Content uploaded by Christian Betancourt
Author content
All content in this area was uploaded by Christian Betancourt on Nov 30, 2022
Content may be subject to copyright.
MILITARY MEDICINE, 187, 11/12:e1422, 2022
Substance Use Relapse Among Veterans at Termination of
Treatment for Substance Use Disorders
LT Christian A. Betancourt, MHA, MSC, USN*; Panagiota Kitsantas, PhD*;
Deborah G. Goldberg, PhD*; LCDR Beth A. Hawks, PhD, MSC, USN†
ABSTRACT
Introduction:
Military veterans continue to struggle with addiction even after receiving treatment for substance use disorders (SUDs).
Identifying factors that may inuence SUD relapse upon receiving treatment in veteran populations is crucial for inter-
vention and prevention efforts. The purpose of this study was to examine risk factors that contribute to SUD relapse upon
treatment completion in a sample of U.S. veterans using logistic regression and classication tree analysis.
Materials and Methods:
Data from the 2017 Treatment Episode Data Set—Discharge (TEDS-D) included 40,909 veteran episode observations.
Descriptive statistics and multivariable logistic regression analysis were conducted to determine factors associated with
SUD relapse after treatment discharge. Classication trees were constructed to identify high-risk subgroups for substance
use after discharge from treatment for SUDs.
Results:
Approximately 94% of the veterans relapsed upon discharge from outpatient or residential SUD treatment. Vet-
erans aged 18-34 years old were signicantly less likely to relapse than the 35-64 age group (odds ratio [OR]
0.73, 95% condence interval [CI]: 0.66, 0.82), while males were more likely than females to relapse (OR 1.55,
95% CI: 1.34, 1.79). Unemployed veterans (OR 1.92, 95% CI: 1.67, 2.22) or veterans not in the labor force
(OR 1.29, 95% CI: 1.13, 1.47) were more likely to relapse than employed veterans. Homeless vs. independently housed
veterans had 3.26 (95% CI: 2.55, 4.17) higher odds of relapse after treatment. Veterans with one arrest vs. none were
more likely to relapse (OR 1.52, 95% CI: 1.19, 1.95). Treatment completion was critical to maintain sobriety, as every
other type of discharge led to more than double the odds of relapse. Veterans who received care at 24-hour detox facilities
were 1.49 (95% CI: 1.23, 1.80) times more likely to relapse than those at rehabilitative/residential treatment facilities.
Classication tree analysis indicated that homelessness upon discharge was the most important predictor in SUD relapse
among veterans.
Conclusion:
Aside from numerous challenges that veterans face after leaving military service, SUD relapse is intensied by risk fac-
tors such as homelessness, unemployment, and insufcient SUD treatment. As treatment and preventive care for SUD
relapse is an active eld of study, further research on SUD relapse among homeless veterans is necessary to better under-
stand the epidemiology of substance addiction among this vulnerable population. The ndings of this study can inform
healthcare policy and practices targeting veteran-tailored treatment programs to improve SUD treatment completion and
lower substance use after treatment.
INTRODUCTION
Veterans have an increasing need for mental health care and
services for substance use disorders (SUDs).1Veterans have
higher rates of SUDs than nonveterans, with estimates rang-
ing from 17.1 to 32.9%, depending on polysubstance use.2
Research evidence indicates that 11% of veterans who sought
care within the Veterans Health Administration (VHA) met
*Department of Health Administration and Policy, George Mason
University, Fairfax, VA 22030-4444, USA
†Department of Preventive Medicine and Biometrics, Uniformed
Services University, Bethesda, MD 20814, USA
The views expressed in this article are those of the author and do not
necessarily reect the ofcial policy or position of the Department of the
Navy, Department of Defense, or the U.S. Government.
doi:https://doi.org/10.1093/milmed/usab280
© The Association of Military Surgeons of the United States 2021.
All rights reserved. For permissions, please e-mail: journals.
permissions@oup.com.
the criteria for an SUD diagnosis.3This problem is further
exacerbated by multiple physical and mental health chal-
lenges4that substance users face in completing SUD treat-
ment programs, thus resulting in higher relapse rates and
return to treatment programs.5Recent estimates show that
76% of veterans experience SUD relapse after receiving treat-
ment.6Substance use disorder relapse constitutes a serious
problem not only for the veterans themselves but also presents
an economic burden to the society. Over a decade ago, nearly a
third of all VHA costs ($12 billion) were attributed to provid-
ing veterans with services for treatment related to substance
use and mental health conditions.7Factors that inuence SUD
relapse upon receiving treatment services and follow-up care
must be closely monitored to ensure that effective targeted
intervention and prevention programs are being adequately
deployed and utilized.
A number of studies have examined factors that may
inuence SUD relapse in veterans within the VHA. Research
e1422 MILITARY MEDICINE, Vol. 187, November/December 2022
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
evidence indicates that Healthy Lifestyle Behaviors, including
recreational, social, coping/spiritual, and substance recovery
activities, have been associated with lower relapse rates in
veterans.8Follow-up on SUD care is as important as residen-
tial treatment itself. A 2016 study found a positive signi-
cant association between inpatient hospitalizations and SUDs
in the year immediately following residential treatment.9
This study shows the importance of enforcing continued and
focused SUD relapse prevention behaviors following residen-
tial treatment. In addition, research evidence indicates that
homeless veterans have a greater difculty in accessing and
receiving adequate treatment for SUD.10 Lack of resources
for this population is associated with disproportionate hos-
pitalization and emergency services utilization rates across
the country.11 Other factors associated with SUD relapse in
veterans include involvement with the justice system.12,13 Vet-
erans represent a signicant proportion of the correctional
population,14 and upon release from incarceration, they may
face homelessness and SUD relapse.12 Many of these individ-
ual factors can impede a veteran’s SUD recovery; however,
several of these elements combined (e.g., homelessness and
incarceration) can present a formidable impact on relapse.
