Introduction
Posttraumatic stress disorder (PTSD) can develop following exposure to one or more traumatic events (Goldstein et al., Reference Goldstein, Smith, Chou, Saha, Jung, Zhang and Grant2016; Koenen et al., Reference Koenen, Ratanatharathorn, Ng, McLaughlin, Bromet, Stein and Kessler2017), with symptoms including intrusions (e.g. distressing memories or dreams related to the trauma), alterations to cognitions and mood (e.g. exaggerated negative beliefs pertaining to self, others, or the world; persistent negative emotional state), heightened arousal and reactivity (e.g. hypervigilance, concentration difficulties), and avoidance of stimuli associated with the traumatic event (APA, 2022). Epidemiological studies in North America have found lifetime PTSD prevalence rates that range from 5.0 to 6.8% in the general population (Kessler, Chiu, Demler, Merikangas, & Walters, Reference Kessler, Chiu, Demler, Merikangas and Walters2005; Statistics Canada, 2022). Despite ongoing efforts to understand and implement effective psychotherapeutic treatments for PTSD, there remain substantial treatment challenges, in part due to the heterogeneity of its presentation, an incomplete understanding of the etiological and maintenance factors, and features of the disorder that hinder treatment, potentially including impulsive decision making (e.g. substance use) leading to premature treatment discontinuation (Back, Waldrop, & Brady, Reference Back, Waldrop and Brady2009; Lewis, Roberts, Gibson, & Bisson, Reference Lewis, Roberts, Gibson and Bisson2020; Schottenbauer, Glass, Arnkoff, Tendick, & Gray, Reference Schottenbauer, Glass, Arnkoff, Tendick and Gray2008; Zoellner, Pruitt, Farach, & Jun, Reference Zoellner, Pruitt, Farach and Jun2014).
Recent efforts have been aimed at developing a psychiatric nosology that emphasizes transdiagnostic processes in the development and maintenance of multiple disorders, offering potential for novel treatment targets (Cuthbert, Reference Cuthbert2022; Dalgleish, Black, Johnston, & Bevan, Reference Dalgleish, Black, Johnston and Bevan2020; Kozak & Cuthbert, Reference Kozak and Cuthbert2016) and greater treatment efficiency (Barlow, Harris, Eustis, & Farchione, Reference Barlow, Harris, Eustis and Farchione2020). One relevant neurocognitive process showing promise as a transdiagnostic factor is delay discounting, a behavioral economic index that assays individual preference for smaller but immediate rewards over those that are larger but delayed. Delay discounting has historically been considered a behavioral index of impulsivity, but there remains controversy about the nature of impulsivity as a psychological construct, how delay discounting fits into such a conceptualization, and the degree to which delay discounting may be truly transdiagnostic (Bailey, Romeu, & Finn, Reference Bailey, Romeu and Finn2021; Levitt et al., Reference Levitt, Oshri, Amlung, Ray, Sanchez-Roige, Palmer and MacKillop2022; Stein, MacKillop, McClure, & Bickel, Reference Stein, MacKillop, McClure and Bickel2023; Strickland & Johnson, Reference Strickland and Johnson2021).
Delay discounting is typically measured using intertemporal choice tasks, in which individuals make decisions between two rewards that vary in magnitude and delay to their receipt (Madden & Bickel, Reference Madden and Bickel2010). Systematic reviews and meta-analyses for continuous association and case-control studies both suggest that delay discounting is positively associated with a variety of mental health symptoms and conditions, including substance use quantity/frequency and addiction severity (Amlung, Vedelago, Acker, Balodis, & MacKillop, Reference Amlung, Vedelago, Acker, Balodis and MacKillop2017; MacKillop et al., Reference MacKillop, Amlung, Few, Ray, Sweet and Munafò2011), behavioral addictions (e.g. gambling and internet gaming disorders; Weinsztok, Brassard, Balodis, Martin, and Amlung, Reference Weinsztok, Brassard, Balodis, Martin and Amlung2021), attention-deficit/hyperactivity disorder (ADHD; Jackson & MacKillop, Reference Jackson and MacKillop2016), and both dysregulated eating (Stojek & MacKillop, Reference Stojek and MacKillop2017) and obesity (Amlung, Petker, Jackson, Balodis, & MacKillop, Reference Amlung, Petker, Jackson, Balodis and MacKillop2016). In a recent meta-analysis of case-control studies for an array of mental health conditions, Amlung et al. (Reference Amlung, Marsden, Holshausen, Morris, Patel, Vedelago and McCabe2019) reported that, relative to controls, elevated rates of delay discounting were found among individuals with major depressive disorder, schizophrenia, borderline personality disorder, bipolar disorder, binge eating disorder, and bulimia nervosa, providing broad support for the transdiagnostic significance of delay discounting. While promising, fewer studies have examined delay discounting in relation to posttraumatic stress, and it remains unclear whether similar associations are present.
