An Initial Exploration of the Daily Associations between Opioid Use and Acute Suicide Risk: The Role of Posttraumatic Stress Symptoms

Abstract

Objective:

Opioid use and suicide frequently co-occur. Proximal posttraumatic stress symptoms (PTSS) may lead to co-occurring opioid use and acute suicide risk. Information from intensive longitudinal methods is needed to inform targeted intervention and policy for opioid use and suicide.

Methods:

Participants were 53 trauma-exposed adults recruited from the community who use opioids and were at elevated risk of suicide (M_age = 45.2 years; 81% white; 72% disabled; 57% men; 38% without stable housing). Participants completed a baseline session and 14 consecutive days of ecological momentary assessment consisting of twice-daily surveys. Community-based participatory research (CBPR) methods centered individuals with lived experiences. Multilevel dynamic structural equation models (DSEM; n = 48) and group iterative multiple model estimation (GIMME; n = 19) frameworks were applied for subsets of the sample.

Results:

PTSS were not associated with next-survey suicidal thoughts; instead, suicidal thoughts were significantly associated with next-survey PTSS (Standardized Fixed Effect Estimate = 0.53, 95% CI[0.25, 0.89]). Within-person opioid use—but not opioid cravings—significantly moderated the association between PTSS and next-survey suicidal thoughts (Standardized Fixed Effect Estimate = 0.40, 95% CI[0.02, 0.79]). Taken together, DSEM and GIMME findings indicated substantial heterogeneity in the momentary associations among PTSS, opioid cravings, and suicidal thoughts across individuals.

Conclusion:

This study offers initial insight in the dynamic relations among PTSS, opioid use, and acute suicide risk. Importantly, findings call for future ecological momentary assessment work in this area to better inform clinical interventions for PTSS, opioid use, and suicide.

The ongoing opioid epidemic is among the worst public health crises in U.S. history. Individuals who use opioids exhibit poor health outcomes including diminished quality of life (De Maeyer et al., 2010), increased healthcare utilization (Sheehan et al., 2022), and more physical health comorbidities (Lake & Kennedy, 2016). Opioid use incurs a staggering national annual cost of more than $1 trillion dollars (Florence et al., 2021). Despite large-scale efforts to decrease the high prevalence and impact of opioid use (Blanco & Volkow, 2019), work is still urgently needed. Research that identifies factors underlying opioid-related consequences is key for preventing adverse outcomes, especially deaths.

There is a critical need for research that improves understanding of the co-occurrence between opioid use and suicide (Gordon & Volkow, 2019). Suicide remains a leading cause of premature death, with rates having risen more than 20% in the past decade despite intervention efforts (Hedegaard et al., 2018). Notably, individuals who use opioids are at three times greater risk of suicide compared to those who do not use opioids (Ali & Dubenitz, 2021). Relative to other substances, opioids are among the most strongly associated with suicide (Wilcox et al., 2004). In the year following a non-fatal opioid overdose, suicide becomes among the highest causes of death (Olfson et al., 2018). Despite the robust link between suicide and opioid-related deaths, there is a scarcity of empirical studies examining the temporality and processes underlying this association. It may be that opioid use results in neurobiological changes leading to dysphoria and myopia that increase risk-taking or impulsive behavior or numbing that lessens fear of death; each of these factors are associated with increased risk for suicide (Rizk et al., 2021). Studies on co-occurring opioid use and suicide almost exclusively rely on cross-sectional data, such as medical records (Bohnert & Ilgen, 2019). However, longitudinal research is needed that examines the specific nature and direction of the relation between opioid use and acute suicide risk. For example, it may be that opioid use and suicidality are causally linked to one another, or this co-occurrence may be better explained by other shared risk factors (Bohnert & Ilgen, 2019).

Posttraumatic stress symptoms (PTSS) may increase proximal risk for opioid use and suicide. PTSS are etiologically tied to trauma and include re-experiencing the trauma (e.g., flashbacks, nightmares), avoidance of trauma cues (e.g., memories, places), negative alterations in mood and cognitions (e.g., negative affect, anhedonia), and alterations in arousal and reactivity (e.g., hypervigilance, startle response; American Psychiatric Association, 2013). PTSS are linked to deleterious mental (Holowka & Marx, 2012) and physical (Pacella et al., 2013) health outcomes. Individuals with PTSS frequently engage in risky or self-destructive behaviors (Rheingold et al., 2004), including opioid use (Bilevicius et al., 2018) and suicidal behavior (Davis et al., 2014). Indeed, among those who use opioids, the prevalence of posttraumatic stress disorder (PTSD) ranges from 14–53% (Ecker & Hundt, 2018), which is higher than the prevalence of lifetime PTSS among the general population (8.3%; Kilpatrick et al., 2013). Among those who have attempted suicide, PTSS emerge as the only anxiety-related predictor (Sareen et al., 2005). Importantly, PTSS, opioid use, and suicide co-occur and are associated with worse outcomes, including more severe opioid use and psychosocial problems (Meshberg-Cohen et al., 2021). While PTSS has emerged as a distal risk factor for opioid use and suicide (Hassan et al., 2017; Stevens et al., 2013), there is a crucial need to identify PTSS as an immediate predictor of real-world opioid use and acute suicide risk. Consistent with the self-medication model (Darke, 2013), trauma-related coping may increase proximal risk for opioid use and contemporaneous suicide. In addition, the pathway from PTSS to suicide risk may be stronger during opioid use. However, this model remains empirically untested.