Identifying factors that may inuence SUD relapse upon
receiving treatment in veteran populations is crucial for inter-
vention and prevention efforts. Limited research, however,
has been conducted on SUD relapse predictors after treat-
ment, especially from data collected outside of the VHA.
The VHA maintains robust veteran datasets along with com-
prehensive health-related and demographic data; however,
the VHA is not the only agency that captures data involv-
ing veteran demographics, health information, and treatment.
Organizations such as the Substance Abuse and Mental Health
and Services Administration (SAMHSA) record and maintain
rich substance use data that report both veteran and nonvet-
eran populations. The SAMHSA coordinates the collection
of Treatment Episode Dataset-Discharge (TEDS-D) data that
provide a unique opportunity to study SUD relapse upon
treatment in veteran populations.
The purpose of this study was to examine risk factors that
contribute to SUD relapse in U.S. veterans upon completion
of treatment using TEDS-D data. In addition, this study uses
both traditional statistical methods of data analysis (e.g., para-
metric regression) and classication trees. The use of machine
learning algorithms such as classication and regression trees
(CART), which constitutes a relatively novel technique in
the eld of SUD, reinforces research ndings by uncovering
clusters of risk factors of SUD relapse.
METHODS
Data Source
Data were used from the 1,661,207 treatment discharges
reported in the 2017 TEDS-D public use les within the
(SAMHSA) data repository. Treatment Episode Dataset-
Discharge data are reported from 47 states, Washington DC,
and Puerto Rico. Georgia, Oregon, and West Virginia did
not report sufcient discharge data for 2017 to be included in
this study.15 The TEDS-D, which is reported annually by the
SAMHSA, is publicly available and provides demographic
and characteristics of substance use treatment discharges
among people aged 12 years and older based on state-licensed
or certied substance abuse treatment centers that receive fed-
eral public funding.16 The TEDS-D represents a compilation
of data collected through the individual data collection sys-
tems of the state agencies for substance use treatment. More
information about data collection procedures are provided in
the TEDS State Instruction Manual.17
Sample
A subsample of the TEDS-D, 2017, was created to analyze
only the veteran sample (n=44,296). The TEDS-D identied
veterans with a “yes/no” question asking the participants if
they had served in the U.S. uniformed services. Discharges
ages 12-17 (n=178) were excluded from the study as indi-
viduals under the age of 17 years cannot join the military.18
Substance use disorder relapse upon treatment discharge had
7.27% missing data (n=3,209), which was excluded from
the total subsample. The nal analytic sample consisted
of 40,909 records. Missing data for demographic variables
ranged from 0% to 2.6%. Three measures, including employ-
ment status at discharge (15.1% missing), living arrange-
ments at discharge (15.6% missing), and arrests (<30 days
before admission) (12.6% missing), had a large percentage
of missing data that were either missing completely, reported
as unknown, not collected, or invalid. The data included
a mixture of outpatient and inpatient residential treatment
episodes.
Measures
The outcome variable of this study was substance use at
the point of discharge (relapse). The original eld response
in the survey was intended to “identify the client’s primary
substance use at admission and discharge” and provided 19
response options including “none” and 18 other types of
substances. This variable was converted into a binary vari-
able identifying discharges that reported either “no relapse”
or “relapse.” This study uses Wesson et al.’s denition of
relapse19 as “a discrete event, which occurs at the moment a
person resumes drug use or as a process which occurs over
time.” Wesson’s denition is appropriate given how SUD
relapse has been measured in TEDS-D outlined above.
Factors that may inuence SUD relapse were selected
based on previous research that assessed interrelationships
among different treatment stages and their effects on the out-
come of alcoholic patients.20 In the present study, 10 variables
were categorized into two sets separating sociodemographic
from treatment-related variables. Sociodemographic vari-
ables included age, gender, race/ethnicity (veteran’s race and
MILITARY MEDICINE, Vol. 187, November/December 2022 e1423
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
specic Hispanic or Latino origin), employment status (full-
time, part-time, unemployed, or not in labor force), education
(no high school, high school, or college), living arrangements
upon discharge (homeless, dependent living, or independent
living), and number of arrests a veteran had 30 days preced-
ing the date of admission to treatment services (none, one,
two, or more), or in the event that the client was in treat-
ment fewer than 30 days, this item refers to number of arrests
during the treatment period only. Treatment-specic variables
were used in this study such as reason for discharge. Options
for this variable included treatment completed, dropped
out of treatment, transferred to another facility, or “other”
(terminated from treatment, incarcerated, death, or other).
