Impulsivity interacts with momentary PTSD symptom worsening to predict alcohol use in male veterans

ABSTRACT

Background: Posttraumatic stress disorder (PTSD) is prevalent among veterans who served post-9/11, and co-occurs with problem alcohol and substance use. Studies using ecological momentary assessment have examined the temporal association between time-varying PTSD symptoms and alcohol use. Results suggest individual differences in these associations. Objectives: We tested hypotheses that alcohol use measured by momentary assessment would be explained by acute increases in PTSD symptoms, and the PTSD–alcohol association would be moderated by trait impulsivity. Methods: A sample of 28 male post-9/11-era veterans who reported past-month PTSD symptoms and risky alcohol use were enrolled. On a quasi-random schedule, participants completed three electronic assessments daily for 28 days measuring past 2-h PTSD symptoms, alcohol, and substance use. At baseline, trait impulsivity was measured by the Barratt Impulsiveness Scale. Past-month PTSD symptoms and alcohol use were measured. Using three-level hierarchical models, number of drinks recorded by momentary assessment was modeled as a function of change in PTSD symptoms since last assessment, controlling for lag-1 alcohol and substance use and other covariates. A cross-level interaction tested moderation of the within-time PTSD–alcohol association by impulsivity. Results: A total of 1,522 assessments were completed. A positive within-time association between PTSD symptom change and number of drinks was demonstrated. The association was significantly moderated by impulsivity. Conclusion: Results provide preliminary support for a unique temporal relationship between acute PTSD symptom change and alcohol use among veterans with trait impulsiveness. If replicated in a clinical sample, results may have implications for a targeted momentary intervention.

KEYWORDS:

• PTSD
• alcohol
• veteran
• momentary assessment
• impulsivity

Introduction

Posttraumatic stress disorder (PTSD) is considered a signature wound of veterans who served in post-9/11 military operations. In this cohort of veterans, the estimated PTSD prevalence rate is astonishingly high at 20% (1). Studies have established a significant positive association between PTSD and problem use of alcohol and other substances (2–4). In a review of national survey data from a representative sample of over 3,000 US veterans, having a lifetime alcohol use disorder (AUD) was associated with four times greater odds of having a current PTSD diagnosis (5). A review of Veterans Affairs (VA) electronic health records of over 450,000 post-9/11-era veterans demonstrated similar risks of comorbidity in that among veterans with an AUD, 63% also had a diagnosis of PTSD (4).

To understand how PTSD and alcohol use covary over time in veterans, ecological momentary assessment (EMA) and other real-time data capture methods have been increasingly utilized. EMA is a behavioral sampling technique that involves administering repeated assessments in real time (6). The particular utility of EMA for alcohol use studies has been detailed (7). Several recent alcohol studies have used momentary assessment to describe patterns of use (8) and demonstrate temporal associations between momentary affect or symptom states and alcohol use (9–11). These types of studies can help to reveal the psychological motives behind drinking, and may inform the design of more precision interventions for veterans with dual diagnoses.

In veterans, EMA has been used to characterize intensive longitudinal patterns of PTSD symptom severity (12), and assess temporal associations between momentary PTSD symptom states and alcohol use (13,14). Patterns of alcohol use closely following an increase in PTSD symptoms are consistent with the self-medication hypothesis, a common clinical explanation for the co-occurrence of PTSD and problem drinking, which posits that alcohol is used to alleviate psychological distress (15–17). With a sample of 143 recent combat veterans, Possemato and colleagues demonstrated that momentary deviations from participants’ mean PTSD symptoms were positively associated with alcohol use within the same 3-h interval, but with less drinking in the immediately subsequent time block (i.e., PTSD symptoms at one occasion predicted less alcohol use at the next occasion) (13). The study demonstrated moderation of this association by avoidance coping; veterans who endorsed more avoidance coping strategies in the moment had less reduction in drinking in the interval following PTSD symptom exacerbation. This study provided new insight into the relationship between momentary PTSD symptom severity and alcohol use in veterans. The findings suggested, as would be expected, that the PTSD–drinking relationship might vary across individuals in predictable ways.

