² In order to balance comprehensiveness of sources cited, while also recognizing brevity of this manuscript format, we did the following to identify sources to inform the SESPM. We searched Pubmed, Medline, Psychinfo, and Psycharticles using combinations of the following phrases: “suicide,” “risk factor,” “protective factor,” “prevention,” “intervention,” “review,” and “meta-analysis” while focusing on articles from 1980 to present. Reviews and meta-analyses were given priority because we aimed to provide a big picture review (see Table 1). We further used Google Scholar to identify pertinent content from the following major professional organizations: American Foundation for Suicide Prevention, Suicide Prevention Resources Center, American Association of Suicidology, World Health Organization, CDC, and Substance Abuse and Mental Health Services Administration. Once a full set of key sources was identified, we dropped sources that were completely redundant with others.

Selection of the final integrated body of existing evidence represented scoping/conceptual summaries (Bryan and Rudd, 2006; Van Orden et al., 2010; Drapeau and McIntosh, 2014; Bernard et al., 2015), systematic reviews/meta-analyses (Serafini et al., 2012; Calear et al., 2016; Chan et al., 2016; Ma et al., 2016; Zalsman et al., 2016; Franklin et al., 2017), mortality risk studies (Björksenstam et al., 2015, 2016; Madsen et al., 2017), measure development (Linehan et al., 1983), and policy analysis (Anestis and Anestis, 2015) into a unified SESPM framework, as presented in Table 1. In doing so, we differentiate factors widely applicable across groups vs. those that tend to demonstrate population-specific associations with suicide (e.g., military veterans, youth, lesbian, gay, bisexual and transgender [LGBT] persons). For example, concerning LGBT youth, literature consistently links population-specific experiences of internalized stigma and victimization as associated with suicide risk; moreover, sexual orientation minority status itself is linked with elevated suicide risk (Haas et al., 2011; Duncan and Hatzenbuehler, 2014). The need for attention to nuance of even strong risk factors varying by population is further illustrated by primary psychiatric diagnoses linked to suicide. For instance, standardized mortality risk (SMR) and other research documents the exacerbated prominence of depression, bipolar, and cluster B personality disorders (e.g., borderline, antisocial) in enhancing risk for death by suicide among psychiatric patients (Björksenstam et al., 2015, 2016; Madsen et al., 2017). To illustrate, cluster B disorders are associated with SMRs in this population as high as 33–34 (Björksenstam et al., 2015; Madsen et al., 2017). Cannabis use and dependence, another diagnostically relevant disorder category, has been shown to be associated with exacerbated suicide risk, especially among adolescents and particularly when the cannabis is associated with experiencing of other psychiatric conditions (e.g., psychosis) (Serafini et al., 2012). Thus, cannabis use or dependence may also serve as poor coping or a pathway to suicide among adolescent youth.

We also note many factors with the strongest, most consistent associations with suicide risk (see Table 1), defining strongest and consistent in terms of effect sizes and odds ratios related to suicide-related thoughts or behaviors, direct associations with suicide (e.g., serving as a mediator), as well as those that are highlighted by clinical and prevention experts as those requiring attention across populations (e.g., depression). In all, SEM levels with the strongest support tend to be individual and interpersonal/relational levels. For instance, at the individual or interpersonal levels we note risk factors with the strongest associations with suicide-related thoughts and behavior such as a prior suicide attempt, diagnosis of depression or bipolar disorders, and suicide contagion. To illustrate, hopelessness provides a clear example of an individual psychological risk factor with considerable support; hopelessness has been identified as an independent risk factor for suicide requiring clinical assessment (Bryan and Rudd, 2006), and empirical data raises the potential that hopelessness may serve as a pathway to suicide-related thoughts and behavior explaining the influence of other risk factors (e.g., thinking styles) (Abramson et al., 1998). Moreover, protective factors such as presence, use and perception of positive social support is denoted as among the strongest factors (see Table 1 for full list of demarcated factors with strongest research support). As such, from a public health education standpoint, the integrated summary may serve to reinforce the key factors to include in dissemination efforts by public organizations.

