Articles

Symptom clusters in adolescent depression and differential

response to treatment: a secondary analysis of the
Treatment for Adolescents with Depression Study
randomised trial
Julia Bondar, Arthur Caye, Adam M Chekroud, Christian Kieling

Summary
Background Better understanding of the heterogeneity of treatment responses could help to improve care for Lancet Psychiatry 2020;
adolescents with depression. We analysed data from a clinical trial to assess whether specific symptom clusters 7: 337–43

responded differently to various treatments. See Comment page 295

Department of Psychiatry,
School of Medicine,
Methods For this secondary analysis, we used data from the Treatment for Adolescents with Depression Study (TADS),
Universidade Federal do
in which 439 US adolescents aged 12–17 with a DSM-IV diagnosis of major depressive disorder and a minimum Rio Grande do Sul,
score of 45 on the Children’s Depression Rating Scale-Revised (CDRS-R) were randomly assigned (1:1:1:1) to treatment Porto Alegre, Brazil (J Bondar,
with fluoxetine, cognitive behavioural therapy (CBT), fluoxetine plus CBT, or pill placebo. Our analysis focuses on the A Caye MD, C Kieling MD); Child
& Adolescent Psychiatry
acute phase of the trial (ie, the first 12 weeks). Groups of co-occurring symptoms were established by clustering scores
Division, Hospital de Clínicas de
for each CDRS-R item at baseline with Ward’s method, with Euclidean distances for hierarchical agglomerative Porto Alegre, Brazil,
clustering. We then used a linear mixed-effects model to investigate the relationship between symptom clusters and Porto Alegre, Brazil (J Bondar,
treatment efficacy, with the sum of symptom scores within each cluster as the dependent measure. As fixed effects, A Caye, C Kieling); Spring
Health, New York, NY, USA
we entered cluster, time, and treatment assignment, with all two-way and three-way interactions, into the model. The
(A M Chekroud PhD); and
random effect providing better fit was established to be a by-subject random slope for cluster based on improvement Department of Psychiatry, Yale
in the Schwarz-Bayesian information criterion. University, New Haven, CT, USA
(A M Chekroud)
Outcomes We identified two symptom clusters: cluster 1 comprised depressed mood, difficulty having fun, irritability, Correspondence to:
Dr Christian Kieling, Child &
social withdrawal, sleep disturbance, impaired schoolwork, excessive fatigue, and low self-esteem, and cluster 2
Adolescent Psychiatry Division,
comprised increased appetite, physical complaints, excessive weeping, decreased appetite, excessive guilt, morbid Hospital de Clínicas de Porto
ideation, and suicidal ideation. For cluster 1 symptoms, CDRS-R scores were reduced by 5·8 points (95% CI 2·8–8·9) Alegre, Rua Ramiro Barcelos
in adolescents treated with fluoxetine plus CBT, and by 4·1 points (1·1–7·1) in those treated with fluoxetine, compared 2350 – 400N, Porto Alegre,
90035-903, Brazil
with those given placebo. For cluster 2 symptoms, no significant differences in improvements in CDRS-R scores were
ckieling@ufrgs.br
detected between the active treatment and placebo groups.

Interpretation Response to fluoxetine and CBT among adolescents with depression is heterogeneous. Clinicians
should consider clinical profile when selecting therapeutic modality. The contrast in response patterns between
symptom clusters could provide opportunities to improve treatment efficacy by gearing the development of new
therapies towards the resolution of specific symptoms.

Funding Conselho Nacional de Desenvolvimento Científico e Tecnológico.

Copyright © 2020 Elsevier Ltd. All rights reserved.

