Veterans and Suicide: A Reexamination of the National Death Index–Linked National Health Interview Survey

Abstract

Objectives. We assessed the risk of suicide among veterans compared with nonveterans.

Methods. Cox proportional hazards models estimated the relative risk of suicide, by self-reported veteran status, among 500 822 adult male participants in the National Death Index (NDI)–linked National Health Interview Survey (NHIS), a nationally representative cohort study.

Results. A total of 482 male veterans died by suicide during 1 837 886 person-years of follow-up (76% by firearm); 835 male nonveterans died by suicide during 4 438 515 person-years of follow-up (62% by firearm). Crude suicide rates for veterans and nonveterans were, respectively, 26.2 and 18.8 per 100 000 person-years. The risk of suicide was not significantly higher among veterans, compared with nonveterans, after adjustment for differences in age, race, and survey year (hazard ratio = 1.11; 95% confidence interval = 0.96, 1.29).

Conclusions. Consistent with most studies of suicide risk among veterans of conflicts before Operation Iraqi Freedom/Operation Enduring Freedom, but in contrast to a previous study using the NDI-linked NHIS data, we found that male veterans responding to the NHIS were modestly, but not significantly, at higher risk for suicide compared with male nonveterans.

In 2008, then Secretary of the Department of Veterans Affairs, James Peake, established the Blue Ribbon Working Group on Suicide Prevention to assess suicide risk among the veteran population.¹ The group issued 8 key findings, the first of which was that the literature was contradictory regarding whether veterans were at higher risk for suicide compared with nonveterans.

The contradiction was traced to a single survey-based cohort study, published in 2007 by Kaplan et al,² which found a greater than 2-fold increased risk of suicide among veterans compared with nonveterans. By contrast, the other studies^3–6 found that suicide risk among veterans as a whole was not higher than that among age-, gender-, and race-matched members of the general population.

Kaplan et al.’s study² differed in 2 important ways from the other work cited by the Working Group.^3–6 First, Kaplan et al. ² relied on self-report to assess veteran status, whereas the other studies ascertained veteran status from databases maintained by the Department of Defense. Second, relative risk estimates in Kaplan et al.² were based on direct comparisons of suicide incidence among veterans to incidence among nonveterans, whereas other studies used standardized mortality ratios (SMRs) that tended to bias veteran risk toward the null because veterans and nonveterans were both included in the comparison group. Two additional studies were published after the Working Group released its findings. The first, a military cohort study,⁷ assessed suicide risk among veterans who served in Operation Iraqi Freedom/Operation Enduring Freedom (OIF/OEF) and who separated from the military before 2006 (i.e., before the unprecedented increase in suicide incidence among soldiers).⁸ The second study ⁹ tracked mortality among male respondents in the Cancer Prevention Study. Both studies failed to find evidence of differential suicide risk attributable to veteran status.

Our study reexamines the question of whether veterans were at increased risk for suicide using public data from the National Health Interview Survey (NHIS; 1986–2000) that have been linked to the National Death Index (NDI) through 2006—the same data source used in the study by Kaplan et al.,² now available for several additional years of baseline interviews and mortality follow-up. Like the previous NHIS–NDI study, the present study was limited to pre-OIF/OEF veterans. Because access to firearms is a risk factor for suicide,^10,11 and because veterans are more likely to own firearms than are nonveterans,¹² the present study examined not only the relation between veteran status and overall suicide risk, as did the original NHIS–NDI study, but also whether any such risk was differentially related to firearm versus nonfirearm suicide.

Because our primary findings were discrepant with those reported by Kaplan et al.,² we conducted sensitivity analyses that restricted data to the survey years (1986–1994) and mortality follow-up (through 1997) used in the study by Kaplan et al.² Covariates included in these comparative analyses were, by design, the same as those used by Kaplan et al.²

METHODS

The study population consisted of 500 822 men, aged 18 years and older, who participated in the NHIS between 1986 and 2000. The NHIS, conducted through the National Center for Health Statistics (NCHS), has been collecting health information on a representative sample of noninstitutionalized civilians in the United States since 1957. Sampling and interviewing are continuous throughout each year. The NHIS uses a multistage probability sampling design that selects primary sampling units from approximately 1900 geographic areas (a county, a small group of contiguous counties, or a metropolitan statistical area) and then samples households within each unit. Face-to-face household interviews are conducted with members of selected households, with response rates of over 90% each year.¹³

The NCHS linked NHIS data with the NDI by matching participants older than 18 years from the date of interview through December 2006 using 12 weighted criteria: social security number, first and last names, middle initial, race, gender, marital status, birth date (day, month, and year), and state of birth and residence.¹⁴

The linked NHIS–NDI file was used to determine the main outcome of the study: death by suicide. Secondary analyses examined the relation between veteran status and suicide by method.

