US suicide rates have risen steadily in the past decade [1], particularly in rural areas [2,3]. Risk of death by suicide is especially high in the months after discharge from inpatient psychiatric treatment [4,5], particularly for people hospitalized because of suicidal thoughts and behaviors (STB) [6]. Despite this well-documented risk, research has made little progress in improving suicide prediction over the past 50 years [7]. Suicide research has been stymied by a focus on between-person factors that differentiate those who have and have not attempted (or died by) suicide, but many of these risk factors are relatively static, such as gender and race; research has lagged in examining dynamic, within-person processes that contribute to shifting risk over time among individuals who are already known to be at highest risk [8].

Leading ideation to action theories have enumerated negative affective experiences thought to contribute to suicidal ideation (SI), such as low connection [9], pain [9,10], hopelessness [9,11], thwarted belongingness [12], burdensomeness [12], defeat [13], and entrapment [13]. Negative affective experiences have been associated with between-person differences in current and past SI [9,12,13], and there is some evidence that within-person fluctuations in negative affective experiences may proximally predict SI [14-16]. Risk factors that are associated with the transition from SI to suicide attempts (SA) have been far less well studied [17]. More broadly, almost no work has examined our hypothesized cognitive, behavioral, and physiological risk factors of SA in daily life. We describe our protocol for a novel, transtheoretical, comprehensive examination of near-term risk factors for SI and SA among adults leaving psychiatric inpatient care, using multimodal ambulatory assessment of affective, cognitive, and physiological processes in an intensive longitudinal design.

Shneidman [10] argued that suicide is fundamentally rooted in unbearable psychache, “hurt, anguish, soreness, aching, psychological pain.” Psychological pain is robustly associated with suicide, whether because of psychiatric disorders [18], stressful life events [19], or physical pain or illness [20,21]. Nonetheless, modern suicide theories differ in the relative importance of negative affective experiences as they relate to suicide. For example, the Interpersonal-Psychological Theory [12,22] focuses on interpersonal facets of pain, namely perceived burdensomeness and thwarted belongingness [23]. The Three-Step Theory formulated by Klonsky [9] takes a broader view of pain as a prerequisite for SI, positing connectedness as a protective factor against severe SI in the context of pain. Both the Three-Step Theory and the hopelessness model described by Beck [11] emphasize hopelessness regarding the likelihood of negative affective experiences improving over time as a key predictor of SI. The Integrated Motivational-Volitional model described by O’Connor [13] argues that SI is prompted by feelings of defeat (consistent with the view expressed by Baumeister [24] of suicide as escape from aversive self-awareness) and entrapment.

Although these models indicate multiple nuanced facets of negative affective experiences, it is unclear whether people experiencing suicidality differentiate these facets and whether they have differential predictive power. Recent network analyses suggest that many of these negative affective experiences cluster together among individuals [25]. However, to date, these studies have not conducted analyses of how these negative affective experiences covary over time within individuals or how they differentially predict outcomes. We aim to test the coherence of negative affective experiences to determine whether these are, in fact, differentially related to SI or whether they are more accurately conceptualized as indicators of the broader construct of emotional pain (eg, psychache).

Although a plethora of cross-sectional studies have shown associations between these negative affective experiences and SI [26-33], little work is available on negative affective experiences as prospective predictors of SI. Longitudinal work has primarily examined negative affective experiences as predictors of SI over months or years [34,35] rather than as predictors of within-person short-term changes in SI. Recent ecological momentary assessment (EMA) research has demonstrated that many negative affective experiences fluctuate markedly at the within-person level, as does SI [15,36-40]; however, these studies have been limited by small sample sizes [15,36], short durations [37,38], and use of analog (community) samples [39-41]. Furthermore, negative affective experiences tell us little about which people with suicidal thoughts will go on to attempt, or die by, suicide, a central issue across ideation-to-action models.

