The vision of the National Action Alliance for Suicide Prevention (NAASP), an initiative launched by the U.S. Department of Health and Human Services in 2010 (actionallianceforsuicideprevention.org), is “a nation free from the tragic experience of suicide.” In pursuit of this vision, the NAASP Research Prioritization Task Force has set out to build a research agenda that, if fully implemented, will reduce suicide deaths and attempts by 20% within five years. This goal translates into preventing about 7,200 suicide deaths and 130,000–220,000 attempts each year by 2018 (1,2).
The NAASP vision cannot be achieved through research alone. Research can point to promising directions, but action—making improvements and monitoring success—is the purview of policy makers, agency heads, and system leaders. The key is in knowing whom to target, with which interventions, and in what order of priority. The approach taken by the National Institute of Mental Health (NIMH) to address these questions is to build upon existing research to stratify risk among subsamples of the population, identify effective interventions (such as clinical interventions, organizational practices, and policies), simulate their impact through modeling exercises, and boost accountability for results when simulations show that large reductions in suicide are possible. Similar approaches have proved effective in reducing mortality from other causes (for example, heart attacks and alcohol-related traffic accidents), suggesting that steep reductions in suicide are also possible (3).
As suicide deaths in civilian and military populations continue to climb (4), limited application of suicide prevention research and other well-intentioned efforts to reverse this trend have proved inadequate. In 2009 deaths classified as suicides in the United States exceeded the number classified as homicides or motor vehicle accidents (1). At the same time, new empirically supported approaches that are being tried outside the United States to intervene with individuals who attempt suicide (5) and boost the quality of care in health systems (6) present opportunities for implementation research in real-world settings. In the United States, the Henry Ford Health System’s Perfect Depression Care program (7) has illustrated that adoption of a “zero-suicide goal” is achievable when organizations apply the best science and demand accountability for high-risk populations.
It is time to strategically apply science and accountability to the mounting public health problem of preventable suicide. The purpose of this Open Forum is to galvanize researchers to help refine this approach and overcome data limitations and methodological challenges in order to prioritize suicide prevention efforts with the greatest potential public health benefit.
To prioritize funding decisions, NIMH often weighs different research pathways for reducing morbidity and mortality associated with mental disorders. Adapted to suicide prevention, this approach suggests four steps: develop a taxonomy of high-risk target subgroups, identify and pair effective practices and policies with specific high-risk groups, estimate the potential impact of implementing effective interventions effects within targeted intervention platforms (for example, health care systems, communities, and workplaces), and estimate time horizons for intervention implementation and future research efforts. [A figure representing this conceptual approach is available online in a data supplement to this report.]
Step 1 requires characterizing the population by concentration of risk and affiliation with intervention platforms. Identifying subgroups with the greatest risk density—in terms of overlapping risk factors and numbers of individuals—can maximize intervention efficiency. Affiliation of subgroups with intervention platforms confers two important benefits. It overcomes problems of case identification and barriers to intervention, which hamper efforts to reduce suicide in the public at large. Equally important, it identifies entities (for example, communities, schools, health care systems, military organizations, and American Indian tribes) with accountability for a subgroup’s health. Thus leaders of these intervention platforms may be called upon to take action when given compelling evidence of the benefits of doing so.
Step 2 involves sorting suicide prevention interventions by strength of research support and applicability to subgroups of interest. Evidence is a matter of degree and pertains to a range of issues, including safety, efficacy in research trials, effectiveness in real-world settings, and implementation feasibility. Applicability to particular subgroups may be facilitated by sorting suicide prevention practices as universal, selected, or indicated (8,9). Universal interventions are designed to reach entire populations and include public health campaigns; education, training (10), and screening (11) programs; and limits on access to lethal means (12). Selected interventions are designed for individuals and groups at increased risk. Indicated interventions for those with a history of suicidal behavior include psychotherapy (13) and medications to treat underlying mental illness (14), among others. All these interventions exist in practice, but they vary in the strength of empirical support.
