RESEARCH DOSSIER
Suicide and Self-Harm Risk Assessment and Prevention
Abstract. Suicide and self-harm research links intentional behavior, distress, developmental and social circumstances, clinical decision-making, and public-health prevention. This research monograph reviews the historical formation of the field, distinctions between suicidal ideation, attempts, nonsuicidal self-injury, and intrusive self-harm thoughts, and the evidence for dynamic psychosocial formulation. Its critical narrative synthesis examines epidemiology, theories of ideation-to-action transition, validated screening instruments, limitations of risk scores and machine-learning prediction, experimental and longitudinal studies, safety planning, psychotherapy, transitions of care, and multilevel prevention. The evidence supports accessible, collaborative assessment, timely care, follow-up, and context-sensitive interventions while identifying substantial uncertainty about individual prediction, cross-population transportability, and implementation effects. Research findings and clinical guidelines are distinguished from personalized recommendations.
Keywords: suicide prevention; suicidal ideation; suicide risk assessment; self-harm; nonsuicidal self-injury; psychosocial assessment; dynamic risk formulation; clinical prediction; safety planning; dialectical behavior therapy; cognitive behavioral therapy; lethal means safety; postdischarge follow-up; public health prevention; clinical psychology
The Scientific Problem of Suicide and Self-Harm
Suicide and self-harm constitute related but distinct objects of psychological, clinical, epidemiological, and public-health investigation. Suicide denotes death resulting from self-directed injurious behavior with an intent to die; a suicide attempt is a nonfatal, self-directed potentially injurious act involving at least some intent to die. Suicidal ideation comprises thoughts about death or ending one's own life and varies in frequency, specificity, emotional meaning, and relation to prospective action. Self-harm is defined differently across research traditions and health systems. The United Kingdom's clinical guideline uses an inclusive definition of intentional self-poisoning or injury irrespective of apparent purpose, while research on nonsuicidal self-injury, or NSSI, ordinarily specifies deliberate self-inflicted tissue damage without suicidal intent. The same patient's history can include several phenomena across time, so the relevant unit of investigation must be explicitly identified NICE (2022) Cipriano et al. (2017).
Risk assessment and prevention ask different scientific questions. Assessment investigates what has occurred, what is happening now, how conditions could change, and what assistance may reduce foreseeable danger; epidemiologic prediction asks how well measurable characteristics distinguish outcomes in a defined population over a specified period. Prevention evaluates changes in suicidal behavior, distress, treatment engagement, exposure, and deaths after interventions at individual, service, and population levels. Valid evidence for one question may be insufficient for another. An interview can establish current thoughts and circumstances without accurately predicting a rare event, and a population policy can reduce rates without identifying which individual would otherwise have died. The long history of weak individual prediction from candidate risk factors demands separation of these purposes Franklin et al. (2017) Carter et al. (2017).
This monograph is a critical narrative synthesis of original trials, cohort studies, measurement validation studies, systematic reviews, public-health analyses, and clinical guidelines. It examines conceptual boundaries, mechanisms of ideation and action, measurement validity, dynamic formulation, intervention effects, and implementation. The clinical literature is treated as a body of population- and setting-specific evidence. When reporting studies, the distinction between measured outcomes, interpretation, and practice recommendations is preserved; where no direct clinical trial establishes a claim, that limitation determines its wording. The work aims to support research literacy and evidence-grounded service design. Individual assessment and treatment remain collaborative professional activities situated in an actual person's circumstances VA/DoD (2024) NICE (2022).
Historical Formation of Suicidology and Prevention Research
Modern suicide research emerged through several intersecting traditions: population statistics, psychiatry, social theory, clinical observation, and experimental psychology. Early sociological accounts examined how the distribution of suicide changes with social integration and regulation. Clinical traditions subsequently developed instruments for suicidal thoughts, histories of attempts, hopelessness, and psychiatric symptoms. Late twentieth-century psychological work investigated the functions of self-injury, cognitive processes associated with suicidal crises, and the distinction between thinking about suicide and acting. These lines of inquiry used different definitions, samples, and inferential methods. Their historical value lies in establishing tractable questions: the distribution of deaths, how people experience crises, why an act occurs in a particular context, and whether preventive actions change observable outcomes Klonsky (2007) Franklin et al. (2017).
An important shift has been from the isolated identification of a stable high-risk type toward the study of trajectories, contexts, and modifiable processes. Postcrisis outreach trials provide early examples. In a randomized study, Motto and Bostrom evaluated ongoing contact after psychiatric hospitalization, making continued human connection a testable preventive intervention rather than a purely descriptive clinical intuition Motto & Bostrom (2001). Later controlled trials examined cognitive treatment following suicide attempts and structured approaches to emotional dysregulation and repeated self-harm Brown et al. (2005) McCauley et al. (2018). The growth of prospective designs also revealed the statistical difficulty of forecasting a relatively rare individual outcome, even when risk factors are strongly associated with the outcome at population level.
Contemporary prevention research incorporates health-system pathways, evidence synthesis, school and community interventions, and means safety. The WHO LIVE LIFE framework organizes national suicide prevention around practical domains including restricting access to highly lethal means, responsible media reporting, socioemotional skills in adolescents, and early identification, management, and follow-up for people affected by suicidal behavior WHO (2021). Methodological advances have not erased historical ambiguities. Definitions of self-harm differ by country and dataset; treatment trials often measure repeat attempts rather than deaths; and legal, cultural, and institutional practices affect which events reach a clinical or statistical record. A responsible historiography therefore follows the evidence and the changing definitions that made it possible.
Terminology, Intent, and Outcome Classification
The most consequential distinction is between an intention to die, an intention to modify an unbearable psychological state, and uncertainty or mixed intent. Researchers may classify a behavior according to contemporaneous self-report, the act's observable circumstances, clinical records, and later testimony. None of these sources is infallible. Intent may be ambivalent, retrospectively reconstructed, hidden because of fear, or difficult to articulate in children or severely distressed adults. A research protocol must document whether it codes an event as a suicide attempt, NSSI, undetermined-intent injury, or all self-harm irrespective of intent. Combining these outcomes can increase the event count while changing the meaning of estimated associations and intervention effects Ribeiro et al. (2016) NICE (2022).
The separation of suicidal ideation, preparation, attempts, and suicide death also matters. Thoughts are commoner than deaths and have different measurement properties. An intervention can reduce reported ideation without a demonstrable effect on attempts, and a study powered for repeated self-harm may contain too few suicide deaths to estimate a mortality effect. For trials, investigators must specify their primary endpoint, assessment horizon, ascertainment procedure, and handling of missing or competing events. Registry-linked mortality data can improve ascertainment but introduce cross-jurisdiction differences. Patient-reported outcomes can capture episodes that never reach clinical attention, yet their recall and willingness-to-disclose assumptions require examination Doupnik et al. (2020) Franklin et al. (2017).
Clinical research also distinguishes acute distress, persistent suicidal thoughts, recurrent NSSI, and specific intrusive images of self-harm. The question is how these experiences are organized for a particular individual: unwanted intrusions can be experienced as alarming and resisted, whereas suicidal intent involves some wish, expectation, or readiness for self-inflicted death. Anxiety, obsessive-compulsive symptoms, depressed mood, substance use, trauma, and psychosis may complicate interpretation. Thought content alone cannot identify intent, and a reported absence of intent at one interview cannot establish an indefinitely safe trajectory. Psychosocial assessment attends to subjective meaning, previous events, current context, protective supports, and change over time NICE (2022) Posner et al. (2011).
Epidemiology, Mortality, and the Denominator Problem
The global burden is substantial and unequally distributed. World Health Organization estimates published in 2025 describe approximately 727,000 suicide deaths worldwide during 2021; most occurred in low- and middle-income countries. These estimates describe a specified reference year and combine variable national registration coverage with statistical estimation. They cannot be treated as a real-time worldwide count for 2026, nor does an international rate measure risk within a particular clinical encounter. In the United States, the Centers for Disease Control and Prevention reports 48,824 suicide deaths in 2024 (a recorded age-adjusted rate of 13.7 per 100,000 population), alongside survey estimates of 14.3 million adults who seriously considered suicide and 2.2 million adults who made an attempt during that year. Counts of deaths and survey-based estimates must retain their distinct statistical sources and ascertainment uncertainties. Deaths, emergency visits, suicide attempts, and self-reported ideation have different denominators and should appear as separate indicators WHO (2025) CDC (n.d.).
Epidemiologic comparisons require attention to age standardization, sex, geography, social conditions, and recording practices. A crude rate can shift when the age structure changes even if age-specific risks remain constant. Different legal and cultural systems may alter whether deaths of ambiguous intent are registered as suicide. Nonfatal self-harm can be undercounted when individuals avoid care or do not disclose the event, while emergency department records oversample people whose acts prompt treatment. Data on adolescents gathered in schools exclude some young people with different circumstances, and results obtained among hospitalized psychiatric patients cannot be generalized directly to all people with suicidal thoughts. These are questions of study population, selection, and measurement rather than minor statistical footnotes WHO (2025) CDC (n.d.).
