Social Psychiatry and Psychiatric Epidemiology

https://doi.org/10.1007/s00127-018-1574-2

REVIEW

The study of effect moderation in youth suicide-prevention studies

Rashelle J. Musci1 · Hadi Kharrazi1 · Renee F. Wilson1 · Ryoko Susukida1 · Fardad Gharghabi1 · Allen Zhang1 ·
Lawrence Wissow1 · Karen A. Robinson1 · Holly C. Wilcox1

Received: 22 February 2018 / Accepted: 30 July 2018

© This is a U.S. government work and its text is not subject to copyright protection in the United States; however, its text may be subject to foreign
copyright protection 2018

Abstract
Purpose  Suicide is now the second leading cause of death among persons between the ages of adolescents and emerging
adults and rates have increased despite more funding and broader implementation of youth suicide-prevention programs. A
systematic review was conducted focusing on identifying youth suicide-prevention studies within the United States. This
paper reports on the methods utilized for understanding possible moderators of suicide-prevention program outcomes.
Methods  We searched six databases from 1990 through August 2017 to identify studies of suicide-preventive interventions
among those under age 26 years. Two independent team members screened search results and sequentially extracted infor-
mation related to statistical methods of moderation analyses.
Results  69 articles were included in the systematic review of which only 17 (24.6%) explored treatment effect heterogeneity
using moderation analysis. The most commonly used analytic tool was regression with an interaction term. The modera-
tors studied included demographic characteristics such as gender and ethnicity as well as individual characteristics such as
traumatic stress exposure and multiple prior suicide attempts.
Conclusions  With a greater emphasis from the federal government and funding agencies on precision prevention, understand-
ing which prevention programs work for specific subgroups is essential. Only a small percentage of the reviewed articles
assessed moderation effects. This is a substantial research gap driven by sample size or other limitations which have impeded
the identification of intervention effect heterogeneity.

Keywords  Suicide prevention · Moderation methods · Systematic review

Context community). Universal suicide-prevention programs, how-

In 2015, suicide was the second leading cause of death
among people between the ages of 15–19 and 20–29 years
old [1]. According to research from the National Youth Risk
Behavior Surveillance system at the Centers for Disease
Control and Prevention (CDC) over one million high school
students are treated for a suicide attempt each year [2, 3].
According to the framework for prevention outlined by
a National Academy of Medicine (formerly Institute of
Medicine) report [3], “universal suicide prevention pro-
grams” target broad populations regardless of risk and are
typically provided to a whole population (e.g., a school or
* Rashelle J. Musci
rmusci1@jhu.edu
1
Johns Hopkins Bloomberg School of Public Health, 624 N.
Broadway Room 831, Baltimore, MD 21205, USA
ever, are hypothesized to have different effects across sub-
sets of the population thus requiring further evaluation and
refinement via statistical methods such as moderation analy-
sis. “Selective prevention programs” target subgroups of the
general population that are at higher risk for suicide, such
as those with psychiatric symptoms or disorders. “Indicated
suicide prevention programs” are directed at those who have
already experienced suicidal ideation or attempts. Modera-
tion can be studied in each type of prevention program, but
suicide and suicide attempts are relatively rare outcomes and
thus few studies are adequately powered to explore subgroup
variation and test moderators within a given analysis.
In the design and analysis of prevention studies, treat-
ment effect heterogeneity is often addressed by identifying
and controlling for moderation effects. Moderation modeling
focuses on the prediction of a dependent variable (e.g., sui-
cide attempt), from an independent variable (e.g., treatment
or prevention program), which differs across levels of a third

