Professional Documents
Culture Documents
MTA Cooperative Group
MTA Cooperative Group
The online version of this article, along with updated information and services, is
located on the World Wide Web at:
http://www.pediatrics.org/cgi/content/full/113/4/754
754
METHODS
Sample
Table 1 shows the demographics of the 540 participants who
were assessed in the 10-month follow-up, which are not different
from the demographic characteristics of the originally randomized
579 MTA participants described elsewhere. The only statistically
significant difference among treatment groups was a trivial difference in age (MedMgt was 0.3 years older than Beh). As a check,
the first analysis below was repeated with age covaried; it made
no difference. In addition to all children meeting criteria for
ADHD-Combined type at baseline, parents were interviewed with
the Diagnostic Interview Schedule for Children16 for other child
psychiatric disorders. The ADHD Diagnostic Interview Schedule
for Children module was completed face-to-face with the childs
principal caregiver by trained research interviewers.
Assessments
On the basis of experience from 14-month analyses and statistical advice, we selected 5 measures from conceptually distinct
domains that were likely to be affected by treatment: 1) parentand teacher-rated Swanson, Nolan, and Pelham Scale (SNAP)17
ADHD symptoms, 2) parent- and teacher-rated oppositional defiant disorder (ODD) symptoms,17 3) Wechsler Individual Achievement Test reading score,18 4) a negative/ineffective discipline
factor,5,6 and 5) parent- and teacher-rated total social skills from
the Social Skills Rating System (SSRS).19
Analyses
Univariate analyses of the 24-month means were performed
with baseline covaried, with 1 main test for each of the 5 symptom
and function domains, using a mixed-effects regression mod-
Participant variables
Age (y; mean [SD])
Male (n [%])
Ethnicity (n [%])
White
Black
Hispanic
Grade (n [%])
First
Second
Third
Fourth
Fifth
WISC-3 IQ (mean [SD])
Verbal
Performance
Total
Parent/family variables
High school graduate (n [%])
Mother
Father
Employed (n [%])*
Mother
Father
Welfare (n [%])
Income (n [%])
0-$20K
20-$50K
$50K
Married (n [%])
Family composition
Parent (n [%])
Totals Whole
Sample
(n 540)
Comb
(n 138)
MedMgt
(n 128)
Beh
(n 139)
CC
(n 135)
Range Across
Sites
(P Value)
8.4 (0.8)
107 (78)
8.6 (0.9)
105 (82)
8.3 (0.8)
110 (79)
8.5 (0.8)
108 (80)
329 (61)
105 (19)
46 (9)
82 (59)
23 (17)
14 (10)
84 (66)
23 (18)
11 (9)
79 (57)
36 (26)
12 (9)
84 (62)
23 (17)
9 (7)
84 (16)
226 (42)
163 (30)
66 (12)
1 (0.2)
20 (14)
60 (43)
43 (31)
15 (11)
16 (13)
47 (37)
48 (38)
16 (13)
1 (0.8)
29 (21)
63 (45)
33 (24)
14 (10)
19 (14)
56 (41)
39 (29)
21 (16)
8.48.6 (.01)
72%88% (.06)
Overall 2 (.0001)
22%81%
4%37%
0%36%
Overall 2 (.01)
6%29%
40%45%
22%41%
8%19%
0%1%
8.4 (0.8)
430 (80)
100.6 (14.9)
101.4 (15.8)
101.0 (14.8)
100.6 (15.6)
101.1 (16.0)
100.7 (15.2)
98.7 (14.0)
100.1 (14.7)
99.1 (13.6)
100.9 (14.4)
101.5 (15.8)
101.1 (14.5)
102.2 (15.2)
103.2 (16.5)
102.8 (15.7)
96.4104.8 (.01)
95.1104.9 (.001)
95.2105.1 (.001)
505/536 (94)
385/423 (91)
131 (96)
106 (94)
116 (91)
88 (86)
130 (95)
95 (91)
128 (95)
96 (92)
87%100% (.01)
79%99% (.0001)
384/537 (72%)
353/418 (84%)
100 (19)
93 (68)
94 (85)
28 (20)
86 (67)
87 (86)
24 (19)
105 (77)
89 (86)
21 (15)
100 (74)
83 (81)
27 (20)
105 (19%)
224 (41%)
200 (37%)
350 (65)
26 (19)
52 (38)
58 (42)
93 (67)
23 (18)
58 (45)
43 (34)
86 (67)
30 (22)
56 (40)
52 (37)
81 (58)
26 (19)
58 (43)
47 (35)
90 (67)
62%79% (.06)
71%95% (.007)
10%40% (.0001)
Overall 2 (.001)
10%34%
35%50%
27%46%
53%72% (.09)
145 (27%)
34 (25)
28 (22)
48 (35)
35 (26)
21%37% (.12)
SD indicates standard deviation. Attrition from MedMgt was marginally greater than from Beh (P .05, with Bonferroni-corrected
significance level being .008). Most of the MedMgt attrition (13 of 16) occurred immediately after randomization, before beginning
treatment.
