NVDRS provides details on suicide incidents such as decedent characteristics, the mechanisms/weapons involved, and the precipitating circumstances (CDC, 2008). From 2005 to 2010, 18 U.S. states participated in NVDRS; therefore, case inclusion for this study was limited to incidents in those states. From 2005 to 2009, NVDRS included 16 states (Alaska, Colorado, Georgia, Kentucky, Maryland, Massachusetts, New Jersey, New Mexico, North Carolina, Oklahoma, Oregon, Rhode Island, South Carolina, Utah, Virginia, and Wisconsin). In 2010, Ohio and Michigan were added. California also collected data from 2005 to 2008; however, these data were excluded because data collection was not ongoing at the time of the study.

Data sources for NVDRS include law enforcement reports, coroner/medical examiner reports, toxicology reports, and death certificates. All sources are linked by incident into a single repository. States manage data collection through state health departments or subcontracted entities. Data are gathered and coded by trained abstractors (CDC, 2008).

NVDRS details on precipitating circumstances of death come from law enforcement and coroner/medical examiner death scene investigation reports. These reported circumstances were used by investigators to help them determine and justify a manner of death at a violent death scene (e.g., homicide, suicide, etc.). The process that law enforcement and coroner/medical examiner investigators use to gather information on precipitating circumstances of a violent death involve open-ended interviews with family members, friends, and others associated with the decedent as well as witnesses to the death incident to assess what these informants believed led to the death (U.S. Department of Justice, 1999). Investigators corroborate statements, identify commonalities, and compare testimonies to other physical evidence such as a decedent confessional in a suicide note to better understand the motive of the death and before drawing opinions about the manner of death. Law enforcement and coroner/medical examiner investigators offer different perspectives than researchers; however, their incident reports or their conclusions on manner of death are regularly used for research (Logan, Hall, & Karch, 2011; Logan et al., 2008).

The DoD's suicide surveillance program (and the Army's similar program that pre-dated the DoDSER program) provides many details on military suicide decedents that are not captured in NVDRS, such as military background, duty status, combat exposure, and military disciplinary history (Gahm et al., 2012). DoDSER submissions are required for all military suicides within 60 days of notification of a confirmed suicide by the Armed Forces Medical Examiner System. In the Army, DoDSERs are completed by behavioral health providers who are required to submit detailed information through a secure web portal. DoDSER data collection requires medical and mental health records, personnel records, responsible investigative agency records, records related to manner of death, and interviews (e.g., coworkers, supervisors, friends, family members, the responsible investigative agency officer, and other professionals such as physicians, behavioral health clinicians, chaplains, and military police officers), as appropriate (Gahm et al., 2012).

We linked DoDSER data to NVDRS data to identify active duty U.S. Army decedents in NVDRS. Data linkage was initiated with DoDSER incident records. There were 154 active duty Army decedents who died in NVDRS states and therefore 154 DoDSER incident records; multivictim suicide incidents (e.g., suicide pacts) were not included. We linked the appropriate NVDRS record to each DoDSER record using a multistep process that compared common variables in both systems because all record were deidentified. For each DoDSER record, we first identified all NVDRS records that had the same manner, location (state), and date of death as well as decedent age and sex. Then, among the pool of selected NVDRS records, we selected the record that had the same decedent race/ethnicity and marital status, had documentation of the decedent “ever serving in the military,” and had documentation that the decedent had a current Army occupation (this documentation is provided as an open text field in NVDRS that lists terms like “soldier,” “Army Sgt”). If we selected an NVDRS record that did not match to its respective DoDSER record on all variables in the second step because of an unknown value, or because of a minor discrepancy, then we additionally reviewed the weapon information and the investigator narratives from both NVDRS and DoDSER records to ensure the stories of the suicide event captured by the two systems corroborated. Each DoDSER record only linked to one NVDRS record. In total, we linked NVDRS-DoDSER data for 141 (92%) Army suicide decedents. No personal identifying information was reviewed or appended.

Each year, NVDRS collects details on over 10,000 civilian suicide incidents. To make a more manageable comparison between the Army versus civilian suicide decedents, we matched three to four NVDRS civilian suicide decedents to every Army decedent based on sex and age (within 1 year), state of death, and year of death to control for potential age, state, and sex specific social and environmental influences. We also considered matching by marital status and race/ethnicity, but small civilian case counts among some of the groups precluded ascertainment of three to four matches. Instead, marital status and race/ethnicity were used as control variables in the comparative analyses. Details on the process for matching civilian suicide decedents to Army decedents are provided in the online version of this article (please see link in the “Supporting Information” section). A total of 563 civilian suicide decedents were included.

DoDSER data were used to describe the Army suicide decedent group with respect to component (Regular, Reserve, National Guard), duty status (active duty, activated Guard/Reserve, active duty for training), pay grade, recent military-related stresses, deployment history, and combat exposure.

