Exploring the Association Between Post-traumatic Stress Disorder Symptoms and Cognitive Functioning: A Network Analysis
M. M. Günak S2182920
Master Thesis Clinical Psychology Supervisor: Dr. E. I. Fried
Institute of Psychology Universiteit Leiden 01-07-2020
Abstract
Background: Post-traumatic stress disorder (PTSD) has been associated with impairments
across cognitive abilities. While some prior work suggests that the PTSD symptom domain of intrusion may be most strongly related to cognitive impairment, little is known about the relation of cognitive functioning with individual PTSD symptoms or other symptom domains, and the temporal stability of such relations. The current study addresses these questions.
Methods: Data were analysed from 1,484 trauma-exposed U.S. military veterans (Mdn=65
years) who participated in the National Health and Resilience in Veterans Study (NHRVS). We estimated four regularised partial correlation networks of DSM-5 PTSD symptoms at baseline (past month or lifetime) and cognitive functioning at baseline and three-year follow-up, respectively. Network comparison tests examined temporal stability, and sensitivity analyses the robustness of the associations.
Results: Across network models, difficulty concentrating and trouble experiencing positive feelings consistently showed unique negative relations to cognitive functioning. Contrary to
expectations, the symptom domains of alterations in arousal and reactivity, as well as cognition and mood were more strongly linked to cognitive functioning than the other two domains. Network structures and overall strength did not significantly differ between cross-sectional and longitudinal networks.
Conclusion: Overall, we highlight the importance of links between PTSD symptoms and
symptom domains on the one hand, and cognitive functioning on the other—relations obfuscated by modelling only PTSD diagnosis or sum score. Given that longitudinal processes between the two constructs appear to be present, we recommend monitoring of cognitive functioning and integrating it into clinical care of PTSD.
Registration Number: 37069 (Aspredicted.org registry)
1. Introduction
Post-traumatic stress disorder (PTSD) is a debilitating condition that arises in response to a traumatic event such as (threatened) death, serious injury or violence (American Psychiatric Association, 2013). Depending on countries, conflicts and sampling strategies, the lifetime prevalence rates of combat-related PTSD among veterans are fairly high, with estimates varying between 6% and 31% (Creamer et al., 2011; Richardson et al., 2010). PTSD
symptoms involve re-experiencing parts of the traumatic event, also known as intrusion (e.g., flashbacks), hyperarousal and avoidance of reminders and distressing memories of the trauma (see Figure 1 for an overview of PTSD symptoms; American Psychiatric Association, 2013). Often, individuals with PTSD seek and receive treatment only many years or decades after disorder onset (Wang et al., 2005). The delay results in a chronic condition that is
accompanied by impairments across a range of domains, including cognitive functioning, daily living and quality of life (Pittman et al., 2012; Qureshi et al., 2011; Ross et al., 2018). [Figure 1]
Over the past two decades, especially cognitive impairment in PTSD has attracted great attention (Eren-Koçak et al., 2009). There are several indicators that the relationship between PTSD and cognitive functioning is of particular importance. PTSD has been viewed as a disorder of memory (McNally, 2006). The recent addition of the symptom cluster alterations in cognition and mood including impaired memory of the trauma, to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) emphasises this (American Psychiatric Association, 2013). The persistence of PTSD is thought to result from cognitive processing alterations in affected individuals, leaving them with a sense of ongoing severe threat (Ehlers & Clark, 2000). Two key processes, in turn, that underlie this sense of threat are extensive negative appraisals of the event and a disrupted autobiographical memory
associated with poor elaboration and contextualization of the trauma. Accordingly, well-established methods that are recommended to treat PTSD include the trauma-focused cognitive behavioural therapy and eye movement desensitization reprocessing (EMDR; Bisson, 2007; NICE, 2018). Both share two essential components and are based on the exposure to the traumatic memory and cognitive processing of the meaning of the trauma (Forbes et al., 2007). Collectively, these features underline the relationship between memory and cognition in PTSD (Wang et al., 2016).
Several studies have shown impairments across cognitive domains in veterans with PTSD compared to those without, including (working) memory, attention, executive functions, learning and processing speed (Hudetz et al., 2010; Koso & Hansen, 2006; Samuelson et al.,
2006; Schuitevoerder et al., 2013; Vasterling et al., 1998, 2012, 2018; Yehuda et al., 2005). Notably, it has been found that memory and cognitive functioning are more impaired in individuals with PTSD compared to those who were exposed to trauma but did not develop PTSD (Friedman et al., 2019; Qureshi et al., 2011; Schuitevoerder et al., 2013). Additionally, more severe PTSD symptoms have been shown to be associated with lower cognitive
functioning in both Vietnam and Iraq War veterans (Vasterling et al., 2002, 2018). Therefore, trauma exposure alone is not sufficient to explain the association with decline in cognitive abilities observed in individuals with PTSD.
Most studies have examined cognitive functioning associated with PTSD as an entity. Research investigating PTSD symptom clusters, also known as domains, and cognitive functioning is limited, mainly involving non-veteran samples. The results of the few existing studies indicate that the symptom domain of intrusion (e.g., flashbacks, nightmares) plays a crucial role for impaired cognitive functioning (Boals, 2008; Bomyea et al., 2012; Clouston et al., 2016; Johnsen et al., 2008; Kivling-Bodén & Sundbom, 2003; Parslow & Jorm, 2007; Vasterling et al., 1998). Both intrusion symptoms and elevated arousal have been suggested to compete for attentional resources with ongoing cognitive processes, which are then disrupted (Boals, 2008; Kolb, 1987), resulting in reduced cognitive functioning. This may be associated with a reduced ability to inhibit reactions to irrelevant information (Vasterling et al., 1998) and regulate the content of conscious awareness referred to as cognitive control (Bomyea et al., 2012; Miyake et al., 2000; Wessel et al., 2008). Yet, prior studies have not supported hyperarousal to be related to impaired cognitive functioning (Bomyea et al., 2012; Clouston et al., 2016). Interestingly, Vasterling and colleagues (1998) found that in Persian Gulf war veterans, such disinhibition was negatively associated with avoidance-numbing symptoms. Other studies found that avoidance was not linked to cognitive functioning (Boals, 2008; Bomyea et al., 2012; Clouston et al., 2016). This may reflect the tendency to avoid (i.e., inhibit), at least superficially, intense trauma-related experiences, which, in turn, might preserve cognitive functioning (Vasterling et al., 1998). Due to its recent introduction, the symptom domain of negative alterations in cognition and mood has not yet been examined to be associated with cognitive functioning. Although the symptom domain of intrusion seems to be particularly associated with cognitive impairment, evidence currently is preliminary and/or inconsistent.
