics 54(2)
ADHD treatment study was considered clinically appropriate. We examined only subjects treated with atomoxetine in double-blind studies, and not the open atomoxetine treatment given afterward to subjects originally randomized to placebo, because the magnitude of improvement in subjects treated openly could be inflated by expectation and/or other aspects of the placebo response. In each of the 7 studies, the pretreatment to post-treatment change score on the ADHD-RS was significant compared with placebo. In all 7 studies, atomoxetine was administered on a weight-adjusted basis to a maximum allowed daily dosage of 1.4 to 2.0 mg/kg, though there were differences in titration rates and dosing frequencies across studies. All studies were performed in compliance with the principles of Good Clinical Practice and were carried out in accordance with the latest version of the Declaration of Helsinki. Study protocols were reviewed by appropriate local ethical committees, and informed consent of the participants was obtained after the nature of the procedures had been fully explained.
Measures Two scales were used to measure response to atomoxetine and thus define the response trajectories: the ADHDRS-Inv and the Clinical Global Impressions– Attention Deficit/Hyperactivity Disorder–Severity (CGI-ADHD-S). The ADHDRS-Inv is an 18-item scale with 1 item for each of the 18 symptoms associated with a diagnosis in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. Each item is scored on a 0 to 3 scale (0 = never or rarely; 1 = sometimes; 2 = often; 3 = very often). The rating scale assessed the symptom severity since the previous visit. The total score was computed as the sum of the scores on each of the 18 items. In addition, the inattention and the hyperactivity-impulsivity subscores were computed. The scale was scored by a qualified researcher based on an interview with the parent.14 The CGI-ADHD-S is a single-item rating of the clinician’s assessment of the severity of ADHD symptoms in relation to the clinician’s total experience with ADHD patients. Severity is rated on a 7-point scale (1 = normal, not at all ill; 7 = among the most extremely ill patients).15,16
Statistical Methods We wanted to explore whether there were groups of patients who differed in their response to atomoxetine. Therefore, to identify differential response profiles to atomoxetine among pediatric patients with ADHD, the growth mixture modeling (GMM)15-17 technique was
applied to ADHDRS-Inv total scores and subscale scores and CGI-ADHD-S scores recorded during the acute phases (6-9 weeks) of the 7 clinical trials described above. For the ADHDRS-Inv hyperactivity/impulsivity subscale score, only data from combined subtype patients were analyzed (subjects of the inattentive subtype were excluded). Data collected at baseline and at the first 4 postbaseline visits were modeled quadratically; data collected later were modeled linearly. Multiple tests for the same outcome variable were adjusted using the Bonferroni method. To test the association between individual baseline characteristics and identified trajectory membership, Chi-square tests were used for categorical variables (gender, ADHD subtype, prior stimulant use, and cytochrome P450 2D6 [CYP2D6] status). Classification and regression tree (CART)18 analyses were performed to identify predictors that can help categorize subjects to different response profiles. Predictors included in the CART analyses were gender, age at diagnosis of ADHD, subtype (not included when analyzing class membership based on hyperactivity/impulsivity subscores), prior stimulant use, baseline Conners’s Parent Rating Scale-Revised: Short Form (CPRS-R:S)19 total score, baseline Child Health Questionnaire Parent Form 50 (CHQ)20 physical summary (PHS) subscore and baseline CHQ psychosocial summary (PSS) subscore, and baseline Life Participation Scale for ADHD total score (LPS).21 ADHDRS total scores, subscale scores, and CGI-ADHD-S scores up to 2 weeks after baseline were also used as predictors. Data were randomly split into 80% training data and 20% test data, and training data were used to build the trees. CARTs were built by recursively splitting the data into 2 subgroups to have similar trajectories of an outcome variable within each resulting subgroup, with the Gini index18 (that measures the degree of discrimination) as the splitting criterion. Typically, the procedure results in very large trees with many predictors, but such a tree may not have very good predictive utility when applied to new data. Therefore, to improve the generalizability of the tree, the original larger trees were pruned to the optimal size as suggested by the 10-fold crossvalidation of the training data. We examined the performance of the trees on test data using prediction error rate, which is the percentage of subjects that were incorrectly classified using a tree, compared with the class membership identified using the GMM analysis. All statistical tests were 2-tailed, with α = .05. GMM analyses were conducted using Mplus software (version 7.11, Muthén & Muthén, Los Angeles, CA) and CART analyses were conducted using R software (version 2.15.3, The R Foundation for Statistical Computing,
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
167
Wietecha et al
Figure 1. Growth mixture modeling of investigator-rated Attention-Deficit/Hyperactivity Disorder Rating Scale total and subscale scores and Clinical Global Impressions–Attention-Deficit/Hyperactivity Disorder–Severity scores: response trajectories with atomoxetine treatment. Abbreviations: ADHDRS-Inv, Investigator-Rated Attention-Deficit/Hyperactivity Disorder Rating Scale; CGI-ADHD-S, Clinical Global Impressions–ADHD–Severity.
