When trauma triggers craving: meta-analytic evidence for the mediating role of negative affect among trauma-exposed individuals who use substances
Cuando el trauma desencadena el craving: evidencia metaanalítica del papel mediador del afecto negativo entre individuos expuestos al trauma que consumen sustancias
EUROPEAN JOURNAL OF PSYCHOTRAUMATOLOGY
S. DEGRACE ET AL.
Department of Psychology and Neuroscience, Dalhousie University, Halifax, NS, Canada
Department of Psychology, The University of British Columbia, Vancouver, BC, Canada
Department of Psychiatry, Dalhousie University, QEII Health Sciences Centre, Halifax, NS, Canada
School of Psychology, University of Sussex, Brighton, UK
Department of Psychology, Neuroscience & Behaviour, McMaster University, Hamilton, ON, Canada
CONTACT Sherry H. Stewart sstewart@dal.ca Department of Psychology and Neuroscience, Dalhousie University, 5909 Veterans' Memorial Lane, 8th Floor, Abbie J. Lane Memorial Building, QEII Health Sciences Centre Halifax, NS, B3H 2E2 CanadaABSTRACT
Background: In experimental studies using the cue reactivity paradigm (CRP), classically conditioned substance craving is well-documented among individuals who use substances with trauma histories. However, the magnitude of this effect has not yet been quantified, nor have moderators of this relationship been identified [DeGrace, S., Romero-Sanchiz, P., Standage, C., & Stewart, S. H. (2022). A scoping review of the literature on trauma cue-induced drug craving in substance users with trauma histories or PTSD. IntechOpen. https://doi.org/10.5772/intechopen.103816].
Objective: While the mechanism driving these trauma cue-induced craving effects has been suggested to involve trauma cue-induced negative affect [Baker, T. B., Piper, M. E., McCarthy, D. E., Majeskie, M. R., & Fiore, M. C. (2004). Addiction motivation reformulated: An affective processing model of negative reinforcement. Psychological Review, 111(1), 33–51. https://doi.org/10.1037/0033-295X.111.1.33], this remains largely untested.
Method: We conducted a meta-analysis to quantify the trauma cue-induced craving effect size, as well as possible moderators of these effects. Moreover, we utilized two-step meta-analytic structural equation modelling (TS-MA-SEM) to test the indirect effect of trauma (vs. neutral) cue exposure on substance craving through negative affect as the mediator. Our literature search returned n = 32 studies involving a total of k = 21 unique datasets that were included in our analyses.
Results: Results supported our hypothesis that trauma cues would induce greater substance craving relative to other cue types/controls (neutral cues, substance cues, pre-cue baseline), with small to medium effect sizes (d = 0.259 – 0.526). We found a few theoretically relevant moderators (CRP method, study quality, and % of sample with PTSD) that varied by comparator condition. Additionally, results supported our proposed mediational model with negative affect indirectly explaining the relationship between trauma cue exposure and substance craving.
Conclusions: Results suggest that trauma cue-elicited substance craving is a robust phenomenon that may be explained through trauma cue-induced negative affect driving increased craving.
Highlights
- In cue reactivity studies, trauma cues significantly heighten substance craving in relation to other comparators (i.e. baseline, neutral, and substance cues).
- Negative affect mediates the relationship between experimental trauma cue exposure and substance craving.
- Moderator effects suggest factors such as study quality and type of cue reactivity paradigm used may influence effect size under specific comparator conditions.
Trans Abstract
Antecedentes: En estudios experimentales que utilizan el paradigma de reactividad de señales (CRP, por sus siglas en inglés), el craving por sustancias, condicionada clásicamente, está bien documentada en personas que consumen sustancias con antecedentes traumáticos. Sin embargo, aún no se ha cuantificado la magnitud de este efecto ni se han identificado los moderadores de esta relación (DeGrace et al., 2022).
Objetivo: Si bien se ha sugerido que el mecanismo que impulsa estos efectos de craving inducidos por señales traumáticas involucra el afecto negativo inducido por señales traumáticas (Baker et al., 2004), esto aún no se comprobado en gran medida.