While research has generally focused on the dichotomy between individuals with a diagnosis of a mental disorder and those without (Armour, Fried, & Olff, 2017), there are 636,120 possible symptom combinations that qualify for a DSM-5 PTSD diagnosis alone
(Galatzer-Levy & Bryant, 2013). The heterogeneity of symptom presentations has led to ongoing debates about the validity and reliability of DSM diagnostic criteria, both in general (Insel, 2013), and specifically of DSM symptom domains for PTSD (Armour, Contractor, et al., 2016; Armour, Műllerová, et al., 2016). This is underlined by the finding that PTSD symptoms are associated with one another across the clusters they are assigned to in the DSM-5 (Armour, Fried, Deserno, et al., 2017). Importantly, prior studies found that different groups of individuals, presenting with diverse PTSD symptom profiles, show different relations with external variables such as risk factors, comorbidities or functional impairment (e.g., Au et al., 2013; Lazarov et al., 2019; Rosellini et al., 2014; Ross et al., 2018). Thus, there may be value in trying to understand the relations of PTSD with cognitive functioning by examining symptoms people actually display, both individually and within their domains, rather than diffuse syndromes. Network analysis lends itself well as a tool to examine the link between PTSD symptoms and cognitive functioning while embracing the complexity of PTSD (Borsboom, 2017; Fried et al., 2017). The network theory of psychopathology is based on the idea that symptoms and other problems directly interact with and cause one another in a dynamic system (i.e., network; Borsboom, 2017; Fried & Cramer, 2017). Thus,
complementing diagnosis-level research by shifting the attention to individual symptoms and variation (Armour, Fried, & Olff, 2017; Fried, 2017; Fried et al., 2017). To date, no network analysis exists that investigates the link between PTSD symptoms and cognitive functioning. Furthermore, most prior studies used a cross-sectional design to examine the link between PTSD and cognitive functioning (Qureshi et al., 2011; Schuitevoerder et al., 2013). Although existing cohort studies indicate longitudinal associations between the two constructs (Parslow & Jorm, 2007; Vasterling et al., 2018), little remains known whether or not PTSD-related cognitive impairment is stable over time. Clarifying the associations of specific PTSD symptoms and symptom domains with cognitive functioning, and their temporal relations, may help to identify veterans with PTSD who are especially susceptible to impaired cognitive functioning and guide personalised treatment planning (Fried et al., 2017; Kivling-Bodén & Sundbom, 2003).
Therefore, the purpose of the study is to elucidate the relationship between PTSD
symptoms and cognitive functioning. We opted to analyse a dataset of veterans because they are a group particularly vulnerable to developing PTSD (Wisco et al., 2016). The objectives of the present study were twofold: a) to identify specific PTSD symptoms and symptom domains that are associated with cognitive functioning; and b) to investigate the temporal stability between PTSD symptoms and cognitive functioning by analysing a second wave of
data three years later. We investigated four research questions (RQ): (1) Is the total PTSD symptom score associated with cognitive functioning in U.S. veterans?; (2) which individual PTSD symptoms are most strongly associated with cognitive functioning; (3) which symptom domain is most strongly related to cognitive functioning?; and (4) do the findings hold over a three-year follow-up? Based on the literature we reviewed above, we hypothesised that PTSD symptoms are negatively associated with cognitive functioning, that the symptom domain of intrusion shows the strongest overall link to reduced cognitive functioning compared to other symptom domains, and that the associations of the estimated network models at baseline (Wave 1) will hold at a three-year follow-up (Wave 2).
2. Methods 2.1 Participants and Procedure
In total, thirteen authors of existing datasets were contacted to inquire about shared access to use for the current study, of which five responded. For the present study, we eventually used data drawn from the second cohort of the National Health and Resilience in Veterans Study (NHRVS), a survey of a nationally representative sample of U.S. veterans (Wisco et al., 2016). The prospective cohort was recruited in September and October 2013 (i.e., baseline; Wave 1) from a research panel of U.S. households that has been developed and maintained by Growth for Knowledge (GfK) Incorporated, a survey research company based in Menlo Park, California (GfK Knowledge Networks, 2020). Panel members were employed through a sampling procedure that includes listed and unlisted phone numbers; telephone,
non-telephone, and cell-phone only households; and households with or without Internet access, allowing coverage of approximately 98% of all U.S. households. Of 1,602 veterans who were in the survey panel when the NHRVS cohort was recruited, 1,484 (92.6%) took part in the NHRVS and completed a confidential 60-min Web-based survey that assessed a range of sociodemographic, psychiatric and health variables. The cohort was re-assessed in September and October 2016 (i.e., follow-up; Wave 2). A total of 713 (48.0%) veterans completed both assessments at Wave 1 and Wave 2. All veterans provided informed consent prior to
participation in the study. The Human Subjects Subcommittee of the Veterans Affairs (VA) Connecticut Healthcare System and VA Office of Research & Development approved the study.
Prior to the analysis of the data, we pre-registered the present study in the Aspredicted.org registry (Registration number: 37069).
2.2 Measures
2.2.1 Demographic Characteristics
A demographic questionnaire assessed age, gender, ethnicity, attained level of education, and current marital and employment status.
2.2.2 Lifetime Exposure to Trauma
The 14-item self-report measure Trauma History Screen (THS) assesses the lifetime exposure to 14 traumatic incidents (Yes/No; Number of times something like this happened; Carlson et al., 2011) in order to establish the criterion A for PTSD of the DSM-5 (i.e., exposure to a traumatic event; American Psychiatric Association, 2013). It includes traumatic experiences across the lifespan including physical or sexual assault, accidents, traumatic incidences during military service, and unexpected loss of a close family member or friend. Additionally, “life-threatening illness or injury” was added as a potentially traumatic event. Veterans who endorsed multiple traumatic experiences were asked, “Which of these experiences was the worst for you?”.
2.2.3 PTSD symptoms
The PTSD Checklist-5 (PCL-5) is a self-report measure that assesses the presence and
severity of PTSD symptoms (Weathers et al., 2013). It comprises 20 items, which are rated on a 5-point Likert scale ranging from 0 (Not at all) to 4 (Extremely). The items on the PCL-5 are in line with the DSM-5 criteria for PTSD and represent clusters B-E (i.e., intrusion, persistent avoidance, negative alterations in cognition and mood, marked alterations in arousal and reactivity; see Appendix A; American Psychiatric Association, 2013; Weathers et al., 2013). In the NHRVS cohort, the PCL-5 was modified in order to include both lifetime and past-month ratings of PTSD symptoms with regards to their self-selected “worst” stressful experience identified on the THS. For the present study, all 20 items were summed up resulting in a possible range of 0 – 80 to provide a total severity score. Internal consistency was excellent at Wave 1, for both past-month PCL-5 (Cronbach’s = .95) and lifetime PCL-5 (Cronbach’s = .95). Probable PTSD was determined as a past-month PCL-5 sum score of ≥ 31, as
recommended by previous evidence (Bovin et al., 2016). While this cut-off score served to identify veterans with probable PTSD to describe the sample, no cut-off score was applied for the analyses in order to examine the relationship between PTSD on a continuum rather than categorically. Thus, all participants in the sample who were exposed to trauma and
consequently filled out the PCL were included in the study. In that way, we aimed to reduce the impact of Berkson’s bias (de Ron et al., 2019).
2.2.4 Cognitive Functioning
One subscale of the Medical Outcomes Study (MOS) assesses cognitive function and is a self-report measure encompassing six Likert-type items on attention, memory, concentration, and reasoning in the past month (Appendix B). The responses to the individual items are
converted and summed up, ranging from 0 (least favourable functioning) – 100 (most
favourable functioning; Revicki et al., 1998). Internal consistency was excellent at Wave 1
(Cronbach’s = .92) and Wave 2 (Cronbach’s = .93). The sum score was used in all further analyses.
2.2.5 Covariates
Gender, age, level of education, depression and alcohol misuse were included covariates. The reason is that research has shown that female sex, older age, low level of education, and having had depression or alcohol abuse are related to cognitive impairment (Barrett et al., 1996; Bhattarai et al., 2019; Laws et al., 2016; Samuelson et al., 2006; Tervo et al., 2004). Simultaneously, female sex (Luxton et al., 2010), although evidence is mixed for veterans (Haskell et al., 2010; Turner et al., 2007), low level of education, depression and alcohol use disorder (Armenta et al., 2019; Carter et al., 2011; Jakupcak et al., 2010; Kessler et al., 1995; Stecker et al., 2010) have also been associated with PTSD (McNally & Shin, 1995; Vasterling et al., 2002). Lifetime history of major depressive episodes and alcohol abuse/dependence were measured with the Mini International Neuropsychiatric Interview (MINI). Adequate test-retest and interrater reliability and concordance with other well-established diagnostic
interviews have previously been shown (Gros & Blake Haren, 2011; Sheehan et al., 1998). 2.3 Statistical Analysis
The statistical software R (version 3.6.2) was used to perform the statistical analyses. Given that the distributions of MOS, past-month and lifetime PCL-5 were skewed, baseline
characteristics of veterans were described by medians as well as means. Baseline and follow-up sample characteristics were compared using the Wilcoxon signed-rank test, a
nonparametric equivalent to the paired t-test. For paired nominal data, the McNemar’s test for dichotomous variables, and the McNemar’s-Bowker test for variables with more than two categories, were used.
