Posttraumatic Stress Disorder Is Associated With Elevated Risk of Incident Stroke and Transient Ischemic Attack in Women Veterans

Abstract

Background

Posttraumatic stress disorder (PTSD) has been associated with ischemic heart disease in women veterans, but evidence for associations with other cardiovascular disorders remains limited in this population. This retrospective longitudinal cohort study evaluated the association of PTSD with incident stroke/transient ischemic attack (TIA) in women veterans.

Methods and Results

Veterans Health Administration electronic health records were used to identify women veterans aged ≥18 years engaged with Veterans Health Administration health care from January 1, 2000 to December 31, 2019. We identified women veterans with and without PTSD without a history of stroke or TIA at start of follow‐up. Propensity score matching was used to match groups on age, race or ethnicity, traditional cardiovascular risk factors, female‐specific risk factors, a range of mental and physical health conditions, and number of prior health care visits. PTSD, stroke, TIA, and risk factors used in propensity score matching were based on diagnostic codes. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs for associations of PTSD with an incident stroke/TIA composite. Subanalyses considered stroke and TIA separately, plus age‐ and race‐ or ethnicity‐stratified analyses were carried out.

The analytic sample included 208 092 women veterans (104 046 with and 104 046 without PTSD). PTSD was associated with a greater rate of developing stroke/TIA (HR, 1.33 [95% CI, 1.25–1.42], P<0.001). This elevated risk was especially pronounced in women <50 years old and in Hispanic/Latina women.

Conclusions

Findings indicate a strong association of PTSD with incident stroke/TIA in women veterans. Research is needed to determine whether addressing PTSD and its downstream consequences can offset this risk.

Nonstandard Abbreviations and Acronyms

PSM

propensity score matching

VHA

Veterans Health Administration

XGB

Extreme gradient boosting

Clinical Perspective.

What Is New?

• Using longitudinal electronic health record data, we demonstrated that posttraumatic stress disorder was associated with a greater rate of developing stroke (particularly ischemic and undefined stroke) and transient ischemic attack in a large sample of women veterans

• Posttraumatic stress disorder emerged as an especially potent risk factor for stroke/transient ischemic attack for women veterans <50 years of age and for Hispanic/Latina women veterans

What Are the Clinical Implications?

• Findings highlight the potential benefit of screening for stroke/transient ischemic attack risk in women veterans with posttraumatic stress disorder and emphasize the importance of integrated mental and cardiovascular health care after trauma

Stroke is the fifth leading cause of death in the United States, with 1 individual estimated to die from stroke every 3.5 minutes. ¹ , ² Stroke prevalence is estimated at 3.2% of the adult US population, with ≈795 000 individuals experiencing a new or recurrent stroke event each year in the United States. ³ The effects of stroke can be long lasting, potentially resulting in physical, functional, mental, or cognitive impairment. ³ , ⁴ , ⁵ , ⁶ Additionally, transient ischemic attacks (TIAs) are increasingly recognized as clinical phenomena that result from ischemia and can produce lasting brain injury. ⁷

Multiple disparities have been observed in rates of stroke/TIA and related outcomes. For example, lifetime risk of stroke is higher in women than in men, ³ and each year, more women die of stroke compared with men. ¹ , ² Women also have worse functional outcomes after stroke than men. ⁸ Additionally, compared with non‐Hispanic White individuals, higher rates of stroke have been observed in non‐Hispanic Black and Hispanic/Latino individuals. ⁹ , ¹⁰ , ¹¹ Differences in the burden of stroke risk factors such as hypertension, dyslipidemia, diabetes, smoking, physical inactivity, and atrial fibrillation may explain, in part, excess stroke vulnerability in women and individuals of minoritized racial and ethnic groups. ⁸ , ¹¹ Female‐specific risk factors for stroke have been identified as well, including gestational hypertension and diabetes, earlier age at menarche and menopause, and use of hormone therapy. ¹²

Growing research has also linked psychosocial factors to stroke, and these factors may be particularly relevant for cardiovascular health in women. ¹³ , ¹⁴ Contingent on experiencing a traumatic event, posttraumatic stress disorder (PTSD) has been linked to risk of cardiovascular disease (CVD), including stroke. ¹⁵ The majority of individuals (50%–89%) will experience trauma over their lifespan, ¹⁶ , ¹⁷ and PTSD is twice as common in women than in men in the general population, with ≈1 in 10 women and 1 in 20 men developing PTSD in their lifetime. ¹⁶ Moreover, PTSD is more prevalent in veterans of wars, especially those involved in combat. ¹⁸ , ¹⁹ , ²⁰ , ²¹

