In the total cohort, the hazard function showed that the rate of suicide increased in a largely monotonic fashion.
Reger MA, Smolenski DJ, Skopp NA, Metzger-Abamukang MJ, Kang HK, Bullman TA, Perdue S, Gahm GA. Risk of Suicide Among US Military Service Members Following Operation Enduring Freedom or Operation Iraqi Freedom Deployment and Separation From the US Military. JAMA Psychiatry. 2015;72(6):561-569. doi:10.1001/jamapsychiatry.2014.3195
Copyright 2015 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
A pressing question in military suicide prevention research is whether deployment in support of Operation Enduring Freedom or Operation Iraqi Freedom relates to suicide risk. Prior smaller studies report differing results and often have not included suicides that occurred after separation from military service.
To examine the association between deployment and suicide among all 3.9 million US military personnel who served during Operation Enduring Freedom or Operation Iraqi Freedom, including suicides that occurred after separation.
Design, Setting, and Participants
This retrospective cohort design used administrative data to identify dates of deployment for all service members (October 7, 2001, to December 31, 2007) and suicide data (October 7, 2001, to December 31, 2009) to estimate rates of suicide-specific mortality. Hazard ratios were estimated from time-dependent Cox proportional hazards regression models to compare deployed service members with those who did not deploy.
Main Outcomes and Measures
Suicide mortality from the Department of Defense Medical Mortality Registry and the National Death Index.
Deployment was not associated with the rate of suicide (hazard ratio, 0.96; 99% CI, 0.87-1.05). There was an increased rate of suicide associated with separation from military service (hazard ratio, 1.63; 99% CI, 1.50-1.77), regardless of whether service members had deployed or not. Rates of suicide were also elevated for service members who separated with less than 4 years of military service or who did not separate with an honorable discharge.
Conclusions and Relevance
Findings do not support an association between deployment and suicide mortality in this cohort. Early military separation (<4 years) and discharge that is not honorable were suicide risk factors.
The US military has traditionally experienced lower suicide rates compared with the general US population.1,2 However, the suicide rate among the active duty US military has increased in the last decade, almost doubling in the Army and Marines Corps.3 Research on the potential effect of deployments to Operation Enduring Freedom or Operation Iraqi Freedom (OEF/OIF) is limited. Department of Defense (DoD) reports show that approximately half of the suicide cases that occur on active duty did not have a history of deployment.4 However, these descriptive DoD studies are generally unable to track mortality among service members after they separate from military service (when DoD jurisdiction ends). Because service members who screen positive for mental health concerns following a deployment are more likely to separate from military service,5 it is important to account for service member suicides that occur after separation from service.
Research on veteran suicide risk factors has increased in recent years,6 but much of the research is limited to veterans who access health care from the Department of Veterans Affairs (VA).7,8 This subgroup represents only approximately 35% of all veterans.9 Furthermore, some epidemiological research relies on death certificates to classify veteran status.10 This approach relies on funeral directors and others, who use methods that vary widely from state to state.11 In addition, length of military service may be an important factor in considering suicide rates among those who have separated from the military because each year thousands of US military personnel fail to complete basic or advanced training or are discharged because of legal problems, adjustment reactions, alcohol and drug–related problems, and other administrative reasons.10
Systematic reviews of military suicide prevention have noted these methodological problems in the epidemiological literature and have called for efforts to link multiple federal databases to address these concerns.10,12 The effect of combat deployments on suicide risk is an issue of national and international importance.13- 15
Two recent DoD studies16,17 have partially addressed these methodological limitations. However, the results on the association between suicide and deployment differed. One study16 included only service members from the regular component of the Army and included only suicides that occurred during military service. The other study17 included only 83 suicides from a survey sample with a low response rate.
The present study examined the association between deployment and suicide among all 3.9 million service members who served in the US military from the beginning of OEF on October 7, 2001, to December 31, 2007. Suicide mortality was followed from October 7, 2001, to December 31, 2009, regardless of separation from military service. To our knowledge, this is the most comprehensive study to date to examine suicide risk in relation to OEF/OIF deployment.
This retrospective cohort study included all uniformed service personnel who were in the military at any time between October 7, 2001, and December 31, 2007. All service members who were in the active or reserve components of the Air Force, Army, Marine Corps, or Navy at any point between October 7, 2001, and December 31, 2007, were eligible for inclusion in the cohort. The initial population included 3 945 099 service members. Data related to service characteristics and demographics were ascertained from records provided by the Defense Manpower Data Center.18 Data on mortality covered the cohort eligibility window and extended through December 31, 2009.