Método: Realizamos un metaanálisis para cuantificar la magnitud del efecto del craving inducida por señales traumáticas, así como los posibles moderadores de estos efectos. Además, utilizamos un modelo metaanalítico de ecuaciones estructurales de dos pasos (TS-MA-SEM por sus siglas en ingles) para evaluar el efecto indirecto de la exposición a señales traumáticas (en comparación con señales neutrales) sobre el craving por sustancias, a través del afecto negativo como mediador. Nuestra búsqueda bibliográfica arrojó n=32 estudios que abarcaban un total de k=21 conjuntos de datos únicos, los cuales se incluyeron en nuestros análisis.
Resultados: Los resultados respaldaron nuestra hipótesis de que las señales traumáticas inducirían un mayor craving por sustancias en comparación con otros tipos de señales/controles (señales neutrales, señales de sustancias, línea base previa a la señal), con tamaños de efecto de pequeños a medianos (d = 0,259 – 0,526). Encontramos algunos moderadores teóricamente relevantes (método CRP, calidad del estudio y porcentaje de la muestra con TEPT) que variaron según la condición de comparación. Además, los resultados respaldaron nuestro modelo de mediación propuesto, donde el afecto negativo explica indirectamente la relación entre la exposición a señales traumáticas y el craving por sustancias.
Conclusiones: Los resultados sugieren que el craving por sustancias provocada por una señal traumática es un fenómeno sólido que puede explicarse a través del afecto negativo inducido por la señal traumática que impulsa un mayor craving.
1.Introduction
Posttraumatic stress disorder (PTSD) can develop following exposure to a traumatic event (e.g. sexual assault, motor vehicle accident, combat) in some individuals and is characterized by significant distress and/or disruptions to an individual’s daily functioning through symptoms such as recurring nightmares, avoidance of trauma reminders, persistent negative emotional state, and hypervigilance (American Psychiatric Association, 2022). PTSD often co-occurs with substance use disorders (SUDs), with comorbid PTSD-SUD associated with greater functional impairment in comparison to PTSD or SUD alone (McCauley et al., 2012). For example, one study found those with (vs. without) cocaine use disorder were 2.18 times more likely to have PTSD in their lifetimes (Saunders et al., 2015). Elevated comorbidity rates are also evident for cannabis use disorder (CUD), with one study reporting those with a PTSD diagnosis were 2.6 times more likely to have CUD compared to those who had experienced a traumatic event but did not have PTSD (Walsh et al., 2014). Similarly, 17.6% of individuals diagnosed with PTSD in the U.S. will also be diagnosed with CUD at some point in their lifetime (Hasin et al., 2016). For alcohol use disorder (AUD), diagnosed individuals are 1.3 times more likely to also be diagnosed with PTSD at some point in their lifetime (Grant et al., 2015) and longitudinal evidence suggests that PTSD confers heightened risk for subsequent AUD onset (Ouimette & Read, 2014).
While SUDs and PTSD frequently co-occur, and while we now have a fair amount of evidence for both mutual maintenance and self-medication models (Hawn et al., 2020; Patton et al., 2024), gaps remain in our understanding of the etiology and maintenance factors of this comorbidity. Self-medication is one mechanism that may help in understanding this relationship (Stewart, 1996). The self-medication hypothesis (Khantzian, 1997) posits that substances are used to alleviate emotional distress and pain, and to regain control; in the context of PTSD, substances are thought to be used to cope with the distress caused by PTSD symptoms. From an operant conditioning perspective, substance use is negatively reinforcing among those with PTSD due to the ability of many substances to acutely relieve distressing PTSD symptoms. In other words, people with PTSD may learn to use substances to self-medicate for PTSD symptoms and associated distress.
Another theoretical mechanism to account for the high PTSD-SUD comorbidity involves classical conditioning. It has long been established that through frequent pairing with substance use, environmental substance-related cues (e.g. a cigarette package for a smoker) can come to elicit substance craving (a strong urge to use) which maintains problematic substance use. For those with PTSD, through their repeated pairing with substance use (due to self-medication), trauma cues (e.g. reminders of the traumatic experience such as loud noises for a combat veteran) may similarly come to elicit conditioned substance craving which contributes to substance use maintenance.
In addition to external trauma cues eliciting conditioned substance craving, another type of cue, namely internal negative affect cues, may serve as conditioned stimuli. Baker and colleagues (2004) proposed that avoidance of/escape from negative affect (i.e. negative feelings/mood) is a strong motivator for substance use through negative reinforcement processes. The consequent repeated pairing of substances with negative affective states would theoretically create strong associations between internal negative affect cues and substance use such that the experience of negative affect would come to elicit substance craving. Given that trauma cues have been shown to elicit conditioned negative affective responses themselves (i.e. due to the pairing of strong negative affect with the original traumatic experience), it is possible that negative affect may wholly or partially mediate (explain) the link of trauma cue exposure with substance craving. In other words, trauma cue exposure may elicit negative affect which itself may trigger conditioned substance craving.