Across several longitudinal studies, PTSD has been found to precede and predict incident stroke or TIA. ²² , ²³ , ²⁴ , ²⁵ , ²⁶ This research has been conducted in large population‐based health registry samples of men and women, ²² , ²⁴ disaster first responders, ²⁶ community‐dwelling women, ²⁵ and predominantly (≈88%) male samples of younger veterans. ²³ , ²⁷ However, despite rapidly growing numbers of women veterans, research on PTSD and stroke/TIA in this population is relatively lacking. This empirical gap is notable, as women veterans have been identified as an important, yet underrepresented, population in cardiovascular research and care. Indeed, there have been calls to action to focus research specifically on the unique set of cardiovascular risks faced by women veterans. ²⁸ Notably, women veterans have rates of traditional and psychosocial cardiovascular risk factors, including trauma exposure and PTSD, that are greater compared with either civilian women or men veterans. ²⁸ Women veterans are also younger and more diverse with respect to race and ethnicity than their male counterparts ²⁹ and thus may carry distinct stroke risks compared with men veterans.

In the current study, we addressed this gap by examining the extent to which PTSD was associated with incident stroke, including ischemic and hemorrhagic subtypes, and TIA specifically in women veterans, the fastest growing group within users of the Veterans Health Administration (VHA) health care system. ²⁹ Furthermore, we performed stratified analyses based on age and race or ethnicity to evaluate potential subgroup differences given the more diverse nature of the women veteran population with respect to these demographic factors.

Methods

Data Source

Because of the sensitive nature of the data presented in this article, on request to the corresponding author, data will be available to approved individuals cleared by the National Data Service Research and Development, and the privacy officer at the Department of Veterans Affairs. In this retrospective longitudinal study, national VHA electronic health records (EHRs), including outpatient, inpatient, and purchased care (ie, services received from external clinicians but paid for by the VHA), were used to identify the study cohort of women veterans and obtain relevant variables (as described in a later section). VHA Health Factor data were also used to define smoking status. This study was approved by the University of California, Los Angeles Institutional Review Board. As a retrospective longitudinal analysis of the medical records, consent was not required. Further, we followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines in preparing this article.

Study Cohort

The VHA Corporate Data Warehouse ³⁰ was used to identify a cohort of women veterans aged 18 years or older who engaged with the VHA health care system from January 1, 2000 to December 31, 2019 (Figure 1). Women who did not pass data quality control (eg, invalid death date or age), and those without any encounters after their first health care encounter during the study period, were excluded. An index visit date marking start of follow‐up was selected as either the first visit with a PTSD diagnosis for women with PTSD or the first visit occurring at least 1 year after the initial health care encounter for women without PTSD. The 12‐month period preceding the index visit date served as a baseline, whereby data on clinical risk factors for stroke and TIA were assessed. Women with <12 months of baseline data were excluded, as were women who diagnosed with PTSD, stroke, or TIA before or during this baseline period. The remaining 630 710 women comprised the prematch sample.

Figure 1. Consolidated Standards of Reporting Trials diagram for creation of the study cohort.

PTSD indicates posttraumatic stress disorder; and TIA, transient ischemic attack.

Variable Definitions

Exposure

Exposure was defined as having a PTSD diagnosis (ever versus never). Consistent with prior VHA EHR research, ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ PTSD was defined using International Classification of Diseases, Ninth Revision (ICD‐9) or Tenth Revision (ICD‐10) diagnosis codes. We required PTSD documentation at 1 or more inpatient or 2 or more outpatient visits. ³¹ , ³² , ³³ , ³⁵ This approach to defining PTSD based on these EHR specifications has been found to have high sensitivity (0.99) and specificity (0.96). ³³

Outcomes

Our primary outcome was a composite of incident stroke (ischemic or hemorrhagic) or TIA. Similar to the approach described for PTSD, stroke and TIA were defined by at least 1 inpatient or at least 2 outpatient documentations of ICD‐9 or ICD‐10 diagnosis codes. This approach increases the precision of identifying events, especially in outpatient records, and has been used to define stroke and TIA in prior VHA EHR research. ²³ In addition to the composite stroke/TIA outcome, we examined incident stroke, including ischemic, hemorrhagic, and otherwise undefined stroke subtypes, and incident TIA separately as secondary outcomes.