13
Vol.:(0123456789)

variable (e.g., severity of psychiatric symptoms) [4, 5]. Mod-
erator variables may influence the strength or direction of the
relationship between the independent variable (the interven-
tion) and the dependent variable (outcome). In this context,
a moderator is a variable/construct that alters the strength or
direction of the relationship between and intervention imple-
mented and an outcome. Thus, moderator variables inform
for whom and under what circumstances interventions have
an impact on suicide-related outcomes. For example, an
intervention could be equally beneficial for boys and girls or
could be effective only in boys, this has implications for tar-
geting limited intervention resources to those who are most
likely to benefit. It could be that those with highest baseline
depression symptoms benefit most from a suicide-prevention
program because they have more room for improvement than
those with lower symptoms of depression. Moderators can
be based on individual characteristics such as age, sex, race,
ethnicity, socioeconomic status, mental health characteris-
tics, or can be based on intervention characteristics such as
whether there was family involvement. Moderators can also
be situational such as the context in which it was imple-
mented (urban/rural; in person, online or telephone) or the
quality of implementation.
Treatment effect heterogeneity is an established concept
in mental health interventions, and an emphasis has been
placed on understanding and modeling the heterogeneity
in suicide prevention efforts rather than just exploring the
main effect of the intervention [6–9]. Not all participants
respond similarly to prevention interventions, and therefore,
an understanding of treatment effect heterogeneity within
the context of suicide prevention programs is essential [6,
10]. Identifying moderators of intervention impact helps to
elucidate which individuals or communities might benefit
from particular preventive programs and under what circum-
stances. Identifying moderators can improve tailoring or tar-
geting interventions to subpopulations (e.g., demographic
or disease groups) and can determine whether interventions
require adaptation for specific high-risk communities (e.g.,
American–Indian populations, substance abuse patients).
In 2014, the National Action Alliance for Suicide Pre-
vention (NAASP), a public–private partnership advancing
the National Strategy for Suicide Prevention, developed a
suicide-prevention research agenda focusing on interven-
tions with the potential to reduce morbidity (attempts) and
mortality (deaths) by at least 20% in 5 years and at least
40% in 10 years [2]. The NAASP has identified understand-
ing existing differences in intervention effectiveness as a
research priority. The Prioritized Research Agenda for Sui-
cide Prevention [2] aimed to identify the state of the science
of suicide-prevention research and it was noted that media-
tors and moderators of interventions are not frequently the
focus.
13
Social Psychiatry and Psychiatric Epidemiology
Hypotheses about moderation are often a fundamental
piece of the theoretical framework that provide the basis for
suicide-prevention programs [11]. There does not exist, to
our knowledge, a systematic review identifying the use of
moderator analyses in suicide-prevention interventions. This
systematic review aims to identify and describe how studies
address moderation within a suicide-prevention context spe-
cifically by identifying which statistical methods were used
for understanding possible moderators in suicide-prevention
programs. We also discuss potential gaps, barriers and chal-
lenges to explore treatment effect heterogeneity in suicide-
prevention interventions.
Methods
Systematic review methods
We assessed all of the studies identified by Wilcox et al. [12]
for whether moderation analyses were conducted. The aim
of the broader systematic review by Wilcox et al. [12] was
to provide an objective description of the state of the science
on data-linkage strategies in suicide-prevention research.
Detailed methods of the broader systematic review, includ-
ing the key questions, analytic framework, search strategies,
inclusion criteria, and study data extraction and quality rat-
ing methods, are available in a separate report [12]. The
protocol was developed using a standardized process with
input from experts and the public.
Data sources and searches
We completed a systematic review of the published litera-
ture to identify youth suicide-prevention studies to identify
prevention studies and potentially linkable external data
systems with suicide outcomes. We searched PubMed, The
Cochrane Library, Campbell Collaboration Library of Sys-
tematic Reviews, Cumulative Index to Nursing and Allied
Health Literature (CINAHL), PsycINFO and Education
Resources Information Center (ERIC) for English language
articles published from January 1990 to December 2015.
An updated search was conducted through August 22, 2017.
Study selection
Two investigators independently reviewed titles and abstracts
first, and then full-text articles against pre-specified eligibil-
ity criteria. We included studies of humans between the ages
of 0 and 25 years, with one of the primary outcomes of
interest (suicide completion, attempt and ideation), which
had taken place in the US. We excluded studies reported