The above theorized conditioned craving mechanisms have been tested experimentally through the use of cue-reactivity paradigms (CRP). Specifically, CRPs expose participants to sensory cues (typically visual and/or auditory) related to a psychologically traumatic event (e.g. a photo of a car wreck for a motor vehicle accident survivor) and a similarly detailed neutral control event (e.g. a photo of a toothbrush and toothpaste; DeGrace et al., 2022). In the context of addiction research, CRPs use substance-related cues with the intention of eliciting responses rooted in motivational processes, such as substance craving (e.g. salivation, subjective reports of craving, Reynolds & Monti, 2013). However, for those with trauma histories who use substances, cues related to the individual’s traumatic experience may also elicit such reactions. Indeed, extant research has shown that exposure to trauma reminders can stimulate both substance craving responses and negative affective responses relative to various control conditions (e.g. DeGrace et al., 2023; Romero-Sanchiz et al., 2022).
A recent scoping review established the breadth and utility of the CRP in PTSD-SUD comorbidity research (DeGrace et al., 2022). However, to date there has been no quantitative review of this literature. A quantitative synthesis of studies in this area would allow for determination of the overall magnitude of the trauma cue-elicited craving effect, permitting an evaluation of its meaningfulness, and providing researchers with a standardized effect size against which to compare those found in individual studies. A meta-analysis of this literature would also allow for a quantification of variables that may moderate this effect, for example, population characteristics, substance type, and/or CRP control condition. Additionally, an examination of the possible mediating role of negative affect would provide a test of Baker et al.’s (2004) theory that negative affective states may be the primary driver of craving. Given the large samples employed in meta-analytic methods, these analyses should be particularly robust, allowing for well-powered tests of moderation and mediation.
Thus, the present meta-analysis sought to quantify the magnitude of trauma-cue induced substance craving responses using the CRP methodology, analyse potential moderators of the trauma-cue induced craving response, and examine if negative affective responses to the trauma cue mediate the relationship between trauma cue presentation and subsequent substance craving. We put forward several hypotheses. First, we hypothesized that, within the CRP framework, craving to trauma cues would be greater in magnitude relative to baseline (pre-cue) craving, and relative to craving to neutral or substance cues (H1). Second, we expected that the percentage of PTSD and SUD symptoms present in the sample, as well as substance type, CRP methodology, and study quality, would moderate the effect of trauma cue-induced craving (H2). Specifically, based on findings from our scoping review (DeGrace et al., 2022) and evidence demonstrating PTSD symptom severity is associated with increased trauma cue-elicited craving (Saladin et al., 2003), we expected to find greater trauma cue-elicited craving effects with samples with higher percentages of participants with PTSD. We expected the same pattern in samples with higher percentages of participants with SUD, as supported by Vafaie and Kober (2022), who identified substance craving as most prominent among those with SUD (compared to those without). We also expected to observe greater substance craving in samples using more sedating (vs. stimulating) types of substances, as variation in cue reactivity responses has been demonstrated across substance types (Coffey et al., 2002) and theoretically learned associations between trauma cues and substance use (and hence conditioned craving) should be strongest for substances most likely to lead to dampening of distress. Finally, based on methodological differences observed in our scoping review (DeGrace et al., 2022), we also expected greater substance craving to be detected in studies using the classic audiovisual cue CRP (vs. other less commonly used forms of CRP), and studies of higher quality (i.e. lower risk of bias). Finally, we expected the relationship between trauma cue exposure (trauma vs. neutral cue) and substance craving would be mediated by negative affect (H3).
2.Method
The present meta-analysis followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021) and was preregistered in the International Prospective Register of Systematic Reviews (PROSPERO; Registration ID: CRD42022304580). The preregistration outlined our research questions, eligibility criteria, search strategy, planned moderator and mediation analyses, and analytic approach.