Demographic and Cardiovascular Risk Factors

Demographic factors (age, race or ethnicity), VHA visit characteristics (year of index visit, number of health care encounters during the 12‐month baseline), traditional CVD risk factors (hyperlipidemia, hypertension, diabetes, smoking), female‐specific risk factors (gestational hypertension and diabetes, placental disorders, preeclampsia), and other mental and physical health conditions (anxiety, depression, schizophrenia, alcohol dependence, nonalcohol drug dependence, obesity, chronic kidney disease, hypothyroidism, hyperthyroidism, adrenal and pituitary disorders) were extracted from the EHR and used in propensity score matching (PSM). Age (continuous, plus grouped as 18–29, 30–39, 40–49, 50–59, 60+ years for stratified analyses) and race or ethnicity (categorized as American Indian or Alaskan Native, Asian or Pacific Islander, Hispanic/Latina, non‐Hispanic Black, non‐Hispanic White, unknown) were self‐reported. Year of index visit, along with the number of health care encounters during the baseline 12‐month period, was also extracted from the EHR. As detailed in our paper on PTSD and ischemic heart disease in women veterans, ³¹ ICD codes were used to define cardiovascular risk factors in the 12 months before and on the index visit date, using the same inpatient and outpatient validation criteria described previously, with 2 exceptions. Obesity was defined by ICD codes or body mass index ≥30, calculated as the most recent weight (in kilograms) at or before the index visit divided by the mean height (in square meters). ³² Smoking status was categorized as current or not current/unknown at index visit based on Health Factors data. ³⁸ , ³⁹ , ⁴⁰

Statistical Analysis

To derive estimates of the marginal association of PTSD with incident stroke and TIA among women with PTSD diagnosis, we used PSM. Propensity scores were estimated for all 630 710 women meeting study criteria outlined in Figure 1 using extreme gradient boosting (XGB), ⁴¹ a tree‐based machine learning algorithm that iteratively models residuals to improve fit. ⁴² Such machine learning techniques have been demonstrated to produce less biased estimates than standard parametric and semiparametric techniques because they are less prone to misspecification (eg, overfitting, assumptions of linearity). ⁴³ To deter overfitting of the propensity model, we randomly partitioned the sample into training and test sets along a 70/30 split. The training set was used to develop and tune the model, which included as predictors the aforementioned demographic and clinical variables, and the test set was used to evaluate the models' performance on heretofore unseen data. K‐fold, specifically 10‐fold, cross‐validation was applied to the training set to derive optimal values of the XGB hyperparameters corresponding to eta, maximum tree depth, ratio of predictors subsampled, percentage of observations subsampled, and boosting iterations. Hyperparameters were optimized via grid search on accuracy of predictions. Once the optimal model hyperparameters were determined through the cross‐validation process, the model was applied to the full training and test sets, and Cohen's kappa agreement between observed and predicted values was calculated for each. According to Cohen, ⁴⁴ kappa values near 0 indicate no agreement, 0.41–0.60 moderate agreement, and 0.81–1.00 almost perfect agreement. As a final step, the XGB model was used to generate propensity scores representing predicted probabilities of PTSD status for the entire matched sample.

Nearest neighbor matching was performed on the XGB propensity scores, using a caliper of 0.25 and restricting matches to women with no more than a 2‐year age difference. Standardized mean differences were used to confirm covariate balance among women veterans with and without PTSD. To assess the marginal association of PTSD with the stroke/TIA composite among women veterans, we conducted Cox survival models that accounted for the shared variance among matched pairs using the grouped jackknife method. ⁴⁵ The Fine–Gray method ⁴⁶ was used to account for the competing risk of death. The proportional hazards assumption was assessed via examination of Schoenfeld residual plots and tests of proportionality, where statistical significance indicates potential violation. We first modeled time until all stroke/TIA events as a function of PTSD status, with censoring applied to patients who died or were not diagnosed with stroke or TIA before end of study. We then modeled stroke and TIA events separately, followed by separate models for stroke subtypes (ischemic, hemorrhagic, undefined). As a final step, we conducted age‐ and race‐ or ethnicity‐stratified analyses of the association of PTSD with the composite stroke/TIA outcome.