Ecological momentary assessment (EMA) methods are useful to inform real-time, naturalistic interventions aimed at PTSS, opioid use, and suicide. EMA entails intensive real-time repeated assessment (Hufford, 2007). Thus, EMA is an ideal method for capturing dynamic processes such as opioid use and suicide (Kowalczyk et al., 2015; Sedano-Capdevila et al., 2021), which may be more likely in specific contexts (Gossop et al., 2002; Kleiman et al., 2017), such as following PTSS. Further, EMA addresses limitations of cross-sectional and retrospective assessments, which are subject to cognitive biases (e.g., more intense or recent experiences) and aggregate symptoms (Brewin, 2014). Leveraging recent advances in personalized statistical models (idiographic) for EMA data is key for more translatable implications (e.g., individualized treatments; Hofmann & Hayes, 2019). Therefore, there is an urgent need to capture inter- and intra-individual differences in momentary relations among PTSS, opioid use, and acute suicide risk to inform our understanding of the temporal sequencing of these processes as well as timely and targeted interventions.

To better understand the associations among PTSS, opioid use, and suicide, the proposed research integrated a community-based participatory research (CBPR) framework among adults who have a history of trauma and report current opioid use and elevated suicide risk. CBPR is a collaborative approach that aims to facilitate authentic, equitable, and sustainable partnerships between researchers and communities (Collins et al., 2018). These methods strategically center the concerns of those directly impacted by research to enhance the effectiveness, acceptability, and feasibility of findings as well as to reduce community opposition and ensure lasting benefits to the community (Kwon et al., 2012). To explicate the unique role of trauma-context (e.g., PTSS) on proximal opioid use and acute suicide risk, we implemented EMA to identify the momentary relations between fluctuations in PTSS, opioid craving/use, and suicidal thoughts. We chose to assess suicidal thoughts as they are one of the strongest predictors of premature death by suicide (Ribeiro et al., 2016) and can be safely and feasibility studied using EMA (Ballard et al., 2021). We hypothesized that the momentary positive association from PTSS to suicidal thoughts would be stronger during moments of elevated opioid craving/use.

Method

Community Advisory Board

Community advisory boards (CABs) have been shown to be an effective CBPR strategy to guide research activities and create infrastructure for community members to voice their priorities that researchers may otherwise neglect (Chené et al., 2005). A CAB was formed of seven community members impacted by PTSS, opioid use, and/or suicide. The CAB met once with the research team in September 2023 and was consulted throughout the study period for recruitment strategies, evaluation of study materials, and solicitation of feedback on the study protocol. The CAB prioritized minimal participant burden and increased participant benefits.

Participants

Participants were recruited from Providence County, Rhode Island using community-based partnerships (e.g., collaboration with harm reduction agencies) and strategies (e.g., posting recruitment materials in community establishments, internet forums). A CONSORT diagram is presented in Figure 1. Eligibility was determined through a phone screen. Inclusion criteria included: (a) age 18 or older; (b) working knowledge of the English language; (c) past month non-prescription opioid use (i.e., at least one self-reported episode per week); (d) past month suicide risk (determined by any suicidal thoughts or behaviors via the Columbia-Suicide Severity Rating Scale; Posner et al., 2011); (e) lifetime exposure to a DSM-5 Criterion A traumatic event (determined by the Primary Care PTSD Screen for DSM-5; Prins et al., 2015); and (e) owned a smartphone. Exclusion criteria included: (a) mania/psychosis or cognitive impairment that interfered with study protocol; and (b) residence in a structured living community (which restricts opioid use). Demographic characteristics for the sample (N = 53) are presented in Table 1. The average age of the sample was 45.15 years (SD = 10.02). Of note, 43.4% identified as a woman, 30.2% identified as a person of color, 9.4% identified as a sexual minority, and 71.7% identified as disabled. A total of 37.7% reported unstable housing in the past 30 days.

Figure 1.

CONSORT Diagram

Table 1.

Demographic Characteristics of the Sample (N = 53)

	M (SD)	Range	n (%)
Age	45.15 (10.02)	27 – 64
Racial/Ethnic Background
Black or African American			7 (13.2%)
White			43 (81.1%)
American Indian or Alaska Native			4 (7.5%)
Hispanic or Latinx			3 (5.7%)
Middle Eastern or North African			1 (1.9%)
Not listeda			1 (1.9%)
Gender
Woman			23 (43.4%)
Man			30 (56.6%)
Sexual Orientation
Heterosexual			48 (90.4%)
Sexual minority			5 (9.4%)
Education
Grades 1–12, no diploma			7 (13.2%)
High school diploma or GED			19 (35.8%)
Some college			14 (26.4%)
Associates degree			9 (17.0%)
Bachelor’s degree			3 (5.7%)
Master’s degree			1 (1.9%)
Disability
Yes			38 (71.7%)
No			15 (28.3%)
Employment
Full time (35+ hours per week)			3 (5.7%)
Part time (<35 hours per week)			3 (5.7%)
Unemployed			34 (64.2%)
Not in labor force			13 (24.5%)
Past-month Stable Housing
No			20 (37.7%)
Yes			33 (62.3%)
Current Opioid Use Treatment
Medication			24 (45.3%)
Alcoholics/Narcotics Anonymous			16 (30.2%)
Therapy			11 (20.8%)
None			22 (41.5%)
Current Opioid Use Disorder			52 (98.1%)
Current PTSD			48 (90.6%)

^aOne participant indicated that their racial background was not listed and self-described as “Atlantic Islander.” PTSD = posttraumatic stress disorder

Procedure

All procedures were reviewed and approved by the University of Rhode Island Institutional Review Board. Data collection occurred from 2023 to 2024. The study entailed a baseline session, an EMA period, and for a subset of participants, an exit interview (not presented here). All participants completed a formal suicide risk assessment and personalized safety plan. Participants were provided a list of community resources and mood improvement activities, which were accessible through the EMA app. A licensed psychologist in the state of Rhode Island (author NHW) was available as needed for any additional trauma, substance, or suicide related support. During EMA, participants were notified that their data collected from the mobile app were not monitored in real-time. At the end of each survey, participants were reminded of their safety plan and emergency phone numbers. Participant EMA data were reviewed daily; outreach calls and suicide risk assessments were conducted as needed (i.e., endorsed suicidal behaviors during EMA).

