Posttraumatic Stress Disorder and Alcohol Dependence: Epidemiology and Order of Onset

Abstract

Objective

Posttraumatic stress disorder (PTSD) and alcohol dependence (AD) frequently co-occur; however, epidemiologic studies of temporal associations between PTSD and AD are limited. The aims of this study were: (1) to investigate the bi-directional associations between PTSD and AD, and (2) to examine demographic and clinical correlates of order-of-onset among individuals with PTSD and AD.

Method

Participants were 11,103 adults (60.6% women; M_age=48.7 years, SD=15.9) from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) who endorsed lifetime alcohol consumption and DSM-IV PTSD Criterion A trauma exposure. Cox proportional hazards models with time-dependent covariates were used to evaluate bidirectional associations between PTSD and AD. Sex differences were assessed using stratified analyses.

Results

After adjusting for demographic, trauma, and alcohol characteristics, PTSD was associated with greater likelihood of subsequent AD (HR=1.359, 95%CI=1.357–1.362), and AD was associated with greater likelihood of subsequent PTSD (HR=1.274, 95%CI=1.271–1.277). Bidirectional associations between PTSD and AD were stronger for women compared to men. Among individuals with PTSD and AD, initial onset of PTSD was associated with younger age of first potentially traumatic event. Initial onset of AD was associated with earlier initiation of alcohol use, earlier onset of heavy alcohol use, family history of alcohol problems, and history of generalized anxiety disorder and social anxiety disorder for women but not men. Initial AD was associated with lifetime panic disorder for men and women.

Conclusions

Etiology of PTSD and AD is heterogeneous, and order of onset may reflect differing risk pathways.

Introduction

High rates of co-occurrence of posttraumatic stress disorder (PTSD) and alcohol dependence are well documented. Approximately half of individuals seeking treatment for alcohol dependence meet current criteria for PTSD (Brown, Stout, & Mueller, 1999), an estimate more than five times greater than the lifetime prevalence in the US population (Kessler, Sonnega, Bromet, Hughes, & Nelson, 1995). Data from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) indicate that about a quarter of individuals (24.2%) with lifetime PTSD meet criteria for lifetime alcohol dependence (AD), compared to 13.7% of those without a history of PTSD (Blanco et al., 2013). This co-occurrence is of clinical concern, given the severity of comorbid PTSD and AD and the impact of this comorbidity on treatment outcomes and prognosis. For example, individuals with co-occurring PTSD report more intense cravings for alcohol (Coffey et al., 2002; Drapkin et al., 2011), improve less during AD treatment, and relapse more quickly than individuals without PTSD upon completion of treatment (Brown et al., 1999; Ouimette, Finney, & Moos, 1999).

The primary, and non-mutually exclusive, etiological theories of the high rates of comorbidity between PTSD and AD are the self-medication model and the shared liability model. The self-medication hypothesis posits that individuals use alcohol and other substances to cope with psychiatric distress, thereby increasing the likelihood of developing dependency (Khantzian, 1999). Individuals with PTSD are more likely than those without PTSD to endorse using alcohol to cope with negative affect (O’Hare & Sherrer, 2011; Waldrop, Back, Verduin, & Brady, 2007). Human laboratory studies have demonstrated that individuals with comorbid PTSD-AD experience increased self-reported and physiological (i.e., salivation) craving in response to personalized trauma cues (compared to neutral cues), even when presented outside of the context of an alcohol cue (Coffey et al., 2002; Coffey et al., 2010). Self-monitoring studies have documented that individuals with PTSD have higher levels of subjective craving for alcohol on days when their PTSD symptoms are more severe (Simpson, Stappenbeck, Varra, Moore, & Kaysen, 2012). Taken together, numerous studies utilizing varied methodologies have garnered support for the self-medication model of comorbidity.

In contrast, the shared liability model suggests that PTSD and AD co-occur due to common familial risk factors (i.e., shared genes, common environment, or both) for both disorders (Krueger & Markon, 2006). Twin studies have largely supported this model. For example, one study examined the factor structure of a number of psychiatric disorders in a sample of over 3,000 male twin pairs from the Vietnam Era Twin Registry (VETR) and found evidence for shared liability between PTSD and alcohol and other substance use disorders (Wolf et al., 2010a), a finding that has been replicated in non-military samples (Xian et al., 2000). Other twin studies have documented shared liability between combat exposure and alcohol consumption (D. S. McLeod et al., 2001), combat exposure and alcohol dependence (Koenen et al., 2003), and PTSD and alcohol consumption (S. McLeod et al., 2001). Although the specific mechanisms are not yet known, it is clear that a portion of the covariance between PTSD and AD is accounted for by shared genetic and environmental risk.