Further study is warranted to understand individual differences in the functional relationship between time-varying PTSD symptoms and alcohol use. One potential moderator is trait impulsivity, a well-documented behavioral correlate of both problem drinking (18–20) and PTSD (21,22). One facet of impulsivity in particular, negative urgency, i.e., liability toward engagement in impulsive behaviors when experiencing negative affect (23), has been the focus of studies examining its links to PTSD and substance abuse. PTSD has been associated with elevated negative urgency impulsivity and higher rates of risky substance use in veterans (24). Experience sampling approaches with college student samples have produced support for the moderating role of negative urgency in negative affect-related alcohol consumption. For example, one study (25) found that higher baseline impulsivity levels were associated with higher drinking levels and alcohol-related problems, and impulsivity moderated the relationship between negative affect and alcohol consumption, and alcohol-related problems. In another study using momentary assessment with college student drinkers (11), mood levels were differentially associated with subsequent drinking amount among students who exhibited poor response inhibition on a behavioral task. However, in an EMA study of post-9/11-era veterans, in a complex model of nighttime alcohol problems measured over a 2-week period, researchers did not find support for hypothesized moderation of the effect of daytime PTSD symptoms by negative urgency impulsivity (14).

To further explore the complex functional relationship among time-varying PTSD symptoms, alcohol use, and impulsivity, in the current study of post-9/11-era male veterans, we tested the hypothesis that, controlling for antecedent alcohol and substance use, an acute increase in PTSD symptom severity relative to the last momentary assessment would be associated with increased alcohol use within time. We proposed that this effect would be stronger among veterans with greater trait impulsivity.

Methods

All study procedures were approved by Institutional Review Boards of VA Connecticut Healthcare System and Yale University.

Sampling and enrollment

Male veterans were recruited by convenience sampling between March 2015 and January 2017 from two Department of Veterans Affairs Medical Center campuses in Connecticut to participate in a pilot study of daily PTSD symptom experiences and risk-taking. Study inclusion criteria required military service after 9/11, self-reported experience of a military-related traumatic event, and ability to reliably receive text messages and complete a web-based assessment every day for 28 days. Participants also were required to have a minimum score of 24/80 on the PTSD Checklist for the Diagnostic and Statistical Manual (DSM) – 5th edition (PCL-5) (26). This inclusive criterion was based on National Center for PTSD guidelines, recommending use of cut-point scores lower than the suggested threshold of 33/80 when a more sensitive screening measure is desired. Because an aspect of the study (not described here) focused on high-risk sexual behavior, inclusion criteria also included reporting a high-risk sexual event in the past 28 days (criterion defined in our previous study) (12). Neither alcohol nor drug use was required for study inclusion.

Participants

A total of 44 male veterans provided written informed consent and completed the baseline assessment. Of those, the 28 veterans who reported at least one binge drinking event, defined as drinking 5 or more standard drinks on a single occasion (27), within the 28 days before the baseline assessment were included in these analyses. Binge drinking is clinically important as a potential early indicator of vulnerability to AUD (28), and is associated with increased risk for accidents, violence, liver disease, and cancer (29). Characteristics of the selected sample are detailed in Table 1.

Table 1. Sample characteristics (n = 28).

	Mean (SD) or n (%)
Age	31.4 (5.7)
Race/ethnicity
White/not Hispanic or Latino	14 (50.0%)
Black/not Hispanic or Latino	8 (28.5%)
Hispanic or Latino	5 (17.9%)
Asian	1 (3.6%)
Married	3 (10.7%)
Years’ education	14.3 (1.4)
Lifetime months’ active military service	65.4 (55.6)
Substance use
Baseline past 28-day number drinks	53.5 (47.2)
Days drinking >4 alcoholic drinks/past 28 days	4.5 (3.6)
Any past 28-day marijuana use	10 (35.7%)
Any past 28-day cocaine use	2 (7.1%)
Any past 28-day opiate use	1 (3.6%)
Any past 28-day other substance use	2 (7.1%)
Percent EMA recording any alcohol use in the past 2 h	16% (15%)
Percent EMA recording >4 alcohol drinks in the past 2 h	3% (4%)
Median number of drinks per drinking occasion (n = 21)	3.4 (2.6)
Percent EMA recording any illicit substance use in the past 2 h	11% (21%)
Mental health
Baseline PCL-5	47.2 (11.4)
Preliminary diagnosis of PTSD (PCL-5 score ≥33)	24 (85.7%)
Lifetime psychiatric hospitalization	14 (50.0%)
Current medication for mental health condition	12 (42.9%)
Traumatic brain injury	6 (21.4%)
Mean momentary PCL-5/SD momentary PCL-5	27.9 (18.6)/8.2 (4.4)
Momentary PCL-5 Mean Squared Successive Differences	114.97 (126.0)
Mean momentary change PCL-5/SD momentary change PCL-5	−0.28 (1.4)/9.39 (5.5)
Barratt Impulsiveness Scale	72.93 (11.9)