Baseline Session

Sessions were conducted in a private office by advanced clinical psychology doctoral students. After providing written informed consent, participants completed study interviews, self-report measures on the computer, and EMA training. Sessions were one hour and participants were compensated $40 via cash.

EMA Period

Participants completed EMA surveys in the morning and evening for 14 days after the baseline session. Research supports EMA as an acceptable (e.g., safe, low attrition) approach for studying PTSS (Lane, Waters, et al., 2019), opioid use (Burgess-Hull & Epstein, 2021), and suicide (Ballard et al., 2021). Surveys took approximately one minute to complete. Participants were reminded by study staff to complete surveys if there was a pattern of non-responding. Participants were compensated $2 per completed EMA survey.

Measures

Baseline Measures

Demographic Information.

Participants self-reported on their age, gender, ethnicity, race, sexual orientation, income, educational level, employment status, housing instability, and disability status.

PTSS.

The Clinician Administered PTSD Scale for DSM-5 (CAPS-5; Weathers et al., 2018) is a semi-structured clinical interview of past-month DSM-5 criteria for PTSD (APA, 2013). It exhibits strong reliability and validity (Weathers et al., 2018).

Opioid Use Disorder.

The Structured Clinical Interview for DSM-5 Opioid Use Module (SCID-5; First, 2014) is a semi-structured clinical interview that assessed past-month DSM-5 diagnostic criteria for opioid use disorder. The SCID-5 demonstrates specificity, reliability (Osório et al., 2019), and concurrent and predictive utility for substance use disorders (Shankman et al., 2018). Participants also self-reported on their history of treatment for opioid use.

Suicidal Thoughts and Behaviors.

The Columbia-Suicide Severity Rating Scale (C-SSRS; Posner et al., 2008) is a semi-structured clinical interview that confirmed study eligibility and assessed past-month suicidal thoughts and behaviors. Prior research has established the utility and sound psychometrics of the C-SSRS across diverse samples (Posner et al., 2011).

Daily Measures

Posttraumatic Stress Symptoms.

An abbreviated four-item version (Price et al., 2016) of the Posttraumatic Stress Disorder Checklist (PCL-5; Weathers et al., 2013) was used to assess PTSS since the last survey based on the index trauma identified at baseline. Items assessed each diagnostic DSM-5 criterion for posttraumatic stress disorder (APA, 2013): intrusions (“Since the last survey was sent, how much were you bothered by: repeated, disturbing, and unwanted memories of the stressful experience?”); avoidance (“Since the last survey was sent, how much were you bothered by: avoiding external reminders of the stressful experience [for example, people, places, conversations, activities, objects, or situations]?”); negative alterations in cognition and mood (“Since the last survey was sent, how much were you bothered by: having strong negative beliefs about yourself, other people, or the world [for example, having thoughts such as: I am bad, there is something seriously wrong with me, no one can be trusted, the world is completely dangerous]?”); and alterations in arousal and reactivity (“Since the last survey was sent, how much were you bothered by: feeling jumpy or easily startled?”). Participants rated the extent to which they were bothered by each symptom using a five-point Likert scale (0 = not at all; 4 = extremely). Items were summed for a total score, with higher scores indicating elevated PTSS. This four-item PCL-5 is a psychometrically valid measure and highly corresponds to the CAPS-5 (Geier et al., 2020; Price et al., 2016). Multilevel reliabilities (Geldhof et al., 2014) in the current sample were good (within-person ω = 0.75, between-person ω = 0.96).

Opioid Cravings and Use.

Participants indicated their strongest opioid cravings since the last survey (“Since the last survey was sent, rate the intensity of your strongest craving to use opioids”) using a Visual Analogue scale from zero (none) to 100 (extreme). Participants also indicated opioid use since the last survey (“Since the last survey was sent, which opioids have you used? [check all that apply]: Heroin, Fentanyl/Carfentanil, Non-Prescribed Methadone/Buprenorphine, Other Non-Prescribed Opioids [e.g., Percocet, Oxy], None”). To facilitate analyses, opioid use was recoded as dichotomous to reflect whether participants used (1) or did not use (0) opioids.

Suicidal Thoughts.

Suicidal thoughts were assessed via a modified version of the C-SSRS (Posner et al., 2011) for EMA. Dichotomous items assessed passive suicidal thoughts (“Since your last survey, have you wished you were dead or wished you could go to sleep and not wake up?”); active suicidal thoughts (“Since your last survey, have you actually had any thoughts of killing yourself?”); and suicidal intent (“Since your last survey, have you had these thoughts and had some intention of acting on them?”). Items were summed, with higher scores indicating greater suicide risk. EMA-assessed suicide risk detects 50% more instances of suicidal thoughts than retrospective reports (Gratch et al., 2021). Multilevel reliabilities (Geldhof et al., 2014) in this sample were acceptable (within-person ω = 0.62, between-person ω = 0.72).

Alcohol and Other Substance Use.

Participants indicated the number of alcoholic drinks consumed since the last survey (“Since your last survey, how many standard alcoholic drinks did you have?”). Participants also indicated other substance use since the last survey (“Since your last survey, did you use any other drugs beyond opioids, such as cannabis, cocaine, or methamphetamine?”), such that participants reported whether they used (1) or did not use (0) substances other than opioids.