Fewer investigations have examined the role of AD in the onset of PTSD, although there is reason to believe that risky alcohol consumption might increase the likelihood of experiencing traumatic events. For example, alcohol use has been associated with increased risk for sexual assault (Kaysen, Neighbors, Martell, Fossos, & Larimer, 2006; Kilpatrick, Acierno, Resnick, Saunders, & Best, 1997; Messman-Moore, Ward, & Brown, 2009). An investigation of survivors of the Oklahoma City bombing found that a pre-existing alcohol use disorder (i.e., abuse or dependence) was related to increased PTSD symptom severity following the event (North et al., 1999). Other work has found that presence of an alcohol use disorder predicted greater PTSD symptom severity in female assault survivors (Kaysen et al., 2006). However, in general, the associations between AD and subsequent trauma and PTSD are not very clear.

Community-based epidemiologic studies are useful for evaluating risk and protective factors for psychopathology because unlike clinic-based studies, they minimize selection and ascertainment bias (Westreich, 2012). Data from the National Comorbidity Survey (using DSM-III criteria) indicated that PTSD was the “primary” diagnosis more often than AD in comorbid individuals, based on the correlation between AD and the timing of individuals’ self-identified worst traumatic event (Kessler et al., 1995). A few related studies of PTSD and substance use disorders (SUD) have been conducted in epidemiologic samples. Chilcoat and Breslau (Chilcoat & Breslau, 1998b) examined a sample of approximately 1,000 adults ages 21–30 and found that PTSD was prospectively related to subsequent SUD, but that SUD did not significantly predict subsequent PTSD; however, the authors noted that the number of individuals with pre-existing SUD was likely too small to detect an effect (n=28). Follow-up analyses of this cohort showed that PTSD and SUD significantly predicted one another, although the magnitude of the effect for PTSD predicting SUD was larger than that of SUD predicting PTSD (HRs=4.5 and 1.6, respectively) (Chilcoat & Breslau, 1998a). A longitudinal study of adolescents (N=627) also indicated that baseline PTSD predicted SUD at follow-up (OR=7.29, 95%CI=1.18–45.25) but not vice versa (Wolitzky-Taylor, Bobova, Zinbarg, Mineka, & Craske, 2012), a finding that was recently replicated with respect to AD in a sample of National Guard soldiers (N=922) who had been deployed to Iraq (Kline et al., 2014). In spite of this substantial body of work, epidemiologic studies specific to PTSD and AD are needed that have sufficient sample size to adequately examine bidirectional associations and that are generalizable to the national population.

Understanding the correlates of the temporal ordering between PTSD and AD also may inform etiologic investigations as to how this comorbidity occurs. To our knowledge, only two studies, both conducted in treatment-seeking samples, have examined correlates of order-of-onset among individuals with histories of both PTSD and AD. Back and colleagues (N=94) found that women were more likely to develop PTSD prior to AD, while men were equally likely to develop PTSD or AD first (Back, Jackson, Sonne, & Brady, 2005). Men with initial PTSD compared to initial AD presented with significantly greater depressive symptoms, and women with initial AD compared to initial PTSD endorsed greater addiction severity. McLean and colleagues (N=165) found that PTSD and AD were equally likely to present first (McLean, Su, & Foa, 2014). Patients with initial AD compared to initial PTSD were more likely to have comorbid antisocial personality disorder or borderline personality disorder. Sexual assault was more common for those with initial PTSD compared to initial AD, whereas physical assault was more likely for those with initial AD.

Building on this research, the primary aim of this study was to utilize survival analyses with time-dependent covariates to examine the association between PTSD and AD in a large (N=11,103) nationally representative sample of adults endorsing lifetime alcohol consumption and trauma exposure. It was hypothesized that: (1) individuals with compared to without PTSD would be more likely to develop subsequent AD; (2) individuals with compared to without AD would be more likely to develop subsequent PTSD; and (3) the association between PTSD and subsequent AD would be stronger than that between AD and subsequent PTSD, as would be consistent with the self-medication model. Models also were run stratified by sex to examine potential sex differences in PTSD-AD associations. The secondary exploratory aim was to compare individuals who developed PTSD first versus AD first among respondents with a history of both disorders with regard to demographic (e.g., sex, ethnicity) and clinical (e.g., rates of other commonly co-occurring axis I disorders, trauma characteristics) correlates to inform hypotheses regarding etiological pathways of this comorbidity.