NVDRS data were used to describe and compare the Army versus civilian suicide decedents with respect to demographic characteristics (age and sex [to verify matching], race/ethnicity, and marital status), place and method of death, as well as the precipitating factors as perceived by law enforcement agents, medical examiners, and their informants. The scope of decedent precipitating factors included the following: current/recent depressed mood or mental health problem; alcohol intoxication or suspected intoxication at the time of death; other substance abuse problems; a precipitating physical health problem (e.g., pain-related problem); a recent crisis (within 2 weeks of death); criminal or civil legal problems; job problems; financial problems; intimate partner problems; and/or other relationship problems. These factors have also been cited elsewhere as being known risk factors for suicide (Allen, Cross, & Swanner, 2005; Boscarino, 2006; Kung, Pearson, & Liu, 2003; Mahon, Tobin, Cusack, Kelleher, & Malone, 2005; Miller, Mogun, Azrael, Hempstead, & Solomon, 2008; Moscicki, 1995; Skopp et al., 2012; Thoresen & Mehlum, 2006). Additionally, we made comparisons based on other relevant preceding circumstances (e.g., prior suicide attempts, current mental health treatment, signs of premeditation). Definitions of all precipitating and other preceding circumstances of suicide in NVDRS are provided in Table1 (CDC, 2008).

We first described the military background of the U.S. Army decedents with the DoDSER data to obtain a better sense of this study group (please note that this study was unable to assess military-related risk factors). We then describe both the U.S. Army and the civilian suicide decedent groups with respect to demographic and incident characteristics as well the precipitating and preceding circumstances (i.e., provide the prevalence of each characteristics among both groups).

We compared the Army and civilian suicide decedent groups using conditional logistic regression. Simple conditional logistic regression models were used to make comparisons by demographic factors. Multivariable conditional logistic regression models were used to make comparisons with respect to location and method of death, the precipitating circumstances of death, and the other preceding circumstances. Matched prevalence odds ratios were used to assess associations. Odds ratios with additional adjustments for race/ethnicity and marital status were also provided. Comparisons made with respect to the precipitating circumstances accounted for all circumstance variables. Differences were considered significant at the 0.05 level.

Most Army decedents (86%) were in the active component (Regular) Army and 88% were in the enlisted ranks of E1 (Private) to E7 (Sergeant First Class) (Table2). A fifth of the Army decedents were identified as having recent military-related stresses. The most common stresses involved article 15 proceedings (nonjudicial proceedings for minor offenses administered by commanding officers), absent-without-leave proceedings, and/or courts-martial proceedings. An estimated 46% of Army decedents had at least one deployment and 19% had experienced combat. At least 19 soldiers were believed to have killed others, witnessed killing, or to have engaged in combat that resulted in casualties or wounded enemies.

The demographic and incident characteristics of the decedents are shown in Table3. The Army and civilian decedent groups were matched by age and sex, and therefore, the two groups did not differ by these characteristics as expected. The majority (88%) of all decedents were aged 18 to 39 years (because of the matching criteria, some civilian decedents were 17 years of age). Most decedents (96%) were males. Also, both groups did not significantly differ with respect to race/ethnicity. Most Army decedents (72%) and civilian decedents (76%) were of non-Hispanic White race/ethnicity.

In contrast, there were clear differences between the two groups with respect to marital status. Over half (58%) of the Army decedents versus only 22% of the civilian decedents were married (p < .05). Furthermore, 64% of the civilian decedents versus 33% of Army decedents were never married (p < .05).

The majority of Army (70%) and civilian (75%) decedents died in personal residences, followed by recreational/commercial/natural areas (10% Army and 12% civilian), and transport areas (9% of each group). A higher proportion of Army versus civilian decedents died in other locations. The most common mechanism of death for both groups was firearm use (63% Army and 56% civilian). Hanging was the second most common mechanism for both groups (18% Army and 29% civilian), but this mechanism was more prevalent among civilian decedents (p < .05).

We did not find any significant between-group differences with respect to precipitating health- and stress-related circumstances based on the adjusted models. The five most common precipitators of suicide for all decedents, based on the investigator reports, were having mental health-related problems (53% Army and 64% civilian), intimate partner problems (54% Army and 42% civilian), alcohol or substance abuse problems (41% Army and 48% civilian), recent crises (37% Army and 33% civilian), and job problems (18% Army and 16% civilian) (see Table4). Also, an estimated 27% of both decedent groups left suicide notes, which often indicates premeditation.

We initially found that a higher proportion of Army versus civilian decedents had precipitating intimate partner problems based on the crude matched odds ratio of 1.62 (p < .05); however, this difference was not significant after accounting for both race/ethnicity and marital status as well as marital status alone (data not shown). Furthermore, there were no differences in the prevalence of intimate partner problems between the two decedent groups in each marital status category (data not shown).

A smaller proportion of Army versus civilian decedents was also initially found to have a known current depressed mood and/or mental health problem; however, this difference was not significant after accounting for race/ethnicity and marital status. Furthermore, we did not find significant differences between the groups with respect to the proportion of decedents who were currently receiving mental health treatment; an estimated 27% of Army and 28% of civilian decedents were receiving mental health treatment near the time of death.

With respect to prior suicidal ideation and behavior, we found a smaller proportion of Army versus civilian decedents had prior suicidal attempts and a smaller proportion of Army versus civilian decedents disclosed suicidal intentions to others (Table4). These differences remained significant across all adjusted regression models (all odds ratios were significant at the 0.05 level).