Research examining the broad domain of impulsivity in relation to STB has been mixed [41,42]. However, emotion-related impulsivity, the tendency to behave impulsively during high-arousal negative or positive affective states [43-46], shows clear associations with STB [47-55], nonsuicidal self-injury (NSSI) [56-60], and self-rated likelihood of future SA [51]. Furthermore, emotion-related impulsivity seems to amplify the association between other suicide risk factors, such as negative affective experiences, and SA [48]. Daily diary research further suggests that high emotion-related impulsivity is related to stronger within-person associations between negative affective experiences and NSSI urges [61], consistent with literature demonstrating that emotion-related impulsivity is characterized by problematic responses to negative affective experiences, rather than by greater affective intensity itself [44,45,62-66]. Emotion-related impulsivity is genetically driven and highly stable over time [67] and may thus serve as a trait-like marker of SA risk among those who experience SI.

Correlational research suggests that deficits in inhibitory control in the context of negative affective experiences are associated with SA and NSSI [47,60,62,68-70], especially among psychiatric populations [49]. Null effects for inhibitory control deficits related to SA have emerged when there was a lack of consideration of affective state [69-71]. That is, meta-analyses indicate that poor inhibitory control is tied to STB when people are tested while experiencing depressed mood or major depressive disorder or when negative stimuli are incorporated into the task [53,72]. This pattern of results highlights the need for integration of affective and cognitive domains jointly to consider STB risk [46,49,62,73].

Sleep disturbances are associated cross-sectionally with SI and SA [74-77], and sleep problems predict later SA in adolescents [78]. Most of this work has used self-report omnibus measures of sleep [75], with limited ability to identify specific sleep domains as they relate to STB or to disentangle effects of objective and subjective sleep disturbances. To our knowledge, only 2 studies have examined prospective associations between actigraphy-assessed sleep and subsequent SI: Littlewood et al [79] found that short sleep duration predicted within-person changes in SI across days, and Bernert et al [80] found that variability in sleep duration predicted later SI above and beyond depression. No published work has examined objective sleep indices as predictors of SA in daily life.

However, there is strong evidence from neuroimaging [81] and behavioral research using naturalistic [82] and experimental [83] paradigms that sleep disturbances diminish inhibitory control [84-87] and that sleep problems are associated with other self-harm behaviors that can occur impulsively, such as NSSI [88]. Furthermore, prior work has shown that sleep disturbances contribute to impulsivity among patients diagnosed with psychiatric disorders [88]. Thus, based on the limited literature, we predict that reduced total sleep time (compared with an individual’s mean sleep duration) will affect likelihood of SA in the context of SI through decays in inhibitory control.

The goal of this study is to integrate examination of affective, cognitive, and physiological risk factors for short-term changes in suicidal thinking and suicidal behaviors among adults known to be at high risk within an ideation-to-action framework. To that end, our hypotheses are as follows:

A visual depiction of these hypotheses is provided in Figure 1.

This is a single-group, repeated-measures (within-persons) observational study following adults with recent STB receiving inpatient psychiatric treatment (N=130). An ethnically diverse sample of participants will be recruited during inpatient psychiatric treatment in Lubbock, Texas, United States, based on past week STB. Eligible and interested participants will then complete the informed consent process, self-report surveys, clinical interviews, a computer task assessing cognitive inhibition, and a debriefing procedure. After discharge from inpatient care, participants will engage in a 4-week (28-day) EMA protocol that involves receipt of 7 signal-contingent self-report survey prompts per day between participants’ self-reported wake times and bedtimes. Participants will also be prompted to complete an abbreviated version of the cognitive inhibitory control task after each survey and asked to wear an actigraphy device for the 28-day follow-up period to assess objective indicators of sleep. Participants will be invited to complete a remote (phone or videoconference) follow-up interview assessing recent STB at 4 weeks and 8 weeks after discharge.