Materials and Methods

Participants

The NESARC is a longitudinal survey of community-dwelling US adults. Details of the study design have been described previously (Grant & Dawson, 2006). The US Census Bureau and the US Office of Management and Budget reviewed and approved the research protocol and informed consent procedures. Briefly, baseline interviews were conducted in 2001/2 (n=43,093, response rate: 81.0%), with a follow-up wave in 2004/5 (n=34,653, response rate: 86.7%). Interviews were conducted in person by trained, lay interviewers. The current investigation is limited to participants who completed both waves of interviews, endorsed lifetime alcohol consumption, and met criteria for lifetime DSM-IV PTSD Criterion A trauma exposure (final N=11,103).

Measures

PTSD and AD were assessed using the NIAAA Alcohol Use Disorder and Associated Disabilities Interview Schedule, DSM-IV version (AUDADIS-IV; Grant, Dawson, & Hasin, 2001). The reliability and validity of the AUDADIS-IV has been evaluated, with good test-retest reliability and moderate agreement against structured clinical interviews (Ruan et al., 2008).

Trauma exposure and PTSD were only assessed at Wave 2. Within the PTSD module, respondents were queried on whether they had experienced a range of 23 potentially traumatic events (PTE). Number of PTEs (used as a covariate in the analytic models) was calculated as the number of trauma categories a participant endorsed (observed range=1–20). Participants also reported on age of first PTE incident within each trauma category, and the age of the earliest PTE was calculated. PTE age data were missing for 3.8% of the sample; therefore, we imputed the missing data based on the median age of onset reported within wave 2 age groups (grouped in 10-year blocks). DSM-IV Criterion A for PTSD was assessed for participants’ self-identified “worst” traumatic event only; all other PTSD symptoms were assessed in reference to this event. The AUDADIS-IV assesses PTSD symptoms based on DSM-IV criteria (e.g., “Did you have physical reactions when reminded of [your worst trauma]?”). Age of onset of PTSD was assessed for those who met diagnostic criteria for lifetime PTSD. One and a half percent of those with lifetime PTSD had missing data for PTSD age of onset; we imputed the missing data based on the median age of onset reported within wave 2 age groups. Descriptive information (e.g., age at first PTE exposure, number of trauma categories endorsed, interpersonal PTE exposure, use of alcohol to cope with PTSD symptoms) was also assessed.

Alcohol consumption and AD were assessed at waves 1 and 2, with wave 2 assessments focusing on the time “since the last interview.” Participants endorsing having ever consumed 1 or more alcoholic beverages in their lifetime at either wave 1 or wave 2 were included in analysis. Lifetime DSM-IV diagnostic criteria for AD were assessed, with individuals meeting a diagnosis at wave 1 or wave 2 being categorized as having a positive history. Age of onset of AD was assessed among those with a lifetime diagnosis. Approximately 19% of individuals with a lifetime AD diagnosis had missing age of onset data; we imputed the missing data based on median age of onset in wave 2 age groups. Descriptive data from the alcohol assessment (e.g., age of alcohol initiation, age of onset of period of heaviest use, family history of alcohol use problems) were also assessed.

Data Analytic Plan

Descriptive analyses were conducted within the full sample, and by sex, to examine the distribution of age of onset of PTSD and AD among individuals with a lifetime history of each disorder. Among individuals with a lifetime history of both disorders, the distribution of a variable reflecting the difference in ages of onset (age of onset of AD – age of onset of PTSD) was used to assess how frequently each disorder preceded the other. Independent samples t-tests were conducted to compare men and women on ages of onset.