This study was reviewed and approved as a minimal risk protocol by the Western Regional Medical Command/Madigan Army Medical Center Institutional Review Board. This protocol was also reviewed by the US Army Medical Research and Materiel Command Office of Research Protections, Human Research Protection Office, and Washington Veterans Affairs Medical Center Institutional Review Board. Because this was a retrospective study, an informed consent requirement was waived by the institutional review boards.
The primary exposure variable was a deployment in support of OEF/OIF between October 7, 2001, and December 31, 2007. We identified deployments in the Contingency Tracking System records provided by the Defense Manpower Data Center. Deployments were defined in accord with standard DoD surveillance practice as follows: (1) a duration of at least 30 days per the parent (primary) start and end dates and (2) a separation of at least 30 days between parent deployments to distinguish 2 deployments as separate deployments. Furthermore, deployments in support of OEF and OIF were identified by having a location assignment at any time during a parent deployment to any of the following land locations: Afghanistan, Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates, Yemen, Djibouti, Iraq, Kyrgyzstan, and Uzbekistan.14 Water locations included designations for the Red Sea, Gulf of Aden, Gulf of Oman, and Arabian Sea.14 Deployments to other locations (or in other years) were not included.
The secondary exposure variable was separation from service, defined as a separation to civilian life from either the active or reserve components without a subsequent record of return to service.19 Therefore, individuals serving in uniform with the National Guard or reserve component (eg, performing weekend training and periodic duty assignments) were considered in the military (not separated), while personnel separated to the Inactive National Guard or the Individual Ready Reserve (minimal annual requirements such as keeping personal contact information current) were considered separated. Time in service was defined as the difference between the date of a final separation from service and the pay entry base date.
The primary outcome for the study was death by suicide. We used data from the Medical Mortality Surveillance Division of the Armed Forces Medical Examiner System (AFMES) as a gold standard source of mortality data for service members.20 The AFMES maintains the DoD Medical Mortality Registry, which tracks all deaths that occur among uniformed service members when in an active status, regardless of geographic location. The National Death Index (NDI) identified deaths in the entire cohort from October 7, 2001, to December 31, 2009, as a secondary source. Death records in the NDI include all deaths that occurred within the United States. This source provided cause-of-death data for cohort members who were not eligible to be included in the AFMES death data because of a separation from service or because National Guard or reserve component personnel died in a nonduty status (outside the jurisdiction of the DoD). International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, codes were used to identify suicides in accord with the National Vital Statistics Reports standard.21
We examined the sensitivity of the NDI to the AFMES records among cohort members who could have been in both sources at the time of death (ie, active duty deaths that occurred in the United States). The sensitivity for the active duty subpopulation who died in the United States was 97.1%, and the overall agreement on suicide as the manner of death between the AFMES and the NDI was 98.2% (κ = 0.94, P < .001). Prior studies17,22 have also successfully identified military suicides with the NDI and AFMES data.
We used extended Cox proportional hazards regression models to estimate the hazard ratios of associations between suicide and the 2 hypothesized determinants of deployment and separation from service. For the primary models, we specified a timescale of calendar time beginning at the onset of the risk window (October 7, 2001) and ending December 31, 2009. Each service member entered the cohort at the beginning of the risk window or at his or her pay entry base date if he or she entered the military after the beginning of the risk window. Exit from the cohort in this model occurred at the date of death from any cause or at the end of the risk window. Deployment and separation from service were specified as time-varying exposure variables. Once a deployment occurred or the service member separated from service, that service member was considered exposed for the remainder of the risk window.
We estimated the following 4 separate models: (1) any deployment, (2) one or more than 1 deployment compared with no deployment, (3) separation from service, and (4) a joint effects model of any deployment and separation from service. Models include time-invariant covariates of sex, age at cohort entry, educational attainment at cohort entry, race/ethnicity, and service branch at cohort entry. Time-varying covariates were rank and component affiliation. We also included marital status at cohort entry for all cohort members. Because we did not have the same precision of changes in marital status as of changes in military characteristics, we did not specify this as time varying. However, we used the last known status at separation from service for the time at risk after separation. We used a likelihood ratio test of effect measure modification between the primary exposure in each model and service branch. To characterize any identified effect measure modification, we estimated the 4 models separately by service branch.
