Prospective Associations of Military Discharge Characterization with Post-active Duty Suicide Attempts and Homelessness: Results from the Study to Assess Risk and Resilience in Servicemembers—Longitudinal Study (STARRS-LS)

ABSTRACT

Introduction

Active duty service members transitioning to civilian life can experience significant readjustment stressors. Over the past two decades of the United States’ longest sustained conflict, reducing transitioning veterans’ suicidal behavior and homelessness became national priorities. However, it remains a significant challenge to identify which service members are at greatest risk of these post-active duty outcomes. Discharge characterization, which indicates the quality of an individual’s military service and affects eligibility for benefits and services at the Department of Veterans Affairs, is a potentially important indicator of risk.

Materials and Methods

This study used data from two self-report panel surveys of the Study to Assess Risk and Resilience in Servicemembers-Longitudinal Study (STARRS-LS) (LS1: 2016-2018, n = 14,508; and LS2: 2018-2019, n = 12,156), which were administered to respondents who previously participated while on active duty in one of the three Army STARRS baseline self-report surveys (2011-2014): the New Soldier Study (NSS), a survey of soldiers entering basic training; All Army Study, a survey of active duty soldiers around the world; and the Pre-Post Deployment Study, a survey of soldiers before and after combat deployment. Human Subjects Committees of the participating institutions approved all recruitment, informed consent, and data collection protocols. We used modified Poisson regression models to prospectively examine the association of discharge characterization (honorable, general, “bad paper” [other than honorable, bad conduct, dishonorable], and uncharacterized [due to separation within the first 180 days of service]) with suicide attempt (subsample of n = 4334 observations) and homelessness (subsample of n = 6837 observations) among those no longer on active duty (i.e., separated or deactivated). Analyses controlled for other suicide attempt and homelessness risk factors using standardized risk indices that were previously developed using the LS survey data.

Results

Twelve-month prevalence rates of self-reported suicide attempts and homelessness in the total pooled LS sample were 1.0% and 2.9%, respectively. While not associated with suicide attempt risk, discharge characterization was associated with homelessness after controlling for other risk factors. Compared to soldiers with an honorable discharge, those with a bad paper discharge had an increased risk of homelessness in the total sample (relative risk [RR] = 4.4 [95% CI = 2.3-8.4]), as well as within subsamples defined by which baseline survey respondents completed (NSS vs. All Army Study/Pre-Post Deployment Study), whether respondents had been separated (vs. deactivated), and how much time had elapsed since respondents were last on active duty.

Conclusions

There is a robust association between receiving a bad paper discharge and post-separation/deactivation homelessness. Policies that enhance transition assistance and access to mental healthcare for high-risk soldiers may aid in reducing post-separation/deactivation homelessness among those who do not receive an honorable discharge.

For many of the 200,000 service members who transition out of the military each year,¹ readjustment to civilian life is associated with a range of stressors, from a shift in identity and loss of social support networks to financial difficulties, unemployment, and housing instability.^1–7 Of particular concern for those making this transition are suicidal behavior and homelessness, which emerged as national priorities during the previous two decades of war.^8–10 Suicide risk during the military-to-civilian transition is significantly increased,^11–13 and veterans continue to be at greater risk of homelessness than other U.S. adults.¹⁴ Unfortunately, it remains a significant challenge to efficiently and effectively identify service members at risk of post-active duty suicidal behavior or homelessness for targeted preventive interventions.

One potentially valuable source of information is the “character of service” designation issued to all service members upon separation (discharge or release from active duty) or, for members of the Reserve Components, deactivation (return to reserve status after a period of active duty service). In addition to providing an indication of service members’ behavior and performance while in the military, discharge characterization affects eligibility for Department of Veterans Affairs (VA) benefits and services, such as disability compensation, education assistance, home loans, and healthcare.¹⁵ Lack of access to benefits and healthcare services may elevate risk of negative outcomes for those transitioning back to civilian life. Indeed, previous research suggests that former service members who did not receive an honorable discharge are at increased risk for psychopathology (e.g., posttraumatic stress disorder and depression), homelessness, and suicidal thoughts and behaviors.^12,16–20

In this report, we use longitudinal survey data from the Study to Assess Risk and Resilience in Servicemembers-Longitudinal Study (STARRS-LS) to prospectively examine the associations of discharge characterization with suicide attempt and homelessness following separation/deactivation from active duty U.S. Army service. Unlike prior studies, we control for the predicted probability of suicide attempt and homelessness based on two previously developed risk indices (described below)^21,22 to examine whether discharge characterization can aid in identifying soldiers at risk of post-separation/deactivation suicide attempt or homelessness beyond other known risk factors for those outcomes.

