Dynamic bidirectional relationships between perceived stress and emotion regulation in emergency medical service clinicians

Abstract

Emergency medical services (EMS) clinicians are first responders who experience recurrent occupational stressors. Cross-sectional research suggests that higher self-regulation of emotions may be related to lower stress, especially in individuals with regular substance use. However, temporal dynamics are unclear. Our objective was to identify real-time dynamics between perceived stress and emotion regulation in EMS clinicians who regularly use substances. Participants were full-time EMS clinicians reporting alcohol and/or cannabis use ≥2x/week. Participants completed five daily ecological momentary assessments (EMAs) at semi-random times for 28 days. We used a continuous-time structural equation model with Bayesian estimation to identify dynamics between perceived stress and emotion regulation (both within-person centered and standardized). The 110 participants completed 12,234 EMAs (81.3% adherence). Higher perceived stress predicted lower future emotion regulation (standardized estimate = −0.68 [−1.05, −0.31]). Inversely, higher emotion regulation predicted lower future perceived stress (standardized estimate = −2.25 [−3.38, −1.15]). We identified bidirectional relationships between perceived stress and emotion regulation in the daily lives of EMS clinicians with regular substance use. While results may not be generalizable to EMS clinicians who do not regularly use substances, we identified emotion regulation as a future interventional target to reduce real-time stress in this highest-risk group.

Introduction

Emergency medical services (EMS) clinicians, including Emergency Medical Technicians (EMTs) and Paramedics, are licensed first responders who deliver advanced care to patients before arrival to the hospital¹. Their job includes initial evaluation of patients, identification of life threats, management of patient vital signs, and emergent delivery of evidence-based interventions². EMS clinicians play a pivotal role in improving long-term patient outcomes, including better survival rates, decreased hospital admissions, and reduced duration of hospitalizations^3,4. However, in this role, EMS clinicians are repeatedly placed in isolated and unpredictable situations with only an ambulance and limited equipment, personnel, and medical oversight to care for patients^5–8. Among all first responders, including career police officers and firefighters, EMS clinicians have been found to be one of the most vulnerable populations for chronic stress, mental health problems, and substance use disorders^7,9.

Stressors from major occupational events, such as the death of a patient, can lead to a stress response, while everyday minor stressors encountered during regular work-related duties among EMS clinicians can lead to an accumulation of stress responses^10,11. Stress accumulation is defined as experiencing repeated stressful responses across a short period without ample time for recovery, such as the potentially stressful events encountered across multiple EMS shifts¹². These stressors can include severe staffing shortages, medical errors, and conflicts with coworkers. Almost 70% of EMS clinicians report insufficient time to recover between experiencing stressful work-related events, which makes understanding the impact of these repeated stressful exposures even more critical¹³. Additionally, a recent meta-analysis found that routine stress exposures in first responders pose a greater risk to the development of probable post-traumatic stress disorder than larger disaster responses (e.g., earthquakes, explosions)¹⁴.

The Lazarus stress theory suggests that individuals first use cognitive appraisal strategies to determine if an event is perceived as threatening, and then, they must decide how to respond to that perceived stress¹⁵. While exposure to stressors is unavoidable in EMS clinicians, perceived stress could be effectively managed through the use of emotion regulation strategies^16–18. Emotion regulation has multiple theoretical definitions, so it was operationalized in this study as a person’s ability to adaptively control and express their emotional arousal, accept their emotions as valid, and understand that emotions are appropriate and normal¹⁹. Importantly, emotion regulation includes the ability to effectively manage emotional experiences in response to stressors²⁰. Among all self-regulation constructs (e.g., mindfulness, sensation seeking, perseverance), stress has the strongest relationship with emotional regulation across other populations of adults^21,22. During daily life, people are constantly required to self-regulate their emotions, which can occur either consciously or unconsciously²³. While some EMS clinicians report frequently using effective emotion regulatory strategies to adaptively cope with their unavoidable, chronic stress, the majority report struggling to cope effectively, which increases the risk of engaging in health-risk behaviors like substance use^18,24–28. In a national study of 3000 first responders, 34% reported using substances to manage stress, and 21% reported that they could not successfully decrease their use²⁹. Therefore, one-third of EMS clinicians may use substances.

Overall, the key relationship between higher stress and lower emotion regulation is often more pronounced in individuals who engage in health risk behaviors^27,30–37. There is a large body of literature suggesting that in individuals with regular substance use, stress might more negatively impact emotional control³⁸. Additionally, in a large systematic review and meta-analysis, individuals with substance use have higher difficulties in managing their emotions³⁹. In first responders with substance use, higher stress has been significantly associated with lapses in effective emotion regulation strategies^{24,26,28,40,41}. Unfortunately, first responders are more likely to engage in substance use compared to the general population of United States adults^42,43, as well as United States adults in other careers⁴⁴. Given the higher rates of substance use in first responders and the negative relationships between stress and emotion regulation in these populations, EMS clinicians with substance use behaviors may be particularly vulnerable^28,29,45. However, we cannot easily decrease daily job-related stressors in EMS clinicians. Therefore, we should identify key modifiable factors that have a significant relationship with stress, which we can then target to decrease substance use among high-risk EMS clinicians. One possible mechanism through which an individual can effectively manage stress is by emotion self-regulatory strategies^16–18.

Emotion regulation has traditionally been considered and studied as a stable characteristic (e.g., a person has either high or low emotion regulation). However, recent studies now suggest that emotion regulation is a dynamic construct that changes within-person across periods of time and situational contexts^26,46. Therefore, an individual’s current levels of emotion regulation at any point in time could be impacted by internal (e.g., perceived stress) and external (e.g., physical location) contexts, and internal contexts can be modificable. Similarly, stress has also been examined as a dynamic process with significant patterns of individual variability across multiple successive days and throughout a single day^47–49.

