The impact of sleep quality on the incidence of PTSD: Results from a 7-Year, Nationally Representative, Prospective Cohort of U.S. Military Veterans

Abstract

Sleep and posttraumatic stress disorder (PTSD) have a complex relationship, with some studies showing that disrupted sleep is associated with subsequent development of PTSD. The purpose of the current study was to examine the relationship between sleep quality and the development of probable PTSD in U.S. veterans surveyed as part of the National Health and Resilience in Veterans Study, a 7-year, nationally representative, prospective cohort study with four waves of data collection. Sociodemographic, military, trauma, and clinical variables were entered into a multivariate analysis to examine independent determinants of new-onset PTSD. A total of 142 (7.3%) veterans developed PTSD over the 7-year study period. Poor/fair sleep quality at Wave 1 was associated with 60% greater likelihood of developing PTSD, with more than twice as many veterans who developed PTSD reporting poor sleep quality at Wave 1 (47.8% vs. 20.7%). Younger age, using the VA as a primary source of healthcare, greater traumas since Wave, and lifetime depression were additionally associated with this outcome. Results of this study underscore the importance of self-reported sleep quality as a potential risk factor for the development of PTSD in the U.S. veteran population.

Disturbed sleep is the most common symptom reported by individuals diagnosed with posttraumatic stress disorder (PTSD; Ohayon & Shapiro, 2000), and problems related to sleep have been in every version of the diagnostic criteria for this disorder. Historically, sleep difficulties were viewed as symptoms of the broader syndrome of PTSD (e.g., Harvey, Jones, & Schmidt, 2003), and have even been described as the “hallmark of PTSD” (Ross, Ball, Sullivan, & Karoff, 1989; Germain, 2013). More recent studies have indicated that the relationship between PTSD and disturbed sleep is more complex. For example, ongoing disrupted sleep has been found to be a residual symptom after otherwise successful treatment of PTSD with evidence-based psychotherapy (e.g., Larsen, Fleming, & Resick, 2019; Zayfert & DeViva,2004). Treatments targeting PTSD symptoms are associated with significant improvements on measures of disrupted sleep, but mean scores on those measures typically remain above the cutoffs for clinically significant sleep problems (Colvonen, Straus, Drummond, Angkaw, & Norman, 2020; Galovski, Monson, Bruce, & Resick, 2009). Taken together, these studies suggest that sleep difficulties may be perpetuated by mechanisms that are independent of other PTSD symptoms (Koffel, Khawaja, & Germain 2016).

In addition, difficulties initiating and maintaining sleep appear to be related to the development of PTSD. Gehrman and colleagues (2013) assessed sleep in a cohort of United States (U.S.) servicemembers and found that greater levels of pre-deployment insomnia symptoms were associated with higher likelihood of developing PTSD after deployment to Iraq or Afghanistan. Thormar, Gersons, Juen, Djakababa, Karlsson, and Olff (2014) found that poorer subjective sleep quality six months after exposure to a natural disaster was associated with higher likelihood of PTSD symptoms 18 months after exposure in a sample of volunteer aid workers. Kobayashi and Mellman (2012) assessed sleep via overnight polysomnography in a sample of men and women admitted to Level 1 trauma centers after life-threatening incidents. Among women, less total sleep time during the first month post-incident was related to higher likelihood of being diagnosed with PTSD two months post-incident.

Studies using ecological momentary assessment have found similar relationships between sleep and subsequent PTSD symptoms at the event level. Short, Allan, and Schmidt (2017) examined sleep and daily PTSD symptoms in a primarily female sample of individuals diagnosed with PTSD and found that a given night’s sleep quality and sleep efficiency were associated with PTSD symptoms the next day. Results indicated that poorer sleep quality and lower sleep efficiency were associated with higher PTSD symptoms the next day. DeViva, Rosen, Cooney, and Black (2020) examined the relationship between hours of sleep (assessed each morning for the previous night) and peak level of PTSD symptoms from thrice-daily assessments using a modified PTSD Checklist for DSM-5 (PCL-5; Weathers, Litz, Keane, Palmieri, Marx, & Schnurr, 2013) that measured symptoms over the past two hours. Results indicated that in a sample of male U.S. veterans, a given night’s sleep was related to the next day’s peak PTSD symptoms, such that less sleep than the individual’s mean amount of nightly sleep was associated with a greater spike in PTSD symptoms the next day. There was no relationship between peak PTSD symptoms on a given day and sleep that night. Similarly, Biggs, Ursano, Wang, Wynn, Carr, & Fullerton, (2020) found that among individuals with probable PTSD, sleep the prior night predicted next-day PTSD symptoms such that a poor night of sleep was associated with worse PTSD symptoms the following day.