There are specific features of PTSD that might contribute to steep discounting of delayed rewards. A defining feature of PTSD is avoidance, whereby individuals make persistent attempts to reduce distress in the short-term by attempting to avoid internal triggers (i.e. thoughts, feelings, physiological sensations) and/or external stimuli associated with the trauma. Short-lived reductions in distress contribute to a negatively reinforced cycle that promotes increasing reliance on avoidance, while simultaneously limiting engagement in activities that are positively reinforcing, disconfirming of trauma-related cognitions, or that may foster habituation or extinction of trauma-related emotions (Foa & Cahill, Reference Foa, Cahill, Smelser and Baltes2001; Foa, Hembree, Rothbaum, & Rauch, Reference Foa, Hembree, Rothbaum and Rauch2019; Olin et al., Reference Olin, McDevitt-Murphy, Murphy, Zakarian, Roache, Young-McCaughan and Peterson2022; Rauch & Foa, Reference Rauch and Foa2006). Thus, like delay discounting, avoidance in PTSD is a short-term strategy to maximize the immediate smaller reward of distress reduction at the expense of long-term symptom recovery. PTSD symptom severity can also be associated with increased perceptions of uncontrollability and unpredictability (Bolstad & Zinbarg, Reference Bolstad and Zinbarg1997), and in previous versions of the Diagnostic and Statistical Manual of Mental Disorders, a characteristic of PTSD was having a sense of foreshortened future (i.e. a sense of foreboding or that one could die at any time; Ratcliffe, Ruddell, and Smith, Reference Ratcliffe, Ruddell and Smith2014). Collectively, these features are consistent with behavioral economic theory and the process of delay discounting.
Despite theoretical and conceptual reasons to expect a positive association between posttraumatic stress and delay discounting, individual studies examining this link have thus far produced mixed results. For example, in a large sample of community adults, Levitt et al. (Reference Levitt, Oshri, Amlung, Ray, Sanchez-Roige, Palmer and MacKillop2022) reported statistically significant correlations between posttraumatic stress symptom severity and delay discounting (rs = 0.14 to 0.16). In contrast, using a smaller sample of active or retired military personnel, Olin et al. (Reference Olin, McDevitt-Murphy, Murphy, Zakarian, Roache, Young-McCaughan and Peterson2022) found a non-significant negative association between similar variables (r = −0.050). Using a case-control design, Morris et al. (Reference Morris, Huffman, Naish, Holshausen, Oshri, McKinnon and Amlung2020) found that delay discounting was significantly elevated among individuals with PTSD relative to those without (d = 0.32), whereas Peck, Nighbor, and Price (Reference Peck, Nighbor and Price2021) found non-significant elevations in delay discounting among individuals with PTSD (without opioid use disorder) relative to healthy controls (d = 0.14) (calculated based on information provided in text).
To address this question, the present study meta-analyzed the primary empirical literature with the goal of characterizing the presence, direction, and strength of the relationship between posttraumatic stress symptoms and delay discounting. Secondary goals were to examine study-level moderating variables that may account for heterogeneity in effects, and to estimate the extent to which findings may be influenced by small study (publication) bias.