Since EMS clinicians experience a high rate of recurrent occupational stressors and have higher rates of substance use compared to the general United States population, this is an exemplary population to study the variability in individuals’ perceived stress and its effects on emotion regulation, as well as the impact of emotion regulation on perceived stress^2,6,7,9,11. Prior studies have identified bidirectional relations between stress and emotion regulation across time in other populations⁵⁰. However, studies of stress in first responders have only used cross-sectional surveys, which do not examine the impact of changes in perceived stress or emotion regulation over time for an individual person during real-world conditions. Cross-sectional studies do not generalize to these within-person relationships over time⁵¹. Therefore, this current study utilized ecological momentary assessments (EMAs), sometimes called ambulatory assessments, where participants completed short surveys delivered to their smartphones throughout the day⁵².

Overall, this current study was the first study to utilize multiple daily measures to test the dynamic relationships between perceived stress and emotion regulation in EMS clinicians with regular substance use. Our methods leveraged EMAs to accomplish our main objective to establish temporal relationships in real-life environments and use intensive longitudinal data analysis to test within-person relations among perceived stress and emotion regulation over time^53–56. Given that these key relationships are often more pronounced in individuals who engage in health risk behaviors^27,30–37, and especially substance use³⁸, our secondary exploratory objective was to examine if these longitudinal relationships are moderated by baseline substance use severity scores in EMS clinicians. The implications of experiencing higher chronic stress are often related to higher substance use severity (e.g., frequency, quantity).

Findings from this first-of-its-kind study will determine how within-person changes in perceived stress impact future emotion regulation strategies and how changes in emotion regulation impact future perceived stress over time in a large national sample of full-time EMS clinicians who report recent, regular substance use. It remains important to identify these relationships in this highest-risk group of clinicians. We hypothesize a bidirectional relationship where higher perceived stress will predict lower future emotion regulation, while higher emotion regulation will predict lower perceived stress. We also hypothesize that baseline substance use severity (e.g., more severe substance use) will moderate the strength of these key relationships.

Methods

Participants

This study received human subjects ethics review and approval from the Dartmouth College Committee for Protection of Human Subjects (IRB #00032966). All participants provided informed, written consent to participate in this research study. The inclusion of participants was limited to EMS clinicians in the United States (1) currently working full-time as an EMS clinician^57,58, (2) ≥18 years of age^57,58, and (3) employed in a 9−1−1 response system⁵⁷, (4) had a regular daytime shiftwork schedule (e.g., scheduled daytime shifts) for the duration of the study, (5) had a personal smartphone that only they use⁵⁷, and (6) reported regular use of cannabis ≥2/week^59–61 and/or alcohol ≥2/week^62,63. We excluded individuals who screened at the highest risk for alcohol use disorder (Alcohol Use Disorders Identification Test ≥20 points)⁶⁴ or cannabis use disorder (Cannabis Use Disorder Identification Test ≥13 points)⁶⁵.

We chose to limit enrollment to full-time working EMS clinicians to ensure ample shifts with possible stress exposures. We required working in 9−1−1 response systems (EMS clinicians who do emergency 911 ambulance calls and not solely community health or vaccination clinics) because these settings have higher exposures to occupational stressors, including caring for critically sick patients with limited tools, staffing, and oversight⁶⁶. Including only daytime shift workers ensured a consistent daytime EMA schedule for high EMA compliance and study power. Studies also suggest that daytime and nighttime EMS clinicians have different risks, which this study was not powered to detect^67,68. Given that EMS clinicians are at high risk for substance use as a stress-coping strategy, and stress more negatively impacts emotional regulation in other adult populations with substance use, our inclusion criteria ensured that we recruited a sample of EMS clinicians at high risk for regular substance use behaviors. We chose to screen for cannabis and alcohol since these are two of the most commonly used psychoactive substances in the United States, with at least 16% and 53% of adults reporting past 30-day use, respectively^69,70). Frequent use of alcohol and cannabis to cope with stress is associated with increased risk of dependence^71,72. This also allowed us to explore moderation by AUDIT and CUDIT scores. We excluded those at high risk for highly problematic use who likely need clinician-facing substance use treatment and would not be the target of our future digital health intervention for EMS clinicians at-risk for substance use as a stress-coping strategy.

Procedures

We utilized two national recruitment strategies. First, we partnered with both the National Registry of EMTs and the National Association of EMTs to sample their members. The National Registry of EMTs maintains the largest database of over 500,000 currently certified EMS clinicians across the United States⁷³, while the National Association of EMTs is the largest professional association of EMS clinicians in the United States⁷⁴. Together, these organizations disseminated a preliminary survey to 73,546 members through email from April to October 2024, allowing individuals to provide their email address for future contact about this current study. Our second strategy included directly emailing leaders from both State EMS offices and individual EMS agencies from across the United States (all 50 states were contacted) through publicly available databases and asking them to share a recruitment flyer and survey link with their EMS clinicians. For both of these recruitment methods, individuals were emailed study information and a link to the online informed consent and screening/baseline survey in Qualtrics Survey Software from February to May 2025.

If eligible and consented, participants received EMAs (short Qualtrics survey links) delivered through text message by SurveySignal⁷⁵ on the day following completion of their baseline survey. Over the next 28 days, participants automatically received 5 daily EMAs during semi-randomly scheduled times (7−10 a.m., 10 a.m.–1 p.m., 1–4 p.m., 4–7 p.m., 7–10 p.m.) and administered at least 1 h apart. Participants received a reminder through text message after 1 h if the last EMA was not completed. The EMAs expired before or upon receipt of the subsequent EMA, which ensured that only one EMA was completed at a time.

EMA completion rates were monitored daily, and participants who missed all 5 EMAs on any given day were contacted by text message and email to encourage participation. Participants received up to $160 in total for completing the EMAs, which were delivered weekly through emailed virtual Amazon gift card codes: $1/EMA completed ($140) and a $5 bonus/week ($20 total) for completing ≥75% of EMAs each week.

Measures

The baseline survey included sociodemographic characteristics of age, gender, race, ethnicity, education, and the occupational characteristics of certification type (e.g., EMT vs. Paramedic), job role (e.g., patient care provider, office administrator), number of years certified as an EMS clinician, service type (e.g., Fire/EMS agency, Hospital-based Service), and state of practice. Three measures assessed baseline mental health symptoms, including risk for PTSD (4-item Short-Form PTSD Checklist-5)^76–78, anxiety (Generalized Anxiety Disorder-7)^79–81, and depression (Patient Health Questionnaire-8)^80,82. Two measures assessed the severity of cannabis use (Cannabis Use Disorder Identification Test—CUDIT)⁶⁵ and alcohol use (Alcohol Use Disorders Identification Test—AUDIT)⁶⁴, with higher total scores indicating a higher severity.