Taken together, these studies indicate that sleep disturbance may be more appropriately conceptualized as a factor that affects the development of and dynamic changes in PTSD than simply as a symptom of the disorder. The studies examining the relationship between sleep and subsequent development of PTSD cited above were conducted on samples that were about to be exposed to trauma or that were recently exposed to trauma. Thus far, there has not been a study of the relationship between sleep and development of PTSD at the general population level. To address this gap, the current study evaluated the relationship between sleep quality and later development of PTSD in a sample of U.S. veterans surveyed as part of a 7-year prospective national cohort study. The main study hypothesis was that poor sleep quality at baseline would be associated with higher likelihood of new onset of PTSD over the 7-year study period.

Method

Participants

Data were drawn from the National Health and Resilience in Veterans Study (NHRVS; Mota et al., 2019), a nationally representative study of U.S. veterans with four waves of data collection (2011, 2013, 2015, and 2018). The NHRVS sample was drawn from KnowledgePanel®, a probability-based, online non-volunteer access survey panel of a nationally representative sample of U.S. adults that covers approximately 98% of U.S. households. This panel of more than 50,000 households was developed and is maintained by GfK Knowledge Networks, Inc. (now Ipsos). Panel members were recruited through national random samples, originally by telephone and then almost entirely by postal mail. KnowledgePanel® recruitment uses dual sampling frames including listed and unlisted telephone numbers, telephone and nontelephone households, and cell-phone-only households, as well as households with and without Internet access (households are provided with access to the Internet and computer hardware if needed). Only persons sampled through these probability-based techniques are eligible to participate on KnowledgePanel®. This procedure results in a more generalizable sample than those recruited through Internet convenience panels, also known as “opt-in” panels, that include only individuals with Internet access who volunteer themselves for research.

Veteran status for the individuals in KnowledgePanel® was assessed using a general demographic questionnaire. Of the 4,750 veterans who were in the GfK Knowledge Networks survey panel at the time the NHRVS was fielded, 3,408 (71.7%) responded to an invitation to participate and completed a screening question to confirm study eligibility. Of these respondents, 3,188 (93.5%) confirmed current or past active military status, and 3,157 (92.6%) completed a confidential 60-minute online survey. A total of 2,157 (68.3% of the Wave 1 cohort) completed Wave 2 in 2013, 1,538 (48.7% of the Wave 1 cohort) completed Wave 3 in 2015, and 1,310 (41.5% of the Wave 1 cohort) completed Wave 4 in 2018. The effective sample for the current study included 2,095 veterans without a positive screen for PTSD at Wave 1 (i.e., Wave 1 PCL-S score<50) and provided data on PTSD symptoms at baseline and at least one of the follow-up assessments (mean=2.2, SD=0.8, range=1-3) over the 7-year study period.

All participants provided informed consent; the Human Subjects Subcommittee of the Veterans Affairs (VA) Connecticut Healthcare System and VA Office of Research and Development approved the study procedures. To permit generalizability of study results to the entire population of U.S. veterans, post-stratification weights were computed by GfK statisticians and applied in inferential analyses. These weights were based on demographic distributions (i.e., age, gender, race/ethnicity, education, Census region, and metropolitan area) of veterans in the GfK Knowledge Networks survey panel and calibrated against distributions from the most contemporaneous (October 2010) Current Population Survey (US Census Bureau, 2010).

Assessments

Demographic information.

Demographic information, including age, gender, race, education level, marital status, household income, employment status, and use of Veterans Affairs (VA) services for health care, was gathered as part of the NHRVS survey. A general demographic questionnaire was used for all members of the survey panel with items added to assess military variables.

Lifetime trauma history.