Service setting at discharge provides an option to report what
type of service was provided for treatment, such as detox
facilities, rehabilitative or residential facility, or ambula-
tory care facilities. Length of stay was measured by dura-
tion of treatment in days; time frames ranged between
1 day (outpatient stay), 2-30 days, 31-90 days, 91-180 days,
181-365 days, and more than 365 days. Stays longer than
365 days may be because veterans receive medication-assisted
treatment for their addictions. Medication-assisted treat-
ment, such as methadone treatment, presents the most
benet to its users when it is administered longer than
12 months.21
Data Analysis
Descriptive statistics were used to describe the sample char-
acteristics. Bivariate analysis with the Pearson’s chi-squared
test was used to test for signicant associations between
the dependent variable (drug/alcohol use upon treatment
discharge, i.e., relapse vs. no relapse) and each of the indepen-
dent variables, including all sociodemographic and treatment-
related variables. A multivariable logistic regression model
was conducted to determine associations between all inde-
pendent variables and SUD relapse after treatment discharge.
Robust standard errors were computed to adjust for poten-
tial departures from assumptions of the logistic regression
model. Statistical analyses were conducted using STATA,
version 16.22
Classication trees were constructed for this sample to
identify subgroups of veterans who are at high risk for sub-
stance use after being discharged from treatment for SUD.
The CART is a nonparametric technique and does not require
assumptions about the distribution of the data.23 This analytic
method can process large datasets with a high number of vari-
ables, and it is not affected by collinearities. Although logistic
regression estimates the associations between risk factors and
SUD relapse, it does not clearly demonstrate how multiple
risk factors interact in creating clusters that predict the depen-
dent variable and identify high-risk subgroups. Classication
and regression trees, however, can uncover multipart variable
associations and categorize high-risk subgroups based on a set
of predictor variables.23 This methodology has been used in
various health-related elds to not only assist researchers in
developing treatment strategies,24 but also as an alternative
method of mimicking actual thinking processes.25 In this
study, the classication tree was constructed based on 70%
of the observations in the sample, and the remaining 30%
of the sample was used to test the accuracy of the classi-
cation tree. The CART analysis was conducted using Minitab
19 computer software.26
RESULTS
Descriptive Statistics and Bivariate Associations
Table I provides descriptive statistics of the sample dis-
charges along with bivariate analysis between SUD relapse
and sample characteristics based on veteran treatment dis-
charges for the year 2017. Most veterans with SUD discharges
were White non-Hispanic (65%), male (89%), and aged
35-64 years (67%). The vast majority of all discharged veter-
ans reported being unemployed or not in the labor force upon
discharge (71%). Most veterans had at least a high school
education (57%) or had attended some college (41%). Nearly
two-thirds (64%) of all veteran discharges were recorded as
living independently upon discharge from treatment. In terms
of involvement with the justice system, 7% of veterans were
arrested at least once within 30 days before admission.
Bivariate analysis demonstrates statistically signicant
associations between all sociodemographic variables and
SUD relapse in veterans (P-value < .001) upon discharge.
Approximately 6% of all veteran discharges abstained from
using drugs/alcohol, with the majority of veterans relapsing
(94%). More than two-thirds (68%) of all veteran discharge
episodes in the 35-64 age group experienced SUD relapse
upon discharge. White non-Hispanic (65%) and males (90%)
represented the vast majority of episodes that resulted in SUD
relapse upon discharge. Approximately, 44% of veteran dis-
charges that did not report having relapsed, left treatment with
some level of employment, while more than two-thirds (71%)
of those who relapsed, reported being either unemployed or
not in the labor force. Among those discharges that reported
an education level of at least a high school education, more
than half (57%) relapsed after a treatment discharge. Home-
lessness was associated with a signicantly higher proportion
of SUD relapse (18%) than those who did not relapse upon
discharge from treatment (5%). Discharges that involved any
veteran arrests within the last 30 days before admission to
treatment demonstrated a slightly higher proportion of SUD
relapse (7%) than those that had no arrests (6%).
Descriptive statistics and associations between SUD
relapse and other treatment-related predictor variables are also
displayed in Table I. About half of all reported treatment dis-
charges resulted in complete treatment (48%). A large portion
