Webinar: Timely and Adaptive Strategies to Optimize Suicide Prevention Interventions
Date
Location
Overview
Therapeutic interventions are now being applied in ways that may enhance timely and tailored strategies to address suicide risk. This approach requires ready access to end-user data that informs proactive intervention strategies. Research strategies must consider 1) which data sources are best equipped to inform intervention strategies, 2) which intervention aspects should involve tailoring strategies vs. simpler approaches that may be just as effective, and 3) how such approaches can be incorporated into providers’ workflows within a health care setting.
This webinar brought together experts in areas that include passive and active data collection methods, the measurement of social contexts related to suicide risk, and the methodological aspects of Just-in-Time Adaptive Interventions (JITAI). The goal is to provide insights into developing and testing JITAIs for suicide prevention and to offer guidance on incorporating these adaptive interventions into healthcare systems. This webinar provided valuable knowledge on advancing and implementing effective suicide prevention strategies.
Recording
Sponsored by
National Institute of Mental Health, Division of Services and Intervention Research
Meeting Summary
Introduction
Stephen O’Connor, Ph.D., Chief, Suicide Prevention Research Program, NIMH
Dr. Stephen O’Connor noted that suicide prevention research is a NIMH priority. Proven strategies to reduce new incidents of suicide attempts and intensity of suicidal thinking, coupled with recent technological advances in understanding the nature and experience of suicide intensity, have led to opportunities to advance suicide prevention efforts. This webinar aimed to explore how to deliver timely and individualized suicide prevention strategies and to better understand how and why interventions work in order to enhance their effectiveness in real world settings.
Overview of Just in Time Adaptive Interventions
Kate Bentley, Ph.D., Mass General Hospital/Harvard Medical School
Dr. Kate Bentley described how traditional interventions often fail to address the rapidly fluctuating nature of suicidal thoughts and behaviors. Specifically, existing interventions lack timely delivery, personalization, and accessibility—underscoring the need for novel approaches that can adapt to an individual’s changing risk level in real time. Just-in-time adaptive interventions (JITAIs) are a promising approach to these challenges, as they are designed to provide the right amount and type of support at the right times. JITAIs can collect data in real-time and be adapted based on active measurement like ecological momentary assessment (EMA) or passive sensor data, providing appropriate support when an individual is most in need and receptive. Dr. Bentley introduced the use of micro-randomized trials (MRTs) to evaluate and optimize JITAIs. MRTs involve sequential randomization of interventions to maximize within-person statistical power and allow investigators to determine which strategies are most effective across different contexts. She emphasized that JITAI is just one model among adaptive interventions, and MRT is just one research design option.
Dr. Bentley shared findings from a pilot MRT involving 87 adults recently discharged from psychiatric inpatient units due to suicide risk, all of whom had developed a safety plan while in the hospital. Participants received EMA prompts six times a day for 28 days, reporting their suicide urge and intent. Those experiencing elevated surge or intent were randomly assigned to either a brief intervention of suggested coping strategies (personalized to match the safety plan, or general in nature) or no intervention. The research team found the intervention was associated with small but significant positive effects on in-the-moment coping. Additionally, they found that personalized messages were more effective than general messages. Dr. Bentley and her research team aim to build on these findings with a larger MRT.
Dr. Bentley concluded with challenges for the field: 1) to understand the optimal data streams, tailored variables, and timescales when using JITAI; 2) to identify the highest priority contexts and populations for JITAIs; 3) how to maximize engagement with JITAIs for suicide prevention; and 4) to determine how and when to integrate JITAIs into existing healthcare systems.
Passive and Active Data Collection Strategies
Randy Auerbach, Ph.D., Columbia University
Dr. Randy Auerbach provided an overview of the use of passive and active data collection strategies. Active assessment, such as EMA, captures what someone is thinking or feeling in the moment as compared to daily or weekly monitoring which captures longer time periods. Passive assessment, such as data collected from wearables or smart phones that measure things such as GPS location, accelerometry and keyboard inputs, capture behavioral phenomena outside of the lab with high temporal resolution—providing continuous data with less participant burden.