Spearman’s correlations between PTSD at baseline (i.e., past-month and lifetime PCL-5 sum scores) and cognitive functioning (i.e., MOS sum scores) at Wave 1 and 2, respectively, were computed in order to see whether increased PTSD symptom severity overall is
follow-up (i.e., RQ1 and RQ4). Participants with missing data were omitted for computing the overall association between PTSD and cognitive functioning.
We followed best practices in the field and estimated network models based on Spearman’s correlations (Epskamp & Fried, 2018). We used pairwise complete observations to deal with missing data when possible with the used R-package. Alternatively, we used listwise deletion. The R codes for network analyses are available in Appendix C.
2.3.1 Network Estimation
For the research questions 2 to 4, of which PTSD individual symptoms and symptom domains are most strongly associated with cognitive functioning and whether it holds through the follow-up, we estimated two cross-sectional and two longitudinal network models: two
network structures were analysed at Wave 1 (i.e., past-month or lifetime PCL-5 with cognitive functioning) and two networks across Wave 1 (i.e., past-month or lifetime PCL-5) and Wave 2 (i.e., cognitive functioning; see Table 1 for an overview). A priori specified covariates were adjusted for in each network. Unique relations among PTSD symptoms, the total score of the MOS and covariates were estimated. In network models, ‘nodes’ represent variables and ‘edges’ between these nodes conditional dependence relations, namely partial correlations, which are associations between nodes after controlling for the influence of all other nodes (i.e., variables; Borsboom, 2017; Epskamp, Borsboom, et al., 2018). If no edge emerges, then nodes are conditionally independent. As the data involves mostly ordinal variables, we estimated the networks by means of the Gaussian Graphical Model (GGMs) with the R-package qgraph (Epskamp et al., 2012). Categorical variables (i.e., level of education, gender, lifetime depression and alcohol use disorder) were treated as ordinal. In order to avoid false-positive findings and reduce the risk of overfitting, we estimated GGMs by using the least absolute shrinkage selection operator (LASSO; Tibshirani, 1996). LASSO shrinks all coefficients (i.e., positive and negative edge weights equally) toward zero and sets small weights exactly to zero. In that way, only the relevant edges remain in the network, resulting in a sparse network structure. The strength of the shrinkage is controlled by varying a tuning parameter λ, which is selected by minimising the Extended Bayesian Information Criterion (EBIC; Epskamp, Borsboom, et al., 2018; Epskamp & Fried, 2018; Foygel & Drton, 2010). The EBIC itself involves γ, a hyperparameter that controls to what extent the EBIC favours simpler models with fewer edges, which was set to 0.5 (i.e., the default setting) for all four network analyses.
Since recent research that identified potential problems with regularisation (Williams et al., 2019), two alternative approaches to estimate network models were used for each a priori
specified network as robustness analyses: 1) without thresholding, which additionally sets coefficients that are lower than the threshold to zero in both the EBIC computation of all considered models and the returned final model (Epskamp, 2018; Epskamp & Fried, 2018; Muthén, 1984); and 2) using a novel network estimation method, ggmModSelect (Epskamp, 2018). The latter entails a model search of unregularised GGM models, where 100 models are re-fitted without regularisation to choose the optimal unregularised GGM according to EBIC. During this selection process, all possible models are tested by adding and removing one edge at a time until the EBIC can no longer be improved.
[Table 1]
2.3.2 Network Inference
In the network models, we estimated node predictability using the mgm package. The networks were re-estimated with Mixed Graphical Models (MGMs) using node-wise regression and listwise deletion. Node predictability quantifies how well a node can be predicted by other nodes it shares an edge with and can be interpreted like R2 (Haslbeck &
Fried, 2017; Haslbeck & Waldorp, 2018).
In order to test RQ3—which PTSD symptom domain is most strongly associated with cognitive functioning—we computed average connectivity (i.e., average strength estimate) with cognitive functioning within each symptom domain, using listwise deletion. That is, edge weights between all PTSD symptoms of a domain and cognitive functioning were summed and then divided by the number of domain items according to the PCL-5.
Differences in average connectivity with cognitive functioning between the PTSD symptom domains were bootstrapped with 1000 iterations using the package bootnet (Epskamp, Borsboom, et al., 2018). We also estimated average connectivity using absolute edge weight values to ensure that this would not lead to different results.
2.3.3 Network Stability
Bootstrapping routines were implemented to estimate edge weight accuracy for each network model (i.e., how precisely parameters were estimated). Therefore, for each network, we calculated 95% confidence intervals around the edge weights based on 2500 bootstrap samples to quantify uncertainty corresponding to all edge-estimates. Based on these bootstrapped samples, edge-weight difference tests were conducted as indicators of edge weight accuracy, testing for significant differences between any two edges of the network (Epskamp, Borsboom, et al., 2018).
2.3.4 Network Comparison
To investigate temporal stability (RQ4), we compared the network estimated with past-month PCL-5 and cognitive functioning at Wave 1 (Network 1), with the network including past-month PCL-5 at Wave 1 and cognitive functioning at Wave 2 (Network 2). We repeated the analyses with lifetime PCL-5 (Network 3 and 4). First, to obtain a coefficient of similarity for the networks, we computed correlations of the adjacency matrices (i.e., network structures) within each pair of networks, using Spearman’s correlations. Second, we formally tested whether network models within each pair differed from one another with regards to structure and level of connectivity (i.e., global strength) with the R package NetworkComparisonTest (NCT; van Borkulo et al., 2016). To this end, we started with a permutation test to examine whether the network structures (i.e., all edges) are identical, with 1000 iterations. If any network structures were significantly different from each other, post hoc tests were conducted to quantify how many and which individual edges differed specifically. Subsequently, the sum of edge weight values was examined in order to test whether the networks had equal
connectivity (i.e., global strength estimates). Third, networks were also compared using absolute edge weight values.
2.3.5 Post Hoc Sensitivity Analyses
We performed several post hoc analyses to assess the robustness of the results. First, we correlated cognitive functioning (i.e., MOS total score) at Wave 1 and at Wave 2; this was followed by repeating the analyses of the two longitudinal networks of past-month and lifetime PCL-5 at Wave 1 with cognitive functioning at Wave 2 (i.e., Network 2 and 4)—but this time adjusting for cognitive functioning at Wave 1, in addition to a priori specified covariates. However, the R package NCT currently cannot compare network models that do not contain an equal number of variables. Hence, the re-estimated longitudinal networks taking cognitive functioning at Wave 1 into account could not be compared with the cross-sectional models (which had one variable less).
Second, we formally compared the cross-sectional (i.e., Network 1 with Network 3), and the longitudinal networks (i.e., Network 2 with Network 4; and Network 2 with Network 4 controlling for cognitive functioning at Wave 1) using NCT.
2.3.6 Network Visualisation
The R-package qgraph (Epskamp et al., 2012; Epskamp, Costantini, et al., 2018) was used to visualize all resulting associations as networks. We only included partial correlations in the graphs with p < .05. The thickness of the edges represents the magnitude of the partial correlations. Blue edges depict positive associations (i.e., conditional dependence relations)
among nodes (i.e., variables) and red edges represent negative relations. Node predictability is illustrated by rings around each node; the more shared variance the node has with its
neighbours, the fuller the ring around the nodes.