Method
Study selection
The methods of this review were pre-registered (CRD42022369226) in the International Prospective Register of Systematic Reviews (PROSPERO). Systematic searches were conducted in PsycINFO, Medline/PubMed, Embase, and Web of Science for articles through November 25, 2022. Boolean terms included the following: (posttraumatic* OR post-traumatic* OR PTSD OR trauma* OR acute stress disorder) AND (delay discounting OR temporal discounting OR intertemporal choice OR impulsive choice). Studies were eligible for inclusion if they met the following criteria: published in a peer-reviewed journal, available in English, human participants, included a delay discounting measure(s), reported on differences between groups (e.g. trauma exposed v. not; PTSD v. no PTSD) or a continuous measure of association between posttraumatic stress symptom severity and delay discounting. Studies were excluded for the following reasons: involved an experimental stress manipulation or a clinical intervention, due to possible confounding (pre-intervention/pre-manipulation associations were eligible, although none of the included studies ultimately met this criteria); were primarily focused on the association between traumatic brain injury and delay discounting, given a different scope; measured trauma using total scores for adverse childhood events (ACEs), to disambiguate traumatic events as per Criterion A of the trauma- and stressor-related disorders in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition Text Revision (DSM-5-TR; APA, 2022) from other forms of adversity included among ACEs. To be maximally inclusive, no restrictions were placed on the type of delay discounting reinforcer, or the measures used to assess posttraumatic stress or delay discounting.
Two authors of the present work (BMB and EEL) independently performed article screening and coding/extraction using the Covidence systematic review software (Veritas Health Innovation), and discrepancies were resolved through consensus ratings. Inter-rater agreement was excellent for title/abstract (92.8% agreement, κ = 0.75) and full text screening (100% agreement). For eligible studies that did not report enough information to derive the effect of interest, the lead author of the present work contacted the corresponding authors of prior studies with a request for the effect size (or information to derive the effect size) of interest. Studies were excluded if a response was not received after two attempts to obtain the requested information.
Characteristics of included studies
The following information was coded or extracted from the included studies: title, year of publication, study design (case-control or cross-sectional with continuous associations), sample sizes after exclusions, percent female, percent White/European ancestry, mean age of the sample, measure of trauma and/or posttraumatic stress, and measure of delay discounting. If age was reported separately by group (e.g. PTSD v. Controls), we took the weighted average of the ages. In one instance, age was binned in various categories, in which case we used the median as an estimate of average age. If an average effect was derived for a study that reported more than one relevant outcome or comparison within the same sample (see meta-analytic approach section), and sample sizes were also reported separately for each outcome (and/or each group), we used the respective average n for meta-analysis, which resulted in a non-discrete sample size for one study. If the percentage of females was not reported for the final sample (i.e. the sample after exclusions), we used the percentage of females in the pre-exclusion sample as a corresponding estimate. In two instances, the authors reported the number of ‘women’ or ‘men’, which are traditionally used as terms referring to gender rather than sex. Given that no separate descriptive statistics were provided for sex, these numbers were used to estimate the percentage of females for the respective samples. In two other instances, the descriptive heading was ‘gender’ but included the number or percentage of ‘male’ or ‘female’ (traditional sex terms) and thus were assumed to be corresponding estimates of biological sex.
Meta-analytic approach
The primary effect of interest was the correlation (Pearson's r) between posttraumatic stress and delay discounting, which is converted to Fisher's Z for analysis (Borenstein, Hedges, Higgins, & Rothstein, Reference Borenstein, Hedges, Higgins and Rothstein2021). Effects from categorical designs were also converted into the same indicator. If the correlation of interest was not reported in the article, effects were initially derived based on available information that allowed an effect to be calculated (see online Supplementary Material Table S2 for information by article). For example, when available, means and standard deviations were extracted for articles reporting on differences between groups. Effects from categorical studies were based on dichotomous categorical variables (e.g. PTSD v. no PTSD group) and continuous delay discounting outcomes. In instances where authors reported multiple statistical models (or examined differences for more than one potential group of relevance), we derived effects based on the least confounded model/group. To avoid issues of non-independence and artificially assigning more statistical weight to an individual sample in the meta-analysis (Borenstein et al., Reference Borenstein, Hedges, Higgins and Rothstein2021), for studies that reported more than one delay discounting outcome within the same sample (k = 5), we averaged the estimates (i.e. average of the associations between posttraumatic stress and delay discounting outcomes) to form a single effect size. In each instance, delay discounting outcomes were from the same task (e.g. small, medium, and large reward magnitudes for the Monetary Choice Questionnaire; Levitt et al., Reference Levitt, Oshri, Amlung, Ray, Sanchez-Roige, Palmer and MacKillop2022). In two studies, correlations were reported between different posttraumatic stress symptom clusters and delay discounting, in which case we preferentially used the correlation for the overall posttraumatic stress symptom index as the most representative estimate.