Public-health analysis also confronts the denominator problem: the large majority of people who have a known risk factor do not die by suicide during a short observation window. When events are rare, even a prediction instrument with reasonable sensitivity and specificity can have a low positive predictive value. The same threshold yields different predictive values at different base rates. Treating a positive classification as an individual forecast converts a population probability into an unjustified statement of certainty. Reviews of clinical risk tools and decades of prospective research show why the appropriate policy objective is accessible, responsive prevention for people who need care rather than confidence in a supposedly infallible sorting algorithm Carter et al. (2017) Franklin et al. (2017).
Nonsuicidal Self-Injury, Functional Analysis, and Clinical Overlap
Nonsuicidal self-injury is commonly investigated as behavior with functions that differ across people and episodes. Functional research has identified intrapersonal processes such as the temporary modulation of overwhelming emotion and interpersonal processes involving communication of distress or changes in social interaction. These categories describe hypothesized consequences and reinforcement histories, not fixed personality types. A patient may experience short-term relief while the longer-term pattern contributes to isolation, shame, and worsening distress. Functional inquiry examines antecedent states, perceived consequences, relationships, and changes in the pattern. Reviews by Klonsky and by Cipriano and colleagues document the conceptual range and limitations of this literature Klonsky (2007) Cipriano et al. (2017).
The absence of suicidal intent during a particular episode does not imply the absence of future suicide risk. Longitudinal research associates a history of self-injurious thoughts and behaviors with later suicidal outcomes, although the predictive strength and causal interpretation depend on design and adjustment. NSSI and suicide attempts may share distress, interpersonal adversities, psychiatric comorbidity, and learned responses while remaining distinguishable behavioral outcomes. Research needs to avoid two errors: reclassifying all NSSI as attempted suicide, and presuming that a person with NSSI has no suicide-related needs because an earlier episode lacked intent. The appropriate clinical question concerns the evolving individual history and current meanings Ribeiro et al. (2016) NICE (2022).
Measurement of self-injury can differ across community surveys, clinical charts, and structured interviews. Some instruments emphasize tissue damage, others include self-poisoning, and studies vary in the observation period and whether indirect forms of self-damaging behavior are included. Differences in definitions can alter reported prevalence, recurrence, and the estimated effect of treatment. Self-report and records may also disagree because events are concealed or never medically assessed. A high-quality study specifies its operational threshold, the source of information, the coding of intent, and the handling of uncertain cases. A high-quality clinical assessment combines respectful inquiry with contextual formulation rather than forcing heterogeneous experiences into a single severity number Cipriano et al. (2017) Witt et al. (2021b).
Intrusive Self-Harm Thoughts, Suicidal Intent, and Phenomenological Assessment
Thoughts about self-injury can emerge as ego-dystonic intrusions in obsessive-compulsive or anxiety-related presentations, as fantasies of escape from distress, as recurrent suicidal thinking, or as images associated with traumatic memories. These experiences require careful inquiry into their meaning and function. In an intrusive obsession, the thought may be experienced as unwanted, inconsistent with the person's wishes, and followed by avoidance, reassurance seeking, or other compulsive responses. Suicidal ideation may instead include a desire to die, anticipation of relief through death, ambivalence, or planning. These descriptive differences can guide further evaluation, but no single adjective, diagnosis, or test score reliably settles an individual's immediate safety NICE (2022) Posner et al. (2011).
A clinician's conceptual task is to clarify the person's own account over time: whether an image or thought is feared or desired; whether it is accompanied by intentions, preparatory action, perceived capability, access to means, or reasons to remain alive; and how these features vary with context. The assessment should also seek information about prior attempts, NSSI, rapid changes in mood, agitation, substance use, sleep disruption, and accessible social support when relevant. Questions must be direct, respectful, and appropriately matched to communication abilities. The literature does not support the commonplace fear that asking about suicidal thoughts will itself implant an idea; a review of available studies instead found no evidence for that general proposition Dazzi et al. (2014) Horowitz et al. (2012).
This distinction has direct methodological consequences. A scale item that records the mere presence of a self-harm thought can conflate obsessional intrusion, passive wishing, and suicidal intent unless the instrument uses meaningful follow-up categories. Studies of measurement validity must test whether items retain the same meaning across diagnoses, age groups, languages, and settings. Clinically, a person with intrusive thoughts can experience intense suffering and require mental-health care even when no wish to die is reported. A person describing ambivalence can experience meaningful risk even without a rehearsed plan. The task is a revisable psychosocial and safety formulation grounded in dialogue and collateral context, not a binary inference from surface wording Posner et al. (2011) NICE (2022).
From Risk-Factor Catalogues to Explanatory Models
Much of the early empirical literature organized suicide research around correlates: mood disorders, previous attempts, hopelessness, exposure to violence, substance-related problems, social isolation, and stressful transitions. Such associations identify populations in which adverse outcomes are more frequent, but they do not themselves explain how an individual progresses from distress to suicidal thinking or from thinking to action. A meta-analysis of research spanning roughly five decades found that many commonly studied risk factors had weak predictive accuracy and that improvement in prediction over time was limited Franklin et al. (2017). The implication is methodological: a theory that accounts for the distribution of risk must identify intervening processes, specify how they vary within a person, and generate testable predictions stronger than a list of diagnostic labels.
Explanatory models commonly distinguish distal vulnerability, proximal motivation, and conditions facilitating or interrupting action. Distal factors can include earlier adversity, enduring illness, socioeconomic exclusion, and prior behavior. Proximal states may involve perceived entrapment, emotional pain, acute interpersonal rupture, or a crisis of expectation. Conditions bearing on action include behavioral capability, access to lethal means, intoxication, rapidly changing circumstances, and availability of immediate help. Each construct requires operational measurement. The separation of ideation from enactment is a key contribution of ideation-to-action frameworks, including the Three-Step Theory and integrated motivational–volitional model Klonsky & May (2015) O’Connor & Kirtley (2018).
These theories should be read as structured explanatory programs rather than validated individual prediction engines. A compelling conceptual map may organize clinical conversation and lead to new research hypotheses, yet its empirical support depends on adequately powered longitudinal designs, replication, measurement independence, and tests against rival models. For example, retrospective self-report about why an attempt occurred may yield important phenomenology but is vulnerable to memory reconstruction. Cross-sectional correlations among hopelessness, entrapment, and ideation cannot establish causal order. The scientific challenge is to investigate temporal transitions and modifying conditions while avoiding the assumption that a clinically plausible story has already been experimentally confirmed Franklin et al. (2017) Ribeiro et al. (2016).
Ideation-to-Action Theories and the Transition Problem
The ideation-to-action framework begins with a central observation: many individuals who experience suicidal thoughts do not attempt suicide. The Three-Step Theory proposes a sequence in which pain and hopelessness can support ideation, the relation between pain and connectedness can modify the intensity of that ideation, and capability becomes relevant to the possibility of action. It is an economical hypothesis about necessary and moderating processes, not a universal sequence observed in every person. The model motivates studies that distinguish predictors of thought intensity from predictors of behavioral transition Klonsky & May (2015).
The integrated motivational–volitional model places greater emphasis on processes that produce a sense of defeat and entrapment and on motivational moderators, volitional moderators, and background vulnerability. Its distinction between a pre-motivational context, the formation of suicidal ideation, and the emergence of suicidal behavior allows hypotheses about what changes at each stage. A supportive relationship may be related to a different causal process than access to a lethal method; an intervention that changes immediate behavior might therefore not change every component of ideation. The model offers a framework for designing temporal research and selecting intervention targets, while the available evidence must determine whether specific hypothesized pathways generalize across populations O’Connor & Kirtley (2018).
A central empirical complication is that the boundary between stages can be unstable. Suicidal thoughts may be transient, hidden, chronic but fluctuating, or accompanied by changing intention. Retrospective classification of the start of an episode may be uncertain. Behavioral outcomes have low base rates in many cohorts, so apparent discrimination can deteriorate when a model is tested in a new setting. Studies need repeated measurement of thoughts, contexts, and transitions, with prospective ascertainment of behavior and explicit treatment of intervening clinical care. The scientific success of an ideation-to-action account rests on its incremental explanation of actual changes rather than on whether a diagram appears intuitively convincing Carter et al. (2017) Franklin et al. (2017).
Psychological Pain, Hopelessness, Entrapment, and Cognitive Narrowing
Psychological pain refers to distress experienced as deeply aversive, difficult to escape, and incompatible with continued functioning as currently imagined. Hopelessness concerns expectations that desired changes are unattainable, while entrapment describes a perception that escape from an intolerable situation is blocked. These constructs overlap with depression and anxiety but are not interchangeable with diagnostic categories. Ideation-to-action models treat their combinations as possible proximal conditions in the formation or intensification of suicidal thoughts O’Connor & Kirtley (2018) Klonsky & May (2015). Empirical studies must determine whether each measure explains variance beyond general distress, previous suicidal behavior, or other measured conditions.