3. Results 3.1 Sample Characteristics
Respondents were predominantly non-combat veterans (61.7%), male (89.4%), White
(81.1%), and older adults, with a median age of 65 years (IQR = 54 – 73 years). Many served in the Army (40.1%) or Navy (25.1%). Of the 1,484 veterans, 1,268 (85.4%) had been
exposed to at least one traumatic incidence at baseline and on average, experienced approximately three such events (see Table 2). The trauma most commonly endorsed was sudden death of a close family member or friend (59.6%; Table 1, Appendix D).
Approximately a third of the respondents stated to have seen someone dying, badly hurt or being killed (36.9%) or seen something horrible specifically during military service (29.2%). Of the respondents, 84.0% received higher education (i.e., college, bachelor’s degree or higher), the minority had a lifetime history of a major depressive episode (9.2%) or alcohol abuse/dependence (36.5%). Past-month and lifetime PCL-5 scores were missing for 350 (23.6%) and 238 (16.0%) participants, respectively, most of whom have not previously experienced a traumatic event according to the THS. Scores of MOS assessing cognitive functioning were missing for 45 participants (3.0%) at baseline and for eight participants (1.1%) at follow-up.
Of those who filled out the past-month PCL-5, the prevalence of probable PTSD was 8.2% at baseline and 5.1% at 3-year follow-up (p = .617). Respondents of the 3-year follow-up differed from those at baseline in terms of ethnicity (p = .046) with more Hispanic veterans at follow-up; reduced cognitive functioning (Z = -3.026, p = .002), likely explained by ageing; marital status (p = .021), with slightly higher rates of being divorced, widowed, or not living with a partner; and employment (p < .001) with more veterans being retired and fewer currently looking for work. Veterans did not significantly differ with regards to PCL-5 score in the past month (Z = -0.420, p = .674), gender (p > .99) or level of education (Z = -1.225, p = .220) between baseline and follow-up.
[Table 2]
3.2 Overall Association between PTSD and Cognitive Functioning
Past-month PCL-5 total score measured at baseline was negatively associated with baseline cognitive functioning (Spearman’s ρ = -0.58, p < .001, n = 1,104). Three years later, the
association was Spearman’s ρ = -0.32 (p < .001, n = 543). Correlations between lifetime PCL-5 total score and cognitive functioning at baseline (Spearman’s ρ = -0.PCL-54, p < .001, n =
1,213), and cognitive functioning at follow-up (Spearman’s ρ = -0.33, p < .001, n = 602) were similar in magnitude. Our hypotheses of a negative relationship between the total PTSD symptom score and cognitive functioning, which held over a three-year follow-up, was supported (RQ1 and RQ4).
3.3 Past-month PCL-5 and Cognitive Functioning
For each network, the three models corresponding to each approach of network estimation were nearly the same, with the highest correlation occurring between the regularised model without thresholding and the two other models (i.e., regularised model with thresholding and the novel network estimation method ggmModSelect; see Appendix D for details). Thus, regularised network models estimated without thresholding (which are the default in the literature) were used for further analyses.
3.3.1 Network Models 1 and 2
Figure 2 shows the estimated networks of past-month PTSD symptoms and cognitive
functioning at baseline (Network 1; panel A) and follow-up (Network 2; panel B), accounting for covariates. Generally, edges mostly emerged between symptoms of alterations in arousal and reactivity, and cognition and mood (RQ2), with similar findings through the follow-up (RQ4). At baseline, out of 325 possible edges, 182 (56.0%) were estimated to be above or below zero, with a mean edge weight of 0.025, and at follow-up, 178 (54.8%) were present (i.e., non-zero), with a mean edge weight of 0.027, which implies a highly similar level of sparsity. In both networks, strong edges, defined as above average edge weight, emerged between cognitive functioning and the two PTSD symptoms difficulty concentrating (E5; see Appendix A), and trouble experiencing positive feelings (D7). In Network 1 (i.e., past-month PTSD symptoms and cognitive functioning at baseline), strong edges were found between cognitive functioning and irritable behaviour, angry outbursts, or acting aggressively (E1),
avoiding memories, thoughts, or feelings related to the stressful experience (C1), trouble falling or staying asleep (E6), feeling jumpy or easily startled (E4), trouble remembering important parts of the stressful experience (D1), and loss of interest in activities (D5).
Additionally, the network revealed edges between cognitive functioning and lifetime depression and alcohol use disorder. In Network 2 (i.e., past-month PTSD symptoms and cognitive functioning at follow-up), strong edges were found between strong negative feelings
about yourself, other people, or the world (D2) and blaming yourself or others for the stressful experience or what happened after it (D3) and cognitive functioning at follow-up.
All weights of edges between cognitive functioning and individual PTSD symptoms were negative, except for the edge with taking too many risks or doing things that could cause
harm (E2), which was below the mean edge weight (i.e., not a strong edge).
[Figure 2]
Node predictability was computed by re-estimating the respective network models using
mgm (n = 1,104 at baseline; n = 543 at follow-up). Spearman’s ρ between the re-estimated
network models and those shown in Figure 2 were 0.69 (Network 1) and 0.58 (Network 2). Predictability of cognitive functioning dropped from 60.1% at baseline to 21.6% at follow-up (which makes sense, given that the latter is a prediction over three years, whereas the former is a contemporaneous prediction). Thus, 60.1% and 21.6% of the variance of cognitive functioning at baseline and follow-up, respectively, could be predicted by the other variables in each network. In other words, at baseline, cognitive functioning was predominantly determined by PTSD symptoms and covariates, whereas over time this was reduced. Covariates that contributed to predictability of cognitive functioning include lifetime depression and alcohol use disorder in Network 1. There were no strong edges between cognitive functioning and covariates in Network 2.
To test RQ3, of which symptom domain is most strongly related to reduced cognitive functioning, we compared the average connectivity with cognitive functioning across the four symptom domains. At baseline, the domain of arousal and reactivity alterations was most strongly associated with cognitive functioning. Over time, the two domains of alterations in arousal and reactivity, as well as cognition and mood, were most strongly associated with reduced cognitive functioning (RQ3 and RQ4). Hence, our hypothesis that the symptom domain of intrusion is most strongly linked to reduced cognitive functioning was not
supported. Specifically, in Network 1, based on bootstrapped confidence intervals, we found that the average connectivity between cognitive functioning and symptoms of intrusion were not significantly greater than that of symptoms of avoidance (95% CI [-0.011, 0.045]) or alterations in cognition and mood (95% CI [-0.007, 0.030]; see Table 3). By contrast, the estimated connectivity within the symptom domain of arousal alterations with cognitive functioning was found to be significantly greater than that within the domain of intrusion (95% CI [0.051, 0.093]). Relations of symptoms of arousal alterations with cognitive functioning, on average, were also significantly stronger than of symptoms of cognition and mood alterations, and of avoidance. Lastly, average connectivity with cognitive functioning within the latter two symptom domains did not significantly differ from each other. Network 2 revealed similar results. Average connectivity with cognitive functioning within domains of
avoidance and intrusion was equal to 0. Thus, bootstrap sampling procedures indicated that in Network 2, relations, on average, were not stronger in the domains of intrusion compared to any other domain. However, both domains of alterations in arousal (95% CI [0.020, 0.055]), and cognition and mood (95% CI [0.006, 0.036]) had significantly greater connectivity with cognitive functioning than both those of intrusion and avoidance, respectively. The amount of average connectivity across domains generally did not change when including absolute edge weight values, apart from the domain of arousal alterations in Network 2, where average connectivity slightly increased in magnitude.