METHOD

Sample and Procedures

Baseline surveys

STARRS-LS is an epidemiological-neurobiological study designed to identify risk and protective factors for suicidal behaviors among U.S. Army soldiers.²³ As described in more detail elsewhere,^23–26 there were three baseline STARRS surveys: (1) the 2011-2012 New Soldier Study (NSS) survey, a cross-sectional survey of n = 38,733 soldiers that included members of the Army Reserve and Army National Guard and was administered when new soldiers reported for Basic Combat Training; (2) the 2011-2013 All Army Study (AAS) survey, a cross-sectional survey of n = 25,088 active duty soldiers serving throughout the world, including in combat deployments in Afghanistan; and (3) the 2012-2014 Pre-Post Deployment Study (PPDS) survey, a four-wave panel survey of n = 8566 soldiers in Brigade Combat Teams deployed to Afghanistan, with a baseline survey administered 2-3 weeks before deployment and subsequent surveys administered between 1 and 9 months after returning from deployment. All study participants provided written informed consent to have their survey data linked to their Army and DoD administrative data and used in deidentified analyses. Human Subjects Committees of the University of Michigan and the Uniformed Services University of the Health Sciences approved all recruitment, informed consent, and data collection protocols. In addition, the Army Medical Research and Materiel Command approved the surveys of soldiers deployed to Afghanistan, who were surveyed as they transitioned through Kuwait for mid-tour leave. As detailed elsewhere,²⁴ post-stratification weighting adjusted the baseline samples for discrepancies with administrative variables available for all soldiers.

Longitudinal follow-up surveys

Longitudinal surveys (LS) were administered to a subset of baseline STARRS survey respondents in 2016-2017 (LS1) and then again in 2018-2019 (LS2). The initial LS sampling frame over-sampled baseline STARRS survey participants who reported a history of mental disorders or suicidality in their baseline surveys. Potential LS1 respondents were mailed a letter inviting them to participate in the survey, with a $50 incentive offered for participation and a link to the web-based survey. Initial nonrespondents were then sent a series of email and text invitations and reminders followed by phone calls. After these recruitment phases, a subsample of nonrespondents was offered an increased incentive of $100 (Fig. S1) before ending recruitment. Final LS1 data were weighted to include the nonresponse adjustment weights developed for the baseline Army STARRS surveys²⁴ along with an additional weight to adjust for over-sampling of baseline respondents with mental disorders and the under-representation of difficult-to-recruit participants in LS1. A total of n = 14,508 respondents completed LS1, with a weighted (for the differential sampling) response rate of 35.6%. This weighted sample was then post-stratified to adjust for differential response related to survey variables available for all baseline Army STARRS survey respondents and all administrative data available for these baseline respondents as of December 31, 2016 (the last date at which Army administrative data were available to the STARRS-LS research team). All LS1 respondents were eligible to complete LS2, which used the same field procedures as LS1 (Fig. S2). The LS2 conditional response rate was 83.7% (n = 12,156) among LS1 participants. The same post-stratification procedures used in LS1 were also used in LS2 to adjust for nonrandom loss to follow-up.

Analytic samples

Results reported here combine data from LS1 and LS2 among respondents who were no longer on active duty at the time of their focal LS survey(s). The term “no longer on active duty” is defined as either (1) separated from active duty service (e.g., administrative or medical discharge; fulfilled service obligation; released from obligation; transferred to Individual Ready Reserve or Inactive National Guard Standby Reserve; retired after 20+ years of qualifying service; reached retired pay eligibility age; medical retirement) or (2) deactivated (i.e., in a Reserve or National Guard Component but no longer on orders or activated [released from active duty in the Selected Reserve, Active Guard Reserve, etc.]). The analysis of suicide attempts was additionally limited to respondents who were last on active duty no less than 12 months ago because we asked only about a 12-month recall period for suicide attempts in the LS surveys. This was not necessary for the analysis of homelessness because the question assessing past 12-month homelessness allowed for responses from those on active duty less than 12 months before the survey (see below). In addition, both the suicide attempt and homelessness analyses excluded LS2 respondents who reported these outcomes in the LS1 survey to avoid double counting any single respondent as having either of these outcomes. This means that, by construction, none of the respondents who were in both the LS1 and LS2 analysis samples reported either a suicide attempt or homelessness in the 12 months before LS1. The full sample included n = 4338 observations for the suicide attempt analysis and n = 6837 for the homelessness analysis.

Measures

Suicide attempts

Self-reported suicide attempts were assessed in the LS surveys using a question adapted from the Columbia-Suicide Severity Rating Scale²⁷: “Did you ever make a suicide attempt (i.e., purposefully hurt yourself with at least some intention to die) at any time since your last survey?” Respondents who responded “yes” were then asked about lifetime number of SAs and recency. We focused on suicide attempts reported to have occurred within 12 months of the focal LS survey.

Homelessness

The LS surveys included a modified question from the Veterans Health Administration’s Homelessness Screening Clinical Reminder^28,29 on how much time over the past 12 months (or since most recently leaving or being released from active duty if less than 12 months ago) respondents were “living in stable housing that you own, rent, or stayed in as part of a household.” If the response was anything other than “all of the time,” a follow-up question asked how many months in the past 12 (or since most recently leaving or being released from active duty) respondents were homeless. A response of one or more months was coded as homeless in the past 12 months.