Introduction symptom scales, or meeting global improvement criteria)

Treatment of depression in adolescents can be has investigated differences in the proportion of patients
challenging. In clinical trials of many therapies, the who respond to treatment and looked for traits that either
frequency of placebo responses is high, and effect sizes predicted improvement across treatments or modified
are small.1 Choice of treatment is thus still largely based the benefit of one treatment relative to the rest.3 Another
on trial and error: which patients derive the greatest study, in which classification of patients into latent
improvement from each available treatment modality, response groups was the outcome, identified several
and under which circumstances, are not known. There is clinical characteristics that were more or less likely to be
no one ideal way to answer these questions, and several linked with distinct trajectories of improve­ment.4 These
research methods have been proposed to investigate studies have helped to identify predictors and moderators
demographic, clinical, and biological markers that affect of therapeutic success in depression.
patient outcomes in adolescent depression.2 A study of A parallel strategy that might be useful in this scenario
total-severity-based measures (such as sum scores on is to probe symptom-based outcomes. Chekroud and

www.thelancet.com/psychiatry Vol 7 April 2020 337

 Articles

Research in context
Evidence before this study Added value of this study
We last searched PubMed with the terms “depression”, In our secondary analysis of data from the Treatment for
“adolescent”, “youth”, “children”, “symptom”, “trajectories”, Adolescents with Depression Study, we identified in an
“clusters”, “symptom profile”, “predictor”, and “moderator” for adolescent sample a symptom cluster that in previous studies
studies published in any language up to Nov 10, 2019, in which in adults was responsible for most improvement of
response to treatment by individual symptoms, symptom depression with treatment. The cluster is composed of
trajectories, symptom groups, and clinical profiles more depressed mood, difficulty having fun, irritability, social
generally was investigated in patients with a primary diagnosis withdrawal, sleep disturbance, impaired schoolwork,
of depression. We found that evidence for patient excessive fatigue, and low self-esteem. A second cluster
characteristics associated with improved responses to (comprising increased appetite, physical complaints,
treatments for depression was available in both adult and excessive weeping, decreased appetite, excessive guilt,
adolescent samples. However, the data were largely for morbid ideation, and suicidal ideation) that is less responsive
total-severity-based measures (ie, sum scores on symptom to treatment was also identified. In this study, unlike the
scales). Multiple factors have been identified as predictors of adult studies of symptom-based outcomes, an arm of
response to treatment through these studies, yet the therapeutic treatment with psychotherapy was included. The
modality most suited for each patient could not be clearly improvement trajectories differed between clusters for
established. In adolescent studies, low cognitive distortions, treatment with fluoxetine with or without cognitive
high family income, and increased depression severity at behavioural therapy, but not for treatment with cognitive
baseline increased the benefit of a type of treatment over the behavioural therapy alone or placebo.
others, but even increased depression severity, arguably the
Implications of all the available evidence
factor most often associated with response prediction,
In samples of both adults and adolescents with depression,
was questioned as a moderator of treatment response in a
a cluster of symptoms can be identified that is responsive to
meta-analysis. Thus, personalisation of treatment choice
pharmacotherapy and combined treatment with
remains difficult in adolescent patients. In adult samples,
pharmacotherapy and cognitive behavioural therapy, and tends
symptom-based outcomes have also been examined for their
to show greater benefit from pharmacotherapy compared with
potential to orient treatment choice. Some studies identified
other symptom clusters. Clinicians and researchers should take
fairly consistent groups of depressive symptoms in adults that
heterogeneity of symptoms into account when assessing
respond to pharmacotherapy differently and thus provide an
individual patients, and also when developing new treatment
opportunity to individualise treatment choice according to
interventions to improve the lives of adolescents with depression.
symptom profile.