All study procedures have been approved by the Texas Tech University Institutional Review Board (IRB2020-713; October 27, 2020). The study procedures were further approved by the first recruitment site (Covenant Health) on March 15, 2021. The study procedures, hypotheses, and all relevant documents were preregistered on the Open Science Framework on August 25, 2021 [89].

Given variability in EMA adherence, we conducted power analyses for 100 participants, while planning to recruit 130 participants to account for those with inadequate compliance. Most study aims will be tested using dynamic structural equation modeling, as described herein. In a recent Monte Carlo simulation study [90], adequate power (>0.80) for modeling of a single process (eg, SI) was demonstrated for moderate effects where n=100 with as few as 50 repeated measures (t=50). Thus, this study (t=196, 7 prompts per day for 28 days) is well powered even with a compliance rate as low as 57.1% (112/196; 4 prompts per day for 28 days) to examine within-person processes (H1a, H1b, and H2b). To provide more conservative estimates of power, a series of simulations with 100 replications were conducted for a random-effects (multilevel) logistic outcome using R software (The R Foundation for Statistical Computing) [91]. The assumptions for these simulations were as follows: (1) n=100, (2) within-person effects for all predictors, (3) standard normal distributions for all predictors, and (4) an intercept variance of 0.3. On the basis of these models, the study is well powered (>0.8) to detect standardized within-person effects as low as 0.1 with compliance as low as 64%.

For H2a, we used G*Power (Heinrich Heine University) [92] to conservatively determine the minimum detectable effect predicting any SA over follow-up with n=100, Cronbach α=.05, and β=.8, controlling for a moderate effect of SI on SA (0.3). We assumed an event proportion of 10% for the outcome based on recent literature [93]. On this basis, the study is powered to detect an odds ratio >2.63 for the effect of baseline emotion-related impulsivity or inhibitory control on follow-up SA risk. By way of comparison, recent research has demonstrated significant predictors of follow-up SA risk with odds ratios from 6 [94] to 13 [93].

For H3a and H3b, the models are identical to those for H1a, H1b, and H2B, except that analyses are conducted at the day level (t=28) to examine the effects of sleep. Previous studies have shown significant effects for the impact of sleep quality on next day SI (β=–.11) [79]. Although examination of within-day fluctuations in inhibitory control is exploratory (H3c), prior work showing large differences in inhibitory control between self-harm and control groups (Cohen d=0.84) [70] allowed us to conservatively estimate how day-level inhibitory control predicts SI and SA. Thus, we repeated our simulations outlined for H1a using t=28 and single predictors for sleep (H3a and H3b) and inhibitory control (H3c) in separate models. Models converged across repetitions, indicating that with n=100, the study is well powered (>0.8) to find standardized regression effects as low as 0.15 (representing small effects) for the within-person regressions of SI and SA on day-level inhibitory control or sleep disturbance.

Although suicide research has proliferated in the past 50 years, this work has not moved the needle substantively on suicide prevention, primarily because of a focus on static, between-person risk factors (who is at risk) rather than modifiable, within-person risk factors (when or why is a particular person at risk). Major suicide theories have proposed key drivers of STB, but very little research has directly compared the relevance of constructs across theories or tested whether these risk factors are related to dynamic fluctuations in suicide risk. In addition, empirical research on hypothesized cognitive and physiological contributors to suicide risk has been limited by an inability to measure these phenomena as they occur and change in daily life.

This study aims to address limitations in the research on SI and behavior in several ways. First, by examining between- and within-person experiences of a multitude of conceptually derived negative affective experiences, the results will clarify whether these are, in fact, uniquely and differentially related to suicide outcomes or whether they are more appropriately construed as indicators of a single latent construct, such as psychache. The results will thus clarify the utility of existing theories in which fine-grained, nuanced distinctions among risk factors may (or may not) map onto experiences that people with STB themselves discriminate among in daily life. Second, by using baseline and ambulatory measures of sleep disturbance and impaired inhibitory control, the results will clarify whether these serve as more stable or trait-like indicators of risk or whether fluctuations in these experiences relate to changes in risk over time. The results will inform the development of interventions designed to modify these risk factors in their most relevant contexts to decrease the odds of suicide among those at high risk.