During an acute crisis, attention may become concentrated on an immediately painful problem and the perceived range of future options may contract. This is a clinically useful description, but a mechanism claim requires direct evidence about attention, decision-making, memory, or expectations under conditions of distress. Repeated assessments can test whether shifts in perceived entrapment precede shifts in ideation and whether interventions change both. Laboratory tasks may contribute evidence about cognitive control, yet their ecological validity depends on connecting task performance to life circumstances. The distinction between a model's psychological vocabulary and the validity of its measurement protects research from circular explanations in which a suicidal act is used to infer an unmeasured internal state that is then invoked to explain the act Franklin et al. (2017) O’Connor & Kirtley (2018).
Clinical formulation treats pain, hopelessness, and entrapment as revisable hypotheses situated in an individual's own account. Financial insecurity, discrimination, chronic illness, loss, interpersonal conflict, and anticipated shame may form part of the situation from which escape seems impossible. Their roles are not determined merely by population averages. A person's present supports, expectations, and ability to engage with care can change during a conversation or across days. The assessment should explicitly note that prediction is uncertain and that the aim is to identify responsive, collaborative care. A reduction of a symptom score can be meaningful without being proof that all future suicidal behavior has been prevented NICE (2022) VA/DoD (2024).
Emotion Regulation and the Functions of Self-Injury
Functional accounts of nonsuicidal self-injury examine the relations among antecedent emotional states, action, short-term consequences, and changes in future behavior. People report different functions, including interruption of distress, self-directed punishment, expression of otherwise incommunicable feelings, or regulation of interpersonal distance. These reports matter because a purely topographic classification of self-inflicted injury obscures the varied psychological processes associated with it. Reviews find substantial support for affect-regulatory functions while also documenting methodological weaknesses in retrospective reports and correlational studies Klonsky (2007) Cipriano et al. (2017).
A learning perspective asks whether a short-lived alteration in affect reinforces an action despite its damaging longer-term consequences. Such an account remains a hypothesis about the consequences of behavior and should be tested through repeated observations of affect before and after events, contextual covariates, and prospective recurrence. Studies comparing individuals with and without NSSI may also be confounded by depression, exposure to trauma, family conflict, or treatment history. A mechanism demonstrated among one group of adolescents may not describe all adults or all types of self-harm. The construct therefore benefits from within-person designs, transparent definitions, and a distinction between measured association and causal reinforcement.
Intervention research can use this understanding to investigate whether training in emotional regulation, distress tolerance, communication, and problem solving changes recurrent self-harm. Dialectical behavior therapy combines skills-based and behavioral strategies within a comprehensive treatment structure, with randomized youth evidence showing a reduction in self-harm-related outcomes in specified clinical populations McCauley et al. (2018). Cochrane syntheses distinguish what is known about interventions for adults from evidence for children and adolescents; treatment labels should not be assumed to imply equivalent effects across ages or outcomes Witt et al. (2021b) Witt et al. (2021a). The clinical objective includes safer ways of managing distress, sustained engagement, and recovery of functioning, while prevention trials must measure their own outcomes precisely.
Temporal Dynamics, Crises, and Within-Person Change
Suicidal thinking and behavior vary across timescales: minutes of acute agitation, days around interpersonal or clinical transitions, months of illness or recovery, and developmental periods marked by changing autonomy and relationships. Risk research often measures exposure at enrollment and outcome months later, leaving the immediate trajectories largely unobserved. A stable baseline factor may explain why one population has a higher event rate but may be less useful for identifying when a particular person's circumstances change. Dynamic formulation therefore requires a time-indexed account of symptoms, recent events, substance-related states, sleep, social contact, and access to support when pertinent. This is the clinical logic behind revisiting safety needs as circumstances evolve NICE (2022) VA/DoD (2024).
Intensive longitudinal methods, repeated clinical contact, and time-stamped service data can illuminate these dynamics, but measurement itself can affect behavior and attrition can be informative. People may omit the highest-distress periods from an ecological assessment because they are unable or unwilling to answer. A prediction rule calibrated on frequent responders may perform poorly for those who disappear from monitoring. Passive digital measures raise additional questions about whether device use is an adequate proxy for affect or intention. Ethical studies must predefine escalation procedures, privacy protections, and the relationship between research measurement and clinical intervention; a participant's data should never be mistaken for an exhaustive representation of their experience.
Time-sensitive studies also need to account for service effects. If individuals assessed as distressed receive more intensive care, their subsequent outcomes reflect the joint influence of underlying vulnerability and treatment. In observational data, this confounding by indication can make an effective service appear associated with higher event rates. Models require careful adjustment, prospective comparison, and explicit causal assumptions. Repeated assessments may improve understanding of change without generating certainty about rare outcomes. The scientific value of dynamic formulation lies in maintaining contact with the evolving person and identifying practical interventions, rather than claiming that real-time surveillance eliminates prediction error Franklin et al. (2017) Spittal et al. (2025).
Outcome Ascertainment, Cohorts, and Causal Inference
Scientific conclusions in suicide risk assessment depend on what investigators classify as an outcome and where they observe it. A prospective cohort can follow people after emergency presentation, hospital discharge, outpatient care, or enrollment in a community sample. Mortality linkage is useful for registering deaths, whereas clinical records primarily ascertain episodes associated with healthcare contact. Interviews and surveys capture experiences missed by medical systems but depend on disclosure and recall. Different ascertainment approaches may produce different rates even in ostensibly comparable populations. Investigators must define the index encounter, follow-up interval, event classification, missing data, and competing risks before interpreting a prediction model or treatment effect Carter et al. (2017) Ribeiro et al. (2016).
Risk-factor analyses also encounter treatment-selection bias. Individuals with more concerning presentations may receive greater clinical attention, and the resulting association between treatment and recurrence need not imply that care caused recurrence. Observational comparisons of service interventions must therefore explain selection mechanisms and baseline differences. Randomized trials offer stronger evidence about the effects of assigned intervention conditions, but can still be influenced by attrition, inconsistent treatment delivery, crossover, and incomplete ascertainment of rare outcomes. Findings about suicidal ideation, self-harm recurrence, suicide attempts, and death should be reported separately. Meta-analyses aggregating incompatible outcomes can conceal important clinical differences Doupnik et al. (2020) D’Anci et al. (2019).
External validity is particularly important because the settings for assessment are heterogeneous. Models built on emergency-department records, military health systems, specialist psychiatric clinics, or school surveys may incorporate selection patterns absent elsewhere. Changing clinical practices, healthcare coverage, demographic composition, or definitions of self-harm can make calibration deteriorate over time. Research-grade prediction evaluation should use temporally and geographically independent test cohorts, report decision thresholds, and inspect performance across groups. A statistical score becomes clinically meaningful only when its use improves a carefully specified decision and does so without unacceptable disparities or missed opportunities for support Spittal et al. (2025) VA/DoD (2024).
Direct Inquiry, Screening, and Structured Assessment Instruments
Direct, nonjudgmental inquiry is the starting point for recognizing suicidal thinking and self-harm in clinical settings. Clinicians examine the nature and course of thoughts, current distress, earlier actions, contextual triggers, available supports, and changes in the person's ability to remain safe. Screening instruments can help standardize the initial identification of people who warrant a fuller conversation. The Ask Suicide-Screening Questions instrument was developed and validated in a pediatric emergency setting, while the Columbia–Suicide Severity Rating Scale has been studied as a structured approach to recording ideation severity and suicidal behavior Horowitz et al. (2012) Posner et al. (2011). Their evidence concerns defined measurement tasks and study samples, and the existence of validation data does not make an instrument a precise individual forecast.
Clinically, a positive screening response should lead to appropriate assessment and connection with support according to the setting's established care pathway. A negative response cannot substitute for attention to a concerning clinical presentation, disclosure elsewhere, or newly emerging circumstances. Instruments can differ in sensitivity, specificity, reference standard, and their handling of ambiguous intent. Their results are shaped by language, privacy, fear of consequences, developmental level, and willingness to speak frankly. An instrument validated for adolescents in an emergency department may need new validation before use in primary care, older adults, or a different cultural setting Horowitz et al. (2012) Carter et al. (2017).
Research on asking about suicide also bears on assessment practice. The concern that direct questions necessarily create suicidal thoughts has not been supported as a general claim in the literature examined by Dazzi and colleagues Dazzi et al. (2014). The relevant ethical requirement is appropriate communication, immediate response capacity, and respect for the person's experience. Screening programs must be coupled to available follow-up services; otherwise identification can fail to translate into help. Routine administration of a questionnaire alone cannot establish that a suicide-prevention program is effective. Investigators should measure the consequences of screening for engagement, distress, later behavior, and service capacity instead of reporting completion rates as health outcomes NICE (2022) VA/DoD (2024).