* Refers to that proportion of sample whose parents hold full- or part-time jobs.
Means proportion of sample with intact, 2-parent families (married or common-law).
Note that treatment groups differed significantly on only 1 variable (age), whereas sites differ significantly on almost all variables.
ARTICLES
755
el.20 24 We set the significance level at P .01 for each of the 5 tests
to maintain an overall P .05 significance level. Within each
domain, we performed orthogonal (nonoverlapping) contrasts
that were developed to decompose the overall effects at the end of
treatment. Three statistically independent questions were addressed4: 1) Do children who were exposed to the MTAs intensive
medication strategy (Comb or MedMgt) show persisting superior
outcomes over children who were not (Beh and CC)the MTA
medication algorithm effect? 2) Do children who were assigned
to Comb show superior outcomes over those who were assigned
to MedMgtthe multimodality superiority effect? 3) Do children who were exposed to intensive Beh show superior outcomes
over those in usual CCthe behavioral substitution effect?
In addition to these 5 domains of effectiveness, we obtained
measures from 4 domains of services used during the follow-up
phase (use and dose of medication, use of specialty mental health
services, and use of special education services) from a structured
interview developed for this purpose, the Services Use by Children and AdolescentsParent Interview (SCAPI). The SCAPI asks
the childs caregivers about any services that the child may have
received since the previous assessment, including medication,
physician or therapy contacts, and school services. We calculated
the percentages of participants in each treatment group who received the respective service between 14 and 24 months and
performed 2 tests to determine whether these percentages differed across the 4 treatment groups. Last, given the highly significant effects found at the 14-month endpoint as a function of the
MTA medication algorithm (Comb and MedMgt groups), we repeated any of the above analyses that showed significant differences between groups at 24 months, this time adjusting for participants medication use in the 14- to 24-month interim and in the
30-day period before the 24-month outcome assessments. These
analyses were done to parse out the extent to which any persisting
outcome differences might be fully or partly explained by participants current medication use, rather than their original group
assignment.
RESULTS
To test whether any differences found were attributable to differential medication use during 14 to 24
months, we repeated the 5 univariate analyses with
interim medication use covaried. We used 2 approaches to test for the effects of medication use,
based on whether the participant took any medication prescribed for ADHD at any time during the
10-month interval or whether the participant was
taking any of these medications in the 30 days before
the 24-month assessment. (Medications in this analysis included not only the psychostimulants, which
constituted 90% of the medications prescribed
across all 4 groups, but also bupropion, tricyclics,
and -agonists.) A third approach was briefly considered but rejected, on the basis of the advice of our
statistical consultant: medication dose, in methylphenidate equivalents at the 24-month assessment.
Examination of the data distribution showed that
this variable was too skewed for valid use.
Table 3 repeats the first 5 analyses from Table 2 but
with medication status covaried: whether the participant took medication for ADHD during the 14- to
24-month interim. In addition, these same analyses
were conducted using the medication variable
whether the participant took medication in the 30
days before the 24-month assessment. There were no
substantive differences between these 2 sets of analyses, so only the first set is shown here. These analyses generally reveal a significant main effect of
medication (the covariate) on ADHD symptoms and
tend toward such an effect on ODD symptoms and
social skills. These data suggest that whether participants took medication mediated significant differences in outcomes and that differential medicationtaking after the intensive 14-month randomized
treatment experience explains at least part of the
persisting advantage of the MTA medication algorithm (Comb and MedMgt). Note, however, even
after controlling for medication use (both for the 14to 24-month interim period and during the 30 days
before the 24-month assessment), the advantage of
Comb/MedMgt over Beh/CC remained significant
(P .002) for ADHD symptoms and almost significant (P .016) for ODD symptoms.