[Table 3]
Accuracy analyses revealed moderate 95% confidence intervals around the edge weights in both networks (see Figure 3 for Network 1). Confidence intervals were overlapping, which indicates that many edge weights likely did not significantly differ from one another. In both network models, the edge-weight difference test found that the association strength between
difficulty concentrating (E5) and cognitive functioning was significantly greater than for all
other edges between specific PTSD symptoms and cognitive functioning, meaning that this individual PTSD symptom had the strongest association with reduced cognitive functioning. Additionally, in Network 1, the edge weight between irritable behaviour, angry outbursts, or
acting aggressively (E1) and cognitive functioning was greater than the weight of edges
between most other PTSD symptoms and cognitive functioning. In Network 2, the edge weight of having strong negative feelings about yourself, other people, or the world (D2) and cognitive functioning was significantly greater than that of the edge between taking too many
risks (E2) and cognitive functioning. Further details in terms of edge weight accuracy and
difference are available in Appendix D. [Figure 3]
3.3.2 Comparison of Network Models 1 and 2
The relationships between PTSD symptoms and cognitive functioning did not significantly change over time. This suggests temporal stability of the associations between PTSD
symptoms and reduced cognitive functioning (RQ4), indicating that certain PTSD symptoms at baseline predict cognitive functioning three years later. First, similarity across the two network models (i.e., past-month PTSD symptoms at baseline associated with cognitive functioning at baseline and follow-up) was high, as indicated by a strong correlation between the two adjacency matrices (Spearman’s ρ = 0.97). Second, we used the NCT to compare the networks. In the corresponding permutation test, Network 1 and 2 did not significantly differ from each other regarding network structure (p = .818) or global strength (p = .645). Global
strength was 10.71 in the first and 10.58 in the second network. Third, repeating the analyses with absolute edge weight values did not affect the results.
3.4 Lifetime PCL-5 and Cognitive Functioning 3.4.1 Network Models 3 and 4
We repeated all of the above analyses, with the difference that PTSD during lifetime, as opposed to past-month was assessed. The estimated networks of lifetime PTSD symptoms at baseline and cognitive functioning at baseline (Network 3; panel A) and three-years follow-up (Network 4; panel B), after controlling for covariates, are visualised in Figure 4. As for the networks including past-month PTSD symptoms (Network 1 and 2), mainly symptoms of alterations in arousal, and cognition and mood were related to cognitive functioning (RQ2 and RQ4). Cross-sectionally, also symptoms of intrusion were found be related to cognitive
functioning. In Network 3, out of 325 possible edges, 201 (61.8%) emerged that were non-zero, with a mean edge weight of 0.025, and in Network 4, 195 (60.0%) non-zero edges were revealed, with a mean edge weight of 0.026, again suggesting great similarity in terms of sparsity. In both networks, strong edges (i.e., above-average weight) appeared between cognitive functioning and the three PTSD symptoms difficulty concentrating (E5), trouble
experiencing positive feelings (D7), and trouble remembering important parts of the stressful experience (D1). Moreover, in Network 3, strong edges were also revealed between cognitive
functioning and irritable behaviour, angry outbursts, or acting aggressively (E1), loss of
interest in activities (D5), avoiding memories, thoughts, or feelings related to stressful experience (C1), strong physical reactions when reminded of stressful experience (B5), feelings or acting as if the stressful experience were happening again (B3), and feeling upset when something reminded of stressful experience (B4). Additionally, cognitive functioning
shared strong edges with gender and lifetime alcohol use disorder. In Network 4, strong edges emerged between cognitive functioning and blaming oneself or someone else (D3) and strong
negative feelings about yourself, other people, or the world (D2). All partial correlations
between PTSD symptoms and cognitive functioning were negative. [Figure 4]
Node predictability was again examined by re-estimating the respective network models using mgm (n = 1,213 at baseline; n = 602 at follow-up). Spearman’s ρ between the re-estimated network models and those shown in Figure 4 was 0.75 (Network 3) and 0.62 (Network 4). Predictability of cognitive functioning again declined over time, from 52.6% at baseline to 17.7% three years later. That is, 52.6% to 17.7% of the variance of cognitive functioning at baseline and follow-up, respectively, could be explained by other variables.
Both the covariates gender and lifetime alcohol use disorder have contributed to predictability of cognitive functioning at baseline (Network 3), but not at follow-up (Network 4).
The symptom domain of arousal and reactivity alterations were most strongly associated with cognitive functioning at baseline, compared to the other symptom domains. At follow-up, the domains of alterations in arousal and reactivity, as well as cognition and mood were most strongly associated with reduced cognitive functioning (RQ3 and RQ4). In Network 3, compared to the symptom domain of intrusion, average connectivity with cognitive
functioning significantly differed within the domain of alterations in arousal (95% CI [0.014, 0.058]) but not within the domain of avoidance (95% CI [-0.033, 0.027]) or alterations in cognition and mood (95% CI [-0.014, 0.024]). Specifically, symptoms of arousal alterations, on average, were significantly more strongly related to cognitive functioning than symptoms of intrusion, cognition and mood alterations, and avoidance. In Network 4, among symptoms of alterations in arousal and reactivity (95% CI [0.009, 0.047) or in cognition and mood (95% CI [0.005, 0.042]), average connectivity with cognitive functioning was significantly greater than among symptoms of intrusion. Associations with cognitive functioning were also significantly stronger, on average, among symptoms of the two former domains compared to avoidance symptoms. Average connectivity estimates with cognitive functioning did not significantly differ between domains of alterations in arousal, and cognition and mood, as well as not between the domains of avoidance and intrusion (see Table 3). For both networks, results did not change when investigating absolute edge weight values, corroborating the results.
The moderate and overlapping confidence intervals around the edge weights in both networks suggest that the rank order of those edges should be interpreted with care (see Figure 5 for Network 3). In both network models, edge weight for difficulty concentrating (E5) was again significantly greater than the edges between any other PTSD symptom and cognitive functioning, meaning that this PTSD symptom is most strongly associated with reduced cognitive functioning. Moreover, in Network 3, the edge weight for trouble
remembering important parts of the stressful experience (D1) was also significantly greater
than all other edges between PTSD symptoms and cognitive functioning, except for irritable
behaviour, angry outbursts, or acting aggressively (E1; see Appendix D for details).
[Figure 5]
3.4.2 Comparison of Network Models 3 and 4
Overall, there were no fundamental differences between the two network models (i.e., lifetime PTSD symptoms associated with cognitive functioning at baseline and at follow-up), again
indicating temporal stability (RQ4) and that certain PTSD symptoms at baseline predict cognitive functioning at follow-up. Similarity across Network 3 and 4 was high, as shown by a strong correlation between the respective adjacency matrices (Spearman’s ρ = 0.97). The permutation test of NCT revealed that both the overall network structures of Network 3 and 4 (p = .872) and global strength (p = .439) did not significantly differ from each other. Global strength was 11.05 and 10.77 for Network 3 and 4, respectively. Repeating the analyses with absolute edge weight values revealed similar results.
3.5 Post hoc Sensitivity Analyses
Cognitive functioning at baseline was positively associated with cognitive functioning three years later (Spearman’s ρ = 0.53, p < .001, n = 529). We re-estimated the longitudinal networks with past-month (Network 2) and lifetime (Network 4) PCL-5 at Wave 1 with cognitive functioning at Wave 2, with additional adjustment for cognitive functioning at Wave 1. The longitudinal networks were robust when accounting for previous cognitive functioning and remained largely unaffected by the additional adjustment (see Figure 10, Appendix D, for a visualisation of the network models). Relative to the Networks 2 and 4 without controlling for cognitive functioning at baseline, the magnitude of edge weights generally was attenuated in the re-estimated network models. Nevertheless, strong edges (i.e., above-average weight) predominantly emerged between the same PTSD symptoms and cognitive functioning and again mainly belonged to the domains alterations in arousal and reactivity, and cognition and mood. In both re-estimated networks, strong edges (i.e., above-average weight) between the three PTSD symptoms difficulty concentrating (E5), having strong negative beliefs about
yourself, other people, or the world (D2), blaming yourself or others for the stressful
experience or what happened after it (D3), and cognitive functioning at Wave 2 emerged. All
of these associations were negative. In the re-estimated Network 2, two additional strong edges with positive edges occurred between cognitive functioning at Wave 2 and taking too
many risks or doing things that could cause harm (E2) and feeling very upset when reminded of the stressful experience (B4). In the re-estimated Network 4, a strong edge emerged
between trouble experiencing positive feelings (D7). Cognitive functioning at baseline explained an additional eight to nine percent of the variance of cognitive functioning at follow-up (i.e., predictability was increased to 29.7% and 26.5% in Network 2 and 4, respectively). The results of which PTSD symptom domain is most strongly related to
cognitive functioning at Wave 2 were not as straightforward in the re-estimated Network 2 as in the previous network models due to the two positive associations. Depending on using signed or absolute values of edge weights, average connectivity with cognitive functioning
was significantly greater in the symptom domain of alterations in arousal, or cognition in mood, compared to the other domains. In the re-estimated Network 4, the symptom domain of cognition and mood alterations is more strongly related to cognitive functioning at Wave 2, compared to the other domains, except for the domain of arousal alterations.