Psychosocial Assessment and Collaborative Risk Formulation
A psychosocial assessment seeks a coherent account of the person's recent experiences, needs, vulnerabilities, and available resources. It brings together the person's own description, relevant clinical history, relationships, cultural context, practical circumstances, and the chronology of suicidal thoughts or self-harm. Assessment also considers physical health and substances when pertinent and whether immediate safeguarding or urgent medical evaluation is needed. NICE NG225 recommends that mental-health professionals undertake a risk formulation as part of every psychosocial assessment and attend to both immediate and longer-term physical and psychological safety NICE (2022). Formulation is an explicit, revisable clinical explanation rather than a prediction expressed as an unchanging rank.
A useful formulation distinguishes background conditions from current pressures and identifies factors that may be amenable to change. It may document a pattern in which a severe mood episode, recent loss, impaired sleep, isolation, and reduced access to care occur together; it may also identify particular relationships or routines that support recovery. The scientific question is whether this representation improves communication, continuity, and timely support compared with unstructured practice. The therapeutic quality of assessment matters because an encounter experienced as dismissive or coercive can interfere with disclosure and later engagement. The person's preferences, communication needs, and account of what has helped previously are therefore part of the relevant evidence NICE (2022) VA/DoD (2024).
Formulation must retain uncertainty. A current denial of suicidal thoughts, a history of earlier self-harm, and a reassuring presentation can coexist, and their significance may change. Where possible and appropriate, clinical teams revisit assessments after important transitions and coordinate relevant observations across care settings without compromising privacy. Documentation should record the findings and the rationale for care decisions rather than implying that a score eliminated responsibility. A collaborative assessment can support an actionable care plan while remaining honest about the limitations of individual prediction. The aim is an informed, proportionate response, with urgent clinical pathways available when immediate danger is apparent NICE (2022) Carter et al. (2017).
The Statistical Limits of Suicide Risk Scores
Risk scores are often presented as convenient ways to transform a complex clinical situation into a category. Their statistical evaluation requires several different concepts. Discrimination asks whether people with an outcome tend to receive higher scores than people without it; calibration asks whether predicted probabilities match observed event frequencies in the population and time horizon. Sensitivity, specificity, and predictive values depend on the selected threshold and on outcome prevalence. A model can discriminate moderately well and still produce many false positive classifications when the event is infrequent. Such a model may also miss individuals who do not resemble the development sample. In the 2017 meta-analysis by Carter and colleagues, the pooled positive predictive value for suicide across examined risk scales was 5.5% (95% confidence interval, 3.9%–7.9%). The estimate is specific to the included studies and their settings, and does not imply that an arbitrary screening program produces the same value Carter et al. (2017) Franklin et al. (2017).
NICE NG225 provides unusually explicit guidance: risk assessment tools and scales should not be used to predict future suicide or repetition of self-harm, to decide who receives treatment or is discharged, or to stratify people globally into low-, medium-, or high-risk categories for these purposes NICE (2022). This guidance reflects the insufficiency of these tools as individual decision rules. It does not prohibit clinical measurement of distress or structured recording of suicidal thoughts. It instead places the emphasis on a person's needs, psychosocial formulation, available support, and safety across time. A categorical label is also vulnerable to institutional misuse if it is interpreted as proof that a crisis has been permanently resolved.
The underlying statistical problem is a mismatch between population association and individual certainty. A risk factor can be causally important in aggregate yet lack specificity for short-term outcomes, while a person without many conventionally measured factors may still encounter an acute crisis. Small development samples, data leakage, incomplete external validation, and changes in care can inflate apparent model performance. Moreover, a clinical decision has benefits and harms that are not summarized by the area under a receiver operating characteristic curve. Decision-curve analysis, calibration assessment, and prospective evaluation of a model's actual effects on care are needed before any proposed clinical utility is established Spittal et al. (2025) Carter et al. (2017).
Machine Learning, Digital Monitoring, and Algorithmic Prediction
Machine learning has been promoted as a way to incorporate hundreds of healthcare variables into suicide prediction. Some models use diagnosis codes, prescription histories, prior service use, sociodemographic characteristics, or features of clinical text. Their ability to identify statistical patterns can exceed that of a small manually defined set of predictors in particular datasets, yet predictive performance is not identical to preventive value. A 2025 systematic review and meta-analysis by Spittal and colleagues examined 53 studies of machine-learning algorithms predicting suicide or hospital-treated self-harm; across the examined studies reported areas under receiver operating characteristic curves ranged from 0.69 to 0.93, alongside substantial variation in design, outcome, and evaluation Spittal et al. (2025). A clinically credible claim must specify the outcome, time horizon, sample, external test, and actual positive predictive value.
Retrospective evaluations risk leakage when predictors contain information recorded after the moment at which a decision would have been made. Model performance can also benefit from features that proxy service access, diagnostic intensity, or socioeconomic disadvantage. If the training cohort disproportionately captures people who reach a hospital, the model may systematically underrepresent those excluded from care. Explainability methods may identify influential variables but do not establish that those variables are causal or that a suggested action will work. Fairness requires subgroup assessment, transparent monitoring, and a defensible policy for handling uncertain scores. Model drift is particularly relevant as health systems, documentation practices, and patient populations change Spittal et al. (2025).
Digital traces and generative artificial systems raise additional questions about autonomy, surveillance, false alarms, and professional responsibility. Private messages, search histories, wearable activity, or conversational outputs may contain signals related to distress, but their use introduces consent and privacy burdens and must be justified by demonstrated benefit. An artificial agent's apparent fluency is not evidence of reliable clinical diagnosis or immediate individual risk prediction. Research needs prospective, independently evaluated studies comparing actual care outcomes, including unnecessary escalation and missed cases, rather than accuracy against retrospectively assigned labels. Within the Artificial Era, responsible psychological research defines the task and evidentiary limits of artificial systems explicitly; clinical governance remains accountable to human patients and health services Spittal et al. (2025) NICE (2022).
Population Differences, Comorbidity, and Clinical Settings
Risk assessment occurs in emergency medicine, primary care, mental-health services, schools, correctional settings, military systems, and community organizations. Each setting has different prevalence, observation windows, resources, and legal constraints. A person arriving with an acute medical consequence of self-harm requires assessment of immediate physical needs in addition to psychological circumstances. A young person identified at school may need developmentally appropriate communication and family or safeguarding coordination. Long-term outpatient care may focus more heavily on recurring patterns and the reliability of continuing support. Guidelines offer pathways adapted to these distinct settings, while the evidence for a particular intervention may come from only one or two study populations NICE (2022) VA/DoD (2024).
Psychiatric diagnoses are relevant clinical information, but their presence or absence does not determine a uniform level of suicide danger. Depressive disorders, bipolar illness, substance use disorders, psychosis, trauma-related symptoms, chronic pain, and certain personality difficulties may overlap with suicidal thoughts or self-injury. Comorbidity complicates causal interpretation because several conditions, adverse experiences, and treatment histories co-occur. Assessment should inquire into the particular processes that matter for the person's presentation, including rapid changes and practical supports, rather than substituting a diagnosis for a complete formulation. The heterogeneity of clinical presentations also limits the portability of population prediction rules Franklin et al. (2017) Carter et al. (2017).
Culture, age, language, and social exclusion affect whether people recognize an assessment as safe and whether a screening item has a comparable meaning. Research needs measurement-invariance testing and inclusion of underrepresented populations. A tool created in one linguistic environment may fail to distinguish culturally specific expressions of hopelessness or fear. Indigenous communities, forcibly displaced people, and marginalized populations may have sound reasons to distrust surveillance or involuntary intervention, which makes respectful partnership and governance relevant to both data quality and ethical implementation. Service programs should evaluate accessibility, acceptability, and downstream outcomes, and should interpret subgroup estimates with their sample-size uncertainty WHO (2021) WHO (2025).
Immediate Care, Emergency Pathways, and Clinical Safety
Emergency responses to self-harm or an acute suicidal crisis bring together immediate medical assessment, psychological care, and decisions about appropriate continuing support. The first requirements concern physical safety, humane communication, and the person's capacity to engage in assessment given possible intoxication, severe distress, injury, or other acute conditions. NICE advises compassionate, person-centered care and access to an age-appropriate psychosocial assessment rather than using the severity of an injury as a proxy for psychological need NICE (2022). Clinical processes should enable urgent evaluation when danger is imminent and attend to practical barriers that could prevent follow-up.
An assessment in an emergency setting is a starting point within a care pathway, not a discrete event that certifies the future. People may have difficulty disclosing thoughts in a crowded setting or may give a different account once a crisis changes. Coordination among emergency clinicians, mental-health professionals, the patient, and appropriate supporters matters, with due regard for consent, safeguarding, and confidentiality. Written plans, transfer arrangements, and scheduled contact can create continuity across services. The VA/DoD guideline provides a structured framework for clinicians managing suicide-related presentations in relevant healthcare systems, while local law and service capacity govern application VA/DoD (2024).
