Procedure
Data were collected via a telephone survey from a sample of community-based U.S. military veterans recruited to assess the health effects of military service.15 All participants were outpatients of Geisinger Clinic, the largest multi-hospital system in central and northeastern Pennsylvania (see: www.geisinger.org), serving more than 3 million residents. Starting in 2007, Geisinger Clinic began to keep records on veteran status and patients were asked to complete a military history questionnaire. Participants were randomly selected using Geisinger’s Electronic Health Record (EHR) system. We used screener questions at the beginning of the survey to exclude potential participants who were institutionalized or incapable of completing a 60-minute interview due to physical, language, or cognitive impairment. Screener questions were also used to identify participants who met our inclusion criteria: being able to complete the survey in English, being between 18 and 75 years old, and having at least one warzone deployment. After obtaining informed consent, trained interviewers administered a structured diagnostic interview which took place between February 2016 and February 2017. The final sample size was 1,730, and the survey cooperation rate was estimated to be approximately 55%.17 The average time for participants to complete the survey was about 65 minutes and participants were offered a $30 incentive for participation. The Institutional Review Boards of Geisinger (IRB #2015-0441) and the US Department of Defense (IRB #A-18989) reviewed and approved the study protocols.
Dependent Variables
The study focused on four outcome variables: PTSD, depression, suicide ideation, and use of mental health services. To assess lifetime and past year PTSD, we used a diagnostic instrument and the PTSD Checklist,18 based on the Association’s Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5). To receive a diagnosis of PTSD, veterans had to meet the DSM-5 diagnostic criteria A through G: trauma exposure (criterion A), intrusive symptoms (criterion B), persistent avoidance (criterion C), negative alterations in cognitions/mood (criterion D), increased arousal (criterion E), and reported impairment/distress related to these symptoms (criterion G). Almost 80% of the veterans in the current study reported that the most significant lifetime stressor they experienced was warzone or combat exposure. Lifetime and past year depression were measured using a 10-item (coded no/yes) version of the Structured Clinical Interview for DSM (SCID) Major Depressive Disorder used in previous studies.15 Consistent with DSM-IV, respondents met criteria for depression (Cronbach’s alpha = 0.87), if they had 5 or more depression symptoms for at least 2-weeks.19, 20 We also used one item from the DSM-IV depression scale to measure Lifetime Suicide Ideation to focus more specifically on this important health outcome. Specifically, the item asked if respondents had ever had thoughts for two weeks or longer about how they would be better off dead or “of hurting yourself in some way” (no vs. yes). Lastly, the survey inquired about the use of eight different mental health services (psychiatrist, general practitioner or other counseling, psychologist, counselor, spiritual advisor, priest, minister, rabbi, etc., social worker, or other types of health professionals or self-help groups) for problems with emotions, nerves, or use of alcohol or drugs (n/y). Use of any of these services over the past year was coded yes (1), otherwise no (0). As with many of our measures, these were used in a number of previous studies.15,16
Independent Variables
Our survey included demographic, military experiences, recent stressors, social connections, and psychological resource independent variables known to affect mental health. Demographic variables were sex, age, race, marital status, and education, which were coded such that female, older (65+), White, married or living together as married, and college graduate were coded as the indicator variable. Our military experience variables included deployment era (Iraq/Afghanistan vs. other eras), multiple combat zone tours (coded two or more vs. one) and deployed as National Guard/Reserve (no vs. yes). We measured combat exposure based on eight items from the Combat Experiences Scale.16, 21-22 A number of military health studies used versions of this scale since the Vietnam war period.23 The 8 items (rated 1 to 4) asked about encountering dead bodies, being wounded by hostile fire, killing enemy combatants, and other combat related events (Cronbach’s alpha = 0.84). We coded the sample into high combat exposure (≥75th percentile) versus lower exposure. Unit support/morale was the mean of six items from the Deployment Risk and Resilience Inventory that inquired about a sense of camaraderie in the unit, trust of other unit members, commanding officers being interested in how they felt, feeling like efforts counted in the military, during deployment, etc.24, 25 (Cronbach’s alpha = 0.78). We coded respondents into those who felt a high sense of support and unit morale versus those who did not using the scale’s 25th percentile. These measures were updated to reflect current conflicts.15 Lastly, concussion history was assessed based on reported concussions experienced during military service (no/yes). We note that all of these measures are important dimensions of military service in combat zones.