Second, we formally compared the cross-sectional (i.e., Network 1 and 3) and longitudinal network models (i.e., Network 2 and Network 4; both with and without adjustment for
cognitive functioning at baseline). Similarity was high within each pair of networks, with Spearman’s ρ of ~0.80 between the respective adjacency matrices. The permutation tests of NCT revealed that global strength or network structure did not differ across networks within each pair (p > .05). Results did not vary when only including absolute values of edge weights.
4. Discussion 4.1 Summary of the Results
We used baseline and three-year follow-up assessments of PTSD symptoms and cognitive functioning in a nationally representative sample of trauma-exposed U.S. veterans to elucidate the relationship between PTSD symptoms and cognitive functioning. To our knowledge, this is the first study to explore network structures of the association between DSM-5 PTSD symptoms and cognitive functioning, cross-sectionally and over time, controlling for
important covariates including age, gender, level of education, lifetime history of depression or alcohol use disorder. We investigated whether PTSD and cognitive functioning are
associated (RQ1), which individual PTSD symptoms (RQ2) and symptom domains (RQ3) are most strongly linked to cognitive functioning, and whether the findings hold over a three-year follow-up (RQ4). We hypothesised that PTSD and cognitive functioning are negatively associated, the symptom domain of intrusion is most strongly related to reduced cognitive functioning relative to the other three PTSD domains, and that results are temporally stable. We performed numerous analyses to ensure robustness of results, including different
assessments of PTSD symptoms (i.e., past month vs. lifetime), and adjustment for cognitive functioning at baseline in the longitudinal networks.
Collectively, four core findings are worth noting. First, as hypothesised, having had PTSD symptoms in the past month or during lifetime was negatively and significantly associated with cognitive functioning, with a correlation of ~0.6 at baseline and ~0.3 at follow-up. Second, across estimated networks, the PTSD symptoms of difficulty concentrating and
trouble experiencing positive feelings were consistently linked to reduced cognitive
functioning included avoiding memories, thoughts or feelings related to the trauma, trouble
remembering important parts of the stressful experience, loss of interest in activities, irritable behaviour, angry outbursts or acting aggressively, and feeling jumpy or easily started.
Through the three-years follow-up, the PTSD symptoms having strong negative feelings
about oneself, other people, or the world as well as blaming yourself or others for the stressful experience or what happened after it were linked to reduced cognitive functioning.
Third, the two symptom domains of alterations in arousal and reactivity, as well as cognition and mood, were predominantly related to reduced cognitive functioning. We were not able to confirm the hypothesis that the symptom domain of intrusion is most strongly associated with cognitive functioning, compared to the other domains. Fourth, the association between PTSD symptoms and reduced cognitive functioning holds over a three-year follow-up, supporting our hypothesis. Although predictability of cognitive functioning decreased in the longitudinal network models, the findings largely replicated through the follow-up. Moreover, findings overlapped greatly for models based on PTSD in the past month or during lifetime. While the magnitude of associations slightly diminished for lifetime PTSD symptoms, our results were found to be robust, underlining that the associations are temporally stable. At baseline, arousal and reactivity alterations were most strongly associated with reduced cognitive functioning. Through the follow-up, two PTSD symptom domains—alterations in arousal and reactivity, and cognition and mood—showed the strongest overall links to reduced cognitive
functioning. Importantly, the results were largely the same in the longitudinal networks when additionally controlling for cognitive functioning at baseline, corroborating the finding that PTSD symptoms predict cognitive functioning three years later.
On the one hand, our findings are generally consistent with prior literature having found that PTSD symptoms are associated with impaired cognitive functioning (Brewin et al., 2007; Schuitevoerder et al., 2013; Scott et al., 2015). In line with previous studies, the association between PTSD symptoms and cognitive functioning seems to be stable over time (Gould et al., 2019; Parslow & Jorm, 2007; Vasterling et al., 2018), with the former predicting the latter. Moreover, the present study investigated and found a relationship assessing everyday
cognitive challenges rather than measuring cognitive functioning experimentally, which is in coherence with findings from other studies (Boals & Banks, 2012; Schweizer & Dalgleish, 2011). Lastly, our results are in line with prior studies that found that the symptom domain of avoidance is not (strongly) associated with reduced cognitive functioning (Boals, 2008; Bomyea et al., 2012; Clouston et al., 2016; Vasterling et al., 1998). On the other hand, our findings are inconsistent with previous evidence with regard to the intrusion symptom domain
being most strongly linked to reduced cognitive functioning (Boals, 2008; Bomyea et al., 2012; Clouston et al., 2016; Johnsen et al., 2008; Kivling-Bodén & Sundbom, 2003; Parslow & Jorm, 2007; Vasterling et al., 1998).
Findings of the extent and nature of change in cognitive function in individuals with PTSD, however, have not been unequivocal and invariant, with some studies not having found a link (Scott et al., 2015). Reasons for the divergent results may be methodological variance among studies, including objective or self-report measures ranging from global to specific cognitive deficits across cognitive domains, participant and trauma characteristics, comparison groups included (i.e., none, non-trauma or trauma-exposed controls), diagnostic tools for PTSD, assessing PTSD overall or specific symptom domains (e.g., only intrusion and avoidance), and covariates (e.g., psychiatric comorbidities) adjusted for. Additionally, performance validity of the tests often has not been tested in prior studies examining cognitive functioning. For example, after excluding participants who did not pass respondent validity testing, Wisdom and colleagues (2014) did not find significant differences in cognitive performance between individuals with and without PTSD symptoms. Such validity testing may also help explaining mixed results.
