The Association Between Recreational Cannabis Use and Posttraumatic Stress Disorder: A Systematic Review and Methodological Critique of the Literature

Abstract

Background:

Given recent changes in the legal status of cannabis, the risks and benefits associated with its use have become an important public health topic. A growing body of research has demonstrated that posttraumatic stress disorder (PTSD) and recreational cannabis use (RCU) frequently co-occur, yet findings are inconsistent (e.g., direction of effect) and methodological variability makes comparison across studies difficult.

Methods:

We conducted a comprehensive systematic review of all studies (N = 45) published before May 2020 regarding etiologic models of co-occurring RCU and PTSD, as well as provided a methodological critique to inform suggestions for future research initiatives.

Results:

Findings indicate that a majority of studies (n = 37) demonstrated a significant association between RCU and PTSD. Findings provide evidence for the self-medication and high-risk models posited to explain co-occurring RCU and PTSD despite variability in assessment of RCU, which includes commonly used non-standardized self-report questions.

Conclusion:

The association between RCU and PTSD is likely bidirectional. Results inform clinicians and researchers working in the mental health and cannabis use fields how the variability in findings on the association between RCU and PTSD may be attributable, in part, to methodological issues that permeate the extant literature pertaining to RCU and PTSD.

1. Introduction

In the United States (U.S.), the National Comorbidity Survey – Replication (NCS-R) from 2001–03 found that 42.5% of the general adult population had engaged in lifetime recreational cannabis use (RCU; cannabis used without medical justification) and 6.8% met the Diagnostic and Statistical Manual of Mental Disorders, 4^th Edition (DSM-IV) diagnostic criteria for posttraumatic stress disorder (PTSD; Cougle et al., 2011). The prevalence rates for current (i.e., past 12-months) RCU and PTSD were 9.5% and 3.6%, respectively. Additionally, the NCS-R found significant associations between lifetime and current PTSD and lifetime, current, and daily RCU (Cougle et al., 2011). Specifically, the prevalence of lifetime, current, and daily RCU among those meeting criteria for current PTSD was 65.1%, 14.0%, and 3.1%, respectively (Cougle et al., 2011). Generally, individuals with PTSD and co-occurring substance use may have a poorer quality of life compared to those without co-occurring substance use (Watson et al., 2011). Further, co-occurring substance use might negatively interfere with the treatment and management of PTSD (McCauley et al., 2012).

To best address treatment and management challenges of co-occurring RCU and PTSD (e.g., higher PTSD treatment dropout, poorer PTSD treatment adherence, worse PTSD treatment outcome; Bedard-Gilligan et al., 2018), we must understand the etiological and maintenance factors contributing to their high co-occurrence. There are several models that have been proposed to explain co-occurring substance use disorders (SUDs) and PTSD broadly (Berenz et al., 2019), but have been applied to co-occurring RCU and PTSD. The shared liability model suggests that RCU and PTSD frequently co-occur due to common familial risk (i.e., genetic factors, shared environmental influences; Krueger & Markon, 2006). The self-medication model suggests that individuals with PTSD are more likely to use cannabis because it serves as a means to cope with trauma-related symptoms (Khantzian, 1997). It has been hypothesized that PTSD may increase risk of RCU through using cannabis to cope with distressing PTSD symptoms (Sarvet et al., 2018). The high-risk model suggests that RCU increases risk for trauma exposure and, consequently, PTSD (Brady et al., 2004). RCU may increase risk of trauma exposure and PTSD by impairing one’s judgment and ability to discern danger cues (Karila et al., 2014), which could increase participation in risky behaviors and/or unsafe situations. Additionally, substance using individuals are more likely to associate with peers who engage in risk-taking behaviors (Fergusson et al., 2002), which may increase risk for trauma exposure and PTSD. Previous research utilizing these three models as a framework to examine the relationship between RCU and PTSD has yielded mixed findings (Cougle et al., 2011; Ehlers et al., 2016; Holliday & Pedersen, 2017; Welsh et al., 2017). Specifically, Holliday and Pedersen (2017) did not find a significant bivariate correlation between PTSD and cannabis misuse; Ehlers and colleagues (2016) did not find support for the high-risk model; Cougle and colleagues (2011) found support for the self-medication model, but Welsh and colleagues (2017) did not. A number of factors likely contribute to discordant results including methodological variability.

1.1. Methodological concerns

The term “cannabis use” has evolved over the past few decades with the literature using “use,” “misuse,” and “abuse” interchangeably as it progressed to the current terms of “RCU,” which includes all use without a prescription (and is the construct of interest in this review), and “medicinal cannabis use,” or prescribed cannabis use. Reasons for specifically engaging in RCU rather than another substance may be associated with differing means of legal status, access, misperceptions of safety, as well as social (e.g., to be more sociable), enhancement (e.g., to get high), and coping motives (Buckner et al., 2019).

Given the perceived safety and social acceptability of RCU (Berg et al., 2015), combined with the increasing number of states with medicinal cannabis, individuals are more likely to have greater access to RCU as a means of coping with PTSD. Turna and colleagues (2017) noted that anxiety and related disorders (i.e., PTSD) are the most common mental conditions affecting U.S. citizens. Notably, given the mixed findings regarding the developmental timeline of co-occurring RCU and PTSD, it is just as likely that due to the perceived safety and social acceptability of RCU, as well as the availability of medical cannabis, the prevalence of RCU may be higher which may put individuals at increased risk for experiencing traumatic events and developing PTSD.

Researchers have noted a number of methodological concerns within the growing RCU literature, including variability in what investigators consider problematic cannabis use or “misuse” (Asbridge et al., 2014). The definition provided by the NSDUH states that any non-medicinal use (i.e., recreational use) within an individual’s lifetime is considered cannabis “misuse” (Substance Abuse and Mental Health Service Administration, 2019). Not only does the focus on recreational use with the NSDUH definition complicate our understanding of temporal relationships between RCU and other variables (e.g., cannot determine whether PTSD precedes RCU per the self-medication model, or whether RCU precedes trauma exposure/PTSD per the high-risk model), any use may hardly capture problematic use worthy of clinical attention. On the opposite end of the RCU severity spectrum, DSM-5 criteria for cannabis use disorder (CUD) are thought to more accurately identify those with problematic use. Unlike RCU, changes from DSM-IV to DSM-5 PTSD did not lead to as much variability in measurement of PTSD (American Psychiatric Association, 2013). Despite these major revisions, results from the National Stressful Events Survey documented similar estimates of DSM-IV lifetime PTSD (10.6%) and DSM-5 lifetime PTSD (9.4%) among a representative group of U.S. adults (Kilpatrick et al., 2013).

Lack of comprehensive assessment of PTSD could contribute to the variability identified within the PTSD and RCU literature. Specifically, inconsistencies could be attributed to restricted consideration of time of onset. When assessing PTSD, it is important to assess time of onset of symptoms and/or the course of the diagnosis in order to help establish a timeline of how co-occurring PTSD and RCU developed. The bidirectional association between PTSD and RCU calls for a comprehensive PTSD assessment, which includes time of onset of symptoms/diagnosis, in order to establish direction of effect.

Given the predominant models of RCU and PTSD co-occurrence are phenotypic predictive models (i.e., PTSD → RCU, RCU → PTSD), longitudinal studies are preferred over cross-sectional studies. Specifically, the goal of this line of research is estimating a direction of effect of one phenotype onto the other. Despite the limitation that no causal inferences can be determined from cross-sectional and observational data, a majority of the literature examining RCU and PTSD is cross-sectional and observational due to ethical dilemmas surrounding investigating the development of trauma exposure/PTSD (Newman et al., 2006).

1.2. Aims

Given the increasing public attention and empirical examination of RCU, focus has been placed on identifying predictors of use, including PTSD. To date, no systematic review of the literature has been initiated examining the intersection of RCU and PTSD. The present systematic review aims to evaluate and summarize the empirical evidence regarding the association between RCU and PTSD, as well as examine the support for each model posited to explain the association as a means of determining what gaps exist in the literature. A secondary aim is to provide a critique of the methodological approaches used to examine co-occuring RCU and PTSD. It is hoped that the present review and methodological critique will aid researchers and clinicians alike as they attempt to create and implement research protocols and interpret research findings regarding the association between RCU and PTSD.