The primary methodological decisions in study design were with respect to the inclusion criteria, intensity of follow-up data collection methods, and level of monitoring and responsivity to indicators of suicide risk during follow-up.

First, we considered whether to limit the sample further (eg, individuals with recent SA) or whether to expand the inclusion criteria to include anyone receiving inpatient psychiatric care, regardless of recent STB. Individuals with a history of SA are at higher risk of later suicidal behavior than individuals with SI only [102], which would increase the odds of suicidal behavior during the follow-up period and plausibly increase statistical power. However, we felt that further restriction of the inclusion criteria would be likely to slow the speed of recruitment, a significant concern for a grant-funded study with a 2-year timeline. Recruiting a more heterogeneous sample will also allow us to examine the generalizability of predictors across a more diverse patient population. By contrast, expanding eligibility to all individuals receiving inpatient psychiatric care would likely reduce the odds of suicidal behavior over follow-up compared with individuals with recent STB [4].

Second, we considered the significant trade-offs between level of detail and frequency of follow-up assessments with the need to reduce participant burden (and, relatedly, the odds of attrition from the study). Prior research has shown that a greater number of surveys per day [103] and a greater number of items per survey [104] predict lower compliance with momentary assessment (EMA) research, which then affects the ability to adequately capture outcomes of interest (STB) during the EMA period. In addition, asking follow-up questions about the nuances of STB based upon yes responses during the EMA may increase the odds of participants choosing to deny these experiences that do occur to avoid additional follow-up questions, which take time and may be sensitive in nature. The duration and frequency of assessments chosen for this study were based on prior research suggesting that these parameters were feasible for the population of interest [105]. Finally, the 4- and 8-week follow-up interviews were added to the study protocol in response to grant reviewer feedback as a check against missed STB that were not captured during the EMA to increase statistical power for between-person models.

Finally, intensive longitudinal studies examining suicide risk and related constructs vary in the strategies used to monitor incoming data and to respond to indicators of elevated suicide risk [106]. These decisions often involve trade-offs of study investigator and staff time, guarantees of anonymity for participants, balancing participant safety with the risks involved in interventions by outside parties, avoidance of unintentional reinforcement of STB through study staff contact, and the goals of observational versus intervention-based research. Although we considered real-time monitoring of participants’ full range of EMA responses concerning suicide risk factors and contacting participants on the basis of these responses, we ultimately determined that this was scientifically unjustified because we do not know precisely what level of SI or other risk factors are related to imminent (proximal) risk for suicide. Follow-up with participants during the EMA protocol was limited to disclosures of suicidal behaviors only, following the aforementioned procedures.

However, we did make several choices across the study to promote truthful responding on suicide risk indicators by participants to ensure that they had adequate access to safety planning and other crisis resources and to avoid development of an intervention study when the study aims were exclusively observational. First, participants will be informed repeatedly at baseline that the study involves many assessments of STB and that many of these experiences will not lead to violations of their confidentiality. Participants will be informed that the confidentiality of information provided during the study will only be broken if the participant indicates imminent (within 1-2 days) desire, intent, and plans to attempt suicide and the research team member is not able to develop a safety plan with the participant to address this risk. Second, crisis hotline information and local, national, and web-based resources related to self-harm and suicide will be provided to participants at the baseline session, at which time all participants will be offered the opportunity to complete a suicide safety plan with the research team; if a safety plan is created, it will be given to the participant to keep and a copy will be retained by the research team. Furthermore, at the start of each EMA survey, participants will be offered live links to call or text 24/7 crisis hotlines, alongside a reminder that responses will not be monitored in real time; the same message will be displayed on the final page of each EMA survey. Each intervening page of questions will have a link at the bottom of the page that opens a full list of resources in a separate web browser if needed. Thus, participation in the research study will involve greater-than-usual exposure to, and reminders of, available crisis supports but will not involve routine additional intervention by the research team based on SI or related experiences.

