Chapter  107:  Preventing Violence and Related Health-Risking Social Behaviors in Adolescents

Purpose of this Review

Over the last two decades of the 20^th century, violence emerged as one of the most significant public health problems in the United States (Administration for Children and Families, 2004). While adults continue to constitute the majority of violent offenses, the decade between 1983 and 1993 was marked by an unprecedented surge of violence, often lethal violence, among young people in the United States (Administration for Children and Families, 2004). This surge of violence left countless young people and their families affected by injuries, disability, and death. Since 1993, there have been encouraging signs that youth violence may be on the decline, a trend that researchers, and the legal and policy communities are attempting to understand. The dramatic rise in youth-centered violence that began in the early 1980s precipitated an urgent and widespread drive among researchers and policy makers across multiple disciplines and sectors to understand the factors that contribute to violence and to develop interventions to address these factors and stem the tide of increasing violence. Science can play an important role in clarifying the scope of the problem, elucidating the responses needed to further reduce and/or eliminate youth violence and related harmful health behaviors, and informing both the development and evaluation of new policies and prevention interventions.

In October of 2004, the National Institutes of Health will convene a State-of-the-Science Conference on “Preventing Violence and Related Health-Risking Social Behaviors in Adolescents.” The purpose of this consensus conference is to provide a forum to present longitudinal and experimental risk factor research and intervention research that has yielded information documenting the role of individual, family, school, community, and peer level influences. In preparation for this meeting, the Office of Medical Applications of Research (OMAR) and the National Institute of Mental Health (NIMH) nominated and supported the topic. The Agency for Healthcare Research and Quality (AHRQ) awarded this project to the Southern California Evidence-Based Practice Center (SC-EPC) and its partner, Childrens Hospital Los Angeles, to conduct a systematic review and analysis of the scientific evidence that exists relative to the prevention of violence and related health-risking social behaviors in adolescence, and to summarize these findings in an evidence report. This systematic review included an evaluation of the factors that contribute to violence during childhood and adolescence as well as the effectiveness of prevention interventions. The findings contained in this report will be presented at the 2004 conference.

Epidemiology of Youth Violence

According to a seminal 2001 report by the Surgeon General, youth violence is one of the Nation's most serious, insidious, and complex problems, influencing nearly every aspect of society (Satcher, 2001). In the decade that extended from roughly 1983 to 1993, an epidemic of violent, often lethal behavior emerged in the United States, resulting in untold injury, disability, and death (Cook & Laub, 1998). Indeed, during that decade, arrests of youth for serious violent offenses surged by 70 percent; more alarmingly, the number of young people who committed a homicide nearly tripled. During that same period of time, the homicide arrest rate, increased 273 percent for adolescents, 14 to17 years (from 7.0 to 19.1 per 100,000), and 65 percent for young adults, 18 to 24 years (from 15.7 to 25.3 per 100,000). In contrast, the homicide arrest rates reported among adults 24 years and older declined by 25 percent (6.3 to 4.7 per 100,000). This increase in homicide arrest rates among adolescents and young adults has largely been attributed to an increase in gang-related activity, an increase in illicit drug use, and the increased availability of guns and other lethal weapons (Hennes, 1998). Among the youth arrested for violent offenses, most are males (84 percent), with males accounting for 94 percent of juveniles convicted for homicide (Federal Bureau of Investigations, 1999).

Yet as we mentioned above, since 1993, the peak year of the epidemic, some encouraging signs have appeared that youth violence is declining. Three important indicators of violent behavior—arrest records, victimization data, and hospital emergency room records—have shown significant downward trends nationally. Despite these encouraging trends, homicide continues to be the second and third leading cause of death in the U.S. for persons aged 15 to 24 years and 5 to 14 years, respectively, and it is the leading cause of death among African American and the second leading cause of death among Latino youth (National Center for Injury Prevention and Control, 2004).

While students are safer in school than out of school, recent shootings in the nation's schools have focused public attention on school-related violence and crime (DeVoe, Peter, 2002). From July 1, 1992 through June 30, 1999, there were 358 school-associated violent deaths in the United States. Overall, school-associated homicide rates appear to have increased in recent years due to an increase in students killed in multiple-victim homicide events. In a 1992-1993 survey of the National School Boards Association, 82 percent of school districts nationwide reported student involvement in violence had increased over the past 5 years (Lowry, Sleet, 1995). Violent behaviors reported by districts included student-on-student assault (78 percent of districts), shootings or knifings (39 percent), and rape (15 percent of districts). An examination of more recent trends in student-on-student assaults shows that the percentages of students who reported fighting on school property declined from 16 percent in 1993 to 13 percent in 2001.

Not all violence reaches the level of homicide. Indeed, one of the most common forms of violence perpetrated by youth is physical fighting. According to the 1999 Youth Risk Behavior Survey, 36 percent of students nationwide reported having been in a physical fight one or more times on 1 or more days during the 12 months preceding the survey, amounting to 105.9 incidents of physical fighting per 100 students. Overall, male students (44 percent) were significantly more likely than female students (27 percent) to have participated in a physical fight; this finding was consistent across racial/ethnic groups and within grades (Kann, Kinchen, 2000). However, encouraging trends have emerged in the percentage of youth reporting involvement in physical fighting, with a significant decrease between 1993 and 1999.

Young males are disproportionately represented as both victims and perpetrators of all forms of violence in the United States. Of the 18,272 homicides committed in the United States in 1998, 35 percent of all victims were younger than age 25 years and 82 percent of these victims were male (Dahlberg and Potter, 2001). Males, 10 to 17, are also significantly more likely to be involved in aggravated assault and robbery than their female peers, while male students in grades 9 through 12 were more likely to report having been in a physical fight and to have engaged in physically aggressive behaviors while at school (Lowry, Sleet, 1995).

Further, the risk of violent death and of committing a violent crime is greater for young people of color and those who are economically disadvantaged. Homicide is the leading cause of death among African-American and Latino youth, 15 to 24. (Dowd, 1998) In 1998, homicide rates for African-American youth were more than twice the rate of Latino youth and more than 13 times the rate of Caucasian, non-Hispanic youth (56.5 vs. 23.3 vs. 4.2 per 100,000, respectively) (Dahlberg and Potter, 2001). In addition to having the highest homicide victimization rate, African-American male youth also have the highest homicide arrest rate. In 1991, African-Americans were 7 times more likely to be arrested for homicide than Whites. Between 1985 and 1994, African-American males, ages 14 to 17 years, had the largest increase in homicide arrest rates (315 percent) (Hennes, 1998). However, racial differences in homicide rates appear to be mediated by both poverty and race-specific homicide victimization; homicide arrest rates become similar after controlling for socioeconomic status (Hennes, 1998).