Measures

PTSD Checklist for DSM-5

The PCL-5 is a 20-item self-report measure of past-month PTSD symptoms based on the diagnostic criteria outlined in the DSM-5(26)

Each PCL-5 item is rated on a scale of 0 (not at all) to 4 (extremely), defining the extent to which the respondent was bothered by the problem. The sum of all PCL-5 items renders a total score of 0–80, measuring overall PTSD symptom severity. The National Center for PTSD guidelines has proposed a cut point of 33 for provisional diagnosis of PTSD, pending additional psychometric evaluation of the PCL-5, but a lower cut point has been recommended to maximize detection of possible cases (30). For the current study, a minimum score of 24 was required for inclusion in order to enroll a more inclusive sample of veterans experiencing PTSD symptoms. The PCL-5 was administered with the Life Events Checklist for DSM-5 (LEC-5) (31).

Timeline Follow-back (TLFB) calendar

The TLFB assessed the number of alcoholic drinks and any use of marijuana, cocaine, opiates, and “other” drugs each day for the 28 days preceding the baseline date(32).

Demographics and personal history questionnaire

A demographics questionnaire assessed personal characteristics, mental health and military service history, and self-reported history of traumatic brain injury (TBI).

Barratt Impulsiveness Scale-11 (BIS-11)

This widely used, 30-item self-report questionnaire assesses general impulsiveness and individual dimensions of impulsiveness: attentional impulsiveness, characterized by inability to concentrate or sustain attention, non-planning impulsiveness, characterized by a lack of careful planning or forethought, and motor impulsiveness, a tendency to act on impulse without planning or thinking. A 4-point Likert scale is used to rate each statement with respect to how often it occurs, from rarely/never to almost always/always. Total scores range from 30 to 120 with higher scores indicating greater impulsivity. In the validation study, a mean score of 64.94 was reported for undergraduate males and a mean of 69.74 was reported for males receiving inpatient psychiatric treatment. In one study, trait impulsivity measured by the BIS-11 (33)

significantly moderated the relationship between stress and alcohol use (34).

EMA procedures

Participants received automated text messages sent to their smart phones three times per day for 28 days. Each text message was randomly timed within one of three 5-h intervals between 7 am and 10 pm (i.e., 7:00 am–12:00 pm, 12:01–5:00 pm, 5:01–10:00 pm). The message prompted participants to complete the linked, web-based assessment at that time. The assessment, created with Research Electronic Data Capture survey software (35), measured PTSD symptom severity, number of alcoholic drinks, and substance use in the past 2 h. Completed assessments were automatically date- and time-stamped and were recorded in a central database. Repeated assessments were linked by participants’ unique study identification numbers.

PCL-5 adapted for EMA

The standard 20-item PCL-5 was adapted for momentary assessment (12) to inquire about Veterans’ experiences of PTSD symptoms in the last 2 h, using the standard 0–4 response scale(26).

Substance use

Questions asked about the number of standard alcoholic drinks consumed and any use of marijuana, cocaine, opiates, and “other” illicit drugs in the past 2 h.

Participant payment

Participants were paid $2.50 for each assessment completed within the same calendar day of the text prompt, and an additional $0.50 for each assessment completed within 15 min of the text (allowing for 5 min’ delivery time), with an additional $1.00 for every day that all three assessments were completed within 15 min of the respective text prompts.

Data analysis

Data screening and total observations

Participants completed a total of 1,757 momentary assessments. Because EMA assessments measured symptoms and behaviors over the last 2 h, to avoid counting data from the same period twice, we analyzed only assessments separated by 2 or more hours from the preceding assessment. This resulted in a total of 1,522 momentary assessments (per-person Mean = 62.8, standard deviation [SD] = 23.4) occurring within 667 days (Mean = 23.82, SD = 7.42). The median within-person lag time between assessments occurring within the same day (58% of all assessments) was 5.00 hours (SD = 2.44), and across consecutive days (39% of assessments) was 15.60 hours (SD = 4.65). Thirty-nine assessments (2.6%) were separated by more than 1 day, with most of those (70%) separated by 2 days (range 2–5 days).