To test the primary hypotheses, two Cox proportional hazards (PH) models with time-dependent covariates were conducted: (1) PTSD predicting AD onset; and (2) AD predicting PTSD onset. The predictor disorder served as the time-dependent covariate, changing from 0 to 1 at the age of onset, with failure time for cases being the age of onset of the criterion disorder or censoring time (for non-cases) being the last observed age (i.e., wave 2 age). Individuals with lifetime histories of both disorders, in which the “criterion” disorder preceded the “predictor” disorder were retained in analyses, with the time-dependent covariate remaining at 0. Treating the predictor disorder as a time-dependent covariate accounts for the fact that the presence of the disorder is not stable over time and capitalizes on all available information to more accurately assess the data. The estimates of the hazard ratio (HR), an approximation of risk at a specific point in time, account for the influence of incomplete information from right-censored observations (e.g., individuals who were at-risk of developing AD [or PTSD] but did not prior to the end of the observation period). All models were adjusted for age (in years), sex (reference group=male), race/ethnicity (reference group=white), education level (reference group=“less than high school”), number of PTE categories endorsed, age of earliest PTE, and age of alcohol initiation. Covariates were selected on the basis of theoretical relevance to the models. Specifically, we wanted to ensure that primary demographic characteristics, as well as other indicators of trauma and alcohol risk, did not better account for the observed associations between PTSD and AD. The NESARC-provided sampling weight, described in detail elsewhere (Grant & Dawson, 2006), was applied to the Cox PH models. Models were also run separately by sex to assess the nature of potential sex differences. To address the secondary aim (i.e., investigation of correlates of order of onset), Chi-square analyses and t-tests were run in the sub-sample of those with lifetime histories of PTSD and AD to determine potential differences in demographic and clinical correlates between those who endorsed initial AD onset (0) compared to initial PTSD onset (1). Individuals endorsing concurrent onset of PTSD and AD were excluded from order-of-onset descriptive analyses (n=69). Statistical significance was defined as p ≤ .05. Analyses were conducted in R 3.0 (Team, 2014), with the Cox PH models being conducted using the coxph function in the survival package (Fox, 2002).

Results

Sample Characteristics and Descriptive Statistics

Table 1 provides sample characteristics. Approximately 19.8% of the total sample (n=2,198) met criteria for lifetime PTSD, with a mean age of onset of 28.5 years (SD=16.0). Men and women with histories of PTSD reported comparable ages of onset (M=28.6, SD=16.4; and M=28.1, SD=15.0, respectively; t(1281)=−0.70, p=.485). Approximately 18.2% of the total sample (n=2,021) met criteria for lifetime AD, with a mean age of onset of 29.9 years (SD=11.3). Men and women with AD reported comparable ages of onset (M=24.7, SD=9.6; and M=25.1, SD=12.4, respectively; t(2001)=0.69, p=.488). Lifetime PTSD and AD was reported in 5.4% (n=597) of the total sample. Difference in age of onset (i.e., age of AD – age of PTSD) was calculated, with a mean difference in onset for the total sample being −0.46 years (SD=14.62). The age of onset difference did not differ by sex (women: M=0.38, SD=14.07; men: M=−1.61, SD=15.28; t(512)=−1.62, p=0.106).

Table 1.

National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) Wave 2 Participant Characteristics: Trauma-Exposed/Alcohol-Exposed Subsample (N=11,103)

	Mean (SD) or %
Sex (% women)	60.6%
Race/Ethnicity (%)
White	60.5%
African American	18.7%
Hispanic/Latino	16.6%
Other	4.3%
Wave 2 Age (years)	48.7 (15.9)
Highest Level of Education Completed (%)
Less than high school degree	12.5%
High school or GED	27.6%
Some college/2-year degree	23.1%
Bachelor’s degree	23.8%
Some graduate school	13.0%
Wave 2 Past Year Household Income
Less than $20,000	19.6%
$20,000 - $39,999	25.0%
$40,000 - $59,999	19.5%
$60,000 - $79,999	13.8%
$80,000 - $99,999	7.6%
$100,000 or greater	14.3%
Number of PTE Categories Endorsed	4.4 (2.5)
“Worst” Criterion A Traumatic Event Endorsed
Serious illness/injury of a loved one	32.5%
Unexpected death of loved one	23.1%
Indirect/direct experience of 9/11 or other terrorist attack	10.8%
Life-threatening illness/injury (to self)	7.2%
Physical assault	5.1%
“Other” traumatic event	3.7%
Childhood violence/neglect	3.2%
Sexual assault	2.5%
Natural disaster	2.0%
Combat	0.4%
Unknown	9.5%
Lifetime PTSD	19.8%
Lifetime AD	18.2%
Lifetime PTSD and AD	5.4%