Over the past two decades, a growing body of research has begun to identify the range of individual, social, environmental and community-level factors that are associated with an increased risk for youth violence, delinquency, and juvenile crime. Researchers generally agree that behavior, including violent and antisocial behavior, is the result of a complex interplay of individual, biological, genetic, and environmental factors that begin to exert their effects during or even prior to fetal development and continue throughout life (Bock and Goode, 1996). Evidence is also emerging regarding developmental precursors in early childhood to youth antisocial and delinquent behavior (McCord, Widom, 2001; Shonkoff and Phillips, 2000). The literature documents the exploration by researchers of potential causes for or contributors to youth violence from early childhood such as child abuse, particular parenting styles, and features of the environment or the community. A wealth of literature also documents investigations of adolescents themselves and their involvement with gangs and other peer groups, the availability of firearms in their community, and their relationships with their sexual partners, all as potential cofactors for violence. Another body of research literature looks at resiliency in an effort to find clues to why the majority of young people with similar individual, familial, and community exposures to these risk factors do not become involved in violence.

Despite this growing evidence, it has been exceedingly difficult to evaluate the strength of this evidence regarding the reported relationships between youth violence and a wide range of risk factors and protective influences. This difficulty is in part due to the fact that numerous disciplines and fields of research, including but not limited to epidemiology and public health, psychology, child development, sociology, anthropology, social work, medicine, education, and public policy, have collectively contributed to this literature, each potentially looking at and operationally defining violence, as well as risk and protective influences, in different ways. Moreover, a wide range of research study designs have been used to evaluate risk and protective influences, with varying degrees of scientific rigor. As a result, the current literature is fragmented in nature, with inconsistent findings often reported across individual studies. While numerous attempts have been made to review the literature (Dahlberg and Potter, 2001; Raine, 2002; Sampson, Morenoff, 2002; Villani, 2001), it remains unclear which risk factors are most salient across different research settings and subject populations. Consequently, it remains relatively unclear which risk and protective factors are most amenable to change through prevention. Moreover, few have attempted to perform a systematic review of the strength of the existing evidence, or to limit their scope to those studies conducted with the greatest scientific rigor. Thus, questions remain about what future research is needed to extend the current literature. It is for all these reasons that the National Institutes of Health commissioned a systematic review of the literature and the strength of the evidence reported in this literature, the results of which are summarized in this report.

Violence Prevention Programs

As the rates of violence began to increase in the 1980s, an entire field of violence prevention emerged with the design and development of many new violence prevention intervention approaches and programs. Prevention and early intervention programs are now in place in cities and regions throughout the country; many target youth violence through early child interventions, others are specifically targeted to adolescents. And while many of these interventions have been evaluated to determine their effectiveness in preventing violent behavior that is perpetrated by youth, the quality and scientific rigor of those evaluations has varied considerably. Moreover, the research on youth violence prevention remains fragmented in nature, in part because of the wide range of interventions approaches used, some better described than others, with the specific targets for the interventions often poorly defined. In addition, given the wide range of program designs and the settings in which they are likely to be delivered, it is difficult to determine what scientific standard should be set and/or methodological approach used to evaluate existing programs with the utmost in scientific rigor. A number of youth violence interventions and prevention programs have been demonstrated to be effective (Satcher, 2001). Unfortunately, few interventions effectively address involvement with delinquent peers and gang membership. Moreover, determining which type of intervention approach might be most effective for which individuals remains largely an unsolved problem, as do sustaining positive outcomes associated with these interventions over time and enacting the kinds of national, state, and local policies that will address the underlying risk and protective factors that are so closely associated with youth violence, delinquency, and other potentially harmful behaviors.

Societal Burden of Youth Violence

The economic costs of violence can be difficult to measure. The cost of violence can be considered in terms of economic, emotional, and social costs, but no reliable estimates exist for expenditures associated with medical care, legal and social investigations, and interventions related either to nonfatal assaults or to homicide. Some estimates for medical care costs do exist. For example, medical treatment for fatal and non-fatal gunshot wounds is estimated to cost one billion dollars per year. The costs of other potential medical consequences of violence including the need for long-term institutional care, rehabilitation services, and support services to victims and their families, have not been estimated. The costs of forensic investigations, court proceedings, incarceration, or processes related to legal execution can also be considered among the economic costs of violence. One unique measure that can incorporate both financial and societal costs is Years of Potential Life Lost. . In 1994, more than 470,000 premature years of life were lost due to the homicide deaths of individuals younger than 25 years old (Dowd, 1998).

Even more difficult to measure than the economic costs of violence are the social and emotional costs. The social and emotional costs of violence include long-term physical and mental disabilities and adverse psychological and behavioral consequences for perpetrators, surviving victims, their families, entire communities, and society as a whole.

Summary

Given this growing yet fragmented knowledge base regarding a critically important public health problem, it is important to conduct a systematic review of the literature in an effort to bring the best available science to bear on future programs and policies. Summarized herein are the findings from a review of the evidence that was conducted on behalf of the National Institutes of Health's Office of Medical Applications and Research (OMAR) and the Agency for Healthcare Research and Quality's Evidence-based Practice Center.

Development of the Project Team

We worked closely with the Director of the Southern California Evidence-Based Practice Center to assemble a team of clinical, behavioral, and methodological experts, most of whom were staff members and faculty at Childrens Hospital Los Angeles. The team included the Task Order Director, with overall responsibility for the project; the Task Order Manager and Synthesis Coordinator, a biostatistician responsible for the methodology of the review and with expertise in conducting evidence based reviews; the Task Order Coordinator, responsible for coordinating activities to ensure effective communication and reporting; four Task Order Literature Reviewers and Synthesizers from the fields of medicine (two pediatricians, one of whom is board certified in Adolescent Medicine) and public health; and a Librarian. Additional researchers, primarily masters and doctoral students from the fields of public health, psychology, and prevention research, joined the team for three months to assist with the primary and secondary reviews. During the first six months, the team met weekly to review and refine the methodology of the task order.

Establishment of the Technical Expert Group (TEG)

In consultation with our Task Order Officer and the NIH Conference Panel Chair, we first created a Technical Expert Group (TEG) comprising nine individuals with both content and methodological expertise in the areas of youth violence and the prevention of youth violence. Specifically, we sought to create a multidisciplinary TEG that represented a range of related fields and disciplines, including early childhood development, adolescent development, juvenile justice, child abuse and neglect, anthropology, psychology, sociology, social work, public health, and public policy. The TEG was drawn from a large pool of potential candidates identified through a review of the literature and solicitation of nominations from researchers, and representatives from related federal agencies and private foundations. The list of potential technical experts and their curricula vitae were submitted to the Task Order Officer for approval. The final roster is provided in Appendix D1. ^*

Defining the Scope of Key Questions

This study was guided by a set of specific research questions that were developed by an NIMH panel of experts and modified in the Task Order. The initial Task Order specified that the team review and examine evidence related to both youth violence and delinquency. The body of published research related to youth violence and violence prevention alone was believed to be massive. Thus, one of the first efforts was to review the scope of the evidence related to youth violence and violence prevention interventions and to delinquency and to refine the key research questions that guided the Task Order. To assess the feasibility of reviewing both sets of literature, we first conducted a preliminary search of relevant databases to obtain an estimate of the number of published articles that might potentially be reviewed. From this search, we learned that MEDLINE® alone contained over 6,000 citations related to youth violence and over 11,000 articles related to youth delinquency, with little overlap between these two bodies of literature. We then determined that given the project timeline and available resources, it would not be feasible to review the literature on youth violence as well as that on delinquency as outcome behaviors. Thus, in consultation with our Task Order Officer and the NIH Panel Chair, we limited the scope of this review to focus specifically on youth violence as an outcome. As a result, the key questions were modified to reflect the revised scope of this review. The following list of questions was used to inform the evidence review:

1. What are the factors that contribute to violence and associated adverse health outcomes in childhood and adolescence?

2. What are the patterns of co-occurrence of these factors?

3. What evidence exists on the safety and effectiveness of interventions for violence?

4. Where evidence of safety and effectiveness exists, are there other outcomes beyond reducing violence? If so, what is known about effectiveness by age, sex, and race/ethnicity?