Respondents were identified as veterans if they answered affirmatively to the questions: “Did you ever serve on active duty in the Armed Forces of the United States?” (NHIS 1986–1996) or “Have you ever been honorably discharged from active duty in the US Army, Navy, Air Force, Marine Corp, or Coast Guard?” (NHIS 1997–2000). Women were excluded from the analytic sample because so few female respondents identified themselves as veterans.

Covariates from baseline interviews that were used in the previous NHIS–NDI study were included in sensitivity analyses in the present study to allow for direct comparison with previous findings. These include age group, race, marital status, living arrangement, education, employment status, region of residence, self-rated health status, body mass index (BMI; defined as weight in kilograms divided by the square of height in meters), family income, and interval since last doctors visit

SUDAAN statistical software program (version 9.0; Research Triangle Park, NC) was used to perform weighted analyses that took into account the multistage sample design employed by the NHIS in the collection of these data. Weighted Cox proportional hazards models were used to estimate the relative hazard of suicide among veterans compared with nonveterans. Primary models were adjusted for age, race, and survey year (dummy variables each covering 3 calendar years). Respondents were censored at death by any cause; survivors were censored at the end of the study window, December 31, 2006. Time to death was measured from the midpoint of the quarter (i.e., 3-month period) of the interview to the exact date of death.

RESULTS

Veterans made up 30% of the sample, were considerably older (72% were aged 45 years and older vs 29% of nonveterans), and more likely to be White (90% vs 83%). Other baseline differences between veterans and nonveterans were greatly attenuated within age-strata (Table 1).

TABLE 1—

Baseline Characteristics Among Male Respondents: National Health Interview Survey, 1986–2000