Clinical Significance
F 3.08 (P .08)
Parent 1.02 (0.029)
Teacher 0.95 (.030)
F 0.94 (P .42)
F 5.04 (P .0019)
F 5.67 (P .018)
Parent 1.08 (.025)
Teacher 1.16 (.025)
F 1.38 (P .25)
F 7.66 (P .0001)
Rater
Rater means
(SD)
Rater*Tx
Tx
Orthogonal
contrasts
1.17 (.29)
1.11 (.29)
1.09 (.28)
1.12 (.28)
24
Months
F 2.48 (P .06)
F 9.99 (P .002)
Parent 1.14 (.012)
Teacher 1.09 (.013)
F 0.52 (P .67)
F 1.16 (P .30)
F 4.08 (P .001)
0.94 (.28)
0.92 (.29)
0.91 (.26)
0.94 (.28)
BL
Social Skills
(SSRS Total P&T;
Mean [SD]*;
n 522;
Higher Better)
24
Months
1.45 (1.59)
0.86 (1.78)
1.05 (1.64)
0.99 (1.47)
F 2.60 (P .05)
F 1.56 (P .08)
F 0.76 (P .58)
0.14 (1.68)
0.24 (1.81)
0.07 (1.55)
0.07 (1.50)
BL
Negative Ineffective
Discipline Factor
(Mean [SD]; n 481; More
Negative Better)
24
Months
98.99 (14.17)
99.17 (15.94)
97.84 (13.96)
96.41 (13.29)
F 1.03 (P .38)
F 1.33 (P .18)
F 1.37 (P .23)
96.50 (14.61)
96.11 (13.82)
95.30 (13.66)
95.15 (14.04)
BL
Reading
(WIAT; Mean [SD];
n 510;
Higher Better)
24
Months
86%
70%
85%
72%
44%
38%
69%
62%
2 Tests Below
1424
Months
30%
32%
26%
34%
BL
Medication Use
(% on Medication)
% Receiving
Special
Education
Services
2 1.15, (P
.80)
35%
36%
38%
41%
1424
Months
% Receiving
Special
Education
Services
2 1.19 (P
.80)
MH Services
30%
35%
31%
34%
1424
Months
% Receiving
Specialty MH
Services
F 13.15
(P 0.0001)
MTA meds vs
not: P
.0001; Comb
vs MedMgt:
P .0013;
Beh vs CC:
P .44
F 0.73
(P .60)
F 1.31
(P .19)
30.43 (14.46)
37.47 (17.70)
25.69 (13.03)
24.00 (13.15)
ANOVA
1424
Months
Last Dose
(in MPHEquivalent
mg;
n 367)
Tx indicates treatment; Sx, symptoms; SSRS, Social Skills Rating System; PT, parentteacher; WIAT, Wechster Individual Achievement Test; MH, mental health; MPH, methylphenidate; BL, baseline;
ANCOVA, analysis of covariance
* The baseline and 24-month item means for the first 3 measures (SNAP and Social Skills) are for average of teacher and parent ratings.
Medication doses last reported during the 14- to 24-month follow-up period for those who took stimulants during that time Comb: 117; MedMgt: 106; Beh: 58; CC: 87), with other stimulants converted
to MPH equivalents (eg, 10 mg of d-amphetamine 20 mg of MPH). Mean doses were not significantly different by site.
Wald 2 was used to test the orthogonal contrasts in a logistic regression model for the percentages taking medication 14 24 months and in the 30 days before the 24-month assessment point.
The medication use percentages on medication show 1) proportion of participants BL on ADHD medication before study entry (not significantly different between groups at BL); 2) proportion of participants
with any ADHD medication use between 14 and 24 months (O2 77.7; P .001), and 3) the percentages using medication in the 30 days before the 24-month assessment.
Significance level for the mixed-effects regression models and ANCOVAs was set at P .01 to adjust for 5 analyses. Only the first 3 analyses have dual raters nested within subject. In the absence of dual
raters, a standard ANCOVA was performed.
