Introduction
Post-traumatic stress disorder (PTSD) is a debilitating mental disorder that can occur in people who have experienced traumatic events such as sexual assault, natural disasters, or terrorist attacks. The lifetime prevalence of PTSD is estimated to be 3.9% globally and 6.8% in the United States (U.S.), and it is approximately twice as common in women than men (Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005; Koenen et al., Reference Koenen, Ratanatharathorn, Ng, McLaughlin, Bromet, Stein and Kessler2017). Compared to men, women with PTSD tend to be exposed to traumatic events at a younger age, may experience more severe symptoms, and are more likely to have comorbid mental disorders, such as depression and anxiety disorders (Olff, Reference Olff2017; Olff, Langeland, Draijer, & Gersons, Reference Olff, Langeland, Draijer and Gersons2007b).
Accumulating evidence has linked PTSD with an increased risk of developing various chronic diseases in both men and women, including cardiovascular diseases (CVD) (Dyball, Evans, Boos, Stevelink, & Fear, Reference Dyball, Evans, Boos, Stevelink and Fear2019; Kubzansky, Koenen, Spiro, Vokonas, & Sparrow, Reference Kubzansky, Koenen, Spiro, Vokonas and Sparrow2007; Sumner et al., Reference Sumner, Kubzansky, Elkind, Roberts, Agnew-Blais, Chen and Koenen2015), type 2 diabetes (Roberts et al., Reference Roberts, Agnew-Blais, Spiegelman, Kubzansky, Mason, Galea and Koenen2015; Vancampfort et al., Reference Vancampfort, Rosenbaum, Ward, Steel, Lederman, Lamwaka and Stubbs2016), and dementia (Desmarais et al., Reference Desmarais, Weidman, Wassef, Bruneau, Friedland, Bajsarowicz and Nguyen2020), as well as all-cause mortality (Boscarino, Reference Boscarino2006; Roberts, Kubzansky, Chibnik, Rimm, & Koenen, Reference Roberts, Kubzansky, Chibnik, Rimm and Koenen2020a). There is also a growing body of literature linking PTSD with poor cognitive function and accelerated cognitive aging, including our previous research in the Nurses' Health Study II (NHS II) (Clouston et al., Reference Clouston, Diminich, Kotov, Pietrzak, Richards, Spiro and Luft2019; Greenberg, Tanev, Marin, & Pitman, Reference Greenberg, Tanev, Marin and Pitman2014; Lawn et al., Reference Lawn, Jha, Liu, Sampson, Murchland, Sumner and Koenen2022; Roberts et al., Reference Roberts, Liu, Lawn, Jha, Sumner, Kang and Koenen2022; Rosen et al., Reference Rosen, Shao, Zhang, Bao, Zhang, Masurkar and Reibman2022; Schuitevoerder et al., Reference Schuitevoerder, Rosen, Twamley, Ayers, Sones, Lohr and Thorp2013; Scott et al., Reference Scott, Matt, Wrocklage, Crnich, Jordan, Southwick and Schweinsburg2015; Sumner et al., Reference Sumner, Hagan, Grodstein, Roberts, Harel and Koenen2017).
While some individuals experience chronic or recurrent PTSD symptoms, others' symptoms may resolve over time, with or without treatment (Galatzer-Levy et al., Reference Galatzer-Levy, Ankri, Freedman, Israeli-Shalev, Roitman, Gilad and Shalev2013; Steinert, Hofmann, Leichsenring, & Kruse, Reference Steinert, Hofmann, Leichsenring and Kruse2015). A meta-analysis of 42 prospective studies on the natural course of PTSD found that, on average, nearly half (44%) of patients with PTSD experienced remission after three or more years from their initial assessment (Morina, Wicherts, Lobbrecht, & Priebe, Reference Morina, Wicherts, Lobbrecht and Priebe2014). However, the estimated remission rate largely varied across studies, likely due to heterogeneity in study design, including the time between the traumatic event and the first assessment, follow-up length and frequency, PTSD assessment (interview-based or self-reported), the definition of remission, types of traumatic events, and whether the participants sought for or received treatment (Morina et al., Reference Morina, Wicherts, Lobbrecht and Priebe2014; Steinert et al., Reference Steinert, Hofmann, Leichsenring and Kruse2015).