Data Analytic Plan

We first used a top-down approach to understand time-based dynamic relations. A multilevel dynamic structural equation modeling (DSEM; Asparouhov et al., 2018; Asparouhov & Muthén, 2020) framework was applied using MPlus Version 8.1. Recent advances in statistical methods and analytic software have led to innovative approaches including DSEM, which allows concurrent testing of both within- and between-person effects over time. Repeated measures (i.e., PTSS, suicidal thoughts, opioid use and cravings) were nested within-person. DSEM provides estimates of dynamic associations and average levels of PTSS, suicidal thoughts, and opioid cravings/use across individuals, allowing us to model sample heterogeneity. Therefore, autoregressive effects, cross-lagged effects, and within-person residual variances were modeled as random effects to capture their variations across individuals. Autoregressive effects indicated how quickly an individual returned to their within-person average levels, with values further from zero reflecting greater length of time. Cross-lagged effects indicated the effect of a variable at the previous time point on another variable at the current timepoint. The residual covariance indicated fluctuations in contemporaneous (same survey) effects among variables (e.g., PTSS, suicidal thoughts). Significant variances for effects indicate meaningful differences in the strength, direction, or presence of the effect across individuals. Statistical significance was determined using 95% credibility intervals (CIs) that did not include zero (Asparouhov et al., 2018). Successful model convergence was determined by the Potential Scale Reduction (PSR) values approaching one. Consistent with recommendations for Bayesian methods for small sample sizes, models were re-run with more than twice the iterations that were necessary for model convergence to ensure stable model convergence (Asparouhov & Muthén, 2021). Missing data in DSEM, which utilizes Bayesian estimation based on Markov Chain Monte Carlo, were well-handled under the assumption of missing at random (Hamaker et al., 2018) and, for restructured data accounting for lagging, via a discrete time Kalman filter.

First, a model was fit to examine whether PTSS predicts suicidal thoughts on a momentary level. Next, the model was extended to examine the influence of opioid cravings and use on the dynamic PTSS-suicidal thoughts relation. The extension of DSEM to capture within-person moderation of within-person effects is a cutting-edge approach, allowing us to uniquely model whether fluctuations in opioid cravings/use interact with increases in PTSS to predict emerging suicidal thoughts (Speyer et al., 2024). Specifically, two separate models were fit for opioid use and cravings. Per current recommendations (Speyer et al., 2024), PTSS and opioid cravings/use were first extracted as factor scores representing Level 1 (within-person) residuals based on latent-mean centering; this allowed for the product term for PTSS and opioid cravings/use to be calculated. DSEM for dichotomous variables (i.e., opioid use) utilizes probit models (McNeish et al., 2024), which provide estimates in the metric of z-scores and assume an underlying standard normal distribution. Of note, we also explored whether fluctuations in alcohol or other substance use impacted the dynamic associations between PTSS and suicidal thoughts.

Next, given significant heterogeneity detected in DSEM (e.g., significant variance suggesting effects may have been significant for a subset of individuals), we used a cutting-edge “bottom-up” approach to identify typologies of person-specific longitudinal networks based on dynamic associations. A group iterative multiple model estimation (GIMME) approach was applied. GIMME techniques provide novel estimation of both individual- and group-level associations, facilitating the assessment of heterogeneity in network patterns between and within individuals (for a full description, see Gates & Molenaar, 2012). It is ideal in this study to use GIMME due to the high person-specific heterogeneity in suicide/opioid risk (Kaurin et al., 2022). GIMME is unique because it fits vector autoregressive models for each person, which incorporate both lagged and contemporaneous relations via unified structural equation models, rather than assuming the same model to fit all individuals (Beltz & Gates, 2017). This method seeks to identify subgroups of people based on similarity of their individually derived networks using a community detection algorithm (e.g., regression paths among variables detected in the majority of individuals). This method provides a directed network of contemporaneous (i.e., how variables are related at the same survey) and lagged (i.e., how variables predict one another across surveys) associations. Of note, the directed contemporaneous associations indicate that one variable explains variability in another variable within the same survey, though the reverse may not be true (Luo et al., 2023).

Using the gimme package in RStudio (Lane et al., 2024), we modeled longitudinal networks of specific PTSS (intrusions, avoidance, negative alterations in cognition and mood, alterations in arousal and reactivity), opioid cravings, and suicidal thoughts. Given that opioid use was coded as a dichotomous variable, we were unable to include opioid use in the networks. Subgroups of networks were estimated using a data-driven approach via an unsupervised subgrouping algorithm (S-GIMME) based on patterns of significant network paths across individuals (Lane, Gates, et al., 2019). Analyses utilized default settings, including freely estimated autoregressive paths, subgrouping based on the Walktrap algorithm, and Bonferroni correction during pruning of non-significant paths (Lane et al., 2024). The group-level criterion was set to 75% of the sample and the subgroup-level criterion was set to 50% of the sample. There was no evidence of significant linear time effects for study variables. GIMME uses full information maximum likelihood estimation for missing data.

To obtain model convergence, GIMME analyses excluded 21 individuals who demonstrated no variability for study variables during the course of EMA (i.e., reported PTSS/opioid cravings/suicidal thoughts that were consistent for each survey during EMA). However, the GIMME model showed less than ideal fit across individuals (CFI_mean=0.84, NNFI_mean=0.67, RMSEA_mean=0.16, SRMR_mean=0.22). Given the effects of missing data on GIMME for short time series (Beltz & Gates, 2017), we further excluded eight individuals who were missing more than 40% of EMA data. As such, the final analytic sample was reduced to 19 individuals. The analytic sample did not significantly differ from the 34 individuals who were part of the larger sample in terms of number of opioid use disorder symptoms endorsed on the SCID (p = 0.07), PTSS severity endorsed on the CAPS-5 (p = 0.53), history of suicidal behaviors (p = 0.18) or thoughts (p = 0.36) endorsed on the C-SSRS, or responses to the interview questions on opioid use and suicide (ps > 0.27).