colleagues5 used data from trials in adult depression to Methods

develop a symptom-cluster-based approach to prediction Study design and participants
of depression outcomes. They clustered commonly used We used the data from the publicly available TADS
symptom checklists into groups of symptoms with database. TADS was a randomised placebo-controlled
similar intensity scores over time and examined whether clinical trial at 13 academic and community clinics in the
the clusters responded differently to various types and USA. 439 outpatients aged 12–17 years with a DSM-IV
doses of antidepressants.5 This approach has not yet diagnosis of major depressive disorder and a minimum
been used to analyse data for adolescent depression. score of 45 on the Children’s Depression Rating Scale-
Depression operates differently (with differences in Revised (CDRS-R) were included. Patients with a
efficacy of treat­
ments and adverse events from treat­ diagnosis of bipolar disorder, severe conduct disorder,
ments) in adolescents and adults. This difference is pervasive developmental disorder, or thought disorder, or
exemplified by the increased risk of suicidality associated with current substance abuse or dependence (according
with the use of antidepressants specifically in young to DSM-IV criteria), were excluded, but those with
people (aged up to 24 years) with depression.1,6–8 In concurrent anxiety disorder, obsessive-compulsive
this study, we analysed data from the Treatment for disorder, disruptive behavior, substance use, and ADHD
Adolescents with Depression Study (TADS), the largest were included. Participants were recruited through
randomised placebo-controlled trial so far in which referral from primary care or mental health professionals
the efficacy of antidepressants or psychotherapy, or and self-referral between spring, 2000, and summer,
both, were compared in adolescents with depression, 2003, and followed up for 21 months. The trial was
to investigate whether data-driven symptom clusters advertised in clinics; via paid and public service
respond differently to specific treatments in this age advertisements in newspapers, on the radio, and on
group. television; by primary care physicians and other mental

338 www.thelancet.com/psychiatry Vol 7 April 2020

 Articles

health clinicians; and in schools and juvenile justice define the number of clusters by identifying the moment
facilities. when cluster formation produces groups that are either
Our analysis focuses on the acute phase of the trial particularly dissimilar or useless for the research
(ie, the first 12 weeks). Participants’ symptoms were question.14 We did sensitivity analyses, in which we varied
assessed with the CDRS-R at baseline. Participants were the clustering method, item rescaling method, and study
randomly assigned in a 1:1:1:1 ratio to fluoxetine timepoint used. The sensitivity analyses are presented as
(10–40 mg perday), cognitive behavioural therapy (CBT), tanglegrams, allowing a visual comparison between the
fluoxetine (10–40mg/day) plus CBT, or a placebo pill. dendrogram from the main analysis and each of its
CDRS-R score was the primary outcome, and was variations (appendix pp 2–4).
measured at 6 weeks and 12 weeks by a blinded evaluator. A linear mixed-effects model was used to investigate
Additional details about the trial and its results have been the relationship between symptom clusters and treat­
described elsewhere.9–11 ment efficacy, with the sum of symptom scores within
The Duke University Medical Center (Durham, NC, each cluster as the dependent measure. As fixed effects,
USA) and the institutional review boards at each site we entered cluster (which was derived from clustering
approved and monitored the protocol; TADS was done at baseline), time (log of days since randomisation),
monitored quarterly by the data safety and monitoring and treatment assignment, with all two-way and
board of the US National Institute of Mental Health. All three-way interactions, into the model. The random
patients and at least one parent per participant provided effect providing better fit was established to be a by-
written informed consent. No additional ethics approval subject random slope for cluster based on improve­ment
or informed consent was required for this secondary in the Schwarz-Bayesian information criterion. Further
analysis because we used publicly available data. details on model development are in the appendix (p 6).
Because of concerns that an increase in suicidality as a
Statistical analysis result of SSRI treatment could bias results, the appendix
The CDRS-R is a semi-structured instrument that is (p 9) shows the overall decrease in both suicidality items.
completed by clinicians based on interviews with the The results provided by the linear model were difficult
patient and their primary caregiver. It comprises to interpret in a clinical context. We therefore calculated
17 items, each of which corresponds to a symptom that effect sizes that reflect the reduction in the number of
is usually graded in intensity from 1 to 7. For our CDRS-R points produced by each treatment in each
analysis, we rescaled two items that are usually graded cluster at the end of the trial (at 12 weeks) by multi­plying
in intensity from 1 to 5 so that they were graded from slope contrasts by the natural log of treatment duration.15
1 to 7 to avoid imbalance. We also excluded the last three All analyses were done in R (version 3.4.2) and all
items on the CDRS-R, which rely on different rating graphs were plotted with the dendextend and ggplot2
procedures (they are scored by the clinician alone R packages.13,14,16 We fitted models with the lme4 package.17
based on the adolescent’s non-verbal behaviour during We report two-sided p-values. The code we used and a
the interview—a scoring method that could induce
increased shared variance and potentially cause the Sleep disturbance
three items to cluster for reasons other than similar
Social withdrawal
symptom behaviour, thereby poten­ tially confounding
Impaired schoolwork
results).12 We also did a sensitivity analysis including all
Excessive fatigue
Cluster 1