2. Methods

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were enlisted during the review process (Moher et al., 2009).

2.1. Databases & search methodology

Two databases (PubMed and PsycINFO) were searched in May 2020 using the following search terms aimed at capturing literature within the intersection of RCU and PTSD: (([cannabis] OR [cannabis use disorder]) OR (“cannabis” OR “cannabis use” OR “cannabis use disorder” OR “marijuana” OR “marijuana use” OR “marijuana use disorder” OR “cannabis abuse” OR “cannabis dependence” OR “marijuana abuse” OR “marijuana dependence”)) AND ([Posttraumatic Stress Disorder] OR (“posttraumatic stress disorder” OR “PTSD”)).

2.2. Inclusion and exclusion criteria

Given the early stage of the literature regarding RCU and PTSD and the secondary aim of evaluating research methodology, attempts were made to balance capturing as much of the literature as possible while maintaining strict inclusion criteria to improve comparability across studies. Thus, prior to beginning the initial search, the following inclusion and exclusion criteria were determined: (1) PTSD (i.e., diagnosis and/or symptoms) must be included in primary analyses related to RCU, (2) RCU must be differentiated from other forms of substance use (i.e., no polysubstance use including cannabis) or describe attempts made to identify RCU specifically, (3) the study cannot evaluate a form of cannabis without THC (e.g., CBD oil), (4) the study cannot be focused on medicinal cannabis use (i.e., prescribed cannabis use), (5) the study cannot be a conceptual article, a review, or qualitative in nature, (6) the study cannot evaluate the effects of an intervention, (7) must be peer reviewed, (8) must use human subjects, and (9) must be in English.

2.3. Identification

During the first stage of the study selection process, searches conducted in PubMed and PsycINFO resulted in 330 and 179 articles being identified, respectively. This resulted in 365 potentially relevant studies after excluding duplicates.

The second stage of the study selection process involved examining each of the 365 articles and excluding those that did not meet the inclusion and exclusion criteria. 320 studies met one or more of these criteria and were excluded from further consideration. This process resulted in a total of 45 relevant studies for inclusion in the present review. Figure 1 presents the flow chart for the selection of the included studies. Two authors (T.A.H and A.J.Z) independently coded each article based on the specific inclusion and exclusion criteria previously mentioned. The authors found no discrepancies during the independent coding process (i.e., 100% agreement).

Figure 1:

Flowchart of systematic review

2.4. Self-medication, high-risk, and shared liability models categorization

Study categorization was based on predictor and outcome variables since the self-medication and high-risk models assume temporal order. If the predictor was PTSD symptoms or diagnosis and the outcome was RCU or CUD symptoms or diagnosis, then the study was classified as testing the self-medication model. If the predictor was RCU or CUD symptoms or diagnosis and the outcome was PTSD symptoms or diagnosis, then the study was classified as testing the high-risk model. If the study was a twin/family study or included genetically-informed shared risk, then it was classified as testing the shared liability model.

3. Results

The results section is separated thematically, and key results are summarized in Figure 2. Basic relations are briefly reviewed, and more weight is applied to the sections pertaining to the methodological quality of the current literature at the intersection of RCU and PTSD, which will provide the basis for further discussion and empirical recommendations in the discussion.

Figure 2:

Infographic displaying results of systematic review of the association between recreational cannabis use (RCU) and posttraumatic stress disorder (PTSD)

3.1. Relationship between RCU and PTSD

Table 1 presents information regarding articles extracted for the present review, specifically, article title, authors, sample information, study design, RCU assessment, PTSD assessment, etiologic model of co-occurrence tested, key findings, and standardized effect size. Thirty-seven out of 45 (82.2%) studies demonstrated significant associations between RCU and PTSD. Eight out of the 45 (17.8%) studies did not demonstrate significant associations between RCU and PTSD.

Table 1.

Study information from the articles included in the present systematic review.