As expected, site differences emerged on 2 measures as a result of differences in local populations. The lack of significant site-by-Tx interaction shows that these did not affect validity of the Tx comparisons.
Similarly, the lack of rater-by-Tx interactions shows that rater differences did not affect Tx comparisons. See text for more on rater effects.
# Because age at baseline was significantly different between MedMgt and Beh (Table 1), this analysis was repeated with age covaried as a check. It made no practical difference. Covarying age resulted in
treatment significance of P .0003 instead of P .0001 and rater significance of P .02, the same as without age covaried. The site effect (P .13), site x treatment interaction (P .16), and rater x treatment
interaction (P .32) remained clearly nonsignificant with age covaried.
F 1.56 (P .083)
0.83 (.70)
0.96 (.76)
1.04 (.81)
1.06 (.79)
24
Months
F 1.44 (P .13)
1.34 (.81)
1.42 (.86)
1.40 (.79)
1.42 (.80)
BL
Site*Tx
1.17 (.66)
1.21 (.68)
1.38 (.69)
1.40 (.68)
24
months
ODD Sx
(SNAP; Mean [SD]*;
n 524;
Lower Better)
F 4.35 (P .0007)
2.01 (.56)
2.06 (.53)
2.05 (.56)
2.02 (.58)
BL
ADHD Sx
(SNAP; Mean [SD]*;
n 526;
Lower Better)
Mixed-effects models
or ANCOVAs
2 or F (P value)
Site
F 2.9 (P .014)
Tx group
Comb
MedMgt
Beh
CC
TABLE 2.
TABLE 3.
Mixed-Effects Models and ANCOVAs of 24-Month Symptomatic and Functional Outcomes With Medication Use Status
During 14 24 Months Covaried (See Table 2 for means and SDs)
ADHD Sx (SNAP)
ODD Sx (SNAP)
Site
F 2.6 (P .025)
F 4.0 (P .001)
Site*Tx
F 1.48 (P .11)
F 1.57 (P .07)
Rater
F 5.75 (P .017)
F 2.58 (P .109)
Rater*Tx
F 1.46 (P .23)
F 1.01 (P .38)
Tx
F 5.13 (P .0017)
24-mo medication
main effect*
Orthogonal
contrasts
F 9.98 (P .0017)
F 3.43 (P .065)
Social Skills
(SSRS Total
P&T)
F 3.73
(P .003)
F 1.19
(P .27)
F 10.66
(P .001)
F 0.73
(P .54)
F 1.66
(P .18)
F 3.98
(P .047)
Negative Ineffective
Discipline Factor
F 0.74 (NS)
F 1.56 (P .08)
F 2.51 (P .06)
F 0.35 (P .56)
Reading
(WIAT)
F 1.37
(P .23)
F 1.32
(P .18)
F 1.02
(P .38)
F 0.00
(P .99)
Significance level for the mixed-effects models and ANCOVAs was set at P .01 to adjust for 5 analyses. NS indicates not significant.
The mixed-effects model was used when dual raters were nested within subject; otherwise, a standard ANCOVA was used.
* The interaction of medication status 14 24 mo with Tx group was not significant for any of the 5 clinical outcomes.
As expected, site differences emerged on 2 measures as a result of differences in local populations. The lack of significant site-by-Tx
interaction shows that these did not affect validity of the Tx comparisons. Similarly, the lack of rater-by-Tx interactions shows that the
significant rater differences did not affect Tx comparisons. See text for more on rater effects.
Pairwise differences computed only in the presence of a significant Tx main effect
(range: 0 3), a standard achieved by 88% of nonclinical classmates.4 After consideration of alternative
methods of measuring normalization, such as calculating movement toward the normal mean,25,26 we
selected this criterion as most clinically meaningful.
Although more stringent than moving half the distance toward the norm, this criterion was attained by
68% of Comb children at 14 months.
The proportion of children with SNAP item means
1.0 (near normalization4 or excellent responders)
at 24 months was 48%, 37%, 32%, and 28%, for Comb,
MedMgt, Beh, and CC, respectively. In a logistic
regression, the medication algorithm contrast
(Comb/MedMgt vs Beh/CC) for these percentages
was significant, but the other 2 contrasts were not.