symptoms irrespective of scoring criteria (appendix p 1). See Online for appendix
Because the CDRS-R item about appetite disturbances Irritability

is scored equally irrespective of the type of appetite Low self-esteem

problem, we added a supplemental item related to Difficulty having fun
whether the teenager had increased or decreased Depressed feelings
appetite, and excluded patients for whom such data Increased appetite
were unavailable. Physical complaints
Groups of co-occurring symptoms were established by Excessive weeping
Cluster 2

clustering scores for each CDRS-R item at baseline—that Decreased appetite

is, symptoms were classified into clusters according to
Excessive guilt
similarity in baseline score across the entire sample.
Morbid ideation
We used Ward’s method with Euclidean distances for
Suicidal ideation
hierarchical agglomerative clustering.13 Unlike some
other clustering methods, hierarchical agglomerative Figure 1: Dendrogram of clustering of Children’s Depression Rating Scale-
clustering does not require a prespecified number of Revised symptoms at baseline
In the clustering procedure, individual symptoms are sequentially grouped
clusters. Rather, symptoms are progressively aggregated
according to similarity of mean scores. The longer the lines between the joining
according to the distance metric and linkage criterion of individual symptoms and of groups of symptoms, the greater the difference
selected, until a cluster is formed. Investigators then between their mean scores.

www.thelancet.com/psychiatry Vol 7 April 2020 339

 Articles

A model dataset are provided in the appendix (pp 12–20).

6 Treatment group TADS is registered with ClinicalTrials.gov, NCT00006286.
Placebo
CBT
Fluoxetine
Role of the funding source
Fluoxetine plus CBT The funder of the study had no role in study design, data
collection, data analysis, data interpretation, or writing of
the report. The corresponding author had full access to
Mean item score

4 all the data in the study and had final responsibility for
the decision to submit for publication.

Results
At baseline, 426 of the 439 patients randomly assigned in
TADS had complete CDRS-R data. 231 (54%) of 426 were
2
female, and the mean age was 14·6 years (SD 1·5). Mean
total corrected CDRS-R scores, which were calculated by
rescaling sleep and appetite items and excluding the
clinician-rated symptoms of the CDRS-R, were similar
B
6
across treatment groups at baseline: 56·1 (SD 9·7) in
the 106 patients in the placebo group, 54·3 (9·1) in the
109 patients in the CBT group, 54·6 (9·6) in the
106 patients in the fluoxetine group, and 56·6 (10·7) in
the 105 patients in the combined treatment group.
Further demographic and clinical characteristics have
Mean item score

4 been described previously.8

The clustering procedure showed that the Euclidean
distance at which symptoms were merged into clusters
tended to be small, except for the final two clusters, for
which the distance was much greater. This finding
suggested that a two-cluster solution fitted the data well,
2
and therefore two symptom clusters (one including eight
symptoms and the other including seven) were derived
from the CDRS-R symptoms (figure 1). Results of
C sensitivity analyses of the clustering procedure in which
6 we used all CDRS-R symptoms, alternative aggregation
methods, and alternativetime points are in the appendix
(pp 1–4). The mixed-effects model showed significant
interactions between symptom cluster and fluoxetine
treatment (1·2 [SE 0·3]; p<0·0001), and between
symptom cluster and combined treatment with fluoxetine
Mean item score