Vienna, Austria). All other analyses were conducted using SAS software (version 9.2; SAS Institute, Cary, NC).
Results A total of 925 patients who had at least 1 postbaseline measurement were identified. Patients were mostly male (73%), of the combined subtype (74%), had been treated with stimulants (56%), and were not slow CYP2D6 metabolizers (95%). The average age of ADHD symptom onset was 4.38 years (SD = 1.35). Baseline ADHD mean severity (SD) was as follows: ADHDRS-Inv total score = 40.68 (8.75), inattention subscale score = 22.37 (3.72), hyperactivity/impulsivity subscale score = 18.31 (6.69), and CGI-ADHD-S = 4.94 (0.78). Results of the GMM analysis are presented in Figure 1. Treatment response trajectories on each of the scales and subscales were characterized by baseline severity and percentage improvement. Baseline ADHDRS-Inv scores were characterized as follows:
total score ≥40 = most severe and
research-article2014
CPJXXX10.1177/0009922814555973Clinical PediatricsWietecha et al
Article
Differential Response Profiles in Children and Adolescents With Attention-Deficit/Hyperactivity Disorder: Treatment With Atomoxetine
Clinical Pediatrics 2015, Vol. 54(2) 164–173 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/0009922814555973 cpj.sagepub.com
Linda A. Wietecha, BSN, MS1, Shufang Wang, PhD1, Keith E. Saylor, PhD2, Kathleen A. Day, RN, MSN1, Sheng Hu Wu3, and Doug Kelsey, MD1
Abstract Atomoxetine has been shown to be safe and effective in the treatment of attention-deficit/hyperactivity disorder (ADHD). The purpose of this post hoc analysis was to examine response trajectories of pediatric patients treated with atomoxetine. Data were pooled from 7 atomoxetine double-blind, placebo-controlled clinical trials conducted in pediatric patients between November 1998 and June 2004. Growth mixture modeling was applied to the investigator-rated ADHD rating scale (ADHDRS-Inv) and Clinical Global Impressions-ADHD-Severity (CGIADHD-S) scores in the randomized acute phase (6-9 weeks) to explore whether there were groups of patients who differed in their response to atomoxetine. Classification and regression tree analyses were performed to identify predictors that can help categorize subjects to different response profiles. Patients (N = 925) were mostly male (73%) and of the combined subtype (74%). Most patients had a response pattern characterized by gradual, modest improvement, while a smaller group experienced early, robust improvement. Keywords atomoxetine, ADHD, growth mixture modeling (GMM) analysis, classification and regression tree (CART) analysis
Introduction Stimulant medications are well established for the management of attention-deficit/hyperactivity disorder (ADHD). In a 2010 meta-analysis of double-blind placebo-controlled studies, the response rate for the stimulant dexamphetamine was 88%, and the response rate for the stimulant methylphenidate was 79%.1 A comparative study of the stimulants methylphenidate and amphetamine concluded that 87% of children with ADHD would respond well to 1 of the 2 stimulant classes.2 Additionally, the American Academy of Pediatrics, an international consensus statement,3 and the Texas Children’s Medication Project4 have recommended stimulants as the first line of treatment for ADHD, particularly in patients without comorbidities. However, atomoxetine, a selective noradrenergic reuptake inhibitor, may be considered as the first medication for ADHD in individuals with an active substance abuse problem, comorbid anxiety, or tics. Atomoxetine is also preferred for patients experiencing severe side effects to stimulants, such as mood lability or tics.5