In this current analysis, we examined the EMA items of shift status (“Are you currently on shift at your EMS job?”), the 4-item Perceived Stress Scale (PSS-4)^83–86 and the 3-item emotional regulation scale²⁰. Similar to other studies of momentary stress and emotion regulation, the items on perceived stress and emotion regulation within these EMAs were anchored to “Since the last survey” ^46,87. The PSS-4 was designed as a brief measure of stress perceptions, including 4 items scored from Never (0) to Very often (4)⁸³. The total score (0–16) was calculated by summing across all four items, with items 2 and 3 reverse-coded⁸³. Higher scores on the PSS-4 indicate higher levels of perceived stress. The psychometric properties are considered sound, and the measure has the strongest test-retest reliability and predictive validity over short time periods, like in this current EMA study. In the current sample of EMS clinicians, the nested internal reliability was excellent between participants (α = 0.90), and acceptable within participants (α = 0.68)⁸⁸.

The 3-item emotion regulation subscale of the larger momentary self-regulation scale measured the ability to adaptively control and express emotional arousal, accept emotions as valid, and understand that emotions are appropriate and normal on a 5-point Likert scale from (1) Not at all like you to (5) Extremely like you²⁰. The items were reverse-coded and then averaged to create a mean emotion regulation score, where higher scores indicated higher average emotion regulation²⁰. This scale was developed through an empirically driven, iterative data analytic and refinement process²⁰. The process utilized exploratory and confirmatory factor analysis, as well as item response theory²⁰. Importantly, this scale was specifically designed to measure momentary emotion regulation dynamically via EMAs in naturalistic settings²⁰. During development, the scale was found to have both intra- and interindividual variability in real-world settings and had strong predictive validity for substance use behaviors in other populations²⁰. Test-retest data from each item were also considered during the scale development process²⁰. After development, the scale has also been used to identify predictors of substance use in multiple other studies^46,89. Additionally, this scale has been administered in EMAs in prior studies examining the impact of emotion regulation across different health risk behaviors^20,46. In this current sample of EMS clinicians, the nested internal reliability was excellent between participants (α = 0.97) and acceptable within participants (α = 0.76)⁸⁸.

We chose to measure emotion regulation abilities (e.g., awareness, acceptance, tolerance, flexibility) versus the use of emotion regulation strategies (e.g., reappraisal, attention deployment, rumination) in this current study⁹⁰. It is generally accepted that emotion regulation abilities are a higher-order process that influences later emotion regulation strategies chosen by an individual in a given situation⁹⁰. Therefore, the acceptance of current emotions as “valid” and “normal” (e.g., abilities), for example, can subsequently increase the use of direct engagement with an emotional experience (e.g., a strategy)⁹⁰. Additionally, prior studies suggest that cognitive-behavioral and acceptance-based interventions may be most efficacious through increased emotion regulation abilities, specifically⁹¹, and that these emotion regulation abilities are more strongly related to substance use than emotion regulation strategies²². While this prompted us to measure emotion regulation abilities, it is important to note that the relationship between these two constructs is likely bidirectional⁹⁰.

Power analysis

We conducted a priori R simulations for power analysis of multi-level EMA studies (ema.powercurve function; EMAtools package). This method was designed for discrete-time multilevel models; however, we performed this analysis as an approximation since our continuous-time models might differ slightly. We determined that 110 participants would allow us to detect a medium (d = 0.50) lagged association between dependent and independent variables at a power of 0.80, assuming a conservative intraclass correlation coefficient (ICC = 0.50) and a conservative estimate of 60% total EMA compliance^92–94. Given strong effect sizes between stress and emotion regulation in prior longitudinal studies, even with up to 40% missingness, the sample size of 110 participants would be sufficient for modeling the within-person analyses^72,95–97.

Statistical analyses

To test the dynamic relationships between perceived stress (continuous score) and emotion regulation (continuous score), we utilized a continuous time structural equation model (CTSEM) using Bayesian estimation with Monte Carlo sampling⁹⁸. CTSEM is ideal as this technique accounts for the differences in time between EMAs that were delivered in this semi-random schedule⁹⁹. CTSEM was preferable in this study versus discrete time (“lagged”) analysis since CTSEM can model our essential theoretical assumption that perceived stress and emotion regulation are continuous processes that unfold over time. Importantly, CTSEM examines the strength of the regression based on the amount of time that has elapsed between the individual measurements⁹⁹.

Specifically, CTSEM provides information about how individual effects change over time, which is necessary for investigating whether individual fluctuations in one variable can predict itself or another variable over different periods of time. In this analysis, time was defined as the number of hours elapsed since receiving the first EMA. Additionally, the Bayesian approach naturally handles missing data by estimating it along with the model parameters, which accounted for unequal measurement occasions due to missing data^100,101. Therefore, CTSEM incorporated the effects from all available observation occasions. Before modeling these relationships, the values for perceived stress and emotion regulation were within-person centered to isolate within-person variance from between-person variance, and then within-person standardized to facilitate comparative interpretation and model convergence¹⁰¹.

Analyses were performed in RStudio version 2024.12.1 using the R package ctsem (v 3.6.0)¹⁰¹. The distributions for perceived stress and emotion regulation were examined graphically and determined to be approximately stable over time within and across random samples of participants (Fig. 2). Given the use of Bayesian estimation, the significance of path parameters was determined using a 95% credible interval, which suggests that there is a 95% probability that the true parameter lies within the interval¹⁰². Since perceived stress and emotion regulation can vary by current working status, the models accounted for on- and off-shift status at each EMA at the within-person level^11,46,103. We did not control for time of day (morning/evening) or weekend status, as EMS clinicians work atypical shift-work schedules and often work late into the evenings and regularly on weekends¹⁰⁴. We used non-informative priors, which are intended to be weakly informative and are typically used in psychological research¹⁰⁵.

Fig. 2. Trajectories of perceived stress and emotion regulation within-person over time.