Research can evaluate crisis pathways through outcomes including treatment engagement, repeat presentations, self-harm events, patient experience, and mortality when suitably measured. The composition and intensity of usual care vary between studies, making intervention comparisons difficult. Observational hospital data may also reflect selection bias: the most unwell people are more likely to be admitted or receive intensive follow-up. Evaluation should measure equity of access and adverse consequences of coercive approaches alongside intended safety outcomes. A clinician-facing evidence synthesis must preserve the distinction between evidence about service pathways and individualized emergency instructions Doupnik et al. (2020) Miller et al. (2017).
Collaborative Safety Planning and Follow-Up Contact
Safety planning is a structured collaborative intervention in which a person and clinician identify relevant warning experiences, personal coping resources, people and services that can offer support, and ways to reduce danger in foreseeable crises. Its value as a research intervention depends on its specific components, usability, adaptation to the person, and integration with actual contact and care. It is not equivalent to a contract requiring a person to promise that they will never self-harm. NICE recommends collaboratively written safety plans for people who self-harm, while acknowledging that plans are one component of a broader care response NICE (2022). Contemporary scholarship treats safety planning as a practical intervention whose efficacy must be assessed against meaningful outcomes.
In a large multisite cohort comparison involving 1,640 patients treated in Veterans Health Administration emergency departments, Stanley and colleagues compared a Safety Planning Intervention combined with structured follow-up against usual care. During six-month follow-up, suicidal behavior was reported for 36 of 1,186 intervention patients (3.03%) and 24 of 454 comparison patients (5.29%); the investigators reported an association approximating 45% fewer suicidal behaviors and higher outpatient engagement in the intervention group Stanley et al. (2018). This was a cohort comparison, not an individually randomized controlled trial. Its findings are consistent with benefit but cannot fully eliminate confounding by site or patient characteristics. A meta-analysis of safety-planning-type interventions offers a broader evidence perspective while revealing variation in intervention packages and study quality Nuij et al. (2021).
Follow-up contact has a distinct evidence tradition. Motto and Bostrom tested a structured postcrisis contact intervention in a randomized study, illustrating the possibility that continuing outreach may alter later outcomes Motto & Bostrom (2001). Subsequent acute-care trials evaluated more complex packages that combined assessment, safety work, and postdischarge engagement Miller et al. (2017). A major methodological issue is attribution: when several elements are delivered together, observed improvement does not establish which component produced it. Future studies need transparent descriptions of fidelity, frequency, contact acceptability, dropout, and the conditions under which outreach becomes intrusive or ineffective. Safety planning should be evaluated as a relational and service-integrated process, not merely as the presence of a completed form.
Dialectical Behavior Therapy, Adolescents, and Repeated Self-Harm
Dialectical behavior therapy integrates behavioral analysis, skills training, emotion regulation, distress tolerance, and collaborative therapeutic engagement. It has been investigated in populations characterized by recurrent self-harm and difficulties managing intense emotion. Its conceptual rationale is compatible with functional accounts in which self-injury temporarily alters an aversive state, although a treatment effect does not by itself prove that any single theorized mechanism is responsible. The intervention is usually delivered as a structured program with multiple interacting elements, and its effectiveness depends partly on access to suitably trained teams and appropriate continuity Klonsky (2007) McCauley et al. (2018).
In a randomized clinical trial, McCauley and colleagues studied dialectical behavior therapy for adolescents at high risk for suicide, providing direct comparative evidence on self-harm-related outcomes in the recruited population McCauley et al. (2018). The Cochrane review of 17 youth trials comprising 2,280 participants identified evidence favoring adolescent dialectical behavior therapy for repeated self-harm: four trials (270 participants) yielded an odds ratio of 0.46 (95% confidence interval, 0.26–0.82) compared with treatment as usual, enhanced usual care, or alternative psychotherapy at postintervention follow-up. The review emphasized uncertainty about other approaches and the still limited number of trials Witt et al. (2021a). The relative maturity of evidence for one intervention should not erase individual preferences or the importance of family and service context. It should guide the interpretation of what has been evaluated, in whom, and against which alternative.
The trial literature also raises important implementation questions. Adolescents differ in developmental capacities, family involvement, school circumstances, social supports, and barriers to attending repeated sessions. Most study samples do not capture every community that needs support. Researchers should examine treatment access, fidelity, attrition, caregiver involvement when appropriate, and outcomes beyond clinical events, including functioning and quality of life. Clinical systems require capacity to respond to acute distress between appointments. The scientific task is to identify the active and transferable elements of effective care while preserving the distinction between a treatment program's tested package and untested substitutions Witt et al. (2021a) NICE (2022).
Care Transitions, Discharge, and Continuity of Treatment
Periods after crisis presentation, hospital treatment, and transfer between services require particularly careful continuity. A change in location or clinician can disrupt access to medication review, psychosocial support, and practical arrangements. A discharge decision is therefore best understood as part of a coordinated plan, not the conclusion of a one-time risk calculation. NICE NG225 specifically cautions against using risk tools to determine discharge or entitlement to treatment and emphasizes appropriate psychosocial assessment and continuing care NICE (2022). For research, the relevant outcomes include whether follow-up was offered, attended, timely, acceptable, and associated with reduced recurrence.
The ED-SAFE research program evaluated a suicide-prevention approach implemented in emergency departments, including screening and follow-up components, and reported improvement in suicide-related outcomes compared with earlier phases or conditions of the study Miller et al. (2017). Evidence syntheses of brief interventions delivered in acute-care settings suggest that postdischarge prevention and linkage can produce beneficial effects, though effect estimates depend on the mix of studies and endpoints Doupnik et al. (2020). Because programs differ in screening, follow-up frequency, behavioral treatment, and community resources, a single pooled estimate should not be presented as a universal expected effect in every hospital.
Continuity also has ethical dimensions. The person must understand available services and have a realistic way to reach them. Plans that presuppose reliable transportation, private telephones, affordable appointments, or stable housing may fail for those who face structural barriers. Families and trusted supporters can facilitate follow-up when involvement is appropriate and consent or safeguarding rules permit it; their participation must respect personal safety and confidentiality. Researchers should measure the consequences of failed referrals and disparities in attendance, and health systems should evaluate actual handoffs rather than the production of discharge paperwork alone. Prevention is best observed across an entire pathway VA/DoD (2024) WHO (2021).
Pharmacotherapy, Somatic Treatment, and Evidence Boundaries
Medication may be essential for particular co-occurring psychiatric or medical conditions, but any claim that a pharmaceutical agent prevents suicide requires a precisely identified population, treatment, comparator, and outcome. Improvements in depression, agitation, sleep, or another symptom are not automatically evidence of fewer suicide deaths. Trials often exclude people with imminent suicide risk, and adverse effects, treatment discontinuation, and monitoring requirements can alter the balance of benefits and harms. The VA/DoD clinical practice guideline reviews relevant intervention evidence for identified patient populations, while systematic reviews examine the differing strengths and limitations of pharmacologic and psychosocial evidence VA/DoD (2024) D’Anci et al. (2019).
Pharmacological research is also vulnerable to confounding. Patients selected for intensive medication management may have greater initial illness severity, and observational associations cannot automatically establish whether an agent causes or prevents suicide-related outcomes. Drug-specific interpretation requires current product labeling, regulatory warnings, age-group considerations, comorbidity, and interaction assessment; a general narrative review cannot serve as a prescribing guide. The Cochrane review of self-harm interventions in youth identified no randomized pharmacological trials meeting its particular criteria, demonstrating why one cannot simply transplant evidence from adult populations or from broader diagnostic treatment trials Witt et al. (2021a).
Longer-term research should combine patient-centered outcomes with careful adverse-event surveillance and transparent characterization of treatment exposure. A person receiving medication can also benefit from targeted psychological treatment, crisis care, and reliable follow-up; complex care packages need study designs capable of separating effects where possible. The goal of this monograph is to locate medication research within the broader evidence architecture of suicide prevention. Treatment choices, emergency stabilization, and medication adjustments belong to appropriately qualified clinicians working with a particular patient under applicable guidance rather than to a general public-facing synthesis VA/DoD (2024) NICE (2022).
Engagement, Recovery, and Lived-Experience Evidence
Successful prevention includes more than avoiding a recorded event during a short study interval. People may value reduced distress, improved relationships, restored daily functioning, dignity during care, and confidence that help remains available. These outcomes should be measured with instruments appropriate to the population and interpreted together with behavioral endpoints. Intervention studies that collect only emergency visits can overlook meaningful improvements in functioning, and they may mistake avoidance of medical services for absence of suffering. A person-centered research framework therefore examines both safety-related outcomes and the experience of treatment, including the possibility that stigmatizing encounters impair engagement NICE (2022) Witt et al. (2021b).