The findings of individual importance of PTSD symptoms domains are unexpected. Relative to the other three symptom domains, symptoms of intrusion had the weakest (to non-existing) association with cognitive functioning. This may likely be because we examined the relationship between PTSD and cognitive functioning using network psychometrics. Although some previous studies have found that alterations in arousal are related to reduced cognitive functioning in both veterans and the general population (Kivling-Bodén & Sundbom, 2003; Parslow & Jorm, 2007; Vasterling et al., 1998), the majority of prior studies indicated that particularly the symptom domain of intrusion most strongly relates to a decline in cognitive functioning (Boals, 2008; Bomyea et al., 2012; Clouston et al., 2016; Johnsen et al., 2008; Kivling-Bodén & Sundbom, 2003; Parslow & Jorm, 2007; Vasterling et al., 1998). 4.2 Proposed Explanations for Findings
A range of potential mechanisms have been proposed to explain the relation between PTSD and cognitive functioning. For instance, biochemical processes may underlie the association between symptom domains of alterations in arousal and reactivity, and cognition and mood with cognitive functioning, such as abnormal hyperactivity in the hippocampus (i.e., relevant for memory) and amygdala (i.e., involved in emotional processing), but hypoactivity in the dorsolateral prefrontal cortex (i.e., associated with working memory and attentional control processes), and the medial prefrontal cortex (i.e., a potential neural marker of reduced
cognitive flexibility and contextualization of stimuli; (Liberzon & Martis, 2006; Patel et al., 2012). This may also be reflected in the PTSD symptom difficulties concentrating. With regards to the PTSD symptom of having trouble experiencing positive feelings, previous studies found that positive feelings enhance cognitive functioning such as working memory or decision making (Carpenter et al., 2013), and people’s ability to think flexibly (Isen, 2004). There are several theories of how positive affect improves cognitive function, such as the broaden-and-build theory: broadening a person’s momentary thought-action repertoire enhances building various long-term personal resources (e.g., psychological and social) (Fredrickson, 2001, 2004). Alternatively, premorbid lower cognitive abilities may also serve as a pre-existing risk factor for PTSD (Amrhein et al., 2008; Gilbertson et al., 2001; Marx et al., 2009; McNally & Shin, 1995; Moore, 2009; Parslow & Jorm, 2007; Vasterling et al., 1997, 2002), suggesting that the pathways between PTSD symptoms and cognitive functioning may well be bidirectional. Given that evidence on PTSD symptom domains related to cognitive functioning currently remains elusive and preliminary, underlying mechanisms are suggestive at best and have to be consolidated by future research. 4.3 Implications
4.3.1 Implications for Practice
The findings of the present study, if replicated in other samples and populations, may have several implications for effective clinical management for individuals with PTSD. The results highlight the importance for clinicians to be aware of the potential of cognitive impairment among patients, and specifically, older aged veterans with PTSD and to monitor cognitive functioning when treating them (Clouston et al., 2016). Based on the results, this may be even more relevant when symptoms of alterations in arousal and reactivity, and cognition and mood are pronounced. The findings that the association held through a three-year follow-up indicates that there may be a longitudinal process going on. That is, throughout treatment, it should be recommended that clinicians regularly assess cognitive functioning, especially when patients complain about difficulties concentrating or experiencing positive feelings. This, in turn, may guide realistic and timely treatment planning tailored to the individual patient (Kivling-Bodén & Sundbom, 2003).
It has been shown that memory deficits accurately predict current occupational and social functioning in veterans with PTSD (Geuze et al., 2009). Moreover, in a recent study,
perceived cognitive difficulties, particularly in concentration and memory, helped to explain quality of life in veterans with PTSD (Silverberg et al., 2017). This further supports that
impaired cognitive functioning may be an important focus of clinical attention in patients with PTSD.
Interestingly, in their meta-analysis, Scott and colleagues (2015) found that individuals currently seeking or undergoing treatment for PTSD are those who most likely show cognitive difficulties. This may suggest that treatment in itself does not prevent or protect from a decline in cognitive functioning, or that patients with impaired cognitive functioning are particularly likely to seek professional help. In any case, it emphasises that
neuropsychological functioning should be integrated into the clinical management for patients with PTSD. Decreased cognitive functioning may also impede effective treatment as it may entail reduced ability to comply with therapeutic regimes and self-management during the treatment (Clouston et al., 2016; Falconer et al., 2013; Wild & Gur, 2008).
4.3.2 Implications for Research
The results of the present study necessitate validation using comprehensive batteries of cognitive tests and diagnostic evaluations of PTSD by trained clinicians (Clouston et al., 2016; Wisdom et al., 2014). It may be useful to examine the relationship between PTSD symptoms and specific cognitive domains to better understand stress-related patterns of neuropsychological functioning (Vasterling et al., 1998). Similarly, it is important to examine whether different types of trauma, severity and duration of PTSD contribute to the association between PTSD and cognitive function. The directionality of the association between PTSD symptoms and cognitive functioning and underlying, inter-related mechanisms and pathways should be clarified in order to develop and implement appropriate prevention and intervention strategies.
It is essential to investigate the extent to which cognitive functioning might affect the developmental course of PTSD, the implementation of and response to specific PTSD treatments (Johnsen et al., 2008). Cognitive-enhancing interventions such as cognitive remediation therapy might be useful for individuals with PTSD and impaired cognitive functioning. It has been shown to improve cognitive performance, which was positively associated with everyday functional skills in patients with treatment-resistant depression (Bowie et al., 2013). Moreover, in light of the PTSD symptom trouble experiencing positive
feelings being consistently associated with reduced cognitive functioning in the present study,
interventions emphasising positive emotions may be beneficial. For instance, loving-kindness meditation, based on the broaden-and-build theory aims at enhancing feelings of compassion and kindness and identifying opportunities and resources. A pilot study found support for this practice for veterans by increasing pleasant and decreasing unpleasant emotions, and
enhancing personal resources such as environmental mastery and self-acceptance (Kearney et al., 2014). Future research should investigate how such treatment approaches may help to improve cognitive functioning and could be used as adjunctive therapy to treating PTSD. Lastly, regarding long-term development, PTSD repeatedly has been associated with an increased risk of developing dementia in veterans and the general population (Günak et al., under review; Rafferty et al., 2018), for which no disease-modifying cure currently exists (Livingston et al., 2017). This potentially indicates a longitudinal mechanism at place that is worth investigating.
4.4 Strengths and Limitations
The present study extends current knowledge by being the first to conduct a network analysis, specifically developed to investigate unique mutual interactions, between PTSD symptoms and cognitive functioning whilst controlling for important covariates. Unlike previous evidence, the present study examined the association in various stages of complexity, with PTSD as an entity, PTSD symptom domains, and individual PTSD symptoms according to the DSM-5 (American Psychiatric Association, 2013). The robustness of the relationship was assessed by testing the impact of five covariates including age, premorbid intelligence
represented by education level, alcohol misuse, and depression, which previously have been linked to PTSD and cognitive decline themselves (Barrett et al., 1996; Bhattarai et al., 2019; Laws et al., 2016; Samuelson et al., 2006; Tervo et al., 2004). Moreover, the study design allowed us to examine temporal stability. Although one can generally not infer causation from observational studies (Sedgwick, 2013), the prospective cohort study design allows to
determine precedence of PTSD symptoms to cognitive functioning. In post hoc sensitivity analyses, we controlled for cognitive functioning at baseline in the longitudinal networks, which led to similar results. This supports that PTSD symptoms predict cognitive functioning over time, independent of cognitive functioning at the time the PTSD symptoms were
assessed (i.e., baseline). Lastly, the present study included a large representative sample along the whole continuum of PTSD severity. Thus, the results likely are generalisable to trauma-exposed older adults without a full-blown PTSD diagnosis.
Despite these strengths, the present study has several limitations and should be interpreted in light of these. First, although representing an important subpopulation of individuals prone to PTSD symptoms (Wisco et al., 2016), the homogeneity of the sample (i.e., predominantly older-aged male U.S. veterans) may hinder the generalisability of the results to the (younger) general population. Previous studies have found that generally, similar results in terms of PTSD and cognitive functioning in veteran samples were found in other and younger
populations (Clouston et al., 2016; Qureshi et al., 2011; Schuitevoerder et al., 2013; Scott et al., 2015). Furthermore, not all of the participants in the present study experienced combat-related trauma but were exposed to a range of other traumatic incidences. We used
regularisation techniques to prevent overfitting and overinterpretation of the estimated networks, which may mitigate the problem of generalisability. Nonetheless, replication studies are vital.
