Between- and within-person associations between opioid overdose risk and depression, suicidal ideation, pain severity, and pain interference

Abstract

Background:

To better understand overdose (OD) risk and develop tailored overdose risk interventions, we surveyed 234 opioid-using veterans residing in New York City, 2014-2017. Our aim was to better understand how predictors of OD may be associated with physical and mental health challenges, including pain severity and interference, depression and suicidal ideation over time.

Methods:

Veterans completed monthly assessments of the Overdose Risk Behavior Scale (ORBS), pain severity and interference, suicidal ideation, and depression for up to two years and were assessed an average of 14 times over 611 days. To estimate between-person and within-person associations between time-varying covariates and opioid risk behavior, mixed-effects regression was used on the 145-person subsample of veterans completing the baseline and at least three follow-up assessments.

Results:

The level of each time-varying covariate at the average of study time (between-person effect) was positively related to ORBS for pain severity and interference, suicidal ideation, and depression. Deviations from individuals’ personal trajectories (within-person effect) were positively related to ORBS for pain severity and interference, suicidal ideation, and depression.

Conclusions:

US military veterans endure physical and mental health challenges elevating risk for opioid-related overdose. When pain severity, pain interference, suicidal ideation and depression were higher than usual, opioid risk behavior was higher. Conversely, when these health issues were less of a problem than usual, opioid risk behavior was lower. Assessing the physical and mental health of opioid-using veterans over time may support the development and implementation of interventions to reduce behaviors that increase the likelihood of overdose.

1. Introduction

The nation remains in a public health crisis involving opioid-related morbidity and mortality (Seth et al., 2018). With their historically high rates of opioid analgesic use to treat chronic pain, U.S. military veterans represent a population at elevated risk for fatal and non-fatal overdose (OD) (Bennett et al., 2017a; Bennett et al., 2017b; Pouget et al., 2017; Wilder et al., 2016). Despite recent reductions in opioid prescribing for acute and chronic pain (Lin et al., 2017), veterans die from opioid-related OD at roughly twice the rate of the general population, (Bohnert et al., 2011; Bohnert et al., 2012) even in the midst of an opioid crisis impacting a broad cross-section of the U.S. population (Rudd et al., 2016). To help stem OD mortality rates among veterans, it is critical to understand how the physical and psychological challenges many veterans face may impact their engagement in the opioid-related behaviors known to elevate OD risk (Pouget et al., 2017).

1.1. Physical Pain and Pain Interference

Veterans’ use of opioids is often iatrogenically initiated in response to service-related injury and subsequent acute and chronic pain (Bohnert et al., 2012; Bohnert et al., 2014; Knapik et al., 2013). Chronic pain is more common and often more severe among veterans than non-veterans, with musculoskeletal pain the most common condition reported (Nahin, 2017). The National Health Interview Survey suggests that 9.1% of veterans report having severe pain, compared to 6.4% of the general population (Nahin, 2017). For veterans serving post-9/11, high rates of opioid prescribing for pain and mental health comorbidities have contributed to high rates of opioid dependence and opioid-related OD. These prescribing practices have since drawn criticism (Clark, 2002; Park et al., 2015), and prescription opioids (POs) are now less widely prescribed (Rosenberg et al., 2018; U.S. Department of Veterans Affairs, 2018). Despite such efforts, some veterans who had been prescribed POs for pain have transitioned to black-market PO or heroin use, which present additional OD risks, particularly when laced with fentanyl-class substances (Banerjee et al., 2016; Jasuja et al., 2018; Ruan et al., 2017). For some veterans, PO misuse or the use of illicit opioids can be a form of self-medication to deal with acute and chronic pain or psychological problems; however, this form of pain self-management can result in heightened OD risk due to opioids’ diminishing analgesic efficacy over time and the related potential for self-initiated dosage increases (Krebs et al., 2018).