The Trauma History Screen (THS; Carlson et al., 2011) is a self-report measure that assesses the occurrence of 13 potentially traumatic life events across the lifespan (e.g., physical or sexual assault during childhood or adulthood, traumatic events during military service, accidents, unexpected loss of a loved one). “Life-threatening illness and injury,” as a potential traumatic event, was added to the THS used in the NHRVS. The THS has strong convergent validity and reliability (Carlson et al., 2011).

PTSD symptoms.

PTSD symptoms were assessed in the first wave of data collection in 2011 using the PTSD Checklist-Specific Stressor version (PCL-S; Weathers, Litz, Herman, Huska, & Keane, 1993). The PCL-S is a 17-item self-report measure that assesses Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV-TR; American Psychiatric Association, 2000) PTSD diagnostic criteria. Scores on the PCL-S range from 17 to 85. A total score of 50 or higher is indicative of probable PTSD (Weathers et al., 1993). The PCL-S has been found to have good test-retest reliability, internal consistency, and validity (Wilkins, Lang, & Norman, 2011). In the current sample, α=0.95. For the subsequent three waves of data collection in 2013, 2015, and 2018, the PCL for DSM-5 (PCL-5) was used to assess PTSD symptoms. The PCL-5 is a 20-item self-report instrument used to assess PTSD symptoms, with scores ranging from 0 to 80. The 20 items correspond to the DSM-5 diagnostic criteria for PTSD (American Psychiatric Association, 2013). A score of 31 or higher has been suggested as being indicative of probable PTSD (Bovin et al., 2015). The PCL-5 has shown good reliability as well as convergent and discriminant validity (Bovin et al., 2015). In the current sample, αs=0.96, 0.95, and 0.95 for PCL-5 items assessed at Waves 2 (2013), 3 (2015), and 4 (2018), respectively. Because the main goal of this study was to examine the relationship between sleep quality at baseline and incident PTSD, the PCL item assessing “trouble falling or staying asleep” (item 13 on the PCL-S and item 20 on the PCL-5) was omitted when calculating PCL scores at each assessment.

Sleep quality.

Self-reported sleep quality was assessed using item 6 of the Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989). This item asks, “During the past month, how would you rate your sleep quality overall?” Response options were 0 “Very good,” 1 “Fairly good,” 2 “Fairly bad,” or 3 “Very bad.” The PSQI has good reliability and validity (Backhaus, Junghanna, Broocks, Riemann, & Hohagen, 2002).

Comorbid diagnoses.

The presence of lifetime history of major depressive disorder, alcohol use disorder, and drug use disorder was assessed using the Mini International Neuropsychiatric Interview (MINI; Sheehan et al., 1998), a widely used structured diagnostic interview instrument. The MINI comprises modules for different diagnostic categories and has good reliability and validity (Sheehan et al., 1998). In the current study, the depression and alcohol and drug use disorder modules were administered.

Data Analyses.

Data analyses proceeded in three steps. First, we computed descriptive statistics to summarize characteristics of the study sample. Second, we conducted independent-samples t-tests and χ² tests to compare sociodemographic, military, and clinical characteristics of veterans who did and did not develop PTSD over the 7-year follow-up period. Third, we conducted a multivariate binary logistic regression with incident PTSD (i.e., PCL-5 score of 31 or higher at the second, third, and/or fourth waves of the NHRVS) as the outcome variable. Variables that were associated with incident PTSD at the p < 0.05 level in bivariate analyses were entered as independent variables in this analysis.

Results

Sample characteristics.

Table 1 shows sociodemographic, military, trauma history, and clinical characteristics of the full sample, and stratified by incident PTSD status. Relative to veterans who did not develop PTSD over the 7-year study period, those who did develop PTSD were younger, and more likely to be female, non-Caucasian, have completed some college or higher education, unmarried/partnered, and not retired. They were also more likely to have served in the Marine Corps, to have been exposed to combat, and to use the VA are their primary source of health care. With regard to trauma history and clinical characteristics, they reported enduring more traumas at and since Wave 1, and were more likely to screen positive for MDD and AUD and/or DUD and to report poor/fair sleep quality at Wave 1.

Table 1.