2.1.Search strategy
Bibliographic searches were completed in August 2022, with a follow-up in January 2025, using the databases PubMed, PsycINFO, the Cochrane Library, and PTSDPubs. Each search was conducted as follows using a Boolean search logic and relevant keywords: (“PTSD” OR “post traumatic stress*” OR “posttraumatic stress*” OR “trauma”) AND (“cue” OR “cue exposure” OR “cue-reactivity” OR “conditioned response” OR “stimuli” OR “cue-induced” OR “cue-elicited”) AND (“substance*” OR “alcohol” OR “drug*” OR “cocaine” OR “cannabis” OR “marijuana” OR “marihuana” OR “THC” OR “CBD” OR “opioid*” OR “opiate*” OR “tobacco” OR “nicotine”) AND (“craving*” OR “urge”). Results from each search were imported into Covidence, a literature screening platform. Covidence first removed duplicates, then, an abstract and title screening was conducted by two independent reviewers to remove studies not meeting criteria. The two raters then completed the full-text screening stage, where studies were read in full by the reviewers and screened based on our inclusion and exclusion criteria. Any conflicts between the two reviewers were discussed and resolved to consensus with the research team.
Our initial literature search in July 2022 produced 91 studies for abstract screening. Of these, 52 were excluded and the remaining 39 studies were assessed at the full-text screening phase. Twenty-eight of these studies were excluded for not meeting all inclusion criteria (see Figure 1), leaving twenty-four studies meeting all criteria for data extraction (excellent inter-rater reliability; κ = 0.80; raters SD and TC).
An updated literature search in January 2025 (constrained from 2022 – onward) produced an additional twenty-seven studies for abstract screening after duplicate removal. Seventeen of these studies were excluded and thus ten new studies were moved to the full text screening phase. Eight out of ten studies were subsequently added to the initial 24 for data extraction (inter-rater reliability; κ = 0.74; raters SD and MEG) as they met inclusion criteria and were not published (or identified) at the time of our initial search. Five of the studies identified in the 2025 search, however, were derived from datasets included in the 2022 search; thus, only three unique datasets from the eight new studies were available from the updated search. A total of 32 studies (21 unique datasets) are included in the present meta-analysis (see Figure 1). When multiple studies used the same sample, we chose to include the study with the largest sample size. When studies using the same dataset had the same sample size, studies with the most relevant information for the present meta-analysis were chosen.
2.2.Inclusion criteria
Studies were included if they used an experimental design, if they included a CRP, and if self-reported craving was assessed post cue-reactivity paradigm. Furthermore, the population of interest had to include individuals who have experienced a past traumatic event.1 Alternatively, studies were included if PTSD symptom data was provided descriptively and data demonstrated participants were a trauma-exposed population. Additionally, it was required that participants report on their substance use or substance use severity to ensure the sample used substances. Finally, the presentation of a trauma cue as well as a comparator condition (i.e. pre-cue baseline, neutral cue, and/or substance cue) were required to be included as a condition within the cue reactivity paradigm.
2.3.Exclusion criteria
We excluded studies that were not written in English, studies testing therapeutic interventions that did not include data at pre-treatment baseline, and studies in which humans were not the research participants. We did not exclude grey literature (e.g. theses/dissertations) to gather the full scope of research in this area and to reduce publication bias.
2.4.Data extraction2
We extracted means and standard deviations for craving scores at baseline and for each cue condition (neutral, trauma, substance). When available, these were transformed into correlation coefficients to align with our chosen effect size metric.3 For H1 and H2, correlations between cue condition (1 = trauma, 0 = control) and craving were extracted directly when reported or derived from available summary statistics. To test H3, if not included in the paper or in published supplemental material, we requested a bivariate correlation table for each author’s dataset of our variables of interest within different cue conditions. Specifically, authors were asked to provide bivariate correlations between negative affect and substance craving/urge across cue conditions (i.e. baseline, neutral, trauma, and/or substance-related cues), and between cue condition and both negative affect and substance craving. These correlations were used to estimate the mediation paths specified in H3: the association between cue condition and negative affect (path a), the association between cue condition and craving (path c), and the association between negative affect and craving controlling for cue condition (path v). The first author contacted all authors and provided an eight-week timeframe to provide the requested data, with reminders sent at the 3-week and 6-week mark. An additional two-week extension was available if authors requested more time. Two authors did not provide the requested data.