Risk indices

Earlier STARRS-LS reports describe the development of risk indices in which baseline Army STARRS survey data and Army/DoD administrative data available before soldiers left active duty were used to predict the subsequent occurrence of suicide attempts²¹ and homelessness²² in the 12 months before the LS surveys among respondents no longer on active duty. Procedures used to develop these risk indices were presented in detail in the original reports but, in brief, involved using ensemble machine learning methods to predict the outcomes from nine known domains of predictors: socio-demographics, Army career history, personality characteristics, physical health problems, psychiatric disorders, self-injurious thoughts and behaviors, chronic stressors, adverse childhood experiences, and other lifetime traumatic events. A composite predicted probability of each outcome was then generated to arrive at each risk index. Importantly, information about discharge characterization was not included in these indices. For purposes of this report, we standardized each risk index to have a mean of 0 and variance of 1 in the total sample, allowing the regression coefficients for the indices to be interpreted as mean differences in relative risk of the outcome associated with 1 SD difference in index scores.

Discharge characterization

The Defense Manpower Data Center administrative records of each separated or deactivated soldier include one of the following “character of service” designations: honorable, general (under honorable conditions), under other than honorable conditions, bad conduct, dishonorable (or dismissal for officers), or uncharacterized.³⁰ Honorable and general (under honorable conditions) are distinguished from “bad paper” discharges (other than honorable, bad conduct, dishonorable) that may lead to ineligibility for any VA benefits or services. An uncharacterized discharge, in which there is no attempt to characterize an individual’s service, is given to soldiers who separate within the first 180 days of beginning military service.³¹ For LS respondents whose discharge characterization was missing, we differentiate Regular Army (Missing-RA) from Guard/Reserve (Missing-GR).

Analysis Methods

To evaluate the associations of discharge characterization with post-separation/deactivation suicide attempts and homelessness, controlling for the predicted probabilities in the risk indices, we used modified Poisson regression equations with robust variance estimation.³² We used the Taylor linearization method to adjust SEs for the clustering and weighting of the LS samples.³³

Given the rarity of both outcomes, results were pooled across the LS1 and LS2 samples. In addition, given that a substantial proportion of LS respondents were new soldiers at baseline and that uncharacterized discharges occur only early in Army careers, models were estimated not only in the total pooled LS sample but also separately for respondents who were originally in the NSS and those who were originally in either the AAS or PPDS.

RESULTS

Sample Characteristics

As noted previously, the total LS1 sample was weighted to make it representative of the population from which the sample was drawn: all soldiers were on active duty at the time of the baseline assessment. LS2 was then reweighted to adjust for minor loss to follow-up. The weighted subsamples of recent separated/deactivated soldiers used in our analysis reflect these characteristics (Table I). Most respondents from the NSS, those who were new soldiers at baseline, were in the age range 18-26, whereas AAS/PPDS respondents, who represented soldiers at all stages of their Army careers, had a wider age distribution. Modal respondents in both the NSS and AAS/PPDS subsamples were male, non-Hispanic White, high school graduates, and had a junior enlisted rank (E1-E4), although the latter was much more common in the NSS than AAS/PPDS subsamples. The two subsamples differed more substantially in marital status, with the majority of NSS respondents never married and most AAS/PPDS respondents married. Given that the NSS was originally based on new soldiers, NSS respondents had a much lower mean number of years in service and a lower probability of ever having a combat deployment than those in the AAS/PPDS. The two subsamples were quite similar in duty military occupational specialty and in assigned commands, other than for a higher proportion of the NSS than AAS/PPDS being in the activated U.S. National Guard/Reserve while on active duty.

TABLE I.

Demographic and Army Career Characteristics of Separated/deactivated Survey Respondents