5. What are the commonalties of the interventions that are effective, and those that are ineffective?

6. What are the priorities for future research?

Development of Causal Pathways and Analytical Framework for Key Questions

Once the scope of the evidence report and the key questions were refined, we adopted definitions for youth violence and violence prevention interventions to further guide the selection and review of the appropriate literature. The definition of youth violence that we chose was developed by the Centers for Disease Control and Prevention Injury Center:

The definition of violence prevention interventions that we chose was developed for and published in the Surgeon General's Report on Youth Violence (Satcher, 2001): “Primary prevention interventions are those that are universal, intended to prevent the onset of violence and related risk factors; secondary prevention interventions are those implemented on a selected scale for children/youth at enhanced risk for youth violence, intended to prevent the onset and reduce the risk of violence; and tertiary prevention interventions are those that are targeted to youth who have already demonstrated violent or seriously delinquent behavior.”

We also generated a list of potential risk- and protective factors that have been found to be associated with youth violence. This list was, in turn, organized by domain – i.e., individual, family, school, peer, community, and social, and macro-level domains – and used to inform data abstraction and synthesis (Appendix E ^*).

Finally, we developed a framework to examine conceptually and analytically the associations between risk factors, violent behavior, and interventions. The Causal Pathways for Violent Behavioral Outcomes During Adolescence (Figure 1) focus on the age of exposure to risk factors. The Conceptual Framework for Risk and Protective Factors by Age of Exposure (Figure 2) depicts the potential opportunities for primary, secondary, and tertiary prevention of youth violence and associated adverse health outcomes.

As these documents were being developed, we shared them with the NIH Panel Chair and our Task Order Officer. We also discussed these documents with members of our TEG during several teleconference meetings, and made numerous revisions based on the feedback that we received.

Literature Search

As specified by the Task Order, the National Library of Medicine (NLM) performed all the searches that were used for this evidence review. Librarians from NLM met with project staff via teleconference to discuss the evidence review, the scope of the review, and the key questions. They also worked with project staff to select the literature databases that were ultimately used and evaluated the search strategies that had been developed by the project team.

In addition, members of the project team worked closely with the NIH Panel Chair and members of the TEG to decide how to further refine the scope of the review and hence, the search strategy. Decisions related to the scope of the review included determining which bodies of literature and search databases we would target and how far back in time we would search for related publications. Ultimately, we chose to limit our review to peer-reviewed published articles, articles retrievable within four search engines – MEDLINE®, PsychInfo, SocioAbstracts, and ERIC, and articles that were published in 1990 or thereafter, recognizing that by doing so, we would exclude a considerable segment of the literature. This decision was made in an effort to reduce the number of citations to be reviewed, given the time and resource constraints of the project and to ensure that our review was focused on the most current literature.

The decision was also made to limit the review to studies that were conducted in the United States, given growing evidence to suggest that numerous risk- and protective factors for violence are country specific, particularly factors that affect youth violence, and because the overarching purpose of the NIH consensus conferences will be to identify gaps and future research needs for research that will largely be conducted in the United States. Even with the limitations placed on the initial search for relevant literature, our first search generated over 10,000 unduplicated citations for review.

We chose to limit the scope of the review further to focus specifically on violent behavior perpetrated by youth, ages 12 through 17 years. We also did not include in our review the increasingly popular topics of verbal aggression, bullying, arson, weapon carrying, externalizing behaviors (e.g., acting out), attitudes about violent behavior, and intent to commit violence. Because our primary focus was on perpetration of violence, we also did not include the extensive literature about childhood and youth victimization. Moreover, this report does not review literature related to youth crime against property or materials (e.g. burglary, theft, vandalism). Thus, these related behaviors and attitudes are considered in this report only to the extent that they appear in the literature as risk factors for violence. In addition, the review did not include studies that examined precursors to violence that occurred in early adulthood (i.e., 18 years and older) or studies on the prevention of violence among young adults (i.e., 18 years of age and older).

As presented in Table 1, the NLM performed four searches in April/May of 2003 using four databases – MEDLINE®, PsychInfo, ERIC, and SocioAbstracts. The specific search strategies and terms used by NLM for these searches are provided in Appendices A1 through A9 ^*. Each time a search was performed, every effort was made to eliminate duplicate citations of articles that were referenced in more than one database.This process included an electronic removal of duplicate citations, first by NLM and then by the project librarian, followed by manual cross-checking of all citations.Table 1 provides a summary of the number of citations generated by each step and across the searches of the various databases provided to us by NLM.After elimination of duplicates, we were left with a total of 10,852 unique citations generated by these searches.A complete record of all titles and abstracts was kept using EndNote.We then performed a preliminary screening of these citations (described below).

To ensure that articles published during the course of this project were included, the NLM conducted a second supplemental search in October of 2003, using the same search strategies and databases. This search yielded an additional 344 citations; thus a total of 11,196 citations were identified during the course of this project.

Development of Data Collection Forms

We developed three data collection forms specifically for this project, including a Title/Abstract Screening Form (Form 1), a Secondary Screening Form for Full-Length Articles (Form 2), and the Study Quality Review Form (see Appendixes B1, B2, and B3 ^*).

The Title/Abstract Screening Form was developed as an initial screening tool to evaluate whether articles were appropriate for the evidence review, given the scope and key questions. The titles and abstracts (when available) were reviewed using six criteria to determine if they were eligible for inclusion within the evidence review. Articles were rejected if 1) they did not report original research findings (e.g., the article was an editorial, letter, discussion of clinical practice, overview, consensus statement, opinion piece, or commentary); 2) violence was not an outcome of the research; 3) the research did not involve human subjects; 4) the study was conducted outside the United States; 5) the age of the study population was 18 years or older; or 6) the study did not focus on youth as perpetrators of violence. If none of the rejection criteria applied, the article was deemed eligible for further review and the full-length article was retrieved.

The Secondary Screening Form of Full-Length Articles was developed to screen full-length articles for their appropriateness, given the scope and key questions. This form included three additional eligibility/rejection criteria, including 1) citation was a duplicate citation; 2) datawere not abstractable; and 3) study did not address one of the key questions. The form was also used to record the type of study design, using coded categories developed for and published in the Agency for Healthcare Research and Quality's Evidence Report Number 47: “Systems to Rate the Strength of Scientific Evidence” (West, King, 2002).

The Study Quality Review Form was used to evaluate the quality of each individual study; this form was adapted from guidelines that had previously been established by the OMAR to evaluate the quality of study designs (ODS and OMAR, 2003). The domains and elements for evaluating individual randomized controlled trials and observational studies are included in Form 3 (Appendix B3 ^*).