Given the fact that suicidology is an ongoing research field, the proposed SESPM is not intended to be exhaustive. The purpose of the SESPM is a guide to move research and prevention forward, as well as to provide a framework for understanding nuance in suicide prevention. To illustrate the latter point, examination of the multi-level organization identifies several levels at which for example firearm-related factors may influence suicide risk. While firearm access or ownership is associated with elevated suicide risk (Anglemyer et al., 2014), this link may be moderated by other individual (e.g., safe storage) (McCarten et al., 2001), relational (e.g., restricted means counseling) (Stanley et al., 2016), and societal (e.g., firearm restriction laws) (Anestis and Anestis, 2015) factors. The SESPM provides a summary of literature in order to build toward better mediation, moderation, and causal research, as well as multi-level prevention efforts.

We adopt the view that the SESPM is both preliminary and fluid, suggesting prevention may need to account for population- or context-specific considerations. For instance, the SESPM itself may vary by population or culture. To illustrate, it is well known that risk factors such as Human Immunodeficiency Virus (HIV) status itself (Carrico et al., 2007) and internalized HIV-related stigma (Cramer et al., 2015) play particularly salient roles in suicide risk, whereas other factors may be less important for this group. Additionally, the SESPM offers a clear organizational approach to future systematic reviews and hierarchical approaches to meta-analysis or regression. In all, with future empirical testing, the SESPM may need refinement or adaptation by population over time. We advocate in a first step to conduct risk and protective factor meta-analyses to develop appropriate SESPM templates for risk groups. In a second step, this quantitative information about the weight of risk factors should be used in population trials consequently (see Figure 1).

Promising suicide prevention education and training programming exists for the public (Teo et al., 2016) and medical/health professionals (Cramer et al., 2017), yet these areas are in need of further study (Zalsman et al., 2016). Moreover, empirically-tested educational prevention strategies often lack consistent structural framing. That is, content and modalities of these trainings vary, often neglecting content such as community and societal level risk factors and prevention efforts. As has been done in the development of other public health prevention approaches such as HIV prevention (Baral et al., 2013), we encourage development of research/data summaries, educational materials, and training content to be organized by SESPM levels. For example, graduate training or continuing education programs may address established suicide prevention-related competencies (Rudd et al., 2008; Cramer et al., 2013) by SESPM level. While a full program-wide review of such competencies is beyond the scope of this piece, using structured training approaches like observed structured clinical examinations (OSCEs), health professions literature (Hung et al., 2012; Cramer et al., 2016) highlights necessary skills for health providers like knowing empirically-indicated risk/protective factors and intervention/support possibilities. Established training models such as patient simulation or online-mediated courses can integrate such skill development and factual content into a SESPM framework. The end goal of such an approach would be that SESPM-educated health clinicians may be able to make better use of their multi-level knowledge and skills in working with at-risk individuals or designing stronger prevention programs.

Historically, public health and health science prevention efforts have lacked adequate theoretical grounding. Recent health professions literature argues that effective prevention efforts requires strong grounding to bolster effective health behavior change (Im, 2015; Prestwich et al., 2015; Krieger, 2016). The advantages of theory-informed public health include conceptualization of multi-level prevention/intervention programming, transdisciplinary communication, and accounting for practical societal and scientific influences (e.g., funding, political issues). Interestingly, social and behavioral science literature focuses on suicide as the subject of theoretical speculation, although there are varying levels of empirical testing and support across these theories.

Although a full review of all contemporary suicide risk theories is beyond the scope of this work, we provide an example theory with short description for each SESPM level for illustrative purposes (see Table 2). From top down, example theories can be seen in societal (e.g., Le Suicide) (Durkheim, 1897), community (e.g., Military Transition Theory) (Castro and Kintzle, 2014), relational (e.g., Interpersonal-Psychological Theory of Suicide) (Joiner, 2005; Van Orden et al., 2010), and individual (e.g., Cubic Model of Suicide) (Shneidman, 1981) level perspectives. In his seminal text Le Suicide, sociologist Emil Durkheim theorized suicide as a reaction to the intersection of social integration (the clustering of people in social groups) and regulation (the extent of rituals and customs being influenced by societal norms) (Lester, 1999). Rooted in a litany of causes for suicide among US veterans, Military Transition Theory highlights suicide risk as a function of factors unique to the community of military personnel reintegrating into civilian life (Castro and Kintzle, 2014). A three-stage transition is posited: approaching, managing, and assessing the transition. These stages imply a degree of multi-level influence in that they require the person to navigate and evaluate individual, familial, work and other challenges.