			Aged 18–44 Yearsa		Aged 45–64 Yearsa		Aged ≥ 65 Yearsa
	Nonveteran (n = 351 689), Weighted %	Veteran (n = 149 133), Weighted %	Nonveterans (n = 245 025; Mean [SD] Age = 30.6 [7.5]), Weighted %	Veterans (n = 40 978; Mean [SD] Age = 35.8 [6.5]), Weighted %	Nonveterans (n = 73 847; Mean [SD] Age = 52.5 [5.6]), Weighted %	Veterans (n = 66 285; Mean [SD] Age = 54.6 [5.8]), Weighted %	Nonveterans (n = 32 817; Mean [SD] Age = 75.2 [6.8]), Weighted %	Veterans (n = 41 870; Mean [SD] Age = 71.6 [5.3]), Weighted %
Age			70.9	27.7	20.4	44.1	8.7	28.2
Race
White	83.2	89.5	83.0	84.0	82.7	91.0	86.0	92.4
Non-White	16.8	10.5	17.1	16.0	17.3	9.0	14.0	7.6
Marital status
Married	62.0	78.4	55.3	71.3	80.3	82.2	73.8	79.5
Widowed/divorced /separated	8.7	13.6	6.1	11.8	12.1	12.9	21.5	16.6
Never married or unknown	29.4	8.0	38.7	16.9	7.6	4.9	4.7	3.9
Education
< 12 y	21.7	16.6	16.9	7.2	26.0	14.1	50.5	29.7
≥ 12 y	77.5	82.9	82.5	92.4	72.9	85.4	47.8	69.5
Unknown	0.8	0.6	0.7	0.3	1.1	0.5	1.7	0.8
Employment status
Has job	78.7	64.5	85.9	90.3	80.5	77.2	15.8	19.3
Looking for work	2.9	1.8	3.5	3.4	1.8	1.7	0.2	0.3
Has no job	18.3	33.8	10.5	6.3	17.6	21.2	84.0	80.4
Unknown	0.10	0.02	0.11	0.02	0.09	0.03	0.03	0.01
Region
Northeast	20.6	19.6	20.1	16.8	21.6	19.9	21.8	21.8
Midwest	24.6	24.6	24.7	25.0	24.0	24.9	25.5	23.8
South	34.3	35.2	34.0	36.7	35.2	34.5	34.7	34.6
West	20.6	20.7	21.3	21.5	19.3	20.7	18.0	19.8
Place of residence population
≥ 1 000 000	40.7	35.9	41.4	36.2	40.4	36.3	36.0	34.8
250 000–999 999	23.1	25.5	23.4	26.6	22.5	25.3	22.5	24.9
≤ 250 000	36.1	38.6	35.2	37.2	37.1	38.4	41.5	40.3
Health status
Poor	9.6	15.2	5.0	6.2	16.3	14.4	31.5	25.3
Good	90.1	84.5	94.7	93.5	83.4	85.3	68.1	74.3
Unknown	0.3	0.3	0.3	0.3	0.3	0.3	0.4	0.3
Activity limitations
Not limited	86.7	77.7	91.7	88.4	80.3	79.1	61.2	65.1
Limited	13.3	22.3	8.3	11.6	19.7	20.9	38.8	34.9
BMI category
Underweight	4.2	2.6	4.4	2.6	2.6	1.9	5.8	3.7
Normal	34.0	30.7	36.2	34.6	25.7	27.7	34.8	31.5
Overweight	31.9	39.0	30.9	38.7	34.7	41.3	33.9	35.6
Obese	14.0	15.0	13.3	14.7	17.4	16.7	12.0	12.6
Missing	15.9	12.7	15.1	9.5	19.6	12.3	13.5	16.6
Family income, $
< 20 000	26.2	22.5	24.8	21.4	20.4	15.6	50.7	34.5
≥20 000	70.4	74.0	72.2	76.5	75.5	81.1	43.9	60.3
Unknown	3.4	3.5	3.0	2.1	4.1	3.3	5.4	5.3
Doctor visits past 12 mo, no.
< 10	92.4	88.3	94.6	92.9	89.4	89.2	81.4	82.4
≥ 10	7.3	11.3	5.1	6.8	10.2	10.4	18.0	17.0
Unknown	0.3	0.4	0.3	0.3	0.4	0.4	0.6	0.6

Note. BMI = body mass index (defined as weight in kilograms divided by the square of height in meters).

^aTo facilitate comparisons of baseline characteristics by veteran status, descriptive statistics are provided not only for veterans as a whole, but also by age groups because veterans during the study period were, in general, much older than nonveterans.

Suicide Rates by Veteran Status

A total of 482 male veteran participants died by suicide during 1 837 886 person-years of follow-up (364 by firearms [76%] and 118 by other methods), and 835 male nonveterans died by suicide during 4 438 515 person-years of follow-up (519 by firearms [62%] and 316 by nonfirearm methods). Crude suicide rates for male veterans and nonveterans were 26.2 and 18.8 per 100 000 person-years, respectively. The risk of suicide was not significantly higher among veterans compared with nonveterans, after adjustment for differences in age, race, and survey year (hazard ratio [HR] = 1.11; 95% confidence interval [CI] = 0.96, 1.29; Table 2).

TABLE 2—

Rates and Adjusted Hazard Ratios for Overall Suicide, Firearm Suicide, and Nonfirearm Suicide Among Male Respondents, According to Veteran Status: National Health Interview Survey, 1986–2000

	All Suicides		Firearm Suicides		Other Suicides
	Veteran	Nonveteran (Ref)	Veteran	Nonveteran (Ref)	Veteran	Nonveteran (Ref)
Suicide rate/100 000 person-y	26.23	18.81	19.81	11.69	6.42	7.12
Crude HR (95% CI)	1.36 (1.19, 1.54)	1.00	1.63 (1.41,1.90)	1.00	0.89 (0.71, 1.13)	1.00
Adjusted HRa (95% CI)	1.11 (0.96, 1.29)	1.00	1.19 (1.01, 1.40)	1.00	0.97 (0.73, 1.29)	1.00

Note. CI = confidence interval; HR = hazard ratio. A total of 482 male veteran participants died by suicide during 1 837 886 person-years of follow-up (364 by firearm and 118 by other methods); 835 male nonveteran died by suicide during 4 438 515 person-years of follow-up (519 by firearm and 316 by nonfirearm methods).