	Suicide attempt sample (n = 4338)a									Homelessness sample (n = 6837)b
	NSS (n = 1844)		AAS/PPDS (n = 2494)		Total (n = 4338)					NSS (n = 3344)		AAS/PPDS (n = 3493)		Total (n = 6837)
	%	(SE)	%	(SE)	%	(SE)	χ2	P	df	%	(SE)	%	(SE)	%	(SE)	χ2	P	df
Demographic characteristics
Age							92.8	<0.001	3							187.5	<0.001	3
18-22	45.2	(2.6)	7.3	(1.4)	23.6	(1.5)				33.7	(1.5)	6.6	(0.9)	19.8	(1.0)
23-26	35.6	(2.3)	21.7	(1.6)	27.7	(1.3)				41.7	(1.5)	19.3	(1.3)	30.2	(1.0)
27-33	14.6	(1.7)	32.2	(2.0)	24.7	(1.4)				21.0	(1.3)	35.4	(1.4)	28.4	(1.0)
34+	4.5	(1.0)	38.8	(2.0)	24.0	(1.4)				3.5	(0.5)	38.7	(1.6)	21.6	(1.0)
Gender							0.9	0.34	1							4.7	0.030	1
Female	19.0	(2.0)	16.5	(1.7)	17.6	(1.3)				19.3	(1.3)	15.5	(1.2)	17.4	(0.9)
Race							1.9	0.13	3							2.4	0.07	3
Non-Hispanic white	66.4	(2.6)	68.5	(2.3)	67.6	(1.7)				65.1	(1.5)	69.4	(1.5)	67.3	(1.0)
Non-Hispanic black	18.1	(2.0)	14.0	(1.4)	15.8	(1.2)				17.1	(1.1)	14.4	(1.1)	15.7	(0.8)
Hispanic	10.5	(1.7)	9.7	(1.3)	10.0	(1.0)				11.8	(1.0)	9.8	(0.9)	10.8	(0.7)
Other	5.0	(1.0)	7.8	(1.5)	6.6	(0.9)				5.9	(0.7)	6.4	(0.9)	6.2	(0.6)
Lifetime max education							34.1	<0.001	2							55.6	<0.001	2
High school diploma	91.1	(1.2)	70.5	(2.1)	79.3	(1.4)				87.2	(0.7)	69.7	(1.4)	78.3	(0.8)
Some college	2.7	(0.7)	4.6	(0.9)	3.8	(0.6)				2.5	(0.4)	5.6	(0.7)	4.1	(0.4)
College or more	6.2	(1.1)	24.9	(1.9)	16.9	(1.2)				10.3	(0.7)	24.7	(1.4)	17.7	(0.8)
Lifetime marital history							62.3	<0.001	2							165.0	<0.001	2
Currently	29.8	(2.2)	63.8	(2.4)	49.2	(1.7)				30.5	(1.4)	66.6	(1.8)	49.0	(1.2)
Previously	2.3	(0.7)	8.1	(1.1)	5.6	(0.7)				0.8	(0.2)	7.4	(0.8)	4.2	(0.5)
Never	67.9	(2.3)	28.1	(2.3)	45.2	(1.8)				68.7	(1.4)	26.0	(1.7)	46.8	(1.2)
Army career characteristics
Separation status							15.1	<0.001	1							88.7	<0.001	1
Separated (vs. deactivated)	79.1	(2.1)	89.0	(1.4)	84.7	(1.2)				63.0	(1.5)	81.3	(1.2)	72.4	(1.0)
Years in service							18.7	<0.001	1							24.4	<0.001	1
Number, mean/SD/t	2.7	(2.6)	10.3	(13.8)	7.1	(8.7)				3.3	(2.7)	10.9	(12.5)	7.2	(10.9)
Lifetime deployment							112.8	<0.001	3							199.9	<0.001	3
0	79.0	(1.8)	20.2	(2.0)	45.5	(1.8)				77.4	(1.4)	15.5	(1.4)	48.6	(1.4)
1	19.6	(1.8)	37.5	(1.9)	29.8	(1.4)				18.7	(1.3)	39.3	(1.5)	45.6	(1.3)
2	0.7	(0.2)	23.9	(1.6)	14.0	(1.0)				2.7	(0.4)	24.2	(1.2)	29.3	(1.0)
3+	0.7	(0.4)	18.3	(1.6)	10.7	(1.0)				1.2	(0.3)	21.0	(1.3)	13.7	(0.7)
Rank							109.1	<0.001	2							180.3	<0.001	2
Junior enlisted (E1-E4)	90.9	(1.5)	46.3	(2.1)	65.5	(1.6)				83.3	(0.9)	42.6	(1.8)	62.4	(1.1)
Senior enlisted (E5-E9)	8.2	(1.4)	40.4	(2.1)	26.6	(1.5)				14.2	(0.9)	43.5	(1.5)	29.2	(0.9)
Officer	0.9	(0.4)	13.3	(1.3)	8.0	(0.8)				2.5	(0.3)	13.9	(1.1)	8.3	(0.6)
Duty-military occupational specialty							10.5	<0.001	3							8.0	<0.001	3
Direct combat arms	32.2	(2.9)	26.1	(2.0)	28.7	(1.7)				28.3	(1.9)	24.5	(1.7)	26.3	(1.3)
Indirect combat arms	4.9	(1.1)	4.1	(0.7)	4.5	(0.6)				4.3	(0.7)	4.8	(0.9)	4.6	(0.6)
Combat support	30.3	(2.4)	20.3	(1.8)	24.6	(1.4)				30.5	(1.4)	22.7	(1.6)	26.5	(1.1)
Combat service support	32.5	(2.1)	49.5	(2.4)	42.2	(1.7)				36.9	(1.7)	48.0	(1.9)	42.6	(1.2)
Command							8.3	<0.001	6							28.1	<0.001	6
FORSCOM	23.7	(2.1)	33.3	(2.0)	29.2	(1.5)				18.8	(1.3)	32.8	(1.5)	26.0	(1.0)
TRADOC	10.5	(1.7)	6.6	(1.7)	8.3	(1.2)				7.4	(0.9)	5.0	(0.8)	6.2	(0.6)
North/South America, Europe/Central/Africa, Pacific	3.6	(0.9)	5.3	(0.9)	4.6	(0.6)				2.9	(0.5)	5.1	(0.5)	4.0	(0.4)
Special operations	0.9	(0.4)	1.6	(0.5)	1.3	(0.3)				1.6	(0.3)	1.9	(0.4)	1.7	(0.2)
MEDCOM	0.6	(0.3)	7.0	(1.3)	4.3	(0.7)				0.7	(0.2)	6.9	(0.8)	3.9	(0.5)
AMC, other, unknown	9.1	(1.6)	11.2	(1.5)	10.3	(1.1)				5.9	(0.6)	10.2	(0.8)	8.1	(0.5)
Guard/Reserve	51.6	(2.6)	35.0	(2.1)	42.1	(1.7)				62.7	(1.6)	38.1	(1.6)	50.1	(1.1)