As a secondary analysis, we examined service characteristics at separation as determinants of suicide rates among those who separated from the military. We used time since separation as the timescale. We estimated the following 2 models: (1) a model with the amount of time in service at the time of separation at the primary exposure and (2) a model with the characterization of service as the primary exposure. For the second model, we restricted the sample to those who reverted to a civilian or undefined status at the time of separation as opposed to retired because they did not have a characterization recorded at their time of separation.
For all models, we examined the Cox proportional hazards regression assumption using graphical methods and goodness-of-fit tests.23 For descriptive rates and all hazard ratio estimates, we evaluated precision and statistical significance using 99% CIs. We chose to use the more stringent α = .01 given the large size of the cohort and potential for statistical significance of otherwise trivial associations. Given the lack of exposure data between 2007 and 2009, we reestimated all models using December 31, 2007, as the end of the risk window to evaluate the consistency of the hazard ratios. We used a software package (STATA, version 12.1; StataCorp LP) to conduct all statistical analyses.24
The final cohort included 3 945 099 service members. A total of 31 962 deaths occurred (of which 5041 were identified as suicides) by December 31, 2009.
Table 1 summarizes the distribution of suicide cases, person-years, and crude rates by several demographic characteristics of the cohort. We observed higher crude rates of suicide for service members who were younger, male, not married, and identified as non-Hispanic white, Native American, or unspecified racial/ethnic category. Crude suicide rates were slightly elevated for active component compared with reserve component service members. Service members in the Army and Marine Corps had higher crude rates of suicide than service members from the Air Force and Navy. Finally, consistent with the baseline age pattern for suicide rates, service members in junior enlisted ranks had the highest rate of suicide.
In the total cohort, the hazard function showed that the rate of suicide increased in a largely monotonic fashion (Figure). We did not observe any differences in the hazard of suicide between those who had deployed and those who had not (Table 2). This was consistent for both specifications of deployment (any deployment or 1 vs >1 deployment). There was no evidence of effect measure modification as a function of service branch affiliation (likelihood ratio χ23 = 0.97, P = .81 for any deployment and χ26 = 2.04, P = .92 for 1 vs >1 deployment). We observed a substantial increase in the hazard of suicide as a function of separation from service. In the joint effects model, we observed a small increase in the hazard of suicide associated with deployment before separation from service, but it was only marginally statistically significant. For those who had separated from service, the hazard of suicide was high, irrespective of deployment history. In contrast to the deployment models, we observed evidence of heterogeneity by service branch for both the separation from service model (likelihood ratio χ23 = 15.82, P = .001) and the joint effects model (likelihood ratio χ29 = 25.45, P = .003).
Table 3 summarizes the results of the 4 main models separately by service branch. Consistent with the results of the likelihood ratio test, the hazard ratios for the 2 deployment models were consistent across service branch and did not indicate an association between deployment and the hazard of suicide. For all 4 service branches, the hazard of suicide increased after separation from service. The smallest hazard ratio was for Marine Corps service members, and the largest hazard ratio was for Navy service members. The results of the joint effects model were similar in kind to those of the overall model, with very small associations with deployment before separation and an elevation in the hazard of suicide for service members after separation from service, irrespective of deployment status. In terms of effect measure modification, the group of service members who deployed and then separated from service had a reversal in the direction of association for the Marine Corps compared with the other service branches. For the same exposure group in the Navy, we saw an increase in the hazard ratio. However, both of these estimates were based on small numbers of suicides.
Given that separation from service carried more evidence as a determinant for the hazard of suicide, we examined characteristics of service at the time of separation as possible explanatory factors. Both time in service and the characterization of service at separation from the military had associations with the hazard of suicide (Table 4). We observed a monotonic decrease in the rate of suicide associated with increased time in service. The hazard of suicide for service members who completed between 4 and 20 years of service was approximately half that for service members who only completed up to 1 year of service before separation. Among the service members who reverted to a civilian status at separation, we examined the characterization of the separation. Compared with service members who left with an honorable characterization, those with characterizations that were not honorable (which includes general/under honorable conditions, bad conduct, other than honorable conditions, and dishonorable) and those with uncharacterized separations (eg, separation initiated following <180 days of military service)25 had an increased hazard of suicide. We did not observe effect measure modification as a function of service branch for either time in service (likelihood ratio χ215 = 14.87, P = .46) or characterization of service (likelihood ratio χ29 = 12.14, P = .20). Estimation of all models reported above using December 31, 2007, as the end of the risk window did not alter the results reported herein.