He reviewed a number of studies involving youth at high risk of suicide. In one study, his research team used EMA to detect changing levels of distress, finding a great deal of fluctuation within individuals, and that higher risk individuals experienced interpersonal contexts as more stressful than a control group. Another study monitoring daily and weekly mood and suicidal thoughts and behaviors showed that worsening in weekly aggregated mood ratings was associated with a threefold increase of clinically significant suicide ideation. His research team also explored whether keyboard inputs (passive assessment) could replace active assessment to reduce participant burden. They found that messages or emojis with more positive sentiment predicted more positive next-day mood, controlling for prior day mood. Additionally, increased use of personal pronouns (e.g., “I”) was associated with the onset of depressive episodes. Another study found that increased home stay was associated with a twofold greater likelihood of experiencing a subsequent suicide event.
Dr. Auerbach addressed the challenge of missingness, noting that sociodemographic factors and clinical acuity did not drive missing data in either active or passive assessments. Dr. Auerbach talked about a recent, promising effort to integrate active and passive data collection approaches into the mental health platform Vira to provide clinician dashboards and deliver nudges that reinforce clinical skills learned in therapy. He concluded by highlighting both the promise and challenges of active and passive assessment strategies to enhance suicide prevention efforts, such as the need to develop appropriate use cases, ensure data reliability, and develop effective interventions that engage an individual when in a distressed state.
Measuring Social Contexts Associated with Suicide Intensity
Leslie Adams, Ph.D., M.P.H., Johns Hopkins University
Dr. Leslie Adams talked about the environmental and societal factors—such as poverty, social isolation, community violence, racism, and discrimination—that influence mental health among Black communities. She reviewed data showing that residential segregation and ongoing poverty left Black Americans with the least desirable housing and lowest-resourced communities in the country. Additionally, unemployment and underemployment (lack of work that makes full use of skills and abilities) rates among Black men remained higher than their White counterparts. These social contexts had significant implications for mental health and suicide risk, including real-time effects.
Dr. Adams reviewed a pilot study that examined how social contexts and daily racialized stressors influenced moment-to-moment suicide risk among young adult Black men who recently received psychiatric care for self-injurious behavior, suicidal ideation, or suicide attempt. Active EMA data was collected from random smartphone prompts, end-of-day diaries, and exit interviews. Passive data was collected from GPS and accelerometer reports. These data were overlaid with a 2015 area deprivation index and the everyday discrimination scale. The primary measures of suicide risk were ratings of belongingness, sense of closeness, and a sense that others would be feeling happier “without me.” EMA data missingness was a challenge, particularly in areas with high deprivation.
The research team found that higher levels of area deprivation were associated with lower odds of feeling belongingness and greater odds of believing that people would be happier without them. Homeownership was associated with a greater sense of belongingness and a lower sense that people would be happier without them. Areas with higher proportions of non-Hispanic White residents were associated with lower feelings of closeness. Areas where people reported high everyday discrimination were associated with areas of high deprivation—demonstrating that deprivation and discrimination may interact to influence suicide risk.
Dr. Adams highlighted the importance of capturing social context in real time to inform JITAI approaches. In particular, she advocated for studies to attempt to capture high-stress moments, to be context-aware, and to have representation of marginalized communities. Smartphone interventions in particular could provide the daily context needed for scalable, low-cost JITAI. Although missingness and data quality were ongoing challenges, the use of social and spatial patterns in JITAIs showed promise for suicide prevention in marginalized communities.
Methodological Considerations for JITAI Studies
Inbal (Billie) Nahum-Shani, Ph.D., University of Michigan
Dr. Billie Nahum-Shani highlighted five common misconceptions that investigators need to consider when designing their JITAI studies, based on her work in substance abuse interventions.