Study Title	Author	Sample	Population	Study Design	RCU Measurement	PTSD Measurement	Model of Comorbidity	Results	Cohen’s d or β
Trends in documented cannabis use disorder among hospitalized adult epilepsy patients in the U.S.	Lekoubou et al., 2020	1,980,707 hospitalized adults with epilepsy	Clinical (adults)	Cross-sectional	ICD-9-CM diagnostic codes for CUDs (lifetime diagnosis excluding those “in remission”)	ICD-9-CM diagnostic codes for PTSD (lifetime diagnosis)	Self-Medication Model	PTSD did not have a significant association with CUD (OR = 1.13; CI: .99–1.31; p = .08)	Cohen’s d = .07
Post-traumatic stress and marijuana outcomes: The mediating role of marijuana protective behavioral strategies	Jordan et al., 2020	7,307 U.S. college students	College students	Cross-sectional	Self-report: MUG; CUDIT-R (past month cannabis use frequency, quantity; past six-month CUD symptoms)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	There was not a significant effect between PTSD symptoms and cannabis use frequency (β = .02) or quantity (β = .04). There was a significant positive direct effect between PTSD symptoms and CUD symptoms (β = .17)	β = .02, .04, .17
PTSD and associated factors among drivers surviving road traffic crashes in SW Ethiopia	Alenko et al., 2019	398 public transportation workers	Ethiopian community	Cross-sectional	Structured or semi-structured interview: WHO’s ASSIST (past 3-month risk problems from current cannabis use)	Self-report: TSQ (past week PTSD symptoms)	High-Risk Model	Individuals with severe-risk cannabis use were at greater risk for developing PTSD (AOR = 2.51, CI: 1.96–7.52, p = .03)	Cohen’s d = .51
Substance use disorders in refugee and migrant groups in Sweden: A nationwide cohort study of 1.2 million people	Harris et al., 2019	1,241,901 refugees, migrants, & Swedish-born adults	Swedish adults	Cross-sectional	ICD-10 diagnostic codes for CUDs (lifetime diagnosis)	ICD-10 diagnostic codes for PTSD (lifetime diagnosis)	Self-Medication Model	PTSD was associated with CUDs (LRT χ2 (2) = 11.7; p = .003). PTSD was associated with CUDs in the Swedish-born population (HR = 11.50; CI: 9.30–14.23; p < .001) than non-refugee (HR: 5.92; CI: 3.58–9.81; p < .001) or refugee migrants (HR: 2.60; CI: 0.63–10.69; p = .19)	Cohen’s d = .01, 1.35, .98, .53
Longitudinal associations between sleep, intrusive thoughts, and alcohol problems among veterans	Miller et al, 2019	361 veterans	Veterans / military personnel	Longitudinal	Self-report: TLFB (past 6-month cannabis use quantity and frequency)	Self-report: IDAS (traumatic intrusions subscale) [past two-weeks PTSD symptoms]	High-Risk Model	Correlations for cannabis use at baseline with PTSD symptoms at baseline, 6 months, and 12 months were correlated with r = .23, .18, and .14, respectively (p < .05)	Cohen’s d = .47, .37, .28
Interactive effects of PTSD and substance use on suicidal ideation and behavior in military personnel: Increased risk from marijuana use	Allan et al., 2019	545 military personnel	Veterans / military personnel	Longitudinal	Structured or semi-structured interview: ASI (past month cannabis use)	Self-report: PCL (past month PTSD symptoms)	High-Risk Model	At high (β = .02; p = .01), but not low (β = .00; p = .96) levels of PTSD symptoms, more days using cannabis predicted increased PTSD symptoms over time	β = .02, .00
Associations of PTSD, chronic pain, and their comorbidity on CUD: Results from an American nationally representative study	Bilevicius et al., 2019	36,309 U.S. adults	U.S. adults	Cross-sectional	Structured or semi-structured interview: AUDADIS (CUD module) [past year CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD]	Self-Medication Model	PTSD+digestive pain (AOR = 6.13, CI: 3.79–9.92; p < .001), PTSD+nerve pain (AOR = 4.21, CI: 2.51–7.07; p < .001), and PTSD+any chronic pain (AOR = 5.08, CI: 3.30–7.83; p < .001) were associated with greater odds of CUD compared to pain alone	Cohen’s d = 1.00, .79, .90
PTSD is a risk factor for multiple addictions in police officers hospitalized for alcohol	Brunault et al., 2019	133 police officers in rehab	First responders	Cross-sectional	Self-report: CAST (past year cannabis use)	Self-report: PCL (past month provisional PTSD diagnosis)	Self-Medication Model	PTSD diagnosis was not a significant predictor of CUD severity (p = 0.82)	Cohen’s d = .12
PTSD symptom severity, cannabis, and gender: A zero-inflated negative binomial regression model	Rehder & Bowen, 2019	536 U.S. college students	College students	Cross-sectional	Self-report: Frequency of cannabis use was assessed via report of the number of days used over the past 60 days	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	The probability of using cannabis was moderated by gender (OR = 0.96, CI: 0.93–0.99, p = .026), such that for males, as PTSD symptom severity increased, likelihood of using cannabis increased	Cohen’s d = −.02
An EMA study examining PTSD symptoms, prenatal bonding, and substance use among pregnant women	Sanjuan et al., 2019	33 trauma-exposed pregnant women	Trauma-exposed adult women	Longitudinal	Self-report: Any cannabis use was assessed via participant report of any use since the previous EMA session (i.e., 24 hours)	Structured or semi-structured interview: CAPS; Self-report: PCL (lifetime and current PTSD; daily PTSD symptoms)	Self-Medication Model	There were moderate associations between greater daily peak PTSD symptoms and cannabis use (β = 0.35, CI: .10–.57) within-subjects	β = .35
Sexual minority women and cannabis use: The serial impact of PTSD symptom severity and coping motives	Walukevich-Dienst et al., 2019	439 Louisiana college students	College students	Cross-sectional	Self-report: MUF (lifetime cannabis use frequency)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	PTSD symptoms did not significantly predict cannabis use frequency (β = −.01; p = .06)	β = −.01
Comorbidity in illness-induced PTSD versus PTSD due to external events in a nationally representative study	Sommer et al., 2018	36,309 U.S. adults	U.S. adults	Cross-sectional	Structured or semi-structured interview: AUDADIS (CUD module) [past year CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD]	Self-Medication Model	Individuals with illness-induced PTSD had significantly greater odds of CUD compared to those with other trauma-induced PTSD (AOR = 1.98, CI: 1.01–3.89, p < .05)	Cohen’s d = .38
Use of protective behavioral strategies among young adult veteran marijuana users	Pedersen et al., 2018	180 veterans	Veterans / military personnel	Cross-sectional	Self-report: Frequency of cannabis use was assessed via report of the number of days used over the past six months; MACQ	Self-report: PCL (past month provisional PTSD)	Self-Medication Model	Participants who screened positive for PTSD reported more cannabis use problems (M = 2.54, SD = 3.13) than participants not screening positive for PTSD (M = 1.54, SD = 2.56), t (176) = 2.20, p = .03)	Cohen’s d = .35
Regular past year cannabis use in women veterans and associations with sexual trauma	Browne et al., 2018	647 women veterans	Veterans / military personnel	Cross-sectional	Self-report: Frequency of cannabis use was assessed via “how often in the past year did you use cannabis” with 7 response options from “not at all” to “every day”	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	Current PTSD symptoms was a significant risk factor for cannabis use above and beyond covariates (OR = 1.02, CI: 1.00–1.04, p < .05)	Cohen’s d = .01
The impact of PTSD clusters on cannabis use in a racially diverse trauma-exposed sample: An analysis from EMA	Buckner et al., 2018	87 trauma-exposed U.S. adults	Trauma-exposed adults	Cross-sectional	Self-report: MPS; TLFB (past 90-days cannabis problems and use frequency)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	Participants with PTSD symptoms reported more cannabis-related problems (χ2 = 27.15, p < .001) than participants without PTSD symptom	Cohen’s d = 1.22
Self- and other-directed forms of violence and their relationship with lifetime DSM-5 psychiatric disorders: Results from the NESARC− III	Harford et al., 2018	36,309 U.S. adults	U.S. adults	Cross-sectional	Structured or semi-structured interview: AUDADIS (CUD module) [lifetime CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD]	Self-Medication Model	Individuals with PTSD had increased risk of also having CUD (OR = 3.6, p < .01)	Cohen’s d = .71
Trajectories of cannabis use beginning in adolescence associated with symptoms of PTSD in the mid-thirties	Lee et al., 2018	674 racial and ethnic minority adults	Racial and ethnic minority adults	Longitudinal	Self-report: Frequency of cannabis use was assessed via “how often have you used cannabis” and “how often have you used cannabis in the past 5 years” with 5 response options from “never” to “once a week or more”	Self-report: PCL (past month PTSD symptoms)	High-Risk Model	The chronic cannabis use group (AOR = 4.68, p < .01), the late quitting group (AOR = 6.18, p < .01), and the moderate cannabis use group (AOR = 3.97, p < .01) were all associated with an increased likelihood of having PTSD symptoms at age 36 compared with the no cannabis use group	Cohen’s d = .85, 1.00, .76
DSM-5 CUD in the National Epidemiologic Survey on Alcohol and Related Conditions-III: gender-specific profiles	Kerridge et al., 2018	36,309 U.S. adults	U.S. adults	Cross-sectional	Structured or semi-structured interview: AUDADIS (CUD module) [past year CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD]	High-Risk Model	Any (AOR = 1.7, CI: 1.12 – 2.57, p < .05), moderate (AOR = 2.0, CI: 1.08 – 3.81, p < .05), and severe CUD (AOR = 3.7, CI: 1.98 – 7.02, p < .05) among men was associated with PTSD. Any (AOR = 1.6, CI: 1.01 – 2.48, p < .05) and moderate CUD (AOR = 3.4, CI: 1.69 – 6.98, p < .05) was associated with PTSD among women	Cohen’s d = .29, .38, .72, .26, .68
Daily-level associations between PTSD and cannabis use among young sexual minority women	Dworkin et al., 2017	114 sexual minority women	Sexual minority adult women	Longitudinal	Self-report: Any cannabis use was assessed via participant report of any use during the past 24 hours	Self-report: PCL (past 24-hours PTSD symptoms)	Self-Medication Model	Women with higher overall daily PTSD scores across time had a higher likelihood of using cannabis on any given day (OR = 2.67, CI: 1.71–4.41, p < .001)	Cohen’s d = .54
Association between substance use diagnoses and psychiatric disorders in an adolescent and young adult clinic-based population	Welsh et al., 2017	483 adolescents and young adults	Adolescents and young adults	Cross-sectional	DSM-IV-TR diagnostic codes for CUDs (lifetime diagnosis)	DSM-IV-TR diagnostic codes for PTSD (lifetime diagnosis)	Self-Medication Model	PTSD and CUD were not significantly associated (AOR = 1.52, CI: .49 – 4.69, p = .47)	Cohen’s d = .23
The mediating roles of coping, sleep, and anxiety motives in cannabis use and problems among returning veterans with PTSD and MDD	Metrik et al., 2016	301 veterans	Veterans / military personnel	Cross-sectional	Self-report: MPS; TLFB; Structured or semi-structured interview: SCID (lifetime and past year CUD; past 90-days cannabis use problems and past 6 months cannabis use frequency)	Structured or semi-structured interview: CAPS (lifetime and past month PTSD)	Self-Medication Model	PTSD was associated with greater cannabis use (OR = 1.75, CI: 1.22, 2.52, p < .01). PTSD predicted the likelihood of being diagnosed with a CUD (OR = 1.71, CI: 1.14, 2.59, p ≤ .01). Individuals with PTSD had more cannabis use problems compared to individuals without PTSD (χ2 = 2.14, p < .05)	Cohen’s d = .31, .30, .32
Lifetime history of traumatic events in a young adult Mexican American sample: Relation to substance dependence, affective disorder, acculturation stress, and PTSD	Ehlers et al., 2016	614 Mexican-American adults	U.S. Mexican-American adults	Cross-sectional	Structured or semi-structured interview: SSAGA (CUDs module) (lifetime diagnosis)	Structured or semi-structured interview: SSAGA (PTSD module) [lifetime diagnosis]	High-Risk Model	Cannabis dependence did not significantly co-occur with PTSD (OR = 2.02, CI: 1.12–3.62, p = .02; p = .01 significance level)	Cohen’s d = .39
Prevalence and correlates of cannabis use in an outpatient VA PTSD clinic	Gentes et al., 2016	719 veterans	Veterans / military personnel	Cross-sectional	Self-report: Frequency of cannabis use was assessed via “how often have you used cannabis in the past 6 months” with 7 response options from “no use” to “daily use”	Structured or semi-structured interview: CAPS (past month PTSD symptoms)	Self-Medication Model	Past 6-month cannabis use was associated with PTSD severity (AOR = 1.30, CI: 1.01–1.66, p < .05)	Cohen’s d = .15