Observed data points of perceived stress and emotion regulation for a random sample of five individual subjects over 28 days. Emotion regulation levels were high and more stable, while perceived stress levels were moderate with more variation over the entire duration of the study.

The Bayesian CTSEM model was configured to identify first-order temporal effects within (autoregressive) and between (cross-regressive) perceived stress and emotion regulation. The model was estimated using 4 chains, each with 2000 total iterations, of which 500 iterations were used as warm-up (i.e., burn-in). This resulted in 6000 post-warmup samples (1500 per chain) used for inference. This included within-subject variation in drift parameters (auto-regressive and cross-regressive effects) and manifest intercepts (“manifest means”), which account for the tendency of perceived stress and emotion regulation to return to a nonzero equilibrium. The continuous drift parameters are time-dependent. Regarding the auto-regressive effects, the magnitude of the effect indicates the strength of the downward pressure to decrease to baseline levels after an increase (positive effects) or the strength of the upward pressure to increase to baseline levels after a decrease (negative effects)¹⁰⁶. The interpretation of the cross-regressive effects is similar to the traditional “lagged” relationships in EMA studies¹⁰⁶.

To explore post hoc moderation of these intensive longitudinal relationships by participants’ baseline AUDIT and CUDIT, we first extracted each participant’s empirical Bayes estimates of the drift parameters (e.g., autoregressive and cross-lagged effects) using the fitted CTSEM model. This method resulted in one drift estimate per participant for each parameter of interest. We then conducted person-level moderation analyses by regressing each participant’s drift estimates (e.g., the cross-lag from emotion regulation to subsequent perceived stress) onto their baseline alcohol use severity (AUDIT), cannabis use severity (CUDIT), or both if participants drank alcohol, used cannabis, or used both substances, respectively. Given the model complexity, this two-stage approach allowed us to examine whether the strength or direction of the within-person continuous-time associations changed based on baseline substance use severity. Each participant contributed one drift estimate and one baseline moderator value per analysis, ensuring independence of observations of the between-person moderator. This exploratory “two-stage” approach is robust for intensive longitudinal data, preserves within-person estimation, and helps avoid pseudo-replication^107,108.

Results

Preliminary analyses

In total, 2174 individuals were screened for eligibility, and 118 (5.4%) were eligible (Fig. 1). There were a total of 15,776 EMAs administered to the 118 enrolled participants. Participants were included in the final analytic sample if they completed EMAs on more than 1 day during the study. In total, 110 participants met this criterion, while 5 participants did not complete any EMAs and 3 participants only completed EMAs on the first day (8 participants total). The final analytical sample included 12,234 completed EMAs (total adherence 81.3%) among 110 participants. The participants in the final analytical sample completed an average of 111 EMAs (Median = 127, SD = 37.6, Min = 5, Max = 140) among the total 140 disseminated EMAs per participant. Only 1.2% (n = 156) and 1.0% (n = 120) of the 12,234 submitted EMAs (M_duration = 1.6 min/EMA) were missing data on the perceived stress or emotion regulation items, respectively, suggesting that participants only skipped a very small number of these EMA items. Participants in the final analytical sample (n = 110) did not differ significantly from those excluded from the sample (n = 8) by age (t = −0.38, p = 0.70), gender (X² = 0.18, p = 0.91), minority status (X² = 0.12, p = 0.73), education (X² = 2.45, p = 0.78), certification (X² = 5.47, p = 0.14), type of agency employed (X² = 3.27, p = 0.67), EMS experience (t = −0.09, p = 0.93), primary job type (X² = 1.70, p = 0.79), GAD-7 (t = −0.73, p = 0.47), PHQ-8 (t = 0.15, p = 0.88), PTSD checklist (t = −0.27, p = 0.78), CUDIT (t = 0.27, p = 0.79), or AUDIT (t = 1.55, p = 0.12).

Fig. 1. Participant recruitment and enrollment flow diagram.

Study participants were screened for eligibility based on eight different inclusion and exclusion criteria. Participants then had to complete more than 1 day of data to be included in the final analytical sample.

Participant characteristics

Participant characteristics are included in Table 1. Participants in the final analytical sample were 38.0 years old on average (SD = 11.4, range = 20–68). Participants were predominantly White (88.2%) and Non-Hispanic (92.7%), with 19 identifying as Non-White/Hispanic (17.3%). Participants were majority men (64.6%) and had an associate’s degree or above (61.8%). These demographic frequencies were all consistent with previous national studies of EMS clinicians^109,110. In regard to occupational characteristics, participants were mostly paramedics (70.0%), had a primary role as a patient care provider (77.3%), and had an average of 14.6 years of experience (SD = 10.7, range = 1–50). Most participants worked for a combined Fire/EMS agency (44.6%). Geographically, participants were enrolled in the study from 33 different states.

Table 1.

Enrolled participant characteristics (N = 110)

Characteristic	Value
Age, years	38.0 (11.4)
Gender, n (%)
Woman	37 (33.6)
Man	71 (64.6)
Nonbinary	2 (1.8)
Race, n (%)
Asian	2 (1.8)
Black or African American	3 (2.7)
Native American or Native Alaskan	2 (1.8)
Native Hawaiian or Pacific Islander	1 (0.9)
White	97 (88.2)
More than one race	5 (4.6)
Ethnicity, n (%)
Hispanic or Latino	8 (7.3)
Not Hispanic or Latino	102 (92.7)
Non-White/Hispanic, n (%)a	19 (17.3)
Education, n (%)
High School Diploma or GED	6 (5.5)
Some College	36 (32.7)
Associate’s Degree	29 (26.4)
Bachelor’s Degree	31 (28.2)
Master’s Degree	6 (5.5)
Doctoral Degree	2 (1.8)
Certification, n (%)
EMT	21 (19.1)
Advanced EMT	12 (10.9)
Paramedic	77 (70.0)
EMS Experience, years	14.6 (10.7)
EMS agency, n (%)
Air Medical	4 (3.6)
Fire-based	49 (44.6)
Government, Non-fire	24 (21.8)
Hospital-based	12 (10.9)
Private	16 (14.6)
Other	5 (4.6)
Primary role, n (%)
Patient Care Provider	85 (77.3)
Administrator/Manager	8 (7.3)
Educator/Preceptor	4 (3.6)
First-line Supervisor	13 (11.8)

Categorical or dichotomous data are reported as n (%). Continuous data reported as mean (SD).