As with all empirical research, this study includes several important limitations to consider. Several of these factors relate to the nature of the specific sample being recruited for the project. First, because of the nature of recruitment from an inpatient psychiatric treatment setting, the study sample is geographically restricted to adults from Lubbock, Texas, and the surrounding areas, and the results may differ from those obtained in larger metropolitan areas and in other parts of the United States, as well as internationally. Second, patients with manic or psychotic symptoms as well as those with cognitive impairment may be underrepresented in the sample, given the exclusion criteria related to the ability to provide informed consent to participate in the study. Third, the decision to base eligibility on recent STB will necessarily limit our ability to make inferences about postdischarge suicide risk among patients with no, or more distal, history of STB, who are at lessened, but nonzero, risk of postdischarge STB [4,5]. Fourth, although adequately powered, the desired sample size (N=130) will limit the ability to examine potential moderators of the study results across demographic and clinical characteristics (eg, race and ethnicity, sexual orientation, and mental health diagnoses).

Limitations also exist with respect to study methodology. Owing to concerns regarding the length and intensity of our baseline session, many study constructs will be assessed using self-report survey instruments, which can be subject to response biases and whose associations with each other may be attributable to shared method variance. Although several strategies have been implemented to improve completion rates of self-report, behavioral, and physiological data collection during the EMA protocol, we expect participants to vary in level of adherence to study procedures, which will also influence data quality and statistical power for planned analyses.

Despite these limitations, the study has numerous strengths. Data will be collected from a population known to be at exceptionally high risk for STB and for whom prior research has been limited by practical challenges involved in collecting data from inpatient psychiatric settings. This particular sample will be drawn from a high-need region where mental health care access is limited, facilitating understanding of risk factors for suicide in an underserved community where mobile interventions may be particularly critical to ameliorate suicide risk. Using a transtheoretical approach to understanding the transition from suicidal thoughts to behaviors, this study will facilitate direct comparison of key hypothesized risk factors as potential real-world contributors to suicide, moving forward both theoretical and empirical understanding of these phenomena. Furthermore, data collection will span a number of methodological approaches, including self-reports, interviewer ratings, behavioral measures, and ambulatory physiological readings, facilitating multi-trait, multi-method examination of STB risk. Finally, and most importantly, the intensive longitudinal design provides a novel opportunity to understand within-person changes in STB risk over time in contrast to prior work that has focused on between-person characteristics that relate to historical experiences of STB.

Numerous steps will be taken to ensure that data collected through this study are used in an effective manner to improve the scientific understanding of STB and the experiences of people with STB interacting with the mental health system. Upon beginning data collection, the study procedures and aims were registered on the Open Science Framework [89] to improve the transparency and reproducibility of the study analyses and findings. Raw (deidentified) data from this study will also be made available to other researchers through the National Institute of Mental Health Data Archive (refer to the Data Availability section). Research outputs generated from this project (eg, conference presentations and scientific manuscripts) will be shared broadly with other STB researchers as well as the general public through translational mechanisms (eg, posters and social media dissemination).

Data from this study will serve as the foundation of future projects to examine tested risk factors for STB in greater detail. First, our data on negative affective experiences as contributors to suicidal thinking will inform the design of future EMA tools tailored to those experiences most strongly predictive of within-person changes in SI. These data will also clarify which therapeutic interventions may be most effective for addressing SI; for instance, if thwarted interpersonal needs are identified as key proximal SI risk factors, interpersonally focused interventions for STB may be more effective than intrapersonally focused interventions. Second, should impaired cognitive control be identified as a key factor in the transition from SI to SA, future research could test whether cognitive interventions are effective in reducing STB risk in this population. Finally, future research may build upon our sleep physiology findings by developing and testing suicide-specific sleep treatments among groups known to be at high risk, such as adults leaving inpatient psychiatric care.