While PTSD is linked to many adverse health outcomes, much less is known about whether the elevated health risks persist after PTSD symptoms resolve. A few studies have investigated health trajectories after PTSD remission, with mixed results across outcomes. For example, in previous research, we found that women with remitted PTSD symptoms did not show significant differences in the risk of CVD compared to individuals with no trauma exposure, while elevated CVD risks were found in those with ongoing PTSD symptoms (N = 49 859) (Gilsanz et al., Reference Gilsanz, Winning, Koenen, Roberts, Sumner, Chen and Kubzansky2017). That study also found a positive association between PTSD symptom duration and CVD incidence (Gilsanz et al., Reference Gilsanz, Winning, Koenen, Roberts, Sumner, Chen and Kubzansky2017). Studies in Veterans Health Affairs (VHA) patients (N = 979–1598) found that clinically meaningful reductions in PTSD symptoms were associated with a lower risk of incident type 2 diabetes (Scherrer et al., Reference Scherrer, Salas, Norman, Schnurr, Chard, Tuerk and Lustman2019) and hypertension (Scherrer et al., Reference Scherrer, Salas, Friedman, Cohen, Schneider, Lustman and Schnurr2020a). Small prospective psychotherapy studies in patients with PTSD (N = 37–235) reported improvements in quality of life (Schnurr & Lunney, Reference Schnurr and Lunney2016), sleep quality (Rousseau et al., Reference Rousseau, Vallat, Coste, Cadis, Nicolas, Trousselard and Khalfa2021), and blood pressure (Schubert et al., Reference Schubert, Schreckenbach, Kirmeier, Gall-Kleebach, Wollweber, Buell and Schmidt2019) as PTSD symptoms reduced and/or remitted. However, some evidence suggests that individuals may still experience adverse health events after PTSD remission. For example, a study of 25 patients with PTSD treated with integrative trauma-focused cognitive behavioral therapy found that hypothalamic–pituitary–adrenal (HPA) axis feedback sensitivity remained unchanged after PTSD symptom improvement (Schubert et al., Reference Schubert, Schreckenbach, Kirmeier, Gall-Kleebach, Wollweber, Buell and Schmidt2019); another study in 1034 traumatically injured patients found that patients with resolved PTSD symptoms had poorer quality of life in various domains compared to those who never developed PTSD (Bryant et al., Reference Bryant, McFarlane, Silove, O'Donnell, Forbes and Creamer2016). In addition, VHA studies (N = 1079–1330) found that patients with a clinically meaningful reduction in PTSD symptoms within 12 months of the initial assessment were no different from those with less or no symptom reduction with regard to subsequent risks of CVD, ischemic heart disease, hyperlipidemia, and weight loss over a 2–7 year follow-up (Scherrer et al., Reference Scherrer, Salas, Friedman, Cohen, Schneider, Lustman and Schnurr2020a, Reference Scherrer, Salas, Schneider, Friedman, van den Berk-Clark, Chard and Cohen2020b).
It remains unclear whether and to what extent cognitive deficits ameliorate with the remission of PTSD symptoms, with findings inconsistent across studies and cognitive measures (Ben-Zion et al., Reference Ben-Zion, Fine, Keynan, Admon, Green, Halevi and Shalev2018; Clouston et al., Reference Clouston, Kotov, Pietrzak, Luft, Gonzalez, Richards and Bromet2016; Eren-Koçak, Kılıç, Aydın, & Hızlı, Reference Eren-Koçak, Kılıç, Aydın and Hızlı2009; Fani et al., Reference Fani, Kitayama, Ashraf, Reed, Afzal, Jawed and Bremner2009; Haaland, Sadek, Keller, & Castillo, Reference Haaland, Sadek, Keller and Castillo2016; Nijdam, Martens, Reitsma, Gersons, & Olff, Reference Nijdam, Martens, Reitsma, Gersons and Olff2018; Vermetten, Vythilingam, Southwick, Charney, & Bremner, Reference Vermetten, Vythilingam, Southwick, Charney and Bremner2003; Walter, Palmieri, & Gunstad, Reference Walter, Palmieri and Gunstad2010). One study of 813 World Trade Center responders found that current but not remitted PTSD was associated with cognitive impairment (Clouston et al., Reference Clouston, Kotov, Pietrzak, Luft, Gonzalez, Richards and Bromet2016). However, other studies have been limited by small samples (N < 100) of relatively younger individuals and failure to control adequately for potential confounders, such as age and education. In addition, most studies were conducted in clinical settings and assessed cognitive change in relation to PTSD symptom improvement over the course of treatment, thus providing limited information on how post-remission cognition compares to that of individuals who never had PTSD after trauma exposure.
The present study leverages the NHS II, a large prospective cohort of U.S. women, to examine how prior endorsement of PTSD symptoms, remitted or unresolved, are associated with cognitive function and changes over time measured six years after the PTSD assessment. We hypothesized that (a) unresolved PTSD symptoms would be associated with poorer cognitive function and worse changes in cognitive function over a 2-year follow-up compared to those who never had PTSD after trauma exposure, and (b) women whose PTSD symptoms remitted would have similar cognitive function and cognitive changes as women who did not have PTSD after trauma exposure.
Methods
Study participants
Data came from 116 429 U.S. female nurses enrolled in the NHS II in 1989, with biennial follow-up ongoing. Women responded to biennial questionnaires about lifestyle and health. In 2008, 60 804 women who returned a 2007 biennial questionnaire were invited to complete a supplemental questionnaire on trauma exposure and PTSD history; of these, 54 224 (89%) responded. In 2014, 43 957 of these responders with known email addresses were invited to an initial (henceforth ‘baseline’) cognitive assessment. A total of 15 138 women completed this assessment, with 14 994 passing data quality checks as previously described (Roberts et al., Reference Roberts, Liu, Lawn, Jha, Sumner, Kang and Koenen2022). We excluded 2724 women who reported no trauma exposure, a requirement for PTSD and subsequent remission, leaving 12 270 participants.
Following the baseline cognitive assessment, women were asked to complete additional cognitive assessments every 6 or 12 months for up to 24 months. Those who did not complete a cognitive assessment at one time point were not invited to the next session. Thus, women completed up to five assessments; over a median follow-up of 1.0 year, 7937 (64.7%) completed at least two assessments, with 625 (5.1%) completing all five. The study was approved by the Institutional Review Board of Brigham and Women's Hospital. Completion of assessments implied consent.
Trauma and PTSD assessment
In 2008, women reported lifetime exposure to 15 traumatic events and any other traumatic event not listed from a modified version of the Brief Trauma Questionnaire (Morgan et al., Reference Morgan, Hazlett, Wang, Richardson, Schnurr and Southwick2001; Schnurr, Vielhauer, Weathers, & Findler, Reference Schnurr, Vielhauer, Weathers and Findler1999). Trauma-exposed women were asked to indicate their worst event. Lifetime (ever) and past-month experience of seven PTSD symptoms with reference to their worst trauma was assessed using the Short Screening Scale for DSM-IV PTSD (Breslau, Peterson, Kessler, & Schultz, Reference Breslau, Peterson, Kessler and Schultz1999). PTSD symptom severity (lifetime and past-month) was defined as the number of symptoms reported. See online Supplementary Materials for more details.