Associations of PTSD symptoms with marijuana and synthetic cannabis use among young adult US Veterans: a pilot investigation	Grant et al., 2016	790 veterans	Veterans / military personnel	Cross-sectional	Self-report: Any and frequency of cannabis use was assessed via report of any lifetime cannabis use and number of days used during the past month, respectively	Self-report: PC-PTSD (past month PTSD symptoms)	High-Risk Model	Cannabis use in the past month was associated with a positive screen for PTSD (OR = 4.19, CI: 2.44, 7.18)	Cohen’s d = .79
Prevalence and mental health correlates of illegal cannabis use among bisexual women	Robinson et al., 2016	262 Canadian bisexual women	Sexual minority adult women	Cross-sectional	Self-report: DUDIT-E (past year cannabis use)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	There was no significant association between cannabis use and PTSD	Cohen’s d = .02, .33
Cannabis use among Navy personnel in Sri Lanka: a cross sectional study	de Silva et al., 2016	671 Navy personnel	Veterans / military personnel	Cross-sectional	Self-report: Any cannabis use was assessed via report of any use during the past 12 months	Self-report: PCL (past month PTSD symptoms)	High-Risk Model	There was a significant association between cannabis use and PTSD (AOR = 4.20, CI: 1.08–16.38)	Cohen’s d = .79
Mental health symptom severity in cannabis using and non-using Veterans with probable PTSD	Johnson et al., 2016	700 veterans	Veterans / military personnel	Cross-sectional	Structured or semi-structured interview: WHO’s ASSIST (past 3-month risk of problems from current cannabis use)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	There was no association between PTSD scores and frequency of cannabis use (OR = 0.99, CI: 0.97–1.01, p = 0.39)	Cohen’s d = −.01
Associations among trauma, PTSD, cannabis use, and CUD in a nationally representative epidemiologic sample	Kevorkian et al., 2015	34,396 U.S. adults	U.S. adults	Cross-sectional	Structured or semi-structured interview: AUDADIS (lifetime cannabis use and CUD modules) [past year CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD]	Self-Medication Model	PTSD showed a significant association with CUD (OR = 1.22, CI: 1.21–1.22, p < .001), but was only marginally associated with lifetime cannabis use (OR = 0.99, CI: .99–.99, p < .001)	Cohen’s d = .11, −.01
The moderating role of experiential avoidance in the relationship between PTSD symptom severity and cannabis dependence	Bordieri et al., 2014	123 adults with substance use disorders	U.S. adults	Cross-sectional	Structured or semi-structured interview: SCID (current diagnosis)	Structured or semi-structured interview: CAPS (past month PTSD symptoms)	Self-Medication Model	There was a conditional relationship between PTSD symptom severity and cannabis dependence when experiential avoidance scores were one SD above the mean (β = 0.05, z = 3.07, p = .002), and at the mean (β = 0.03, z = 2.50, p = .01)	β = .05, .03
Marijuana, expectancies, and post-traumatic stress symptoms: A preliminary investigation	Earleywine & Bolles, 2014	653 veterans	Veterans / military personnel	Cross-sectional	Self-report: Quantity of cannabis use was assessed via report of “how much marijuana do you use in a month” with 14 response options from “less than ¼ ounce” to “more than 3 ounces”	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	There was a significant indirect effect of PTSD symptoms on cannabis use via PTSD expectancies (β = 0.04, p < .01)	β = .04
Negative cognitions as a moderator in the relationship between PTSD and substance use in a psychiatrically hospitalized adolescent sample	Allwood et al., 2014	188 U.S. adolescents	Clinical (adolescents)	Cross-sectional	Structured or semi-structured interview: K-SADS-PL (SUD module) [lifetime SUD symptoms including age of onset]	Structured or semi-structured interview: K-SADS-PL (PTSD module) [lifetime PTSD including age of onset]	Self-Medication Model	PTSD was significantly associated with CUD symptoms (β = .73, p < .001) at high (+1 SD) negative view of the future. Average age of onset of PTSD was 10.65 years (SD = 3.47). Average of onset of SUD was 14.67 years (SD = 1.43)	β = .73
Patterns of drug and alcohol use associated with lifetime sexual revictimization and current PTSD among three national samples of adolescent, college, and household-residing women	Walsh et al., 2014	1,763 girls; 2,000 college women; and 3,001 household-residing women	Adolescents and young adults	Cross-sectional	Self-report: Non-experimental cannabis use was assessed via participant report of cannabis use at least four times during the past year	Structured diagnostic interview with yes/no responses indicating the presence of DSM-IV PTSD symptoms during the past 6 months	Self-Medication Model	For adolescents, past 6-month PTSD was associated with increased odds for cannabis use (OR = 2.50, CI: 1.20–4.80). For household women, past 6-month PTSD was associated with increased odds for cannabis use (OR = 3.50, CI: 2.24–5.45)	Cohen’s d = .51, .69
Trauma exposure, PTSD and risk for alcohol, nicotine, and marijuana dependence in Israel	Walsh et al., 2014	1,317 Jewish household residents	Jewish community	Cross-sectional	Structured or semi-structured interview: AUDADIS (CUD module) [past year CUD]	Structured or semi-structured interview: AUDADIS (PTSD module) [past year PTSD symptoms and diagnosis]	Self-Medication Model	PTSD symptoms were associated with increased odds of cannabis dependence (OR =1.1, CI: 1.04–1.2). A PTSD diagnosis was associated with increased odds of cannabis dependence (OR = 2.6, CI: 1.2–5.9)	Cohen’s d = .05, .53
High school students’ posttraumatic symptoms, substance abuse and involvement in violence in the aftermath of war	Schiff et al., 2012	4,151 U.S. high school students	U.S. adolescents	Cross-sectional	Self-report: Any cannabis use was assessed via report of any lifetime cannabis use	Self-report: UCLA PTSD-RI (past month PTSD symptoms)	Self-Medication Model	PTSD symptoms increased odds of lifetime cannabis use (OR = 1.04, CI: 1.03–1.05, p < .001)	Cohen’s d = .02
PTSD and cannabis use in a nationally representative sample	Cougle et al., 2011	5,672 U.S. adults	U.S. adults	Cross-sectional	Self-report: Any and frequency of cannabis use was assessed via report of lifetime cannabis use and past 12 months use with 6 response options from “none” to “nearly every day”, respectively	Structured or semi-structured interview: WHO’s WMH-CIDI (lifetime and past month PTSD)	Self-Medication Model	Lifetime PTSD was associated with lifetime (OR = 2.4, CI: 1.7 – 3.5, p < .01), current (OR = 1.4, CI: 1.0 – 2.1, p < .01), and daily cannabis use (OR = 1.9, CI: 1.1 – 3.2, p < .05). Current PTSD was related to lifetime (OR = 2.4, CI: 1.6 – 3.5, p < .01), current (OR = 2.0, CI: 1.4 – 3.1, p < .01), and daily cannabis use (OR = 2.1, CI: 1.1 – 3.9, p < .05).	Cohen’s d = .48, .19, .35, .48, .38, .41
Interpersonal victimization, PTSD, and change in adolescent substance use prevalence over a ten-year period	McCart et al., 2011	7,329 U.S. adolescents	U.S. adolescents	Longitudinal	Self-report: Non-experimental cannabis use was assessed via participant report of cannabis use at least four times during their lifetime	Structured or semi-structured interview: National Women’s Study PTSD module (lifetime PTSD)	Self-Medication Model	PTSD was significantly associated with both lifetime cannabis use (OR = 3.06, CI: 1.62 – 5.81, p < .001) and lifetime non-experimental cannabis use (OR = 3.53, CI: 1.72 – 7.22, p < .001) over the 10-year period	Cohen’s d = .62, .70
The relationship between substance use and PTSD in a methadone maintenance treatment program	Villagonzalo et al., 2011	80 individuals in a methadone maintenance treatment program	Clinical (adults)	Cross-sectional	Self-report: Frequency and quantity of cannabis use was assessed via report of how often used with 5 response options from “only used it once” to “every day”, and how much used per session with 5 response options from “less than .1 g” to “more than 4.0 g” (lifetime cannabis use)	Self-report: PCL (past month PTSD symptoms)	Self-Medication Model	Severity of cannabis use was associated with total PTSD symptoms (F2,65 = 3.71, p < .05)	Cohen’s d = .44
Association between psychopathology and substance use among school-going adolescents in Cape Town, South Africa	Saban et al., 2010	939 South African adolescents	South African adolescents	Cross-sectional	Self-report: Any cannabis use was assessed via report of any lifetime cannabis use	Self-report: HTQ (lifetime PTSD)	Self-Medication Model	Grade 11 students had a 2.3% increased risk of cannabis use with every one-point increase in the PTSD total score	Cohen’s d = .01
Substance use, childhood traumatic experience, and PTSD in an urban civilian population	Khoury et al., 2010	587 urban primary care patients	Urban community	Cross-sectional	Self-report: KMSK (lifetime CUD)	Self-report: PSS (past 2-week PTSD symptoms)	Self-Medication Model	Lifetime cannabis dependence was associated with PTSD symptoms (F = 10.12, p <.01) and all PTSD symptoms (Intrusive: F = 4.16, p <.05; Avoidance: F = 11.25, p <.01; and Hyperarousal: F = 7.72, p <.01).	Cohen’s d = .26, .17, .28, .23
Identifying risk factors for marijuana use among veterans affairs patients	Goldman et al., 2010	5,512 veterans	Veterans / military personnel	Cross-sectional	Self-report: Any cannabis use was assessed via report of any use during the past year	Structured or semi-structured interview: MINI (current PTSD)	Self-Medication Model	PTSD increased odds of past year cannabis use (OR = 1.64, CI: 1.4–2.0)	Cohen’s d = .27
Impact of Hurricane Rita on adolescent substance use	Rohrbach et al., 2009	280 U.S. adolescents	U.S. adolescents	Longitudinal	Self-report: Frequency of cannabis use was assessed via report of how often used over past 30 days with 8 response options from “zero” to “more than 100 times”	Self-report: UCLA PTSD-RI (past month PTSD symptoms)	Self-Medication Model	The PTSD severity score was positively associated with increases in cannabis use at 7-month follow up (β = .11, SE = .06, p < .05)	β = .11
PTSD contributes to teen and young adult CUDs	Cornelius et al., 2010	693 U.S. teens and young adults	U.S. adolescents and young adults	Cross-sectional	Structured or semi-structured interview: SCID (lifetime CUD)	Structured or semi-structured interview: SCID (lifetime PTSD)	Self-Medication Model	PTSD is directly associated with CUD (β = 0.19, z = 4.40, p < 0.001)	β = .19
Exposure to the tsunami disaster, PTSD symptoms and increased substance use–an Internet based survey of male and female residents of Switzerland	Vetter et al., 2008	2,921 Switzerland residents after a natural disaster	Switzerland community	Cross-sectional	Self-report: Any increase in cannabis use was assessed via report of any increase in cannabis use since the tsunami	Self-report: PTSS (past few days PTSD symptoms)	Self-Medication Model	PTSD symptoms were significantly associated with increases in cannabis use following the tsunami for both men (OR = 1.15, CI: 1.06 – 1.25) and women (OR = 1.10, CI: 1.07 – 1.14)	Cohen’s d = .08, .05
Consumption of cigarettes, alcohol, and marijuana among New York City residents six months after the September 11 terrorist attacks	Vlahov et al., 2004	1,570 New York residents after 9/11 terrorist attacks	New York community	Longitudinal	Self-report: Frequency of cannabis use was assessed via participant report of the number of times they used cannabis during the past week	Structured or semi-structured interview: National Women’s Study PTSD Module (lifetime PTSD)	Self-Medication Model	Rates of PTSD were statistically similar for individuals who increased cannabis use vs. those who did not (4.0% vs. 1.4%, p = 0.25)	Cohen’s d = .06
PTSD and substance use in inner-city adolescent girls	Lipschitz et al., 2003	54 inner-city adolescent girls	U.S. Urban adolescents	Cross-sectional	Structured or semi-structured interview: K-SADS-PL (CUD module) [lifetime CUD symptoms including age of onset]	Self-report: PCL (past month provisional PTSD including age of onset)	Self-Medication Model	PTSD increased likelihood of using cannabis regularly (χ2 = 9.33, df = 1, p = 0.002). Age of onset of PTSD symptoms ranged from 5–17 years (M = 13.5, SD = 3.2). Age of onset of cannabis abuse ranged from 13–18 years (M = 15.0, SD = 1.5)	Cohen’s d = .71