Descriptive analyses

Descriptive statistics summarized sample characteristics measured at baseline. The PTSD symptoms measured by momentary assessments were aggregated to summary measures reflecting overall severity (within-person mean), range (SD), and instability (Mean Squared Successive Difference). Time-varying alcohol and substance use were summarized by three separate measures: the proportion of completed momentary assessments in which any alcohol use, 5 or more drinks (i.e., binge drinking) (27), and any illicit substance use were reported. Within participants recording alcohol use, the median number of drinks per drinking occasion was calculated.

The proportion of assessments with any alcohol use recorded was calculated separately for weekends (Friday–Sunday) and weekdays, and separately for nighttime (7:00 pm–5:00 am) and daytime across all days of the week.

Missing data

To account for participant differences on variables associated with the number of momentary assessments completed that might confound alcohol modeling results, we estimated correlations between the number of momentary assessments completed and a priori-selected participant characteristics that may be relevant to alcohol use. These variables included age, race (White vs. other race), ethnicity (Hispanic vs. other ethnicity), education, history of TBI, and any substance use reported at baseline (baseline alcohol use was already controlled for in all models). Variables significantly correlated with the number of completed assessments using α = .05 (a more inclusive, uncorrected Type-1 error rate), or with a correlation coefficient ≥ |.30|, regardless of statistical significance, were included as covariates in multilevel models.

Multilevel modeling

Using three-level multilevel modeling, specifying a Poisson-distributed outcome, the number of alcoholic drinks recorded at time t was modeled as a function of person-level characteristics (level-3), day-varying covariates (level-2), and moment-varying covariates (level-1). The modeling procedure allows for unequally spaced and missing observations by use of maximum likelihood parameter estimation, which derives unbiased estimates under the assumption that data are missing at random (MAR). To increase the plausibility of the MAR assumption, variables associated with rates of response were included in the analytical model (as described above). All participants had complete data on level-3 (person-level) variables.

Person-level predictors were trait-level measures of PTSD, alcohol use, and impulsivity and included baseline (past-month) PCL-5 sum scores, the within-person mean of all momentary PCL-5 sum scores, number of alcoholic drinks recorded for the 28-day period preceding baseline, and BIS total scores. The number of assessments completed, and variables correlated with numbers of assessments completed were included as covariates. All person-level predictors were centered at the sample means of those variables.

A weekend indicator was included as a day-level predictor.

Moment-varying covariates included a nighttime (post-7 pm) indicator, time since last momentary assessment, antecedent (i.e., measured at time t − 1) number of alcoholic drinks and use of any illicit substance.

Testing the hypothesis that acute increases in symptom severity would be associated with more alcohol use, change in PCL-5 score since the last momentary assessment was included.

Intercepts at levels 2 and 3 were specified to vary randomly, and variance components for all level-1 and level-2 predictors were estimated. Variances significantly different from zero, representing significant differences in the effects of level-1 and/or level-2 predictors across persons, were retained in the model.

To test the hypothesis that veterans with greater trait impulsivity would drink more in response to an increase in PTSD symptoms, a subsequent model specifying a cross-level interaction tested moderation of the effect of momentary change in PCL-5 score by the BIS total score. To further explore the nature of the interaction, post hoc analyses tested moderation by each BIS subscale individually.

Results

Most participants (85.7%) had baseline PCL-5 scores meeting the criterion for provisional diagnosis of PTSD. Among the 21 participants who recorded any alcohol use during the sampled intervals, a total of 211 alcohol use events were recorded (Mean = 10.05, SD = 5.55), with 48 events (23% of all events) involving ≥5 drinks (Mean = 2.29 events, SD = 2.55). At least one binge drinking event was recorded for 16 of the 21 participants recording any drinking during the EMA period, with the total number of binge-drinking events among those 16 participants ranging from 1 to 9. Alcohol use events were more likely to be recorded during nighttime intervals (27% of all nighttime assessments vs. 9% of all daytime assessments) and during the weekend (23% of all weekend assessments vs. 15% of weekday assessments). The mean number of alcohol use events recorded per day, among those reporting any drinking, was 0.46 (SD = 0.25) (i.e., alcohol use was recorded approximately every 2 days, on average), and the median within-person number of alcoholic drinks recorded per drinking occasion was 3.4 (SD = 2.6). Of the 21 participants recording any alcohol use event, 11 (52%) recorded more than one event on the same day at least once.