2.5.Study coding
Studies included in the meta-analysis were coded on sample size, mean age, sex, percentage of the sample with PTSD, percentage of the sample with SUD, as well as measures used to assess negative affect and substance craving. Studies were also coded based on the type of CRP they employed (1 = personalized audio-recorded script paired with a trauma-related photograph; 2 = personalized audio-recorded script paired with in-vivo cues, specifically water or alcohol stimuli; 3 = other paradigms characterized by authors as trauma cue reactivity paradigms, such as non-personalized trauma-related videos,4 structured interviews, written trauma recall tasks, and physiological induction of hyperventilation). Substances were coded as 1 = alcohol, 2 = cannabis, 3 = nicotine, 4 = other drug (e.g. cocaine, opioids).
2.6.Study quality assessment
Two raters (SD and TC) independently completed study quality assessments on all studies included in the meta-analysis using the Cochrane Risk of Bias tool (ROBINS-I; Sterne et al., 2016). Based on the ROBINS-I assessments, each study was categorized into one of four levels of risk of bias: (1) low risk, (2) moderate risk, (3) serious risk, or (4) critical risk. Discrepancies in ratings were resolved through discussion to reach consensus. Inter-rater agreement prior to consensus was 88%. Cohen’s kappa was κ = .36; although low, nearly all ratings fell in the lowest risk category, making kappa likely an underestimate of inter-rater consistency. The prevalence-adjusted, bias-adjusted kappa (PABAK) was .77, indicating substantial agreement beyond chance under conditions of low category variance.
3.Results
3.2.H1
We first conducted a traditional meta-analysis using R (v. 4.4.2; metafor package, Viechtbauer, 2010). Cue condition was treated as a within-subjects variable, consistent with the design of the included studies in which participants were exposed to multiple cue types (e.g. neutral, trauma, substance). Raw means and standard errors for substance craving within each cue condition were extracted where available, and, when needed, correlation coefficients were derived from summary statistics. These values were then transformed into Fisher’s z-scores. Using this data, we correlated the dichotomous within-subjects variable representing cue type (trauma cue versus comparator) with the degree of substance craving present under that cue condition (see Table 2 for full meta-analyses breakdown). We ran individual meta-analyses using random effects (REML) models by comparator with the correlation coefficients transformed into Fisher's z to ensure we met the normality assumption. Using baseline craving as the comparator, the overall effect of cue type on substance craving (trauma cue-induced craving compared to baseline craving) was significant (r = 0.129, 95% CI [0.086–0.251], p = .035; k = 8; see Figure 2) but small in magnitude (transformed to Cohen’s d = 0.259). The heterogeneity analysis (Qr = 16.96, p = .017) indicated significant variability across studies. Using neutral cues as the comparator, the overall effect of cue type on substance craving (trauma cue-induced craving vs. substance craving to a neutral cue) was significant (r = 0.260, 95% CI [0.179–0.341], p < .0001; k = 17; see Figure 3), with the effect representing a medium sized effect (transformed Cohen’s d = 0.526). The test for total heterogeneity suggested significant variability across studies (Qr = 27.41, p = .037). Finally, using substance cues as the comparator, the overall effect of cue type on substance craving (trauma cue-induced craving vs. substance cue-induced craving) was significant (r = 0.137, 95% CI [0.043–0.231], p = .004; k = 6; see Figure 4) with the effect representing a small-medium effect size (transformed Cohen’s d = 0.466). The heterogeneity test indicated no substantial variability across studies (Qr = 3.63, p = .810).
k N Estimate 95% CI p I2 Qr Trauma – Baseline 8 830 0.129 [0.086–0.251] .035* 62.70% 16.96* Trauma – Neutral 17 1244 0.260 [0.179–0.341] <.0001*** 42.69% 27.41* Trauma – Substance 6 479 0.137 [0.043–0.230] .004** 3.12% 3.63
3.3.H2
Our meta-analyses of trauma (vs. neutral) cue-induced craving and trauma (vs. baseline) cue-induced craving indicated significant heterogeneity, suggesting the presence of possible moderators. Thus, we re-ran our meta-analyses for these two models across several potential moderators. Specifically, we tested the moderating effects of the percentage of participants with PTSD (%) and SUD (%), as well as CRP type, study quality, and type of substance used by the sample. In the trauma-baseline analysis, PTSD (%) emerged as a moderator of the trauma cue type effect on craving (r = 0.007, 95% CI [0.00–0.02], p = .05; k = 7; see Table 3) with the trauma cue-induced craving effect proving stronger in samples with a larger proportion of participants with PTSD. In the trauma-neutral analysis, the CRPs assigned the category ‘other’ (e.g. structured interviews, writing tasks) were significantly less effective in eliciting trauma cue-induced substance craving compared to traditional audio-visual CRPs (r = −0.174, 95% CI [−0.345–0.003], p = .046; k = 17; see Table 3). Further, higher study quality (i.e. lower risk of bias) marginally predicted greater substance craving in the trauma-neutral analysis (r = −0.201, 95% CI [−0.431–0.029], p = .086; k = 17; see Table 3).