Results

Preliminary Findings

Opioid Use and Alcohol and Other Substance Use

Most (94.3%; n = 50) participants met criteria for severe (6+ symptoms) opioid use disorder, followed by moderate (4–5 symptoms; 1.9%; n = 1), and mild (2–3 symptoms; 1.9%; n = 1) opioid use disorder; one participant (1.9%) did not meet criteria for opioid use disorder. In the past month, 38 (n = 71.7%) participants used fentanyl, 23 (n = 43.4%) participants used prescription opioids without a prescription, 5 (n = 9.4%) participants used methadone without a prescription, and 14 (n = 26.4%) participants used heroin. Most (58.5%) participants reported that they were currently receiving treatment for opioid use. During the EMA period, the most commonly used form of opioid was fentanyl (278 instances), followed by prescription opioids used as non-prescribed (213 instances), heroin (146 instances), and methadone used as non-prescribed (118 instances). In total, participants endorsed using opioids on 66.7% of surveys (755 instances). Additionally, during the EMA period, participants endorsed using alcohol on 18.8% of surveys (257 instances) and drinking an average of 1.1 drinks (SD = 2.3) since the last survey. Participants endorsed using substances other than opioids on 35.5% of surveys (477 instances).

Suicide Risk

Per study inclusion criteria, all participants endorsed some suicidal thoughts or behaviors in the past 30 days. Per the C-SSRS, over two-thirds of the sample (77.4%; n = 41) reported active thoughts of killing themselves in the past three months. Nearly all participants (94.3%; n = 50) endorsed active thoughts of killing themselves in their lifetime. Over half (64.2%; n = 34) reported a lifetime history of a suicide attempt, with an average of 3.7 (SD = 4.4) attempts; six (11.3%) participants reported one suicide attempt in the past three months. During the EMA period, there were 430 reported instances of suicidal thoughts: 205 instances of passive suicidal thoughts, 100 instances of active suicidal thoughts, and 38 instances of suicidal intent.

PTSS

Most (90.6%; n = 48) participants met criteria for posttraumatic stress disorder. The most common index trauma reported was sexual assault (37.7%; n = 20), followed by physical assault (17.0%; n = 9), sudden accidental death (15.1%; n = 8), assault with a weapon (7.5%; n = 4), combat or exposure to a war zone (3.8%; n = 2), sudden violent death (3.8%; n = 2), fire or explosion (3.8%; n = 2), transportation accident (3.8%; n = 2), other unwanted or uncomfortable sexual assault (1.9%; n = 1), a serious accident at work/home/recreational activity (1.9%; n = 1), life-threatening illness or injury (1.9%; n = 1), and serious injury/harm/death caused to someone else (1.9%; n = 1). During the EMA period, all participants (n = 48) reported PTSS on at least one survey. Participants endorsed an average of 3.9 (SD = 0.4) PTSS per survey.

EMA Compliance and Missing Data

A total of 48 participants completed the EMA portion of the study (See CONSORT diagram). The compliance rate for completed EMA surveys was 71.2%, such that participants (n = 48) completed an average of 19.9 (SD = 7.9) out of 28 surveys. Compliance rates were not associated with PTSS, suicidal thoughts, opioid cravings, or opioid use during EMA (ps > 0.12).

Primary Analyses

DSEM

Dynamic Associations between PTSS and Suicidal Thoughts.

The model showed stable convergence (PSR value = 1.05) after 400 iterations. The fixed and random effects are shown in Table 2. A significant positive cross-lagged effect was found for suicidal thoughts predicting next-survey PTSS (Standardized Fixed Effect Estimate = 0.53, 95% CI[0.25, 0.89]), though there was significant variation for this effect across individuals. The model suggested comparable results (Standardized Fixed Effect Estimate = 0.55, 95% CI[0.20, 0.90]) when the model was estimated using 2,000 iterations. There was no significant cross-lagged effect for PTSS predicting next-survey suicidal thoughts. However, a significant variance of the random effect suggests that for some individuals PTSS may have been associated with next-survey suicidal thoughts. There were significant positive fixed and random autoregressive effects for suicidal thoughts and PTSS, suggesting that both processes had a lingering nature from previous timepoints that also varied across individuals. The mean-level of the residual covariance was not significantly different from zero. However, the variance of the residual covariance was significant (i.e., random effect). This indicates there was substantial variability across people which may suggest that some individuals may have had a contemporaneous (same survey) association between PTSS and suicidal thoughts while others did not. Of note, we conducted separate exploratory models to assess the effects of alcohol use and other substance use on PTSS-suicidal thoughts; next-survey suicidal thoughts were not significantly influenced by the dynamic PTSS-alcohol use interaction (Standardized Fixed Effect Estimate = −0.03, 95% CI[-0.24, 0.20]) nor the dynamic PTSS-other substance use interaction (Standardized Fixed Effect Estimate = −0.10, 95% CI[-0.20, 0.01]).

Table 2.