PTSD symptoms were highly variable within person over the course of EMA. Although the mean change in PCL-5 scores between adjacent assessments was close to 0 (−0.28), reflecting fluctuations in symptoms in both directions over time, the within-person SD of change scores was 9.39 points. Characteristics of the full sample are summarized in Table 1.

Multilevel modeling

Model results are reported in Table 2. The only significant person-level predictor was the number of drinks recorded at baseline. Age was entered as the only tested covariate that was associated with number of momentary assessments completed (r = .32, p = .09), but it did not significantly predict number of drinks, and was removed from the model.

Table 2. Multilevel poisson regression of time-varying number of alcoholic drinks.

	PTSD main-effect model		PTSD BIS-moderated model
Fixed effect	b	IRR (95% CI)	b	IRR (95% CI)
Intercept	−2.68**	0.07 (0.03, 0.13)	−2.69**	0.07 (0.03, 0.13)
BL PCL-5 sum	0.04	1.04 (0.98, 1.11)	0.04	1.04 (0.98, 1.11)
Assessments completed	−0.02	0.98 (0.96, 1.01)	−0.02	0.98 (0.96, 1.01)
Mean momentary PCL-5	−0.01	0.99 (0.96, 1.02)	−0.01	0.99 (0.96, 1.02)
BL sum alc drinks	0.02**	1.02 (1.01, 1.03)	0.02**	1.02 (1.01, 1.03)
BIS	0.03	1.03 (0.98, 1.08)	0.02	1.02 (0.97, 1.07)
Weekend	0.79**	2.21 (1.52, 3.20)	0.81**	2.24 (1.54, 3.25)
Lag-1 Alcohol use	−0.15**	0.86 (0.80, 0.93)	−0.15**	0.86 (0.80, 0.93)
Lag-1 Substance use	−1.05*	0.35 (0.15, 0.84)	−1.05*	0.35 (0.15, 0.84)
Change in PCL-5	0.03*	1.03 (1.003, 1.05)	0.02	1.02 (1.00, 1.04)
Moderated by BIS	—	—	0.003**	1.003 (1.001,1.005)
Time since last observation	0.01	1.01 (0.97, 1.06)	0.01	1.01 (0.96, 1.06)
Night	1.07**	2.93 (1.64, 5.23)	1.09**	2.96 (1.68, 5.23)

Notes. CI, confidence interval; IRR, incident rate ratio; PCL-5, Posttraumatic Stress Disorder Checklist for DSM-5; BIS, Barratt Impulsiveness Scale-11. Predictors in boldface included random effects.

**p ≤ .01; *p < .05.

Nighttime and weekend indicators were both significantly associated with higher rates of drinking. Controlling for differences in the lag time since last assessment, having drunk alcohol, or used an illicit substance at the time of last momentary assessment predicted lower rates of drinking at the next assessment. As hypothesized, an increase in PTSD symptom severity from the last momentary assessment was positively associated with number of drinks in the subsequent assessment. This effect was significantly moderated by trait impulsiveness such that participants with higher scores on the BIS experienced greater increases in drinking in association with the same increase in PTSD symptom severity. For example, relative to a participant at the sample mean of BIS-11 for whom a 5-point increase in PTSD symptom severity was associated with a 10% higher rate of drinking (relative to no change in PTSD symptoms under the same conditions), a participant with impulsivity of 1 SD (i.e., 11.9 points) above the sample mean on BIS-11 experiencing the same (5-point) increase in PTSD symptom severity would experience a 28% higher rate of drinking. In this example, the rate increase effected by a 5-point increase in PTSD symptoms is equivalent to an increase of one standard drink when the base (i.e., no change in PTSD symptoms) drinking rate is approximately 3.5 drinks.

Post hoc analyses revealed that the Non-planning impulsiveness subscale, but not Attention or Motor impulsiveness subscales, moderated the effect of PTSD symptom change on rate of drinking.