The first was that JITAIs and adaptive interventions are the same. While both involve adaptation, they differed in terms of timeframe. Standard adaptive interventions adapt on a slow timescale of weeks or months, and typically involve human-delivered components such as coaching sessions. In contrast, JITAIs adapt on a fast timescale of seconds, minutes, hours, or days to address rapidly changing conditions. JITAIs use digital components, such as personalized messages, to deliver an intervention.
The second misconception was that a JITAI is an experimental design. Dr. Nahum-Shani said that a JITAI is an intervention design to provide guidelines on how to deliver an intervention. In contrast, an experimental design involved systematic manipulation and randomization to answer specific scientific questions. Intervention designs guide practical implementation, while experimental designs evaluate and optimize interventions.
The third misconception was that MRTs are always suitable for developing digital interventions. Dr. Nahum-Shani emphasized that the choice of experimental design should be driven by the specific scientific questions. For example, a sequential multiple assignment randomized trial (SMART) design might be appropriate if an investigator was uncertain about how to initiate an intervention or how to sequence intervention components. Conversely, an MRT may be appropriate if an investigator wanted to evaluate a digital JITAI and examine different strategies that are happening rapidly, including those that occur on a daily basis. A hybrid design combining both SMART and MRT would be appropriate when an intervention involved both human-delivered and digital components adapted over both slow and fast timescales.
The fourth misconception was that JITAIs should always intervene when the participant experienced a state of vulnerability. Dr. Nahum-Shani cautioned that individuals might not always be receptive to an intervention during their vulnerable moments and may even experience increased stress or cognitive overload. Whether to intervene or not when people are in a vulnerable state is an open scientific question.
Finally, the fifth misconception was that JITAIs always promote intervention engagement. She noted that engagement needed to be clearly defined. For example, while JITAIs often aim to promote engagement with a specific health behavior, there may be multiple, interconnected stimuli and tasks involved (like completing an EMA survey and also completing the suggested prompt), all of which need to be engaging.
Integrating JITAI into Healthcare Systems
Ewa Czyz, Ph.D., University of Michigan
Dr. Ewa Czyz reviewed three ongoing studies demonstrating how adaptive designs could optimize brief interventions for individuals at risk for suicide within healthcare settings. All three use a SMART or MRT design to inform adaptive interventions.
The first study is a SMART design focused on adolescents in a psychiatric hospital setting. All participants receive a standard safety planning intervention during their hospitalization. Post-discharge, they are randomized to either receive daily supportive text messages or standard care for four weeks. Their responses to daily electronic check-ins are assessed at the end of the first and second weeks, and these are used to determine who might need additional support. Non-responders are re-randomized to receive either booster phone calls with a counselor or asynchronous communication with a counselor via a web-based portal. The SMART trial tests four different embedded adaptive interventions that have differing combinations of post-discharge support while also considering practical implementation challenges such as integration into the clinical setting. For instance, automated text messages are tested first, with more resource-intensive interactions requiring staff time of counselors reserved for non-responders.
The second ongoing study is a randomized controlled trial (RCT) with an embedded MRT that targeted parents of adolescents who received emergency department services for suicide risk. Parents are randomized to either a six-week texting intervention or a control. The intervention group receives one of two intervention arms—one that only encourages parental engagement with adolescent suicide prevention activities (“adolescent-centered) and one that adds texts focused on parental wellbeing (“parent-centered). Dr. Czyz noted that this second component was added based on parent feedback, emphasizing the importance of engaging them in intervention development. To study how to best implement the parent-centered texts, an MRT was embedded within the RCT to test different timing and number of texts. The study aims to assess the feasibility, acceptability, and optimal timing and frequency of support.
The third ongoing study is a pilot RCT focused on adults at risk for suicide who were discharged from emergency departments. The intervention combines an electronic safety planning tool delivered in the emergency department, followed by a post-discharge text-based support program. Participants are randomized to received two daily text prompts focused on coping strategies and safety plan use, or to a control group. The study also includes an embedded MRT to determine if and when the text messages were most beneficial, as well as the optimal degree of message personalization.