of veterans (58%) were seen at ambulatory facilities followed
by rehab/residential facilities (24%) after discharging from
SUD treatment. About half the treatment episodes lasted more
than 1 day, but no more than 30days (51%). Of the veteran
e1424 MILITARY MEDICINE, Vol. 187, November/December 2022
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
TABLE I. Sample Characteristics and Associations of Sociodemographic and Treatment-Related Factors, and Relapse among Veterans
upon Discharge from Treatment
Sociodemographic variables
Entire sample
n=40,909
n(%)
Substance use disorder relapse
upon discharge
n(%)
38,381 (93.8%)
No substance use disorder
relapse upon discharge
n(%)
2,528 (6.2%) P-value
Age 40,909 .000
18-34 11,314 (27.7) 10,474 (27.3) 840 (33.2)
35-64 27,447 (67.1) 25,904 (67.5) 1,543 (61)
≥65 2,148 (5.2) 2,003 (5.2) 145 (5.7)
Gender 40,905 .000
Male 36,501 (89.2) 34,344 (89.5) 2,157 (85.3)
Female 4,404 (10.8) 4,033 (10.5) 371 (14.7)
Race/ethnicity 39,839 .000
White non-Hispanic 25,737 (64.6) 24,079 (64.5) 1,658 (66.2)
Black non-Hispanic 7,831 (19.8) 7,96 (19.8) 495 (19.8)
Hispanic 3,769 (9.5) 3,494 (9.4) 275 (11)
Othera2,442 (6.1) 2,367 (6.3) 75 (3)
Employment status (at discharge) 36,942 .000
Full-time 8,472 (22.9) 7,779 (22.2) 693 (36.6)
Part-time 2,418 (6.6) 2,276 (6.5) 142 (7.5)
Unemployed 12,415 (33.6) 11,987 (34.2) 428 (22.6)
Not in labor force 13,637 (36.9) 13,006 (37.1) 631 (33.3)
Education 40,006 .000
No high school 740 (1.8) 701 (1.9) 39 (1.6)
High school 22,783 (57) 21,237 (56.6) 1,546 (62)
College 16,483 (41.2) 15,577 (41.5) 906 (36.4)
Living arrangements (at discharge) 36,677 .000
Homeless 6,176 (16.8) 6,083 (17.5) 93 (4.9)
Dependent living 6,926 (18.9) 6,566 (18.9) 360 (19)
Independent living 23,575 (64.3) 22,135 (63.6) 1,440 (76.1)
Arrests (<30 days before admission) 37,768 .000
None 35,218 (93.2) 32,878 (93.2) 2,340 (94.4)
One 2,172 (5.8) 2,063 (5.8) 109 (4.4)
Two or more 378 (1) 348 (1) 30 (1.2)
Reason for discharge 40,909 .000
Treatment completed 19,672 (48.1) 18,319 (47.7) 1,353 (53.5)
Dropped out of treatment 8,628 (21.1) 7,916 (20.6) 712 (28.2)
Transferred to another facility 8,932 (21.8) 8,614 (22.4) 31 (12.6)
Otherb3,677 (9) 3,532 (9.2) 145 (5.7)
Service setting (discharge) 40,909 .000
Detox 7,365 (18) 6,983 (18.1) 382 (15.1)
Rehab/Residential facility 9,855 (24.1) 9,276 (24.2) 579 (22.9)
Ambulatory 23,689 (57.9) 22,122 (57.6) 1,567 (62)
Length of stay 40,909 .000
Outpatient stay (1 day) 5,117 (12.5) 4,919 (12.8) 198 (7.8)
2-30 days 15,909 (38.9) 14,902 (38.8) 1,007 (39.8)
31-90 days 8,566 (20.9) 8,090 (21.1) 476 (18.8)
91-180 days 5,808 (14.2) 5,303 (13.8) 505 (20)
181-365 days 3,326 (8.1) 3,143 (8.2) 183 (7.2)
>365 days 2,183 (5.3) 2,024 (5.3) 159 (6.3)
aNon-Hispanic: Alaska Native/American Indian, Asian/Native Hawaiian/Pacic Islander, other single race, or two or more races.
bTerminated from treatment, incarcerated, death, or other.
treatment episodes that resulted in SUD relapse, more than
half (58%) received treatment in an ambulatory setting after
discharge. Of the veteran treatment episodes that resulted in
relapse, 60% had treatment periods of 2-90 days, 14% stayed
in treatment between 91 and 180 days, and 8% stayed in
treatment 180-365 days.
MILITARY MEDICINE, Vol. 187, November/December 2022 e1425
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
TABLE II. Multivariable Logistic Regression Model (Odds Ratios
[ORs] and 95% Condence Intervals [CIs]) for the Association
between Sociodemographic and Treatment-Related Factors, and
Relapse among Veterans
Relapse
OR (95% CI) P-value
Age
35-64 Reference
18-34 0.73 (0.66, 0.82) .000
≥65 0.95 (0.75, 1.20) .655
Gender
Female Reference
Male 1.55 (1.34, 1.79) .000
Race/Ethnicity
White non-Hispanic Reference
Black non-Hispanic 0.89 (0.79, 1.01) .077
Hispanic 1.17 (0.98, 1.41) .084
Othera2.31 (1.71, 3.13) .000
Employment status (discharge)
Full-time Reference
Part-time 1.27 (1.05, 1.54) .016
Unemployed 1.92 (1.67, 2.22) .000
Not in labor force 1.29 (1.13, 1.47) .000
Education
High school Reference
No high school 1.29 (0.86, 1.93) .219
College 1.43 (1.29, 1.59) .000
Living arrangements (discharge)
Independent living Reference
Homeless 3.26 (2.55, 4.17) .000
Dependent living 0.99 (0.87, 1.12) .834
Arrests
None Reference
One 1.52 (1.19, 1.95) .007
Two or more 0.72 (0.45, 1.14) .160
Reason for discharge
Treatment completed Reference
Dropped out of treatment 2.87 (2.44, 3.39) .000
Transferred to another
facility
2.50 (2.16, 2.90) .000
Otherb2.64 (2.15, 3.24) .000
Service setting (discharge)
Rehab/Residential facility Reference
Detox 1.49 (1.23, 1.80) .000
Ambulatory 0.73 (0.63, 0.83) .000
Length of stay
2-30 days Reference
Outpatient stay (1 day) 1.13 (0.94, 1.37) .191
31-90 days 1.49 (1.30, 1.72) .000
91-180 days 1.00 (0.87, 1.16) .991
181–365 days 1.78 (1.46, 2.17) .000
>365 days 1.81 (1.39, 2.36) .000
aNon-Hispanic: Alaska Native/American Indian, Asian/Native Hawai-
ian/Pacic Islander, other single race, or two or more races.
bTerminated from treatment, incarcerated, death, or other.