as meeting abstracts only and papers without original data.
We did not limit study inclusion by study population size or
Social Psychiatry and Psychiatric Epidemiology
study design. Studies published prior to 1990 were excluded
as, according to the NAASP, suicide prevention became a
priority in the United States in the mid-1990s with the pub-
lication by the US. Department of Health and Human Ser-
vices of the Report of the Secretary’s Task Force on Youth
Suicide in 1989 and the Surgeon General’s Call to Action
to Prevent Suicide in 1999 [13]. Disagreements about arti-
cle eligibility that could not be resolved by two reviewers
were resolved by the team. A majority of the 4000 articles
originally flagged to be included in the broader review were
excluded because the study did not take place in the United
States. Other reasons include not including an intervention,
having participants outside the proposed age range (ages
0–25 years), and not having original data [12].
Data extraction
Two team members sequentially extracted data on study
characteristics, participant characteristics, intervention char-
acteristics, and suicide outcomes. We also abstracted the
statistical tests used in the study, and analyses performed.
We extracted variables used or controlled for in the analyses.
Disagreements were resolved by consensus. Further, each
prevention program was classified into universal, selected
or indicated based on the National Academy of Medicine
Prevention framework [3].
Role of the funding source
The AHRQ funded the review [13], and a working group
convened by the NIH assisted in developing the review’s
scope and key questions. Neither AHRQ nor NIH had a role
in study selection, quality assessment, or synthesis. The
investigators are solely responsible for the content.
Results
The literature search identified 4,198 unique citations, of
which 69 articles were eligible for the broader review. Sev-
enteen studies (24.6%) explored the effect of moderators in
suicide prevention efforts; all moderation relationships were
stated as a priori hypotheses (3 of the 17 studies included
was captured during the updated literature search). Ten of
these studies used a form of regression, either linear or logis-
tic, depending on the outcome of interest, and interaction
terms to explore moderation (Table 1). 5 of the programs
were classified as universal, whereas 12 have been classified
as selected/indicated.
The most commonly used approach for moderation anal-
ysis among suicide prevention studies was general linear
modeling [14–19]. The second most commonly used method
for moderation analysis among suicide prevention studies
was hierarchical linear modeling, used by 4 of the 17 stud-
ies [20–23]. Hierarchical methods require the creation of an
interaction term to be included in the model. The next most
common method used in these suicide prevention studies
was mixed effects modeling, used in two of the included
studies [24, 25]. Other methodologies included Chi-square
tests [26] and multivariate analysis of variance (MANOVA)
[27]. Three studies used growth models [28–30]. One
study utilized interaction terms within a discrete time sur-
vival analysis framework [31]. Further, 50% of the stud-
ies reported some kind of effect size (adjusted odds ratios,
Cohen’s d, eta-squared, f squared), which is useful for future
meta-analysis.
The most common moderator explored was gender, which
was explored in 7 of the 17 studies [16, 19, 21, 24, 26–28].
Other demographic variables (i.e., age, income, and ethnic-
ity) were explored frequently [14, 16, 19, 27, 28]. Finally,
psychiatric symptoms and diagnoses [14, 15, 25], a history
of sexual trauma [20], history of prior suicide attempts [18,
24], as well as problem-solving abilities and other cogni-
tive issues [17, 22] were tested as moderators in a number
of studies. Uniquely, Vidot et al. [29] explored parent–ado-
lescent communication as a moderator of their universal
intervention (Familias Unidas). Results suggested that the
intervention had a significant, positive impact among those
adolescents with low parent–adolescent communication.
Universal prevention programs
Five of the 17 articles were identified as focusing on univer-
sal prevention programs [18, 19, 27, 29, 31]. Hawkins et al.
examined the main effect of the Seattle Social Development
Project as well as the moderating effects of sex, poverty,
and ethnicity [27]. Results suggested a moderating role of
gender, with females responding better to the intervention
compared to men. No moderating effects were found for
childhood poverty. Some significant ethnic interactions were
found, primarily comparing African–American participants
to White participants. Wilcox et al. explored the impact of
two universal school-based prevention programs (the Good
Behavior Game and Mastery Learning) on suicide ideation
and attempt [31]. Moderation of the universal interventions
was explored with baseline aggressive behavior as well as
depressive and anxious symptoms. Results demonstrated no
significant moderation effects.
Selective and indicated prevention programs
Curry et al. [14] used logistic regression to explore the
role of family income, depressive symptoms, and cognitive
distortions in the treatment for major depressive disorder.
Kaminer et al. [15] used general linear models to explore
the main effects of cognitive behavioral group therapy for
13