More in-depth analyses of normalization will be explored in a subsequent paper.
DISCUSSION
continuing to do well after stopping intensive MedMgt, whereas others deteriorated. An obvious partial
explanation for such individual differences in trajectory might come from some participants stopping
medication, a possibility explored in the companion
paper. Because medication use does not seem to
explain all of the continuing benefit, we should note
that not only did the MTAs 0- to 14-month medication approach for the Comb and MedMgt groups
result in higher medication doses by 14 months than
CC-treated participants (31.2 and 37.7 mg vs 22.8
mg/day in methylphenidate equivalents, respectively),1,2 but also during its implementation, the medication algorithm included an initial double-blind,
placebo-controlled titration (a method to yield optimal symptom reduction and minimal side effects,
tailored for each child); performed sequential testing
of other medications if the child did not respond well
to methylphenidate; provided monthly half-hour office visits with the pharmacotherapist to review caregiver concerns, evaluate progress, and provide advice and support; communicated regularly with the
childs teachers via monthly telephone calls by the
pharmacotherapist; and readjusted medications if
the child was not doing well. Apart from the higher
doses produced in Comb and MedMgt participants
by the 14-month treatment endpoint, these activities
are likely also to have offered supportive benefits to
families. Once these nonspecific supports were withdrawn after the MTA treatments ceased, some loss of
benefit need not be surprising. Our analyses controlling for 14- to 24-month medication dose do not (and
cannot) take these factors into account. Such nonspecific but potentially therapeutic influences can be
appreciated visually by inspecting the differences
between the 14- and 24-month endpoints (ie, loss of
benefit) for Comb and MedMgt participants, a topic
of the accompanying report.28 Nonetheless, research
in other fields suggests that such factors can and do
function as a type of social support, even providing
an incremental part of the therapeutic benefit of an
efficacious medication program (via the therapeutic
alliance with the physician).29 In sum, these findings
might suggest that some children and families receive maximum medication benefit only when it is
accompanied by fairly intensive support and regular
contact with their doctor and/or a behavior therapist, whereas others may continue to benefit even
after this support is withdrawn.
At follow-up, MedMgt participants dose levels (in
methylphenidate equivalents) were significantly
higher than in Comb participants (Table 2). These
interesting results suggest the possibility that early
Comb interventions might allow reducing overall
medication requirements during later periods, consistent with findings that others have reported.30
This deserves future study as a possible way to keep
doses lower to avoid side effects.
Results should also be understood in the context of
the actual medications taken by participants in each
of the groups over the course of the study. Although
level of use of medication differed as a result of the
MTA medication algorithm versus no structured approach in the Beh and CC groups, group differences
ARTICLES
759
REFERENCES
1. MTA Cooperative Group. A 14-month randomized clinical trial of
treatment strategies for attention-deficit/hyperactivity disorder. Arch
Gen Psychiatry. 1999;56:10731086
2. MTA Cooperative Group. Moderators and mediators of treatment response for children with attention-deficit /hyperactivity disorder. Arch
Gen Psychiatry. 1999;56:1088 1096
3. Conners CK, Epstein J, March JS, et al. Multimodal treatment of ADHD
(MTA): an overall measure of treatment outcome. J Am Acad Child
Adolesc Psychiatry. 2001;40:159 167
4. Swanson JM, Kraemer HC, Hinshaw SP, et al. Clinical relevance of the
primary findings of the MTA: success rates based on severity of symptoms at the end of treatment. J Am Acad Child Adolesc Psychiatry. 2001;
40:168 179
5. Hinshaw SP, Owens EB, Wells KC, et al. Family processes and treatment outcome in the MTA: negative/ineffective parenting practices in
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
Many doctors and the drug industry promoted use of hormone therapy a few
years ago despite the lack of evidence from clinical trials. But now experts say that
experience has taught doctors and society a lesson in the need to exert extreme
caution in introducing new therapies.
Altman LK. New study links hormones to cancer risk. New York Times. August 8, 2003
Submitted by Student
ARTICLES
761
References
This article cites 26 articles, 6 of which you can access for free
at:
http://www.pediatrics.org/cgi/content/full/113/4/754#BIBL
Citations
Subspecialty Collections
Reprints