1.2. Psychological Risk Factors and OD

Veterans also exhibit high rates of mental health disorders (e.g. depression, anxiety, suicidality, posttraumatic stress disorder) (Andrews et al., 2007; Goebel et al., 2011; Institute of Medicine, 2012; Kimerling et al., 2016; Lin et al., 2019; McCarthy et al., 2009; Vazan et al., 2013) and may use opioids and other substances as a means of coping with emotional pain and trauma. Veterans experiencing chronic non-cancer pain can also have depression, anxiety and substance use disorders (SUD), and research has shown that longer periods of ineffectively treated chronic pain are associated with a greater negative relationship between mental health disorders and physical disability (Gatchel, 2004). Certain mental health challenges that are commonly reported by veterans, such as anxiety, depression and difficulties relating to civilians (Baria et al., 2018; Schell et al., 2011), may heighten their OD morbidity and mortality risk, especially if opioid use is a solitary activity (Latkin et al., 2004) or involves concomitant use of benzodiazepine-class anti-anxiety medications (Gressler et al., 2018). Compounding these vulnerabilities, opioid-using veterans may avoid treatment for substance-related and mental health problems due to stigma associated with seeking help for personal or psychological issues (Aikins et al., 2015; Hoge et al., 2004; Kim et al., 2011), concerns about the implications of a mental health diagnosis for future employment (Brown and Bruce, 2016), as well as fear of losing opioid prescriptions.

1.3. Social Risk Factors for OD

Many social factors – such as difficulties managing relationships, finding stable housing and employment, and interactions with the legal system – can also function as risk factors for both accidental and intentional OD in some opioid-using populations. Research has found that some veterans feel socially disconnected and alienated from more affluent, non-veteran populations (Badiaga et al., 2008; Baria et al., 2018; Bennett et al., 2017a; Bennett et al., 2017b; Elliott et al., 2018). Additionally, the settings in which many unstably housed veterans reside (such as shelters, single room occupancies or transitional housing) often involve easily accessible networks of illicit substance use. Incarceration experience is common among unstably housed veterans (Cusack and Montgomery, 2017; Greenberg and Rosenheck, 2008), and may function to heighten veterans’ OD risk after release, particularly for those with mental illness and/or substance use issues (Binswanger et al., 2012). Veterans also report difficulties in obtaining and maintaining employment, in part due to mental health challenges (Adler et al., 2011). Additionally, recent research conducted in NYC found residents of very high poverty neighborhoods had a higher rate of overdose death (33.8 per 100,000) than residents of high, medium, and low poverty neighborhoods (19.0, 14.8, and 14.7 per 100,000 respectively; NYC Office of the Chief Medical Examiner and DOHMH, 2018).

The objective of this study is to assess predictors of opioid overdose-related risk behavior among a community sample of post-9/11 U.S. military veterans recruited in New York City. Our aim is to better understand how physical and mental health challenges, including pain severity and interference, depression and suicidal ideation may be associated with overdose risk behaviors over time. The study’s distinctive focus on tracking veterans’ prospective overdose risk, in an urban, community-based setting contributes to a broader and more comprehensive understanding of veterans’ overdose risk behavior.

2. Method

2.1. Participants

Participants screened as eligible were post-9/11 military veterans who reported any licit or illicit opioid use (e.g., POs, heroin, methadone) within the 30 days prior to enrollment. Participants were recruited throughout NYC using venue-based (e.g. homeless shelters, veteran-specific residences) and chain-referral sampling methods. Veteran status was established via DD-214 (DoD discharge form) and/or VA or veterans’ housing identification. For those lacking these forms of identification, including year of separation, staff (three of whom were themselves post-9/11 veterans) queried them about their Military Occupational Specialty (MOS), boot camp location, and service/deployment experiences. Those unable to respond to questions involving military jargon or specifics about their service were not enrolled in the study, thanked for their time, and given a subway fare card. In total, 234 participants were enrolled between August 2014 and June 2016.

2.2. Procedures

After providing written informed consent, participants completed a roughly 2-hour baseline assessment about their military service history, alcohol, PO and other drug use, current pain, housing, relationship and mental health status, including significant life events experienced in the past 30 days. Baseline assessments were administered face-to-face, with a trained and experienced interviewer entering data on a tablet computer. During enrollment, participants also completed their first monthly assessment of past 30-day OD risk behaviors and biopsychosocial predictors of OD risk. Participants received $60 at the completion of the baseline assessment and first monthly OD risk survey and were compensated $20 for each monthly follow-up assessment completed over 24 months. After enrollment, monthly SMS and emails were sent to participants at 30-day intervals. Using links sent via SMS and email, participants completed monthly surveys on phones or internet enabled computers (home, libraries, shelters, public internet kiosks). Monthly surveys took ~ 20 minutes to complete. Procedures were approved by the Institutional Review Board of the host institution.