After developing these three screening/reviewing instruments, we conducted pilot testing and training with members of the project team to ensure the reliability and validity of the screening and review of data. For the primary screening, the entire team was given the same set of 10 titles and abstracts to review. The Task Order Manager and Coordinator then reviewed the results with the team to ensure that all reviewers were in agreement about criteria for rejection and inclusion. When discrepancies were identified or questions about key terms emerged, they were brought to the team for discussion and resolution. For the secondary review, the team was similarly trained. The Task Order Manager thoroughly reviewed the screening instrument and answered questions, particularly those regarding new reasons for rejection. Reviewers were instructed to submit questions to the Task Order Manager by email, and these questions were shared with the entire team and discussed at weekly meetings.

Screening of Retrieved Titles/Abstracts Against Inclusion/Exclusion Criteria

When the searches had been completed and the duplicates had been removed, the citations were exported from EndNote into Microsoft Word so that each individual title and abstract could be printed on its own page for screening purposes. Excel spreadsheets were also created to record the results of the screening.

Two members of the team independently screened each citation. One screener was a member of the faculty with specific expertise related to adolescent development and/or youth violence, and the other screener had a masters degree in public health or was a doctoral student in the field of psychology, public health, or prevention research. The Task Order Manager or the Task Order Coordinator compared the screening results of the two screeners, resolved discrepancies, and recorded the decisions in the Excel master file. The citations for which full-length articles were to be pulled were forwarded to our librarian for retrieval. For the rejected citations, the reason for rejection was recorded (i.e., the first reason for rejection that was identified by the screeners). This protocol was followed throughout all screening processes.

Many citations identified through the initial and supplemental searches did not include an abstract or had a limited abstract. Thus, information was sometimes inadequate to perform the initial screening. In these cases, the full-length article was retrieved and the articles were screened using the secondary screening procedures (further described below).

At each staff meeting, the team reviewed the rules and instructions for screening and discussed any questions that arose during the initial round of screening. Most of the questions that arose during this round focused on whether or not terms used in the titles or abstracts corresponded to behaviors that fell within the definition of violence employed by the Task Order. These terms included, but were not limited to, conduct disorder, verbal aggression, oppositional defiant disorder, and externalizing behavior. Since many of these terms are not used uniformly in the literature and the title or abstract generally does not provide a full description of the behavior, many of these abstracts were accepted for further review.

Retrieval and Review of Full-Length Articles

The titles/abstracts identified for further review were forwarded to the librarian for full article retrieval. Libraries at both Childrens Hospital Los Angeles and the Keck School of Medicine at the University of Southern California were the primary sources of the articles. Articles not found at either library were sought through Inter-Library Loan requests. We were able to retrieve all 1,612 full-length articles.

Two independent reviewers used the Secondary Screening Form to screen all the full-length articles; the same project staff and screening procedures used for the primary screening were used for the secondary screening. As with the title/abstract screening, the Task Order Manager compared the screening results provided by the two screeners, resolved discrepancies through consensus, and recorded the decisions on the Excel master file. This process resulted in one of two outcomes: The article was either accepted for data abstraction or rejected. As was done with the primary screening, the first reason for rejection identified by the two screeners was recorded for all rejected articles.

Data Abstraction into Evidence Tables

For each article that was deemed eligible for inclusion, data were abstracted by a member of the project team and subsequently checked by the Task Order Manager. Two sets of evidence tables were constructed for this project: the Evidence Abstraction Table for Risk Factors (Evidence Table #1) which was specifically designed to address Key Questions #1 and #2, and the Evidence Abstraction Table for Interventions (Evidence Table #2) which was designed to address Key Questions #3–#5. The Task Order Manager presented drafts of the format, coding, and recording instructions for the two Evidence Tables to the team for comment and discussion. Based on this discussion, the Evidence Tables were modified prior to abstraction. The format for Evidence Table #1 and Evidence Table #2 are included in Appendixes B4 and B5 ^*.

To pilot test the forms, we assigned each reviewer the same two articles for abstraction: one article about risk factors and one about an intervention. As a result of this pilot testing, minor modifications were made to the abstraction forms. To ensure quality control over time, the team met regularly to discuss and review terms, procedures, and the abstraction process. In addition, reviewers submitted questions by email to the Task Order Manager, and the answers were shared with the entire review team. Most of the questions that arose during this period addressed missing data in the articles. For example, reviewers wanted to know whether or not to abstract an article when only the median age was provided (rather than the age range) and whether or not to abstract articles when the outcome reflected only attitudinal change or skill development but not a change in behavior. The Task Order Manager later checked all evidence tables for consistency and accuracy.

For the articles that addressed Key Questions #1 and #2, the following data were abstracted using the Evidence Abstraction Table for Risk Factors: 1) article identifier information, including internal record number, first author's last name, year of publication, and journal of publication; 2) characteristics of the study, including study design, study quality, sample size, age, gender and race or ethnicity description of the study population, time period, location, setting, study population, and inclusion and exclusion criteria; 3) risk factor information, including main independent risk factor(s), instrument(s) used to measure factors, and other factors studied; 4) outcome definitions and characteristics, including outcome measure(s), definition of the outcome, instrument used to measure outcome(s), type of instrument, circumstance or situational context in which the violence measured occurred, e.g. whether the violent incident was proactive or reactive, weapon used, and victim-offender relationship; 4) findings reported in the article; 5) any adverse health outcomes; and 6) whether the study was theory-driven and the theory that was used.

For the 32 articles that addressed Key Questions #3 through #5, the following data were abstracted using Evidence Abstraction Table for Interventions: 1) article identifier, including internal record number, first author's last name, year of publication, and journal of publication; 2) characteristics of the study, including study design, study quality, sample size, age, gender and race or ethnicity description of the study population, time period, location, setting, study population, and inclusion and exclusion criteria; 3) moderating and or mediating variables reported in articles; 4) outcome definition and characterization, including outcome measure(s), definition, instrument used to measure outcome, type, circumstance or situational context in which the violence measured occurred, e.g. whether the violent incident was proactive or reactive,, weapon used, and victim-offender relationship; 5) whether the program represented a primary, secondary, or tertiary intervention, the kind of intervention (such as behavioral, skill building, etc.), how the intervention was delivered, target population, setting where intervention was delivered, setting where subjects were recruited, professional background of the individual(s) performing intervention (e.g. teacher, psychologist, graduate student), duration of the program, and/or frequency of intervention; 6) study findings; 7) intervention effectiveness, and 8) any negative outcomes attributed to the intervention.

