Publications (see right side for filters!)

Researchers developed and validated EHR‑based metrics, including natural language processing of clinical notes, to track key components of the Stanley‑Brown Safety Planning Intervention across six health systems. They found that professional contact information and lethal‑means counseling for firearms and medications were most commonly documented, and that structured templates and NLP tools can reliably monitor safety‑planning uptake in Zero Suicide evaluations.

Linking state death records with EHR data from three health systems, this report compares patterns of health care use in the year before firearm suicide versus other suicide deaths. Most decedents in both groups had recent health visits, but people who died by firearm were less likely to have received mental health care, suggesting that firearm‑specific prevention must extend beyond specialty mental health encounters.

During the COVID‑19 pandemic, three health systems delivered a virtual version of the Guiding Good Choices parenting program to families recruited through primary care and assessed implementation fidelity. While adherence, quality, and engagement were high, attendance was lower than expected, indicating that virtual delivery in primary care is feasible but requires additional strategies to retain families.

Qualitative interviews with health care workers in the ASPIRE trial show strong consensus that pediatric clinicians have a responsibility to promote firearm safety during well‑child visits. S.A.F.E. Firearm’s brief, universal, nonjudgmental approach and emphasis on family autonomy facilitated acceptability, but successful scale‑up depends on clear communication of the program’s harm‑reduction focus and simple, flexible training.

Analyzing nearly 4.7 million psychiatry and addiction visits across six health systems, this study examines racial‑ethnic variation in receipt of suicide screening, risk assessment, and intervention under Zero Suicide. After adjustment, Black patients were slightly less likely to be screened but somewhat more likely to receive evidence‑based interventions, while American Indian/Alaska Native patients had the lowest unadjusted intervention rates, pointing to areas for targeted equity work.

In this quality improvement study, ZS model implementation was associated with a reduction in suicide attempt rates among patients accessing outpatient mental health care at most study sites, which supports widespread efforts to implement the ZS model in these settings within US health systems.

Among adolescents with depression who attempted self‑harm or died by suicide within 30 or 90 days of PHQ‑9 screening, nearly one‑third reported “not at all” to item 9 on suicidal thoughts. Older teens, those screened in primary care, and youth with prior inpatient or mental health diagnoses were more likely to deny ideation, indicating that clinicians must look beyond item 9 when assessing suicide risk.

This review summarizes current evidence on youth suicide epidemiology and highlights pediatric primary care as a critical setting for identification and intervention. It outlines practical strategies—routine screening, structured risk assessment, brief interventions, and coordinated follow‑up—that can be integrated into primary care workflows to better detect and manage suicide risk in children and adolescents.

In this cluster randomized trial among 47 307 well-child visits at 30 clinics in Michigan and Colorado, the chance of receiving the firearm storage program was significantly higher in clinics in the nudge+ group than in the nudge group. The findings suggest that an EHR strategy combined with facilitation may be more effective at increasing delivery of an evidence-based firearm storage program in pediatric primary care than an EHR strategy alone.

Using EHR data from a community health system, researchers built random‑forest and penalized regression models to predict PHQ‑9 outcomes and response after starting antidepressant treatment. Model performance was modest, with poor discrimination for favorable response, leading the authors to conclude that current EHR‑based prediction is not accurate enough to guide antidepressant selection and that early symptom monitoring remains essential.

This mixed‑methods study describes integrating the Lock to Live web‑based firearm safety decision aid into an integrated health system’s EHR safety‑planning workflows. Patient and clinician interviews informed implementation strategies—such as open, validating conversations and embedding L2L links and QR codes in templates—leading to rising adoption over two years, particularly in mental health clinics.

In this randomized trial of high‑risk patients identified by suicide prediction models, a brief patient‑portal message inviting use of the Lock to Live decision aid modestly shifted firearm storage readiness toward preparation and action, but did not change actual storage behaviors or medication storage. The findings suggest that low‑touch digital outreach alone is insufficient and should be paired with stronger interventions, such as clinician counseling nudges.

Across six diverse health systems, this study measures how often patients received suicidal‑ideation screening, formal risk assessment, and lethal‑means counseling between 2010 and 2019. Screening and assessment increased over time but varied widely by site and setting, with lower rates in primary care than specialty mental health, highlighting opportunities to broaden screening and ensure risk assessment with lethal‑means counseling after positive screens.