DSEM Unstandardized Fixed and Random Effects

	Fixed effects (means)			Random effects (variances)
Parameter	Est	95% CI: Lower	95% CI: Upper	Est	95% CI: Lower	95% CI: Upper
PTSS intercept	4.97 *	3.66	6.26	20.24 *	10.98	37.68
Suicidal thoughts intercept	0.23 *	0.12	0.39	0.09 *	0.02	0.28
Autoregressive effects
PTSS (t-1) → PTSS (t)	0.22 *	0.07	0.34	0.13 *	0.05	0.27
Suicidal thoughts (t-1) → Suicidal thoughts (t)	0.16 *	0.04	0.27	0.10 *	0.04	0.22
Cross-lagged effects
PTSS (t-1) → Suicidal thoughts (t)	−0.002	−0.02	0.03	8.90 *	2.81	24.03
Suicidal thoughts (t-1) → PTSS (t)	1.53 *	0.77	2.90	0.003 *	0.001	0.01
PTSS residual variance	0.63 *	0.10	1.22	3.71 *	2.09	6.57
Suicidal thoughts residual variance	−2.68*	−3.56	−1.81	9.56 *	5.71	16.51
Residual covariance (Suicidal thoughts & PTSS)	0.02	−0.25	0.31	.002 *	.001	.004

PTSS = posttraumatic stress symptoms. Fixed effects indicate the average within-person associations across participants; random effects indicate how much these associations vary across participants.

^*Bolded indicates that the 95% credibility interval does not contain 0.

The between-person associations among random effects are shown in Table 3. Individuals who had greater variation in suicidal thoughts over time tended to have greater person-level means of suicidal thoughts and PTSS, as well as greater variation in PTSS over time. Individuals who tended to have suicidal thoughts predicting next-survey PTSS tended to have less suicidal thoughts and less variation in suicidal thoughts over time. Individuals who tended to have greater variation in PTSS over time tended to have PTSS predicting next-survey suicidal thoughts and PTSS that co-occurred with suicidal thoughts.

Table 3.

DSEM Standardized Between-person Correlations among Random Effects

Parameter	1	2	3	4	5	6	7	8	9
1. PTSS intercept	—
2. Suicidal thoughts intercept	0.28	—
3. PTSS (t-1) → PTSS (t)	−0.01	0.35	—
4. Suicidal thoughts (t-1) → Suicidal thoughts (t)	−0.01	0.41	0.25	—
5. PTSS (t-1) → Suicidal thoughts (t)	−0.26	0.43	0.10	0.09	—
6. Suicidal thoughts (t-1) → PTSS (t)	−0.30	−0.67*	−0.21	−0.35	0.07	—
7. PTSS residual variance	−0.01	0.32	0.23	−0.24	0.45 *	−0.002	—
8. Suicidal thoughts residual variance	0.37 *	0.86 *	0.36	0.30	0.24	−0.81*	0.42 *	—
9. Residual covariance (Suicidal thoughts & PTSS)	−0.12	0.32	0.20	−0.07	0.54	−0.01	0.77 *	0.39	—

PTSS = posttraumatic stress symptoms. Effects indicate how individual differences in average levels and within-person dynamics are related.

^*Bolded indicates that the 95% credibility interval does not contain 0.

Interaction of PTSS and Opioid Cravings on Suicidal Thoughts.

The model showed stable convergence (PSR value = 1.05) after 200 iterations. There was no significant influence of the dynamic PTSS-opioid cravings interaction on next-survey suicidal thoughts (Estimate = −0.001, 95% CI[-0.004, 0.003]; Standardized Fixed Effect Estimate = −0.08, 95% CI[-0.45, 0.31]). However, there was a significant variance of the random effect, suggesting that for some individuals, PTSS and opioid cravings may have interacted to predict next-survey suicidal thoughts. The model suggested comparable results (Standardized Fixed Effect Estimate = −0.06, 95% CI[-0.40, 0.30]) when the model was estimated using 2,000 iterations.

Interaction of PTSS and Opioid Use on Suicidal Thoughts.

The model showed stable convergence (PSR value = 1.05) after 600 iterations. There was a significant positive influence of the dynamic PTSS-opioid use interaction on next-survey suicidal thoughts (Estimate = 0.10, 95% CI[0.01, 0.20]; Standardized Fixed Effect Estimate = 0.40, 95% CI[0.02, 0.79]). There was significant variation for this effect across individuals. The model suggested comparable results (Standardized Fixed Effect Estimate = 0.41, 95% CI[0.01, 0.83]) when the model was estimated using 2,000 iterations.

GIMME

Group-level Model.

The GIMME model converged normally. Results for the group-level model across the total analytic sample (n = 19) are presented in Figure 2. Fit indices showed acceptable fit across individuals (CFI_mean=0.93, NNFI_mean=0.81, RMSEA_mean=0.09, SRMR_mean=0.11). Example networks for selected study participants who demonstrated good model fit indices are presented in Figure 3. Networks contained an average of 3.33 significant paths, excluding autoregressions. The number of significant paths ranged from one to eight. The autoregressions (i.e., how strongly a variable at the prior survey was associated with itself at the current survey) were significant for all variables; no other group-level paths were retained in the model (i.e., they were not significant for at least 75% of the sample).

Figure 2.

GIMME Networks for the Group-Level Model (n = 19)

This model presents the network based on path counts: black lines represent group-level paths; grey lines represent individual-level paths; and green lines represent sub-group level paths. Dashed lines indicated lagged relations; solid lines indicate contemporaneous relations. ReExp = posttraumatic stress re-experiencing symptoms; Avoid = posttraumatic stress avoidance symptoms; Hyper = posttraumatic stress hypervigilance symptoms; NACM = posttraumatic stress negative alterations in cognition and mood symptoms; Crav = cravings for opioids; and Suic = suicidal thoughts.

Figure 3.