Some studies of delay discounting report results using an area under the curve (AUC) analysis—which quantifies distances between observed values (subjective value of a reward) and delay to their receipt (Myerson, Green, & Warusawitharana, Reference Myerson, Green and Warusawitharana2001). With higher (steeper) delay discounting, AUC values decrease. Frequently, other studies use metrics that are positively associated with discounting of delayed rewards. For example, delay discounting can be modeled with a hyperbolic function (V = A/(1/ + kD), where V is the current value of the delayed reward, A is the amount of the delayed reward, D is the reward delay, and k is a derived parameter that indicates the discounting rate, with higher k values indicating steeper discounting of delayed rewards (Mazur, Reference Mazur, Commons, Mazur, Nevin and Rachlin1987). One study from the current meta-analysis used AUC analysis, and thus the effect from this study had its sign reversed prior to inclusion (given inverse association between AUC values and k indices).
Effect size conversion, and the meta-analysis itself, were principally performed with Comprehensive Meta Analysis V 4.0 (Borenstein, Hedges, Higgins, & Rothstein, Reference Borenstein, Hedges, Higgins and Rothstein2022). When the studies in the meta-analysis are not functionally identical, the random effects model is often recommended over the fixed-effects model as the former accounts for both within-study sampling error and between-study variance, while guarding against inflated effect estimates (Borenstein et al., Reference Borenstein, Hedges, Higgins and Rothstein2021; Schmidt, Oh, & Hayes, Reference Schmidt, Oh and Hayes2009). Given that true effect sizes are likely to vary with the different sample characteristics and study designs, a random effects meta-analytic approach was employed. Heterogeneity in effects was quantified with Cochran's Q test (a test of the null hypothesis that the included studies share a common effect size) and the I2 statistic (the proportion of observed variance that reflects true differences in effect rather than sampling error; Higgins and Thompson, Reference Higgins and Thompson2002). A one-study-removed analysis was also conducted to determine the extent to which individual studies disproportionately affected the summary effect. When there were sufficient sample sizes, moderator analyses were conducted to determine if categorical differences (sub-groups analysis) or continuous variables (meta-regression) accounted for a significant portion of effect heterogeneity.
Risk of bias was assessed with multiple indices. Rosenthal's classic fail-safe n (Rosenthal, Reference Rosenthal1979) is an estimate of the number of effects averaging an effect size of 0 that would be required to nullify the summary effect of the meta-analysis. The funnel plot for the summary effect was also visually inspected for symmetry of effect sizes around the mean effect. Examination of the funnel plot was supplemented by (1) the Begg-Mazumdar rank correlation test of the association between the treatment effect and the standard error (Begg & Mazumdar, Reference Begg and Mazumdar1994) and (2) Egger's one-tailed test of the regression intercept (Egger, Davey Smith, Schneider, & Minder, Reference Egger, Davey Smith, Schneider and Minder1997), both of which provide indications of whether small sample studies are more likely published due to having larger effect sizes. Lastly, an adjusted estimate of the summary effect was computed using the trim-and-fill approach (Duval & Tweedie, Reference Duval and Tweedie2000), which re-calculates the summary effect after imputing studies that are likely to be missing to the left of the summary effect in the forest plot.
Results
Preliminary analyses
The Preferred Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram (Page et al., Reference Page, McKenzie, Bossuyt, Boutron, Hoffmann, Mulrow and Moher2021) appears in Fig. 1. Following exclusion of non-eligible studies, and three studies whose authors did not respond to data requests, the meta-analysis sample included a total of 13 studies, 14 independent effect sizes, and a combined sample size of N = 6897.
PRISMA flow diagram for study selection.
Note. DD, Delay discounting; ACEs, Adverse childhood events.