To our knowledge, this is the first study designed to examine the association between deployment and suicide among all 3.9 million US military personnel who served during the first 6.25 years of OEF/OIF. These data showed that service members who deployed in support of OEF/OIF were not at increased risk of suicide compared with service members who never deployed in support of these conflicts. The association between deployment and suicide may be complex. Important factors to consider include differences by military service, combat exposure, number of deployments, and combat injuries (among others). We were able to examine some of these variables in the context of this study. None of the military services showed a significantly higher suicide rate among those who had an OEF/OIF deployment history compared with those who had not deployed. In addition, examining those with 1 deployment or 2 or more deployments separately did not reveal differential results compared with those with no deployments. Service members undergo predeployment health screenings, including mental health screening, to ensure that they are fit to deploy.26 Therefore, the cohort that deploys may be healthier than the general military population. Future research is needed to address other variables that may influence the complex relationship between deployment and suicide. Additional research with the OEF/OIF cohort may replicate findings from prior war eras that showed increased suicide risk among subgroups of veterans with risk factors such as having mental health conditions or being wounded in combat.27,28
These results are consistent with a recent prospective longitudinal study17 that included 83 suicides from the Millennium Cohort Study. They did not find increased suicide risk associated with any deployment-related factors such as combat experience, number of days deployed, or number of deployments.
The results differ in some ways from a recent study16 from the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS), although the methods and goals of the 2 studies were different. The Army STARRS reported an elevated suicide risk among currently and previously deployed soldiers. The discrepancy in results is likely due to methodological differences. The Army STARRS included only Army personnel, excluded Army National Guard and Army Reserve service members, included only suicides that occurred during military service, and included data on exposure and mortality from 2004 to 2009. Our study included all military services (Air Force, Army, Navy, and Marines Corps), included National Guard and reserve component service members, and used data on exposures from 2001 to 2007 and on mortality from 2001 to 2009. Analytically, the Cox proportional hazards regression models compared time to suicide, whereas the discrete time survival analysis used by the Army STARRS assessed risk of suicide based on the presence or absence of deployment.
The results of the present study showed that those who separated from military service were at increased risk of suicide compared with those who had not separated. Among those who had separated from service, both those who deployed and those who had not deployed showed similarly elevated risks for suicide. In contrast, in most analyses those who deployed but were not separated did not have higher rates than those who did not deploy and were not separated.
Consistent with research conducted with the Armed Forces in the United Kingdom,15 risk for suicide was highest among those who separated after shorter periods of military service. Compared with those with 4 or more years of military service, individuals with less than 4 years of service had an increased rate of suicide. There are several possible explanations for these findings. It is possible that the transition from the military to civilian life may have increased risk for suicide. Loss of a shared military identity, difficulty developing a new social support system, or unexpected difficulties finding meaningful work may contribute to a sense that the individuals do not belong or are a burden on others.29 It is possible that individual characteristics, experiences, or other factors that existed before military service increased suicide risk for this cohort.30 Alternatively, it is possible that military experiences that we are currently unable to identify in our data conferred risk for suicide. For some, it is possible that the factors that led to an early military discharge may continue to be problematic (eg, legal problems, mental health disorders, medical problems, disciplinary issues, and disability). For example, Iversen and colleagues31 found that veterans with mental health problems were more likely to be discharged early and were at elevated risk for unemployment following discharge. We found that service members discharged under not honorable conditions had modestly higher rates of suicide than those discharged under honorable conditions. The Army STARRS16 reported elevated suicide rates among soldiers who were demoted within the prior 2 years. Additional research is needed to clarify what the circumstances are surrounding early discharges and how these factors may be related to suicide. Protective factors may also differ for some who are discharged early. For example, individuals who separate with a dishonorable discharge are generally not entitled to VA services.