Logistic Regression Analysis: Sociodemographic
and Treatment-related Predictors
Table II presents the results of the multivariable logistic
regression. The area under the receiver operating character-
istic was 0.713 and the overall rate of correct classication
was about 94%. Younger veterans between 18 and 34 years
old were 27% less likely to relapse than those who were
35-64 years old (odds ratio [OR] 0.73, 95% condence
interval [CI]: 0.66, 0.82). Male veteran discharge episodes
were associated with 1.55 (95% CI: 1.34, 1.79) higher odds
of SUD relapse upon discharge than their female counter-
parts. “Other” race/ethnicity (e.g., Alaska Native/American
Indian, etc.) were 2.31 times more likely (95% CI: 1.43, 2.34)
to relapse than White non-Hispanics. All discharge episodes
of veterans who did not have full-time employment upon dis-
charge were more likely to relapse that those who were fully
employed (part-time [OR 1.27, 95% CI: 1.05, 1.54], unem-
ployed [OR 1.92, 95% CI: 1.67, 2.22], not in labor force [OR
1.29, 95% CI: 1.13, 1.47]). College-educated veterans were
43% more likely to relapse after discharge from treatment
than those who only possessed a high school education (OR
1.43, 95% CI: 1.29, 1.59). Homeless veteran episodes were
more than three times more likely to relapse after treatment
discharge compared to those living independently (OR: 3.26,
95% CI: 2.55, 4.17). Veteran discharges that had at least one
arrest before admission were 52% more likely to relapse than
those who were never arrested (OR 1.52, 95% CI: 1.19, 1.95).
Discharges involving veterans who either dropped out of
treatment (OR 2.87, 95% CI: 2.44, 3.39), were transferred
to another facility (OR: 2.50, 95% CI: 2.16, 2.90), or left
for other reasons (terminated from treatment, incarcerated,
or death) (OR: 2.64, 95% CI: 2.15, 3.24) all had a higher
likelihood of SUD relapse upon treatment discharge than
those who successfully completed treatment. Participants who
received treatment at a 24-hour detox facility were more
likely to relapse than those who received care at a rehabili-
tative/residential treatment facility (OR 1.49, 95% CI: 1.23,
1.80); however, veterans at ambulatory healthcare facilities
fared better than those at rehabilitative/residential treatment
facilities in terms of treatment outcome. Statistically signif-
icant ndings in duration of treatment (days) revealed that
veterans who received treatment 31-90days (OR: 1.49, 95%
CI: 1.30, 1.72), 181-365 days (OR: 1.78, 95% CI: 1.46, 2.17),
and greater than 365 days (OR: 1.81, 95% CI: 1.39, 2.36) all
had higher odds of SUD relapse upon discharge, compared to
treatment for 2-30 days.
Classication and Regression Trees Analysis
The CART analysis originally produced a 59-terminal node
tree (terminal nodes are made of a group of cases that share
similar characteristics and cannot be further split); however,
for the purpose of clarity, a 6-terminal-node tree was cre-
ated (Fig. 1). This tree yielded a sensitivity of 57% and a
specicity of nearly 68% with an area under the receiver
operating characteristic of 65%. Figure 1 shows the vari-
able splits that represent six terminal nodes. Every node
displayed on the tree includes the number of veteran dis-
charge episodes and the percentage of each outcome after
discharge (None indicates no drug use at discharge vs. Sub-
stance Use which indicates substance use at discharge). The
e1426 MILITARY MEDICINE, Vol. 187, November/December 2022
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
FIGURE 1. Classication tree for veterans suffering from substance use disorders, where each node includes estimates (%) for no substance use (none) upon
discharge from treatment and substance use upon discharge from treatment.
CART analysis for veterans suffering from SUDs revealed
that homelessness upon discharge from treatment increased
the percentage of relapse among this population to almost
99% (terminal node 6) from about 94% (root node) and
MILITARY MEDICINE, Vol. 187, November/December 2022 e1427
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
FIGURE 2. Relative feature/measure importance based on the classication tree analysis of veteran relapse after discharge from treatment.
showed to be the most important predictor among all other
independent variables (Figs. 1 and 2). Among those veterans
who were not homeless (lived independently or dependently)
or veterans who were transferred to another facility to con-
tinue treatment or were discharged for “Other” reasons, nearly
96% of them relapsed at discharge (terminal node 5). How-
ever, 99% of the veterans who were not homeless, dropped out
of treatment or completed it with a length of stay ranging from
1 day, 31-90days, to 181-365 days, and received treatment at
a 24-hour detoxication facility experienced a relapse upon
discharge (terminal node 4). Receiving ambulatory or rehabil-
itative/residential SUD treatment upon discharge from initial
treatment and being unemployed or not being in the labor
force was also associated with relapse to substance use upon
discharge (94%) (terminal node 3). Among the same subgroup
of veterans receiving similar care, a slightly lower percentage
of employed veterans relapsed upon discharge from treatment
(91%) (terminal node 2). The group with the lowest rate of
relapse (90%) comprised veterans who were stably housed,
completed treatment or dropped out of treatment voluntar-
ily, and spent 2-30 days, 3-6 months, or more than a year in
treatment (terminal node 1).
In summary, the most important features as predictors of
SUD relapse generated from the classication tree analysis are
shown in the “Relative Variable Importance” chart (Fig. 2).