Estimates reflect weighted data.

Abbreviations: NSS = New Soldier Study; AAS = All Army Study; PPDS = Pre-Post Deployment Study; FORSCOM = Forces Command; TRADOC = Training and Doctrine Command; MEDCOM = Medical Command; AMC = Army Materiel Command.

^aEstimates reflect weighted data to make the n = 4338 respondents considered here representative of the n = 12,296 in the full eligible LS1-LS2 pooled sample. The smaller sample is due to our use of a case-control sampling scheme to develop the Super Learner model.

^bEstimates reflect weighted data to make the n = 6837 respondents considered here representative of the n = 16,807 in the full eligible LS1-LS2 pooled sample. The smaller sample is due to our use of a case-control sampling scheme to develop the Super Learner model.

Distribution of Discharge Characterization

The distribution of discharge characterization was quite different in the NSS and AAS/PPDS subsamples (Table II). This was true partly because uncharacterized discharge was much more common among respondents who were originally in the NSS (16.6% and 12.8% in the suicide attempt and homelessness subsamples, respectively) than the AAS/PPDS (1.0% and 0.6%), but an even larger difference was that the proportion of respondents with missing discharge characterization data from the Guard and Reserve was much higher in the NSS (26.4% and 40.2%) than the AAS/PPDS (5.6% and 8.7%). Because of these differences, the proportion of respondents with an honorable discharge was smaller in the NSS than AAS/PPDS (41.4% and 33.5% NSS vs. 80.6% and 75.4% AAS/PPDS). However, the proportion with a general discharge was higher in the NSS (10.5% and 6.3%) than the AAS/PPDS (5.2% and 4.3%), as was the proportion with a bad paper discharge (1.2% and 0.7% NSS vs. 0.3% and 0.2% AAS/PPDS).

TABLE II.

Distribution of Discharge Characterization

	Suicide attempt sample (n = 4338)						Homelessness sample (n = 6837)
	NSS (n = 1844)		AAS/PPDS (n = 2494)		Total (n = 4338)		NSS (n = 3344)		AAS/PPDS (n = 3493)		Total (n = 6837)
	%	(SE)	%	(SE)	%	(SE)	%	(SE)	%	(SE)	%	(SE)
Discharge characterization
Honorable	41.4	(2.5)	80.6	(2.1)	63.8	(1.8)	33.5	(1.5)	75.4	(1.4)	55.0	(1.2)
General	10.5	(1.6)	5.2	(1.1)	7.5	(0.9)	6.3	(0.9)	4.3	(0.8)	5.3	(0.6)
Uncharacterized	16.6	(1.9)	1.0	(0.4)	7.7	(0.9)	12.8	(1.0)	0.6	(0.2)	6.5	(0.5)
Bad papera	1.2	(0.6)	0.3	(0.2)	0.7	(0.3)	0.7	(0.2)	0.2	(0.1)	0.5	(0.1)
Missing—RA	3.9	(0.9)	7.2	(1.3)	5.8	(0.8)	6.5	(0.7)	10.8	(0.9)	8.7	(0.5)
Missing—GR	26.4	(2.5)	5.6	(1.1)	14.5	(1.3)	40.2	(1.5)	8.7	(1.2)	24.0	(1.0)

Estimates reflect weighted data.

Abbreviations: NSS = New Soldier Study; AAS = All Army Study; PPDS = Pre-Post Deployment Study; RA = Regular Army; GR = Guard/Reserve.

^aThe 22 bad paper discharges in the suicide attempt sample include: other than honorable (n = 19) and bad conduct (n = 3). The 42 bad paper discharges in the homelessness sample include: other than honorable (n = 39), bad conduct (n = 2), and dishonorable (n = 1).