The analysisincluded three measures of stress based on previous work.14, 16 Stressful events in the past year was the sum of 8 experiences (e.g., spouse/mate die, serious injury, problems at work, etc.), which has been used in past research.16 Lifetime trauma was the sum of 12 experiences (e.g., natural disaster; being attacked with gun, knife, weapon; situation where you were seriously injured or suffered physical damage; forced sexual contact, etc.) that could have happened to the respondent in their lifetime. Finally, the Adverse Childhood Experiences measure was the sum of 9 events (e.g., parent swear or indult you; parent push, grab, slap, or push you) that could have happened to the participant before they were 18.26 This measure of childhood abuse and neglect has been used in many studies, showing good validity and reliability.26, 27 For all three of these measures, we divided the sample into low vs high exposure (high = 2+ events).
Lastly, the analyses included several psychological, social, and physical health factors that could help explain sex differences in well-being. Psychological resilience was assessed by the 5-item version of the Connor-Davidson Resilience Scale,28 with respondents who fell below the 25th percentile defined as having low resilience.15 The Cronbach’s alpha for this scale was 0.99.The social support scale (e.g., someone available to help you if you were confined to bed) used in this study was based on four questions (Cronbach's alpha = 0.84) that inquired about emotional, informational, and instrumental support, coded 1-4 (“none of the time” to “all the time”).29 This scale has been used in previous trauma studies and is considered a reliable and valid measure of current social support.15,30-32 Low social support was defined as cases falling at or below the 25th percentile.30 Self-rated physical health was assessed using one item (fair/poor vs. good to excellent). The survey inquired about past year heavy drinking, which we operationalized as 40 or more drinks per month (coded 1), and self-esteem, using 5-items from Rosenberg’s Self-Esteem scale (Cronbach’s alpha = .87), which we divided into low vs. high categories, using the 75th percentile as the cut-point. These measures have beed used in other studies and show good validity.15-16
Finally, we assessed several variables related to health services use. VA service use was assessed using single item questions inquiring about current and lifetime use of VA healthcare services. We also used a single item to ask about current VA disability status. In addition, the survey asked about the use of psychotropic medications in the past year. More specifically, the survey inquired about anti-depressants, tranquilizers, sleeping pills or other medicines for problems with emotions, nerves, concentrating, sleeping or coping with stress over the past 12 months. All these measures were used in previous studies.15, 16
Analytic Strategy
We present descriptive statistics and bivariate differences between male and female veterans (Table 1). Given the low number of females in our sample, we conducted preliminary analyses focused on lifetime disorders and retained variables that predicted these outcomes using multivariate logistic regression (Table 2). All variables in the multivariate models had complete data, except for age which had two missing values. We dropped these cases from the analyses. To further examine the relationship between gender and our outcomes, we performed propensity score matching at 1:1, 1:3, and 1:5 ratio of female to male using nearest neighbors and compared these results from propensity matched cohorts to conventional multivariate logistic regression.33
Table 1: Demographic, Deployment, and Well-Being Measures for Total Sample and by Sex (N=1727-1730)
|
(N)
|
% Total
|
Sex
|
|
|
Study Variables
|
%Male
|
%Female
|
χ2
|
p-value
|
Age: 18-64
|
(751)
|
43.5
|
40.8
|
95.3
|
97.75
|
<0.001
|
White Race
|
(1655)
|
95.7
|
95.9
|
90.6
|
5.56
|
0.018
|
Married
|
(1340)
|
77.5
|
78.8
|
50.6
|
36.96
|
<0.001
|
College Graduate or Higher
|
(429)
|
24.8
|
23.6
|
47.1
|
23.56
|
<0.001
|
Iraq/Afghanistan Veteran
|
(396)
|
22.9
|
21.3
|
54.1
|
49.39
|
<0.001
|
Multiple Tours
|
(686)
|
39.7
|
40.3
|
29.4
|
3.97
|
0.046
|
Deployed Guard/Reserve
|
(665)
|
38.4
|
37.0
|
65.9
|
28.45
|
<0.001
|
High Childhood Abuse/Neglect
|
(288)
|
16.6
|
16.5
|
20.0
|
0.72
|
0.395
|
High Combat Exposure
|
(408)
|
23.6
|
24.7
|
2.4
|
22.36
|
<0.001
|
Low Unit Support
|
(364)
|
21.0
|
20.5
|
31.8
|
6.19
|
0.013
|
High Stressful Events Past Yr.