Relative importance is dened as a percentage improvement
to the top predictor. In this CART model, the most important
variable in predicting relapse is the veteran’s living arrange-
ment upon discharge from treatment (100%). Relative to
“living arrangements,” the second-most important variable
in predicting relapse was a veteran’s “reason for discharge”
(61%) from treatment.
DISCUSSION
The ndings of this study indicate that almost 94% of the
veterans had an SUD relapse upon treatment discharge. This
estimate is higher compared to estimates generated from
longitudinal studies that show a 76% SUD relapse rate in
veterans after receiving treatment.6Most importantly, this
study identied risk factors associated with SUD relapse
in this population using both traditional statistical methods
and classication trees. Classication tree analysis revealed
high-risk subgroups for SUD relapse that further conrmed
the importance of a veteran’s living arrangements and ser-
vice setting after SUD treatment discharge. Simply being
homeless indicated the highest likelihood of relapsing after
treatment. Although the number of homeless veterans have
decreased over the last decade,27 this does not negate the
increasing proportion of homeless veterans who live in a per-
petual recovery–relapse cycle. Chronically homeless veterans
tend to experience a broad variety of physical and mental
illnesses,10,28 which can contribute signicantly to their abil-
ity to successfully complete SUD treatment programs and
remain sober upon discharge. It is crucial that treatment pro-
grams incorporate Healthy Lifestyle Behaviors which have
been associated with lower relapse rates,8provide resources
that can lead to stable housing, and follow-up treatment to
prevent SUD relapse in this population.
Classication tree analysis also revealed a high likelihood
of SUD relapse among veterans who were not homeless;
however, they either dropped out of treatment or completed
treatment ranging from 1 day, 1-3 months, to 6-12 months,
and received treatment at a 24-hour detox facility after dis-
charge. Research evidence shows that those veterans who
dropped out of SUD treatment early were more likely to
e1428 MILITARY MEDICINE, Vol. 187, November/December 2022
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
relapse 6 months after treatment than those who nished all
treatment sessions.29 Length of treatment and type of treat-
ment facilities along with the treatment plans may need further
examination in identifying factors that increase the likelihood
of SUD relapse in veterans.
The ndings of this study also show that veteran discharges
which were either prematurely terminated, or were transferred
to another treatment location, were at a higher risk of relapse.
Treatment completion is vital for reducing relapse after dis-
charge; research evidence shows that those who complete
substance use treatment have fewer relapses and are more
likely to maintain abstinence after completing treatment.30
Transfers from detox treatment facilities typically occur when
patients have exceeded their level of treatment at that loca-
tion, thus requiring follow-on residential care in longer-term
SUD treatment facilities. Once a patient is discharged from
detox, they are at high risk of relapse and therefore vulnera-
ble to treatment incompletion and failure.31 Studies that used
VHA patient record data found that veterans who completed
treatment had a lower SUD relapse prevalence (30% vs. 54%)
than in publicly available data such as TEDS-D.6However,
VHA studies concur with the high risk of relapse during the
early stages of recovery.32 In general, studies performed with
VHA data are longitudinal studies; therefore, they tend to rep-
resent ndings that are captured over an extended period of
time.33,34 Since this is a cross-sectional study, relapse gures
are only indicative of the year they were reported (2017).
This study also demonstrates that unemployment plays a
signicant role in relapse. Logistic regression and classica-
tion tree analyses demonstrated a high likelihood of relapse
among unemployed veterans after receiving treatment for
SUD. Research evidence shows that individuals who are sud-
denly unemployed may return to substance use simply by
having more spare time that comes with job loss or by sur-
rounding themselves with people who are also chronically
unemployed.35
Veterans who are involved with the criminal justice system
represent a particularly vulnerable population who experience
high rates of both posttraumatic stress disorder and SUDs.36
The ndings of this study show that veterans with at least one
arrest within the last 30 days before admission have a greater
likelihood of SUD relapse. This indicates that treatment ser-
vices may need to be tailored to specic populations such as
those with prior involvement with the justice system.
Limitations
There are several limitations in this study. First, the facili-
ties reporting TEDS data receive state drug/alcohol agency
funds for the provision of drug/alcohol treatment services,
and no data are reported on facilities operated by federal
agencies, including the Bureau of Prisons, the Department
of Defense, and the Department of Veterans Affairs. How-
ever, some facilities operated by the Indian Health Service are
included. Another limitation is that, in many states, TEDS-
D data may include multiple discharges for the same patient.
This in turn makes it challenging to adjust the statistical anal-
yses for potential dependencies in the data. In addition, the
data in this study represent admissions and not patients. Due to
the small sample sizes for racial/ethnic groups in the “Other”
category for race/ethnicity, we were not able to conduct mean-
ingful analyses and therefore did not examine these groups
separately related to SUD relapse. Also, most of the substance
use and living arrangement data collected by facilities are
self-reported by persons admitted for treatment; thus, social
desirability bias may have affected participant responses.
This study also utilized cross-sectional data and thus causal-
ity cannot be inferred. Regardless of these limitations, this
study offers a unique perspective of correlates associated with
veteran relapse after discharge from SUD treatment using
classication trees and a national-level sample.