Using over 25 million mental health visits from seven health systems, this study compares random forests, neural networks, and ensemble models with logistic regression models for predicting suicidal behavior. All approaches performed well, but complex machine‑learning models offered only minimal gains over a simpler regression model with about 100 predictors, suggesting that more transparent parametric models may be preferable for routine clinical use.

This case‑control study of women aged 15–44 in nine health systems found that suicide deaths were strongly associated with prior mental health and substance use disorders and recent emergency department use. Non‑Hispanic White and perinatal women were less likely to die by suicide, indicating that reproductive‑age women with behavioral health conditions, ED encounters, and from minority racial and ethnic groups may benefit from enhanced screening and follow‑up.

This editorial responds to updated USPSTF guidance that supports depression screening but finds insufficient evidence for universal suicide‑risk screening, arguing that the key question is how to respond to suicidal ideation and risk scores rather than whether to ask. The authors review evidence on predictive value of PHQ‑9 item 9 and EHR‑based risk models, caution against coercive or high‑harm responses, and call for collaborative, respectful, process‑focused approaches to suicide prevention in health care.

This study evaluated how accurately ICD‑10‑CM injury and poisoning codes capture self‑harm events among high‑risk patients reporting frequent suicidal ideation across seven health systems. Review of clinical text showed that nearly 90% of encounters coded as self‑harm had clear documentation of self‑harm intent, while smaller proportions of encounters coded as undetermined, accidental, or with no intent code reflected misclassified self‑harm. These findings suggest that relying on ICD‑10‑CM self‑harm intent codes identifies most true self‑harm events but will miss a modest number that are embedded in other injury and poisoning categories.

This narrative review summarizes recent text‑mining applications to suicide research using electronic health records, social media posts, and death records. It shows how natural language processing has been used to uncover suicide risk factors in both general and specific populations, improve EHR‑based risk prediction, and monitor shifts in suicidal discourse after events like COVID‑19 or celebrity deaths. The authors call for future work that links multiple data sources, tests NLP‑driven interventions, standardizes reporting of model performance, and uses implementation science to understand how these tools can be feasibly and ethically integrated into real‑world suicide prevention efforts.

This descriptive study examined where the medications used in suicidal overdoses came from among people who died by intentional opioid or psychotropic/hypnotic poisoning across nine health systems. Most individuals who died by intentional opioid poisoning had filled opioid prescriptions in the prior year, and most who died by psychotropic/hypnotic poisoning had recently and repeatedly been dispensed those same medication classes. The findings suggest that suicidal poisonings often involve recently prescribed drugs, supporting medication‑focused lethal‑means counseling, quantity limits, naloxone distribution, and safer packaging (such as blister packs) as key suicide prevention strategies.

Among Veterans with opioid use disorder, clinically recognized suicidality was associated with a slightly higher likelihood of starting medications for OUD but a lower likelihood of continuing them the following year. The findings point to an opportunity to both expand initiation and strengthen retention in MOUD for patients with suicidality.

Interviews with clinicians and leaders from three health systems found general support for EHR‑based suicide risk prediction models but highlighted the importance of early buy‑in, clear role expectations, and adequate resources. Participants wanted transparent information about how models estimate risk, efficient EHR interfaces, and explicit guidance for responding to risk alerts without overburdening workflows.

This study evaluated prediction models that use electronic health record data at baseline to forecast psychotherapy outcomes for depression, including follow-up PHQ‑9 scores and treatment response. The models showed limited ability to predict individual outcomes accurately, leading the authors to caution against relying heavily on such tools for clinical decision-making.

This cost-effectiveness study evaluated the Mindful Mood Balance (MMB) web program added to usual depression care for adults with residual depressive symptoms. Patients receiving MMB plus usual care had 29 more depression-free days over 12 months than those receiving usual care alone, at an incremental cost of about $432, yielding incremental costs of roughly $10–15 per depression-free day gained depending on which care costs were included. The authors conclude that MMB is a cost-effective adjunctive web-based intervention for reducing residual symptoms and preventing relapse and recommend that health systems consider adopting it alongside traditional services.

Using updated CDC mortality data through 2020, this brief report shows that firearm‑related injuries have surpassed motor vehicle crashes as the leading cause of death among US children and adolescents. The sharp increase in firearm homicides and concurrent rise in drug poisoning deaths during the first pandemic year underscore the urgency of sustained investment in firearm‑violence prevention and youth safety initiatives.