Step 3 involves statistical modeling to estimate the reductions in suicide that could be achieved by implementing specific interventions with particular subgroups on particular intervention platforms. Estimation modeling has been used in other areas of public health but less so with suicide (15). In step 3, gaps in knowledge required to build estimation models (for example, robust effect size estimates) will become apparent and can be added to the emerging research agenda.
Step 4 is to assess time horizons for implementation. Some interventions will be better suited for broad-scale implementation in the near term given factors such as agency staffing, system infrastructure, available technology, financing, and political will. Similarly, research studies will vary in their short- and long-term promise for informing suicide prevention efforts.
Steps 1 and 3 in this approach—develop an optimal taxonomy of high-risk subgroups and estimate the impact of interventions—may be the most challenging. We turn to them now to describe the basic processes involved, illustrate how these steps might be taken, and discuss challenges that must be addressed for this work to yield useful, compelling results.
Efforts to identify individuals at greatest risk of suicide often compare suicide rates on the basis of demographic characteristics; risk factors, such as mental illness or substance abuse; geographic region; suicide setting; or means of death. Each comparison suggests a hierarchy of intervention targets for lowering the national suicide rate, but none is sufficient. A systematic analysis of various combinations of these characteristics for individuals affiliated with feasible intervention platforms is needed to identify large, accessible subgroups with highly concentrated risk in order to strategically target suicide prevention efforts.
Consider an example using data on suicide attempts from the 2008–2009 National Survey on Drug Use and Health (NSDUH) (2,16). According to survey data, more than two million adults were treated each year (2008 and 2009) in specialty facilities for substance use (2,16). Of those, about 395,000 (17%) reported suicidal thoughts, and of these, 106,000 (5% of those in treatment) attempted suicide in the past year. [A figure illustrating these subgroups is available in the online supplement to this report.] Presence of a substance use disorder is a known risk factor for suicide (17), borne out by the fact that almost 5% of adults receiving specialty care for substance use attempted suicide in the past year, compared with .5% of the U.S. adult population (1). This constitutes about a tenfold concentration of suicide risk for adults treated in specialty facilities for substance use. Among adults in treatment who had suicidal thoughts, the suicide attempt rate was 27%. Among those who made a suicide plan, 61% attempted suicide within the year. It is feasible that routine screening for suicidal thoughts and plans could be implemented in substance abuse treatment facilities to identify high-risk cases that would benefit from assessment and intervention. If, for example, an intervention with 50% effectiveness was fully implemented for adults with suicidal thoughts in substance abuse treatment, 53,000 suicide attempts could be averted (assume for simplicity’s sake that suicide is equally likely before and after substance abuse treatment and that the intervention is equally effective for individuals with suicidal thoughts who do and who do not subsequently attempt suicide). The intervention efficiency ratio would be 13% (53,000 suicide attempts averted through intervention with 395,000 individuals).
Introducing a second risk factor—serious mental illness—can increase intervention efficiency by improving targeting specificity. NSDUH respondents’ serious mental illness status was estimated. Of the 395,000 individuals in substance use treatment who had suicidal thoughts, 238,000 (60%) were estimated to have a serious mental illness, 80,000 (20% of those in specialty care for substance use who had suicidal thoughts) of whom attempted suicide in the past year. [A figure illustrating these subgroups is available in the online supplement to this report.] If an intervention with 50% effectiveness was implemented for adults with serious mental illness and suicidal thoughts in substance abuse treatment, 40,000 suicide attempts could be averted. In this case, the intervention efficiency ratio rises to 17% (40,000 suicide attempts averted by intervening with 238,000 individuals). The trade-off for higher intervention efficiency is decreased reach. With two-factor targeting (serious mental illness and suicidal thoughts), only 75% of those in treatment for a substance use disorder who would subsequently attempt suicide would receive the intervention; 25% would be excluded because they did not have a serious mental illness. Thus a consequence of adding serious mental illness to increase risk concentration might be failure to avert 13,000 attempts.