EMT Emergency Medical Technician.

^aMinority status is defined as race/ethnicity of Hispanic/Non-White individuals.

Mental health and substance use characteristics

Participants in this sample had moderate levels of anxiety^79–81 (M_anxiety = 8.7, SD = 5.1, range 0–21) and depression symptoms^80,82 (M_depression = 9.3, SD = 5.4, range = 0–24), since scores ≥10 points represent cut points for clinically significant anxiety^79–81 and depression^80,82 symptoms. Additionally, participants had mild PTSD symptoms (M_PTSD = 5.7, SD = 3.5, range = 0–15)^76–78. Regarding the severity of substance use, the participants also had moderate severity of cannabis⁶⁵ (M_cannabis = 3.7, SD = 4.0, range = 0–13) and alcohol⁶⁴ (M_alcohol = 6.8, SD = 4.6, range = 0–18) use. In this sample of EMS clinicians, 64 (58.2%) used cannabis at least monthly, 102 drank alcohol (92.7%) at least monthly, and 56 (50.9%) used both cannabis and drank alcohol at least monthly.

Descriptive EMA statistics

Participants were on shift during 33.9% (n = 4144) of the 12,234 total completed EMAs. The grand mean (e.g., the overall mean of all within-person means) of perceived stress was 5.65 (iSD = 0.47), and the grand mean of emotion regulation was 4.33 (iSD = 0.47) among participants during the study. Participants reported moderate perceived stress (range = 0–16) and high emotion regulation (range = 1–5), on average, over the full duration of the study (Fig. 2).

Initial model

The ICC was 0.68 (95% CI: 0.67, 0.70) for perceived stress and 0.65 (95% CI: 0.63, 0.67) for emotion regulation. Therefore, 32.0% of the variation in perceived stress and 35.0% of the variation in emotion regulation were attributed to individual differences within participants over time. Inversely, 68.0% of the variation in perceived stress and 65.0% of the variation in emotion regulation could be attributed to differences in trait-level characteristics between participants.

Continuous time structural equation model

Regarding continuous time auto-regressive effects, changes in emotion regulation [standardized estimate = −1.02, 95% CI (−1.59, −0.57)] persisted longer than changes in perceived stress [standardized estimate = −1.83, 95% CI (−2.77, −1.08)] (Table 2). Since auto-regressive effects were plausibly non-null, the negative parameters suggest that both return to null relationships over time¹¹¹. Through interpretation of these relationships graphically, both parameters were most positively related to their state in the past at shorter time intervals (approximately 1 h), but this relationship was found to be null over longer time intervals (Fig. 3). Regarding the continuous time cross-regressive effects, the estimates of perceived stress on emotion regulation were below zero and plausibly non-null [standardized estimate = −0.68, 95% CI (−1.05, −0.31)] (Table 2). This suggests that higher levels of perceived stress predicted lower future levels of emotion regulation. Inversely, the estimates of emotion regulation on perceived stress were also below zero and plausibly non-null [standardized estimate = −2.25, 95% CI (−3.38, −1.15)] (Table 2). This suggests that higher levels of emotion regulation also predicted lower future levels of perceived stress. The effect size of emotion regulation on perceived stress was larger than perceived stress on emotion regulation. Through examining these relationships graphically in Fig. 3, both parameters were most negatively related to the other at shorter time intervals (approximately 1–5 h), but this relationship was null over longer time intervals. The largest effect of emotion regulation on perceived stress peaked around 1 h in time (Fig. 3).

Table 2.

Continuous time parameter estimates of perceived stress and emotion regulation

Continuous drift parameter	Estimate	SD	[95% CI]
Auto-regressive effects
Perceived stress → Perceived stress	−1.83	0.43	[−2.77, −1.08]a
Emotion regulation → Emotion regulation	−1.02	0.27	[−1.59, −0.57]a
Cross-regressive effects
Perceived stress → Emotion regulation	−0.68	0.20	[−1.05, −0.31]a
Emotion regulation → Perceived stress	−2.25	0.57	[−3.38, −1.15]a

^aAll effects not including the value of zero in the 95% CI are considered to be significant. Continuous drift parameters include the time-dependent autoregressive and cross-regressive estimates from the drift matrix of the Bayesian continuous-time structural equation model. Values for perceived stress and emotion regulation were within-person centered to eliminate between-person variance, and within-person standardized to facilitate comparison of the effects.

Fig. 3. Continuous time auto-regressive and cross-regressive standardized effects of perceived stress and emotion regulation in emergency medical services clinicians.

Standardized auto-regressive effects of perceived stress and emotion regulation (on themselves) and cross-regressive effects of perceived stress and emotion regulation (on each other) from a Bayesian continuous time model. Data are plotted as a posterior median (solid line) with a 95% credibility interval (dotted lines). Time intervals are in hours; auto-regression of emotion regulation (red); auto-regression of perceived stress (purple); cross-regression of emotion regulation on perceived stress (green); cross-regression of perceived stress on emotion regulation (blue).

Exploratory moderation analyses

In the post-hoc exploratory moderation analysis, the within-person effects of perceived stress on emotion regulation were not significantly modulated by either baseline AUDIT scores [effect = −0.01, 95% CI (−0.01, 0.01)] among the 102 participants who endorsed drinking alcohol (Fig. 4A) or CUDIT scores [effect = 0.01, 95% CI (−0.01, 0.02)] among the 64 participants who endorsed using cannabis (Fig. 4B). Additionally, among the 56 participants who endorsed using both alcohol and cannabis in the prior month, the within-person effects of perceived stress on emotion regulation were not significantly modulated by either baseline AUDIT [effect = 0.00, 95% CI (−0.01, 0.01)] or CUDIT scores [effect = 0.00, 95% CI (−0.01, 0.02)]. The within-person effects of emotion regulation on perceived stress was also not significantly modulated by AUDIT [effect = −0.01, 95% CI (−0.01, 0.01)] (Fig. 4C), CUDIT [effect = 0.00, 95% CI (−0.01, 0.01)] (Fig. 4D) or both scores together [effect = 0.00, 95% CI (−0.01, 0.01)] among those who drank alcohol, used cannabis, or used both of these respectively.