3.2. Support for etiologic models of co-occurring RCU and PTSD

Thirty-seven of the 45 (82.2%) studies tested the self-medication model, with 30 (81.1%) reporting significant associations of PTSD with RCU (Table 1). Eight of the 45 (17.8%) studies tested the high-risk model, and seven (87.5%) were supportive, finding associations between RCU with PTSD. None of the 45 (0%) studies tested the shared liability model, highlighting a critical gap to fill in this developing literature.

3.3. RCU assessment

As shown in Table 1, the most common assessment technique for RCU was self-report without the use of a questionnaire (i.e., RCU frequency [number of times used over x amount of time], dichotomous RCU [any use, lifetime use, non-experimental use]) and 90% and 100% of the studies that used continuous RCU frequency and dichotomous RCU were supportive based on significant findings, respectively.

The second most common assessment technique for RCU was a structured or semi-structured interview (i.e., Alcohol Use Disorder and Associated Disabilities Interview Schedule [AUDADIS], Structured Clinical Interview for DSM-IV [SCID-IV], World Health Organization’s Alcohol, Smoking, and Substance Involvement Screening Test [ASSIST], Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version [K-SADS-PL], Semi-Structured Assessment for the Genetics of Alcoholism [SSAGA], Addiction Severity Index [ASI]) and 86.7% of the studies that used structured or semi-structured interviews were supportive based on significant findings.

The third most common assessment technique for RCU was a self-report questionnaire (i.e., Timeline Followback [TLFB], Marijuana Problems Scale [MPS], Cannabis Use Disorders Identification Test – Revised [CUDIT-R], Marijuana Use Grid [MUG], Marijuana Use Form [MUF], Marijuana Consequences Questionnaire [MACQ], Kreek-McHugh-Schluger-Kellogg Scale [KMSK], Cannabis Abuse Screening Test [CAST], Drug Use Disorders Identification Test – Extended [DUDIT-E]) and 66.7% of the studies that used self-report questionnaires were supportive based on significant findings.

The least common assessment technique for RCU was diagnostic codes from medical records (i.e., ICD and DSM diagnostic codes), and 33.3% of the studies that used ICD and DSM diagnostic codes were supportive based on significant findings, respectively.