^aAdjusted for age (as a continuous variable), race, and survey year.

The rate of firearm suicides was significantly higher among veterans (19.8/100 000 person-years) than among nonveterans (11.7/100 000 person-years); increased risk of firearm suicide persisted after adjusting for age, race, and survey year (HR = 1.19; 95% CI = 1.01, 1.40). Veterans and nonveterans did not differ significantly in their risk of suicide by nonfirearm methods.

Sensitivity Analyses

Sensitivity analyses restricted to the survey and follow-up years used in the original NHIS–NDI study² produced HRs similar to those obtained using the longer study period (Table 3). In the restricted study period analysis, there were 209 suicides among 106 166 male veterans and 298 suicides among 220 993 nonveteran men. The age- and race-adjusted suicide risk among male veterans over this abbreviated study period was modestly, but not significantly, higher among veterans than among nonveterans (HR = 1.21; 95% CI = 0.97, 1.50). Models that included all the covariates used in the previous NHIS study (“extensively” adjusted models) showed a significantly elevated suicide risk among veterans compared with nonveterans (HR = 1.33; 95% CI = 1.03, 1.71). When age was entered into the extensively adjusted model as a continuous variable, rather than as the categorical age groupings used by Kaplan et al.,² the HR and significance associated with veteran status were attenuated further (HR = 1.23; 95% CI = 0.96, 1.58). Additional sensitivity analyses that excluded persons with missing values for baseline covariates produced results virtually identical to those presented in Table 1.

TABLE 3—

Suicide Rates and Hazard Ratios Among Male Respondents, According to Veteran Status: National Health Interview Survey, 1986–1994

			Firearm Suicides, Unweighted %		Person-Years of Follow-Up		Suicide Ratea			HR (95% CI) Adjusted for Age (Continuous) and Race	Multivariable Adjusted HR (95% CI)b
	Veterans, No.	Nonveterans, No.	Veterans	Nonveterans	Veterans	Nonveterans	Veterans	Nonveterans	Crude HR (95% CI)
Current reanalysis	106 166	220 993	209 (77%)	298 (59%)	735 744	1 542 018	28.4	19.3	1.50 (1.23,1.84)	1.21 (0.97, 1.50)	1.33 (1.03, 1.71)
Kaplan et al.2	104 026	216 864	197 (77%)	311 (59%)	NAc	NAc	NAc	NAc	NAc	NAc	2.13 (1.14, 3.99)

Note. CI = confidence interval; HR = hazard ratio; NA = not available. Study period restricted to years analyzed in the original study by Kaplan et al.² (i.e., deaths through December 31, 1997). Multivariate models that used all 11 covariates in the study by Kaplan et al.,² but characterized age as a continuous variable, produced an attenuated HR of 1.23 (95% CI = 0.96, 1.58).

^aSuicide rate was suicides per 100 000 person-years.

^bMultivariate analysis included the covariates as described in the original study by Kaplan et al.²: age at the time of the interview (18–44, 45–64, and ≥ 65 years), race (White or non-White), marital status (married; widowed, separated or divorced; never married), living arrangement (alone or with others), education (< 12 years, ≥ 12 years, or unknown), employment status (has job, looking for work, has no job, unknown), region of residence (Northeast, Midwest, South, or West), self-rated health status (good—defined as excellent, very good or good; poor—defined as fair or poor; or unknown), body mass index (underweight, normal, overweight, obese, or missing), family income (< $20 000, ≥ $20 000, or unknown), and interval since last doctors visit.

^cData omitted because Kaplan et al.² did not provide person-years of follow-up or crude suicide rates by veteran status.