Fig. 4. Post hoc moderation of the dynamic bidirectional relationships between perceived stress and emotion regulation over time by baseline alcohol and cannabis use severity scores.

The alcohol use sample (blue) included 102 participants who endorsed drinking alcohol and the cannabis use sample (red) included 64 participants who endorsed cannabis use. The null relationships between perceived stress on emotion regulation (A) and emotion regulation on perceived stress (B) are visualized in the alcohol use sample. The null relationships between perceived stress on emotion regulation (C) and emotion regulation on perceived stress (D) are visualized in the cannabis use sample. The solid lines represent the estimated regression lines and the gray shaded regions represent the 95% CI.

Discussion

In this first-of-its-kind study, we explored the dynamic, momentary relationships between perceived stress and emotion regulation among 110 full-time EMS clinicians who regularly engaged in substance use and were recruited from across the United States. The demographics of our sample were consistent with previous studies from the National EMS database, including women and minorities^109,110. Additionally, participants had high adherence to the repeated EMA measures. We utilized robust and novel Bayesian CTSEM to identify key momentary reciprocal relationships between perceived stress and emotion regulation in high-risk EMS clinicians. Since stress is an unavoidable occupational exposure among EMS clinicians, findings suggest that emotion regulation might be a key modifiable momentary target for a future health intervention aimed to increase emotion regulation to decrease stress levels^16–18.

First, we found that higher momentary perceived stress predicted lower momentary emotion regulation in these high-risk EMS clinicians. To date, most studies of stress and emotion regulation in other populations are cross-sectional, which cannot determine the temporal dynamics between these key factors. However, cross-sectional studies across populations found that higher levels of self-reported stress are associated with lower self-regulation, including worse impulsivity¹¹², failures with goal pursuit^113,114, and less attention¹¹⁵. Cross-sectional studies in first responders, specifically, including firefighters, police officers, and EMS clinicians, suggest that higher levels of self-reported stress are associated with lower levels of emotion regulation^116,117. Our within-person, intensive longitudinal results expand upon these cross-sectional findings among high-risk first responders, suggesting that when an individual EMS clinician reports experiencing higher momentary perceived stress than their usual, this moderately predicts that they will experience lower future momentary emotion regulation.

Second, we found that higher momentary emotion regulation strongly predicted lower momentary perceived stress over time in this high-risk population of EMS clinicians. To date, most prior intensive longitudinal studies have only examined the dynamic impacts of emotion regulation on emotional distress and not perceived stress. However, prior work has demonstrated a significant relationship between stress levels and emotional distress¹¹⁸. In one study deploying EMAs over 7 days in college students experiencing academic stress, results suggest that higher momentary emotion regulation at one time was weakly associated with decreased levels of emotional distress at a later time (discrete “time-lagged associations”)¹¹⁹. A meta-analysis of 37 EMA studies and 39 daily diary studies (surveys completed once daily) suggested that lower emotion regulation was associated with higher levels of emotional distress¹²⁰. However, 53% of these studies recruited solely college student samples, and most only deployed EMAs over 1–2 weeks¹²⁰. Our within-person, intensive longitudinal results extend beyond solely college students, suggesting that when an individual EMS clinician reports experiencing higher momentary emotion regulation than their usual, this strongly predicts that they will report experiencing lower future momentary perceived stress, especially over short periods of time from the reported stress.

In this current study, we also examined these key bidirectional relationships between momentary perceived stress and momentary emotion regulation in high-risk EMS clinicians with recent, regular substance use. Most prior work has only conceptualized stress as a predictor of future emotion regulation. However, a prior study using EMAs also found a bidirectional relationship between stress and emotion regulation in college students across time⁵⁰. Similar to our results between perceived stress and emotion regulation, the prior study identified small cross-lagged effects for academic-specific stress on self-regulation, but larger effect sizes for self-regulation on academic stress⁵⁰. Specifically, higher self-regulation capacity significantly predicted lower academic stress⁵⁰. In another longitudinal study of college students, relationships were examined between two surveys with a time-lag of 1 year, and results suggested that higher perceived stress significantly predicted lower future self-efficacy¹²¹. Similarly, the effect size for higher self-efficacy on lower future perceived stress was largest¹²¹. In our current study, we found that the effect of momentary emotion regulation on momentary perceived stress was over three times larger than that of stress on emotion regulation, which suggests that higher emotion regulation might outweigh the negative effects of high stress. Together, this suggests that emotion regulation might serve as a key interventional target to alleviate stress, which would be critical for this sample with unavoidable stressful exposures¹²². When examining these key relationships graphically, results suggest that emotion regulation might have the largest effect on perceived stress at time intervals of just a few hours (Fig. 3).

Lastly, our exploratory moderation analysis suggests that higher substance use severity did not change how perceived stress or emotion regulation predicted each other in EMS clinicians with regular substance use. It is important to note that participants in this study had moderate levels of alcohol and cannabis use, on average, and we excluded individuals who screened at the highest risk for alcohol use disorder⁶⁴ or cannabis use disorder⁶⁵. Those individuals likely need clinician-facing treatment and would not be the target of a standalone digital health intervention. Additionally, while alcohol and cannabis use can impact physiological, neurochemical, and behavioral stress responses, this has only been observed in individuals with more significant substance use risks³⁸. Future studies could include individuals with higher severity; however, meta-analyses suggest that even moderate substance use severity modulates relations between emotion regulation and daily substance use¹²³. Therefore, future analyses can test if severity scores moderate relationships between emotion regulation and daily substance use in EMS clinicians. Overall, this exploratory baseline moderation analysis was likely underpowered, so we cannot definitely state that significant relationships do not exist. Future research should include individuals with a wider range in AUDIT and CUDIT scores, as our inclusion criteria were limited to individuals with truncated scores below the threshold for probable substance use disorder. Future studies should also examine how momentary substance use at the time of each EMA (e.g., use since the last EMA) moderates real-time perceived stress and emotion regulation links as CTSEM packages expand.