We defined remission as reporting lifetime but no past-month PTSD symptoms; women who reported past-month symptoms were identified as having unresolved symptoms at the time of the assessment. Trauma-exposed women were classified into three groups indicating PTSD status as of 2008: (a) no PTSD: trauma exposed but no PTSD symptoms at any point in their lifetime, (b) remitted PTSD: any lifetime symptoms, but no past-month PTSD symptoms, and (3) unresolved PTSD: any past-month PTSD symptoms. In addition, we utilized a cutoff score of four on the PTSD screener to create more granular categories that incorporate symptom severity. In a validation study of this screener, a score of ⩾4 identified PTSD cases with sensitivity = 85% and specificity = 93% (Breslau et al., Reference Breslau, Peterson, Kessler and Schultz1999). Based on this cutoff, we further classified trauma-exposed women into five groups: (a) trauma/no PTSD symptoms, (b) remitted mild (1–3) PTSD symptoms, (c) remitted moderate-severe (4–7) PTSD symptoms, (d) unresolved mild (1–3) PTSD symptoms, (e) unresolved moderate-severe (4–7) PTSD symptoms.
Cognitive assessment
The Cogstate Brief Battery was used to assess cognitive function beginning in 2014. Cogstate is a self-administered online tool comprised of four tasks measuring psychomotor speed, attention, visual learning, and working memory (Fredrickson et al., Reference Fredrickson, Maruff, Woodward, Moore, Fredrickson, Sach and Darby2010; Koyama et al., Reference Koyama, Hagan, Okereke, Weisskopf, Rosner and Grodstein2015). This computerized instrument has good construct and criterion validity (Hammers et al., Reference Hammers, Spurgeon, Ryan, Persad, Barbas, Heidebrink and Giordani2012; Maruff et al., Reference Maruff, Thomas, Cysique, Brew, Collie, Snyder and Pietrzak2009), test-retest reliability (Fredrickson et al., Reference Fredrickson, Maruff, Woodward, Moore, Fredrickson, Sach and Darby2010; Lim et al., Reference Lim, Jaeger, Harrington, Ashwood, Ellis, Stöffler and Maruff2013a; Maruff et al., Reference Maruff, Lim, Darby, Ellis, Pietrzak and Snyder2013), acceptability and efficiency in large community-based studies of older individuals (Fredrickson et al., Reference Fredrickson, Maruff, Woodward, Moore, Fredrickson, Sach and Darby2010; Koyama et al., Reference Koyama, Hagan, Okereke, Weisskopf, Rosner and Grodstein2015), and clinical utility in identifying cognitive impairments and dementia (Hammers et al., Reference Hammers, Spurgeon, Ryan, Persad, Barbas, Heidebrink and Giordani2012; Lim et al., Reference Lim, Ellis, Harrington, Ames, Martins and Masters2012; Maruff et al., Reference Maruff, Lim, Darby, Ellis, Pietrzak and Snyder2013, Reference Maruff, Thomas, Cysique, Brew, Collie, Snyder and Pietrzak2009). Standardized z scores were calculated at all assessments using means and standard deviations at baseline for each task. Two composite scores were generated and analyzed as primary outcomes: psychomotor speed/attention composite and learning/working memory composite. For both composites, higher scores reflect better cognitive function. These two composites have been previously validated in our sample using confirmatory factor analysis (Sumner et al., Reference Sumner, Hagan, Grodstein, Roberts, Harel and Koenen2017) and have been found to be sensitive measures of cognitive decline (Lim et al., Reference Lim, Pietrzak, Ellis, Jaeger, Harrington, Ashwood and Maruff2013b; Maruff et al., Reference Maruff, Lim, Darby, Ellis, Pietrzak and Snyder2013). See online Supplementary Materials for more details.
Covariates
Analyses were adjusted for potential confounding factors for the association between PTSD status and cognition, including sociodemographic factors, biobehavioral factors, and health conditions. These included age at Cogstate baseline, self-identified race/ethnic identity (Non-Hispanic White, Black, Hispanic, Asian, other), parental education at participant's birth (high school or less, some college, 4 + years of college), and participant's highest education (associate's, bachelor's, master's, doctorate). Biobehavioral factors included body mass index [BMI, kg/m2, calculated from self-reported height and weight (Rimm et al., Reference Rimm, Stampfer, Colditz, Chute, Litin and Willett1990)], cigarette smoking (nonsmoker, former smoker, current smoker of 1–14, 15–24, or 25 + cigarettes/day), alcohol consumption (0, 1 − <5, 5 − <10, 10 − 20, 20 + grams/day), diet quality measured using the Alternative Healthy Eating Index (excluding alcohol consumption) (Chiuve et al., Reference Chiuve, Fung, Rimm, Hu, McCullough, Wang and Willett2012), ascertained in 2007, and physical activity (<3, 3 − <9, 9 − <18, 18 − <27, 27 + metabolic equivalent hours/week), ascertained in 2005. Lifetime history of physician-diagnosed hypertension, type 2 diabetes, stroke, and myocardial infarction was ascertained based on women's reports on biennial questionnaires from 1989–2007. While the assessment of biobehavioral factors and health conditions predated the PTSD assessment in 2008, some of these factors may be on the pathway between PTSD remission and cognition, as the exact timing of PTSD remission is not known. For example, good diet quality may positively impact the remission of PTSD symptoms, or it may be the result of PTSD remission. It was not possible to determine the sequence of PTSD remission and these health factors; thus, we cannot conclusively distinguish if they are serving as confounders or as mediators between the PTSD status in 2008 and cognition assessed six years later.