4 plus CBT (1·0 [0·3]; p=0·0008), but not between

symptom cluster and CBT (0·4 [0·3]; p=0·18).
The trajectories of the total corrected CDRS-R score in
each treatment group followed an identical pattern to
those described in the original TADS study, in which the
highest improvement rates (measured as magnitude of
2
improvement in CDS-R scores) occurred in the combined
fluoxetine and CBT group and in the fluoxetine group.11
The CBT and placebo groups in TADS did not differ
0 4 8 12
Time (weeks) significantly from each other in terms of improvement
rate. We noted similar patterns for symptom cluster 1 in
Figure 2: Mean total corrected scores on the CDRS-R (A), and mean scores in our study, despite the higher mean item scores and the
symptom cluster 1 (B) and symptom cluster 2 (C) wider gap between the most and least efficacious
The total corrected score was obtained by rescaling sleep and appetite items and
excluding the clinician-rated items of the CDRS-R, such that this score is similar treatments in cluster 1 compared with the total corrected
to scores in the clusters. The shaded regions represent 95% CIs. scores (figure 2). By contrast, for cluster 2 symptoms,
CDRS-R=Children’s Depression Rating Scale-Revised. CBT=cognitive behavioural mean baseline scores were lower, trajectories were flatter,
therapy. and treatment effects became indistinguishable compared
with total corrected scores (figure 2). We identified no

340 www.thelancet.com/psychiatry Vol 7 April 2020

 Articles

benefits of active treatment compared with placebo for Cluster

cluster 2 symptoms (figure 2). Individual mean cluster 1
2
item scores per timepoint and all the model’s slopes are
in the appendix (pp 5–6). 5
To assess the clinical significance of these findings, we
calculated effect sizes in terms of the difference in score

Effect size
reduction between treatments during the first 12 weeks
of treatment. In cluster 1, the effect size was 4·1 CDRS-R 0
points (95% CI 1·1–7·1) for fluoxetine versus placebo and
5·8 CDRS-R points (2·8–8·9) for combined CBT plus
fluoxetine versus placebo (figure 3). For cluster 2, none of
the effect sizes obtained with active treatments were
−5
significant compared with placebo (figure 3). Effect sizes Fluoxetine plus Fluoxetine plus Fluoxetine Fluoxetine Fluoxetine plus CBT vs placebo
CBT vs placebo CBT vs CBT vs placebo vs CBT CBT vs fluoxetine
for all possible comparisons between treatments and
clusters are in the appendix (p 7). Figure 3: Between-group differences in improvements in CDRS-R scores at 12 weeks
Error bars represent 95% CIs. Improvement is expressed as a reduction in CDRS-R points. Effect sizes reflect the
Discussion number of additional CDRS-R points reduced by the first treatment compared with the second. Effect sizes are
given for each cluster and are derived from mixed-effects model estimates multiplied by treatment duration.
Our secondary analysis of data from TADS represents, to CDRS-R=Children’s Depression Rating Scale-Revised. CBT=cognitive behavioural therapy.
our knowledge, the first attempt to identify symptom
clusters associated with differential treatment responses
in adolescent depression. Two symptom clusters emerged showed that gender is neither a predictor nor a moderator
related to the CDRS-R. The first cluster included of treatment response,20 another cast doubt on one of the
symptoms for which patients had higher mean baseline most consistently reported findings previously (ie, that