Random effects

The effects of antecedent alcohol use, antecedent substance use, and the weekend indicator did not differ significantly across persons, and random effects for these predictors were trimmed from the model. Significant between-person variance was observed on the effects of change in PCL-5 scores, time since last observation, and night, and these random effects (indicated in boldfaced font) were retained.

Discussion

Although this study’s inclusion criteria permitted enrollment of male veterans with subthreshold PCL-5 scores, the majority of participants met criteria for a provisional PTSD diagnosis. The severity of mental health issues experienced by this sample was also suggested by half having experienced a lifetime psychiatric hospitalization and a substantial minority of participants (42.9%) taking a prescribed medication for a mental health condition at the time of baseline assessment.

Participants’ scores on the BIS were slightly higher than scores reported in the instrument’s validation study for men receiving inpatient psychiatric services, suggesting this sample was particularly impulsive. The scores reflect the nature of this select sample of fairly young, mostly single men with PTSD and recent high-risk behavior (24).

The means of participants’ momentary PCL-5 scores were notably lower than their baseline PCL-5 scores. This difference may represent differences in assessment, or may reflect heuristics used to describe past-month PTSD severity, whereby recalled severity is influenced by peak symptom experiences (Black et al., manuscript in preparation). Over the course of the EMA period, participants experienced substantial variability in symptom severity. This is consistent with illustrated symptom patterns from our group’s previous study (12).

Although alcohol use was recorded during the EMA period by only 21 of 28 participants, the number of alcohol use events recorded was fairly remarkable, given that only three 2-h periods per day were sampled, and only one of those occurred after 5 pm each day. This sampling strategy captured binge drinking events for 16/28 participants, and captured a drinking occasion in 16% of all sampled intervals, on average. Hypothesized associations between PTSD symptoms and alcohol use were supported: the number of alcoholic drinks consumed at a given time was accounted for by experienced increases in PTSD symptom severity from the last momentary assessment, and this effect was stronger for participants with a higher trait impulsiveness. The post hoc finding that moderation by impulsiveness was driven by the Non-planning impulsiveness subscale is consistent with a self-medication interpretation of the PTSD–alcohol use association. Veterans with greater tendency toward a short-term focus might be more likely to seek immediate amelioration of their worsening symptoms with less regard for the longer-term detrimental effects of this strategy. Nevertheless, given the correlational nature of the results, and lack of temporal ordering between PTSD symptom change and alcohol use, we cannot rule out the possible, less explored, explanation that alcohol use exacerbated PTSD symptoms within time, and that this effect was stronger among veterans with greater impulsiveness.

The study includes several limitations that must be considered. This was a small convenience sample of male veterans who used services at a VA Medical Center. Many participants recorded relatively low levels of alcohol use during the EMA period. It is unclear how these study results may generalize to a more representative sample of veterans, those not connected with VA medical services, and to those with more severe alcohol use.

Nevertheless, the results illustrate contextual risk factors for alcohol use in male veterans experiencing PTSD symptoms. If these results are replicable in a larger, clinical sample of veterans, they may have important implications for clinical practice. One suggested direction of these preliminary results is to provide veterans with targeted momentary supports to manage acutely worsening PTSD symptoms. These may include prompts delivered in real time to implement practiced stress-reducing exercises and to avoid alcohol-related cues or triggers at those times (i.e., avoid purchasing alcohol or going to a bar). Such an intervention could be particularly effective for veterans with problematic alcohol use and greater impulsivity. Results also suggest the utility of screening veterans for impulsivity upon intake for PTSD or substance use treatment, and implementing interventions that target impulsivity more generally that may reduce PTSD symptom-associated alcohol use.

The increasing availability of technology capturing real-time data has afforded new insights into overtime patterns of mental health symptoms and their functional association with alcohol and other substance use. Emerging patterns across studies provide an empirical basis for developing targeted and precisely timed momentary interventions. Studies are now needed that test the feasibility and effectiveness of ecological momentary interventions addressing symptom-associated substance use in individuals with PTSD.

Declaration of interest

The authors report no relevant financial conflicts.

Additional information

Funding

This work was supported by R21 DA039038 (ACB), V1CDA2014-27 (ACB), VA HSR&D Pain Research, Informatics, Multi-morbidities, and Education (PRIME) Center CIN 13-407, and the VISN 1 Mental Illness Research Education and Clinical Center (MIRECC).