Past-week depressive symptoms and history of physician-diagnosed depression were analyzed in a secondary analysis, as they could be a cause or consequence of unresolved PTSD. Past-week depressive symptoms were assessed in 2008 using the 10-item Center for Epidemiologic Studies Depression (CES-D-10) scale. The CES-D-10 has excellent psychometrics and performs similarly to the 20-item CES-D (Andresen, Malmgren, Carter, & Patrick, Reference Andresen, Malmgren, Carter and Patrick1994). History of physician-diagnosed depression was ascertained from biennial questionnaires in 2003–2007.
Less than 5.0% of all covariates were missing, except for participant's education (24.4%), which was assessed by a supplemental 2018 questionnaire in a subset of participants. We addressed missing values in all covariates using multiple imputations (MI) with a fully conditional specification method (Brand, Reference Brand1999; van Buuren, Reference van Buuren2007), generating 10 imputed datasets. See online Supplementary Materials for more details.
Statistical analysis
For descriptive analyses, we summarized sociodemographic and health characteristics across PTSD groups using means and standard deviations and frequencies where appropriate. To comprehensively examine the relationship between PTSD status, symptom severity, and cognition, we modeled PTSD in three different ways in the association analysis. Firstly, we used the 3-level categorical variable to compare remitted and unresolved PTSD with those who had no PTSD ever (reference group). Secondly, we used a continuous score of lifetime and past-month PTSD symptoms to assess the linear relationship between symptom severity and cognition among women with remitted and unresolved PTSD symptoms, respectively. Thirdly, we utilized a 5-level categorical variable that combined both PTSD status and symptom severity, with trauma/no PTSD symptoms as the reference group.
To estimate the association between PTSD in 2008 and baseline cognition in 2014 (henceforth ‘cross-sectional analysis,’ with prospective measures of outcome), we used multivariate linear regression (PROC GLM in SAS v9.4). The beta coefficients for PTSD were reported, representing the estimated differences in baseline cognition compared to the reference group or per one unit increase in symptom severity. We also estimated cognitive differences associated with one year of aging by fitting linear regressions on cognitive composites with age as the predictor, adjusted for race/ethnicity (Roberts et al., Reference Roberts, Sumner, Koenen, Kubzansky, Grodstein, Rich-Edwards and Weisskopf2020b; Sumner et al., Reference Sumner, Hagan, Grodstein, Roberts, Harel and Koenen2017). To estimate the association between PTSD in 2008 and the rate of cognitive change during follow-up (henceforth ‘longitudinal analysis’), we used linear mixed-effects models (PROC MIXED in SAS v9.4), including time since baseline in years, PTSD, the interaction between time and PTSD, and covariates. Dependence between repeated measurements within study participants was accounted for via the inclusion of random intercepts and random slopes for time. The beta coefficients for the time-PTSD interaction terms were reported, representing the estimated differences in the rate of cognitive change compared to the reference group or per one unit increase in symptom severity. These models were fit to each of the 10 imputed datasets, with results combined using Rubin's rule (Rubin, Reference Rubin1987). Estimated associations with a p value less than 0.05 were considered statistically significant.
In both cross-sectional and longitudinal analyses, we started with a crude model with no covariates. Next, we included covariates in increasingly adjusted models. In the first adjusted model, we accounted for possible sociodemographic confounders (e.g. parental education). In the second adjusted model, we examined the impact of biobehavioral factors (e.g. BMI, smoking) on a possible association of PTSD status with cognition. Finally, to assess the impact of co-morbidities on the association, the third model was additionally adjusted for health conditions (e.g. diabetes, stroke).
We conducted several secondary analyses. First, since PTSD and depression are highly co-morbid (Rytwinski, Scur, Feeny, & Youngstrom, Reference Rytwinski, Scur, Feeny and Youngstrom2013), we further adjusted for depressive symptoms in 2008 and history of depression during 2003–2007 to examine if a PTSD-cognition association was accounted for by depression. Next, to evaluate the potential impact of practice effects in the cognitive tasks in the longitudinal analysis, we further adjusted for prior practice (Vivot et al., Reference Vivot, Power, Glymour, Mayeda, Benitez, Spiro and Gross2016). Moreover, to evaluate whether results were affected by the imputation of missing values, particularly participant's education, we explored several alternative imputation methods (e.g. single imputation, hot-deck imputation) and compared results. Finally, to examine potential bias due to selection into the Cogstate sample and loss to follow-up, we conducted inverse probability weighted (IPW) analyses. The 5-level categorical PTSD variable that incorporated both PTSD status and symptom severity was assessed in secondary analyses. See online Supplementary Materials for more details.
Results
Participant characteristics and PTSD status
A total of 12 270 trauma-exposed women were included in this study. Most were non-Hispanic White (95.9%), with age at baseline cognitive assessment ranging from 50 to 71 years (mean = 61). The prevalence of any PTSD symptoms was 67.3% (N = 8253); 2997 (24.4%) and 784 (6.4%) women had unresolved mild and moderate-severe symptoms at the time of PTSD assessment in 2008, respectively; 3347 (27.3%) and 1125 (9.2%) women reported remitted mild and moderate-severe symptoms, respectively. Participant characteristics by PTSD status are shown in Table 1. Compared to women whose PTSD symptoms had remitted as of 2008, those with unresolved PTSD symptoms had lower levels of parental education, had higher BMI, were more likely to be current smokers, and had higher levels of depressive symptoms. The distribution of trauma types and lifetime PTSD symptoms by PTSD status is shown in online Supplementary Table S1.
Participant characteristics by PTSD status (N = 12 270)
MET, metabolic equivalent of tasks; sx, symptoms.