DISCUSSION

Consistent with retrospective cohort studies that used standardized mortality ratios to measure suicide risk among veterans relative to the general population,^5,7,15–17 and with 1 of 2 previous survey-based cohort studies of suicide risk among veterans,⁹ we found that the risk of suicide among male veterans was modestly, but not significantly, higher than that among age-, race-, and period-matched male nonveterans (HR = 1.11; 95% CI = 0.96, 1.29). By contrast, our results were at odds with those reported by Kaplan et al.,² who, using the same underlying data source as the present study, found a 2-fold increased risk of suicide among male veterans compared with male nonveterans (adjusted HR = 2.13; 95% CI = 1.14, 3.99).

Sensitivity analyses restricted to the same years of survey administration and mortality follow-up as in the study by Kaplan et al.² produced relative risk estimates similar to those we observed over our entire study period—but considerably more modest than the 2-fold risk ratio reported by Kaplan et al.² Our attempts to replicate the study by Kaplan et al.² were unsuccessful, despite baseline data that appeared very similar (e.g., total male suicides 508 vs 507; veterans constituted 15.7% of all respondents; and 32% of all male respondents in our analysis vs 15.7% of all respondents in the study by Kaplan et al.²). Attempts to replicate the findings of Kaplan et al.² produced 95% CIs with standard errors much smaller than those reported in that study and, importantly, did not include the point estimate they reported. Although Kaplan et al.² analyzed the restricted-use mortality data, whereas we used the public-use file, it was unclear why the 2 files yielded such contrasting findings.

Salient differences between our primary analyses and those in the previous NHIS–NDI study were that the present study covered more person-years of follow-up, observed more than twice as many suicides (1317 vs 508), adjusted for period effects by including indicator variables, and adjusted for age as a continuous variable (the previous study controlled for age using 3 broad age groupings). In addition, whereas we reported both crude suicide rates and age- and race-adjusted suicide risk among veterans and nonveterans, Kaplan et al.² reported only multivariable HRs that adjusted for several characteristics ascertained years after separation from the military. Consequently, although we could report that (1) the crude suicide rate among our veterans (26.2/100 000 person-years) was similar to that reported in previously cited retrospective military cohort studies,^3–6 and (2) the age- and race-adjusted suicide risk was not significantly higher among veterans compared with nonveterans, no such comparison could be ascertained from data published by Kaplan et al.²

Although veterans in our study were not at significantly elevated risk for suicide, they were significantly more likely to die by suicide involving firearms. Male veterans are more likely than are male nonveterans to live in homes with firearms,¹² and the presence of firearms in the home is a well-recognized risk factor for suicide.^{10,11,18–24} As such, suicide rates among veterans, although not significantly higher than those among male nonveterans, might be higher than they would be if veterans were no more likely than nonveterans to live in homes with firearms. This counterfactual possibility was consistent with all studies of current, active duty military personnel before the current conflict in Iraq and Afghanistan, which consistently found evidence of a “healthy warrior effect” for suicide among military personnel (i.e., markedly lower suicide rates compared with the age-, race- and gender-matched civilian population).^25–34

Our findings should be interpreted with several limitations in mind. First, we did not have information about deployment during wartime or combat-related injuries, aspects of exposure that might be relevant to suicide risk.^4,7 Second, as in other cohort studies, suicides were identified on the basis of death certificates. Although death certificate data are believed to underreport suicide, death certificates have nevertheless been widely accepted as a valid mortality source for epidemiological research on suicide.^35,36 Third, all veterans in our study were men, most were identified at a time remote from military service, and all had separated from the military before the unprecedented rise in suicide rates among soldiers of the OIF/OEF era. Our findings did not, therefore, address the risk of suicide among women veterans, veterans of the recent conflicts in Iraq and Afghanistan, or veterans recently separated from military service—a period that some,³ but not all,^7,15 previous studies suggested might be one of heightened suicide risk.

Despite these limitations, and in view of consistent evidence of a “healthy warrior” effect for suicide risk among military personnel of every conflict before,^25–34 but not including, the recent conflict in Iraq and Afghanistan,⁸ findings from the present study underscore the need for population-based suicide prevention policies that focus on (1) cohorts of veterans likely to be at greatest risk, such as those recently returning from Iraq and Afghanistan, and (2) suicide risk factors observed commonly among veterans, such as combat-related injuries and easy access to firearms, rather than on the notion that veteran status per se is a risk factor for suicide.