National Institutes of Health Pathways to Prevention Workshop: Advancing Research to Prevent Youth SuicideFREE

Table 1 shows recommendations for improving data systems that focus on better identifying persons at risk for suicide and advancing knowledge of risk factors. The availability of effective data systems for examining risk and outcomes is limited. For example, the accompanying systematic evidence review (2) identified only 6 of the 153 suicide prevention studies that linked suicide data from multiple sources, which limited researchers' capacity to study determinants, mediators, and moderators of suicidal behaviors and suicides. Poor documentation of interventions (for example, scarcity of usable data dictionaries and comprehensive clinical records) compounds the problem of inadequate data systems. The lack of comprehensive linked data resources makes it difficult to identify at-risk persons.

Because not all states mandate use of cause-of-injury codes, the ability to understand the magnitude of suicide, suicide attempts, and risk factors of suicide is hampered when they are not used by medical departments or on insurance claims. Obtaining a complete picture of suicide and related factors would require federal mandates for health care providers to use cause-of-injury codes. Improved standardization and consistent use of such codes would enable researchers to conduct more accurate studies of suicide risk and prevention. Without additional mandatory coding, it is not possible to determine whether the death was a homicide, suicide, or accident. Implementing mandatory external cause-of-injury coding would enable public health officials and prevention science researchers to capitalize on these data to identify means of suicide. This recommendation is paramount because nearly half of reported suicides involve firearms as the means.

Additional surveillance on suicides is paramount to obtain a complete picture. The National Violent Death Reporting System provides an example of a surveillance system that could yield insights into the causes of and context for suicides by linking data from death certificates, law enforcement reports, crime laboratories, and medical examiner reports. These data are not available in all states, and the reporting only addresses deaths from suicide. However, surveillance systems should be expanded to capture suicide attempts and related behaviors. Linking surveillance and administrative data should be encouraged because it would better inform suicide risk and behaviors that could lead to suicide. Researchers need training to identify and obtain permission to use data sources found in school, municipal, state, and federal records (for example, to make sure consent forms explicitly ask permission to link with other data sources).

Policies at the state and community levels share a role in improving our ability to understand suicide and suicide attempts. State all-payer claims databases could provide communities with local data about suicide and suicide attempts. Under the Patient Protection and Affordable Care Act, accountable care organizations or health information exchanges could provide local communities with population data. Claims databases can be cumbersome, require extensive cleaning and specialized expertise to analyze, and are not always timely. Some of these limitations can be addressed through coordinated efforts at the state level. By improving these systems, states have the opportunity to be innovative with syndromic surveillance data, which could be used to identify patients who need better care management or to help target community-level interventions.

Some policy and practice issues are difficult to rectify because of social stigma, governance, conflicting legal goals, a fragmented death scene investigation system, silos of isolated research teams, and unique data systems. Until suicide and mental health issues are destigmatized, reporting will be inaccurate. Reporting and tracking suicidal behavior and its precursors are hampered by disincentives embedded in policies and practices from the federal to the local level. Families and medical providers are often reluctant to label events as suicides or suicide attempts for many reasons, including legal concerns, cultural issues, community referral patterns, and the lack of standard procedures for investigating suicide death scenes.

Table 2 depicts the panel's recommendations for improving research design and analysis of complex systems. Descriptions of several innovative and promising techniques follow.

Conduct measurement at multiple levels. Measurement is an especially ripe area for improving data systems in suicide prevention research. Measuring risk and protective processes at several levels—including the individual, family, peer group, school, and community—facilitates the process of investigating and understanding the complex factors central to suicide risk across diverse populations. At the lowest level, information on biomarkers and biological processes is important for advancing the continuum of suicide research, from surveillance to basic research and then prevention studies.

Novel ways of integrating neurobiological measures into the science of suicide prevention research are needed. Schools could collect electronic data on student head injuries, for example, and link these to school and health records to help identify youth at increased risk for suicide. Biological measures may improve the effectiveness of evaluation research. Evidence for the protective effects of mindfulness and meditation practices would be strengthened by including biological measures, such as cortisol. Incorporating these measures into studies of suicide risk and prevention will help identify potential treatment approaches for ameliorating the adverse effects that trauma and stress have on youth suicide risk.