Review and Assessment of Study Quality

For this Task Order, we were expected to use the criteria set forth in the Procedures for EPC Reports for Office of Dietary Supplements (ODS) and the Office of Medical Applications of Research (OMAR) (ODS and OMAR, 2003) to evaluate the quality of individual articles. Thus, to assess the quality of individual articles for Key Questions #1 and #2, we first evaluated the relevance of seven criteria previously developed for use with observational studies: 1) baseline comparability, 2) concurrent controls, 3) follow-up rate greater than or equal to 80 percent, 4) valid and reliable instruments used for assessments, 5) equal application of instruments for assessment, 6) important outcomes considered, and 7) appropriate control of confounders. Because all the prospective longitudinal cohort studies included in our review satisfied criteria #1, #2, #5, and #6 in the same ways, we used the three remaining criteria (i.e., #3, #4, and #7) to evaluate the quality of articles addressing risk and protective factors associated with youth violence. In addition, although it was not considered as a criterion, we examined the participation rate of each cohort study as well as the retention rate because we believe that it is important to maximize both the participation rate and follow-up rate to achieve an unbiased study sample for prospective longitudinal cohort studies. Large participation and retention rates are especially important for outcomes such as violence, because risk factors that are likely to contribute to youth violence are also likely to be associated with both participation and attrition rates.

For the Key Questions related to effectiveness of interventions, Key Questions #3, #4, and #5, we used the criteria set forth by OMAR for randomized controlled trials and observational studies (Appendix B3 ^*).The rating system used with randomized controlled trials consists of “Good”, “Fair” or “Poor.” A “Good” rating was assigned if the study fulfilled all the OMAR criteria without uncertainty, which means that comparable groups were assembled initially through adequately concealed randomization and maintained throughout the study (follow-up at least 80 percent) and an intention-to-treat analysis was used.Intention-to-treat analysis was performed for randomized controlled trials. Intention-to-treat is a strategy for analyzing data from randomized controlled trials that compares participants according to the groups towhich they were originally randomly assigned. This type of analysis is generallyinterpreted as including all originally enrolled participants in the final analysis, regardless of the treatment they actuallyreceived, whether they subsequently withdrew, or some other deviation from the protocol (Hulley, Cummings, 2001).A study was graded “fair” if any or all of the following problems occurred: the groups assembled initially were generally comparable but some questions remained whether some (although not major) differences occurred with follow-up due to differential attrition; some but not all important outcomes were considered; and some but not all potential confounders were accounted for.A “Poor” rating was assigned if the study failed to (at least partially) fulfill most criteria or if any of the following were reported: lack of comparability of groups assembled initially or failure to maintain them throughout the study; failure to mask outcome assessment; little or no attention given to key confounders; andlack of use of intent-to-treat analysis.. The rating system used with observational studies also used the “Good”, “Fair” or “Poor” categories. A “Good” rating was assigned when the following criteria were satisfied: 1) comparable groups were assembled initially and maintained throughout the study (follow-up at least 80 percent); 2) reliable and valid measurement instruments were used and applied equally to the groups; 3) all important outcomes were considered; and 4) appropriate attention was given to confounders in analysis.A study was rated as “fair” if any or all of the following problems occurred:1) generally comparable groups were assembled initially but some question remained whether some (although not major) differences occurred with follow-up, 2) measurement instruments were acceptable (although not the best) and generally applied equally, 3) some, but not all, important outcomes were considered, and 4) some, but not all, potential confounders were accounted for.A “Poor” rating was assigned if any of the following was reported: 1) groups assembled initially were not similar or comparable, or were not maintained throughout the study; 2) unreliable or invalid measurement instruments were used to assess exposure or outcomes or not applied equally among groups, and 3) key confounders were given little or no attention.

The rating system used with case-control studies also used the same three categories: “Good”, “Fair” and “Poor”. A “Good” rating was assigned when the following criteria were satisfied: 1) there was an appropriate ascertainment of cases and a nonbiased selection of case and control participants (i.e., cases and controls were drawn from the same population); the exclusion criteria were applied equally to cases and controls; 2) the response rate was equal to or greater than 80 percent; 3) diagnostic procedures and measurements were accurate and applied equally to cases and controls; and 4) appropriate attention was paid to confounding variables. A “Fair” rating was assigned for studies where: 1) there was no obvious subject recruitment or selection bias, 2) the retention rate was less than 80 percent, and 3) some attention was given to possible confounding variables. A “Poor” rating was assigned if: 1) significant sampling biases were evident, 2) the response rates were less than 50 percent, or 3) insufficient attention was given to important confounding variables.

Procedures to Reduce Bias, Enhance Consistency, and Check Accuracy

To reduce selection bias, we assigned two reviewers — one faculty member with relevant expertise and one masters- or doctoral-level intern — to screen and review titles/abstracts and full-length articles at every stage of the selection process. For data abstraction, one member of the project team with at least a masters degree in public health, psychology, or prevention research abstracted the data onto evidence tables, and the Task Order Manager reviewed all of the evidence tables for data abstraction and recording accuracy.

Analysis of the Scientific Evidence

We describe here our methods of data synthesis for the two sets of key questions: Questions #1 and #2, the risk factor questions; and Questions #3, #4, and #5, the intervention questions.

Rating the Strength of Scientific Evidence

According to the OMAR guidelines, EPCs are not required to make judgments about the overall strength of a body of evidence. The rating of the strength of scientific evidence remains the prerogative of the Consensus Panel. However, we conducted two sensitivity analyses to assist the Consensus Panel to assess the strength of the scientific evidence in our review. The first sensitivity analysis addressed the adequacy of number of subjects studied. We reanalyzed the data excluding the studies with sample size below the thresholds set at 1100 for the general population and 500 for the at-risk population. The second sensitivity analysis addressed the quality of studies. We re-assessed the findings excluding the studies that did not meet all OMAR study quality criteria.

Priorities for Future Research (Key Question #6)

At the outset of the project, we established a conceptual framework (a road map of causal pathways); frameworks to categorize exposure factors, interventions, and violence outcomes; an analytical plan to assess the evidence according to key questions; and tools to assess study quality and rate the strength of the evidence. These items were used as yardsticks to measure the adequacy of the existing literature to address the key questions and to identify gaps in relevant research.

We used the findings from our rating of the overall strength of a body of evidence to identify gaps and potential areas for future research in three domains: quality, quantity, and consistency.

For quality, we addressed the extent to which the design, conduct, and analysis displayed by a body of research minimized selection-, measurement-, and confounding biases.

For quantity, we referred to the strength of the relationship between the exposure factor being evaluated and the outcome being measured, as well as to the amount of information supporting that relationship. Three main factors contributed to quantity: the magnitude of effect (i.e., estimated effects such as mean differences, odds ratio, relative risk, or other comparative measure); the number of studies performed on the topic in question (e.g., only a few versus perhaps a dozen or more); and the number of individuals studied, aggregated over all the relevant and comparable investigations, which provides the width of the confidence limits for the effect estimates.

For consistency, we referred to the degree to which a body of scientific evidence was in agreement with itself and with outside information. A body of evidence is said to be consistent when numerous studies performed in different populations using different study designs to measure the same relationship produce essentially similar or compatible results. In addition, consistency addresses whether a body of evidence agrees with externally available information about the topic. It is important to note, however, that consistency is not possible without a uniform approach to defining and operationalizing the independent and dependent variables studied.

While the first area of recommendations address the quality of the study, the second area of recommendations address the quality of the publication. We addressed the adequacy of description of the characteristics of the study such as study questions, conceptual framework, study design, description of study population, randomization procedures if any, blinding procedures, data collection procedures and instruments, validity of data collection instruments, definition of and rationale for choice of exposure factors and outcomes, analytical approaches, statistical analysis, and publication of findings.