Utilizing this mechanism–based approach, this line of research can inform the development of Just-in-Time Adaptive Interventions (JITAIs) for EMS clinicians. These novel interventions can provide support to the right people at the right moments by adapting to changes in an individual’s internal and external environments¹²⁴. JITAIs have the potential to collect near real-time data through self-assessments and passive sensing from a smartwatch to help determine when an EMS clinician has experienced a spike in stress and might benefit from a brief, evidence-based intervention¹²⁴. In the context of our current study and results, we have helped determine the potential to which increasing emotion regulation might decrease perceived stress. This can then inform the design of future JITAIs to improve engagement with emotion regulation strategies in times of high stress.

Given the negative impact of chronic stress on overall health across different populations, numerous interventions have been designed to help reduce stress and improve health outcomes^125–127. There are over 10,000 applications on the Apple and Google Play stores designed to promote behavioral health¹²⁸. Unfortunately, most of these apps are disseminated on the market without any scientific evaluation of their underlying mechanisms of behavior change¹²⁸. Therefore, we employed robust scientific methodology to examine the relationships between momentary perceived stress and emotion regulation in daily life among this high-risk population of EMS clinicians with regular substance use. In this current study of a national sample of EMS clinicians with regular cannabis use or alcohol drinking, we identified that momentary emotion regulation is a key factor associated with momentary stress through significant bidirectional relationships. Specifically, the reciprocal effects are strongest at lower time intervals of just a few hours, where higher emotion regulation had the largest effects on lower perceived stress over time (Fig. 3). Our brief emotion regulation measure assessed the frequency that EMS clinicians failed to adaptively express emotions, accept that their emotions were valid, or believe that their emotions were appropriate and not normal for them to feel. These negative viewpoints on emotions and emotional expression predicted higher stress levels.

Improving emotion regulation skills and enabling individuals to better express and control their emotions has produced desirable effects in other populations¹²⁹. Numerous therapies, such as dialectical behavioral therapy, cognitive behavioral therapy, and mindfulness-based cognitive therapy, can help improve emotion regulation and reduce rumination and self-criticism¹³⁰. Additionally, digital technologies and tools, such as health apps and web-based interventions, have been widely and effectively used to deliver these therapies to other populations, improving adaptive emotion regulation strategies and helping to reduce negative mental health outcomes¹³⁰.

Currently, the EMS profession primarily takes a different approach to mitigation, where an EMS manager must determine that an individual EMS clinician has experienced a stressful situation and then place them “out of service” for 1–2 h to conduct a critical incident stress debriefing (CISD)¹³¹. This intervention includes post-incident support and mitigation strategies from peers to process the stressful situation¹³¹. This approach does indeed seem efficacious, as a study of firefighters found that 90% reported at least one “critical incident” within the prior year, with an average of six of these incidents per year¹³². However, in a study of EMS clinicians, only about half report CISD as helpful, and these interventions were not associated with decreased stress, PTSD, or burnout¹³³. Additionally, CISD requires identification of a potentially stressful event; however, EMS clinicians often experience recurrent stressors, making this intervention less feasible^10,11. While CISD is intended to be delivered once the immediate stressor (e.g., stressful event) is resolved, this intervention might be difficult in an environment with such an accumulation of stress responses^10,11. Therefore, future JITAIs might help EMS clinicians recover from stressors until they reach a point where CISD might be more efficacious.

Our current study suggests that the timing of brief interventions may be critically important, as the effects of emotion regulation on perceived stress peak around 1 h. Therefore, future studies should consider testing JITAIs that bolster emotion regulation strategies in EMS clinicians and improve engagement with these emotion regulation strategies in times of high stress. JITAIs might provide a more effective way to monitor individual EMS clinician’s stress levels in near real-time and adapt to these changes in the individual’s stress levels¹²⁴. However, we do not want to prompt an intervention when an EMS clinician is in a moving ambulance or taking care of a patient, so future studies can explore pairing self-assessments with passive sensing data from a smartphone or smartwatch. These devices collect information on heart rate variability, an indicator of stress responses, as well as general location¹³⁴. We can determine when an EMS clinician has experienced a spike in their stress levels, but is at a time and place where they are available to receive a brief, evidence-based intervention right on their own smartphone. These interventions might include strategies such as brief mindfulness exercises, cognitive reappraisal prompts, or even a generative AI conversational agent designed for in-the-moment coaching¹³⁵.

A systematic review found that mindfulness-based stress reduction interventions are efficacious at decreasing stress levels in other populations of healthcare professionals¹³⁶. These studies also found that abbreviated mindfulness interventions are just as effective as longer 8-week interventions, and suggest that momentary interventions embedded within working days were even more efficacious at improving symptom severity in healthcare professionals¹³⁶. Additionally, cognitive reappraisal is an evidence-based strategy to decrease negative emotions during high-intensity situations by providing effective strategies to self-regulate negative emotions during times of high stress^137,138. Future JITAIs for EMS clinicians could consider incorporating brief cognitive reappraisal prompts to improve self-regulation of emotions in times of high stress.

There is also growing evidence suggesting that interpersonal emotion regulation, such as regulating emotions extrinsically with a peer or possibly a generative AI conversational agent, is effective at improving wellbeing and decreasing mental health symptom severity in both the individual and peer^{135,139–141}. In prior studies, 84% of EMS clinicians endorsed spending their downtime with peers after experiencing a critical incident, which suggests that future JITAIs should also incorporate interpersonal emotion regulation interventions¹³³. Future JITAIs could first identify when an individual has experienced a peak in stress levels and are back at the EMS station with their peers, and then prompt the EMS clinician to work with their peers or a conversational agent, if preferred, to (1) identify how they are feeling through individualized prompts, (2) set a short-term affective goal, (3) select a brief emotion regulation strategy to use at that current moment, and (4) implement the strategy to achieve their goals¹⁴².