This multi‑system case‑control study used EHR data to identify over 200 diagnosis and utilization indicators of suicide death among adults with and without mental health diagnoses. It found that most indicators were associated with suicide, including malignant cancers in those without mental health diagnoses and distinct psychiatric indicators in those with mental health diagnoses, while several benign conditions and preventive care patterns appeared protective. Latent class models revealed five subgroups with different risk profiles, showing that the highest‑risk groups in both strata had heavy, complex healthcare use and the lowest‑risk groups were younger women who frequently used preventive services, highlighting diverse opportunities for targeted suicide prevention in health systems.

This large pragmatic randomized trial tested two low‑intensity outreach programs delivered mainly via patient portal messaging for adults reporting frequent suicidal thoughts in four US health systems. A care‑management program did not reduce risk of fatal or nonfatal self‑harm compared with usual care, and a brief dialectical behavior therapy–based skills training program was associated with a modest but significant increase in self‑harm events. The findings suggest that these specific low‑touch, remote interventions are not effective—and may be harmful—for suicide prevention in high‑risk outpatients, underscoring the need for more intensive or differently designed approaches.

This article describes a stepwise process to embed health equity into a hybrid effectiveness‑implementation trial of the S.A.F.E. Firearm program in pediatric primary care. Using the Health Equity Implementation Framework, pilot data and clinician interviews were used to surface potential inequities by race, ethnicity, and sex assigned at birth and to plan equity‑focused adaptations and monitoring in the full trial.

his article highlights the rising burden of firearm deaths among children and adolescents in the United States and the protective role of triple‑safe storage (locked, unloaded, ammunition stored separately). It argues that pediatric health care providers are uniquely positioned to counsel families about secure firearm storage but often lack training, confidence, and shared language with firearm‑owning families. The authors synthesize existing evidence and expert guidance to propose a practical, culturally respectful framework and suggested terminology that pediatric clinicians can use to engage caregivers in patient‑centered firearm safety conversations.

This mixed‑methods quality‑improvement study describes adding an encounter‑based suicide‑risk flag from a prediction model to existing PHQ‑9/C‑SSRS workflows in an integrated health system. The flag rarely triggered additional risk assessments when patients denied suicidal ideation, and interviews revealed clinician and patient concerns about workflow burden, model accuracy, and potential coercive responses, underscoring the need for careful implementation design.

This study evaluated ways to deliver Lock to Live, a web‑based decision aid that helps patients at suicide risk choose safer firearm and medication storage options. Among outpatients reporting suicidal ideation on the PHQ‑9, population‑based electronic outreach (especially EHR portal messages plus email reminders) achieved modest uptake but reached far more patients overall than traditional provider referral, which had perfect uptake but very limited reach. Provider interviews highlighted that reminders, training, clear policies, quality metrics, and leadership support could help integrate L2L into routine care, suggesting that scalable digital outreach combined with system supports may broaden access to lethal‑means counseling tools.

This study asked whether EHR‑based machine‑learning suicide risk models mainly flag unexpected patients or the same individuals already known to be high‑risk. Using more than 27 million outpatient visits with mental health diagnoses, the authors found that visits in the highest predicted‑risk percentiles almost always involved patients with traditional risk markers, such as prior self‑harm diagnoses or recent mental‑health–related emergency or hospital care. These results indicate that model‑generated suicide risk alerts largely coincide with patients clinicians would already view as high‑risk at those visits, rather than uncovering large numbers of people without documented risk factors.

This study tested whether adding detailed opioid‑related information to existing EHR‑based suicide risk prediction models improves their performance. Incorporating more than 600 opioid variables did not meaningfully change model accuracy for suicide attempts or deaths in mental health specialty care and produced mixed effects in primary care, with a modest gain for predicting suicide deaths but slightly worse performance for attempts. Overall, the findings suggest that, among patients already identified with mental health diagnoses, mental health predictors carry most of the predictive signal and extensive opioid‑related data provide little additional benefit for suicide risk prediction.

Through qualitative interviews with adults who completed a self‑administered mental health questionnaire and clinicians responsible for safety planning, this study explores experiences with a single standardized item on firearm access. Both groups saw value in the question for prompting conversations, but emphasized the need for nonjudgmental, patient‑centered language and system supports to make firearm discussions more acceptable and effective.

In interviews across three health systems, most patients endorsed using EHR‑based suicide risk algorithms to identify people who might need help, especially when followed by empathetic conversations with experienced clinicians. Concerns centered on possible anxiety, stigma, or coercive treatment, suggesting that implementation should prioritize transparent communication, voluntary engagement, and monitoring for unintended harms.