Psychological and developmental processes also play key roles in suicide risk and prevention. Workshop participants noted the need for psychometric work addressing the measurement of personal characteristics, such as sexual orientation and identity, and for processes displaying universal prominence and culture- and context-specific importance across diverse populations. Rather than adapting existing measures to new cultural contexts, direct development of theoretically informed measures for a given cultural context is warranted.

For persons identifying as sexual and gender minorities, measures of peer and self behaviors and attitudes salient to gender identity and sexual orientation can help to better identify the correlates of suicide risk. Research indicates higher rates of suicide among transgender youth than gay or lesbian youth. General methodological improvements in measurement, such as visual analogue scales, computerized adaptive testing, and multiform questionnaire protocols to collect data, would increase the generalizability of findings.

Perhaps the least extensively investigated domain concerns measuring the settings and contexts beyond the individual and family levels. Ecometrics, the measurement of environmental contexts, is essential to accommodate the multilevel analytic approaches needed for this field of research. Innovation in this area includes direct assessment of constructs, such as climate and aggregate indicators that can come from linkable administrative data (for example, police records).

Assess developmental and longitudinal change. Despite the importance of dynamic change processes in youth suicidal behavior, few studies have addressed how changes and reciprocal influences among risk and protective factors influence youth suicidal behaviors across multiple time scales (short- and long-term changes). Longitudinal information is lacking about the processes and provider practices occurring among suicidal youth, particularly right before suicide. Our capacity to design interventions aimed at preventing suicide depends on longitudinal research that can better capture the complex interplay among imminent and long-term factors in predicting suicide ideation and attempts.

By incorporating a broader repertoire of predictors into a longitudinal context—whether in a single study or through the use of linked studies—we are better positioned to understand the mediating, moderating, and reciprocal mechanisms underlying suicidal behavior. Studies integrating qualitative and quantitative data on these mechanisms will then better inform approaches to optimally time and maximize prevention efforts. Children who question their gender identity, for example, may be rejected by parents, teachers, or peers; rejection, in turn, may increase a child's social isolation and depressive symptoms that can further escalate rejecting behaviors.

Measurement and design strategies that facilitate the study of changes over time and across developmental periods will help inform the timing and targets for interventions that can interrupt the recursive cycle of negative social interaction. Growing evidence points to the potentially powerful effects of short-term predictors (for example, insomnia, exposure to coping or self-regulation skills, peer support, intervention efforts from teachers, and real-time sharing for care management) on longer-term suicidal prevention processes. Such cascading, multiple time-scale effects offer a renewed way to conceptualize and test mediation and moderation. To evaluate the long-term effects of intervention programs, we recommend that researchers collect and integrate measurements from multiple time scales, including measures of likely mechanisms of change. Integrating evidence-based results with theory and methods will help ensure high-quality, effective suicide prevention efforts.

These modeling efforts are enhanced by using latent variable approaches to test critical assumptions, such as the psychometric equivalence of constructs across time and subgroups, and adjust important estimates for various sources of measurement and sampling error. At group and network levels, powerful methods exist for modeling important effects, such as diffusion, contagion, selection, and socialization as well as propagation of risk or protective factors and associated processes.

Model multilevel structure. Compelling evidence exists for the multilevel nature of factors and processes tied to youth risk for suicidal behavior. Research rarely assesses or analyzes the interrelated and nested social processes and structures tied to suicide risk, particularly at higher levels of influence, such as the school, neighborhood, and community. Settings at a higher, more distal level (for example, community) can have a cascade of effects on youth suicide risk by shaping family and individual functioning. Studies of multilevel effects on suicide risk suggest that interventions addressing community- and family-level factors may affect many persons to a greater extent than typical individual-level interventions alone. Variations of universal intervention programs sensitive to multilevel structures should be designed and evaluated to effectively target individual- and higher-level risk factors. Multilevel analytic techniques help adjust for issues of known clustering (for example, families nested in communities) and thus can capture the heterogeneity across multiple levels and cross-level mediation or moderation effects. Several methodological challenges must be addressed to estimate multilevel effects on youth suicide risk. These challenges also manifest in the measurement and design needs to adequately represent the various levels of a multilevel structure.