Characteristics of the included samples are depicted in Table 1. Mean ages ranged from 8.3 to 76.2 years (mean weighted age across samples = 35.7, median = 37.1 years). Study design was evenly split between case-controls (trauma exposure v. controls, k = 3; groups with PTSD v. controls, k = 4) and continuous associations (k = 7). Traumatic exposure and/or posttraumatic stress severity was most-commonly indexed using self-report (or in one instance, parent-report) measures (k = 10), followed by clinician-administered semi-structured interviews (k = 4). Delay discounting was indexed with measures that employ pre-configured items (Monetary Choice Questionnaire [MCQ] or an adapted version thereof, k = 5; brief MCQ, k = 2; incentivized choice task, k = 1) or an adjusting amount task (ED50, k = 3; other, k = 3). See Table 1 for a list of measures and tasks. Nearly 80% of studies addressed possible influence of a skewed delay discounting distribution, either by using transformed delay discounting indicators (k = 9), by using a non-parametric test (k = 1), or by reporting that skewness and kurtosis were within normal limits (k = 1). Three studies did not report whether transformations were used or required.
Characteristics of studies included in the meta-analysis
Note: ± indicates the presence/absence of a group characteristic; group characteristics and sample sizes reflect those for which effect sizes were extracted or derived and thus may differ from the original study; DD, Delay discounting; PTSD, Posttraumatic Stress Disorder; SA, Suicide Attempt; MDD, Major Depressive Disorder; HC, Healthy Controls; TE, Trauma Exposure; HED, Heavy Episodic Drinking; SUD, Substance Use Disorder; OUD, Opioid Use Disorder; AUD, Alcohol Use Disorder; tx, treatment; SCID for DSM-IV, Structured Clinical Interview for the DSM–IV (First & Gibbon, Reference First, Gibbon, Hilsenroth and Segal2004); CAPS for DSM-IV and DSM-5, Clinician-Administered Posttraumatic Stress Disorder Scale for the DSM–IV (Blake et al. Reference Blake, Weathers, Nagy, Kaloupek, Gusman, Charney and Keane1995) and DSM-5 (Weathers et al. Reference Weathers, Blake, Schnurr, Kaloupek, Marx and Keane2013); CTQ-SF, Childhood Trauma Questionnaire Short Form (Bernstein et al. Reference Bernstein, Stein, Newcomb, Walker, Pogge, Ahluvalia and Zule2003); SSAGA-II, The Semi-Structured Assessment for the Genetics of Alcoholism-II (Hesselbrock, Easton, Bucholz, Schuckit, & Hesselbrock, Reference Hesselbrock, Easton, Bucholz, Schuckit and Hesselbrock1999); PC-PTSD-5, The Primary Care PTSD Screen for the DSM–5 (Prins et al. Reference Prins, Bovin, Smolenski, Marx, Kimerling, Jenkins-Guarnieri and Tiet2016); MCQ 21, Monetary Choice Questionnaire 21-item version (Kirby & Maraković, Reference Kirby and Maraković1996); MCQ, Monetary Choice Questionnaire (Kirby, Petry, & Bickel, Reference Kirby, Petry and Bickel1999); DDT, Delay Discounting Task (note that ‘DDT’ was used to indicate general delay discounting tasks that were not named in the original publications, rather than the name of a particular task); bMCQ, Brief MCQ (also referred to as Brief Delay Discounting Task; Gray, Amlung, Acker, Sweet, and MacKillop, Reference Gray, Amlung, Acker, Sweet and MacKillop2014); ICT, Incentivized Choice Task (Angerer, Lergetporer, Glätzle-Rützler, & Sutter, Reference Angerer, Lergetporer, Glätzle-Rützler and Sutter2015); ED50, Effective Delay 50 (also referred to as the 5-item DDT; Koffarnus and Bickel, Reference Koffarnus and Bickel2014).
a Wittmann, Leland, and Paulus (Reference Wittmann, Leland and Paulus2007).
b Parent reported child exposure to trauma (for the current meta-analysis, we used an average estimate from ‘witnessing someone being swept away by a tsunami’ and ‘saw a dead body’ to align most closely with the definition of a Criterion A traumatic event in the DSM-5-TR); the non-discrete N reflects the sum of average Ns for witnessed (N = 6.5) and did not witness (N = 123) a traumatic event.
c Only medium magnitude reward levels used.
d Young (Reference Young2017).
e Swiss-German version (Forstmeier & Maercker, Reference Forstmeier and Maercker2011).
f Du, Green, and Myerson (Reference Du, Green and Myerson2002).
Meta-analysis results
Meta-analytic results, in addition to a forest plot, appear in Fig. 2. The analysis revealed a statistically significant positive summary effect of small magnitude (r = 0.135, p < 0.0001, 95% CI [0.074, 0.195]), with high heterogeneity across studies (Q = 67.974, p < 0.001; I2 = 81%). The one-study-removed analysis revealed that the summary effect was not unduly influenced by any single effect size (estimates = 0.112–0.151, all ps < 0.001; see online Supplementary Material Table S1).