Associations of Discharge Characterization with Future Suicide Attempts

Twelve-month prevalence of self-reported suicide attempts in the total pooled LS sample was 1.0%, including 1.2% in the NSS and 0.8% in the AAS/PPDS. The RR of the standardized suicide attempt risk index in predicting suicide attempts was a statistically significant 6.2 in the total sample, 6.3 in the NSS, and 5.9 in the AAS/PPDS. This means that an increase of 1 SD in predicted probability of suicide attempt was associated with roughly a 6-fold increased relative risk both in the total sample and in each of these two subsamples (Table III). None of the respondents with a bad paper discharge reported a suicide attempt. In the total sample, RR was 0.9-1.3, in comparison, for soldiers with general and uncharacterized discharges, neither of which was statistically significant either individually or as a set (F₂ = 0.8, P = 0.44). RR for these same two predictors was 0.6-1.2, again neither significant either alone or as a set (F₂ = 1.8, P = 0.16), in the NSS subsample. Given the rarity of uncharacterized discharge in the AAS/PPDS subsample, there were no suicide attempts among soldiers with that type of discharge in that subsample. The RR among soldiers with a general discharge was 1.4, which, although elevated, was not statistically significant (F₁ = 0.7, P = 0.40).

TABLE III.

Relative Risk of Discharge Characterization with Suicide Attempt (N = 4338)^a

	Suicide attempt sample (n = 4338)
	NSS				AAS/PPDS				Total
	Prevalence		(n = 1844)		Prevalence		(n = 2494)		Prevalence		(n = 4338)
	%	(SE)	RR	(95% CI)	%	(SE)	RR	(95% CI)	%	(SE)	RR	(95% CI)
Suicide attempt predicted riskb	–	–	6.3*	(5.0-8.0)	–	–	5.9*	(4.0-8.9)	–	–	6.2*	(4.9-7.9)
Discharge characterization
Honorable (reference)	1.2	(0.3)	1.0	–	0.9	(0.2)	1.0	–	1.0	(0.1)	1.0	–
General	0.9	(0.5)	0.6	(0.2-1.4)	2.2	(1.0)	1.4	(0.6-3.3)	1.4	(0.5)	0.9	(0.5-1.7)
Uncharacterized	2.2	(0.6)	1.2	(0.7-2.2)	0.0	(0.0)	0.0	–	2.1	(0.6)	1.3	(0.8-2.2)
Bad paperc	0.0	(0.0)	0.0	–	0.0	(0.0)	0.0	–	0.0	(0.0)	0.0	–
Missing—RA	0.6	(0.6)	2.2	(0.3-15.9)	0.4	(0.2)	1.5	(0.4-5.4)	0.5	(0.2)	1.8	(0.6-5.4)
Missing—GR	0.9	(0.3)	1.2	(0.6-2.6)	0.0	(0.0)	0.0	–	0.7	(0.2)	1.2	(0.6-2.2)
F 2,1,2	–	–	1.8	0.16	–	–	0.7	0.40	–	–	0.8	0.44
Total	1.2	(0.2)			0.8	(0.1)			1.0	(0.1)

Estimates reflect weighted data.

Abbreviations: NSS = New Soldier Study; AAS = All Army Study; PPDS = Pre-Post Deployment Study; RR = Relative Risk; CI = Confidence Interval; RA = Regular Army; GR = Guard/Reserve.

^*Significant at the 0.05 level, two-sided test.

^aRelative Risk estimates were computed from survey-design based modified Poisson regression models with robust variance estimates.

^bSuicide attempt predicted risk was computed from the predicted probability of suicide attempt based on a previously developed machine learning model using information available during active duty, including the demographic and Army career characteristics presented in Table I. The predicted probability was standardized to a mean equal to zero and SD equal to 1 to convert to a predicted risk score.

^cIncludes the following discharge characterizations: other than honorable, bad conduct, and dishonorable.

The majority (84.7-72.4%) of LS respondents were separated rather than deactivated. Replication of the same analysis in the subsample of respondents who were separated found no significant association of general or uncharacterized discharges, relative to honorable discharges, with suicide attempts among either the total sample (RR = 1.0-1.4, F₂ = 1.1, P = 0.35) or NSS respondents (RR = 0.6-1.3, F₂ = 1.9, P = 0.16). The association of general discharge with suicide attempts was also nonsignificant among AAS/PPDS respondents (RR = 1.6, F₁ = 1.3, P = 0.28). Neither of these RRs was significant when we looked separately at respondents who were last on active duty either 1-2 or 3+ years before their LS survey (Table S1).

Associations of Discharge Characterization with Future Homelessness

Twelve-month prevalence of self-reported homelessness in the total pooled LS sample was 2.9% (n = 443; 30 of whom also reported a 12-month suicide attempt) and higher in the NSS (3.2%) than the AAS/PPDS (2.7%). The RR of the standardized homelessness risk index in predicting homelessness was a statistically significant 5.4 in the total sample, 4.1 in the NSS, and 5.1 in the AAS/PPDS. This means that an increase of 1 SD in the predicted probability of homelessness was associated with roughly a 4- to 5-fold increased relative risk both in the total sample and in each of these two subsamples (Table IV). Respondents with a bad paper discharge had a dramatically higher prevalence of subsequent homelessness than other respondents in the total sample (23.9%) as well as separately in the NSS (25.1%) and AAS/PPDS (17.4%) subsamples. There was a significantly elevated RR of homelessness among soldiers with a bad paper discharge in the total sample and the two major subsamples even after controlling for the standardized homelessness index (RR = 4.2-4.8). General and uncharacterized discharges were not associated with significantly elevated risks of homelessness compared with honorable discharges either in the total sample or the NSS or AAS/PPDS subsamples (RR = 1.1-1.5). We found consistent results when repeating these analyses using a more conservative definition of homelessness that required ≥3 months of reported homelessness in the past 12 months (Table S3).