The third area of recommendations addressed the methods we used to assess the evidence on this topic of youth violence. The criteria driving this evidence review have been effectively applied to a review of the literature relating to the treatment of illness and disease. However, we question whether these exact criteria and methods can be applied effectively in a review of research that examines such a complex social problem such as youth violence. We summarized what this evidence assessment has and has not contributed to the field and made suggestions of how future assessments of evidence could be approached.

It is our hope that this evidence report will provide a basis for future research not only in the area of youth violence, but also in the area of quality of research, quality of publications, and quality of evidence assessment methodology.

Identification of Peer Reviewers

To identify a group of Peer Reviewers, we solicited nominations from our Technical Expert Group, our Panel Chair, and national associations recommended by our Project Officer (including the American Academy of Pediatrics, the American Public Health Association, the American Association of Health Plans, the American Academy of Family Physicians, the American Society of Internal Medicine, the American Psychological Association, and the American College of Physicians, and the Society of Adolescent Medicine). The role of Peer Reviewers is to provide independent feedback about the report. As a result of these solicitations, we received nominations for 24 individuals. These individuals represented federal agencies, academia, philanthropy, clinical practice, and managed care. From this list, the Task Order Project Director invited eight individuals — representing a variety of expertise and geography — to participate. This list of peer reviewers was approved by the Task Order Officer.

Peer Review Process

A copy of the draft evidence report was mailed to each peer reviewer, along with an instruction sheet (Appendix B10 ^*) for reviewing the draft evidence report.A copy of the draft evidence report was also mailed to the members of the Technical Expert Group.All reviewers were asked to respond within three weeks. Six of the eight peer reviewers, six of the nine technical experts, and one AHRQ-appointed peer reviewer provided comments.Appendix D2 ^* lists the names and affiliations of the six peer reviewers who submitted their comments.

Upon receipt of all responses from the peer reviewers and technical experts, the project staff compiled a summary of the comments and changes and revised the draft evidence report accordingly. We submitted a complete copy of each reviewer's comments, together with the report of disposition of those comments to the Task Order Officer for review and approval.

Overview

Of the 11,196 titles and abstracts from the initial and supplemental literature searches, 1,612 (14.4 percent) met our criteria for eligibility for retrieval and secondary review. The results of the primary screening are summarized in Table 3. The reasons for rejecting the remaining 9,584 citations are summarized in Table 4. We were able to retrieve all 1,612 full-length articles. Secondary screening resulted in the acceptance of a total of 466 (28 percent) articles for data abstraction. Of these articles, 404 addressed either the first or second key question for this evidence review (i.e., these articles examined risk factors associated with youth violence), and 66 articles addressed one of the other key questions (i.e., these articles considered outcomes associated with a violence prevention intervention). Four of the accepted articles addressed both sets of questions (i.e., risk factors for youth violence and intervention effectiveness). A summary of the reasons why the other 1,146 articles were rejected is provided in Table 5. During the abstraction process, 201 articles were rejected because, on further review, they did not provide data that could be abstracted; the study outcome was not violence as defined by the project; or the research did not address one of the evidence review's Key Questions. The reasons for rejecting articles during this tertiary review are presented in Table 6. Of the 265 remaining articles, 67 were included in our evidence assessment: 35 for the risk factor questions and 32 for the intervention questions. Figure 3 presents the screening and review process used for the task order. In the following sections, we present the findings of our analysis for each of the five Key Questions.

Key Question #1: What are the factors that contribute to violence and associated adverse health outcomes in childhood and adolescence?

The 35 articles that were included in our assessment reported findings from 23 prospective longitudinal cohort studies; the number of articles per cohort study ranged from 1 to 6. As pointed out in Chapter 2, we used the cohort as the unit of analysis so as to assign the same weight to studies whose findings were published in a single article and studies whose findings were published in multiple articles. A list and description of the 23 prospective cohort studies is provided in Table 7. Additional information about the study subjects and study design for each prospective study is provided in Table 8, including age, gender and race/ethnicity of the study sample, duration of follow-up assessment, the sample size used in reported statistical analyses, and the retention rates.

As noted in Table 8, the 23 cohort studies showed considerable variability with respect to the age at which subjects were first enrolled in the study, ranging from birth to 19 years; the duration of follow-up, which ranged from 1 to 18 years; the sample size, which varied from 86 to 14,358 subjects; and the retention rate, which ranged from a low of 33 percent to a high of 100 percent. We categorized the prospective studies according to sample characteristics, including population types (general population of children or adolescents vs. an at-risk population), gender, and racial/ethnic group; this information is summarized in Table 9. Table 10 summarizes the various outcomes reported in each of the published articles by type of study population, the setting from which subjects were recruited, and sample size.

From the sample-size column in Table 10, we can determine the adequacy of sample size for each subpopulation. Based on the sample size and power considerations in Table 2, a cohort size of 1100 would be needed for the general population and 500 would be needed for the at-risk population, to achieve an 80 percent power to detect an odds ratio 1.5 or higher at 0.05 level of significance. Based on these thresholds, articles that deal with cohorts #8, #9, #12, #14, #22, and #23 for the general population and cohorts #1, #2, #17, and #21 for the at-risk population would not have adequate power to identify risk- or protective factors leading to youth violence. As a result, we lacked adequate statistical power for three of the six subpopulations of the general population, all of which are ethnic subpopulations (A-3: Male, African-American; A-4: Male, White; and A-6: Female, African-American. The three subpopulations with adequate power all represent multiple races/ethnicities. For the at-risk population, only two ethnic subpopulations were studied, each in a single cohort study. Subpopulation B-2, African-American males and females, studied 867 subjects and subpopulation B-4, White males, studied 195 subjects.

We then examined the other descriptor information in Table 10: outcome descriptor, recruitment setting, and age, and observed that no two prospective cohort studies were alike with respect to the type of study outcome descriptor, the recruitment setting, and the age at enrollment. Thus, we did not consider it to be scientifically sound to pool data across the heterogeneous cohort studies using meta-analysis. Instead, we summarized the study findings by population groups according to adequacy and consistency, defined as follows. First, at least two cohort studies must have reported findings for a specific risk- or protective factor: the evidence for a risk- or protective factor was considered inadequate when it was reported in only one cohort study. Second, we classified the risk- or protective factors into three categories of consistency: those consistently reported as being significantly associated with violence (defined as at least 75 percent of the studies that tested an association reporting a statistically significant result); those consistently reported as being not significantly associated with violence (similarly defined as 75 percent or more of studies); and those where studies reported mixed findings. A finding was considered to be statistically significant if the article reported a p-value of less than 0.05. In this summary, we did not use the effect size as a criterion because of the heterogeneity in the number and type of covariates or confounding factors included in the analytic model and the inconsistency in the way effect size was reported in the literature.