Analyses were conducted using R, version 4.1.2. XGB and k‐fold cross‐validation were conducted using the caret package for R, version 6.0–94. Nearest neighbor PSM was conducted using the MatchIt package, version 4.5.3, and Cox models were conducted using the survival package, version 3.5–5. A 2‐sided P value<0.05 indicated statistical significance.

Results

Participant Characteristics

Baseline characteristics and standardized mean differences for the prematch and matched samples of women veterans with and without PTSD are displayed in Table 1. In contrast to the prematch sample, the matched sample achieved good balance across all covariates, with corresponding standardized mean differences <0.10. The final XGB model used to produce propensity scores for matching demonstrated moderate fit on both the training and test sets (both Cohen's kappas=0.51), thus indicating good generalizability of the model. Number of visits, year of index visit, depression, and anxiety conveyed the most importance in the propensity model (Figure 2).

Table 1.

Baseline Demographics and Clinical Features of the Prematch and Matched Samples of Women Veterans With and Without PTSD

	Prematch sample			Matched sample
	No PTSD	PTSD		No PTSD	PTSD
	(n=486 851)	(n=143 859)	SMD	(n=104 046)	(n=104 046)	SMD
Age, y	43.24±15.76	40.29±11.61	−0.213	39.13±11.67	39.16±11.66	0.003
Age group, y
18–29	123 376 (25.3%)	34 287 (23.8%)	−0.035	29 806 (28.7%)	29 623 (28.5%)	−0.004
30–39	100 803 (20.7%)	40 939 (28.5%)	0.181	28 239 (27.1%)	28 355 (27.3%)	0.003
40–49	113 560 (23.3%)	35 664 (24.8%)	0.034	24 880 (23.9%)	25 009 (24.0%)	0.003
50–59	82 261 (16.9%)	25 517 (17.7%)	0.022	16 410 (15.8%)	16 298 (15.7%)	−0.003
60+	66 851 (13.7%)	7452 (5.2%)	−0.295	4711 (4.5%)	4761 (4.6%)	0.002
Race or ethnicity
Asian or Pacific Islander	11 310 (2.3%)	3274 (2.3%)	−0.003	2691 (2.59%)	2542 (2.4%)	−0.009
Non‐Hispanic Black	115 746 (23.8%)	45 981 (32.0%)	0.183	31 143 (29.9%)	31 458 (30.2%)	0.007
Hispanic/Latina	31 648 (6.5%)	13 041 (9.1%)	0.096	9692 (9.3%)	9772 (9.4%)	0.003
Unknown	77 394 (15.9%)	7802 (5.4%)	−0.344	5601 (5.4%)	6348 (6.1%)	0.031
American Indian or Alaska Native	3612 (0.7%)	1793 (1.3%)	0.051	1203 (1.2%)	1273 (1.2%)	0.006
Non‐Hispanic White	247 141 (50.8%)	71 968 (50.0%)	−0.015	53 716 (51.6%)	52 653 (50.6%)	−0.020
Baseline conditions
Hyperlipidemia	24 117 (5.0%)	24 234 (16.9%)	0.389	10 023 (9.6%)	11 496 (11.1%)	0.047
Hypertension	40 683 (8.4%)	26 296 (18.3%)	0.295	12 606 (12.1%)	13 780 (13.2%)	0.034