Table 1  Summary of moderator variables and analytic methods in studies that assessed effects of moderators (n = 17)
Author, year Moderator variables Brief results Setting Analytic methods

13
Universal
 Hawkins, 2005 Gender, poverty, ethnicity Intervention effects significantly differ by gender but School MANOVA
not by childhood poverty. White participants in the
full-intervention group showed more constructive
engagement compared with those in the control
group
 Wilcox, 2008 Aggressive and disruptive behavior, depressive No significant moderation was found with the School Discrete time survival analysis
symptoms, anxious symptoms Good Behavior Game and baseline behavior and
symptoms
 Schilling, 2016 History of suicide attempt History of suicide attempt and intervention signifi- School Generalized estimating equations
cantly interacted to predict suicide planning at
post-test. No significant moderation was found
with suicide attempt or ideation at post-test
 Vidot, 2016 Parent–adolescent communication The universal intervention significantly reduces Family Growth curve model
suicidal behavior among adolescents with low par-
ent–adolescent communication
 Aseltine, 2007 Gender, race, grade, wave Intervention effects did not differ significantly by any School General linear models
demographic characteristics
Selective/indicated
 Kaminer, 2006 Internalizing disorders, externalizing disorders, Results of the intervention did not significantly differ Clinical General linear models
substance use disorders, suicide ideation by gender, DISC internalizing disorders, DISC
substance use disorders, and baseline SIQ scores
 Kerr, 2014 Addition of intervention modules Suicide ideation decreased significantly more among Clinical Hierarchical linear growth models
those receiving the additional intervention models
as compared to those receiving the multidimen-
sional treatment foster care (MTFC)
 Diamond, 2012 Sexual trauma History of sexual trauma did not moderate treatment Clinical Hierarchical generalized linear model
outcome for attachment-based family therapy
 King, 2012 Receipt of public assistance, gender Adolescents whose families did not receive public Clinical ANOVAs and logistic regressions
assistance had higher levels of suicide ideation if
assigned to in-person follow-up
 King, 2009 History of multiple suicide attempts, gender Youth-nominated support team intervention effects Hospital Mixed effects models
were moderated by a history of multiple suicide
attempts, demonstrating more rapid decrease in
ideation for those with multiple attempts
 King, 2006 Gender Adolescent girls who received the Youth-nominated Hospital Chi-square, Fisher’s exact test
support team intervention demonstrated greater
decreases in self-reported suicide ideation as com-
pared with girls who did not get the intervention
 Wingate, 2005 Problem-solving appraisal Participants with poor problem-solving appraisal at Clinical Hierarchical multiple regression
baseline responded better than participants with
good problem-solving appraisal
Social Psychiatry and Psychiatric Epidemiology
 Social Psychiatry and Psychiatric Epidemiology

alcohol use disorders and suicide ideation. Furthermore,

Emergency room Linear mixed effects regression model

Hierarchical linear growth modeling
this study explored treatment differences based on gender,
Mixed effects growth models internalizing disorders, substance use disorders, and base-
line suicide ideation scores.
Shamseddeen et al. [17] explored the role of medication

Multivariate regression
along with cognitive behavior therapy to aide in treatment-

Logistic regression
Analytic methods

resistant depression among adolescents. Moderation anal-

yses found a significant interaction between adolescents
having school difficulties and timing of treatment. Their

ANOVA analysis of variance, DISC The Diagnostic Interview Scale for Children, MANOVA multivariate analysis of variance, SIQ Suicidal Ideation Questionnaire
results suggested that participants who had school diffi-
culties and were ending their treatment during the active
school year had significantly lower treatment responses
compared to others [16].
Diamond et al. [20] used general linear modeling to
Clinical

Clinical

Clinical

explore whether sexual trauma history moderates the

Setting

School

impact of attachment-based family therapy for adolescents

with suicide ideation. Huey et al. used generalized linear
The intervention had the greatest impact on maternal

Family income, depressive symptoms, and cognitive

distortions all moderated the acute outcome of the
ing suicide attempts as compared with emergency
hospitalization, this relationship varied by ethnic-
Multisystemic therapy was more effective at reduc-

Personal control mediated the relationship between

mixed models analysis to assess the moderating role of

cents having school difficulties and ending treat-
Lowest rates of response occurred among adoles-
emotional distress and family cohesion among

age, gender, and ethnicity moderated the impact of multi-

support resources and suicide risk behavior

suicide attempters who were symptomatic

systemic therapy [28].

Additionally, the other indicated prevention program
analyses used a variety of statistical techniques to analyze
ment during the active school year

the moderator effect on treatment or prevention (Table 1).

King et al. used a randomized control trial to explore the
impact of a youth-nominated support team to reduce risk
for suicide attempt and suicide ideation in hospitalized
ity, gender, and age

adolescents [26]. This team used Chi-square analyses,

Fisher’s exact tests, and repeated measures ANOVA to
TADS study
Brief results

explore main effects and interaction effects. They reported

a significant three-way interaction between time, sex, and
intervention. The same team, in a second manuscript on
youth-nominated support team for hospitalized adoles-
cents [24], found no treatment effect heterogeneity based
Family income, depressive symptoms, cognitive

on multiple attempt status using ANOVA [24]. King et al.

explored the role of communication on type of screening
in response to a self-report suicide risk screen and focused
on the moderating role of receiving public assistance [16].
Results showed a significant interaction between follow-
up condition and receipt of public assistance suggesting
that in families who did not receive public assistance and
Gender, age, ethnicity

 Rotheram-borus, 2000 Psychiatric symptoms

who were told they would receive an in-person follow-up

Moderator variables

School difficulties

reported greater suicide ideation. Rotheram-borus et al.

explored the impact of an emergency room intervention,
distortions

and found a significant interaction of the intervention

Gender

condition with the participants’ initial level of psychiatric

symptoms [25]. Thompson et al. focused on an indicated
prevention program for adolescents at risk for suicide [21].
This study was accomplished using structural equation
 Shamseddeen, 2011
Table 1  (continued)

 Thompson, 2000

modeling and found moderating effects of gender. Win-

gate et al. explored the moderating role of problem-solving
 Curry, 2011
 Huey, 2004
Author, year

skills on treatment outcomes [22]. Results suggested that

individuals with poor problem-solving skills responded
better than those with good problem-solving skills.