scores than in cluster 2, including cardinal symptoms of baseline severity is a predictor of response in
major depressive disorder, that tended to improve more depression),21 and the third showed differential response
from treatment with fluoxetine with or without CBT than to treatment based on symptom clusters.5
from treatment with placebo or CBT only. The second Academic literature on depressive symptom clusters in
cluster included symptoms for which patients had lower adults describes a set that includes core emotional or
baseline scores than in cluster 1 (including suicidal general depression items (eg, sad mood, anhedonia, low
symptoms), which did not improve significantly with any self-worth). This cluster had a better response to pharma­
active treatments compared with placebo. Our results cotherapy than other clusters in adults.5,22 However, the
confirm that response to treatment is heterogeneous and effect of psychotherapy on this cluster has not been
suggest the clinical profile for which a given therapeutic assessed. Several symptoms in symptom cluster 1 in our
modality might be more or less beneficial. The contrast in study are analogous to symptoms in the adult core
response patterns between symptom clusters could emotional cluster: depressed feelings, difficulty having
provide unique opportunities to improve treatment fun, low self-esteem, and impaired schoolwork from the
efficacy by gearing the development of new therapies CDRS-R—which are similar to mood (sad), loss of
towards the resolution of specific symptoms. interest, feelings of worthlessness, and concentration or
Most efforts to identify predictors and moderators of decision making from the QIDS. This core emotional and
treatment response have focused on adult depression. general depression cluster from the adult literature
One study18 has shown that improvement of a symptom shares several symptoms with the adolescent cluster 1
dimension from the Quick Inventory of Depressive found in our secondary analysis; both clusters decrease in
Symptoms (QIDS) at week 2 can predict remission under magnitude to a greater extent than the other clusters
citalopram. This dimension is composed of sad mood, (atypical for adults and cluster 2 for adolescents) through
negative self-view, low energy, restlessness, and feeling treatment, meaning they exhibit better response to
slowed down. Another study19 used factor analysis to treatment.
reclassify symptoms from the Hamilton Depression Research about the grouping of symptoms in the
Rating Scale (HAM-D), Montgomery Asberg Depression second cluster is less consistent. An analysis of QIDS5
Rating Scale (MADRS), and Beck Depression Inventory clustered suicidal ideation, psychomotor agitation,
(BDI) into three categories (mood, cognitive, and neuro­ psychomotor slowing, and hypersomnia, whereas the
vegetative symptoms) and found mood and cognitive same analysis of the HAM-D suggested that suicidal
symptoms improved more in patients treated with ideation, psychomotor agitation, psychomotor slowing,
escitalopram, whereas neurovegetative symptoms hypochondriasis, and reduced libido were atypical
improved more in patients treated with nortryptiline. symptoms. A meta-analysis of factor structures of the
Three studies5,20,21 stand out because they included HAM-D23 separated the same symptoms into three
multiple samples, the best statistical methods were used, different factors. None of these clusters perfectly match
and individual patient data meta-analysis was done. One the CDRS-R cluster 2 that we identified in this study,