Trauma – Baseline k N Estimate 95% CI p I2 QM PTSD (%) 7 726 0.007 [−0.001–0.015] .051† 52.42% 3.78 SUD (%) 5 389 −0.004 [−0.105–0.058] .501 77.38% 0.452 CRP (Type; Other-AV) 8 830 −0.100 [−0.350–0.149] .430 62.87% 0.430 Study Quality 8 830 −0.177 [−0.420–0.065] .152 59.86% 2.05 Substance (Alcohol-Cannabis) 8 830 0.014 [−0.201–0.230] .896 68.91% 0.017 Trauma – Neutral k N Estimate 95% CI p I2 Qr PTSD (%) 14 1081 0.002 [−0.007–0.062] .120 33.09% 2.40 SUD (%) 14 1081 −0.000 [−0.004–0.033] .770 47.62% 0.084 CRP (Type) 17 1244 28.36% 5.03† In-vivo – AV 0.021 [−0.161–0.202] .825 Other – AV −0.174 [−0.345 – – 0.003] .046* Study Quality 17 1244 −0.201 [−0.431–0.029] .086† 38.04% 2.93 Substance 16 1098 46.25% 0.694 Alcohol – Cannabis −0.071 [−0.296–0.154] .535 Alcohol – Cocaine −0.787 [−0.291–0.134] .468
3.4.H3
Our third aim was to conduct a mediational meta-analysis using meta-analytic structural equation modelling (MA-SEM) to evaluate the indirect effect of cue condition (trauma vs. control) on substance craving through negative affect. To do so, we synthesized correlational data capturing the relationship between cue condition (X) and negative affect (M), and craving (Y).
Using this data, we used two-step meta-analytic structural equation modelling (TS-MA-SEM) using the shinyapp webMASEM (Jak et al., 2021). Specifically, we used correlations between negative affect and substance craving following trauma cues (coded as 1) versus neutral cues (coded as 0), as well as point-biserial correlations between cue type and negative affect, and between cue type and craving (pooled N = 854). We specified a fully saturated model including a direct path from cue type to craving, with negative affect entered as a mediator in the indirect pathway from cue type to craving. In step one, study heterogeneity for the pooled correlation matrix was significant (Qr = 234.59, p < .0001) and was high for intercept one (I2 = 92.4%) and intercept three (I2 = 64.2%) but not intercept two (I2 = 0%). The high variability in intercepts one and three suggest that trauma cues may affect negative affect (intercept one) and craving (intercept three) differently across studies, though the relationship between negative affect and craving (intercept two) was consistent across studies. In step 2, results of the TS-MA-SEM did not support the direct pathway between cue type and craving (r = −0.05, 95% CI [−0.144–0.044], p = .28). As hypothesized, results provided support for the indirect pathway (see Figure 5), with the trauma cue condition (vs. the neutral cue condition) significantly predicting negative affect (r = 0.39, 95% CI [0.235–0.545], p < .0001) and negative affect in turn significantly predicting substance craving (r = 0.42, 95% CI [0.328–0.512], p < .0001). This indirect effect was significant (r = 0.164, 95% CI [0.090–0.238], p < .0001), indicating that the trauma condition (0 = neutral, 1 = trauma) was associated with higher negative affect, which in turn was associated with higher substance craving.
4.Discussion
The present meta-analysis was the first to quantify the magnitude of trauma cue-induced substance craving among trauma-exposed individuals who use substances. Consistent with our first hypothesis, based on predictions of classical conditioning theory (Romero-Sanchiz et al., 2022; Stasiewicz & Maisto, 1993), trauma cues elicited significantly greater craving than both neutral cues and than the degree of craving at pre-cue baseline with effects proving medium and small in magnitude, respectively. Interestingly, trauma cues elicited significantly greater substance craving even when compared to substance cues (a small to medium effect) suggesting the classical conditioning to trauma cues is even stronger than the well-established phenomenon of substance cue-induced craving (Drobes & Tiffany, 1997) among those who use substances with trauma histories.