Diabetes	10 049 (2.1%)	8484 (5.9%)	0.197	3199 (3.1%)	3809 (3.7%)	0.033
Smoking	33 221 (6.8%)	42 208 (29.3%)	0.612	20 753 (20.0%)	23 067 (22.2%)	0.055
Obesity	63 803 (13.1%)	56 119 (39.0%)	0.618	29 999 (28.8%)	31 962 (30.7%)	0.041
Chronic kidney disease	801 (0.2%)	938 (0.7%)	0.077	305 (0.3%)	372 (0.4%)	0.011
Neuroendocrine disorders	13 475 (2.8%)	10 318 (7.2%)	0.204	4618 (4.4%)	5240 (5.0%)	0.028
Female‐specific risk factors	473 (0.1%)	1434 (1.0%)	0.122	327 (0.3%)	570 (0.6%)	0.036
Depression	8382 (1.7%)	34 980 (24.3%)	0.713	7457 (7.2%)	10 278 (9.9%)	0.097
Bipolar disorder	3271 (0.7%)	8958 (6.2%)	0.308	2239 (2.2%)	3309 (3.2%)	0.064
Anxiety	7434 (1.5%)	27 068 (18.8%)	0.597	6413 (6.2%)	8680 (8.3%)	0.084
Schizophrenia	1266 (0.3%)	1271 (0.9%)	0.083	434 (0.4%)	563 (0.5%)	0.018
Alcohol use disorder	1348 (0.3%)	5591 (3.9%)	0.255	1139 (1.1%)	1706 (1.6%)	0.047
Drug use disorder	900 (0.2%)	4726 (3.3%)	0.239	796 (0.8%)	1325 (1.3%)	0.051
Number of baseline health care visits	4.98 (9.64)	17.78 (22.21)	0.747	11.33 (14.43)	12.46 (16.45)	0.073
Year of index visit date
2000	8 (0.0%)	0 (0.0%)	−0.006	1 (0.0%)	0 (0.0%)	−0.004
2001	48 441 (10.0%)	988 (0.7%)	−0.422	848 (0.8%)	986 (1.0%)	0.014
2002	29 389 (6.0%)	1532 (1.1%)	−0.271	1390 (1.3%)	1503 (1.4%)	0.009
2003	25 634 (5.3%)	1947 (1.4%)	−0.220	1751 (1.7%)	1838 (1.8%)	0.006
2004	23 720 (4.9%)	2803 (2.0%)	−0.162	2662 (2.6%)	2612 (2.5%)	−0.003
2005	21 687 (4.5%)	3469 (2.4%)	−0.112	3060 (2.9%)	3060 (2.9%)	0.000
2006	20 678 (4.3%)	3716 (2.6%)	−0.092	3194 (3.1%)	3224 (3.1%)	0.002
2007	20 838 (4.3%)	4823 (3.4%)	−0.048	4338 (4.2%)	4073 (3.9%)	−0.013
2008	21 425 (4.4%)	5621 (3.9%)	−0.025	4891 (4.7%)	4689 (4.5%)	−0.009
2009	22 905 (4.7%)	6450 (4.5%)	−0.011	5231 (5.0%)	5206 (5.0%)	−0.001
2010	24 030 (4.9%)	6964 (4.8%)	−0.004	5657 (5.4%)	5606 (5.4%)	−0.002
2011	24 132 (5.0%)	7708 (5.4%)	0.018	6410 (6.2%)	6150 (5.9%)	−0.010
2012	25 415 (5.2%)	8704 (6.1%)	0.036	6797 (6.5%)	6830 (6.6%)	0.001
2013	25 177 (5.2%)	10 214 (7.1%)	0.080	7970 (7.7%)	7820 (7.5%)	−0.005
2014	25 764 (5.3%)	11 469 (8.0%)	0.108	8513 (8.2%)	8514 (8.2%)	0.000
2015	25 821 (5.3%)	13 100 (9.1%)	0.147	8950 (8.6%)	9108 (8.8%)	0.005
2016	28 631 (5.9%)	13 479 (9.4%)	0.132	8417 (8.1%)	8336 (8.0%)	−0.003
2017	27 639 (5.7%)	13 412 (9.3%)	0.139	7746 (7.4%)	8113 (7.8%)	0.013
2018	25 288 (5.2%)	14 403 (10.0%)	0.183	8326 (8.0%)	8568 (8.2%)	0.009
2019	20 229 (4.2%)	13 057 (9.1%)	0.199	7894 (7.6%)	7810 (7.5%)	−0.003