Twenty-three of 27 (85.2%) studies that used only self-report measures were supportive based on significant findings (Table 1). Twelve of 14 (85.7%) studies that enlisted the use of only structured or semi-structured interviews were supportive based on significant findings. One of three (33.3%) studies that used only DSM or ICD diagnostic codes was supportive based on significant findings. The one study that used both self-report and a structured or semi-structured interview was supportive based on significant findings (Metrik et al., 2016).

3.4. PTSD assessment

As shown in Table 1, the most common assessment technique for PTSD was a self-report questionnaire (i.e., PTSD Checklist [PCL], Primary Care PTSD Screen [PC-PTSD], PTSD Symptom Scale [PSS], University of California, Los Angeles PTSD Reaction Index [UCLA PTSD-RI], Trauma Screening Questionnaire [TSQ], Inventory of Depression and Anxiety Symptoms Scale [IDAS], Harvard Trauma Questionnaire [HTQ], Posttraumatic Stress Scale [PTSS]) and all were 100% supportive except studies that used the PCL, which were 76.5% supportive based on significant findings.

The second most common assessment technique for PTSD was a structured or semi-structured interview (i.e., AUDADIS, Clinician-Administered PTSD Scale [CAPS], National Women’s Study PTSD module, SSAGA, K-SADS-PL, World Health Organization’s World Mental Health Composite International Diagnistic Interview [WHO-WMH-CIDI], Mini International Neuropsychiatric Interview [MINI], SCID-IV, unspecified structured diagnostic interview) and 88.9% of studies that used structured or semi-structured interviews were supportive based on significant findings.

The least common assessment technique for PTSD was diagnostic codes from medical records (i.e., ICD and DSM diagnostic codes), and 33.3% of the studies that used ICD and DSM diagnostic codes were supportive based on significant findings, respectively.

Twenty of the 24 (83.3%) studies that used only self-report measures were supportive based on significant findings (Table 1). Fifteen of the 17 (88.2%) studies that enlisted the use of only structured or semi-structured interviews were supportive based on significant findings. One of the three (33.3%) studies that used only DSM or ICD diagnostic codes was supportive based on significant findings. The one study that used both self-report and a structured or semi-structured interview was supportive based on significant findings (Sanjuan et al., 2019).

3.5. Cross-sectional versus longitudinal studies

Thirty-seven of 45 (82.2%) studies used a cross-sectional design to test the self-medication and high-risk models. The remaining eight of 45 (17.8%) studies utilized a longitudinal study design to test the self-medication and high-risk models. Seven of eight (87.5%) longitudinal studies demonstrated significant associations between RCU and PTSD. Both the self-medication and high-risk models assume that causes precede their effects across time, which is difficult to capture with a cross-sectional study design. However, six of 37 (16.2%) cross-sectional studies tried to address the limitation of time by assessing age of onset of RCU and PTSD, or by using lifetime RCU to predict current PTSD or lifetime PTSD to predict current RCU (Alenko et al., 2019; Allwood et al., 2014; Cougle et al., 2011; de Silva et al., 2016; Lipschitz et al., 2003; Metrik et al., 2016). Specifically, 2 studies assessed age of onset of RCU or CUD and PTSD (Allwood et al., 2014; Lipschitz et al., 2003), 2 studies used lifetime PTSD to predict current RCU (Cougle et al., 2011; Metrik et al., 2016), and 2 studies used lifetime RCU or CUD to predict current PTSD (Alenko et al., 2019; de Silva et al., 2016). All six of these cross-sectional studies (100%) demonstrated significant associations between RCU and PTSD. Out of the remaining 31 cross-sectional studies that did not try to establish a temporal relationship between RCU and PTSD, 23 studies (74.2%) demonstrated significant associations between RCU and PTSD.

3.5.1. Self-medication model

Thirty-two of 45 (71.1%) studies used a cross-sectional design to test the self-medication model. Twenty-six of those 32 (81.3%) cross-sectional studies demonstrated significant associations between RCU and PTSD. Five of 45 (11.1%) studies used a longitudinal design to test the self-medication model. Four of those five (80.0%) longitudinal studies demonstrated significant associations between RCU and PTSD.

3.5.2. High-risk model

Five of 45 (11.1%) studies used a cross-sectional design to test the high-risk model. Four of those five (80.0%) cross-sectional studies demonstrated significant associations between RCU and PTSD. Three of 45 (6.7%) studies used a longitudinal design to test the high-risk model, and all 3 demonstrated significant associations between RCU and PTSD.

3.6. Study samples

Significant associations between RCU and PTSD supporting the self-medication and high-risk developmental models were found across multiple study samples, with the exception of first responders. Of the two clinical adult studies, only one reported significant associations between RCU and PTSD (50%). The only study of a clinical adolescent sample found significant associations between RCU and PTSD (100%). There were three studies of college samples, and two found significant associations between RCU and PTSD, while one did not (66.7%). The studies with only adolescents (n=5) all found significant associations between RCU and PTSD (100%). One study of adolescents and young adults combined did not find significant associations between RCU and PTSD, while two studies did find that they were significantly associated (66.7%). All five studies with community samples found significant associations between RCU and PTSD (100%). The one study using a nationwide sample found significant associations between RCU and PTSD (100%). All but one of the 10 of studies with veterans and military personnel found significant associations between RCU and PTSD (90%), as did the one study of women veterans (100%). Seven United States adult survey samples found significant associations between RCU and PTSD, while one study did not (87.5%). Both studies of trauma-exposed individuals found significant associations between RCU and PTSD (100%). The one study investigating a sample of racial and ethnic minority adults found significant associations between RCU and PTSD (100%). One study with women who identified as sexual minoriies found significant associations between RCU and PTSD, and one did not (50%). No studies with first responders found significant associations between RCU and PTSD (0%).

4. Discussion

This review aimed to synthesize the limited RCU and PTSD literature, which is an emerging area of empirical interest given the potential health-related consequences that have been associated with RCU (Cohen et al., 2019). Following review of the studies fitting inclusion criteria (N = 45), the majority of studies (n = 37) demonstrated a significant association between RCU and PTSD. The variability in findings may be attributable, in part, to methodological issues that permeate the extant literature pertaining to RCU and PTSD. Thus, the second aim of this review was to critically examine the literature to identify such concerns. Discussion of these concerns is separated thematically and recommendations regarding how to approach methodological issues are provided within each section.

4.1. RCU and PTSD co-occurrence

Although the majority of studies identified a significant association between RCU and PTSD, eight studies did not. Each study included in the review had at least one methodological limitation that may have created bias, including the eight studies demonstrating non-significant results. Notably, two of the eight studies that did not find a significant association between RCU and PTSD used a measure assessing PTSD symptoms on a continuum (i.e., PCL, PC-PTSD Screen) and created a dichotomous PTSD diagnosis variable (Brunault et al., 2019; Walukevich-Dienst et al., 2019). Dichotomizing continuous variables can lead to several problems, such as losing variability and reducing statistical power to detect a relation between variables (Altman & Royston, 2006). Two studies that did not find a significant association between RCU and PTSD used diagnostic codes from medical records to assess lifetime CUD and PTSD (Lekoubou et al., 2020; Welsh et al., 2017). Both studies tested the self-medication model and it is possible that the self-medicating relationship between PTSD and CUD is more immediate and thus needs to be assessed closer in time opposed to within a lifetime. One study that did not demonstrate associations between RCU and PTSD used a more stringent significance threshold (i.e., p = .01) to control for multiple comparisons, but the association between cannabis dependence and PTSD would have been considered significant under different circumstances with a p-value equal to .02 (Ehlers et al., 2016). One study that did not demonstrate associations between RCU and PTSD excluded cannabis using individuals from their sample for other drug use or alcohol dependence (Johnson et al., 2016). Cannabis users who were excluded from the study were psychiatrically more complex (Johnson et al., 2016), and it is possible that other drug and alcohol use should be taken into account as confounding variables because research shows that nicotine and alcohol use are strongly associated with RCU (Fite et al., 2019; Yurasek et al., 2017). One study that did not demonstrate associations between RCU and PTSD used a measure that has shown to perform poorly with their bisexual sample (MacLeod et al., 2015; Robinson et al., 2016). The other study that did not demonstrate significant associations between RCU and PTSD curated a question with free response to assess RCU frequency during the past week (Vlahov et al., 2004). The unknown psychometric properties of this item may have contributed to not finding an effect.