Given the identification of these key relationships between perceived stress and emotion regulation in EMS clinicians with regular substance use, future research should investigate these mechanisms on substance use behavioral outcomes using the “experimental medicine” approach. This is a mechanism-focused way to examine targets of health behavioral outcomes outlined by the National Institutes of Health¹⁴³. The overall goal of this framework is to investigate the relationship between a modifiable mechanism and health behavior outcomes to find common, predictable targets of health risk behaviors¹⁴³. Then, later steps include incorporating these mechanistically informed targets into the development of replicable and scalable behavioral health interventions to reduce health risk behaviors¹⁴³. Importantly, we have completed the first step in this process to understand relationships between stress and emotion regulation targets, and future work should examine the impact of these potentially modifiable targets on substance use outcomes within these high-risk EMS clinicians.

This novel, intensive longitudinal study was the first to demonstrate key within-person relationships among perceived stress and emotion regulation in EMS clinicians who engaged in regular substance use behaviors. Despite our robust methodology, the results should still be interpreted while considering several potential limitations. First, several inclusion and exclusion criteria limit the generalizability of our findings. In this study, we limited enrollment to full-time EMS clinicians working in 9 − 1 − 1 “emergency” response systems to first explore these relationships in EMS clinicians with the highest exposure to occupational stressors⁶⁶. Therefore, results may not be generalizable to part-time EMS clinicians or those who work in more community-based settings, including local vaccine clinics or home-health visits.

Second, we included only daytime shift workers to ensure a consistent daytime EMA schedule with high EMA completion rates, and since nighttime EMS clinicians have different risks, which this study was not powered to detect^67,68. Future studies could replicate these methods in a larger, stratified study of both day and night shift workers designed specifically to compare these two samples. Additionally, all participants had to meet the substance use criteria of regular cannabis or alcohol use. Including only EMS clinicians at risk for engaging in health risk behaviors limits the generalizability of the findings, which may not be representative of EMS clinicians with lower or higher substance use risks.

Fourth, despite potential for interpersonal emotion regulation, we did not include EMA measures on social support or workplace resources as potential protective factors. Prior studies suggest that social contexts also impact perceived stress and emotion regulatory strategies^144–146. However, adding additional EMA measures would increase the length of EMAs, which may be too burdensome. We also believed that these workplace/work-unit resources are unlikely to vary throughout the day momentarily.

Additionally, our EMA measures did not assess substance use motives, and therefore, we do not know why participants used substances momentarily in this study (e.g., stress coping, social contexts, use disorder). A recent systematic review of 64 different EMA studies suggests that assessing substance use motives is likely biased by retrospective recall, and the authors suggest further research and measure development^147,148. Unfortunately, key aspects (e.g., utility and validity) of this motivational model for substance use have not been systematically examined or have very little evidence in EMA studies^148,149. In another EMA study of adults with alcohol and/or cannabis disorders, participants reported an “unknown motive” in 40% of the assessments, and this was the only option selected in 77% of EMAs¹⁵⁰. Given these key methodological concerns and frequent reporting of “unknown” motives, we believe that further research needs to be done to better develop and validate momentary motivational measures before deployment.

First responders may be using alcohol and/or cannabis to promote both maladaptive (e.g., avoidance, dissociation) and adaptive (e.g., social bonding, interpersonal re-processing) coping strategies. For example, the fire service, specifically, has a pro-social organizational structure with regular social gatherings and community events^151,152. Firefighters often report alcohol drinking as a key part of this occupational and social culture^153,154. In prior studies of first responders, stress coping accounted for the most variation in alcohol use; however, socially-motivated use accounted for the second most variation¹⁵⁵. In prior studies, alcohol drinking behaviors facilitated increased social bonding within group settings¹⁵⁶, and therefore, these substances might be used to help promote post-shift social bonding between coworkers.

Regarding limitations during data analysis, while we utilized robust CTSEM methods, the emotion regulation scale had a large grand mean of 4.33 points (out of 5 total). This may increase the likelihood of a “ceiling effect,” which can hinder parameter estimates in longitudinal data¹⁵⁷. Additionally, we did not power the current analysis to test for cross-level moderation. Moderation analyses are considered exploratory to inform future work.

Lastly, we utilized the brief 3-item emotion regulation subscale of the momentary self-regulation scale, which measured how participants thought about and expressed their emotions since the last EMA²⁰. However, this self-regulation subscale did not measure the strategies that participants commonly used to effectively regulate their emotions. Future research should explore what emotion regulation strategies are most helpful before, during, and after their EMS shifts, which would allow for further personalization of potential JITAIs. We kept these EMA prompts short (<2 min) to ensure high completion.

This first-of-its-kind intensive longitudinal study examined the momentary relationships between perceived stress and emotion regulation in the daily lives of 110 full-time EMS clinicians who regularly engaged in substance use and were recruited from across the United States. Through utilizing robust and novel Bayesian CTSEM, we identified key momentary reciprocal relationships between perceived stress and emotion regulation in these EMS clinicians at high-risk for substance use behaviors. Specifically, higher emotion regulation had the strongest effects on lower perceived stress over time. Therefore, momentary emotion regulation may be a key construct associated with momentary stress. Subsequent studies can test interventions aimed to improve momentary emotion regulation. Since stress is a repeated occupational exposure in EMS clinicians, we must identify modifiable momentary mechanisms as future interventional targets to reduce perceived stress levels, which could then help decrease health risk behaviors in high-risk EMS clinicians^16–18.

Author contributions

E.G.P.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, visualization, writing—original draft preparation. M.R.F.: investigation, methodology, visualization, supervision, writing—review & editing. N.C.J.: conceptualization, formal analysis, funding acquisition, investigation, methodology, software, visualization, supervision, writing—review & editing. J.J.G.: conceptualization, investigation, methodology, writing—review & editing. A.R.P.: conceptualization, investigation, methodology, resources, writing—review & editing. P.J.W.: conceptualization, investigation, methodology, writing—review & editing. L.A.M.: conceptualization, formal analysis, funding acquisition, investigation, methodology, resources, visualization, supervision, writing—review & editing.

Data availability

The datasets generated and/or analyzed during the current study are not yet publicly available as they form part of an ongoing doctoral dissertation at Dartmouth College. However, data are available from the corresponding author upon reasonable request.

Competing interests

L.A.M. has an affiliation with Square2 Systems, Boehringer Ingelheim, and Click Therapeutics. These relationships are extensively managed by her employer, Dartmouth College. The other authors do not have a competing interest.