Data are reported as mean±SD or /requency (%). PTSD indicates posttraumatic stress disorder; and SMD, standardized mean difference.

Figure 2. Variable importance from extreme gradient boosting propensity model.

 

Women veterans in the analytic sample (N=208 092) had a mean age of 39.15 years, with <5% of the study population aged 60 years or older. The population was diverse with respect to race or ethnicity as well (51% non‐Hispanic White, 30% non‐Hispanic Black, 9.3% Hispanic/Latina, 2.5% Asian or Pacific Islander, 1.2% American Indian or Alaskan Native).

PTSD and Incident Stroke/TIA

The Cox models all demonstrated appropriate assumptions of proportional hazards according to Schoenfeld residual plots and tests of statistical significance (P≥0.16). The mean length of follow‐up for women veterans with PTSD was 6.42 years (SD=4.49) and 5.82 years (SD=4.58) for women veterans without PTSD. During that time, 1544 (1.5%) matched controls died before end of follow‐up, as did 1555 (1.5%) women with PTSD. Over follow‐up, there were 3883 stroke or TIA events; 1573 (1.5%) were among women veterans without PTSD, and 2310 (2.2%) were among women veterans with PTSD (Table 2). For both women veterans with and without PTSD, the majority of these events were strokes (64.6% and 71.9%, respectively). Illustration of the survival analysis for patients with PTSD versus those without PTSD (Figure 3) presents Kaplan–Meier survival curves for the stroke/TIA composite (A), stroke (B), and TIA (C) for women veterans with and without PTSD. PTSD was significantly associated with an elevated rate of developing stroke or TIA (hazard ratio [HR], 1.33 [95% CI, 1.25–1.42]; Table 2). In separate analyses, PTSD was also significantly related to incident stroke (HR, 1.19 [95% CI, 1.10–1.29]) and incident TIA (HR, 1.66 [95% CI, 1.48–1.87]). Analyses of stroke subtypes revealed that PTSD was significantly associated with a heightened rate of developing ischemic and undefined stroke but not hemorrhagic, stroke (Table 2).

Table 2.

Associations of PTSD With Incident Stroke and TIA in Women Veterans

	No PTSD	PTSD	Cox regression results
Outcome	No. (%) events	No. (%) events	HR (95% CI)	P value
Stroke/TIA composite	1573 (1.5%)	2310 (2.2%)	1.33 (1.25–1.42)	<0.001
Stroke	1131 (1.1%)	1493 (1.4%)	1.19 (1.10–1.29)	<0.001
Hemorrhagic	117 (0.1%)	125 (0.1%)	0.96 (0.75–1.24)	0.75
Ischemic	806 (0.8%)	1068 (1.0%)	1.19 (1.09–1.30)	<0.001
Undefined	208 (0.2%)	300 (0.3%)	1.31 (1.10–1.57)	0.003
TIA	442 (0.4%)	817 (0.8%)	1.66 (1.48–1.87)	<0.001

HR indicates hazard ratio; PTSD, posttraumatic stress disorder; and TIA, transient ischemic attack.

Figure 3. Survival analysis for patients with PTSD versus those without PTSD.

A, Comparison of stroke/TIA between the PTSD and no‐PTSD cohorts. B, Comparison of stroke between the PTSD and no‐PTSD cohorts. C, Comparison of TIA between the PTSD and no‐PTSD cohorts. HR indicates hazard ratio; PTSD, posttraumatic stress disorder; and TIA, transient ischemic attack.

Stratified Analyses of PTSD and Incident Stroke/TIA: Age and Race or Ethnicity

The results of analyses stratified by age and race or ethnicity are reported in Table 3. PTSD was associated with elevated rates of the incident stroke/TIA composite for all age groups except women veterans aged 60 years and older. However, the latter population comprised <5% of the total study population, and the PTSD effect estimate still conveyed elevated risk. In analyses stratified by race or ethnicity, PTSD was associated with significantly elevated rates of developing stroke/TIA for Hispanic/Latina (HR, 2.27 [95% CI, 1.66–3.12]), non‐Hispanic White (HR,1.40 [95% CI, 1.28–1.52]), non‐Hispanic Black women (HR, 1.12 [95% CI, 1.00–1.25]), and women of other/unknown race or ethnicity (HR, 1.40 [95% CI: 1.04–1.89]). In contrast, no significant associations were observed for American Indian or Alaskan Native, or Asian or Pacific Islander women.

Table 3.

Age‐ and Race‐ or Ethnicity‐Stratified Analyses of Association of PTSD With Incident Stroke/TIA in Women Veterans

		No PTSD	PTSD	Cox regression results
	No.	No. (%) events	No. (%) events	HR (95% CI)	P value
Age, y
18–29	59 429	76 (0.3%)	133 (0.4%)	1.54 (1.16–2.04)	0.003
30–39	56 594	190 (0.7%)	333 (1.2%)	1.56 (1.31–1.87)	<0.001
40–49	49 889	532 (2.1%)	812 (3.2%)	1.41 (1.27–1.58)	<0.001
50–59	32 708	556 (3.4%)	761 (4.7%)	1.29 (1.15–1.43)	<0.001
60+	9472	219 (4.6%)	271 (5.7%)	1.10 (0.92–1.32)	0.28
Race or ethnicity
Asian or Pacific Islander	5233	21 (0.8%)	23 (0.9%)	0.99 (0.55–1.79)	0.98
Non‐Hispanic Black	62 601	547 (1.8%)	664 (2.1%)	1.12 (1.00–1.25)	0.049
Hispanic/Latina	19 464	53 (0.5%)	135 (1.4%)	2.27 (1.66–3.12)	<0.001
Other/Unknown	11 949	65 (1.2%)	129 (2.0%)	1.40 (1.04–1.89)	0.026
American Indian or Alaskan Native	2476	13 (1.1%)	30 (2.4%)	1.89 (0.99–3.60)	0.054
Non‐Hispanic White	106 369	874 (1.6%)	1329 (2.5%)	1.40 (1.28–1.52)	<0.001