13

Discussion
This manuscript reports a component of a broad systematic
review of suicide-prevention data-linkage strategies, focus-
ing on studies that explored moderation in some form. We
sought to identify the statistical methods used for mod-
eration and whether they were appropriately utilized in
these articles. Just over 20% of the studies included in the
original systematic review explored moderators despite the
essential effect of heterogeneity on treatment or preven-
tion [3]. The majority of studies that evaluated moderators
of impact on suicide outcomes were exploring indicated
prevention programs.
Moderators of suicide prevention studies include non-
modifiable factors such as demographics (e.g., age, gender,
race/ethnicity), genetic factors, trauma history, LGBTQ
status or modifiable characteristics such as psychosocial
factors (e.g., impulsive aggression, depression, social con-
nectedness and support), implementation factors such as
intervention fidelity, access to lethal means, nongenetic
biological factors (e.g., DNA methylation, immune system
function). A number of moderators commonly included in
models of suicide were not included in the studies identi-
fied: genetic and nongenetic biological factors (e.g., DNA
methylation, immune system function), LGBTQ status,
social connectedness and support, family history of sui-
cide, implementation factors such as intervention fidelity,
access to lethal means [32].
To date, the emphasis of prevention programs that
have examined suicide outcomes has been on overall
impact. The relative lack of testing moderation could
impede the knowledge base on for whom and under what
circumstances interventions have had an impact on sui-
cide-related outcomes. This lack of studies examining
moderation might be partially explained by the common
challenges in suicide research of working with low base
rate outcomes (suicide attempts and suicide death) and the
sensitivity and safeguards needed to conduct this type of
clinical research. In the age of big data, it is possible that
this will change. We urge the suicide prevention field to
study for whom and under what circumstances interven-
tions have had an impact on suicide-related outcomes.
Several of the studies utilized linear models with interac-
tion terms—a pragmatic analytical technique [5]—to explore
moderation. This analytic technique has been shown to be a
valid approach for studying moderation in youth suicide-pre-
vention programs [5]. However, none of the studies graphi-
cally depicted the moderation results which could facilitate
interpretation [33]. Further, more novel techniques to study-
ing moderation (i.e., propensity scores) have been underu-
tilized in suicide-prevention work, and mental health work
more broadly, but could easily be applied [8].
13
Social Psychiatry and Psychiatric Epidemiology
There could be several reasons why researchers did not
focus on moderation effects. For example, having sufficient
power to detect moderation effects may not have been possi-
ble, particularly in complex prevention intervention designs
which focus on indicated programs with limited sample
sizes. Sufficient sample size is needed to provide adequate
statistical power to identify the impact of moderators on
suicide deaths, as well as attempts. Many of the studies
identified in our review were small, for example, one study
included 98 participants [22]. The sample size required for
detecting moderation effects is significantly more than the
sample size needed to detect a main effect [5]. According to
the NAASP’s prioritized research agenda for suicide preven-
tion (2014), statistical power remains a substantial issue to
combat, particularly with the low base rates of suicide death
and attempt.
Additionally, the prevention interventions explored may
be too novel to hypothesize moderation effects, such that
researchers are still focused on understanding the main
effects of their programs. Further, the focus of researchers
may also be primarily on understanding the main effects of
the preventive intervention and not on exploring treatment
effect heterogeneity. And finally, the challenges that arise
in multi-trial data harmonization may explain lack of mod-
eration exploration may be a result of very few of the stud-
ies that utilized methods to harmonize data across multiple
prevention intervention trials [11]. Data harmonization may
provide a useful way to increase sample size and add to the
low base rate of some moderators (e.g., LBGTQ status). This
may not have been done due to a lack of common data ele-
ments that would support harmonization. Moving forward,
funding agencies may wish to require common data elements
across treatment trials to allow for harmonization across tri-
als without complex statistical models.
Methodology used to examine moderation effects is in
need of more development. The current analytic techniques
may not be the best suited for moderators that are rare, or
outcomes that are rare. This is especially relevant for sui-
cide-prevention programming, as both suicide attempt and
suicide completion are relatively rare. These outcomes and
moderators require analytic tools that have not yet been
applied in the current set of studies. An additional problem
may be the presence of multiple moderators, which can com-
plicate the analytic methodology and requires more complex
analytic tools still in development (e.g., a combination of
variables interact to lead to moderation) [34].
There are two broad classes of methodologies that war-
rant further investigation: outcome-model-based approaches
(e.g., random forests and Bayesian additive regression trees),
and propensity score approaches. These approaches are