Cox PH Models

Table 2 shows the results of the Cox proportional hazards models. After adjusting for demographic, PTE, and alcohol initiation characteristics, PTSD was significantly associated with subsequent onset of AD (HR=1.359, 95% CI=1.357–1.362, p<.001). The association between PTSD and subsequent AD was nominally stronger for women (HR=1.370, 95% CI=1.366–1.373, p<.001) compared to men (HR=1.290, 95% CI=1.286–1.294, p<.001; Z=29.35, p<.01). After adjustment for covariates, AD also was significantly related to subsequent onset of PTSD (HR=1.274, 95% CI=1.271–1.277, p<.001). The association between AD and subsequent PTSD was significantly stronger for women (HR=1.503, 95% CI=1.499–1.507, p<.001) compared to men (HR=1.110, 95% CI=1.106–1.113, p<.001; Z=143.94, p<.01). Comparison of the HR’s in the full sample indicated that PTSD was a stronger predictor of AD than the reverse association (Z=42.37, p<.01).

Table 2.

Cox proportional hazards models (N=11,103)

	HR (95% CI)	p-value
PTSD predicting AD
PTSDa	1.359 (1.357–1.362)	< .001
Wave 2 age	0.959 (0.959–0.959)	< .001
Sex (0=male, 1=female)	0.512 (0.511–0.512)	< .001
Race/ethnicity (0=white, 1=nonwhite)	0.817 (0.815–0.818)	< .001
Education (1: high school/GED)b	0.921 (0.919–0.923)	< .001
Education (2: some college)	0.964 (0.962–0.966)	< .001
Education (3: Bachelor’s degree)	0.736 (0.734–0.737)	< .001
Education (4: some graduate school)	0.584 (0.582–0.586)	< .001
Number of PTE categories endorsed	1.145 (1.145–1.145)	< .001
Age of first PTE	1.001 (1.001–1.001)	< .001
Age of alcohol initiation	0.960 (0.960–0.960)	< .001
AD predicting PTSD
ADa	1.274 (1.271–1.277)	< .001
Wave 2 age	0.975 (0.975–0.975)	< .001
Sex (0=male, 1=female)	1.984 (1.981–1.987)	< .001
Race/ethnicity (0=white, 1=nonwhite)	1.115 (1.113–1.117)	< .001
Education (1: high school/GED)b	0.795 (0.793–0.797)	< .001
Education (2: some college)	0.715 (0.713–0.716)	< .001
Education (3: Bachelor’s degree)	0.654 (0.652–0.655)	< .001
Education (4: some graduate school)	0.624 (0.623–0.626)	< .001
Number of PTE categories endorsed	1.160 (1.159–1.160)	< .001
Age of first PTE	0.995 (0.995–0.995)	< .001
Age of alcohol initiation	0.998 (0.998–0.998)	< .001

Note:

^aTime-dependent covariate;

^bReference group = “less than high school”

Characteristics associated with Order of Onset

In the sub-sample with histories of both PTSD and AD, group comparisons of clinical characteristics were conducted separately for women and men on the basis of which disorder presented first (Table 3). Approximately 60% endorsed initial presentation of AD compared to 40% endorsing initial presentation of PTSD. With regard to trauma-related characteristics, men and women endorsing initial PTSD onset compared to initial AD onset endorsed a younger age of the first PTE (9 years versus 12–13 years, respectively). There were no significant differences in trauma type by order of onset. In terms of alcohol characteristics, women with initial AD onset compared to initial PTSD onset endorsed earlier age of initiation of alcohol use (17 and 18 years, respectively), earlier age of heaviest alcohol use (24 and 26 years, respectively), and greater likelihood of having a positive family history for alcohol use problems (72% and 60%, respectively). Men did not differ in any alcohol characteristics on the basis of order of onset. All individuals were equally likely to endorse using alcohol to cope with PTSD symptoms. Finally, women with initial AD onset compared to initial PTSD onset were more likely to endorse a history of generalized anxiety disorder, social anxiety disorder, and panic disorder. Men with initial AD onset compared to initial PTSD onset were more likely to endorse a history of panic disorder but no other diagnoses. Order of onset was not associated with lifetime major depression for men or women, although a history of depression was quite common for the entire sub-sample (>60%).

Table 3.