Although biases were present in the studies that did not find a significant association between RCU and PTSD, it is critical to note that many of these same limitations were also present in the studies that did identify a significant association. For example, Harris and colleagues (2019) used diagnostic codes from medical records to assess lifetime CUD and PTSD when investigating the self-medication model. However, Harris and colleagues (2019) used ICD-10 codes as opposed to the ICD-9-CM codes Lekoubou and colleagues used (2020). In transitioning from ICD-9 to ICD-10, it is unknown why there were decreases in rates of substance use disorders or whether the decrease is related to coding errors (Yoon & Chow, 2017). Additionally, five studies created a dichotomous PTSD diagnosis variable from a measure assessing PTSD symptoms on a continuum and found a significant association between RCU and PTSD despite losing variability (Alenko et al., 2019; de Silva et al., 2016; B. F. Grant et al., 2016; Metrik et al., 2016; Pedersen et al., 2018). Thus, it is likely that there is a relationship between RCU and PTSD that explains their greater-than-chance co-occurrence, but the eight studies that did not find significant associations between RCU and PTSD were likely influenced by another source of error in addition to measurement error. Effect sizes in these eight studies did not vary substantially, ranging from Cohen’s d of −.01 to .39 and β of −.01. Comparitively, effect sizes in the studies that did find significant associations between RCU and PTSD were higher, ranging from Cohen’s d of −.01 to 1.35 and β of .02 to .73, but many were modest with 13 of the 37 studies finding an effect that was considered less than small (i.e., effect size < .20).

4.2. Support for etiologic models of co-occurring RCU and PTSD

The self-medication model presumes that individuals with PTSD use cannabis to cope with distressing symptoms (Khantzian, 1997). The self-medication model is attractive in that it is intuitive to clients with mood and anxiety disorders who report using cannabis without a prescription to self-medicate (Sarvet et al., 2018). Furthermore, there has been an increase in the number of states with medicinal cannabis (35 as of November 2021) that include PTSD as an approved condition for medicinal cannabis. Medicinal cannabis laws may be associated with greater acceptance of the therapeutic value of cannabis, leading individuals to self-medicate. Therefore, it is unsurprising that the majority of studies in the present review tested only the self-medication model and that it has received the most theoretical attention and empirical support (Allwood et al., 2014; Bilevicius et al., 2019; Bordieri et al., 2014; Browne et al., 2018; Buckner et al., 2018; Cornelius et al., 2010; Cougle et al., 2011; Dworkin et al., 2017; Earleywine & Bolles, 2014; Gentes et al., 2016; Goldman et al., 2010; Harford et al., 2018; Harris et al., 2019; Jordan et al., 2020; Kevorkian et al., 2015; Khoury et al., 2010; Lipschitz et al., 2003; McCart et al., 2011; Metrik et al., 2016; Pedersen et al., 2018; Rehder & Bowen, 2019; Rohrbach et al., 2009; Saban et al., 2010; Sanjuan et al., 2019; Schiff et al., 2012; Sommer et al., 2018; Vetter et al., 2008; Villagonzalo et al., 2011; Walsh, Elliott, et al., 2014; Walsh, Resnick, et al., 2014). Notably, the high-risk model was supported by seven of the eight studies in the present review (Alenko et al., 2019; Allan et al., 2019; de Silva et al., 2016; S. Grant et al., 2016; Kerridge et al., 2018; Lee et al., 2018; Miller et al., 2019). Surprisingly, there were no twin/family studies or genetically-informed shared risk studies included in the present review that tested the shared liability model. It is impossible to fully understand the association between RCU and PTSD from both the self-medication and high-risk models’ perspective without first understanding their shared underlying risk. Future research should continue to investigate potentially predictive associations between RCU and PTSD since seven of the eight longitudinal studies in the present review found support for a potentially predictive relationship (4 of 7 self-medication model studies and 3 of 7 high-risk model studies) between the two phenotypes (Allan et al., 2019; Dworkin et al., 2017; Lee et al., 2018; McCart et al., 2011; Miller et al., 2019; Rohrbach et al., 2009; Sanjuan et al., 2019). However, if a longitudinal study found support for the self-medication model or high-risk models, then they did not simultaneously test the opposing direction of effect. Therefore, future research should simultaneously test both the self-medication and high-risk models with longitudinal studies and more sophisticated modeling, such as cross-lagged panel analyses, since they are not mutually exclusive. It may be the case that individuals experience risk from both phenotypes, and it may be that different models explaining the development of co-occurring RCU and PTSD are more relevant for certain individuals. Genetically informed studies can be expensive, but a more feasible genetically-informed option using the shared liability model framework would be to test how shared genetic risk for RCU and PTSD influences these phenotypes in individuals, which could be done both cross-sectionally and longitudinally.

4.3. RCU assessment

Studies that utilized self-report and semi-structured or structured interviews to assess RCU were more likely to find significant results compared to studies that used ICD and DSM diagnostic codes from medical records, which is likely due to increased variability in responses and the limitation that electronic health record data may lead to underestimates of cannabis use (Palamar & Le, 2022). A majority of studies found small to small-medium inter-relations of RCU and PTSD. Notably, studies that used the TLFB and K-SADS-PL to assess RCU were more likely to find medium and large effects compared to other assessment tools.

Cannabis used to be seen primarily as a recreational drug, but medicinal cannabis has rapidly increased over the past couple decades (Carliner et al., 2017). As the motives behind RCU evolve (e.g., enhancement, coping, social, medicinal; Buckner et al., 2019), assessment methods should also evolve. This evolution of RCU assessment is reflected in most of the studies included in the present review.

Despite the concern regarding the variability in assessment present within the RCU literature (e.g., experimental use, regular/frequent use, problematic/disordered use), the majority of studies (6 of 8; 75%) included in the present review that were conducted from 2000–2010 assessed any RCU or frequency of RCU without the use of a standardized self-report assessment tool (Goldman et al., 2010; Lipschitz et al., 2003; Rohrbach et al., 2009; Saban et al., 2010; Vetter et al., 2008; Vlahov et al., 2004), which does not attempt to capture motives behind RCU or assess problematic or disordered RCU. The majority of studies (19 of 37; 51.2%) conducted from 2011–2020 assessed problematic and/or disordered RCU. Although more studies between 2011–2020 compared to 2000–2010 (51.2% vs. 25%) assessed RCU in a way that attempted to capture problematic and/or disordered use, the use of non-standardized self-report assessment tools did not decrease a study’s chance of finding significant associations between RCU and PTSD in either decade.

The majority of studies from 2011–2020 included in the present review used some form of structured or semi-structured interview or standardized self-report measurement tool. While structured or semi-structured interviews may be more ideal for assessment of problematic and/or disordered behaviors compared to self-report measures, they are not always the most feasible option. Therefore, it is equally as important for psychometrically valid self-report measures assessing problematic and/or disordered RCU to be developed. A recent review of screening tools for CUDs and their adaptation for DSM-5 revealed that 5 self-report questionnaires (i.e., Problematic Use of Marijuana [PUM], CAGE – cannabis, Severity of Dependence Scale [SDS], CAST, CUDIT-R) were the only ones that had good characteristics for a short screening tool (Artigaud et al., 2020). Additionally, the Screen of Drug Use (SoDU) was found to be a promising screening tool for CUD in primary care settings (Tiet et al., 2019). Two studies in the present review utilized measures from their list (Brunault et al., 2019; Jordan et al., 2020). Surprisingly, the increased quality of information gathered from assessment tools, such as the CUDIT-R and CAST which attempt to capture problematic and/or disordered recreational use, did not increase the chance of finding significant results or a larger effect compared to non-standardized and other self-report assessment techniques. The standardized assessment tool that was more likely to find a larger effect was the TLFB, which does not capture problematic and/or disordered recreational use and instead asks individuals to retrospectively estimate the number of days of RCU anywhere from 7 days to 2 years prior (Robinson et al., 2014). Lifetime or frequent RCU may not prove clinically meaningful; however, problematic or disordered recreational use that does not meet the threshold of CUD may still be relevant. Future research should focus on CUD opposed to RCU in general and include continuous CUD symptom scores to further explicate these associations. Additionally, future research investigating the self-medicating association between RCU and PTSD should strongly consider assessing trauma-specific, coping-related use.