Values are means (s.d.) for continuous variables or percentages (N) for categorical variables. Values of categorical variables may not sum to 100% due to rounding or missing data.
a Alternate Healthy Eating Index without alcohol consumption was used in the analysis; higher scores reflect better diet quality (possible range = 0–100).
b Center for Epidemiologic Studies Depression (CES-D) scale, short form (possible range = 0–30).
c History of clinician-diagnosed depression reported at the 2003–2013 questionnaires.
d History of clinician-diagnosed health conditions reported at the 1989–2013 questionnaires.
Cross-sectional analysis
Compared to trauma/no PTSD symptoms, unresolved PTSD symptoms at the time of 2008 assessment were associated with significantly lower scores in both cognitive composites assessed six years later. Specifically, in analyses without adjustment, women with any unresolved PTSD symptoms showed mean differences of −0.10 standard deviation (s.d.) (95% CI −0.14 to −0.06) in psychomotor speed/attention and −0.10 s.d. (95% CI −0.13 to −0.06) in learning/working memory, compared to women with no PTSD symptoms ever (online Supplementary Table S2). Among women with unresolved symptoms, one unit increase in past-month PTSD symptom severity was associated with −0.02 (95% CI −0.04 to 0.001) and −0.04 s.d. (95% CI −0.05 to −0.02) mean differences in psychomotor speed/attention and learning/working memory, respectively (Table 2). Using a cutoff of four, we found mean differences between women with unresolved mild (1–3) PTSD symptoms and women with trauma/no PTSD symptoms of −0.09 s.d. (95% CI −0.13 to −0.05) in psychomotor speed/attention and −0.07 s.d. (95% CI −0.11 to −0.04) in learning/working memory; while for women with unresolved moderate-severe (4–7) PTSD symptoms, these mean differences were −0.14 s.d. (95% CI −0.21 to −0.07) in psychomotor speed/attention and −0.18 s.d. (95% CI −0.23 to −0.12) in learning/working memory (Table 3, online Supplementary Fig. S1). To help interpret these differences, we found that each additional year of age was associated with a −0.04 s.d. decrease in psychomotor speed/attention and a −0.03 s.d. decrease in learning/working memory; thus, unresolved mild and moderate-severe PTSD symptoms were associated with the equivalent of approximately two and five years of cognitive aging, respectively.
Association between PTSD symptom severity and Cogstate composite scores at baseline
Beta coefficients of PTSD groups from linear regression models are shown, representing the difference in baseline Cogstate composites for every unit increase in symptom severity.
Model 0: no covariates.
Model 1: age at baseline cognitive assessment, race/ethnicity, parental education, participant education.
Model 2: Model 1 + body mass index, smoking status, alcohol use, physical activity, diet quality.
a Analysis was restricted to women with remitted PTSD symptoms as of 2008. Lifetime PTSD symptom severity was modeled as the predictor.
b Analysis was restricted to women with unresolved PTSD symptoms as of 2008. Past-month PTSD symptom severity was modeled as the predictor.
Association between PTSD status (5 levels) and Cogstate composite scores at baseline
Beta coefficients of PTSD groups from linear regression models are shown, representing the difference in baseline Cogstate composites compared to trauma-exposed women with no PTSD. sx, symptoms.
Model 0: no covariates.
Model 1: age at baseline cognitive assessment, race/ethnicity, parental education, participant education.
Model 2: Model 1 + body mass index, smoking status, alcohol use, physical activity, diet quality.
We observed no significant difference in baseline cognition between women with remitted PTSD symptoms as of 2008 compared to the women with no PTSD symptoms ever (online Supplementary Table S2, Table 3). Among women with remitted PTSD symptoms, increase in lifetime PTSD symptom severity was not associated with baseline cognition (Table 2). Above results on unresolved and remitted PTSD were unchanged after adjusting for sociodemographic and biobehavioral factors (Tables 2, 3, online Supplementary Table S2) and health conditions (data not shown).
Longitudinal analysis
Compared to women with no PTSD symptoms, worse rates of changes in learning/working memory were observed in both women with unresolved PTSD (mean difference = −0.03 s.d./year, 95% CI −0.06 to −0.01) and remitted PTSD (mean difference = −0.02 s.d./year, 95% CI −0.05 to 0.0005) (online Supplementary Table S3). Specifically, unresolved mild (1–3) symptoms at the time of PTSD assessment had a marginally significant association with worse rates of changes in learning/working memory (mean difference = −0.03 s.d./year, 95% CI −0.05 to 0.0004) that further reached statistical significance when adjusted for sociodemographic and biobehavioral factors (Table 4). In addition, unresolved moderate-severe (4–7) PTSD symptoms were significantly associated with worse rates of changes in learning/working memory (mean difference = −0.05 s.d./year, 95% CI −0.09 to −0.01), and the results remained unchanged with the adjustment of covariates (Table 4, online Supplementary Fig. S1). Among women with unresolved PTSD symptoms, increase in past-month symptom severity was not linearly associated with the rate of change in learning/working memory (online Supplementary Table S4).
Association between PTSD status (5 levels) and rate of change in Cogstate composite scores
Beta coefficients of the time × PTSD interaction terms from the linear mixed-effects models are shown, representing the difference in 1-year change in Cogstate composites compared to trauma-exposed women with no PTSD. sx, symptoms.
Model 0: no covariates.
Model 1: age at baseline cognitive assessment, race/ethnicity, parental education, participant education.
Model 2: Model 1 + body mass index, smoking status, alcohol use, physical activity, diet quality.