Atomoxetinewas first approved on November 26, 2002, for use in children and adolescents 6 to 17 years of age and adults as a nonstimulant treatment for ADHD. Atomoxetine has been shown to be effective (and have a favorable safety profile) in the treatment of ADHD in children and adolescents.6-8 However, even though the mean change from baseline and improvement relative to placebo were statistically significant in these clinical trials, there were also many patients who did not respond to treatment with atomoxetine. For example, Spencer et al7 reported the results of 2 studies of atomoxetine in patients with ADHD. They found, by using a criterion of ≥25% reduction in the ADHD Rating Scale-IV-Parent Version: Investigator-Administered and Scored (ADHDRS) total 1
Eli Lilly and Company, Indianapolis, IN, USA NeuroScience, Inc, Herndon, VA, USA 3 Eli Lilly Asia, Shanghai, People’s Republic of China 2
Corresponding Author: Linda Wietecha, US Medical Neuroscience, Eli Lilly and Company, Lilly Corporate Center, DC 4135, Indianapolis, IN 26285, USA. Email: [email protected]
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
165
Wietecha et al Table 1. Patients.
Study
Week Postbaseline in Which Data Were Collected (Visit Is Indicated)
Patients, n (%); Total = 925
0
1
2
3
4
206 (22.3) 81 (8.8) 65 (7.0) 63 (6.8) 208 (22.5) 126 (13.6) 176 (19.0)
V3 V2 V3 V3 V2 V2 V2
V4 V3 V4 V4
V5 V4 V5 V5 V3
V6
V7 V5 V7 V7 V4 V4
B4Z-MC-LYAC B4Z-MC-LYAT B4Z-MC-HFBD B4Z-MC-HFBK B4Z-MC-LYBI B4Z-MC-LYBG B4Z-US-LYCC
V3 V3
V6 V6
V4
5
6
V8 V8
V8 V6 V9 V9 V5
7
8
9
Reference
Michelson et al (2001)8 Michelson et al (2002)22 V11 V12 Spencer et al (2002)7 V11 V12 Spencer et al (2002)7 Newcorn et al (2008)24 V5 Kelsey et al (2004)13 Block et al (2009)12 V9
V10 V10
V5
Abbreviation: V, visit.
score, that response rates were 64.1% for study 1 and 58.7% for study 2. There are some indications that subpopulations vary in their response to atomoxetine.9 Clinical experience and exploratory research findings suggest that there is a subgroup of responders who are much improved with atomoxetine (≥40% decrease on ADHDRS total score), some of whom show preferential response to atomoxetine over stimulants.9,10 Newcorn et al9 found that the distribution of clinical responses to atomoxetine was bimodal, with most subjects being either responders who were much improved (47% of the sample) or nonresponders (40% of the sample). The reasons for this heterogeneity of response are unclear. It is possible that baseline symptom severity or patient characteristics play a role. For example, it is known that the rate of clearance of atomoxetine in cytochrome P450 2D6 poor metabolizers is approximately one tenth that of extensive metabolizers11; therefore, poor metabolizers are exposed to greater concentrations of the drug and run a greater risk of adverse reactions. Predictors of response could provide useful information that would lead to a more appropriate treatment for the patient and minimize the time spent exploring a treatment that is unlikely to be successful for that individual. Because the time course for response to atomoxetine can be up to 6 weeks,7,8,12,13 we conducted this post hoc analysis to explore potential differential response profiles to atomoxetine among pediatric patients. The purpose of this post hoc analysis was to (a) examine response trajectories of pediatric patients treated with atomoxetine to determine if there are identifiable groups of patients with distinctly different patterns of response to atomoxetine over time and (b) determine if there is a clearly definable robust response to atomoxetine, to attempt to identify those patients with a high probability of having a good response to atomoxetine before treatment begins or very early in treatment.