This case series examines how individuals who died by suicide responded to a survey question regarding access to firearms before death.

This editorial discusses a machine‑learning study that uses California handgun purchase data to predict firearm suicide within one year of a transaction. It explains that the random‑forest model identified a small absolute number of suicides, but concentrated many of them in the highest‑risk purchasers, and highlights how adding richer clinical data could further improve prediction. The authors emphasize the need to interpret low positive predictive value cautiously, warning against using such models to mandate point‑of‑sale restrictions or legal actions, and instead suggest they may be better suited to guide low‑burden, voluntary prevention and education efforts among firearm retailers and purchasers.

Drawing on NIH Collaboratory demonstration projects, this paper describes how concurrent quality‑improvement initiatives can overlap with embedded pragmatic trials and complicate interpretation of intervention effects. The authors share case‑based strategies—such as ongoing monitoring of QI, close collaboration with health‑system leaders, and flexible trial design—to keep research and operational improvement aligned in real‑world health systems.

Across six large health systems, embedded researchers cataloged how Zero Suicide practices for identification, engagement, treatment, and care transitions were being implemented and measured. Identification practices (screening and risk stratification) were the most defined and tracked, whereas engagement, treatment, and transition supports were less standardized and rarely measured, pointing to priorities for strengthening Zero Suicide implementation.

This pilot tested a low‑intensity, web‑based dialectical behavior therapy skills program with secure‑message coaching for patients flagged via PHQ‑9 suicidal‑thought responses in the EHR. Almost all invitees opened the message and nearly half consented, with participants engaging repeatedly with the site and describing the skills as helpful, supporting feasibility for a larger randomized trial.

When an integrated system rapidly transitioned mental health care from in‑person to telehealth during COVID‑19, completion of standard screening questionnaires and structured suicide‑risk assessments dropped sharply but later recovered after EHR‑based portal workflows and alerts were introduced. The experience shows that patient‑portal tools can help maintain systematic suicide‑risk identification in virtual care.

This study re-evaluated previously published EHR-based suicide risk prediction models using over 10 million mental health and general medical visits from 2014–2017 across seven health systems. Models remained well calibrated and accurately predicted suicide attempts, with c‑statistics around 0.86 and nearly half of attempts and about one‑third of deaths occurring after visits with risk scores in the top 5%. Performance for suicide death prediction was modestly lower (c‑statistics about 0.80), but the authors conclude that these models still perform similarly to historical estimates and can inform current clinical practice and patient care.

This practical article outlines six key actions for system‑wide suicide‑risk screening within the Zero Suicide framework, including leadership buy‑in, cross‑disciplinary workgroups, validated tools, EHR‑embedded workflows, clear clinical responsibilities, and ongoing performance feedback. Together, these steps are presented as a roadmap for integrating routine suicide‑risk screening into large health systems.

This cross-sectional study examined how often adult patients disclose firearm access when a single standardized question is added to a routine mental health questionnaire in primary care and outpatient mental health clinics. Among more than 128,000 patients, 83.4% in primary care and 91.8% in mental health answered the firearm question, and of those who responded, 20.9% and 15.3%, respectively, reported having firearm access. The authors conclude that most patients will answer a standard firearm access question, and that using such questions alongside counseling and decision-making resources could strengthen suicide prevention practice and guideline development.

This study tested whether baseline EHR data and machine learning models can accurately predict outcomes of psychotherapy for depression. Models using demographics, diagnoses, medications, and prior encounters performed poorly, with inaccurate predictions of follow-up PHQ‑9 scores and low discrimination for treatment response (AUC around 0.6). The authors conclude that even with rich EHR data and advanced methods, prediction of individual depression outcomes was not accurate, so health systems should be cautious about deploying such models without rigorous validation.

This column describes how health systems in the Mental Health Research Network function as learning health care systems that integrate suicide prevention research into routine practice. It highlights specific implementation efforts and discusses practical challenges of conducting embedded suicide prevention research.

This large multi‑system study examined whether PHQ‑9 item 9 (thoughts of death or self‑harm) predicts suicide attempts and deaths among outpatients with substance use disorders. Higher item‑9 scores, especially “nearly every day,” were strongly associated with increased 90‑day risk of suicide attempt and death, but nearly half of subsequent attempts occurred after visits where patients answered “not at all,” indicating that low scores do not eliminate suicide risk and should not preclude preventive intervention.