By contrast, we observed a significant linear association between increase in lifetime symptom severity and changes in learning/working memory among women with remitted PTSD symptoms (online Supplementary Table S4). Specifically, one unit increase in lifetime symptom severity was associated with a mean difference of −0.01 s.d./year (95% CI −0.02 to −0.004) in the rate of change in learning/working memory. Using a cutoff of four, we observed significantly worse cognitive changes in learning/working memory for women with remitted moderate-severe (4–7) PTSD symptoms (−0.06 s.d./year, 95% CI −0.10 to −0.03) compared to women with no PTSD symptoms, but not with remitted mild (1–3) PTSD symptoms (−0.01 s.d./year, 95% CI −0.03 to 0.02) (Table 4, online Supplementary Fig. S1). No significant association was observed between PTSD status or symptom severity and psychomotor speed/attention (Table 4, online Supplementary Tables S3, S4). Results were unchanged after controlling for sociodemographic and biobehavioral factors (Table 4, online Supplementary Tables S3, S4) and health conditions (data not shown).
Secondary analyses
Adjustment for depression
In the cross-sectional analysis, the estimated decrease in cognitive composite scores for women with unresolved PTSD symptoms, compared to those with no PTSD, remained statistically significant after adjusting for depression, but estimates were attenuated by 25.0 to 53.8%, depending on the model (online Supplementary Table S5). In the longitudinal analysis, including the main effects of depression did not change the results (online Supplementary Table S6). However, when further adjusted for time × depression interactions, the associations of unresolved PTSD symptoms with changes in learning/working memory attenuated by 22.5 to 29.5% and were no longer statistically significant. The association of remitted moderate-severe PTSD symptoms with the rate of change in cognition was similar to that in the unadjusted analyses (online Supplementary Table S6).
Adjustment for practice effects
Longitudinal results were consistent after adjusting for practice effects and were robust to the different parameterization of practice effects, including the categorical and the continuous number of prior tests, the square root of the number of prior tests, and the indicator for baseline v. post-baseline assessments (online Supplementary Table S7). Cognitive composite scores decreased over time in all PTSD groups after adjusting for the number of prior tests (categorical) (Fig. 1); the estimated rate of decline in learning/working memory was 0.07 s.d./year among women with trauma/no PTSD, and nearly twice as fast among women with remitted or unresolved moderate-severe PTSD symptoms (0.13 s.d./year and 0.12 s.d./year respectively).
Fitted linear cognitive trajectories by PTSD status (5 levels) in 2008. Plots were based on results from models adjusted for age at baseline cognitive assessment, race/ethnicity, parental education, participant education, and categorical indicator of the number of prior cognitive tests. Fitted linear trajectories were calculated for Non-Hispanic White women ages 61 at baseline, with the highest education of Bachelor's degree and parental education of high school. sx, symptoms.
Alternative imputation of participant education
Results were consistent across different imputation methods, including group-specific mode, hot-deck imputation, missing indicator, and single imputation (online Supplementary Tables S8, S9). As single imputation results were similar to MI, we conducted the following IPW analyses in one of the imputed datasets.
Selection into Cogstate and IPW
Women in this study were similar to those in the original PTSD cohort (N = 44 573) from which the cognitive samples were drawn in many characteristics, although the former had slightly better health behaviors and fewer diseases (online Supplementary Table S10). Accounting for the probability of selection from the original PTSD cohort into the Cogstate sample using IPW, overall results remained the same except for a mild attenuation in the association between unresolved mild PTSD symptoms and the rate of change in learning/working memory in the longitudinal analysis (−0.02 s.d./year, 95% CI −0.05 to 0.00, online Supplementary Table S11).
Loss to follow-up during Cogstate assessments and IPW
During the longitudinal assessment period, women who continued to the next assessment had higher cognition on average than those who did not respond to the next assessment, particularly between the first and the second assessments and in learning/working memory (online Supplementary Fig. S2). PTSD status was not associated with the probability of dropout (online Supplementary Fig. S3). Sample characteristics were similar across women who completed different numbers of cognitive assessments (online Supplementary Table S12). After accounting for loss to follow-up using IPW, the difference in the change of learning/working memory composite compared to women with trauma/no PTSD remained significant for women with remitted or unresolved moderate-severe PTSD symptoms; however, it was attenuated and no longer significant for women with unresolved mild PTSD symptoms (p = 0.14, online Supplementary Table S13).
Discussion
In this cohort study, unresolved PTSD symptoms as of 2008 were associated with significantly poorer cognitive function at baseline in 2014 in psychomotor speed/attention and learning/working memory and faster cognitive decline during follow-up in learning/working memory, compared to the reference group of women with trauma/no PTSD. In women whose PTSD remitted by 2008, while we found no significant difference in baseline cognition compared to women with trauma/no PTSD, increase in lifetime PTSD symptom severity was significantly associated with a faster decline in learning/working memory. The observed associations were independent of sociodemographic, biobehavioral, and health factors, and were robust to adjustment for practice effects, imputation of missing values, sample selection, and loss to follow-up. Co-morbid depression partially accounted for the associations between PTSD and cognition, potentially due to the overlapping symptoms between the two conditions. Our findings add to the literature on the cognitive sequelae of PTSD and provide a more granular evaluation of the dynamic association between PTSD and cognition.