Methods Studies Data were pooled from 7 atomoxetine double-blind, placebo-controlled, randomized clinical trials conducted in pediatric patients (ages 6-18 years) between November 1998 and June 2004 (Table 1).7,8,12,13 Only patients treated with atomoxetine were included in this analysis (regardless of how well they responded to treatment). However, patients who were primarily of the hyperactive/impulsive subtype were excluded (because of small sample size). In addition to meeting the clinical criteria for ADHD diagnosis, all patients were required to meet a symptom severity threshold: a score of at least 1.5 standard deviations (SDs) above age and gender norms on the ADHDRS-Inv, for the total score or either of the inattention or hyperactivity/impulsivity subscales. These studies examined the effect of atomoxetine on symptom reduction as measured by the ADHD Rating Scale (ADHD-RS) total score as the primary outcome measure. The study designs, inclusion and exclusion criteria, and assessments in each of the 7 trials presented here were similar and have been published previously. Importantly, subjects taking any other psychotropic medication were excluded as were subjects who had a documented history of bipolar I or bipolar II disorder or any history of psychosis. Other studies that used a relapse-prevention design or recruited subjects with a specific comorbid disorder (ie, major depressive disorder, anxiety disorders, or enuresis) were not included in the present analysis either because the study methodology was different or because adding smaller and potentially heterogeneous samples might have limited our ability to detect predictors of response. However, subjects with comorbid disorders (except bipolar I or II disorder or any psychotic disorder) were permitted to enroll in the 7 trials included in this analysis, provided that their primary disorder was ADHD and enrolling in an
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
166
Clinical Pediatrics 54(2)
ADHD treatment study was considered clinically appropriate. We examined only subjects treated with atomoxetine in double-blind studies, and not the open atomoxetine treatment given afterward to subjects originally randomized to placebo, because the magnitude of improvement in subjects treated openly could be inflated by expectation and/or other aspects of the placebo response. In each of the 7 studies, the pretreatment to post-treatment change score on the ADHD-RS was significant compared with placebo. In all 7 studies, atomoxetine was administered on a weight-adjusted basis to a maximum allowed daily dosage of 1.4 to 2.0 mg/kg, though there were differences in titration rates and dosing frequencies across studies. All studies were performed in compliance with the principles of Good Clinical Practice and were carried out in accordance with the latest version of the Declaration of Helsinki. Study protocols were reviewed by appropriate local ethical committees, and informed consent of the participants was obtained after the nature of the procedures had been fully explained.