Characteristics of individuals with lifetime PTSD and AD (n=555) based on order of onset

	Men (n=230)			Women (n=325)
	AD first (n=137)	PTSD first (n=93)	Group difference Test	AD first (n=198)	PTSD first (n=127)	Group difference test
Number of PTE categories endorsed	6.52 (3.23)	7.67 (3.54)	t(224)=−2.51, p=.013	6.68 (3.14)	7.03 (3.33)	t(320)=−.96, p=.340
Age of first PTE	12.83 (12.23)	8.99 (6.04)	t(228)=2.81, p=.005	11.89 (9.36)	9.39 (6.75)	t(323)=2.61, p=.009
Endorsement of interpersonal PTE	83.9%	92.4%	X2=3.55, p=.059	90.9%	88.2%	X2=0.63, p=.428
Age of first alcoholic beverage	16.05 (3.31)	16.71 (4.54)	t(218)=−1.26, p=.209	17.34 (4.28)	18.43 (5.41)	t(316)=−2.00, p=.046
Age of onset of period of heaviest alcohol use	24.68 (8.37)	26.43 (10.47)	t(218)=−1.38, p=.169	24.35 (8.89)	26.41 (9.08)	t(315)=−2.00, p=.047
Family history of alcohol problems	65.1%	61.6%	X2=0.26, p=.608	72.1%	59.7%	X2=5.09, p=.024
Endorsement of alcohol use to cope with PTSD symptoms	43.8%	51.6%	X2=1.36, p=.244	35.5%	38.6%	X2=0.31, p=.578
Lifetime illicit drug use (0=no; 1=yes)	60.3%	63.4%	X2=0.23, p=.631	59.4%	61.9%	X2=0.20, p=.652
Lifetime major depression	62.0%	59.1%	X2=0.20, p=.658	79.3%	77.2%	X2=0.21, p=.649
Lifetime generalized anxiety disorder	32.1%	28.0%	X2=0.45, p=.501	41.9%	29.9%	X2=4.77, p=.029
Lifetime social anxiety disorder	29.2%	24.7%	X2=0.56, p=.456	35.4%	25.2%	X2=3.71, p=.054
Lifetime panic disorder (with or without agoraphobia)	30.7%	12.9%	X2=9.72, p=.002	41.4%	15.7%	X2=23.67, p<.001

Discussion

The primary aim of this study was to examine the bidirectional relationship between PTSD and AD in a nationally representative sample of adults endorsing lifetime alcohol consumption and trauma exposure. As hypothesized, PTSD and AD were each associated with subsequent risk for the other, although the effect sizes were more modest than those reported in prior studies. The smaller effect sizes may be due to differences in sample selection (e.g., large representative sample compared to treatment-seeking or community samples). As predicted, the association between PTSD and subsequent AD was stronger than the reverse relationship. This pattern of findings could be considered evidence for the self-medication model, although it is important to note that causal inferences cannot be made. Indeed, it is possible that common genetic and environmental factors are responsible for high rates of co-occurrence of PTSD and AD, as consistent with the shared liability model (Krueger & Markon, 2006). It has been established that approximately 30–40% of variance for PTSD (Stein, Jang, Taylor, Vernon, & Livesley, 2002; True et al., 1993) and 40–60% of that for AD (Agrawal & Lynskey, 2008; Knopik et al., 2004) is due to genetic factors, with the genetic risks for PTSD and AD being moderately correlated (Sartor et al., 2011). However, specific genetic factors accounting for shared heritability are unknown. Further research also is needed to understand shared environmental risk for PTSD-AD, including but not limited to non-traumatic psychosocial stressors, peer influences, social support, and personality factors (e.g., neuroticism).

With regard to the sex-specific models, the associations between PTSD and AD were stronger for women compared to men, regardless of the direction of association. It is possible that PTSD and AD exhibit stronger causal links and/or a larger portion of shared liability among women compared to men. Further mechanistic research is needed to understand this finding. Among women, the pathway from AD to PTSD was significantly stronger than that from PTSD to AD. It is possible that women who develop AD at an earlier age are at particularly heightened risk for PTSD due to increased risk for new incidents of sexual assault or other forms of interpersonal-type trauma under conditions of alcohol intoxication (Kaysen, Neighbors, Martell, Fossos, & Larimer, 2006; Kilpatrick, Acierno, Resnick, Saunders, & Best, 1997; Messman-Moore, Ward, & Brown, 2009). Interpersonal trauma, such as sexual assault, compared to accidental trauma is associated with greater risk for PTSD (Kessler et al., 1995). Longitudinal studies during high-risk periods for both trauma exposure and the development of risky alcohol use patterns (e.g., adolescence and early adulthood) would inform this work. Among men, the pathway from PTSD to AD was stronger than that from AD to PTSD. The self-medication model may be particularly relevant for men, although further work is needed to evaluate causal associations between these disorders.