HR indicates hazard ratio; PTSD, posttraumatic stress disorder; and TIA, transient ischemic attack.

Discussion

This study represents the largest and most comprehensive investigation of the association of PTSD with incident stroke/TIA in women veterans, a growing population with high level of trauma exposure and cardiovascular risk that nevertheless has been underrepresented in research on the link between PTSD and CVD. ²⁸ Our investigation has several important findings. First, PTSD was a risk factor for incident stroke/TIA based on a composite outcome after matching for a comprehensive set of cardiovascular risk factors. Second, in separate subanalyses, PTSD was also significantly associated with heightened rates of developing stroke, TIA, ischemic stroke, and undefined stroke. Third, PTSD was a significant risk factor for developing stroke/TIA across all age groups, except for women veterans aged 60 and older. Fourth, differences in the PTSD‐stroke/TIA relation emerged for different racial and ethnic groups, with significantly elevated HRs observed for Hispanic/Latina, non‐Hispanic Black, and non‐Hispanic White women and women of other race or ethnicity. These findings are consistent with our previous study demonstrating a significant association between PTSD and incident ischemic heart disease in women veterans. Together, these data suggest that women veterans with PTSD may be at elevated risk of developing various cardiovascular disorders and that this risk may be pronounced for certain subgroups.

Our results extend growing work demonstrating a link between PTSD and stroke/TIA in various trauma‐exposed samples, including population‐based registry members, community‐based samples, and veterans. ²² , ²³ , ²⁵ , ²⁶ , ²⁷ Despite differences in covariates and analytic approaches across investigations, the effect sizes in our sample of women veterans were generally similar to those detected in women in these other samples. ²² , ²⁵ Further, we identified links between PTSD and incident ischemic and undefined (but not hemorrhagic) stroke, as in some, ²³ , ²⁴ but not all, ²² , ²⁷ prior work. In general, hemorrhagic stroke is less prevalent than ischemic stroke, ⁴⁷ and hemorrhagic stroke comprised the smallest proportion of stroke events in the current study. Thus, we may have been underpowered for this subtype analysis in particular.

Understanding the mechanisms linking PTSD with incident stroke/TIA can inform potential risk screening and intervention efforts. PTSD has been associated with a number of behavioral risk factors for stroke/TIA, including poor diet, physical inactivity, smoking, insomnia, and medication nonadherence. ³⁴ , ⁴⁸ , ⁴⁹ , ⁵⁰ PTSD is also highly comorbid with psychiatric disorders that have also been linked to stroke (eg, depression), ⁵¹ , ⁵² and pharmacotherapy commonly prescribed for individuals with PTSD or depression has been associated with elevated stroke risk. ²⁷ PTSD is also characterized by numerous physiological processes that can contribute to stroke/TIA, including dysregulation of the hypothalamic–pituitary–adrenal axis, sympathetic‐adrenal‐medullary system, and immune system. ⁵³ , ⁵⁴ To date, research in community‐dwelling women found that a variety of behavioral (eg, physical activity, diet quality, smoking) and medical (eg, obesity, hypertension, type 2 diabetes) risk factors may explain only ≈50% of the PTSD–CVD relationship (based on a composite of myocardial infarction and stroke), ²⁵ suggesting that comprehensive approaches are needed to fully delineate relevant pathway variables. Furthermore, although our groups of women veterans with and without PTSD were balanced with respect to a variety of psychiatric disorders (anxiety, depression, schizophrenia, alcohol dependence, and nonalcohol drug dependence) through PSM, more research is needed to better understand the manifestations of posttraumatic psychopathology that may be most relevant to subsequent risk of CVD, including stroke/TIA. For example, PTSD and depression have a number of overlapping symptoms (eg, anhedonia, sleep and concentration difficulties), and research that considers cross‐cutting dimensions of psychopathology may help to identify particular symptoms to address in more targeted screening and intervention efforts.