increasing in popularity to explore treatment effect heteroge-
neity, for example, propensity score methods are commonly
used in exploring educational programs and outcomes [35].
Social Psychiatry and Psychiatric Epidemiology
Further, these methodologies may offer a unique opportu-
nity to model complex moderation models within preven-
tion intervention trials [36]. These methods have not been
specifically developed for rare outcomes such as suicide.
Therefore, more research is needed to understand the right
analytic framework to use when the outcome is rare.
Clinical implications
Suicide is such a tragic preventable outcome, it is ideal for
clinicians to have the best information available for clinical
decision making in determining who is at highest risk or
who will benefit most from specific treatments. As always,
when clinicians go to the literature to look for interventions
they might apply with individual patients or new programs,
they should make sure that the studies read were conducted
in similar populations to the ones that they serve. Even when
there seems to be a match between the study population and
individuals served by clinicians, the lack of exploration of
heterogeneity means that it is still not clear that the results
are applicable to any given patient. There is always a need to
partner with patients and their families to explore the accept-
ability of interventions and to explore a range of choices that
might be acceptable. Further, to the extent that clinicians get
involved in recruiting for studies, standing up for both the
inclusion of relevant populations as well as the importance
of exploring treatment effect heterogeneity can make a big
contribution.
Limitations
This paper was limited by what was collected in the broader
systematic review as only this pool of included studies was
examined for moderation analyses. The systematic review
was primarily focused on understanding the state of the sci-
ence related to youth suicide-prevention studies and national,
state, and community level data systems that could be used
within suicide-prevention studies. For example, studies that
did not fit the inclusion criteria of the broader systematic
review (e.g., conducted in the US, on those ages 0–25, with
a suicide ideation, suicide attempt or suicide death outcome)
were not included in this review. Therefore, we may be miss-
ing some suicide-prevention programs that explored modera-
tion effects if they focused on adults or were outside the US.
Further, we were limited to information extracted from
the published work. We did not focus on evaluating the qual-
ity of the moderators or the research used to support select-
ing them. A potential publication bias may have limited the
methodologies used to explore moderation in this study. It
is possible that more studies explored moderation effects but
chose not to publish non-significant findings.
Conclusions
This systematic review identified that only 22% of youth
suicide-prevention studies explored treatment effect hetero-
geneity using statistical approaches of moderation analysis.
Though most of the studies may have collected variables
that may moderate the relationship between the intervention
and the outcome, few explicitly tested the effects of these
variables.
With a growing emphasis on understanding mechanisms
of intervention effects, researchers should take advantage
of the data already collected to study the prevention effect
heterogeneity. The broader research community should focus
on developing analytic methods to understand the role of
moderator variables in suicide-prevention studies. Data har-
monization methods, which allow the integration of data
systems into existing trials may aid in increasing power and
availability of moderator variables. The National Strategy
for Suicide Prevention emphasizes the importance of evalu-
ating the effectiveness of suicide-prevention interventions.
This objective requires the use of moderation methods to
understand prevention effect heterogeneity.
Acknowledgements  This project was funded under Contract no.
HHSA29020150000XI Task Order 290-2012-00007-I from the Agency
for Healthcare Research and Quality (AHRQ), U.S. Department of
Health and Human Services (HHS). The authors of this manuscript
are responsible for its content. Statements in the manuscript do not
necessarily represent the official views of or imply endorsement by
AHRQ or HHS. This topic a Pathways to Prevention Project and was
nominated by the Office of Disease Prevention of the National Institutes
of Health (NIH) and selected by AHRQ for systematic review by an
EPC. A representative from AHRQ served as a Contracting Officer’s
Technical Representative and provide technical assistance during the
conduct of the full-evidence report and provided comments on draft
versions of the full-evidence report. AHRQ did not directly participate
in the literature search, determination of study eligibility criteria, data
analysis or interpretation, or preparation, review, or approval of the
manuscript for publication.
Compliance with ethical standards 
Conflict of interest  The authors declare that they have no conflict of
interest.
References
1. Prevention C for DC (2015) Web-based injury statistics query and
reporting system (WISQARS). http://www.cdc.gov/injur​y/wisqa​
rs/. Accessed 14 Jul 2015
2. National Action Alliance for Suicide Prevention, R. T. F. (2014) A
prioritized research agenda for suicide prevention: an action plan
to save lives. National Institute of Mental Health and the Research
Prioritization Task Force, Rockville
3. Institute of Medicine (US) Committee on Pathophysiology and
Prevention of Adolescent and Adult Suicide; Goldsmith SK,
13