Similar to findings from previous research in mixed-gender samples (Clouston et al., Reference Clouston, Kotov, Pietrzak, Luft, Gonzalez, Richards and Bromet2016; Eren-Koçak et al., Reference Eren-Koçak, Kılıç, Aydın and Hızlı2009), our cross-sectional analysis found that cognitive function did not differ in women with remitted PTSD compared to those who never had PTSD. These findings may be interpreted in two ways. First, PTSD is associated with adverse cognitive outcomes that can become alleviated as PTSD symptoms improve. The DSM-5 criteria for PTSD include cognitive symptoms such as attention and memory problems (American Psychiatric Association, 2013), and previous research has shown that these cognitive symptoms improve as individuals recover from PTSD (Ben-Zion et al., Reference Ben-Zion, Fine, Keynan, Admon, Green, Halevi and Shalev2018; Nijdam et al., Reference Nijdam, Martens, Reitsma, Gersons and Olff2018). In addition, PTSD remission may be accompanied by behavioral and biological changes that contribute to improved cognition, such as improved quality of life (Schnurr & Lunney, Reference Schnurr and Lunney2016), sleep quality (Rousseau et al., Reference Rousseau, Vallat, Coste, Cadis, Nicolas, Trousselard and Khalfa2021), blood pressure (Schubert et al., Reference Schubert, Schreckenbach, Kirmeier, Gall-Kleebach, Wollweber, Buell and Schmidt2019), and hormone levels (Olff, de Vries, Güzelcan, Assies, & Gersons, Reference Olff, de Vries, Güzelcan, Assies and Gersons2007a), as well as reduced DNA strand breakage (Morath et al., Reference Morath, Moreno-Villanueva, Hamuni, Kolassa, Ruf-Leuschner, Schauer and Kolassa2014). Some neuroimaging studies of PTSD also found changes in functional activation and neurocircuitry mechanisms in response to psychotherapies (Cisler et al., Reference Cisler, Sigel, Kramer, Smitherman, Vanderzee, Pemberton and Kilts2016a, Reference Cisler, Sigel, Steele, Smitherman, Vanderzee, Pemberton and Kilts2016b; Helpman et al., Reference Helpman, Marin, Papini, Zhu, Sullivan, Schneier and Neria2016; Simmons, Norman, Spadoni, & Strigo, Reference Simmons, Norman, Spadoni and Strigo2013), and specific changes in brain structural network among individuals with remitted PTSD that had been linked to improved cognitive processing (Sun et al., Reference Sun, Davis, Haswell, Swanson, LaBar, Fairbank and Morey2018).
Second, we cannot rule out the possibility that lower cognition preceded and predicted PTSD remission. Pre-trauma or pre-PTSD cognitive ability has been shown to be associated with PTSD onset and severity (Aupperle, Melrose, Stein, & Paulus, Reference Aupperle, Melrose, Stein and Paulus2012; Scott et al., Reference Scott, Matt, Wrocklage, Crnich, Jordan, Southwick and Schweinsburg2015), as well as PTSD trajectories (Samuelson et al., Reference Samuelson, Newman, Abu Amara, Qian, Li, Schultebraucks and Marmar2020) and treatment outcomes (Haaland et al., Reference Haaland, Sadek, Keller and Castillo2016; Nijdam, de Vries, Gersons, & Olff, Reference Nijdam, de Vries, Gersons and Olff2015; van Rooij, Kennis, Vink, & Geuze, Reference van Rooij, Kennis, Vink and Geuze2016). However, in this study, cognitive function was only measured after PTSD assessment; thus, we could not determine whether cognitive function predicted PTSD remission. Future research with pre- and post-PTSD assessments of cognition will be necessary for a comprehensive characterization of the cognitive trajectory over the course of PTSD.
While women with remitted PTSD showed no difference in cognitive function assessed six years later from the reference group of women who never had PTSD, our longitudinal analysis indicated that women who had previously experienced high levels of PTSD symptoms were still prone to a more rapid decline in learning/working memory after symptom remission, at a rate comparable to that observed in women with unresolved symptoms of similar levels. These findings indicate that there may be lingering cognitive effects even after the remission of PTSD symptoms and suggest the potential need for continued support among women who have experienced high levels of PTSD symptoms in the past. PTSD is associated with dysregulated physiological and behavioral responses, including sleep problems (U.S. Department of Veterans Affairs, 2021), substance abuse (Brady, Back, & Coffey, Reference Brady, Back and Coffey2004), and many other chronic illnesses. These adverse health outcomes are more severe among individuals with higher levels of PTSD or persistent PTSD (D'Andrea, Sharma, Zelechoski, & Spinazzola, Reference D'Andrea, Sharma, Zelechoski and Spinazzola2011); thus, such individuals may experience more potent, longer-lasting effects. Indeed, one study found that remitted PTSD patients still experienced poorer physical, psychological, and social functioning than those who never developed PTSD (Bryant et al., Reference Bryant, McFarlane, Silove, O'Donnell, Forbes and Creamer2016). Furthermore, there is also evidence from previous research suggesting that PTSD may leave biological traces that can persist following symptom remission. For example, a neuroimaging study in U.S. military veterans showed that not all brain network architecture changes associated with PTSD returned to the level of trauma-exposed controls after PTSD remission (Sun et al., Reference Sun, Davis, Haswell, Swanson, LaBar, Fairbank and Morey2018); another study found that HPA axis feedback sensitivity remained unchanged after PTSD symptoms abated (Schubert et al., Reference Schubert, Schreckenbach, Kirmeier, Gall-Kleebach, Wollweber, Buell and Schmidt2019). Together with previous evidence, our findings imply that the adverse health consequences of PTSD may not fully resolve following PTSD remission, which may leave individuals with a greater likelihood of an impaired aging process. However, no other longitudinal studies have examined cognitive decline during aging in relation to remitted and unresolved PTSD. Therefore, additional studies are warranted to replicate our findings.
This study builds upon previous research indicating that lifetime PTSD is associated with poorer cognition and faster cognitive decline by providing a more detailed characterization of this relationship regarding the remission of PTSD symptoms. Furthermore, in this study, PTSD was associated with greater impairment in learning/working memory compared to psychomotor speed/attention and accelerated decline in learning/working memory only. As age, mild cognitive impairment, and Alzheimer's disease (AD) also associate with greater impairments in learning/working memory compared to psychomotor speed/attention (Lim et al., Reference Lim, Ellis, Harrington, Ames, Martins and Masters2012; Maruff et al., Reference Maruff, Lim, Darby, Ellis, Pietrzak and Snyder2013), it is possible that cognitive domains that represent earlier or more severe deficits in the aging process are particularly susceptible to the effects of PTSD. Further research is needed to understand the biological basis of these PTSD-related cognitive differences and how they may predict the risk of AD.