Using these criteria, we summarized the findings for each of the study populations that had at least two cohort studies and an adequate number of study subjects together with the combined findings for all 11 population groups (Table 11). As described in Chapter 2, we summarized the findings for each risk- or protective factor by including, first, the number of cohort studies that showed statistically significant findings, followed by the number of cohort studies that showed non-significant findings, both set off by parentheses (a finding that showed a p-value of less than 0.05 was considered statistically significant). We further summarized the findings presented in Table 11 for all study populations, the general population, and the at-risk population in Tables 12, 13, and 14, respectively.

When all population groups were considered, findings for a total of 151 single factors were examined: 85 factors (56 percent) related to the individual risk domain, 40 factors (26 percent) related to the parental/familial risk domain, 11 factors (7 percent) related to the peers risk domain, three factors (2 percent) related to the school risk domain, and 12 factors (8 percent) related to the community risk domain. In addition, 12 multiple factors or pathways examined in this set of literature fell in one or more domain.

As presented in Table 12, seven individual domain risk factors were found to be consistent predictors of youth violence, including male gender, antisocial behavior, alcohol use, alcohol and drug use, drug dealing, weapon carrying, and non-violent delinquency. Of the parental/familial risk factors, only lack of parental attachment was found to be a significant predictor of youth violence. Numerous risk factors across all risk domains were either consistently not associated with youth violence or were associated only in some studies, and many of the remaining factors were examined in only one cohort study [i.e., 47 of 85 (55 percent) factors in the individual risk domain, 22 of 40 (55 percent) factors in the parental/family risk domain, six of 11 (55 percent) factors in the peer risk domain, all (100 percent) factors in the school risk domain, and nine of 12 (75 percent) factors in the community domain]. Consequently, we judged that the strength of evidence was insufficient to be able to determine the predictive power of these factors.

While Table 12 presents findings for all 11 study populations, Table 13 and Table 14 present findings for the three subpopulations with adequate sample sizes for the general population and for the at-risk population, respectively. For the general population (Table 13), findings on 48 risk or protective factors were examined from seven articles based on five cohort studies. Among the 48 factors, 10 factors (20 percent) were examined among two or more cohorts. Male gender (Komro, Williams, 1999; Roitberg and Menard, 1995; Saner and Ellickson, 1996) and alcohol or drug use (Kaplan, Tolle, 2001; Komro, Williams, 1999) were consistently significant risk factors; low socio-economic status (Roitberg and Menard, 1995; Saner and Ellickson, 1996) and living in an urban setting (McNulty and Bellair, 2003; Roitberg and Menard, 1995) were consistently reported as not being significant risk factors. We should point out that Roitberg and Menard, using data from the first five years (1976-1980) of the National Youth Survey of 1,725 Americans who were 11 to17 years old in 1976, reported that although the influence of socioeconomic status (SES) was significant in the first year of observation, the influence of SES was not significant in the subsequent three years of observation. Even for the first year, after adjusting for multiple comparisons, the influence of SES was not statistically significant. The effects of White ethnicity, alcohol use, illicit drug use, occupational strain, unstable family financial base, and low parental education were inconclusive. The evidence for the remaining 38 of 48 (79 percent) factors was derived from only one cohort study, and thus was considered inadequate as a basis for drawing conclusions.

Among boys in the general population, anger (Felson, 1992; Foshee, Linder, 2001), cigarette smoking (Dornbusch, Lin, 1999; Ellickson, Tucker, 2001) and non-violent delinquency (Becker and McCloskey, 2002; Saner and Ellickson, 1996) were consistently reported as significant risk factors for violence. Low socio-economic status (Brezina, 1999; Saner and Ellickson, 1996) was consistently reported as non-significant. Findings were mixed for the seven remaining risk factors with two or more cohort studies, including age, depression, physical aggression, illicit drug use, non-intact family structure, low parental education, and physical abuse by caretakers. Evidence for the remaining 47 of 58 (81 percent) factors were considered inadequate for assessment as they were investigated in one cohort study only.

For girls in the general population, non-violent delinquency (Becker and McCloskey, 2002; Herrera and McCloskey, 2003; Saner and Ellickson, 1996) was consistently reported as a significant risk factor. Illicit drug use (Dornbusch, Lin, 1999; Saner and Ellickson, 1996) and low parental education (Dornbusch, Lin, 1999; Saner and Ellickson, 1996) were consistently reported as non-significant. Age, cigarette smoking, and non-intact family structure showed mixed effects, and the remaining 44 of 50 (88 percent) factors did not have adequate evidence for assessment.

For at-risk youth (Table 14), male gender (in four of five cohort studies) was consistently reported to be a significant risk factor for violence (Foshee, Bauman, 2000; Herrenkohl, Guo, 2001; McCloskey and Lichter, 2003; Rivera and Widom, 1990). One cohort study found that male gender was a significant risk factor for peer aggression and dating aggression but not for aggression toward parents (McCloskey and Lichter, 2003). Low SES was consistently reported as a non-significant risk factor (Herrenkohl, Egolf, 1997; Herrenkohl, Guo, 2001; Herrera and McCloskey, 2001). However, it should be noted that Herrenkohl and colleagues reported a significant influence of SES in bivariate analysis using data from the Lehigh longitudinal study in 1976 and data from the Seattle Social Development Project in 2001, but the influence of SES was not statistically significant after controlling for the influence of other factors in multivariate analysis. Findings on age, depression, empathy, antisocial behavior, individual involvement in pro-social activities, family's pro-violence attitude, parental violence, deviant peers, and the community's low neighborhood attachment were mixed. Finally, evidence for the remaining 41 of 52 (79 percent) risk factors was inadequate for assessment. One cohort study (Herrera and McCloskey, 2001) found increasing age as a significant risk factor, while two other cohort studies (Herrenkohl, Egolf, 1997; Rivera and Widom, 1990) found age to be a non-significant risk factor.

Among high-risk males, Latino ethnicity and repeated physical aggression (Loeber, Wei, 1999; Loeber, Wung, 1993) were consistently reported as significant risk factors. Impulsivity/attention deficit/hyperactivity disorders were reported as non-significant risk factors in two studies (Becker and McCloskey, 2002; Beyers, Loeber, 2001). Findings regarding an association with age, African-American race, positive attitude toward problem behavior, lack of guilt, having previously engaged in sexual intercourse, weapon carrying, violent behavior, poor academic performance, parental supervision or monitoring, and poor familial communication patterns were all found to be inconclusive. Evidence for the remaining 45 of 58 (78 percent) factors were judged inadequate for assessment, while findings for age were found to be mixed, with two cohort studies reporting it as a significant risk factor (Becker and McCloskey, 2002; Zhang, Loeber, 1997) and one study reporting it as a non-significant finding (Beyers, Loeber, 2001).

Among high-risk females, no factors were consistently reported as significant. Findings for 2 of the 32 risk or protective factors were mixed, and the evidence for the remaining 30 (94 percent) was judged inadequate for assessment.

In summary, although many risk factors were studied in the 23 prospective cohorts reviewed, 58 percent (87 of 151) of the risk factors were examined within only a single cohort study; 34 percent (52 of 151) of the risk factors had mixed findings; and only 8 percent (12 of 151) of the risk factors were consistently reported to be significantly associated with youth violence. Further, those factors that were consistently reported as significant or non-significant in this report were considered without regard to the type of violent outcome, the age at enrollment in the cohort, the type of at-risk population, and the type of analysis. Thus, additional research is needed to assess whether the associations are affected by these factors. Although we considered using meta-analysis techniques to pool findings, we found that the published data were too inconsistent and/or inadequate to allow the use of this technique.