Lived-experience research can reveal contextual factors that standardized scales omit: reasons disclosure feels unsafe, practical barriers to follow-up, expectations created by previous treatment, and what makes a safety plan usable. Qualitative interviews provide detailed mechanisms and hypotheses but do not automatically estimate prevalence or average intervention effects. Combining qualitative evidence with prospective follow-up and controlled comparisons can strengthen the interpretation of treatment trajectories. Study populations need to include people who disengage from services as well as those who complete interventions, because dropout is often correlated with unmet need. Researchers must preserve confidentiality and prevent participation from becoming a substitute for clinical care NICE (2022).
Recovery is usually nonlinear. There may be recurring distress, renewed support needs, and changes in protective circumstances. A narrow account that treats a person as permanently safe or permanently dangerous after one event undermines the logic of longitudinal care. Appropriate evaluation investigates treatment accessibility, meaningful choice, collaborative follow-up, and improvement across time. Where clinical claims are tentative, the paper should say so; optimism about recovery is compatible with precise assessment of uncertainty. The evidence supports designing systems in which people can receive and re-enter care without being defined by a historical risk label WHO (2021) VA/DoD (2024).
Prevention Across Universal, Selective, and Indicated Levels
Suicide prevention involves interventions at several levels of a population. Universal approaches address conditions or environments affecting broad groups; selective approaches focus on populations experiencing elevated exposure or vulnerability; indicated interventions engage people with identifiable suicidal thoughts, attempts, or self-harm. These categories describe the intended reach of a program, not an immutable classification of individuals. A national prevention strategy can combine surveillance, equitable service access, schools, community outreach, and clinical follow-up. The WHO LIVE LIFE implementation framework emphasizes a multisectoral response and practical interventions whose feasibility depends on local institutions WHO (2021). The scientific evaluation of each level requires attention to the mechanism and to the outcome that the program could plausibly change.
Universal prevention can alter environmental opportunities for harm, public knowledge, and service accessibility without relying on individual screening. Selective services may improve engagement among groups facing particular social or clinical pressures. Indicated care can address current distress and the aftermath of suicidal behavior. Evidence from a specialized clinical trial should not be used to infer the effectiveness of an unrelated universal campaign; conversely, an ecological trend following a public policy does not demonstrate benefit for each exposed person. The appropriate designs differ: individual randomization for feasible clinical interventions, cluster randomization for some schools or service systems, and carefully controlled quasi-experiments or interrupted time-series analyses for larger policies Zalsman et al. (2016) WHO (2021).
A prevention system's effectiveness depends on coverage, quality, persistence, and fair access. High-performing services in a small city may have little national effect if most people cannot reach them. Outreach that increases identification without treatment capacity can generate demand without adequate response. Conversely, a well-integrated care pathway may produce benefit through multiple mechanisms that a single component trial cannot capture. Evaluation should measure implementation fidelity, unintended consequences, socioeconomic disparities, and changes in actual attempts or deaths where sufficiently powered. Prevention in the Artificial Era adds possible digital routes for information and contact, but their value depends on demonstrated outcomes and responsible governance rather than novelty alone WHO (2021) Doupnik et al. (2020).
Environmental Means Safety and Responsible Public Communication
Limiting access to highly lethal means during periods of vulnerability is a central public-health prevention strategy. Its conceptual basis is temporal and environmental: the immediate circumstances of a crisis may change, and reducing availability of lethal options can create opportunities for interruption, support, and recovery. The details of any intervention belong to the relevant environment, law, professional practice, and safety governance. Evidence syntheses identify means restriction among the prevention approaches supported by research, while implementation studies must establish whether reductions in one form of death correspond to reductions in overall suicide mortality and whether effects persist Zalsman et al. (2016) WHO (2021).
Public communication also influences the conditions in which suicidal behavior is understood and discussed. Responsible reporting seeks accurate, contextual information about suicide, attention to available support, and avoidance of sensational presentation or unnecessarily detailed descriptions of harmful methods. These principles apply to journalism, educational materials, search systems, and generative digital services. The WHO LIVE LIFE framework explicitly identifies interaction with the media as a population-level prevention component WHO (2021). Evaluation must separate changes in reporting practices from changes in population outcomes and examine whether media interventions are adopted across outlets and languages.
Such environmental approaches are complementary to individual clinical care. A person may need urgent medical attention and continuing psychological support even when a community has effective population prevention policies. In parallel, a clinically responsive service operates in a world shaped by exposure, public communication, and material resources. The scientific challenge is to estimate the added value of policies in different contexts while respecting ecological inference limits. Reducing a national rate does not prove the causal pathway for an individual death, and a short period of stable rates does not necessarily invalidate an intervention in a setting with low statistical power WHO (2025) Zalsman et al. (2016).
School-Based Prevention and Developmentally Responsive Programs
Schools create an opportunity to provide broad preventive education and pathways to support young people, but they also carry distinctive safeguarding and privacy responsibilities. Programs may address emotional literacy, recognition of distress, help seeking, communication, and access to qualified services. Their evaluation should test outcomes beyond knowledge gains: a student may learn vocabulary about mental health without experiencing a change in later behavior. Cluster-randomized designs can reduce contamination across classrooms or schools while requiring statistical analysis that accounts for correlation within clusters Wasserman et al. (2015).
The Saving and Empowering Young Lives in Europe project investigated school-based approaches in a large cluster-randomized trial across multiple countries. Its Youth Aware of Mental Health program was associated with reductions in certain suicidal behavior outcomes at follow-up compared with control conditions Wasserman et al. (2015). This finding supports further evaluation of structured school programs, but transfer to another educational system requires consideration of age, language, program fidelity, local clinical referral capacity, and baseline exposure. A trial's effectiveness estimate is a property of its program, implementation, comparison, and population together.
Targeted clinical care remains necessary for students experiencing self-harm, active suicidal thoughts, or other serious difficulties. School staff have responsibilities for recognizing need and facilitating access to appropriate professional support, while qualified care is needed for comprehensive clinical assessment and treatment. Reviews of youth interventions show why prevention education and specialized therapy should be examined separately Witt et al. (2021a). Research also needs to test interventions for young people outside formal schooling, including those experiencing exclusion, displacement, or unstable living conditions. A comprehensive prevention architecture joins education, accessible services, confidential communication, and reliable referral systems WHO (2021).
Healthcare Systems, Implementation Quality, and Equity
Population health is shaped by what happens between first recognition of distress and sustained engagement with care. Service pathways include opportunities to disclose concerns, timely assessment, therapeutic communication, access to evidence-supported interventions, and re-entry after treatment interruptions. A person may traverse emergency services, primary care, specialist mental-health care, social services, and family or community resources. The quality of handoffs among these points is an empirical property of the system. Studies of acute-care interventions and postdischarge contact show the importance of evaluating complete pathways rather than isolated encounters Miller et al. (2017) Doupnik et al. (2020).
Implementation research asks whether interventions proven under trial conditions remain effective when adopted in ordinary care. Training intensity, caseload, staff turnover, supervision, referral availability, and fidelity all influence what is actually delivered. Health systems should track meaningful safety and engagement outcomes, avoid incentives to underreport adverse events, and evaluate patient experience alongside administrative metrics. Performance measures such as the number of screened patients or completed plans are process indicators; they do not alone establish fewer suicidal behaviors. Adaptation to rural areas, underfunded services, and different health insurance structures should be studied prospectively WHO (2021) NICE (2022).
Equity is a condition of credible generalization. When intervention trials enroll mostly people who can attend multiple visits, the findings may not describe those living far from care or facing unstable housing, stigma, language barriers, or economic precarity. Aggregate improvement can coexist with worsening outcomes for excluded groups. Researchers should report recruitment, dropout, and outcomes by relevant characteristics while guarding against stigmatizing interpretations of group differences. Community participation and lived-experience perspectives can identify barriers that routine clinical records conceal. An equitable prevention system assesses both who benefits from an intervention and who never receives it WHO (2025) VA/DoD (2024).
Evidence Quality, Heterogeneity, and Replicability
The evidence base for suicide and self-harm prevention is heterogeneous. Study populations range from adolescents with recurrent self-harm to adults after a recent attempt, veterans receiving emergency care, and entire school communities. Primary outcomes may be ideation, repeat self-harm, suicide attempts, care linkage, or suicide deaths. Follow-up durations range from weeks to years. A pooled effect can summarize compatible studies, yet it may conceal different mechanisms, measurement rules, or service conditions. The Cochrane reviews of adult and youth interventions and broader systematic reviews are useful precisely because they distinguish outcomes and evaluate study quality Witt et al. (2021b) Witt et al. (2021a) D’Anci et al. (2019).
Trials in populations with elevated event frequency can efficiently study recurrence but may not generalize to all people at risk. Studies powered to detect changes in suicide mortality require much larger samples and longer observation than many psychotherapy trials can obtain. Attrition, selective outcome reporting, incomplete blinding, and variation in usual care can bias effect estimates. Publication bias may favor apparently successful interventions. Interpretation should consider absolute event rates, confidence intervals, preregistration, the validity of endpoint adjudication, and independent replication. The statistical presence of an effect in one trial does not establish identical clinical impact in another health system Doupnik et al. (2020) McCauley et al. (2018).