Forest plot of studies included in the meta-analysis. Box size is proportional to study weight. Black diamond depicts the summary effect, indicating a positive meta-analytic association between posttraumatic stress and delay discounting.
Risk of bias
The fail-safe N indicated that an additional 353 studies would be required to nullify the significant summary effect. The Begg-Mazumdar test (Kendall's τ = 0.07, one-tailed p = 0.37), and Egger's test of the regression intercept (intercept = 0.11, p = 0.47) were both non-significant. Examining the funnel plot revealed symmetric distribution around the mean effect for studies with moderate to larger samples, with two smaller sample studies located to the right of the mean effect (see online Supplemental Materials Fig. S1). The trim and fill technique (Duval & Tweedie, Reference Duval and Tweedie2000) suggested the possibility that one study was missing to the left of the mean effect, although the random effects point estimate remained of similar magnitude at 0.128 (95% CI 0.07–0.19). Collectively, the various indices suggest that there is minimal evidence or influence of publication bias.
Moderator analysis
For categorical moderators, the effect size was statistically significant for continuous association studies (r = 0.092, p = 0.027) and those employing case-control designs (r = 0.179, p < 0.001), with no statistically significant difference between the two (Q = 1.98, p = 0.160). Similarly, the effect size was statistically significant for studies assessing traumatic exposure and/or PTSD symptoms with self-report instruments (includes Matsuyama et al. (Reference Matsuyama, Fujiwara, Sawada, Yagi, Mashiko and Kawachi2020), which assessed trauma using parent report) (r = 0.107, p < 0.001) and clinician-assessed with semi-structured interview (r = 0.249, p = 0.028), with no statistically significant difference between the two (Q = 1.53, p = 0.216). The effect size was also statistically significant for studies employing adjusting amount delay discounting tasks (r = 0.166, p = 0.011) and pre-configured items delay discounting tasks (r = 0.132, p < 0.001), with no statistical difference between the two (Q = 0.24, p = 0.625).
Meta-regression for continuous moderators produced significant estimates for year of publication (coefficient = −0.052, p = 0.017) and mean sample age (coefficient = 0.008, p = 0.009), suggesting that effect sizes tended to decrease over time and increase with the mean age of the sample. Caution is warranted, however, given the relatively low number of samples, and the restricted range for year of publication (2015–2022). Percent female (coefficient = 0.003, p = 0.295) and percent White/European ancestry (coefficient = −0.002, p = 0.327) were not significantly associated with effect size.
Discussion
The current meta-analysis provides the first synthesis of literature on the association between posttraumatic stress and delay discounting. The principal analysis identified a small but significant positive association between posttraumatic stress and delay discounting. Given the heterogeneity in associations, it is possible that steep delay discounting is a feature of posttraumatic stress, but not a defining one, and thus may be variably present. It is worth considering that delay discounting can be influenced by one's context, including stimulating social environments (Bixter & Luhmann, Reference Bixter and Luhmann2021; Gilman, Curran, Calderon, Stoeckel, & Evins, Reference Gilman, Curran, Calderon, Stoeckel and Evins2014; Martínez-Loredo, Reference Martínez-Loredo2023). For some individuals, PTSD-related avoidance may reduce their exposure to such environments that otherwise might enhance delay discounting. The extent to which this contributes to variability in the associations examined in the current study, and whether other factors (e.g. the number and nature of traumatic events experienced) play important roles, requires further study. More generally, the overall effect size here is consistent with recent arguments that complex psychological phenomena—which are likely determined by a multitude of causes—may have small but important effects that contribute to a cumulative psychological science (Götz, Gosling, & Rentfrow, Reference Götz, Gosling and Rentfrow2022).
Various metrics indicate that the summary effect was unlikely to be unduly influenced by publication bias. Examination of key candidate moderators revealed no statistical difference between study designs (case–control v. continuous association studies) or methodology (traumatic exposure and/or posttraumatic stress severity measured with self-report instrument v. clinician-administered semi-structured interview; adjusting amount v. pre-configured item delay discounting tasks). The effect size did tend to decrease with year of publication and increase with mean age of the sample, but interpretation is limited by the relatively small number of samples in the analysis and the relatively recent publication of articles. As additional studies accumulate, it may be useful for future meta-analytic work to examine age and year of publication as moderators of effect size.