GIMME Example Networks for Individual-Level Models

Example networks for four participants who demonstrated good model fit indices. Blue lines represent negative associations; red lines represent positive associations. Dashed lines indicated lagged relations; solid lines indicate contemporaneous relations. ReExp = posttraumatic stress re-experiencing symptoms; Avoid = posttraumatic stress avoidance symptoms; Hyper = posttraumatic stress hypervigilance symptoms; NACM = posttraumatic stress negative alterations in cognition and mood symptoms; Crav = cravings for opioids; and Suic = suicidal thoughts.

Results for lagged and contemporaneous paths are presented in Table 4. In terms of lagged paths, the most frequent lagged path was PTSS negative alterations in cognition and mood to suicidal thoughts, which occurred in 42% of networks (n = 8). The next most frequent lagged path was opioid cravings to PTSS re-experiencing, which occurred in 37% of networks (n = 7). Opioid cravings to PTSS hypervigilance, suicidal thoughts to PTSS avoidance, and suicidal thoughts to PTSS re-experiencing each occurred in 32% of networks (n = 6). In terms of contemporaneous paths, the most frequent directed contemporaneous path was opioid cravings to PTSS avoidance, which occurred in 42% of networks (n = 8). The directed contemporaneous path from suicidal thoughts to PTSS hypervigilance occurred in 26% of networks (n = 5).

Table 4.

GIMME Percentages and Path Counts for Lagged and Contemporaneous Network Paths in the Group-level Model (n = 19)

	Lagged Associations						Contemporaneous Associations
	Re-exp	Avoid	NACM	Hyper-vigilance	Opioid Cravings	Suicidal Thoughts	Re-exp	Avoid	NACM	Hyper-vigilance	Opioid Cravings	Suicidal Thoughts
Re-exp	100% (n=19)	16% (n=3)	21% (n=4)	11% (n=2)	37% (n=7)	32% (n=6)	—	11% (n=2)	26% (n=5)	5% (n=1)	11% (n=2)	16% (n=3)
Avoid	26% (n=5)	100% (n=19)	16% (n=3)	21% (n=4)	11% (n=2)	32% (n=6)	5% (n=1)	—	16% (n=3)	11% (n=2)	42% (n=8)	16% (n=3)
NACM	11% (n=2)	16% (n=3)	100% (n=19)	11% (n=2)	11% (n=2)	11% (n=2)	5% (n=1)	16% (n=3)	—	26% (n=5)	5% (n=1)	11% (n=2)
Hyper-vigilance	21% (n=4)	11% (n=2)	16% (n=3)	100% (n=19)	32% (n=6)	16% (n=3)	21% (n=4)	5% (n=1)	16% (n=3)	—	16% (n=3)	26% (n=5)
Opioid Cravings	26% (n=5)	11% (n=2)	11% (n=2)	16% (n=3)	100% (n=19)	26% (n=5)	21% (n=4)	5% (n=1)	21% (n=4)	16% (n=3)	—	16% (n=3)
Suicidal Thoughts	16% (n=3)	21% (n=4)	42% (n=8)	16% (n=3)	16% (n=3)	100% (n=19)	16% (n=3)	16% (n=3)	16% (n=3)	21% (n=4)	16% (n=3)	—

Re-exp = posttraumatic stress re-experiencing symptoms; Avoid = posttraumatic stress avoidance symptoms; Hypervigilance = posttraumatic stress hypervigilance symptoms; NACM posttraumatic stress negative alterations in cognition and mood symptoms.

Subgroup-level Models.

Results demonstrated two subgroups, with one subgroup emerging that had significant subgroup level paths (i.e., they were significant for at least 50% of the sample). Specifically, these significant subgroup level paths were for directed contemporaneous paths from PTSS re-experiencing to PTSS avoidance as well as PTSS avoidance to PTSS hypervigilance.

Discussion

This study advances our understanding of the dynamic, real-time associations among PTSS, opioid use, and acute suicide risk among community members with lived experiences of trauma. Findings suggest that EMA is a feasible approach among this population and offers key implications for future research. Through the application of novel longitudinal data analytic methods, our results offer preliminary insight into the nuances in the associations among PTSS, opioid use, and suicidal thoughts among a high-risk sample.

Contrary to our hypothesis, we did not find evidence that PTSS was associated with next-survey suicidal thoughts in the DSEM model. Instead, suicidal thoughts were significantly associated with next-survey PTSS. This finding is unexpected given prior literature detecting contributions of PTSS on subsequent suicidal thoughts (Sareen et al., 2005; Schafer et al., 2022). One possible explanation for this discrepancy is that there is a scarcity of extant research applying EMA methods to assess PTSS and suicidal thoughts, particularly among those who use opioids. One recent study found that day-level data may be insufficient to predict increases in suicidal thoughts based on PTSS (Zhu et al., 2022). Our findings suggest a need for a more fine-grained EMA approach to understanding bidirectional associations between PTSS and suicidal thoughts among those who use opioids. There may be factors (e.g., cognitive biases and rumination; Law & Tucker, 2018; Rogers et al., 2024) underlying suicidal thoughts that contribute to PTSS (also characterized by cognitive processing alterations; Buckley et al., 2000; Ehlers & Clark, 2000). Alternatively, there may be important contextual factors (e.g., stressors; behaviors such as avoidance) not considered in our model that help to explain how suicidal thoughts can exacerbate daily PTSS. Future research should explore these questions given that suicidal thoughts leading to PTSS is an unanticipated finding.