Consistent with our second hypothesis, we did find some moderators of the degree of trauma cue induced craving. Specifically, the percentage of the sample with PTSD moderated the degree of trauma cue-induced substance craving relative to craving at pre-cue baseline, with the magnitude of trauma cue induced craving increasing as the proportion of the sample with PTSD increased. These results suggest that classically conditioned craving to trauma cues may be a particularly important mechanism to explain substance use in those with PTSD as opposed to among those with trauma histories, per se. Also consistent with hypotheses, study quality was an observed moderator with the magnitude of trauma cue-elicited craving relative to craving to neutral cues proving greater for higher (vs. lower) quality studies. This result points to the importance of strong study design in the conduct of CRP research in the PTSD-SUD field to ensure detection of trauma cue induced craving effects when they exist. A final observed moderator was CRP type with cue induced craving to trauma vs. neutral cues proving stronger in the case of traditional audiovisual CRPs relative to CRPs classified in the ‘other’ category (e.g. expressive writing). This suggests that when testing the efficacy of a new CRP methodology for eliciting trauma cue induced craving, researchers may be best to compare their effects against the use of the traditional audiovisual cue. Perhaps audiovisual cues are most effective in eliciting trauma cue-induced craving as they use external cues (e.g. imaginal or written cues might be less effective for those with poorer imaginal abilities) and/or because they combine two sensory modalities to maximize effect. Overall, though, it is important to note the general lack of detection of moderators (and lack of significant heterogeneity in the case of trauma vs. substance cue induced craving); this finding suggests that trauma cue-induced craving is a robust phenomenon that is relatively invariant among trauma exposed individuals who use substances (e.g. applying across different substances of abuse and across those who use substances with and without clinical SUD). However, given the moderate heterogeneity observed in other comparisons and the relatively small number of studies within levels of each moderator, this interpretation should be taken with caution.
Finally, results provided support for our third hypothesis, with negative affect acting as a significant mediator in the relationship between cue exposure (trauma vs. neutral) and substance craving. Indeed, the fact that no direct effect between cue exposure and substance craving remained after accounting for the indirect effect through negative affect, points to a very strong role of trauma cue-elicited negative affect in explaining how exposure to trauma cues comes to elicit substance craving. This finding supports approaches to PTSD-SUD treatment that focus on reducing negative affect to trauma cues (e.g. exposure-based therapies) as a method for reducing the trauma cue-induced craving that may contribute to substance use maintenance. Interestingly, our findings are contrary to a recent meta-analysis that identified that daily fluctuations in negative affect were not associated with increased daily alcohol use (Dora et al., 2023). This discrepancy may reflect important differences in study design and outcomes. Indeed, while Dora and colleagues (2023) utilized daily diary data, the present meta-analysis examined studies using lab-based trauma cue reactivity paradigms and examined craving rather than substance use. It may be that trauma cue-elicited negative affect exerts an effect on craving in these acute contexts, without necessarily translating into increased substance use at a daily level. Importantly, Dora et al. (2023) highlight that the relationship between negative affect and substance use may vary by a number of situational factors; perhaps trauma cue-elicited negative affect represents a particularly potent and specific trigger for craving that might not be captured by momentary changes in affect. Indeed, given Dora and colleagues (2023) found an association between daily changes in positive (but not negative) affect and substance use, this discrepancy might reflect processes operating in broader or non-traumatized samples. In contrast, in trauma-exposed individuals, trauma-specific motivational processes (e.g. avoidance, relief-seeking) may be more activated by trauma reminders, thereby linking acute negative affect to craving in different ways from general affect-use situations observed in daily life. This is partially supported by other work by Dora et al. (2024), who found in an exploratory analysis that a higher demand for alcohol consumption on the Alcohol Purchase Task was elevated following negative (vs. neutral) mood inductions.
Our study has several strengths. It is the first to quantify the degree of substance craving across a variety of CRPs (DeGrace et al., 2022) in samples of trauma-exposed substance individuals who use substances (Renaud et al., 2021). Importantly, the quantification of the trauma cue induced craving effect across several potential moderators gives researchers an understanding of typical effect sizes in this population and how such effects vary across variables such as comparator condition, PTSD status, study quality, and type of CRP. Furthermore, the results of our mediation model provide support for the theorizing of Baker et al. (2004), whosuggests that the negative affect elicited by the trauma CRP (Read et al., 2017), rather than the CRP itself, is the primary driver behind trauma cue-induced substance craving.