www.thelancet.com/psychiatry Vol 7 April 2020 341

 Articles

partly because some symptoms were unique to each to all treatment regimens.3 A review25 of predictors and
scale.24 Still, suicidal ideation was a constant, and despite moderators of psychotherapy in child and adolescent
content differences, both adult atypical clusters and the depression also showed that increased baseline symptom
symptom cluster 2 in adolescents were the poorest severity and comorbid anxiety were associated with
responders among the groups of symptoms.5 decreased response. However, the authors cautioned that
Our findings can be generalised only to adolescents the associations identified were present in few studies,
with moderate-to-severe depression who were from a which usually had small samples, and the review omitted
high-income country, being treated as outpatients, and important information about statistical methods. Thus,
had not previously taken fluoxetine, and are subject to all understanding of predictors and moderators of treatment
the TADS trial’s biases. Whether our findings can be responses in adolescent depression is limited. All previous
replicated in different populations is unclear. Responses findings came from studies in which total-severity-based
to CBT were particularly low in TADS compared with measures were used, and that assessed mostly demo­
findings in other studies. The control method used in graphic characteristics. Now that evidence is available for
TADS (ie, placebo) might have contributed to this finding. symptom-based outcomes too, high-powered studies
A meta-analysis25 of moderators of the effects of psycho­ should be developed to establish the extent to which these
therapy on depression in children and adolescents aged markers of response co-occur and interact and which
4–18 years showed that psychotherapy was associated with intervention, if any, is useful to each individual patient.
larger effect sizes when compared with inactive control Contributors
conditions (eg, waitlist) than when compared with an All authors conceived the study and interpreted the results. JB, AC, and
active control (eg, usual care, placebo medication). Future AMC analysed the data. JB and CK wrote the Article, which was revised
by AC and AMC.
studies should account for such limitations and the
difficulties posed by the low amount of content overlap Declaration of interests
AMC holds equity in Spring Care, Fitbit, and UnitedHealthcare, is the
between common depressive scales24—our data are in lead inventor on three patent submissions relating to treatment for
agreement with this finding. Specifically, the appendix major depressive disorder (US Patent and Trademark Office docket
(p 8) contains a table in which we compare the same number Y0087.70116US00 and provisional application numbers
symptoms across three depression scales (CDRS-R, 62/491 660 and 62/629 041), has done paid consultancy for Fortress
Biotech about antidepressant drug development, and provides unpaid
HAM-D, and QIDS) to see if they are always contained in advisory services to health care technology startups. All other authors
analogous clusters. It is therefore important to reproduce declare no competing interests.
results on other scales to gain a clearer understanding of Acknowledgments
the factors that affect response and remission. One might The authors are extremely grateful to the participants who took part in
also consider that items on these scales are differentially the TADS trial, as well as to the original study’s investigators. They also
sensitive to change, which might affect results, although thank Claudia Buchweitz for her input to an earlier version of the
manuscript. Data and research tools (dataset identified 116260) used to
baseline differences were accounted for in our mixed prepare this Article were obtained and analysed from the controlled
model.26 access datasets distributed from the US National Database for Clinical
Our study had several strengths. We used a sample Trials, a collaborative informatics system created and supported by the
US National Institute of Mental Health to provide a national resource to
from a high-quality randomised contolled trial in which
support and accelerate discovery related to clinical trial research in
response to various treatment types was closely monitored. mental health. This study is based on data from TADS, which was
Another strength was the performance of data-driven supported by a grant from the US National Institute of Mental Health to
symptom grouping, a complementary technique to the Duke University Medical Center (Durham, NC, USA). Lilly provided
fluoxetine and matched placebo under an independent educational grant
oft-used theory-driven or DSM-driven symptom
to Duke University (Durham, NC, USA). JB received a Research
aggregation approaches. Initiation Scholarship from Conselho Nacional de Desenvolvimento
To understand the potential role of symptom presen­ Científico e Tecnológico (CNPq; a Brazilian public funding agency).
tation in the prediction of treatment outcome, available CK is a CNPq researcher and a UK Academy of Medical Sciences
Newton Advanced Fellow.
evidence for all other factors that seem to affect response
in adolescents should be reviewed. Curry and colleagues,3 References
1 Cipriani A, Zhou X, Del Giovane C, et al. Comparative efficacy and
who also worked with the TADS sample, found three tolerability of antidepressants for major depressive disorder in
moderators of treatment among various baseline children and adolescents: a network meta-analysis. Lancet 2016;
demographic and clinical characteristics. High annual 388: 881–90.
2 Perlis RH. Abandoning personalization to get to precision in the
family income increased the benefit of CBT compared pharmacotherapy of depression. World Psychiatry 2016; 15: 228–35.
with placebo, and decreased that of fluoxetine compared 3 Curry J, Rohde P, Simons A, et al. Predictors and moderators of
with placebo.3 Both high severity and low cognitive acute outcome in the Treatment for Adolescents with Depression
Study (TADS). J Am Acad Child Adolesc Psychiatry 2006; 45: 1427–39.
distortions scores at baseline increased the benefit of
4 Scott K, Lewis CC, Marti CN. Trajectories of symptom change in the
fluoxetine such that fluoxetine alone was equivalent to Treatment for Adolescents with Depression Study.
combined treatment with fluoxetine and CBT.3 The J Am Acad Child Adolesc Psychiatry 2019; 58: 329–18.
presence of other severity markers, comorbid anxiety 5 Chekroud AM, Gueorguieva R, Krumholz HM, Trivedi MH,
Krystal JH, McCarthy G. Reevaluating the efficacy and predictability
disorder, hopelessness, not expecting much improvement of antidepressant treatments: a symptom clustering approach.
from therapy, and increased age predicted worse response JAMA Psychiatry 2017; 74: 370–78.