Pellmar TC, Kleinman AM, Bunney WE (eds) (2002) Reducing
suicide: a national imperative. National Academies Press, Wash-
ington, DC
4. Belsky J (1997) Theory testing, effect-size evaluation, and dif-
ferential susceptibility to rearing influence: the case of mother-
ing and attachment author(s): Jay Belsky Published by: Wiley on
behalf of the Society for Research in Child Development Stable.
Child Dev 68:598–600
5. Fairchild AJ, Mackinnon DP (2009) A general model for testing
mediation and moderation effects. Prev Sci 10:87–99. https​://doi.
org/10.1007/s1112​1-008-0109-6
6. Angrist JD (2004) Treatment effect heterogeneity in theory and
practice. Econ J 114:C52–C83
7. Dunn G, Bentall R (2007) Modelling treatment-effect heteroge-
neity in randomized controlled trials of complex interventions
(psychological treatments). Stat Med 26:4719–4745
8. Green KM, Stuart EA (2014) Examining moderation analyses in
propensity score methods: application to depression and substance
use. J Consult Clin Psychol 82:773
9. Kraemer HC, Wilson GT, Fairburn CG, Agras WS (2002) Media-
tors and moderators of treatment effects in randomized clinical
trials. Arch Gen Psychiatry 59:877–883
10. Imai K, Ratkovic M (2013) Estimating treatment effect het-
erogeneity in randomized program evaluation. Ann Appl Stat
7:443–470
11. Brown CH, Sloboda Z, Faggiano F et al (2013) Methods for syn-
thesizing findings on moderation effects across multiple rand-
omized trials. Prev Sci 14:144–156. https:​ //doi.org/10.1007/s1112​
1-011-0207-8
12. Wilcox HC, Kharrazi H, Wilson RF et al (2016) Data linkage
strategies to advance youth suicide prevention: a systematic
review for a National Institutes of health pathways to prevention
workshop. Ann Intern Med. https​://doi.org/10.7326/M16-1281
13. Wilcox H, Wissow L, Kharrazi H, Wilson R, Musci R, Zhang
A, et al (2016) Data linkage strategies to advance youth suicide
prevention. Evidence Report/Technology Assessment no. 222.
(Prepared by the Johns Hopkins University Evidence-based Prac-
tice Center under contract no. 290-2012-00007-I.) Agency for
Healthcare Research and Quality, Rockville
14. Curry J, Silva S, Rohde P et al (2011) Recovery and recurrence
following treatment for adolescent major depression. Arch Gen
Psychiatry 68:263–270. https​://doi.org/10.1001/archg​enpsy​chiat​
ry.2010.150
15. Kaminer Y, Burleson JA, Goldston DB, Burke RH (2006) Sui-
cidal ideation among adolescents with alcohol use disorders dur-
ing treatment and aftercare. Am J Addict 15:43–49. https​://doi.
org/10.1080/10550​49060​10061​54
16. King CA, Hill RM, Wynne HA et al (2012) Adolescent suicide
risk screening: the effect of communication about type of follow-
up on adolescents’ screening responses adolescent suicide risk
screening : the effect of communication about type of follow-up
on adolescents’ screening responses. J Clin Child Adolesc Psychol
41:508–515. https​://doi.org/10.1080/15374​416.2012.68018​8
17. Shamseddeen W, Clarke G, Wagner KD, Ryan ND, Birmaher
B, Emslie G, Asarnow JR, Porta G, Mayes T, Keller MB, Brent
DA (2011) Treatment-resistant depressed youth show a higher
response rate if treatment ends during summer school break. J
Am Acad Child Adolesc Psychiatry 50:1140–1148. https​://doi.
org/10.1016/j.jaac.2011.07.022
18. Schilling EA, Aseltine RH, James A (2016) The SOS suicide pre-
vention program: further evidence of efficacy and effectiveness.
Prev Sci 17:157–166. https:​ //doi.org/10.1007/s11121​ -015-0594-3
19. Aseltine RH, James A, Schilling EA, Glanovsky J (2007)
Evaluating the SOS suicide prevention program: a replica-
tion and extension. BMC Public Health 7:1–7. https​ : //doi.
org/10.1186/1471-2458-7-161
13
Social Psychiatry and Psychiatric Epidemiology