Implications
This study proles SUD relapse among the veteran popula-
tion in community-based treatment programs. The ndings
highlighted several risk factors that clinicians and hospital
administrators can focus on to identify veterans with a high
likelihood of relapse. These factors include homelessness and
unemployment upon discharge, arrest history, the use of a
24-hour detox facility, treatment length of stay, and the rea-
son for discharge. Unemployment and homelessness are two
of the most pervasive risk factors of substance use, even after
receiving treatment for SUDs. It is imperative for state and
federal funding to focus on treatment programs that incorpo-
rate stable housing and employment after discharge from such
programs. State programs must continue to focus efforts on
housing veterans through programs like the VHA’s Housing
First initiative. Lastly, treatment centers that provide sub-
stance use treatment must also be adequately staffed with job
placement or work therapy specialists in addition to substance
use clinicians. Future research needs to test whether apply-
ing these principles may result in lower SUD relapse among
veterans. In addition, qualitative research studies in the form
of interviews or case studies are needed as they may uncover
unobserved obstacles that preclude veterans from maintaining
sobriety well beyond treatment completion.
CONCLUSION
Substance use among veterans is already a public health con-
cern that needs to be addressed. This problem is exacerbated
by compounding other risk factors such as homelessness,
unemployment, and inadequate length and type of SUD treat-
ment. Further research should be conducted to identify what
motivates veterans to stay sober and what triggers substance
use after treatment. Studies focusing on the identication of
efcient length of stay in rehabilitative treatment based on
the patient characteristics would provide valuable informa-
tion for SUD treatment programs. Findings from this study
can be used to promote policies related to the implementa-
tion of veteran-specic treatment and prevention programs
MILITARY MEDICINE, Vol. 187, November/December 2022 e1429
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
that can increase relapse prevention efforts. The ndings of
this study also show the need for healthcare providers who
work with the veteran population to design veteran-specic
treatment plans that are specically geared to improve SUD
treatment completion rates and lower substance use after
treatment.
ACKNOWLEDGMENT
None declared.
FUNDING
None declared.
CONFLICT OF INTEREST STATEMENT
None declared.
DATA AVAILABILITY
The data underlying this article are available in the SAMHSA online data
repository at https://www.datales.samhsa.gov/dataset/teds-d-2017-ds0001-
teds-d-2017-ds0001.
REFERENCES
1. Pemberton MR, Forman-Hoffman VL, Lipari RN, Ashley
OS, Heller DC, Williams MR: Prevalence of past year sub-
stance use and mental illness by veteran status in a nationally
representative sample. CBHSQ 2016; (November): 2–4. Available at
https://www.samhsa.gov/data/sites/default/les/NSDUH-DR-Veteran
Trends-2016/NSDUH-DR-VeteranTrends-2016.pdf.
2. Boden MT, Hoggatt KJ: Substance use disorders among veterans in
a nationally representative sample: prevalence and associated func-
tioning and treatment utilization. J Stud Alcohol Drugs 2018; 79(6):
853–61.
3. Teeters J, Lancaster C, Brown D, Back S: Substance use disorders in
military veterans: prevalence and treatment challenges. Subst Abuse
Rehabil 2017; (8): 69–77.
4. Albright DL, McDaniel J, Kertesz S, et al: Small area estimation and
hotspot identication of opioid use disorder among military veter-
ans living in the Southern United States. Subst Abuse 2021; 42(1):
116–22.
5. Witkiewitz K, Marlatt GA: Relapse prevention for alcohol and drug
problems: that was Zen, this is Tao. Am Psychol 2004; 59(4): 224–35.
6. Decker KP, Peglow SL, Samples CR, Cunningham TD: Long-term
outcomes after residential substance use treatment: relapse, morbidity,
and mortality. Mil Med 2017; 182(1): e1589–95.
7. Watkins KE, Pincus HA, Paddock S, et al: Care for veterans with
mental and substance use disorders: good performance, but room to
improve on many measures. Health Aff (Millwood) 2011; 30(11):
2194–203.
8. LePage JP, Garcia-Rea EA: The association between healthy lifestyle
behaviors and relapse rates in a homeless veteran population. Am J
Drug Alcohol Abuse 2008; 34(2): 171–6.
9. Chang G, Martin KB, Tang M, Fleming JA: Inpatient hospitalization
for substance use disorders one year after residential rehabilitation:
predictors among US veterans. Am J Drug Alcohol Abuse 2016; 42(1):
56–62.
10. Montgomery AE, Hill LL, Kane V, Culhane DP: Housing chronically
homeless veterans: evaluating the efcacy of a housing rst approach
to HUD-VASH: housing chronically homeless veterans. J Community
Psychol 2013; 41(4): 505–14.
11. Dunne EM, Burrell LE 2nd, Diggins AD, Whitehead NE,
Latimer WW: Increased risk for substance use and health-related
problems among homeless veterans. Am J Addict 2015; 24(7):
676–80.
12. Simmons MM, Fincke BG, Drainoni M-L, et al: A two-state com-
parative implementation of peer-support intervention to link veterans
to health-related services after incarceration: a study protocol. BMC
Health Serv Res 2017; 17(1): 1–10.
13. Bhalla IP, Stefanovics EA, Rosenheck RA: Social determinants of
mental health care systems: intensive community based care in the
Veterans Health Administration. BMC Public Health 2020; 20(1):
1–13.
14. Blodgett JC, Avoundjian T, Finlay AK, et al: Prevalence of men-
tal health disorders among justice-involved veterans. Epidemiol Rev
2015; 37(1): 163–76.