The Interpersonal-Psychological Theory of Suicide (IPTS) posits that suicide thinking is a function of self-perceptions in relation to others; negative interpersonal cognitions occur in two forms: thwarted belonging and perceived burdensomeness (Van Orden et al., 2010). Ideation transitions to an attempt when the individual has developed sufficient habitation to pain and fearlessness of death in order to commit the act. In this way, the IPTS may be considered both individual and relational in nature. Finally, Shneidman articulated an individual theory of impulsive suicide (Shneidman, 1981). The impulsive act, calculated to be fatal by the attempter, is thought to occur in the presence of acute psychological states of stress, agitation and psychache (i.e., emotional pain).

Relying on the SESPM organizational framework of risk and protective factors (see Table 1), we believe a valuable next step in the theoretical development and testing is a multi-level or social-ecological theory of suicide in both population-based and clinical trials, as well as in interventions (see Figure 1) in order to inform theory development and prevention programming. The “x” lines refer to the idea that suicide research suffers from the problem of repeated over simplification of studies bypassing a comprehensive use or development of a multi-level model. Therefore, as illustrated in Figure 1, it is critical to understand that the proposed SESPM, a multi-level organizational framework, does not constitute a true theory of suicide by itself. However, following the pathways outlined in the Figure, a consistent and causal theory of suicidal behavior should be deductible from such a framework, and therefore be empirically testable (Horvath, 2016). A social-ecological theory of suicide, for instance, may specify individual attitudes, traits and mental health symptoms as primary, direct predictors of imminent suicide risk. Complementing that testable hypotheses, societal or community level factors, as well as sub-population variation, may serve as directly influence chronic risk, while also playing moderating roles concerning how individual and relational factors affect imminent or acute risk.

Despite a proliferation of suicide risk assessment tools (Lotito and Cook, 2015), recent evidence suggests limited ability to predict future suicidal behavior (Chan et al., 2016; Large et al., 2016). Self-report instruments suffer an identical limitation much of suicide-related theory and risk/protective factor summaries have: a lack of accounting multi-level understanding. For example, the most commonly used self-report tools often assess the frequency and nature of past/present/future ideation and attempts, strongly correlated mental health symptoms (e.g., depression, hopelessness), or protective factors (e.g., reasons for living). While such information is clinically useful to fill in gaps not otherwise captured in interview (Lotito and Cook, 2015), it is still limited in scope.

Several assessment structures exist in the literature, sometimes hinting at the need to address multi-level facets. Such attempts have led to different recommendations for the development of suicide risk assessment tools, but none of these strategies have seen effective implementation yet. For example, suicide risk assessment is often based on lists of symptoms without an integrated perspective (Kral and Sakinofsky, 1994), therefore proposing a model that comprises both background and subjective suicide risk factors. The former are the socio-demographic indices associated with increased risk which are based on different populations and cultures, and are prone to change over time. The latter, background risk factors, can inform the clinician about a patient's general level of risk, while the assessment of individual factors focusses on emotions, cognition, idiosyncratic meanings, general mental state, and experience. Suicide risk assessment methods can also be based on factors falling into (1) individual, (2) clinical, (3) interpersonal, (4) situational, and (5) demographic categories, thus encompassing some of the SESPM levels suggested herein (Simon, 2011).

In total, the variation in recommended methods suggests that a one-size-fits-all solution to suicide risk estimation is an ill fit. Such methodological complexities might be responsible for the result of the most recent meta-analysis of suicide risk assessment scales, which concluded that there is insufficient evidence to support the use of risk scales and tools in clinical practice due to the rather low positive predictive value (PPV) of the scales, which ranged between only 1.3 and 16.7% (Chan et al., 2016), with 87% false positives, a clinically imprecise, economically intensive and unnecessarily stigmatizing proportion. We argue that the heterogeneity and confusion about suicide risk assessment methods has its primary origin in the lack of a unified and empirically testable theory of suicidal behavior.