|
(375)
|
21.7
|
21.3
|
29.4
|
3.15
|
0.080
|
High Lifetime Trauma
|
(357)
|
20.6
|
20.6
|
21.2
|
0.90
|
0.891
|
Low Psych Resilience
|
(439)
|
25.4
|
24.3
|
45.9
|
19.85
|
<0.001
|
Low Current Social Support
|
(314)
|
18.2
|
17.9
|
23.5
|
1.74
|
0.187
|
Fair/Poor Current Health
|
(633)
|
36.7
|
37.2
|
26.2
|
4.16
|
0.041
|
Concussion in Service
|
(491)
|
28.4
|
29.1
|
14.1
|
8.95
|
0.003
|
Heavy Drinking Past 30 Days
|
(208)
|
12.0
|
12.5
|
2.4
|
7.90
|
0.005
|
Low Self-Esteem
|
(400)
|
23.1
|
22.7
|
30.6
|
2.80
|
0.094
|
PTSD Past Year
|
(132)
|
7.6
|
7.3
|
14.1
|
5.34
|
0.021
|
PTSD Lifetime
|
(216)
|
12.5
|
11.6
|
29.4
|
23.44
|
<0.001
|
Current Depression Disorder
|
(143)
|
8.3
|
7.8
|
17.6
|
10.38
|
0.001
|
Lifetime Depression Disorder
|
(381)
|
22.0
|
20.8
|
45.9
|
29.63
|
<0.001
|
Recent Suicidal Thoughts
|
(94)
|
5.2
|
5.2
|
9.4
|
2.75
|
0.1.33
|
Ever Suicidal Thoughts
|
(196)
|
11.3
|
10.5
|
27.1
|
22.02
|
<0.001
|
Currently Using VA Service
|
(864)
|
49.9
|
50.3
|
43.5
|
1.47
|
0.225
|
Current VA Disability
|
(629)
|
36.4
|
37.0
|
24.7
|
5.25
|
0.022
|
Use Psych Services Past Yr.
|
(406)
|
23.5
|
22.4
|
43.5
|
20.03
|
<0.001
|
Ever use of Psych Services
|
(832)
|
48.1
|
47.1
|
67.1
|
12.88
|
<0.001
|
Use Psychotropics Past Yr.
|
(384)
|
22.2
|
21.4
|
37.6
|
12.36
|
<0.001
|
N (%)
|
|
|
1645(95.1)
|
85(4.9)
|
|
|
Table 2: Odds Ratios and 95% Confidence Intervals for Mental Health Outcomes Regressed on Demographic, Deployment, Drinking, and Psychological Resource Variables (N=1728)
Independent
Variables
|
Lifetime PTSD
OR (95% CI)
|
Lifetime Depression
OR (95% CI)
|
Ever Suicidal Thoughts
OR (95% CI)
|
Lifetime Psych Services
OR (95% CI)
|
Sex (Female)
|
5.16 (2.61-10.20)***
|
3.06 (1.74-5.49)***
|
2.62 (1.40-4.93)***
|
1.68 (0.98-2.87)
|
Age (65+)
|
0.71 (0.48-1.03)
|
0.50 (0.38-0.67)***
|
0.72 (0.51-1.03)
|
0.52 (0.41-0.66)***
|
College Graduate or Higher
|
0.83 (0.54-1.26)
|
0.81 (0.58-1.13)
|
1.00 (0.68-1.49)
|
0.95 (0.74-1.22)
|
Married
|
1.41 (0.92-2.14)
|
1.05 (0.75-1.45)
|
1.21 (0.81-1.81)
|
0.66 (0.51-0.87)**
|
High Child, Abuse/Neglect
|
1.64 (1.10-2.43)*
|
2.13 (1.54-2.95)***
|
2.25 (1.56-3.24)***
|
1.87 (1.38-2.55)***
|
High Stress past Yr.