Sociodemographic, military, and clinical characteristics of full sample, and stratified by incident PTSD status over the 7-year study period

	Full sample	No Incident PTSD	Incident PTSD	Test of difference	p
	n=2,095	n=1,953 weighted 92.7%	n=142 weighted 7.3%
	Weighted mean (SD) or n (weighted %)	Weighted mean (SD) or n (weighted %)	Weighted mean (SD) or n (weighted %)
Age	62.8 (12.2)	62.3 (13.7)	51.1 (14.6)	9.13	<0.001
Male gender	1,894 (91.2%)	1,784 (91.9%)	110 (81.8%)	16.13	<0.001
Caucasian race/ethnicity	1,782 (78.5%)	1,674 (79.7%)	108 (61.6%)	24.91	<0.001
Some college of higher education	1,804 (69.6%)	1,677 (69.0%)	127 (77.4%)	4.22	0.040
Married/partnered	1,667 (75.7%)	1,569 (76.4%)	98 (66.7%)	6.63	0.010
Retired	1,003 (46.3%)	1,000 (48.6%)	142 (16.1%)	54.21	<0.001
Household income $60K+	1,147 (47.5%)	1,081 (47.9%)	66 (42.3%)	1.57	0.21
Branch of service				16.40	0.003
Army	844 (38.0%)	789 (38.9%)	55 (27.0%)
Air Force	548 (24.7%)	515 (24.9%)	33 (21.9%)
Navy	482 (23.4%)	449 (21.1%)	33 (27.7%)
Marines	149 (9.8%)	132 (9.2%)	17 (17.5%)*
Other	68 (4.05)	64 (3.9%)	4 (5.8%)
Combat veteran	707 (31.3%)	651 (30.7%)	56 (39.4%)	4.45	0.035
VA is primary source of health care	327 (17.2%)	285 (15.8%)	42 (35.0%)	32.97	<0.001
Number of traumas at Wave 1	3.1 (2.5)	2.9 (2.4)	4.7 (3.8)	7.82	<0.001
Traumas since Wave 1	2.0 (1.4)	1.9 (1.1)	3.6 (2.1)	15.73	<0.001
Lifetime MDD	289 (13.4%)	226 (11.5%)	63 (37.2%)	72.61	<0.001
Lifetime AUD or DUD	406 (19.9%)	356 (19.1%)	50 (30.7%)	10.61	0.001
Poor/fair sleep quality	451 (22.6%)	383 (20.7%)	68 (47.8%)	52.98	<0.001

^*Statistically significant difference after Bonferroni correction.

Determinants of Incident PTSD.

A total of 142 (weighted 7.3%) veterans screened positive for incident PTSD over the 7-year study period. Table 2 shows results of a binary logistic regression analysis of determinants of incident PTSD. Results of this analysis revealed that younger age, VA as a primary source of health care, greater number of traumas since Wave 1, lifetime MDD, and poor/fair sleep quality were associated with increased likelihood of developing PTSD. There was also a significant interaction between MDD and poor/fair sleep quality; among veterans who reported poor/fair sleep quality, those who screened positive for MDD were more likely than those who did not to develop PTSD (Wald X²=4.05, p=0.044; Relative Risk Ratio [RRR]=2.72, 1.03-7.22).

Table 2.

Results of multivariable binary logistic regression model predicting incident PTSD over the 7-year study period (Nagelkerke R²=0.32)

	Wald χ2	p	RRR (95%CI)
Age	6.44	0.11	0.98 (0.96-0.99)*
Male gender	1.50	0.22	1.48 (0.79-2.76)
Caucasian race/ethnicity	3.50	0.061	0.64 (0.41-1.02)
Some college or higher education	1.13	0.29	1.31 (0.80-2.14)
Married/partnered	0.01	0.97	0.99 (0.62-1.58)
Retired	2.96	0.085	0.58 (0.32-1.08)
Marine Corps vs. Other Branch	0.33	0.56	1.19 (0.66-2.16)
Combat veteran	0.23	0.63	0.89 (0.56-1.42)
VA is primary source of health care	9.93	0.002	2.11 (1.33-3.36)**
Number of traumas at Wave 1	0.32	0.57	0.97 (0.89-1.06)
Traumas since Wave 1	77.19	<0.001	1.82 (1.59-2.08)***
Lifetime MDD	5.44	0.020	1.83 (1.10-3.05)*
Lifetime AUD or DUD	1.34	0.25	1.32 (0.82-2.13)
Poor/fair sleep quality	4.52	0.034	1.60 (1.04-2.47)*

Note. RRR=relative risk ratio; 95%CI=95% confidence interval; MDD=major depressive disorder; AUD=alcohol use disorder; DUD=drug use disorder.