In addition to demonstrating that PTSD was associated with incident stroke/TIA in the overall sample of women veterans, we identified age‐ and race‐ or ethnicity‐related subgroups for whom this risk was particularly prominent. Similar to other research on PTSD and CVD in women, ¹³ , ¹⁴ , ²⁵ , ³¹ younger women with PTSD had the greatest risks for incident stroke/TIA. Indeed, the PTSD‐stroke/TIA link was most pronounced for women <50 years old. These findings may have multiple explanations. For example, CVD risks accumulate with advancing age, and the relative impact of PTSD as a CVD risk factor may be attenuated at older ages when other risk factors are also prevalent. At younger ages, when there are fewer other CVD risk factors, the PTSD‐stroke/TIA relation may thus appear more pronounced. Furthermore, with advancing age, it is possible that PTSD may resolve naturally or due to treatment, thereby reducing its impact on stroke/TIA. There is also initial evidence that PTSD symptom improvement or treatment may be associated with reductions in a CVD risk factor (hypertension), ⁵⁵ , ⁵⁶ and PTSD remission (versus ongoing symptoms) has been associated with reduced risk of CVD onset. ⁵⁷ Future research is needed to directly test these hypotheses.

Differences in the PTSD‐stroke/TIA relation also emerged for women veterans of different race or ethnicity. PTSD was associated with significantly elevated rates of developing stroke/TIA for Hispanic/Latina, non‐Hispanic Black, and non‐Hispanic White women and women of other/unknown race or ethnicity, with Hispanic/Latina women with PTSD having >2‐fold greater risk of stroke/TIA. No significant associations were observed for American Indian or Alaskan Native or Asian or Pacific Islander women. However, caution should be used in interpreting the results for these latter populations due to the relatively small subgroup size and low absolute number of events. Nevertheless, our findings parallel some of the larger literature on racial and ethnic disparities in stroke/TIA, with particularly heightened risk observed in Hispanic/Latina women. ¹¹ Research on the factors underlying these differences is needed. However, current research points to differences in the prevalence of CVD risk factors, access to care, treatment adherence, and systemic or provider‐based bias as potential explanations. ²⁸ , ⁵⁸ , ⁵⁹

Limitations

The present results are subject to limitations characterizing retrospective analyses of administrative diagnostic data, including misclassification, misestimation of onset time, ascertainment bias, and issues with representativeness. Although PSM, particularly when implemented with machine learning, minimizes biases due to measured confounds, estimates are subject to bias from unmeasured confounders. For instance, due to limitations of the EHRs, not all potential risk factors, such as menopausal status, diet, or physical activity could be assessed. Furthermore, we did not account for pharmacotherapy as a factor relevant to stroke/TIA. Such analysis requires incorporating various information (eg, medication class, dosage, length of therapy, adherence) and was beyond the scope of this project. We also did not have information about trauma exposure, PTSD severity, or receipt of interventions. Moreover, women with PTSD who were not matched to controls, numbering 39 813 (27.7%), were not represented in our sample, thus limiting generalizability. Relatedly, our results are based on a large cohort of women veterans and may not be generalizable to other women and men. However, our study also had many strengths. This is the largest investigation of the association of PTSD with stroke/TIA in women veterans, analyzing over 1 million years of health data. In addition, we matched the cohorts with respect to a comprehensive list of potential risk factors in addition to traditional risk factors. Finally, because of the large sample size, we were able to perform stratified analysis based on age and race or ethnicity, thus uncovering subgroups with particularly elevated risk.

Conclusions

Our study suggests that PTSD is an important risk factor for stroke/TIA in women veterans, especially at younger ages and in the Hispanic/Latina population. These results reinforce our call to action to improve the cardiovascular care of women veterans. ²⁸ Furthermore, our findings highlight the potential benefit of screening for stroke/TIA risk in women veterans with PTSD and emphasize the importance of integrated mental and cardiovascular health care after trauma. Additional research is required to assess the extent to which PTSD predicts a broader range of cardiovascular disorders and whether interventions targeting PTSD can offset the elevated cardiovascular risk.

Sources of Funding

This research was funded through a grant from the Department of Defense Office of Congressionally Directed Medical Research Programs Proposal Number PR171210, Award Number W81XWH‐18‐1‐0725).

Disclosures

None.

Supporting information

Data S1.