15. Substance Abuse and Mental Health Services Administration.
Treatment Episode Data Set (TEDS): 2017 Admissions to and
Discharges from Publicly-Funded Substance Use Treatment.
Annual report. Center for Behavioral Health Statistics and Quality,
2019.
16. Center for Behavioral Health Statistics and Quality: Treatment
Episode Data Set Discharges (TEDS-D) 2017 Codebook. 2017.
https://www.datales.samhsa.gov/dataset/teds-d-2017-ds0001-teds-
d-2017-ds0001.
17. Ofce of Applied Studies: Substance Abuse and Mental Health
Services Administration: Treatment Episode Data Set (TEDS)
State Instruction Manual: Discharge Data with National Outcomes
Measures (NOMS). Department of Health and Human Services;
2010. Available at https://www.samhsa.gov/data/sites/default/les/
teds_disch_manual_feb10.pdf. February 2010; accessed May 24,
2021.
18. Department of Defense: Qualication standards for enlistment,
appointment, and induction: instruction, 1304.26. 2015. Available at
https://www.esd.whs.mil/Portals/54/Documents/DD/issuances/dodi/
130426p.pdf?ver=2018-10-26-085822-050; accessed April 15, 2020.
19. Wesson DR, Havassy BE, Smith DE: Relapse and Recovery in
Drug Abuse: Theories of Relapse and Recovery and Their Implica-
tions for Drug Abuse Treatment. Vol 72. Department of Health and
Human Services, Public Health Service, Alcohol, Drug Abuse, and
Mental Health Administration, National Institute on Drug Abuse;
1986.
20. Cronkite RC, Moos RH: Determinants of the posttreatment function-
ing of alcoholic patients: a conceptual framework. J Consult Clin
Psychol 1980; 48(3): 305–16.
21. National Institute on Drug Abuse: Principles of Drug Addiction Treat-
ment: A Research-Based Guide. 3rd ed. National Institute on Drug
Abuse, National Institutes of Health; 2018.
22. StataCorp: Stata Statistical Software: Release 16. StataCorp LLC;
2019.
23. Breiman L, Friedman JH, Olshen RA, Stone CJ: Classication and
Regression Trees. Wadsworth International Group; 1984. Available at
https://catalog.hathitrust.org/Record/000121804; accessed December
1, 2020.
24. King MW, Resick PA: Data mining in psychological treatment
research: a primer on classication and regression trees. J Consult Clin
Psychol 2014; 82(5): 895–905.
25. Ciampi A, Negassa A, Lou Z: Tree-structured prediction for cen-
sored survival data and the Cox model. J Clin Epidemiol 1995; 48(5):
675–89.
26. Minitab: Minitab 19 Statistical Software. Minitab, Inc.; 2020. Avail-
able at www.minitab.com; accessed December 2, 2020.
e1430 MILITARY MEDICINE, Vol. 187, November/December 2022
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022
Substance Use Relapse Among Veterans
27. Council of Economic Advisers (U.S.): The State of Homelessness in
America. The Council of Economic Advisers; 2019.
28. Evans WN, Kroeger S, Palmer C, Pohl E: Housing and urban
development–veterans affairs supportive housing vouchers and vet-
erans’ homelessness, 2007–2017. Am J Public Health Wash 2019;
109(10): 1440–5.
29. Wallace AE, Weeks WB: Substance abuse intensive outpatient treat-
ment: does program graduation matter? J Subst Abuse Treat 2004;
27(1): 27–30.
30. Stahler GJ, Mennis J, DuCette JP: Residential and outpatient treatment
completion for substance use disorders in the U.S.: moderation analy-
sis by demographics and drug of choice. Addict Behav 2016; 58(July):
129–35.
31. Spear SE: Reducing readmissions to detoxication: an interorganiza-
tional network perspective. Drug Alcohol Depend 2014; 137(April):
76–82.
32. Hawkins EJ, Grossbard J, Benbow J, Nacev V, Kivlahan DR:
Evidence-based screening, diagnosis, and treatment of substance use
disorders among veterans and military service personnel. Mil Med
2012; 177(suppl_8): 29–38.
33. Hagedorn HJ, Noorbaloochi S, Bangerter A, Stitzer ML, Kivlahan D:
Health care cost trajectories in the year prior to and following
intake into Veterans Health Administration outpatient substance
use disorders treatment. J Subst Abuse Treat 2017; 79(August):
46–52.
34. Moos RH, Moos BS: Treated and untreated alcohol-use disorders:
course and predictors of remission and relapse. Eval Rev 2007; 31(6):
564–84.
35. Badel A, Greaney B: Exploring the link between drug use and
job status in the U.S. Reg Econ 2013; 21(3). Available at https://
www.stlouisfed.org/~/media/les/pdfs/publications/pub_assets/pdf/
re/2013/c/drug-use.pdf.
36. Stimmel MA, Rosenthal J, Blue-Howells J, et al: The impact of sub-
stance use disorders on treatment engagement among justice-involved
veterans with posttraumatic stress disorder. Psychol Serv 2019; 16(4):
564–71.
MILITARY MEDICINE, Vol. 187, November/December 2022 e1431
Downloaded from https://academic.oup.com/milmed/article/187/11-12/e1422/6323256 by guest on 30 November 2022