Second, we defined “strong edges” as edges with weights above the mean edge weight of the network. We could have also used the median or mode as a cut-off point. However, across networks, both of these were lower than the mean weights. While any cut-off score is rather arbitrary, using mean edge weight as the criteria for strong edges was the conservative choice. Third, many different scales to test cognitive functioning or impairment exist. The present study used a self-report measure of cognitive functioning without objective cognitive testing or testing performance validity. Although subjective memory complaints have been linked to objectively tested cognitive decline in veterans (Chao, 2017), there is the risk of assessing subjective rather than objective cognitive impairment (Binder, 1999) and therefore, response bias cannot be eliminated (Wisdom et al., 2014). Similarly, while internal consistency was excellent, corroborating that the scale used in the present study measured cognitive
functioning, it is a subscale and encompasses six items only. Therefore, overall cognitive functioning was examined as opposed to each individual aspect. Based on the present study, it cannot be precluded that certain PTSD symptoms are more or less strongly related to specific aspects of cognitive decline. It is open if the findings will replicate with different scales or cognitive tests.
Fourth, in order to prevent Berkson’s bias leading to spurious correlations when analysing a subset of the sample only, data from the entire sample was analysed. The minority of the sample screened positive for PTSD and results may not generalise to clinical populations of individuals with PTSD but rather may exhibit the average network structure of the broader population of trauma-exposed veterans (von Stockert et al., 2018).
Lastly, individuals with PTSD diagnoses regularly exhibit very different symptoms (Galatzer-Levy & Bryant, 2013) and in the present study, a self-report measure was used rather than clinical interviews. It should be noted that the PCL-5 is a widely used tool to screen for PTSD symptoms and has been found to be comparable with interview-based assessments of PTSD (Wortmann et al., 2016). Nonetheless, the validity of diagnostic criteria has been debated (Fried & Cramer, 2017; Miller et al., 2014). PTSD is highly heterogeneous with a wide range of symptom combinations qualifying for a PTSD diagnosis (Galatzer-Levy
& Bryant, 2013), which is why PTSD symptoms were considered both in domains and individually.
4.5 Conclusion
Our results indicate that in U.S. veterans not all PTSD symptoms are equally important in the relationship between PTSD and cognitive impairment, both cross-sectionally and over time. Certain individual PTSD symptoms as well as the symptom domains of alterations in arousal and reactivity, and cognition and mood are more strongly related to reduced cognitive
functioning than symptom domains of intrusion or avoidance. The results of the present study aim at stimulating new research as much remains unknown with regard to this striking
relationship, and may have important implications for effective clinical care of patients with PTSD.
References
American Psychiatric Association. (2013). American Psychiatric Association, 2013.
Diagnostic and statistical manual of mental disorders (5th ed.). In American Journal of
Psychiatry. https://doi.org/10.1176/appi.books.9780890425596.744053
Amrhein, C., Hengmith, S., Maragkos, M., & Hennig-Fast, K. (2008). Neuropsychological characteristics of highly dissociative healthy individuals. Journal of Trauma and
Dissociation. https://doi.org/10.1080/15299730802226332
Armenta, R. F., Walter, K. H., Geronimo-Hara, T. R., Porter, B., Stander, V. A., &
LeardMann, C. A. (2019). Longitudinal trajectories of comorbid PTSD and depression symptoms among U.S. service members and veterans. BMC Psychiatry, 19(1), 396. https://doi.org/10.1186/s12888-019-2375-1
Armour, C., Contractor, A., Shea, T., Elhai, J. D., & Pietrzak, K. R. H. (2016). Factor structure of the PTSD Checklist for DSM-5: Relationships among symptom clusters, anger, and impulsivity. Journal of Nervous and Mental Disease.
https://doi.org/10.1097/NMD.0000000000000430
Armour, C., Fried, E. I., Deserno, M. K., Tsai, J., & Pietrzak, R. H. (2017). A network analysis of DSM-5 posttraumatic stress disorder symptoms and correlates in U.S. military veterans. Journal of Anxiety Disorders.
https://doi.org/10.1016/j.janxdis.2016.11.008
Armour, C., Fried, E. I., & Olff, M. (2017). PTSD symptomics: network analyses in the field of psychotraumatology. European Journal of Psychotraumatology, 8(sup3), 1398003. https://doi.org/10.1080/20008198.2017.1398003
Armour, C., Műllerová, J., & Elhai, J. D. (2016). A systematic literature review of PTSD’s latent structure in the Diagnostic and Statistical Manual of Mental Disorders: DSM-IV to DSM-5. Clinical Psychology Review, 44, 60–74.
https://doi.org/10.1016/j.cpr.2015.12.003
Au, T. M., Dickstein, B. D., Comer, J. S., Salters-Pedneault, K., & Litz, B. T. (2013). Co-occurring posttraumatic stress and depression symptoms after sexual assault: A latent profile analysis. Journal of Affective Disorders. https://doi.org/10.1016/j.jad.2013.01.026 Barrett, D. H., Green, M. L., Morris, R., Giles, W. H., & Croft, J. B. (1996). Cognitive
functioning and posttraumatic stress disorder. American Journal of Psychiatry. https://doi.org/10.1176/ajp.153.11.1492
Bhattarai, J. “Jackie,” Oehlert, M. E., Multon, K. D., & Sumerall, S. W. (2019). Dementia and Cognitive Impairment Among U.S. Veterans With a History of MDD or PTSD: A
Retrospective Cohort Study Based on Sex and Race. Journal of Aging and Health, 31(8), 1398–1422. https://doi.org/10.1177/0898264318781131
Binder, L. (1999). Subjective Cognitive Complaints, Affective Distress, and Objective Cognitive Performance in Persian Gulf War Veterans. Archives of Clinical
Neuropsychology, 14(6), 531–536. https://doi.org/10.1016/S0887-6177(98)00047-X
Bisson, J. I. (2007). Post-traumatic stress disorder. In Occupational Medicine. https://doi.org/10.1093/occmed/kqm069
Boals, A. (2008). Short communication: Intrusive thoughts and everyday cognitive failures in Holocaust survivors. Stress and Health. https://doi.org/10.1002/smi.1191
Boals, A., & Banks, J. B. (2012). Effects of traumatic stress and perceived stress on everyday cognitive functioning. Cognition & Emotion, 26(7), 1335–1343.
https://doi.org/10.1080/02699931.2011.651100
Bomyea, J., Amir, N., & Lang, A. J. (2012). The relationship between cognitive control and posttraumatic stress symptoms. Journal of Behavior Therapy and Experimental
Psychiatry, 43(2), 844–848. https://doi.org/10.1016/j.jbtep.2011.12.001
Borsboom, D. (2017). A network theory of mental disorders. World Psychiatry. https://doi.org/10.1002/wps.20375
Bovin, M. J., Marx, B. P., Weathers, F. W., Gallagher, M. W., Rodriguez, P., Schnurr, P. P., & Keane, T. M. (2016). Psychometric properties of the PTSD checklist for diagnostic and statistical manual of mental disorders-fifth edition (PCL-5) in veterans.
Psychological Assessment. https://doi.org/10.1037/pas0000254
Bowie, C. R., Gupta, M., Holshausen, K., Jokic, R., Best, M., & Milev, R. (2013). Cognitive Remediation for Treatment-Resistant Depression. The Journal of Nervous and Mental
Disease, 201(8), 680–685. https://doi.org/10.1097/NMD.0b013e31829c5030
Brewin, C. R., Kleiner, J. S., Vasterling, J. J., & Field, A. P. (2007). Memory for emotionally neutral information in posttraumatic stress disorder: A meta-analytic investigation.
Journal of Abnormal Psychology, 116(3), 448–463.
https://doi.org/10.1037/0021-843X.116.3.448
Carlson, E. B., Smith, S. R., Palmieri, P. A., Dalenberg, C., Ruzek, J. I., Kimerling, R., Burling, T. A., & Spain, D. A. (2011). Development and Validation of a Brief Self-Report Measure of Trauma Exposure: The Trauma History Screen. Psychological
Assessment. https://doi.org/10.1037/a0022294
Carpenter, S. M., Peters, E., Västfjäll, D., & Isen, A. M. (2013). Positive feelings facilitate working memory and complex decision making among older adults. Cognition and