We posit that the next meaningful steps in suicide risk assessment tool development may lie in two areas: (1) a psychometrically-validated structured professional judgment (SPJ) of key multi-level risk factors, such as the Screening Tool for Assessing Risk of Suicide (STARS) protocol (Hawgood and De Leo, 2016) (which does not fully account for multi-level influences) and (2) a multi-level suicide risk assessment algorithm. For example, although a potentially time, resource and funding intense project, we recommend development of a suicide risk assessment tool for use by mental and medical health professionals that addresses risk and protective factors across all four layers of the SESPM. Such a new approach may be translated into a SPJ tool, an approach to mental health assessment that provides semi-structured rating forms to be used by trained health professionals. In addition, post-interview, more rigorous interviewer-rated checklists, could help to refine and validate the SPJ tool. Nowadays, online implementation and translation in different languages allows for the development of a globally available instrument for suicide risk assessment within an SESPM model for ongoing refinement. Violence risk literature provides examples of well-validated SPJs accounting for three SEM levels, accounting for empirically-indicated risk and protective factors for interpersonal violence, including individual (e.g., affective stability, substance use), relational (e.g., treatment compliance, personal/social support), and community (e.g., living situation, professional services) level issues (Douglas et al., 2013). Community level risk is further accounted for by the relevance of these factors to the setting of evaluation (e.g., inpatient hospital vs. outpatient clinic). Using the SESPM, it is plausible that a suicide SPJ could be developed by first identifying and testing a lengthy set of risk of factors, exposing the preliminary instrument testing in emergency room, outpatient clinic and inpatient hospital settings. After initial reduction and psychometric evaluation of the instrument, further testing would be required for longitudinal and cross-cultural validation. Likely, a culturally-adapted version of such an instrument would be quite useful.

Alternatively, it may be beneficial to develop a risk assessment algorithm, which is stratified by available knowledge (i.e., different sets of risk factors might be relevant for males vs. females, adolescents vs. elderly, etc.). As such, a computer-based suicide risk estimation algorithm may be developed in which a clinician can collect comprehensive multi-level patient and situational information, entering the information into a weighted equation. The first step in such effort is mathematical identification of a weighted formula, likely based on a validated multi-level theory of suicide that has to be developed. First approaches in the direction of a suicide risk algorithm have been provided; however, the presented tool was limited to six individual level factors (e.g., age, self-harm history) suitable for depressed persons for the prediction of suicidal ideation only (Liu et al., 2016). With simultaneous systematic review of multi-level risk and protective factors for suicide ideation, attempts, and completed suicide, identification of relative weights for an SESPM-based algorithm based suicide risk assessment tool should be attainable.

First and foremost, SESPM-based research is needed to inform best practices for prevention programming. Thus, our recommendations for development of multi-level programming are provided with the caveat that further research and theory development are required. As a starting point, multi-level suicide prevention should address (1) general practitioner education concerning depression and suicide; (2) increased access to care for high-risk groups, and; (3) emphasis on restricting access to lethal means (van der Feltz-Cornelis et al., 2011). While we agree with the importance of these practical recommendations, we further posit that design of an ideal multi-level approach to suicide prevention would possess the following characteristics: (1) incorporation of the risk and protective factors with strongest empirical support relative to the population (e.g., general population vs. high risk psychiatric inpatients); (2) use of prevention strategies at each SESPM level; (3) inclusion of a multi-level program evaluation strategy including data gathered from patients and other stakeholders (e.g., therapists, policy makers, etc.)—patient data would include suicide and self-harm, whereas additional patient and stakeholder information could cover subjective and objective patient-oriented outcomes more generally; (4) grounding in relevant theory to inform mechanisms of change; and (5) presence of prevention efforts using at least primary and secondary prevention techniques where possible.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.