Dr. Czyz emphasized that the research design needs to fit the scientific questions being asked, as well as the clinical population or setting. She highlighted practical challenges such as integrating technology into healthcare workflows and the need to collaborate with IT experts to navigate any limitations or approval processes. She also emphasized the importance of engaging stakeholders through the process to ensure that interventions were tailored and practical for real world settings.
Moderated Discussion
Moderator: Mary Rooney, NIMH
Discussants:
Matthew Nock, Ph.D., Harvard University
Evan Kleiman, Ph.D., Rutgers University
Nick Allen, Ph.D., University of Oregon
Walter Dempsey, Ph.D., University of Michigan
Question: What are the hurdles the field needs to overcome in order to embed JITAI into real-world practice and community settings? What can investigators do to proactively address these hurdles when designing their studies?
Answers: Dr. Matthew Nock said that traditional suicide prevention focused on weekly or monthly meetings with people, but suicide risk does not wait for those timelines. With more information about the in-between times, there is now an ability to intervene across a range of settings and contexts. NIMH has been funding the right studies, but more are needed. Larger datasets are also needed because suicide outcomes are rare.
Dr. Nick Allen talked about the term ubiquitous liability, which refers to the responsibilities involved in responding to someone at risk when one has been collecting longitudinal or continuous data on that person. There is a tension between wanting to know more about a person’s risk and being able to act on it. It is important for investigators to consider how to implement their interventions in ways that are not only acceptable to patients, but also practical for clinicians.
Dr. Evan Kleiman added that there was always a concern about the cost of an intervention and that it was important for investigators to think ahead about whether it could be implemented in a no-cost or revenue-generating way.
Dr. Walter Dempsey said that JITAIs are effective, but often targeted to a particular population and might not translate broadly. More work is therefore needed on how to adapt JITAIs. Additionally, investigators who use machine learning algorithms should consider the robustness and sensitivity of their algorithms and ensure there are no unintentional inequities baked in.
Question: When is the right time to engage? When a person is vulnerable and in crisis or earlier, when they might be more receptive?
Answers: Dr. Nock said there was an abundance of human learning research showing that people needed to practice skills before they are in a high-stress situation.
Dr. Dempsey added that JITAIs are often thought of as a specific intervention but should instead have multiple intervention components to address different states of vulnerability or receptivity.
Question: How does one sustain engagement with JITAIs?
Answers: Dr. Allen said that a human touch somewhere in the intervention can dramatically increase engagement. There may be a way to scale interventions with chatbots, now that large language models have made them more sophisticated.
Dr. Nahum-Shani added that many investigators pay participants for completing an assessment, which increases engagement, but becomes costly and less scalable in the real world.
Dr. Auerbach noted the challenge of keeping pace with technological advances so that an intervention is not dated before deployment.
Dr. Kleiman added that this required moving research quickly into practice, having a faster regulatory process, improving the user interface, and educating the consumers about the technology.
Dr. Nock emphasized that the suicide prevention technology should not only be engaging but also scientifically sound.
Dr. Dempsey added that the iterative process involved in developing and optimizing an intervention is very slow, particularly with small populations.
Question: What are the challenges involved in making a truly personalized intervention?
Answers: Dr. Nock said there was a statistical challenge of building a person-specific model, particularly when the field is not yet good at identifying high-risk individuals. There are several layers of complexity involved in determining which interventions are right for an individual. Dr. Dempsey added that what might be right for an individual can change over time and determining within-person change is a major challenge.
Dr. Kleiman said that some data sources require a lot of computational power on a person’s smartphone, which was a computer science and engineering problem.
Dr. Allen noted that privacy and security were also important considerations—considering whether to store data on the cloud or on devices, for instance, which requires additional computational power. He noted that who is at risk at the individual level is a fundamental clinical question but that now we have new data to bring to the question.