Measures Two scales were used to measure response to atomoxetine and thus define the response trajectories: the ADHDRS-Inv and the Clinical Global Impressions– Attention Deficit/Hyperactivity Disorder–Severity (CGI-ADHD-S). The ADHDRS-Inv is an 18-item scale with 1 item for each of the 18 symptoms associated with a diagnosis in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. Each item is scored on a 0 to 3 scale (0 = never or rarely; 1 = sometimes; 2 = often; 3 = very often). The rating scale assessed the symptom severity since the previous visit. The total score was computed as the sum of the scores on each of the 18 items. In addition, the inattention and the hyperactivity-impulsivity subscores were computed. The scale was scored by a qualified researcher based on an interview with the parent.14 The CGI-ADHD-S is a single-item rating of the clinician’s assessment of the severity of ADHD symptoms in relation to the clinician’s total experience with ADHD patients. Severity is rated on a 7-point scale (1 = normal, not at all ill; 7 = among the most extremely ill patients).15,16
Statistical Methods We wanted to explore whether there were groups of patients who differed in their response to atomoxetine. Therefore, to identify differential response profiles to atomoxetine among pediatric patients with ADHD, the growth mixture modeling (GMM)15-17 technique was
applied to ADHDRS-Inv total scores and subscale scores and CGI-ADHD-S scores recorded during the acute phases (6-9 weeks) of the 7 clinical trials described above. For the ADHDRS-Inv hyperactivity/impulsivity subscale score, only data from combined subtype patients were analyzed (subjects of the inattentive subtype were excluded). Data collected at baseline and at the first 4 postbaseline visits were modeled quadratically; data collected later were modeled linearly. Multiple tests for the same outcome variable were adjusted using the Bonferroni method. To test the association between individual baseline characteristics and identified trajectory membership, Chi-square tests were used for categorical variables (gender, ADHD subtype, prior stimulant use, and cytochrome P450 2D6 [CYP2D6] status). Classification and regression tree (CART)18 analyses were performed to identify predictors that can help categorize subjects to different response profiles. Predictors included in the CART analyses were gender, age at diagnosis of ADHD, subtype (not included when analyzing class membership based on hyperactivity/impulsivity subscores), prior stimulant use, baseline Conners’s Parent Rating Scale-Revised: Short Form (CPRS-R:S)19 total score, baseline Child Health Questionnaire Parent Form 50 (CHQ)20 physical summary (PHS) subscore and baseline CHQ psychosocial summary (PSS) subscore, and baseline Life Participation Scale for ADHD total score (LPS).21 ADHDRS total scores, subscale scores, and CGI-ADHD-S scores up to 2 weeks after baseline were also used as predictors. Data were randomly split into 80% training data and 20% test data, and training data were used to build the trees. CARTs were built by recursively splitting the data into 2 subgroups to have similar trajectories of an outcome variable within each resulting subgroup, with the Gini index18 (that measures the degree of discrimination) as the splitting criterion. Typically, the procedure results in very large trees with many predictors, but such a tree may not have very good predictive utility when applied to new data. Therefore, to improve the generalizability of the tree, the original larger trees were pruned to the optimal size as suggested by the 10-fold crossvalidation of the training data. We examined the performance of the trees on test data using prediction error rate, which is the percentage of subjects that were incorrectly classified using a tree, compared with the class membership identified using the GMM analysis. All statistical tests were 2-tailed, with α = .05. GMM analyses were conducted using Mplus software (version 7.11, Muthén & Muthén, Los Angeles, CA) and CART analyses were conducted using R software (version 2.15.3, The R Foundation for Statistical Computing,
Downloaded from cpj.sagepub.com at East Carolina University on April 25, 2015
167
Wietecha et al
Figure 1. Growth mixture modeling of investigator-rated Attention-Deficit/Hyperactivity Disorder Rating Scale total and subscale scores and Clinical Global Impressions–Attention-Deficit/Hyperactivity Disorder–Severity scores: response trajectories with atomoxetine treatment. Abbreviations: ADHDRS-Inv, Investigator-Rated Attention-Deficit/Hyperactivity Disorder Rating Scale; CGI-ADHD-S, Clinical Global Impressions–ADHD–Severity.
Vienna, Austria). All other analyses were conducted using SAS software (version 9.2; SAS Institute, Cary, NC).
Results A total of 925 patients who had at least 1 postbaseline measurement were identified. Patients were mostly male (73%), of the combined subtype (74%), had been treated with stimulants (56%), and were not slow CYP2D6 metabolizers (95%). The average age of ADHD symptom onset was 4.38 years (SD = 1.35). Baseline ADHD mean severity (SD) was as follows: ADHDRS-Inv total score = 40.68 (8.75), inattention subscale score = 22.37 (3.72), hyperactivity/impulsivity subscale score = 18.31 (6.69), and CGI-ADHD-S = 4.94 (0.78). Results of the GMM analysis are presented in Figure 1. Treatment response trajectories on each of the scales and subscales were characterized by baseline severity and percentage improvement. Baseline ADHDRS-Inv scores were characterized as follows:
total score ≥40 = most severe and