342 www.thelancet.com/psychiatry Vol 7 April 2020

 Articles

6 Fergusson D, Doucette S, Glass KC, et al. Association between 18 Sakurai H, Uchida H, Abe T, et al. Trajectories of individual
suicide attempts and selective serotonin reuptake inhibitors: symptoms in remitters versus non-remitters with depression.
systematic review of randomised controlled trials BMJ 2005; J Affect Disord 2013; 151: 501–13.
330: 396. 19 Uher R, Maier W, Hauser J, et al. Differential efficacy of escitalopram
7 Cipriani A, Furukawa TA, Salanti G, et al. Comparative efficacy and and nortriptyline on dimensional measures of depression.
acceptability of 12 new-generation antidepressants: a multiple- Br J Psychiatry 2009; 194: 252–59.
treatments meta-analysis. Lancet 2009; 373: 746–58. 20 Cuijpers P, Weitz E, Twisk J, et al. Gender as predictor and moderator
8 Hazell P, Mirzaie M. Tricyclic drugs for depression in children and of outcome in cognitive behavior therapy and pharmacotherapy for
adolescents. Cochrane Database Syst Rev 2013; 6: CD002317. adult depression: an “individual patient data” meta-analysis.
9 March J, Silva S, Petrycki S, et al. Treatment for Adolescents with Depress Anxiety 2014; 31: 941–51.
Depression Study (TADS): rationale, design, and methods. 21 Weitz E, Hollon S, Twisk J, et al. Baseline depression severity as
J Am Acad Child Adolesc Psychiatry 2003;42: 531–42. moderator of depression outcomes between cognitive behavioral
10 Silva S, Petrycki S, Curry J, Wells K, Fairbank J. The Treatment for therapy vs pharmacotherapy. JAMA Psychiatry 2015; 72: 1102–09.
Adolescents with Depression Study (TADS): demographic and 22 Tokuoka H, Takahashi H, Ozeki A, et al. Trajectories of depression
clinical characteristics. J Am Acad Child Adolesc Psychiatry 2005; symptom improvement and associated predictor analysis:
44: 28–40. an analysis of duloxetine in double-blind placebo-controlled trials.
11 March J, Silva S, Petrycki S, et al. Fluoxetine, cognitive-behavioral J Affect Disord 2016; 196: 171–80.
therapy, and their combination for adolescents with depression: 23 Shafer AB. Meta-analysis of the factor structures of four depression
Treatment for Adolescents with Depression Study (TADS) questionnaires: Beck, CES-D, Hamilton, and Zung. J Clin Psychol
randomized controlled trial. JAMA 2004; 292: 807–20. 2006; 62: 123–46.
12 Ameena I, Bernstein I, Trivedi M, Mayes T, Kennard B, Emslie G. 24 Fried EI. The 52 symptoms of major depression: lack of content
Childhood depression subscales using repeated sessions on overlap among seven common depression scales. J Affect Disord
Children’s Depression Rating Scale-Revised (CDRS-R) scores. 2017; 208: 191–97.
J Child Adolesc Psychopharmacol 2014; 24: 318–24. 25 Eckshtain D, Kuppens S, Ugueto A, et al. Meta-analysis: 13-year
13 Murtagh F, Legendre P. Ward’s hierarchical agglomerative clustering follow-up of psychotherapy effects on youth depression.
method: which algorithms implement Ward’s criterion? J Classif 2014; J Am Acad Child Adolesc Psychiatry 2020; 59: 45–63.
313: 274–95. 26 Hieronymus F, Emilsson JF, Nilsson S, Eriksson E.
14 Norušis MJ. Cluster analysis. PASW statistics 18 advanced statistical Consistent superiority of selective serotonin reuptake inhibitors
procedures companion. Upper Saddle River: Prentice Hall, over placebo in reducing depressed mood in patients with major
2010: 361–391. depression. Mol Psychiatry 2016; 21: 523–30.
15 Rosenthal R, Rosnow RD, Rubin D. Contrasts and effect sizes in 27 Nilsen TS, Eisemann M, Kvernmo S. Predictors and moderators of
behavioral research. New York: Cambridge University Press, 2000. outcome in child and adolescent anxiety and depression:
16 Galili T. dendextend: an R package for visualizing, adjusting, and a systematic review of psychological treatment studies.
comparing trees of hierarchical clustering. Bioinformatics 2015; Eur Child Adolesc Psychiatry 2013; 22: 69–87.
31: 3718–20.
17 Bates D, Mächler M, Bolker BM, Walker SC. Fitting linear mixed-
effects models using lme4. J Stat Soft 2015; 67: 1–48.

www.thelancet.com/psychiatry Vol 7 April 2020 343