There was a significant interaction between PTSS and opioid use—but not opioid cravings—predicting next-survey suicidal thoughts. Opioid use may be followed by neurobiological effects such as dysphoria, diminished reward sensitivity, and cognitive impairments (Baldacchino et al., 2012; Rizk et al., 2021) that compound with PTSS to produce heightened vulnerability to suicidal thoughts. Short-term effects of opioid use, such as increases in stress and negative mood (Preston et al., 2018), may uniquely interact with various facets of PTSS that have been linked to suicidal thoughts, such as trauma-related shame (Ollivier et al., 2022) and cognitive distortions (Whiteman et al., 2019). Although opioid cravings are a momentary predictor of subsequent opioid use (Burgess-Hull & Epstein, 2021), our finding that PTSS and opioid cravings did not interact to predict suicidal thoughts suggests that trauma-related suicide risk may be more directly linked to opioid use rather than cravings alone. Larger scale studies applying EMA methods are needed to identify mechanisms linking co-occurring moments of PTSS and opioid use with acute suicide risk. This information could inform the development of just-in-time adaptive interventions (Nahum-Shani et al., 2018) that provide trauma-related coping skills to prevent opioid use, and in turn, mitigate acute suicide risk. Our findings integrate two emerging lines of research showing that the treatment of opioid use (Watts et al., 2022) and PTSS (Rozek et al., 2022), separately, reduces suicide risk. If replicated, our findings may indicate that concurrent PTSS and opioid use treatment (Flanagan et al., 2016) could improve suicide prevention outcomes.

Heterogeneity observed in our sample suggested the utility of an idiographic approach to better understand the associations among PTSS, opioid use, and suicide across individuals. Our preliminary findings yielded key information regarding the sample and application of GIMME to address our research question. Specifically, findings indicate substantial heterogeneity in PTSS, opioid use, and acute suicide risk associations across individuals in our sample. Nearly half (42%) of the analytic sample was characterized by negative alterations in cognition and mood predicting next-survey suicidal thoughts. Over a third of the analytic sample (37%) was characterized by opioid cravings predicting next-survey re-experiencing. Additionally, nearly half of the analytic sample was characterized by contemporaneous (same survey) effects of opioid cravings on avoidance. Taken together, these results highlight nuances in the bidirectional associations among distinct facets of PTSS and both suicidal thoughts and opioid cravings. The high degree of heterogeneity in network patterns aligns with the remarkable degree of heterogeneity in PTSS presentations following trauma (Galatzer-Levy & Bryant, 2013). Similarly, our results align with prior GIMME research detecting heterogeneity in suicide risk over time (Coppersmith et al., 2024). This study extends this work to opioid use, suggesting that pathways to risk for co-occurring opioid use and suicide following trauma cannot be assumed to apply to all individuals. Personalized treatments that accommodate this heterogeneity may be needed to improve clinical outcomes for opioid use and suicide prevention following trauma (e.g., personalized treatments; Hofmann & Hayes, 2019). The present study may propel further work on the intersections between PTSS, opioid use, and acute suicide risk that vary across individuals. For instance, one consideration of GIMME is the exclusion of individuals who demonstrate low variance in the study variables to obtain stable model convergence. This highlights the challenges of predicting opioid use and suicide risk among individuals who experience severe and chronic symptoms as well as among those who do not have sufficient instances of symptoms to facilitate prediction, and thus the need for idiographic approaches to understand experiences of PTSS, opioid craving, and suicidal thoughts among these individuals.

Results from the current study provide novel initial evidence to suggest important associations among PTSS, opioid use, and acute suicide risk. However, it is necessary to consider findings within the context of study limitations. First, given the preliminary nature of our study, we had a small sample size and findings should be interpreted with caution. PTSS and suicidal thoughts may be fluctuating rapidly over shorter time scales than our twice daily EMA surveys, limiting conclusions about the nature and direction of momentary effects. Our preliminary results support the urgent need to replicate this study using a larger sample size with more frequent sampling to better capture momentary dynamics in PTSS, opioid use, and acute suicide risk. Second, we were unable to assess self-injurious behaviors using EMA due to the limited occurrences. There may be important nuances differentiating the function and impact of thinking about suicide relative to engaging in self-injurious behaviors following PTSS/opioid use, consistent with ideation-to-action frameworks for suicide (Klonsky et al., 2018). Future research in this area among higher-risk clinical samples (e.g., following psychiatric hospitalization discharge) and utilizing longer study periods (e.g., 30 days) to capture self-injurious behavior (e.g., preparatory behaviors, non-suicidal self-harm) is warranted. Third, a strength of this study is the sample, which represents a vulnerable and understudied population of individuals using opioids and experiencing PTSS and elevated suicide risk. However, it is unclear how findings may generalize to individuals who use opioids but do not also experience PTSS/suicidal thoughts as well as to other populations (e.g., individuals who use other forms of substances). Future research should replicate this study among more diverse samples, including racial/ethnic diversity and other trauma-exposed populations (e.g., Veterans). Fourth, there may be factors not captured by our EMA design that help to explain how suicidal thoughts contribute to PTSS, such as behavioral responses to suicidal thinking that in turn exacerbate PTSS (e.g., self-isolation; Vlachos et al., 2020). Findings call for future EMA work to explore mechanisms driving potential bidirectional effects among PTSS and suicidal thoughts.

Despite these limitations, initial findings from the present study underscore the need to examine the momentary associations among PTSS, opioid use, and acute suicide risk in tandem. In particular, opioid use in the context of heightened PTSS may confer elevated suicide risk. Importantly, findings suggest substantial heterogeneity in these associations, such that pathways to risk for PTSS, opioid use, and suicide vary across individuals. Tailoring trauma-informed assessment and interventions to the individual may improve opioid use and suicide outcomes.

Public Health Significance Statement.

This study offers initial evidence that co-occurring posttraumatic stress symptoms and opioid use may lead to heightened risk for suicide. Findings highlight a need to consider how the pathways among posttraumatic stress symptoms, opioid use, and suicide risk vary across individuals to better inform clinical interventions.