This large EHR-based study of more than 37,000 adults with bipolar disorder examined whether PHQ‑9 scores predict suicidal behavior. Patients reporting nearly daily suicidal ideation had a 4.21% 90‑day suicide attempt risk versus 0.74% among those reporting none and were almost four times more likely to attempt suicide and about fourteen times more likely to die by suicide; recent self-harm also conferred very high risk. The authors conclude that PHQ‑9 suicidal ideation is a robust short‑term indicator of suicide risk in bipolar disorder and should prompt thorough clinical evaluation.

This case-control study compared BMI changes over the year before death among 387 adults who died by suicide and 1,935 matched controls in integrated health systems. People who died by suicide had a small but significant BMI decrease (−0.72 kg/m²) versus a slight increase (0.06 kg/m²) in controls, and each one‑unit BMI drop was associated with an 11% higher odds of suicide after adjusting for demographics, depression, and comorbidities. The authors conclude that weight loss in the year before death, including among people without diagnosed depression, is associated with increased suicide mortality and may serve as a clinical warning sign.

Qualitative interviews with primary‑care patients who had been asked a standard firearm‑access question revealed concerns about privacy, autonomy, and stigma, as well as uncertainty about how the information would be used. Participants suggested that framing firearm questions within broader injury prevention, clearly explaining their purpose, and explicitly linking them to suicide‑risk discussions could make disclosure feel safer and more patient‑centered.

This quasi-experimental study of adults in Kaiser Permanente Colorado who reported suicide ideation on the PHQ‑9 found that only one-third received documented counseling about access to lethal means. Lethal means assessment was associated with a substantial reduction in suicide attempts or deaths within 180 days (from 3.3% to 0.83%), supporting broader use of this counseling alongside other suicide prevention practices.

Using linked AUDIT‑C and PHQ‑9 data from mental health visits, this study found that patients who reported suicidal ideation had markedly higher 90‑day suicide‑attempt risk at every level of drinking, especially with high consumption or frequent heavy episodic use. Among patients denying suicidal thoughts, alcohol‑use patterns were not associated with attempts, highlighting the importance of offering alcohol‑related care when ideation is present and improving ways to detect risk when it is not.

The study examined whether following benzodiazepine prescribing guidelines in patients with anxiety or sleep disorders is associated with suicide risk. It found that, among patients with anxiety disorders, guideline‑concordant benzodiazepine use—short to moderate duration and combined with antidepressants or psychotherapy—was associated with lower odds of suicide, while a similar pattern in sleep disorders did not reach statistical significance.

Using PHQ‑9 data from 5,554 new psychotherapy episodes in two health systems, this study compared four metrics of depression treatment success: response, remission, effect size, and severity‑adjusted effect size. Effect size–based measures were most often achieved but were strongly influenced by baseline symptom severity, whereas response (≥50% PHQ‑9 reduction) was not biased by starting severity, was clinically meaningful, and was simple to calculate; the authors therefore recommend response as the preferred outcome measure for comparing depression treatment results.

This study examined how often self-harm events are missed when using ICD‑10‑CM injury and poisoning codes that label encounters as accidents, undetermined intent, or give no intent code among high‑risk patients in a suicide prevention trial. By reviewing clinical notes for these encounters, investigators found that only a small fraction represented misclassified self-harm—adding all of them would increase identified self-harm events by about 10%—suggesting ICD‑10‑CM self-harm coding in this setting is reasonably accurate for surveillance and evaluation.

This study used natural language processing of electronic health records to measure how often clinicians documented assessing access to lethal means (firearms and medications) among adults reporting suicidal ideation or behavior in outpatient and emergency settings. Only about one-third of these patients had documented firearm or medication assessments, indicating that lethal means assessment is infrequently recorded in this nonacademic health system and that NLP can help monitor this underused prevention practice until more structured EHR documentation is adopted.

This brief commentary in Psychiatric Services argues that commonly used depression outcome metrics (like response and remission thresholds) are more nuanced than they appear, and that contextual factors and measurement choices can substantially affect how treatment outcomes are interpreted.

This randomized trial in Kaiser Permanente Colorado tested Mindful Mood Balance (MMB), an 8‑session web-based mindfulness‑based cognitive therapy program, added to usual depression care for adults with residual depressive symptoms. Compared with usual care alone, MMB plus usual care led to greater reductions in depressive symptoms, higher remission rates, lower relapse risk over 15 months, less anxiety, and better mental functioning, suggesting a scalable option to manage lingering depression.