This case involves a definable subgroup of substantial size, with concentrated risk and affiliation with a viable intervention platform. Despite its limitations, this simplified example illustrates the balance that must be struck in identifying an optimal set of target subgroups given trade-offs between concentration of risk and intervention efficiency, targeting specificity, and reach. Much work is required to systematically consider the scientific evidence in regard to various subgroups’ suicide rates, relative size in the general population, and availability of care in order to construct an optimal taxonomy of subgroups for strategic intervention.
The purpose of estimating the impact of effective interventions with high-risk subgroups is to find leverage points for reducing suicide. Complete discussion of this process is beyond the scope of this Open Forum, but another example can illustrate its potential utility. The 2008–2009 NSDUH data indicate that almost 62 million adults per year were treated in an emergency department for any reason one or more times in the past year (2,16). Had these individuals been asked about suicide, some 686,000 (1%) would have reported a suicide attempt in the past year (2,16). We don’t know from these data how many of these attempters would subsequently die by suicide, but we can estimate on the basis of other studies. A meta-analysis by Owens and colleagues (18) of studies of repetition of self-harm suggested that about 2% of individuals seen in hospitals for self-harm will die by suicide within one year. The meta-analysis involved individuals receiving care specifically for self-harm, whereas the NSDUH data include individuals seen for any reason, which may mean that the NSDUH sample is lower risk than the meta-analysis sample. Therefore, we apply a more conservative estimate of suicide reattempt to the NSDUH data—say, 1%–2%—and estimate that 6,860–13,720 of the 686,000 prior attempters seen in emergency departments could be expected to die by suicide within one year. If, however, these past attempters were routinely identified and treated when they are seen in emergency departments, some proportion of lives presumably may be saved and subsequent suicide attempts averted.
Brown and colleagues (19) tested the benefit of cognitive-behavioral therapy (CBT) over treatment as usual for preventing suicide reattempts among adults seen in emergency departments after an index suicide attempt. The authors found that 42% of those receiving usual care during the 18-month follow-up made at least one subsequent suicide attempt, whereas only 24% of adults treated with CBT did so. Combining these findings with those of Owens and colleagues and applying them to the NSDUH data for a simple illustration, we can extrapolate that about 1,201–2,402 lives a year may be saved by delivering CBT to adults seen in emergency departments for suicide attempts. This represents 3%–6% of the 36,035 suicide deaths that occurred in the United States in 2008. [A table showing these calculations is available in the online supplement to this report.]
This example has limitations. The populations and time frames used differ in ways likely to affect estimates of reattempts, deaths, and lives saved. Moreover, although a suicide attempt and subsequent death are associated, the nature of that association is complicated. Rigorous statistical modeling in step 3 must attend to these issues and include more parameters than are used here and with greater specification. However, this example suggests the potential of more rigorous modeling to produce knowledge that decision makers can use for strategically targeting suicide prevention resources.
Simulation of the impact of suicide intervention can inform research, practice, and policy, but the ultimate utility of such modeling depends on whether data limitations and methodological challenges can be addressed. Some of these are discussed here.
Step 1 depends on the timeliness, scope, and quality of population-based data on correlates of suicidal behavior. Existing data are limited. There is a delay of three or more years in availability of national mortality data. Data on correlates of suicide death come mainly from death certificates, and thus such correlates are limited to demographic characteristics. Additional detail is available from 18 states participating in the National Violent Death Reporting System, but data from this system are also limited for the purposes discussed here. More generally, there are no universally used standard criteria for defining death as suicide on death certificates, and the process for determining manner of death varies considerably across jurisdictions.
The richest detail on correlates of suicide comes from “psychological autopsy” studies, but these are not designed for rapid or broad surveillance. National “followback” research— for example, the 1993 National Mortality Followback Survey, which sampled from death certificates for greater detail—could fill gaps in knowledge about correlates of suicide. Oversampling of suicide deaths in 1993 allowed several studies to examine additional risk factors for suicide (20). Linkage of death certificates with health insurance or medical records could fill other knowledge gaps.