Examine known and unknown subgroups. Subgroups and subpopulations contribute to the heterogeneity of study cohorts. These subgroups can have differential effects and patterns of change. Methods to model known and unknown heterogeneity can identify and explicitly delineate these differential effects. When suicide risk groups are known (for example, those defined by gender identity and orientation), membership can be explicitly compared as multiple groups to examine various influences, including moderation by group membership. When not explicitly known, the various suicide risk subgroups that are often embedded in universal programs can be identified. Mixture modeling identifies subgroups of persons whereby predictors and outcomes of group membership can inform the differential effects and outcomes of suicide prevention research.

Integrate and link data across studies. Another recommendation involves coordinating efforts in the broader research community. Integrative data analysis uses a set of common measures across 2 or more studies to link the data. These linked studies can be combined as an integrated data set that allows greater overall power to identify hard-to-detect mediating and moderating mechanisms and greater representation of suicides, which are infrequent in any given study or setting. Including common measures and linking items across projects, coupled with principled treatment of the missing data, would expand the power and validity of the larger research portfolio sponsored by funding agencies. The data archive of the National Institute of Mental Health (http://rdocdb.nimh.nih.gov) is an important sharing platform for integrative data analysis, but the linking information must be coordinated and highlighted.

Use stronger inference strategies. Most randomized, controlled trials (RCTs) are characterized by strict exclusion criteria that limit generalizability for suicide-related research. Recent methodological advances, however, offer alternative methods to strengthen valid interpretations from non-RCT data. Propensity score methods can be used to study group differences, probe for unmeasured confounding effects, control selection effects in mediation analyses, and infer aggregate effects in studies where random assignment is not possible. Further, quasi-experimental designs, such as the regression discontinuity design, are relevant for suicide prevention research. Cross-design synthesis also can be used to help combine RCT and observational data. These designs facilitate valid inferences based on targeted variables and can account for the effects of moderators across the range of studies. Of note, multidisciplinary collaborations can integrate the strengths of multiple techniques to overcome weaknesses and the restrictive assumptions of any single technique or study.

Table 3 provides recommendations for building and strengthening collaborative efforts among researchers, methodologists, and practitioners. Building a coordinated research and practice community would foster data linking, translating research to practice, and disseminating aggregated data needed for community planning. The research community would need to coordinate at multiple levels, including developing requests for proposals, having preaward discussions with program officers, and sharing across projects among various principal investigators of funded research. The practice community would also need to coordinate the coupling of administrative data.

Interdisciplinary collaboration is critical to identify highest-risk persons for inclusion in targeted prevention efforts (both universal and indicated approaches). Youth who die by suicide may not have had prior contact with mental health providers; however, they may have interacted with educators, coaches, medical providers, and other community members. Research identifying effective policies, such as gun control, to prevent the means to suicide events is needed. Similarly, population-based efforts can and should draw on cross-sector collaborations (for example, schools, law enforcement, parks and recreation departments, and faith-based organizations) to strengthen protective factors in individuals, families, and communities. Recognizing the broader costs and effects of youth suicide is a critical part of policy agenda that can be addressed only by strengthening the larger community of researchers, practitioners, and stakeholders.

Participation in education and training opportunities is also needed to build and expand the research and practice infrastructure. Providers, agencies, families, and communities need education on the importance of removing the stigma associated with suicide. Improving messaging and social norms around mental health and suicide may help destigmatize suicide and promote connectedness within families and communities. Training in the advanced design and analysis techniques described here must be made readily available, and all members of collaborative teams should be given access to these training opportunities. Broadening the understanding of the merits of using the workshop presenters' recommended procedures is critical to bringing these procedures into the realm of standard practice.

As researchers and practitioners, we must unite to stop youth suicide in order to circumvent its associated economic cost and devastating pain and suffering. We must build and strengthen both coordination and collaboration among all members of the larger policy, practice, and research communities. We need to improve and coordinate the various surveillance and administrative data systems across these sectors. We must also elevate the level of rigor and breadth of methods used in studies of suicidal behavior. Adherence to the recommendations summarized herein provides us with a roadmap to our ultimate goal—eliminating suicide.