TABLE IV.

Relative Risk of Discharge Characterization with Homelessness (N = 6837)^a

	Homelessness sample (n = 6837)
	NSS				AAS/PPDS				Total
	Prevalence		(n = 3344)		Prevalence		(n = 3493)		Prevalence		(n = 6837)
	%	(SE)	RR	(95% CI)	%	(SE)	RR	(95% CI)	%	(SE)	RR	(95% CI)
Homelessness predicted riskb	–	–	4.1*	(3.1-5.4)	–	–	5.1*	(3.4-7.6)	–	–	5.4*	(3.9-7.4)
Discharge characterization
Honorable	2.9	(0.5)	1.0	–	2.6	(0.4)	1.0	–	2.7	(0.3)	1.0	–
General	7.6	(2.0)	1.5	(0.8-3.1)	7.9	(2.4)	1.1	(0.5-2.4)	7.7	(1.5)	1.4	(0.8-2.3)
Uncharacterized	4.8	(0.9)	1.3	(0.8-2.3)	0.0	(0.0)	0.0	(0.0-0.0)	4.6	(0.9)	1.4	(0.9-2.3)
Bad paperc	26.1	(10.9)	4.2*	(2.0-8.6)	17.3	(13.2)	4.8*	(1.3-17.9)	23.9	(8.7)	4.4*	(2.3-8.3)
Missing—RA	1.9	(0.7)	0.8	(0.4-1.5)	1.5	(0.4)	0.7	(0.3-1.3)	1.7	(0.4)	0.7	(0.4-1.1)
Missing—GR	2.0	(0.3)	1.0	(0.6-1.6)	1.4	(0.4)	1.1	(0.5-2.1)	1.9	(0.2)	1.1	(0.8-1.5)
F 3,2,3	–	–	4.9*	0.003	–	–	2.7	0.07	–	–	11.7*	<0.001
Total	3.2	(0.3)			2.7	(0.3)			2.9	(0.2)

Estimates reflect weighted data.

Abbreviations: NSS = New Soldier Study; AAS = All Army Study; PPDS = Pre-Post Deployment Study; RR = Relative Risk; CI = Confidence Interval; RA = Regular Army; GR = Guard/Reserve.

^*Significant at the 0.05 level, two-sided test.

^aRelative Risk estimates were computed from survey-design based modified Poisson regression models with robust variance estimates.

^bHomelessness predicted risk was computed from the predicted probability of homelessness based on a previously developed machine learning model using information available during active duty, including the demographic and Army career characteristics presented in Table I. The predicted probability was standardized to a mean equal to zero and SD equal to 1 to convert to a predicted risk score.

^cIncludes the following discharge characterizations: other than honorable, bad conduct, and dishonorable.

Further disaggregation showed that the significantly elevated RR of bad paper discharge existed among soldiers who were on active duty either 0-1 (RR = 2.3) or 2+ (RR = 5.2) years before their LS survey. In addition, AAS respondents on active duty 0-1 years before their LS survey had a significantly elevated RR of homelessness if they had a general discharge (RR = 8.7). However, this association became nonsignificant among respondents who had been out of active duty 2 or more years before their LS survey (Table S2).

DISCUSSION

This novel prospective study of the association of discharge characterization with suicide attempt and homelessness, two public health concerns of national significance, produced a noteworthy finding: that risk of homelessness was more than four times as high among soldiers with a bad paper discharge as those with an honorable discharge, even after controlling for a previously developed homelessness risk index. This was a robust finding that was true not only in the total sample, but also in both the NSS and AAS/PPDS subsamples, among respondents who separated (vs. deactivated), and regardless of the time elapsed between leaving active duty and participating in the LS survey. In contrast to previous studies based on VA data,^17,18 the present study included individuals with dishonorable discharges, more fully representing the range of bad paper service characterizations. It would be valuable for future studies to examine whether this association is present among both women and men, and across other important segments of the population. We did not find an association between discharge characterization and suicide attempt after controlling for our previously developed suicide attempt risk index. Although none of the respondents with a bad paper discharge reported a suicide attempt, the fact that both bad paper discharges and suicide attempts are rare occurrences means that the absence of any suicide attempts in this subgroup is not a reliable result, highlighting the need for research with larger samples.