To avoid confusion, we did not use the term veteran to describe those who separated from military service in our cohort. Prior Veterans Health Administration (VHA) studies32- 34 typically define veterans as those individuals who are eligible for VA benefits, including some individuals who are still serving in the reserve component. Our findings are similar to prior research on US veteran populations from prior war eras. Such research indicated that, when veterans are considered as a whole, their rate of suicide was not significantly elevated compared with that of the general population.33 More recent research on VHA patients shows a 42% to 66% increase in suicide rates compared with control subjects,35,36 but these studies did not include the majority of veterans who do not access VHA services.9 Veterans who seek care from the VHA likely represent a vulnerable subpopulation. A strength of the present research is that it provided suicide risk estimates inclusive of the entire population of veterans within our study cohort. Additional research is needed to describe deployment risk among veterans eligible for VHA benefits.
The multifaceted and idiosyncratic nature of suicide renders suicide prevention a complex challenge. Taken as a whole, this study suggests that the deployed OEF/OIF cohort is not at increased suicide risk compared with the nondeployed OEF/OIF cohort members. We did not have access to data on combat exposure or mental health status. Because combat experiences and postdeployment mental health concerns vary widely among the deployed cohort, additional research is needed to inform related prevention efforts. Our findings indicate that preventionists should consider opportunities to intervene among those who separate from service early, especially those with less than 4 years of service.
Strengths of this study include a comprehensive analysis of suicide in relation to deployment among a very large, well-defined military cohort that served during OEF/OIF. Limitations include the use of death certificates to classify suicides, which may result in a misclassification bias.37 However, this is a widely used and accepted approach to classifying suicides.38,39 It is possible that suicides were underestimated because some service members with suicidal intentions may place themselves in harm’s way. In addition, some time-varying covariates (eg, marital status) used baseline values given a lack of temporal precision in changes over time. However, for marital status we were able to account for a change in status that was recorded before separation from service.
In summary, the accelerated rate of suicide among members of the US Armed Forces and veterans in recent years is concerning. Although there has been speculation that deployment to the OEF/OIF combat theaters may be associated with military suicides, the results of this research do not support that hypothesis. Future research is needed to examine combat injuries, mental health, and other factors that may increase suicide risk. It is possible that such factors alone and in combination with deployment increase suicide risk.
Submitted for Publication: July 24, 2014; final revision received November 12, 2014; accepted December 16, 2014.
Corresponding Author: Mark A. Reger, PhD, National Center for Telehealth and Technology, Joint Base Lewis-McChord, 9933 W Hayes St, Tacoma, WA 98431 (firstname.lastname@example.org).
Published Online: April 1, 2015. doi:10.1001/jamapsychiatry.2014.3195.
Author Contributions: Drs Reger and Smolenski had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Reger, Smolenski, Skopp, Metzger-Abamukang, Kang, Bullman, Gahm.
Acquisition, analysis, or interpretation of data: Reger, Smolenski, Skopp, Perdue.
Drafting of the manuscript: Reger, Smolenski, Skopp, Metzger-Abamukang, Perdue.
Critical revision of the manuscript for important intellectual content: Reger, Smolenski, Kang, Bullman, Perdue, Gahm.
Statistical analysis: Smolenski, Kang, Bullman, Perdue.
Obtained funding: Reger, Skopp, Kang, Gahm.
Administrative, technical, or material support: Skopp, Metzger-Abamukang, Gahm.
Study supervision: Reger, Skopp, Gahm.
Conflict of Interest Disclosures: None reported.
Funding/Support: This research was supported by grant W81XWH-08-MOMRP-SPCR from the US Army Medical Research and Materiel Command Military Operational Medicine Research Program–Suicide Prevention and Counseling Research.
Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The views expressed are those of the authors and do not reflect the official policy or position of the Department of Defense or the US Government.
Additional Contributions: Tim Lineberry, MD (Telemedicine and Advanced Technology Research Center), Paul Amoroso, MD, MPH (MultiCare Institute for Research & Innovation), and Greg Reger, PhD (VA Puget Sound Health Care System) critically reviewed the preliminary results. Elissa Thomas, LPN, CCRC, assisted in managing this project, Daniel Schwesinger aggregated and managed enterprise data sources, and David Luxton, PhD, and Nigel Bush, PhD, assisted in conceptualizing the project (all affiliated with the National Center for Telehealth and Technology). The Armed Forces Medical Examiner System provided data for enterprise death verification. Ms Thomas and Mr Schwesinger received compensation for their contributions.