|
3.29 (2.28-4.77)***
|
2.27 (1.67-3.08)***
|
1.23 (0.84-1.79)
|
2.05 (1.54-2.74)***
|
High Lifetime Trauma
|
2.38 (1.64-3.45)***
|
1.64 (1.20-2.25)**
|
1.20 (0.82-1.76)
|
1.55 (1.16-2.05)***
|
High Combat Exposure
|
3.00 (2.03-4.44)***
|
1.86 (1.34-2.58)***
|
1.20 (0.81-1.79)
|
1.69 (1.29-2.22)***
|
Concussion in Service
|
2.27 (1.56-3.31)***
|
1.50 (1.10-2.04)**
|
1.13 (0.77-1.64)
|
2.05 (1.59-2.66)***
|
Heavy Drinking
|
1.66 (1.01-2.72)*
|
1.09 (0.71-1.64)
|
1.17 (0.72-1.91)
|
0.89 (0.64-1.24)
|
Low Self-Esteem
|
3.06 (2.09-4.49)***
|
3.41 (2.51-4.63)***
|
4.27 (2.95-6.18)***
|
2.37 (1.77-3.18)***
|
Low Resilience
|
2.26 (1.54-3.31)***
|
2.33 (1.72-3.15)***
|
1.95 (1.35-2.81)***
|
2.03 (1.54-2.67)***
|
Low Current Social Support
|
1.69 (1.12-2.54)*
|
1.62 (1.16-2.27)**
|
1.34 (0.91-1.98)
|
0.95 (0.70-1.27)
|
Constant
|
0.011***
|
0.076***
|
0.034***
|
0.11***
|
Logistic Regression: OR-Odds Ratio CI-Confidence Interval
Significance levels: * p < .05 * p < .01 *** p < .001
Propensity score matching
For propensity score matching we first included the confounding covariates listed in Table 2, including age, college graduate, married status, anti-social disorder, high combat exposure, serving on multiple tours, ever used mental health services, low psychological resilience, high neglect/abuse history, high current life stress, high lifetime trauma, history of concussion, low self-esteem, and low social support. In addition to these variables, we also added history of ADHD (doctor told respondents they had this health problem, no vs. yes), Iraq/Afghanistan service, low unit support during deployment, military rank (officer vs. enlisted), white race, assessment of stable emotions, based on the 5-factor personality scale34, and current reported VA service use to estimate the propensity score for the variable, female sex. Then our matching procedure, was executed using 1:1 nearest neighbor matching without replacement where a single female participant was matched to a single male participant who had the most similar estimated propensity score with a caliper of 0.2. In addition, as the sample sizes of the female and male participants vary greatly, we performed the one to many matchings where a single female participant was matched to more than one male participant (such as 1: 3 and 1: 5 matching) using nearest neighbor based on propensity scores.33 There were 85 females and 1644 males in original dataset. After propensity score matching, there were 85 females and 85 males selected for the 1: 1 matching; 85 females and 255 males selected for the 1: 3 matching; 85 females and 425 males selected for the 1: 5 matching. Multivariate logistic regression models were then conducted for the 1 : 1, 1 : 3 and 1 : 5 nearest neighbor matching to evaluate the sex differences in predicting lifetime PTSD, lifetime depression, suicidal thoughts, and use of psychological services (Table 3). The propensity scores matching procedures were conducted in RStudio Version 1.2.1335, the “MatchIt” package. 35, 36 As a further test of model adequacy, we performed a sensitivity analysis using the Wilcoxon Signed Rank Test, where the value of Gamma can be interpreted as the odds of the model suffering from hidden bias due to omitted (unmeasured) variables.37
Table 3: Multivariate Logistic Regression Results using Propensity Score Matching for Lifetime PTSD, Depression, Suicide Ideation, and Psychological Services
Dependent Variable
|
OR (95% CI)
|
Pr(>|z|)
|
Matching: PTSD
1 to 5 female vs. male
1 to 3 female vs. male
1 to 1 female vs. male
|
5.19 (2.43 11.09)
5.13 (2.32 11.34)
11.55 (3.06 43.63)
|
0.00002***
0.000523***
0.000309***
|
Matching: Major Depression
1 to 5 female vs. male
1 to 3 female vs. male
1 to 1 female vs. male
|
3.27 (1.77 6.06)
3.04 (1.61 5.74)
2.64 (1.19 5.84)
|
0.000162***
0.000635***
0.0171*
|
Matching: Suicide Ideation
1 to 5 female vs. male
1 to 3 female vs. male
1 to 1 female vs. male
|
2.48 (1.30 4.75)
2.82 (1.41 5.63)
3.99 (1.53 10.37)
|
0.006106**
0.003367**
0.004533**
|
Matching: Psych Services
1 to 5 female vs. male
1 to 3 female vs. male
1 to 1 female vs. male
|
1.91 (1.09 3.38)
1.96 (1.08 3.57)
1.46 (0.71 2.99)
|
0.024818*
0.027033**
0.299672
|
Significance levels: *** <0.001 ** <0.01 * <0.05.