As noted, a defining feature of PTSD is avoidance of stimuli associated with the trauma, likely contributing to a negatively reinforced cycle of short-lived reductions in distress and increasing reliance on avoidance. Past work has also shown that relative to trauma-exposed controls without PTSD, individuals with PTSD show lower future specificity (i.e. the generation and description of possible future events in one's life) in response to positive (Kleim, Graham, Fihosy, Stott, & Ehlers, Reference Kleim, Graham, Fihosy, Stott and Ehlers2014) and neutral word cues (Brown et al., Reference Brown, Root, Romano, Chang, Bryant and Hirst2013). PTSD may thus be at least partly characterized by a relatively greater focus on the short-term and/or a pessimistic future perspective, which may be reflected in the positive association between posttraumatic stress and delay discounting.
Notably, the findings here are broadly consistent with evidence for a link between delay discounting and other psychiatric disorders or measures of psychiatric symptomology, such as addiction severity and substance use disorders, major depressive disorder, bulimia nervosa and binge eating disorder, ADHD, schizophrenia spectrum disorders, bipolar disorder, and borderline personality disorder (Amlung et al., Reference Amlung, Marsden, Holshausen, Morris, Patel, Vedelago and McCabe2019; Reference Amlung, Vedelago, Acker, Balodis and MacKillop2017; Jackson & MacKillop, Reference Jackson and MacKillop2016; MacKillop et al., Reference MacKillop, Amlung, Few, Ray, Sweet and Munafò2011; Stojek & MacKillop, Reference Stojek and MacKillop2017). Further, findings are of similar magnitude to prior meta-analysis of continuous associations, specifically (e.g. r = 0.14 for addiction severity/quantity-frequency and delay discounting; Amlung et al., Reference Amlung, Vedelago, Acker, Balodis and MacKillop2017).
Limitations and future directions
Despite using a relatively large aggregated sample for the current report, the literature on the association of posttraumatic stress with delay discounting remains small relative to other psychiatric disorders (e.g. substance use). Further, although candidate moderators for design and methods were not statistically significant, results suggest the possibility that as further studies accumulate, it may reveal larger effects for case-control over continuous association studies, and/or for those employing clinician-administered semi-structured interviews over self-report measures of posttraumatic stress symptom severity. As the literature develops, it will be useful for future meta-analyses to more closely examine these potential effect moderators, which may help clarify and explain the significant heterogeneity that was observed.
Other aspects of methodological heterogeneity remain important considerations for future work. Most studies transformed the delay discounting variable, which is frequently skewed, but in other instances, it was unclear whether these steps were performed or necessary. It is recommended that future studies on PTSD symptoms and delay discounting report whether transformations were performed prior to analysis. With one exception, the studies in the current meta-analysis also focused on samples of adults. While current diagnostic criteria for PTSD in children is similar (in children 6 years or younger) or the same (in children older than 6 years) to those for adults, some symptom expression may differ (APA, 2022). Additionally, the extent to which individuals discount delayed rewards can change across development (Klein, Collins, & Luciana, Reference Klein, Collins and Luciana2022). It remains unknown to what degree traumatic exposure and/or PTSD symptoms may have a differential association with delay discounting across developmental stages, representing a fruitful area for future research.
Additional consideration can be given to the measurement of posttraumatic stress. Only two studies examined associations between PTSD cluster scores and delay discounting (Olin et al., Reference Olin, McDevitt-Murphy, Murphy, Zakarian, Roache, Young-McCaughan and Peterson2022; Olson, Kaiser, Pizzagalli, Rauch, & Rosso, Reference Olson, Kaiser, Pizzagalli, Rauch and Rosso2018), and the results were inconsistent. Olson et al. (Reference Olson, Kaiser, Pizzagalli, Rauch and Rosso2018) found that an avoidance symptom cluster was most strongly positively correlated (controlling for age and sex) with delay discounting (r = 0.388) relative to hyperarousal (r = 0.320) or re-experiencing (r = 0.144). Olin et al. (Reference Olin, McDevitt-Murphy, Murphy, Zakarian, Roache, Young-McCaughan and Peterson2022), on the other hand, found weak, non-significant correlations between delay discounting and clusters of avoidance (r = −0.05), intrusions (r = −0.04), negative alterations in cognition and mood (r = −0.07), and alterations in arousal and reactivity (r = 0.02). Nevertheless, some previous work has found differential associations between individual PTSD cluster scores and behaviors associated with delay discounting, such as substance use (Livingston, Farmer, Mahoney, Marx, & Keane, Reference Livingston, Farmer, Mahoney, Marx and Keane2022; Sullivan & Holt, Reference Sullivan and Holt2008). As more studies employing this parsing appear in the literature, it may therefore also be useful for future meta-analytic work to directly examine associations between PTSD symptom clusters and delay discounting.