Significant association:

^*p<0.05;

^**p<0.01;

^***p<0.001

Discussion

This study adds to the growing literature indicating that sleep quality is related to increased risk for the development of PTSD. Results indicated that in a nationally representative sample of United States military veterans, self-rated fairly bad or very bad sleep quality at baseline was associated with 60% increased risk for developing PTSD over the next seven years, compared to veterans who self-rated their sleep quality as fairly or very good. Of note, this association was independent of a broad range of sociodemographic, military, and clinical determinants of PTSD risk, including trauma burden and depression.

There are several possible explanations for the finding that fairly bad/very bad sleep quality may increase risk for the development of PTSD. Ben Simon, Rossi, Harvey, and Walker (2020) reported that anxiety reduction occurs during sleep and noted that sleep-deprived individuals can report anxiety at clinically significant levels. It is possible that after trauma exposure, individuals with poor sleep have difficulty reducing anxiety back to baseline, maintaining posttraumatic responses above pathological levels. This may be related to disruption of inhibitory learning; research has shown that sleep disturbance either before or after fear extinction learning interferes with extinction recall (Seo et al,. 2018; Straus, Acheson, Risbrough, & Drummond, 2017). It may also be that lower levels of sleep or lower quality of sleep leaves the trauma-exposed individual with fewer resources to manage their response to stress and trauma (e.g., Matteson-Rusby, Pigeon, Gehrman, & Perlis, 2010). For example, lack of sleep compromises immune system functioning, and may affect overall pathophysiological pathways that facilitate stress tolerance (Pederson, Troxel, Shih, Pinder, Lee, & Guyer, 2015). It is also possible that some unmeasured third variable accounts for the relationship between sleep quality and later development of PTSD. For example, sleep apnea (Zhang, Weed, Ren, Tang, & Zhang, 2017) and anxiety sensitivity (Babson, Boden, Woodward, Alvarez, & Bonn-Miller, 2013) may, at least in part, account for this relationship, as both have been linked to poor sleep quality as well as PTSD.

Among other variables assessed, exposure to number of traumas since Wave 1 of the study was most strongly associated with new onset PTSD, which suggests that greater cumulative trauma burden may increase the likelihood of developing PTSD. Participants using VA as their primary health care source were more likely to develop PTSD. Enrollment in VA care may serve as a proxy variable for other factors that may increase the likelihood for PTSD; for example, in the current sample new cases of PTSD were less likely to be retired, and 47.5% of the current sample reported an annual household income of over $60,000. According to data from 2017, Veterans enrolled in VA care were over twice as likely to be unemployed as veterans in general (9.2% versus 3.7%), and 46.8% reported an annual household income below $35,000 (Huang, Kim, Muz, & Gasper, 2018). It is also possible that use of VA care is driven by poor access to other care, as 20.0% of Veterans enrolled in VA care in 2017 reported having private insurance other than Medicare, Medicaid, or Tricare, with 20.2% reporting no coverage (Huang et al., 2018). Alternatively, it is possible that when considering options for care for PTSD, veterans chose to receive care in a system more familiar with military culture. The relationship with lifetime MDD, as well as the interaction between sleep quality and lifetime MDD, are consistent with prior work demonstrating high comorbidity of MDD with PTSD (Armenta et al., 2019), and suggest that fair/poor sleep quality in combination with MDD may especially increase risk for the development of PTSD. Research has also shown that problems with sleep may increase the likelihood of MDD (Becker, Jesus, João, Viseu, & Martins, 2017; Pemberton & Tyszkjewicz, 2016), and poor sleep is a diagnostic criterion for both PTSD and MDD. The relationship between younger age and greater likelihood of new-onset PTSD is consistent with other evidence that the rate of PTSD is lower among older adults (e.g., Pietrzak, Goldstein, Southwick, & Grant, 2012). However, this may be due to avoidance or deficits in recall in older populations (McCarthy, Cook, & Thorp, 2020).