Critical synthesis also requires evaluating contradictory and null findings. A poorly powered study can yield an inconclusive result despite a clinically important possible effect; a large observational association may arise from confounding. Reviews should avoid smoothing such uncertainty into a claim of universal consensus. The literature on prediction illustrates this danger: decades of identified risk factors have yielded limited individual predictive accuracy, while targeted interventions have shown benefits for particular populations and outcomes. The scientifically productive conclusion is to pursue better causal studies, service integration, and transparent replication without treating accurate personal forecasting as a prerequisite for prevention Franklin et al. (2017) Spittal et al. (2025).
Ethics, Autonomy, Confidentiality, and Institutional Responsibility
Suicide prevention must balance immediate safety, autonomy, privacy, relational trust, and institutional duties. Clinical encounters can involve difficult decisions about disclosure, consent, family involvement, and access to urgent support. Ethical reasoning is sensitive to capacity, developmental stage, jurisdiction, the seriousness and immediacy of danger, and the least restrictive feasible response. The scientific literature can evaluate whether different policies affect engagement and outcomes, while professional practice and law establish the authority for specific interventions. A respectful assessment includes the person's account and preferences wherever possible NICE (2022) VA/DoD (2024).
Confidentiality is especially important in settings where disclosure of suicidal thoughts may produce fear of punishment, exclusion, or surveillance. A service that records sensitive information needs clear access controls, retention procedures, and policies for when information may be shared. Research ethics likewise requires a plan for responding to disclosures made during participation. Data from clinical records and digital platforms can carry indirect identifiers that reveal intimate experiences even when explicit names are removed. The purpose of data collection, the limits of consent, and the burden of monitoring should be justified against demonstrated benefit. Predictive uncertainty strengthens the case for transparency and accountability in decisions that affect liberty or access to care Spittal et al. (2025) NICE (2022).
Artificial systems used for screening, triage, or conversational support introduce additional governance questions. An automated inference may be wrong, reflect a biased training distribution, or be presented with unjustified confidence. Its clinical role needs explicit validation, human accountability, mechanisms for correction, and testing of unintended harm. The independent philosophical discussion of Artificial agency in the wider Aisentica corpus does not supply evidence of diagnostic reliability in a health setting. In this monograph, artificial systems are evaluated by observable performance, privacy protection, and demonstrable changes in human outcomes. Their use should be proportionate to the quality and context of the evidence Spittal et al. (2025) Carter et al. (2017).
Research Agenda: Dynamic Mechanisms, Transportability, and Prevention Effects
The field needs studies that distinguish the development of suicidal ideation from behavioral transitions and that measure relevant processes at timescales appropriate to each hypothesis. Prospective designs should record distress, social circumstances, support, care access, and intervention exposure without substituting highly frequent measurement for ethical responsiveness. Research can test whether within-person changes in entrapment, emotional regulation, or connectedness precede changes in behavior, while explicitly addressing informative dropout and the possible effects of observation itself. Mechanism research should compare rival models and avoid inferring causal pathways from cross-sectional correlations O’Connor & Kirtley (2018) Klonsky & May (2015).
Prediction research needs honest external validation, calibration at clinically relevant horizons, and decision-focused evaluation. A statistically impressive area under the curve is insufficient if a model produces excessive false alarms, misses people in distress, or changes access to care unfairly. Future studies should preregister methods, report subgroup performance where feasible, and evaluate whether deploying a score improves outcomes compared with responsive assessment and established services. Prediction may support service planning under constrained conditions, but it must be assessed as a real intervention in a system rather than as an abstract ranking exercise Spittal et al. (2025) Carter et al. (2017).
Prevention research should expand access to high-quality trials and policy evaluations across low- and middle-income settings, underrepresented groups, and ordinary service environments. The strongest next studies will link treatment fidelity, participant experience, economic feasibility, and meaningful behavioral or mortality outcomes. Multi-country comparisons should document differences in reporting systems and local implementation. The WHO strategy offers a shared policy vocabulary while requiring rigorous local evaluation WHO (2021) WHO (2025). A mature evidence architecture connects conceptual precision, causal mechanisms, culturally valid assessment, ethically sustainable services, and measurable reductions in harm without promising that every individual crisis can be forecast.
Conclusion
Suicide and self-harm research establishes a scientific field whose objects must be distinguished with care: thoughts about death, suicidal intent, attempts, nonsuicidal self-injury, ambiguous self-harm, and suicide mortality differ in definition, measurement, and causal interpretation. Their relationship unfolds across psychological mechanisms, social conditions, development, clinical care, and public policy. The history of research shows progress from descriptive associations toward prospective study, specific intervention trials, and multilevel prevention. It also reveals durable limits in translating population associations into precise forecasts about an individual. The scientific synthesis therefore places psychosocial formulation, responsive care, and outcome-focused research at the center of the field Franklin et al. (2017) NICE (2022).
The evidence supports several complementary directions: structured and compassionate assessment; collaborative safety planning integrated with follow-up; selected cognitive-behavioral and dialectical behavior treatments evaluated in appropriate populations; attentive transitions after crisis care; and population prevention addressing environmental and service conditions Brown et al. (2005) Stanley et al. (2018) McCauley et al. (2018) WHO (2021). These findings have specific evidentiary boundaries. A cohort comparison is not a randomized trial, improved engagement is not identical to reduced mortality, and a favorable screening metric does not establish individual predictive certainty. A responsible monograph retains such distinctions while bringing the literature into a coherent scientific account.
The unresolved tasks are substantive: stronger longitudinal identification of dynamic mechanisms, more representative and independently replicated trials, better cross-cultural validation, improved service implementation, and rigorous assessment of digital and artificial tools. The historical horizon of the Artificial Era makes these issues increasingly consequential because automated systems may shape how distress is detected, classified, and communicated. Their legitimate role depends on demonstrable benefit, transparent uncertainty, patient privacy, and accountable care. The central research commitment is to advance prevention through evidence about actual people, environments, and outcomes while preserving the dignity and complexity of every person who seeks support Spittal et al. (2025) VA/DoD (2024).
References
- Brown, G. K., Ten Have, T., Henriques, G. R., Xie, S. X., Hollander, J. E., & Beck, A. T. (2005). Cognitive therapy for the prevention of suicide attempts: A randomized controlled trial. JAMA, 294(5), 563–570. https://doi.org/10.1001/jama.294.5.563
- Carter, G., Milner, A., McGill, K., Pirkis, J., Kapur, N., & Spittal, M. J. (2017). Predicting suicidal behaviours using clinical instruments: Systematic review and meta-analysis of positive predictive values for risk scales. The British Journal of Psychiatry, 210(6), 387–395. https://doi.org/10.1192/bjp.bp.116.182717
- Centers for Disease Control and Prevention. (n.d.). Suicide data and statistics. National Center for Injury Prevention and Control. https://www.cdc.gov/suicide/data/index.html
- Cipriano, A., Cella, S., & Cotrufo, P. (2017). Nonsuicidal self-injury: A systematic review. Frontiers in Psychology, 8, Article 1946. https://doi.org/10.3389/fpsyg.2017.01946
- D’Anci, K. E., Uhl, S., Giradi, G., & Martin, C. (2019). Treatments for the prevention and management of suicide: A systematic review. Annals of Internal Medicine, 171(5), 334–342. https://doi.org/10.7326/M19-0869
- Dazzi, T., Gribble, R., Wessely, S., & Fear, N. T. (2014). Does asking about suicide and related behaviours induce suicidal ideation? What is the evidence? Psychological Medicine, 44(16), 3361–3363. https://doi.org/10.1017/S0033291714001299