PTSD is highly comorbid with several other psychiatric disorders, including those with well-established links with delay discounting. It remains largely unknown to what degree comorbidities are relevant or indeed responsible for PTSD associations with delay discounting. Many of the studies included in the current meta-analysis did not systematically measure and incorporate (i.e. control for) other psychiatric conditions in a manner that would allow for meta-analysis, making the specificity of the relationship of delay discounting to posttraumatic stress less clear. It thus remains possible that the association between posttraumatic stress and delay discounting is a function of an unmeasured third variable (e.g. PTSD is highly comorbid with substance use disorder [Pietrzak, Goldstein, Southwick, and Grant (Reference Pietrzak, Goldstein, Southwick and Grant2011)], which is strongly associated with delay discounting [MacKillop et al. (Reference MacKillop, Amlung, Few, Ray, Sweet and Munafò2011)]). As additional studies become available, future work may benefit from examining the degree to which the presence or number of psychiatric comorbidities affects the strength of association with delay discounting, or whether broader disorder categories (e.g. internalizing v. externalizing) show differential associations with delay discounting. It seems plausible that delay discounting would be further elevated among individuals with greater levels of psychiatric comorbidity. Determining the extent to which this is true will require a direct test in future work.
To the degree that PTSD is partly characterized by an excessive focus on the short-term (e.g. avoidance and short-term distress relief) and deficits in prospective cognition, important questions can be raised about whether modifying one's focus to be more future-oriented may be useful for such populations. One candidate of rapidly growing interest is episodic future thinking (EFT), which refers to the simulation or imagination of possible future events (Atance & O'Neill, Reference Atance and O'Neill2001). Meta-analytic work has recently shown that interventions aimed at improving EFT can reduce delay discounting, particularly when the future event is positively valenced (Ye et al., Reference Ye, Ding, Cui, Liu, Jia, Qin and Wang2022). The extent to which expanding the temporal horizon out of the present toward the future would be useful for individuals with PTSD, in addition to whether modifications would be required to overcome potential interference from traumatic memories, remains an empirical question. Further questions can be raised about the extent to which interventions for PTSD—with or without co-occurring conditions—may lead to clinical change, in part, due to indirect effects of future thinking experiential exercises (e.g. setting values-consistent, long-term goals, as in acceptance and commitment therapy; Meyer et al., Reference Meyer, Walser, Hermann, La Bash, DeBeer, Morissette and Schnurr2018) on delay discounting.
Conclusion
The current meta-analysis provides the first quantitative synthesis showing that traumatic exposure and posttraumatic stress are significantly and positively associated with delay discounting. The findings are consistent with previous meta-analytic findings for other psychiatric disorders and provide additional support for delay discounting as a likely transdiagnostic process. Although further refinement of these relations is needed, especially regarding potential confounders, these results nonetheless support investigating future orientation as a novel clinical target for PTSD.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723003069
Funding statement
BMB's role was partially supported by a postdoctoral fellowship from the Canadian Institutes of Health Research (MFE 181842) and by the Peter Boris Centre for Addictions Research at McMaster University and St. Joseph's Healthcare Hamilton. SHS's role was partially supported by a Tier 1 Canada Research Chair in Addiction and Mental Health. JM's role was partially supported by the Peter Boris Chair in Addictions Research, and a Tier 1 Canada Research Chair in Translational Addiction Research.
Competing interests
JM is a principal and senior scientist in Beam Diagnostics, Inc. and consultant to Clairvoyant Therapeutics, Inc. The authors have no other disclosures.
Open access