20. Creed T, Gillham JE, Diamond G et al (2012) Sexual trauma his-
tory does not moderate treatment outcome in attachment-based
family therapy (ABFT) for adolescents with suicide ideation
sexual trauma history does not moderate treatment outcome in
attachment-based family therapy (ABFT) for adolescen. J Fam
Psychol. https​://doi.org/10.1037/a0028​414
21. Thompson EA, Eggert LL, Herting JR (2000) Mediating effects
of an indicated prevention program for reducing youth depression
and suicide risk behaviors. Suicide Life Threat Behav 30:252–271
22. Wingate LR, Orden KA, Van TEJ Jr et al (2005) Comparison of
compensation and capitalization models when treating suicidality
in young adults. J Consult Clin Psychol 73:756–762. https​://doi.
org/10.1037/0022-006X.73.4.756
23. Diamond G, Creed T, Gillham J et al (2012) Sexual trauma history
does not moderate treatment outcome in attachment-based fam-
ily therapy (ABFT) for adolescents with suicide ideation. J Fam
Psychol 26:595–605. https​://doi.org/10.1037/a0028​414
24. King CA, Klaus N, Kramer A, Venkataraman S, Quinlan PG
(2009) The youth-nominated support team for suicidal adoles-
cents—version II: a randomized controlled intervention trial. J
Consult Clin Psychol 77:880–893. https​://doi.org/10.1037/a0016​
552.The
25. Rotheram-borus MJ, Piacentini J, Cantwell C et al (2000) The
18-month impact of an emergency room intervention for adoles-
cent female suicide attempters. J Consult Clin Psychol Psychol
68:1081–1093. https​://doi.org/10.1037//0022-006X.68.6.1081
26. King CA, Kramer A, Preuss L et al (2006) Youth-nominated
support team for suicidal adolescents (version 1): a randomized
controlled trial. J Consult Clin Psychol 74:199–206. https​://doi.
org/10.1037/0022-006X.74.1.199
27. Hawkins JD, Kosterman R, Catalano RF et al (2005) Promoting
positive adult functioning through social development interven-
tion in childhood. Arch Peditric Adolesc Med 159:25–31
28. Huey SJ, Henggeler SW, Rowland MD (2004) Multisystemic ther-
apy effects on attempted suicide by youths presenting psychiatric
emergencies. J Am Acad Child Adolesc Psychiatry. https​://doi.
org/10.1097/01.chi.00001​01700​.15837​.f3
29. Vidot DC, Huang S, Poma S et al (2016) Familias Unidas’ cross-
over effects on suicidal behaviors among hispanic adolescents:
results from an effectiveness trial. Suicide Life Threat Behav
46:S8–S14. https​://doi.org/10.1111/sltb.12253​
30. Kerr DCR, DeGarmo DS, Leve LD, Chamberlain P (2014) Juve-
nile justice girls’ depressive symptoms and suicidal ideation
9 years after multidimensional treatment foster care. J Consult
Clin Psychol 82:684–693. https​://doi.org/10.3174/ajnr.A1256​
.Funct​ional​
31. Wilcox HC, Kellam SG, Brown CH et al (2008) The impact of two
universal randomized first- and second-grade classroom interven-
tions on young adult suicide ideation and attempts. Drug Alcohol
Depend. https​://doi.org/10.1016/j.druga​lcdep​.2008.01.005
32. Gould MS, Greenberg TED, Velting DM, Shaffer D (2003)
Youth suicide risk and preventive interventions: a review of the
past 10 years. J Am Acad Chld Adolesc Psychiatry. https​://doi.
org/10.1097/01.CHI.00000​46821​.95464​.CF
33. Rosnow RL, Rosenthal R (1989) Definition and interpretation of
interaction effects. Psychol Bull 105:143–146
34. Kent DM, Rothwell PM, Ioannidis JPA et al (2010) Assessing
and reporting heterogeneity in treatment effects in clinical trials:
a proposal. Trials 11:1–10
35. Morgan PL, Frisco ML, Farkas G, Hibel J (2010) A propensity
score matching analysis of the effects of special education ser-
vices. J Spec Educ 43:236–254
36. Sparapani RA, Logan BR, McCulloch RE, Laud PW (2016) Non-
parametric survival analysis using Bayesian Additive Regression
Trees (BART). Stat Med 35:2741–2753