- Doupnik, S. K., Rudd, B., Schmutte, T., Worsley, D., Bowden, C. F., McCarthy, E., Eggan, E., Bridge, J. A., & Marcus, S. C. (2020). Association of suicide prevention interventions with subsequent suicide attempts, linkage to follow-up care, and depression symptoms for acute care settings: A systematic review and meta-analysis. JAMA Psychiatry, 77(10), 1021–1030. https://doi.org/10.1001/jamapsychiatry.2020.1586
- Franklin, J. C., Ribeiro, J. D., Fox, K. R., Bentley, K. H., Kleiman, E. M., Huang, X., Musacchio, K. M., Jaroszewski, A. C., Chang, B. P., & Nock, M. K. (2017). Risk factors for suicidal thoughts and behaviors: A meta-analysis of 50 years of research. Psychological Bulletin, 143(2), 187–232. https://doi.org/10.1037/bul0000084
- Horowitz, L. M., Bridge, J. A., Teach, S. J., Ballard, E., Klima, J., Rosenstein, D. L., Wharff, E. A., Ginnis, K., Cannon, E., Joshi, P., & Pao, M. (2012). Ask Suicide-Screening Questions (ASQ): A brief instrument for the pediatric emergency department. Archives of Pediatrics & Adolescent Medicine, 166(12), 1170–1176. https://doi.org/10.1001/archpediatrics.2012.1276
- Klonsky, E. D. (2007). The functions of deliberate self-injury: A review of the evidence. Clinical Psychology Review, 27(2), 226–239. https://doi.org/10.1016/j.cpr.2006.08.002
- Klonsky, E. D., & May, A. M. (2015). The Three-Step Theory (3ST): A new theory of suicide rooted in the “ideation-to-action” framework. International Journal of Cognitive Therapy, 8(2), 114–129. https://doi.org/10.1521/ijct.2015.8.2.114
- McCauley, E., Berk, M. S., Asarnow, J. R., Adrian, M., Cohen, J., Korslund, K., Avina, C., Hughes, J., Harned, M., Gallop, R., & Linehan, M. M. (2018). Efficacy of dialectical behavior therapy for adolescents at high risk for suicide: A randomized clinical trial. JAMA Psychiatry, 75(8), 777–785. https://doi.org/10.1001/jamapsychiatry.2018.1109
- Miller, I. W., Camargo, C. A., Jr., Arias, S. A., Sullivan, A. F., Allen, M. H., Goldstein, A. B., Manton, A. P., Espinola, J. A., Jones, R., Hasegawa, K., & Boudreaux, E. D. (2017). Suicide prevention in an emergency department population: The ED-SAFE study. JAMA Psychiatry, 74(6), 563–570. https://doi.org/10.1001/jamapsychiatry.2017.0678
- Motto, J. A., & Bostrom, A. G. (2001). A randomized controlled trial of postcrisis suicide prevention. Psychiatric Services, 52(6), 828–833. https://doi.org/10.1176/appi.ps.52.6.828
- National Institute for Health and Care Excellence. (2022). Self-harm: Assessment, management and preventing recurrence (NICE guideline NG225; reviewed 2024). https://www.nice.org.uk/guidance/ng225/chapter/recommendations
- Nuij, C., van Ballegooijen, W., de Beurs, D., Juniar, D., Erlangsen, A., Portzky, G., O’Connor, R. C., Smit, J. H., Kerkhof, A., & Riper, H. (2021). Safety planning-type interventions for suicide prevention: Meta-analysis. The British Journal of Psychiatry, 219(2), 419–426. https://doi.org/10.1192/bjp.2021.50
- O’Connor, R. C., & Kirtley, O. J. (2018). The integrated motivational–volitional model of suicidal behaviour. Philosophical Transactions of the Royal Society B: Biological Sciences, 373(1754), Article 20170268. https://doi.org/10.1098/rstb.2017.0268
- Posner, K., Brown, G. K., Stanley, B., Brent, D. A., Yershova, K. V., Oquendo, M. A., Currier, G. W., Melvin, G. A., Greenhill, L., Shen, S., & Mann, J. J. (2011). The Columbia–Suicide Severity Rating Scale: Initial validity and internal consistency findings from three multisite studies with adolescents and adults. American Journal of Psychiatry, 168(12), 1266–1277. https://doi.org/10.1176/appi.ajp.2011.10111704
- Ribeiro, J. D., Franklin, J. C., Fox, K. R., Bentley, K. H., Kleiman, E. M., Chang, B. P., & Nock, M. K. (2016). Self-injurious thoughts and behaviors as risk factors for future suicide ideation, attempts, and death: A meta-analysis of longitudinal studies. Psychological Medicine, 46(2), 225–236. https://doi.org/10.1017/S0033291715001804
- Rudd, M. D., Bryan, C. J., Wertenberger, E. G., Peterson, A. L., Young-McCaughan, S., Mintz, J., Williams, S. R., Arne, K. A., Breitbach, J., Delano, K., Wilkinson, E., & Bruce, T. O. (2015). Brief cognitive-behavioral therapy effects on post-treatment suicide attempts in a military sample: Results of a randomized clinical trial with 2-year follow-up. American Journal of Psychiatry, 172(5), 441–449. https://doi.org/10.1176/appi.ajp.2014.14070843
- Spittal, M. J., Guo, X. A., Kang, L., Kirtley, O. J., Clapperton, A., Hawton, K., Kapur, N., Pirkis, J., & Carter, G. (2025). Machine learning algorithms and their predictive accuracy for suicide and self-harm: Systematic review and meta-analysis. PLOS Medicine, 22(9), Article e1004581. https://doi.org/10.1371/journal.pmed.1004581
- Stanley, B., Brown, G. K., Brenner, L. A., Galfalvy, H. C., Currier, G. W., Knox, K. L., Chaudhury, S. R., Bush, A. L., & Green, K. L. (2018). Comparison of the Safety Planning Intervention with follow-up vs usual care of suicidal patients treated in the emergency department. JAMA Psychiatry, 75(9), 894–900. https://doi.org/10.1001/jamapsychiatry.2018.1776
- U.S. Department of Veterans Affairs & U.S. Department of Defense. (2024). VA/DoD clinical practice guideline for assessment and management of patients at risk for suicide (Version 3.0). https://www.healthquality.va.gov/guidelines/MH/srb/
- Wasserman, D., Hoven, C. W., Wasserman, C., Wall, M., Eisenberg, R., Hadlaczky, G., Kelleher, I., Sarchiapone, M., Apter, A., Balazs, J., Bobes, J., Brunner, R., Corcoran, P., Cosman, D., Guillemin, F., Haring, C., Iosue, M., Kaess, M., Kahn, J.-P., ... Carli, V. (2015). School-based suicide prevention programmes: The SEYLE cluster-randomised, controlled trial. The Lancet, 385(9977), 1536–1544. https://doi.org/10.1016/S0140-6736(14)61213-7
- Witt, K. G., Hetrick, S. E., Rajaram, G., Hazell, P., Taylor Salisbury, T. L., Townsend, E., & Hawton, K. (2021a). Interventions for self-harm in children and adolescents. Cochrane Database of Systematic Reviews, 2021(3), CD013667. https://doi.org/10.1002/14651858.CD013667.pub2
- Witt, K. G., Hetrick, S. E., Rajaram, G., Hazell, P., Taylor Salisbury, T. L., Townsend, E., & Hawton, K. (2021b). Psychosocial interventions for self-harm in adults. Cochrane Database of Systematic Reviews, 2021(4), CD013668. https://doi.org/10.1002/14651858.CD013668.pub2
- World Health Organization. (2021). LIVE LIFE: An implementation guide for suicide prevention in countries. World Health Organization. https://www.who.int/publications/i/item/9789240026629
- World Health Organization. (2025). Suicide worldwide in 2021: Global health estimates. World Health Organization. https://www.who.int/publications/i/item/9789240110069
- Zalsman, G., Hawton, K., Wasserman, D., van Heeringen, K., Arensman, E., Sarchiapone, M., Carli, V., Höschl, C., Barzilay, R., Balazs, J., Purebl, G., Kahn, J.-P., Sáiz, P. A., Bursztein Lipsicas, C., Bobes, J., Cozman, D., Hegerl, U., & Zohar, J. (2016). Suicide prevention strategies revisited: 10-year systematic review. The Lancet Psychiatry, 3(7), 646–659. https://doi.org/10.1016/S2215-0366(16)30030-X
Social Relationships, Development, and Structural Context
Suicide and self-harm take place within relationships, institutions, and unequal social conditions. Family conflict, bereavement, violence, peer experiences, exclusion, unemployment, and barriers to care may be associated with distress and suicidal behavior, but their effects depend on context and the measurements used. Social connection can create access to practical help, protection from isolation, and opportunities to revise a perceived hopeless situation. Schools, workplaces, and healthcare institutions can also shape whether individuals reveal difficulties or encounter a respectful response. A mechanistic analysis asks which relational processes, material conditions, and service characteristics alter trajectories, rather than treating social context as a static background covariate WHO (2021) Zalsman et al. (2016).
Development changes both the organization of distress and the opportunities for prevention. In adolescence, family relationships, peer networks, educational contexts, identity development, and varying capacities for autonomous consent interact with mental-health needs. Research on school-based prevention and adolescent treatment provides distinct evidence about population-level psychoeducation and clinical intervention Wasserman et al. (2015) McCauley et al. (2018). Adult experiences may involve work and caregiving strain, disability, chronic health conditions, or transitions after hospital discharge. Older adults may face changing health and social roles. Population comparisons require measurements that are valid in each group; a questionnaire validated in a pediatric emergency department cannot automatically be transferred to a geriatric primary-care cohort.
Structural prevention attends to the conditions that determine access to safe services and exposure to preventable harms. Rural residence, poverty, discrimination, displacement, and healthcare shortages can affect both the occurrence of crises and the probability that they are counted in official data. An individual-risk score cannot supply housing, timely clinical follow-up, or affordable care. The WHO population framework therefore includes policy actions alongside health-system identification and support WHO (2021). Evaluation must use designs that can identify whether a policy changed behavior or mortality rather than merely whether public awareness increased. Local adaptation should preserve the core mechanism under study while acknowledging cultural differences in language, stigma, consent, and help seeking.