Key Question #2: What are the patterns of co-occurrence of these factors?

While Key Question #1 was intended to identify independent risk factors that have a high likelihood of leading to youth violence, Key Question #2 was intended to identify clusters of risk factors that may lead to youth violence. The analytic approach to the two key questions is different. Very frequently, a factor that is found to be statistically significant in a univariate or bivariate analysis becomes non-significant after adjusting for other factors in the model. For example, low SES or low family income was reported to be a significant risk factor associated with youth violence in bivariate analysis. The association disappeared (became non-significant) after controlling for the effect of other risk factors in the multivariate model. (Roitberg and Menard, 1995; Saner and Ellickson, 1996; Herrenkohl, Egolf, 1997; Brezina, 1999; Herrenkohl, Guo, 2001; Herrera and McCloskey, 2001) In contrast, Key Question #2 purports to identify clusters of risk factors that occur simultaneously (co-occur) and appear to predispose to youth violence. Unlike the independent factors identified in Key Question #1, the factors in a cluster are likely to be highly correlated and if we subject them to adjustment in multivariate analysis, many will likely become non-significant. Therefore, analytically, while we are looking for independent risk or protective factors in Key Question #1, we are looking for dependent risk or protective factors that occur simultaneously in Key Question #2.

Operationally, we defined co-occurrence as the simultaneous presence of two or more risk or protective factors that predict violence in an individual. Of the 23 longitudinal prospective cohort studies included in our assessment, only five articles from four cohort studies (Beyers, Loeber, 2001; Borowsky, Ireland, 2002; Dornbusch, Lin, 1999; Herrenkohl, Maguin, 2000; Piquero and Tibbetts, 1999) examined different types of co-occurrence of risk- or protective factors..

Using data collected from the National Longitudinal Study of Adolescent Health (Add Health), Dornbusch and colleagues (Dornbusch, Lin, 1999) examined the relationship between young people's use of drugs and involvement in violence and found polydrug use was significantly and positively associated with increased involvement in violence among both boys and girls. This same association was not found between use of a single illicit drug and violence. Also using Add Health data, Borowsky and colleagues (Borowsky, Ireland, 2002) reported the protective nature of three factors, including parent-family connectedness, school connectedness/parental presence, and grade point average; all three were found to be significantly and negatively associated with violence among both males and females. As part of the Seattle Social Development cohort study, Herrenkohl and colleagues (Herrenkohl, Egolf, 1997) reported that youth exposed to multiple risk factors were significantly more likely than non-exposed youth to engage in violence. However, in this study, only the number of risk factors and not the type of factors that co-occurred were reported. In a study that focused on repeated incidence of youth violence among high risk males recruited for the Pittsburgh Youth Study, Beyer and colleagues (Beyers, Loeber, 2001) reported that two combinations of risk factors were significantly associated with repeated violence: 1) low SES neighborhood, lack of guilt, early sexual activity, carrying hidden weapons, and poor communication at home, and 2) high SES neighborhood and physical aggression on the part of the youth. Importantly, whereas SES was consistently not reported as a significant “independent” risk factor in Key Question #1, SES was a significant risk factor when it co-occurred with other risk factors as determined in our analysis for Key Question #2. Data from the nationwide Collaborative Perinatal Project, which followed a cohort of African-American children from birth, Piquero and Tibbetts (Piquero and Tibbetts, 1999) reported that pre/perinatal disturbances — when combined with a disadvantaged familial environment at age seven years — were associated with increased risk for criminal offense during early adulthood among a high-risk, inner-city sample from Philadelphia.

Viewing the findings for both Key Questions #1 and #2, it is evident that both the dependent and independent nature of the risk or protective factors must be properly assessed and clearly differentiated. Until this is done, controversies regarding the significance or non-significance of risk or protective factors will persist.

Key Question #3: What evidence exists on the safety and effectiveness of interventions for violence?

A total of 32 articles were selected to address Key Questions #3, #4, and #5, the questions that address safety and effectiveness of interventions. Table 17 provides the numbers of articles by intervention level (primary, secondary, and tertiary prevention, according to the definitions provided in Chapter 2) and by study design. Two articles reported findings for primary and secondary prevention interventions in the same article. Thus, a total of 34 intervention studies are summarized in this table. Table 18 lists the unit of randomization for randomized controlled studies (RCTs) and the name of the intervention. We provide a description of each intervention program and its findings for the five primary prevention interventions conducted by RCT in Table 19, for the 10 primary prevention interventions conducted using other study designs in Table 20, for the secondary prevention interventions conducted by RCT in Table 21, for the four secondary prevention interventions conducted using other study designs in Table 22, for the two tertiary prevention interventions evaluated by RCT in Table 23, and for the five tertiary interventions evaluated using other types of study designs in Table 24.

For this assessment, we considered an intervention program effective when at least one violent outcome indicator was found to change significantly at the p<0.05 level after the intervention. When no significant change in violent outcome indicators occurred at the p<0.05 level, we considered the program ineffective.

Of the 32 intervention studies, 13 were evaluated using a RCT, and 19 were evaluated using other study designs. Of the 13 RCTs, five incorporated primary prevention interventions (Bosworth, Espelage, 2000; Farrell, Meyer, 2003; Foshee, Bauman, 1998; Foshee, Bauman, 2000; Orpinas, Kelder, 2000; Perry, Komro, 2003); six incorporated secondary prevention interventions (Foshee, Bauman, 1998; Foshee, Bauman, 2000; Friedman, Terras, 2002; Hanlon, Bateman, 2002; Ludwig, Duncan, 2001; Moore, Armsden, 1998; Simon, Sussman, 2002); and two incorporated tertiary prevention interventions (Henggeler, Clingempeel, 2002; Scott, Tepas, 2002). Of the five RCTs used to evaluate primary prevention interventions, one (Farrell, Meyer, 2003) was reported to be effective (20 percent). Of the six RCTs for secondary prevention interventions, three (Hanlon, Bateman, 2002; Ludwig, Duncan, 2001; Moore, Armsden, 1998) were reported to be effective (50 percent). And of the two RCTs for tertiary prevention interventions (Henggeler, Clingempeel, 2002; Scott, Tepas, 2002), both were found to be effective (100 percent).

Of the 19 interventions using other study designs, 10 evaluated a primary prevention intervention, four evaluated secondary prevention interventions, and five evaluated tertiary prevention interventions. Four primary prevention interventions (40 percent) (DuRant, Treiber, 1996; Hawkins, Catalano, 1999; O'Donnell, Stueve, 1999; Reynolds, Temple, 2001), one secondary prevention intervention (25 percent) (Hammond and Yung, 1991), and three tertiary prevention interventions (60 percent) (Borduin, Mann, 1995; Morrissey, 1997; Stein, 1999) were reported to be effective. The findings of one tertiary intervention (Hagan, King, 1994) were reported to be inconclusive.