The current study adds to a growing body of evidence suggesting that improving sleep quality may help mitigate risk for the development of PTSD. These results have notable clinical implications. Given that sleep quality appears to be related to increased likelihood of later development of PTSD, it would be beneficial at minimum for health care providers routinely to assess sleep quality in trauma-exposed individuals and in individuals at high risk for exposure to trauma, such as service members and police officers. Regardless of other symptoms, insomnia is associated with significant impairment in functioning as well as cost to society (Matteson-Rusby et al., 2011; Zee et al., 2014), so it should be independently assessed, diagnosed, and treated when present. Increasing screening for sleep disorders is also congruent with Veteran Affairs and Department of Defense clinical practice guideline recommendations for the management of chronic insomnia disorder and obstructive sleep apnea (Department of Veterans Affairs and Department of Defense, 2019). Assessment of sleep before and during PTSD treatment is also important. The association between poor sleep and extinction learning suggests that poor sleep has the potential to interfere with PTSD treatments that rely on extinction, and there is some evidence that this is the case (López, Lancaster, Gros, & Acierno, 2017). The gold-standard treatment for insomnia is cognitive-behavioral therapy for insomnia (Department of Veterans Affairs and Department of Defense, 2019), which has show to be effective with veterans (Karlin, Trockel, Taylor, Gimeno, & Manber, 2013) and which is available in a variety of modalities (e.g., Hermes, Tsai, & Rosenheck, 2015).

Over 20% of this sample rated their sleep as “Fairly bad” or “Very bad.” Colvonen, Almklov, Tripp, Ulmer, Pittman and Afari (2020) found that 57.7% of a sample of post-9/11 veterans enrolling in care at VA likely had insomnia disorder, and of those, 93.3% also likely had PTSD. Mustafa, Erokwu, Ebose, and Strohl (2005) reported that a sample of veterans being seen in a VA primary care clinic were judged to be at high risk for insomnia at about twice the rate of the current study (41%). The sample in the present study was recruited from the general population, not a healthcare setting, and only about a sixth of the current sample used VA as their primary health care provider, which is less than the approximately 30% of all who veterans who used VA health care services in 2017 (National Center for Veterans Analysis and Statistics, 2020). Thus, results of the current study may be best generalized to the broader population of veterans than those who use VA care.

Sleep quality was assessed with a single item taken from an established measure of insomnia (“During the past month, how would you rate your sleep quality overall?”). Given that there are a number of different factors that may affect the subjective rating of sleep quality (e.g., insomnia or other sleep disorders, life circumstances, medical conditions), use of a single-item measure is a notable limitation of this study in that specific conclusions cannot be made regarding aspects of sleep that might be most strongly related to the development of PTSD. Due to changes in the DSM criteria for PTSD during the study period, PTSD screening status at baseline was assessed using the DSM-IV version of the PCL, while the DSM-5 version was administered at all follow-up assessments. Replication of the results using structured diagnosed interviews of sleep disorder and PTSD are needed.

Notwithstanding these limitations, results of this study suggest that U.S. veterans who describe their sleep as fairly bad or very bad at baseline were 60% more likely to develop new-onset PTSD over the subsequent seven years relative to veterans who described their sleep as fairly good or very good. This association was significant even after adjustment for a broad range of sociodemographic, military, and trauma-related risk factors for PTSD, thus underscoring the importance of self-rated sleep quality as a potential risk factor for the development of PTSD in veterans. Accordingly, these results underscore the importance of regularly assessing sleep quality in veterans at increased risk for PTSD. Though the study used only a single self-reported item to assess sleep quality, the association of ratings on this single item with later development of PTSD suggests that it can potentially be useful for everyday practice. These results also suggest that assessment, monitoring, and early interventions to address sleep difficulties may help mitigate the risk of subsequent development of PTSD in at-risk veterans. Further research is needed to replicate these findings using multidimensional and more nuanced measures of sleep quality; characterize biopsychosocial mechanisms linking poor sleep quality to increased risk for PTSD; and evaluate the efficacy of sleep interventions in mitigating risk for PTSD in veterans and other traumatized populations.

Highlights.

• Poor sleep is associated with a 60% greater likelihood of subsequent PTSD development

• Over 22% of a nationally representative sample of veterans report poor sleep quality

• Lifetime depression is associated with higher likelihood of PTSD development