Our meta-analysis is not without its limitations. First, the studies included in our meta-analysis contained little variation in methodology, potentially limiting our ability to detect moderation in our analyses. Some moderators contained too little variability across studies to examine statistically. For example, we had originally planned to examine whether greater trauma cue-elicited craving effects are observed for personalized than generic trauma cues (e.g. a combat veterans’ personal trauma story vs. a generic video depicting a traumatic combat scene), but only two studies used generic trauma cues precluding analysis of this moderator. Second, study participants included in this meta-analysis were from North American and Europe; it is unclear whether such results would be replicated in non-Western countries/cultures. Third, the affective measures used across studies (see Table 1) capture related but distinct affective constructs, including but not limited to general negative affect, arousal and situational distress measures. As such, the mediator in our model reflects a broad affective response. Future work that employs more standardized affect measures, or that examines whether effects differ by the specific affect measure used, would allow for more precise tests of this mediating mechanism. Finally, the observed moderators did not replicate across comparator conditions (trauma vs. neutral cue; trauma cue vs. pre-cue baseline) complicating their interpretation. However, it is important to note that the moderator analyses are not directly comparable across comparator conditions as they contained some different studies (i.e. not all studies included all comparator conditions).
Our results contribute to the cue-reactivity literature in a trauma-exposed, substance using population, by demonstrating the magnitude of substance craving is greatest following trauma cues compared to all other cue types/control conditions. It is unclear why using neutral cues as the comparator yielded stronger trauma cue induced craving effect than when using pre-cue baseline craving levels as the comparator. It is possible that anticipatory anxiety at baseline when awaiting the trauma cue may have contributed to dampening of the trauma cue-induced craving effect given prior findings that even anticipating exposure to a trauma reminder can induce negative affect (Grillon et al., 2009). Practically, our findings suggest that in future, trauma CRP studies in the PTSD-SUD field consider using neutral cues as the comparator to maximize the opportunity to observe trauma cue-induced craving if it is present. Clinically, our results may represent meaningful considerations for understanding comorbid PTSD-SUD. Indeed, while our moderator analyses should be interpreted cautiously given the lack of variability in the included studies, further work should determine if individuals with more severe PTSD and/or SUD symptoms experience particularly strong trauma cue-elicited craving. For individuals with elevated symptom burden, trauma reminders may function as more potent cues for relapse (Romero-Sanchiz et al., 2022). Future researchers may also wish to examine clinically relevant outcomes that might moderate trauma cue-elicited craving and represent modifiable intervention targets, such as emotional regulation (Tull et al., 2018), coping motives (Atasoy et al., 2023) or distress tolerance (Vujanovic et al., 2018). Finally, our examinations of mediators and moderators were conducted separately in the present review, however, future work could integrate these approaches using moderated mediation models to test whether individual differences influence specific pathways linking trauma exposure to craving. For example, perhaps trauma cues lead to greater anxiety (and in turn, greater craving) for those with more severe PTSD symptoms (moderation on the a path by PTSD symptoms); or that trauma cue-induced anxiety may increase trauma cue induced craving more so for those with greater coping motives for substance use (moderation on the b path by coping motives).
In summary, our meta-analysis is the first to quantify cue reactivity effects among trauma-exposed individuals who use substances. Results suggest that trauma cues elicit greater levels of substance craving compared to neutral cues, substance cues, and pre-cue baseline levels, respectively. Moreover, our mediational analysis demonstrated that the increase in substance craving following trauma (vs. neutral) CRP exposure is indirect, mediate through greater trauma (vs. neutral) cue-induced negative affect. The present meta-analysis yields trauma cue-elicited craving effect sizes that may prove useful for researchers who wish to compare the size of their study effects to those in the extant trauma cue reactivity literature. Moreover, our findings provide support for Baker et al.’s (2004) affective processing model as applied to PTSD-SUD comorbidity, which suggests that trauma cue-induced negative affect is a primary driver of substance craving in response to trauma cues – a process which may maintain substance use in this population.
Open scholarship
This article has earned the Center for Open Science badges for Open Data and Preregistered. The data and materials are openly accessible at osf.io/5pb7v and https://www.crd.york.ac.uk/PROSPERO/view/CRD42022304580.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
The data supporting the findings of this study are available on Open Science Framework (osf.io/5pb7v).