Among adults in outpatient mental health care, high‑level alcohol consumption and daily or near‑daily heavy drinking episodes on the AUDIT‑C were associated with substantially higher odds of suicide attempt within 90 days. The study supports using routine alcohol‑use screening results to help flag patients who might benefit from intensified suicide‑prevention efforts.

Surveying pediatric primary care physicians and clinic leaders in two health systems, this study finds high acceptability and fairly common but inconsistent use of firearm screening and safe‑storage counseling. Providing firearm locks was viewed more neutrally and rarely done, indicating that implementation efforts should focus on normalizing routine screening and counseling while addressing practical and attitudinal barriers to distributing locking devices.

A Monte Carlo simulation using national data estimated how many youth firearm suicides and unintentional deaths could be prevented if more families stored all guns locked. Even modest improvements in safe storage were associated with meaningful reductions in youth firearm mortality, suggesting that widely implemented safe‑storage interventions could avert a substantial share of preventable deaths.

The study examined whether following benzodiazepine prescribing guidelines in patients with anxiety or sleep disorders is associated with suicide risk. It found that, among patients with anxiety disorders, guideline‑concordant benzodiazepine use—short to moderate duration and combined with antidepressants or psychotherapy—was associated with lower odds of suicide, while a similar pattern in sleep disorders did not reach statistical significance.

This paper describes the development of NowMattersNow.org, a free, video‑based website that teaches dialectical behavior therapy skills to help people manage suicidal thoughts and intense emotions. In surveys from thousands of users, including many who arrived with completely overwhelming suicidal thoughts, visitors reported small but significant reductions in the intensity of suicidal ideation and negative emotions while on the site. These findings provide preliminary support for using NowMattersNow.org as a brief, scalable resource that clinicians and health systems can offer to patients for short‑term coping with suicidal crises.

Interviews with primary‑care patients who completed PHQ depression screening found that most viewed questions about suicidal thoughts as appropriate and valuable. Some, however, described a mismatch between their experiences and item 9 wording and weighed hopes for help against fears of stigma or consequences, with trusting, empathetic providers making disclosure easier.

This case-control study took place across eight health systems comparing medical, mental health, and demographic factors for people who died by firearm suicide vs other methods. They found specific general medical conditions that strongly increase suicide risk and supporting lethal means counseling.

This correspondence examines the sensitivity limitations of the PHQ-9 item 9—a widely used self-report measure of suicidal ideation—in the context of systematic suicide risk identification under the national Zero Suicide prevention framework. The authors highlight that over one-third of suicide attempts and deaths occur among patients who report thoughts of death or self-harm 'not at all,' raising concerns about sole reliance on this item for risk stratification. The piece calls for reevaluation of screening approaches to better capture at-risk patients who may not endorse suicidal ideation on standard questionnaires.

The study used data from 2,674 suicide decedents in eight health systems to examine firearm suicides among people with mental illness, substance use diagnoses, or prior suicide attempts. It found that most firearm suicides occurred among individuals without these recorded clinical risk factors, indicating that prevention efforts must extend beyond traditional psychiatric populations.

This paper presents the protocol for a multisite randomized controlled trial—conducted across Group Health Cooperative, HealthPartners, and Kaiser Permanente Colorado—designed to evaluate two population-based outreach programs to prevent suicide attempts among at-risk outpatients identified via routine PHQ-9 screening. One intervention arm offered Care Management with systematic risk assessment and EHR-integrated care pathways, while the other provided Skills Training in Dialectical Behavior Therapy delivered via secure patient portal messaging. The pragmatic design, including population-based enrollment and intent-to-treat analysis, reflects the real-world scale needed to meaningfully assess population-level suicide prevention strategies.

This cross-sectional study examined how often adult patients disclose firearm access when a single standardized question is added to a routine mental health questionnaire in primary care and outpatient mental health clinics. Among more than 128,000 patients, 83.4% in primary care and 91.8% in mental health answered the firearm question, and of those who responded, 20.9% and 15.3%, respectively, reported having firearm access. The authors conclude that most patients will answer a standard firearm access question, and that using such questions alongside counseling and decision-making resources could strengthen suicide prevention practice and guideline development.

This column describes how health systems in the Mental Health Research Network function as learning health care systems that integrate suicide prevention research into routine practice. It highlights specific implementation efforts and discusses practical challenges of conducting embedded suicide prevention research.