Our study has several limitations. First, our definitions of remitted and unresolved PTSD were limited by the data available. PTSD can take various courses across individuals, including stable remitting, fluctuating, or chronic (Galatzer-Levy et al., Reference Galatzer-Levy, Ankri, Freedman, Israeli-Shalev, Roitman, Gilad and Shalev2013; Marmar et al., Reference Marmar, Schlenger, Henn-Haase, Qian, Purchia, Li and Kulka2015; Steinert et al., Reference Steinert, Hofmann, Leichsenring and Kruse2015). However, information about PTSD symptom onset and fluctuations over time was unavailable. Thus, PTSD status measured at a single time point in 2008 may not accurately reflect the duration or chronicity of PTSD symptoms or PTSD status at the time of cognitive assessment. In addition, not all the symptoms that make up the diagnostic criteria of PTSD were assessed, and limited information on treatment-seeking made it challenging to distinguish between natural remission v. remission from treatment. Future research that integrates more frequent and comprehensive assessment of PTSD symptoms will allow for a thorough investigation of the dynamic changes in symptom severity over time. Second, this sample consisted of individuals in the early years of aging, and the follow-up was relatively short, which may have limited our ability to detect modest associations between remitted PTSD and cognition. Thus, further research with longer follow-ups to older ages will be needed. Third, the presence of emotional symptoms, such as loss of interest and feeling jumpy, among women with unresolved symptoms at the time of the PTSD assessment could be a sign of early developing underlying neurodegenerative diseases; however, we could not directly assess this possibility due to the lack of data on neurodegenerative disease status. Fourth, observational studies are subject to confounding and selection bias. While our analysis was robust to a series of covariate adjustment and secondary analyses, unmeasured confounders or competing risks may partially account for the observed results. Fifth, while the traumatic experiences included in this study (such as sexual harassment and the death of a close family member or friend) are not specific to nurses, the NHS II cohort is predominantly white professionals; further research is needed to determine whether the findings generalize to more diverse populations.
Our study has many strengths. First, by leveraging data from a large, prospective cohort, we were able to analyze the largest sample to date, to the best of our knowledge, assessing the relationship of PTSD symptom remission with cognition. Second, we conducted both cross-sectional and longitudinal analysis, providing a comprehensive evaluation of the dynamic changes in cognition in relation to PTSD status. Third, the study was comprised of women. While women have a higher risk of PTSD than men (Christiansen & Berke, Reference Christiansen and Berke2020), recover slower from PTSD (Breslau, Reference Breslau2009), and experience higher rates of dementia (Snyder et al., Reference Snyder, Asthana, Bain, Brinton, Craft, Dubal and Carrillo2016), they have been underrepresented in PTSD-cognition research.
To summarize, we found that unresolved but not remitted PTSD was associated with worse cognitive function assessed six years later and that accelerated cognitive decline was observed during follow-up among women with either unresolved or remitted high levels of PTSD symptoms. These findings suggest the value of integrating cognitive assessments and interventions into routine mental health care for PTSD patients and providing resources and support for managing cognitive decline in this population. The observed accelerated cognitive decline may imply poorer cognitive aging and highlights the importance of screening and prevention of dementia not only for women experiencing persistent PTSD symptoms but also for those who have previously experienced high levels of PTSD symptoms. Co-occurring depression partially explained the link between unresolved PTSD symptoms and cognition, emphasizing the importance of monitoring and managing depression in PTSD treatment. Future research is needed to illuminate the pathways underlying the association of PTSD with worse cognitive trajectories and to determine whether effective PTSD treatment can prevent cognitive decline and other long-term adverse health outcomes associated with PTSD.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723001915
Acknowledgements
We acknowledge the Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital, and Harvard Medical School for managing the Nurses' Health Study II (NHS II).
Financial support
This study was supported by the National Institute of Mental Health (NIMH) grants R01MH101269 (to K Koenen and L Kubzansky), R01MH078928 (to K Koenen), 1R21MH102570 (to K Koenen and F Grodstein), as well as the National Institute of Health (NIH) grant U01 CA176726 (for NHS II infrastructure). L Sampson was supported by NIH grant T32 HL098048. Participant education was collected under R01MH101269 and used in the analysis of this manuscript. Data on this variable only were submitted to the NIMH Data Archive (NDA). NDA is a collaborative informatics system created by the NIH to provide a national resource to support and accelerate research in mental health. Dataset identifier: 10.15154/ph8a-3e55. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or of the Submitters submitting original data to NDA.
Competing interest
Dr Sumner received grants from the National Heart, Lung, and Blood Institute of the National Institutes of Health (NIH) and the US Department of Defense outside the submitted work. Dr Kubzansky received grants from the NIH during the conduct of the study and grants from the NIH and the US Department of Defense and royalties from Oxford University Press outside the submitted work. Dr Koenen received grants from Cohen Veterans Bioscience and the NIH during the conduct of the study; personal fees from BakerHostetler, Capita, Converys, Discovery Vitality, the European Central Bank, and the US Department of Justice; and royalties from Guilford Press and Oxford University Press outside the submitted work. Dr Roberts received grants from the NIH during the conduct of the study outside the submitted work. Dr Kang received grants from the NIH and Pfizer, Inc during the conduct of the study outside the submitted work. The authors do not have any other conflicts of interest to declare.