4.4. PTSD assessment

Studies that utilized self-report and semi-structured or structured interviews to assess PTSD were more likely to find significant results compared to studies that used ICD and DSM diagnostic codes from medical records, which is likely due to increased variability in responses. A majority of studies found small to small-medium inter-relations of PTSD and RCU. However, the study that used the K-SADS-PL was more likely to find a medium effect between PTSD and RCU. Additionally, the K-SADS-PL assesses the onset of PTSD symptoms, which was wisely used in two cross-sectional studies included in this review to help establish the onset of PTSD prior to RCU (Allwood et al., 2014; Lipschitz et al., 2003).

The use of semi-structured or structured interviews are beneficial in the assessment of PTSD and about 40% of the studies in the present review utilized such methods. Moreover, the remaining studies implemented standardized self-report measures of PTSD with the exception of the 3 studies that used ICD or DSM diagnostic codes (Harris et al., 2019; Lekoubou et al., 2020; Welsh et al., 2017). One study used the PC-PTSD, which is a brief screening tool used in primary care settings to identify individuals likely meeting diagnostic criteria for PTSD (Cameron & Gusman, 2003). However, the tool only has four items and although psychometrically sound, provides little detail regarding symptom severity beyond presence of at least one symptom within each PTSD symptom cluster. More continuous measures of PTSD would likely add more detailed insight into the relations between PTSD and RCU. Twenty-four studies took this approach and measured PTSD symptoms using a standardized self-report measure, semi-structured, or structured interview without dichotomizing into a PTSD diagnosis. Four studies used the CAPS (Weathers et al., 2018), which is the gold standard in PTSD assessment, and all found significant associations between greater PTSD symptoms and RCU (Bordieri et al., 2014; Gentes et al., 2016; Metrik et al., 2016; Sanjuan et al., 2019). Sanjuan and colleagues (2019) utilized both the PCL and the CAPS in their longitudinal study investigating the self-medication model and found significant associations between daily peak PTSD symptoms and RCU. The PCL and the CAPS are the most widely used, psychometrically sound PTSD assessment tools and future research should continue to utilize them when investing the association between PTSD and RCU.

The PCL is a self-report assessment tool that can be used to capture PTSD symptoms, as well as make a provisional PTSD diagnosis, more accurately than other self-report assessment tools since its questions correspond directly to the DSM-IV (Weathers et al., 1993) or DSM-5 (Weathers et al., 2013) symptom criteria for PTSD. The PCL was utilized by 17 studies included in this review and 13 of which found significant results. Therefore, studies that used the PCL were more likely to find significant results when investigating the association between PTSD and RCU in adult samples. In comparison, the next most used self-report assessment tool was the UCLA PTSD-RI, which was used by two studies included in this review and both found significant results. However, the UCLA PTSD-RI is only validated for children and adolescents (Kaplow et al., 2020), and should continue to be used when investigating the association between PTSD and RCU in children and adolescent samples.

The CAPS is a structured interview that can be used to capture PTSD symptoms over the past week, as well as make a current (past month) or lifetime diagnosis (Weathers et al., 2018). The CAPS was utilized by four studies included in this review and all found significant results. However, three of those studies were cross-sectional and did not fully utilize the quality of information gathered by the gold standard in PTSD assessment. Similar to the K-SADS-PL, the CAPS also assesses time of onset of symptoms, which could be used to help establish if PTSD developed before or after RCU when investigating the association between the two. However, the K-SADS-PL is only validated for children and adolescents (Kaufman et al., 1997). A version of the CAPS for children and adolescents ages 7 and above is available (Pynoos et al., 2015). Thus, the CAPS should continue to be used when investigating the association between PTSD and RCU and future studies should utilize the time of onset of symptoms information to help establish the direction of effect or predictability.

4.5. Cross-sectional versus longitudinal studies

A majority of studies in this review utilized a cross-sectional opposed to a longitudinal study design to investigate the association between RCU and PTSD. The self-medication and high-risk models presume that RCU or PTSD comes before the other, thus a longitudinal study design is more desirable to investigate the direction of effect. All except one of the longitudinal studies included in this review found support for a potentially predictive relationship between RCU and PTSD regarding the self-medication model, which should be given more weight than the results of cross-sectional studies included in this review. Most cross-sectional study results are just an association rather than predictive. However, all of the cross-sectional studies included in this review that tried to establish PTSD or RCU prior to the other found support for a potentially predictive relationship between the two phenotypes. A majority of the remaining cross-sectional studies included in this review found support for an association between RCU and PTSD regardless of the various ways RCU and PTSD were assessed regarding time. For example, a study testing the self-medication model used past month PTSD symptoms to predict past 6-month CUD symptoms and found a significant association. Similarly, another study testing the self-medication model used past 2-weeks PTSD symptoms to predict a lifetime CUD diagnosis and found a significant association. Regardless of study design, RCU and PTSD frequently co-occur, and the self-medication and high-risk models posited to explain the association appear to be well-supported (Allan et al., 2019; Dworkin et al., 2017; Lee et al., 2018; McCart et al., 2011; Miller et al., 2019; Rohrbach et al., 2009; Sanjuan et al., 2019).

4.6. Conclusions and future directions

The intersection of RCU and PTSD is a growing field of interest; however, the existing literature has several methodological concerns. As RCU increases in prevalence and potentially contributes to negative physical and mental health outcomes, it is critical to take a more hierarchical approach to the investigation of RCU, particularly as it relates to PTSD. Overall, the results of this systematic review suggest that the literature to date does provide evidence for the self-medication and high-risk models posited to explain the development of co-occurring RCU and PTSD across multiple populations (e.g., clinical, community, veterans, adolescents, adults), but that empirically rigorous investigation of the self-medication, high-risk, and shared liability model is lacking due to numerous methodological limitations. First, future studies should use longitudinal data thoroughly, which includes the use of time of onset of RCU and PTSD in order to further establish how co-occurring RCU and PTSD develops in reference to the self-medication and high-risk models. Additionally, using longitudinal data thoroughly involves simultaneously testing both the self-medication and high-risk models because they are not mutually exclusive and it is likely that RCU and PTSD interact to maintain each other once developed. Second, future studies should continue using conceptually and psychometrically valid self-report and structured or semi-structured interviews in the assessment of RCU and PTSD to capture the full RCU spectrum (i.e., any use through CUD symptom severity) and PTSD symptom severity, respectively. Additionally, future research that seeks to provide sound evidence for the self-medication model should focus on incorporating an assessment tool to measure the motivation of someone’s RCU that includes coping with symptoms as a separate indicator. Assessment of polysubstance use (i.e., alcohol, tobacco/nicotine) is also warranted as alcohol, tobacco, and cannabis consumption is associated with increased odds of same-day substance co- and tri-use (Roche et al., 2019). Third, although the current evidence supports a bidirectional developmental relationship between RCU and PTSD, the current literature is not definitive since samples have not been free of RCU and PTSD at the start of the studies included in this review. Future studies may benefit from using samples that have not been trauma exposed or initiated cannabis use prior to the study in order to investigate how RCU and trauma/PTSD develop when one has truly not been experienced before the other. Fourth, future studies should include genetically informed studies and genetically-informed shared risk studies testing the non-mutually exclusive shared liability model, which suggests that RCU and PTSD co-occur due to common familial risk factors (i.e., shared genes, common environment, or both). Taken together, future research will be able to obtain a more comprehensive understanding of how RCU and PTSD may be associated with one another.