Available data on nonfatal suicide events are also limited. The NSDUH data used here are broad and nationally representative, but suicidal behavior is based on self-report, and there is no way to observe subsequent mortality or to relate nonfatal events and other characteristics with death. Representative longitudinal surveys, such as Add Health or the Health and Retirement Study, generally have too few participants for robust assessment of suicide. Some population-based data on nonfatal suicide events are available from health care delivery systems. It is possible to identify self-harm in medical claims or encounter data via “E” diagnosis codes from the International Classification of Diseases; however, these codes are used inconsistently in many clinical settings. Electronic health record systems may provide richer information on nonfatal suicide events, but health care data are only relevant for cases in which individuals seek or receive treatment.
Intervention research literature
Few studies have evaluated outcomes of suicide prevention strategies (21); fewer still have looked at specific population subgroups (22). Given the low base rates of suicide attempts and deaths, studies rarely have power to detect the impact of an intervention on these outcomes. Consequently, many studies use proxy outcomes, typically suicidal ideation, making it difficult to model population-level impacts on attempts and deaths. Effective treatment of psychiatric and substance use disorders holds promise for reductions in suicidal behavior (7); unfortunately, most studies are powered only to examine posttreatment effects on primary outcomes of interest (for example, reductions in depression or substance use) and few studies report longer-term outcomes, including suicide attempts or deaths. Although the inclusion of suicidal individuals in psychiatric clinical trials can be done ethically (23), their exclusion from treatments for depression, schizophrenia, anxiety, and substance use disorders continues to hamper conclusions about the impact of these treatments on suicide.
Suicide prevention studies also vary in method and rigor within and across population subgroups. Inconsistencies in assessment and reporting of participant characteristics hamper application of study results to specific subgroups. Differences in the operationalization of suicide outcomes complicate synthesis of information across studies, and reviewers struggle to account for variation in methodological rigor when integrating results (24).
Additional study characteristics have implications for the generalizability of results to community practice settings. Effect sizes observed in efficacy studies usually exceed those in effectiveness studies, which involve more typical patients, clinicians, and settings (25). Moreover, effect sizes in research trials generally surpass those in practice settings (26). Commonly hypothesized sources of the research-to-practice “voltage drop” in effect sizes are the many differences between study settings and real-world practice settings (25–27). Thus modeling the impact of evidence-based interventions on a given subgroup requires estimation of key parameters that can affect the generalizability of results.
These limitations can be addressed in future efforts to model the impact of interventions in preventing suicide. Modeling exercises would benefit from development of methods for extrapolating results from research studies to specific subpopulations, including strategies for estimating the similarity between research samples and target populations and developing weights for simulations. Similarly, methods for identifying clinician- and setting-level characteristics that have an impact on intervention fidelity and outcomes and approaches for modeling these influences could also yield more accurate estimates of effect sizes and corresponding confidence intervals to be anticipated in community practice settings (28).
Rapid, substantial reductions in suicide will require targeted delivery of effective interventions to well-defined, relatively large subgroups of high-risk individuals affiliated with intervention platforms. A four-step approach can produce the knowledge needed to take these strategic actions: develop a taxonomy of target subgroups, identify and pair effective practices with specific high-risk groups, simulate intervention effects in targeted care systems, and estimate time horizons for intervention implementation and future research efforts. This process will reveal leverage points for prevention interventions to be exploited today and generate research questions to further improve outcomes in the future. Taken to scale via strategic implementation of existing evidence-based interventions, rapid development of new interventions to close gap areas, and measures to increase accountability for quality of care, this approach holds promise for reducing the morbidity and mortality burden of suicide.
The views expressed here do not necessarily represent the views of the National Institute of Mental Health, the National Institutes of Health, the U.S. Department of Health and Human Services, or the U.S. government.
The authors report no competing interests.