Mental illness and substance use disorders are among the strongest and most consistent predictors of homelessness among veterans.^14,34,35 These same factors are likely associated with misconduct and other behaviors that may lead to a bad paper discharge.^18,36 The U.S. Government Accountability Office study found that 62% of service members administratively separated for misconduct had received a mental health diagnosis within the two years before separation.³⁷ Although soldiers with a bad paper discharge have historically been ineligible for VA mental healthcare, including the period during which the data for the present study were collected, there have been a number of recent developments in VA that increased access. For one, mental healthcare eligibility was expanded for certain service members with an other-than-honorable discharge.³⁸ A second development is the recent passage of two laws^39,40 that now authorize VA to fund legal services for veterans by awarding grants to legal service providers. These legal service providers will help veterans with various civil legal issues, including military discharge upgrades, which may make VA services more accessible. A third area of development specific to homeless veterans is the National Defense Authorization Act for Fiscal Year 2021,³⁹ which included provisions to expand eligibility for the U.S. Department of Housing and Urban Development-VA Supportive Housing (HUD-VASH) program which has housed more than 100,000 veterans over two decades.⁴¹ The new provisions allow veterans with other-than-honorable discharges and bad conduct discharges due to a special court-martial to participate in HUD-VASH. It will be important for future research to examine whether these efforts reduce or eliminate the association of bad paper discharges with homelessness found in this analysis, as well as the overall rate of homelessness among former service members. Given the association between veteran homelessness and suicide risk,⁴² such efforts may also help prevent suicides after separation/deactivation.

Although soldiers who have served for less than 180 days are generally eligible for VA benefits and services,³¹ they are not eligible for the Transition Assistance Program, which requires at least 180 days of continuous active service.⁴³ However, the DoD funds a program called inTransition that provides coaching (e.g., helping participants locate and connect with mental health providers and community resources) and is available to all service members transitioning within or discharging from the military, regardless of time in service, time since discharge, or discharge characterization.⁴⁴ Targeted efforts to encourage use of this program by soldiers with a bad paper discharge may aid in reducing risk of future homelessness.

This study has three noteworthy limitations. First, the sample was limited to soldiers who participated in Army STARRS surveys in 2011-2014 and could then be traced and resurveyed in 2016-2019, raising the possibility of sample bias even though we adjusted for nonrandom response. For example, participation of homeless individuals in the LS survey may have been reduced by the potential challenges around locating and recruiting such individuals. Second, our outcomes were based exclusively on self-reported suicide attempts and homelessness. We do not know whether administratively documented suicide attempts and/or homelessness (e.g., in VA medical records) would have different associations with discharge characterization. Third, it is not clear why the proportion of missing discharge characterization data for Guard/Reserve soldiers was so high in the NSS. Had a discharge characterization been issued to those soldiers, it may have influenced our findings to an unknown extent.

With those limitations in mind, our prospective findings indicate that soldiers who receive a bad paper discharge are at increased risk for homelessness over and above other known risk factors for homelessness. Policies and programs that enhance transition assistance and access to mental healthcare for soldiers at greatest risk, regardless of discharge characterization, may aid in reducing post-separation/deactivation homelessness.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Military Medicine online.

FUNDING

Army STARRS was sponsored by the Department of the Army and funded under cooperative agreement number U01MH087981 with the U.S. Department of Health and Human Services, National Institutes of Health, National Institute of Mental Health (NIH/NIMH). Subsequently, STARRS-LS was sponsored and funded by the Department of Defense (USUHS Grant Nos. HU00011520004 and HU0001202003). The grants were administered by the Henry M. Jackson Foundation for the Advancement of Military Medicine Inc. (HJF).

CONFLICT OF INTEREST STATEMENT

In the past 3 years, Dr. Kessler was a consultant for Datastat, Inc., Holmusk, RallyPoint Networks, Inc., and Sage Therapeutics. He has stock options in Mirah, PYM, and Roga Sciences. In the past 3 years, Dr. Stein received consulting income from Actelion, Acadia Pharmaceuticals, Aptinyx, atai Life Sciences, Boehringer Ingelheim, Bionomics, BioXcel Therapeutics, Clexio, EmpowerPharm, Engrail Therapeutics, GW Pharmaceuticals, Janssen, Jazz Pharmaceuticals, and Roche/Genentech. Dr. Stein has stock options in Oxeia Biopharmaceuticals and EpiVario. He is paid for his editorial work on Depression and Anxiety (Editor-in-Chief), Biological Psychiatry (Deputy Editor), and UpToDate (Co-Editor-in-Chief for Psychiatry). In the past 3 years, Dr. Marx received royalties from Guilford Press and the American Psychological Association. The other authors report no conflicts of interest.

ROLE OF THE FUNDER/SPONSOR

As a cooperative agreement, scientists employed by the NIMH and U.S. Army as liaisons and consultants collaborated to develop the Army STARRS study protocol and data collection instruments and to supervise data collection. The funders had no further role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation or approval of the manuscript; and decision to submit the manuscript for publication.