The secondary aim of this study was to examine correlates of order-of-onset of PTSD and AD among individuals with a history of both disorders. Men and women presenting with initial PTSD compared to initial AD reported a younger age of first PTE; however, all individuals were equally likely to endorse one or more interpersonal PTEs (e.g., sexual or physical assault). Women presenting with initial AD compared to initial PTSD reported a younger age of alcohol initiation, younger age of heaviest period of alcohol use, and greater odds of having a family history of alcohol use disorder. Women with initial AD compared to initial PTSD also were more likely to meet criteria for histories of generalized anxiety disorder, social anxiety disorder, and panic disorder, whereas this effect was observed in men only with respect to panic disorder. It may be the case that a greater proportion of familial liability for PTSD (and other internalizing disorders) and AD overlaps for women compared to men. Past studies are unable to address this question, as the existing twin studies of PTSD and AD have been conducted in all-male or all-female samples only (McLeod et al., 2001; Wolf et al., 2010).

It also is possible that engaging in heavy drinking at a younger age interferes with healthy brain development during key developmental periods (e.g., adolescence), impairing individuals’ cognitive and affective regulation abilities and placing them at heightened risk for psychopathology more broadly. Animal studies have found that rats exposed to alcohol during late adolescence exhibit impairments in fear extinction, lending some preliminary support for this notion (Broadwater & Spear, 2013). In humans, data indicate that women with AD experience greater deficits in cognitive functioning as a result of heavy alcohol use than men (Flannery et al., 2007). Future research efforts examining the effects of adolescent alcohol use on brain structure and function as related to trauma/PTSD and emotion regulation phenotypes are warranted. Finally, order of onset was not associated with likelihood of endorsing alcohol use as a means of coping with PTSD, indicating that individuals with PTSD and AD are using alcohol as an emotional avoidance strategy, regardless of the potential role of self-medication in the onset of AD.

The current findings should be considered in light of a number of limitations. First, the primary survival analyses examined temporal associations between PTSD and AD, but the models were not able to accommodate additional time-dependent variables, such as alcohol initiation and exposure to multiple PTEs. However, we included these factors as non-time dependent covariates in the models. Second, we could not examine mechanisms underlying potential risk pathways among individuals with PTSD preceding AD compared to those with AD preceding PTSD. Multi-method investigations of diverse PTSD-AD risk trajectories are needed. Third, the current study relied on retrospective reports of ages of onset of PTSD and AD. Longitudinal studies with multiple assessment time points, which will undoubtedly include smaller sample sizes than that afforded by the NESARC, will allow for more fine-tuned assessment procedures, in spite of the findings being less generalizable to the US population. Fourth, PTEs and PTSD were only assessed in wave 1 of the study, which may have introduced measurement bias in the current study. Fifth, there were high rates of missingness for some of the variables in the study (e.g., age of onset of AD). We did our best to include all possible cases by applying data imputation methods, but future studies focused on consistently assessing such constructs would be useful. Sixth, the current study was unable to account for non-traumatic stressors, which may influence risk for PTSD and/or AD. Future studies assessing a range of stressful life events are needed. Seventh, the current study evaluated sex differences in PTSD-AD associations; however, pathways of risk for PTSD-AD also should be evaluated as a function of race/ethnicity, as well as other demographic factors. Finally, the current study was focused on PTSD and AD, but these conditions are associated with numerous additional comorbidities (e.g., major depression). Future efforts to account for such comorbidities in etiological models would be informative. This study also had a number of strengths, including the large representative sample, the use of validated structured instruments for assessing psychopathology, and the application of analytic methods to examine the temporal ordering of these two conditions.

In summary, the current study provided novel insight into bidirectional associations between PTSD and AD, as well as sex differences in PTSD-AD associations and correlates of order-of-onset, in a large, nationally representative sample of trauma- and alcohol-exposed adults. Findings indicate that multiple pathways to comorbidity exist and that these pathways may differ as a function of sex. Future studies focused on key developmental periods, such as adolescence and young adulthood, as well as longitudinal studies, would be useful for elucidating risk pathways for PTSD-AD comorbidity.