2.3. Measures

2.3.1. Overdose Risk Behaviors Scale (ORBS)

Opioid overdose risk behavior was measured using 24 items from the ORBS instrument (Pouget et al., 2017). Items were grouped into five subscales: adherence; alternate administrations; solitary use; non-prescribed use; and concurrent use subscale. Individual items asked about the number of days in the past 30 on which the risk behavior occurred. Subscale scores were calculated by averaging the items within each scale, and a total score was calculated by averaging the five subscale scores. As reported in earlier work (Pouget et al., 2017), ORBS subscales were positively correlated (Pearson correlation range: .45 - .72) and the total scale score had good internal consistency reliability (Cronbach’s alpha = .84). Strong positive correlations between the total scale score and both loss of consciousness (r = .56) and calling for medical help (r = .44) were observed, suggesting the total scale score is related to actual overdose events. To avoid multiple testing issues, we used the ORBS total score as an overall summary of overdose risk and focus on that score in analysis.

2.3.2. Patient Health Questionnaire (PHQ)-9

Severity of depression was measured using the Patient Health Questionnaire - PHQ-9 (Kroenke et al., 2001). Nine items asked about the frequency of symptoms of depression within the past two weeks and were scored from 0 (not at all) to 3 (nearly every day). The nine items were summed, yielding a score which ranged from 0 to 27.

2.3.3. Suicidal Ideation

Eight suicidal ideation items were scored using a proportion of maximum possible (POMP) approach (Cohen et al., 1999). The eight items were the following: 1) “Have you wished you were dead or wished you could go to sleep and not wake up? Or have you thought that it might better if you weren’t alive anymore?”; 2) “How often have you had these thoughts in the last 30 days?” [less than once a week; once a week; 2-5 times a week; daily or almost daily; many times each day]; 3) “Have you actually had any thoughts of killing yourself?”; 4) “How often have you had these thoughts in the last 30 days?” [less than once a week; once a week; 2-5 times a week; daily or almost daily; many times each day]; 5) “Have you been thinking about how you might do this?”; 6) “Have you had these thoughts and had some intention of acting on them?”; 7) “Have you started to work out or worked out the details of how to kill yourself?”; and 8) “Do you intend to carry out this plan?”. A proportion score of zero indicates no thoughts or wishes to be dead in the past 30 days. A proportion score of one, the highest possible, indicates frequent suicidal ideation with a detailed plan and an intention to carry out the plan. To facilitate interpretation of coefficients in regression analysis, the proportion score was transformed to a 0-10 scale by multiplying the proportion by 10.

2.3.4. Brief Pain Inventory

Pain severity and pain interference were measured with the Brief Pain Inventory (Cleeland, 2014; Cleeland and Ryan, 1994). Three items (worst, least, and average pain in the past month) were averaged to create a pain severity score. Pain interference (the extent to which pain interferes with various domains of functioning in the past month) was measured with the 7-item BPI pain interference sub-scale. Scores for both the severity and interference scales could range from 0 to 10.

2.4. Analysis Strategy

Of 234 recruited veterans, 145 completed an initial risk assessment and at least three follow-ups over time. Because the primary goal of the analyses is to estimate between-person and within-person associations with opioid risk behavior, only the 145 veterans with a baseline and at least three follow-up assessments were included in longitudinal analysis. Among the 145 veterans in the longitudinal analysis sample, the average number of assessments over time was 14 (sd = 6.2; maximum = 25). The final assessment among members of the longitudinal sample was an average of 611 days after baseline (median = 710; sd = 162; minimum = 166; maximum = 735).

Linear mixed-effects regression models were used to estimate quadratic growth models for opioid risk behavior and each of four time-varying covariates (depression severity, suicidal ideation, pain severity, and pain interference). These growth models describe the average trajectory over the study period. Variables representing the linear and quadratic components of trajectories were formed by applying an orthogonal polynomial transformation to years since baseline. In each growth model, fixed effects for the intercept and linear and quadratic components of time also had corresponding random effects, and correlations among random effects were estimated. Our primary motivation for growth modeling was detrending. Prior to specifying quadratic growth models, we visualized the individual trajectories of all veterans in the analysis sample for overdose risk, depression, suicidal ideation, pain severity and pain interference. We wanted to get a sense of how complex the growth pattern needed to be to capture the broad strokes of trends over the study period. Based on the individual trajectories, we determined that quadratic growth models with random effects for intercepts, linear change, and curvature fit our detrending purpose well.

Linear mixed-effects regression models also were used to estimate the between-person and within-person effects of time-varying covariates on opioid risk behavior. Following Curran and Bauer (Curran and Bauer, 2011), a quadratic growth model was fit to each veteran in the longitudinal sample. From these quadratic growth models, the intercept coefficient, which captures differences among veterans in the level of the growth trajectory at the average of study time (about 11 months after baseline), was used to represent the between-person effect of the time-varying covariate. Also from the individual quadratic growth models, residuals, which capture deviations of each veteran from their own growth trajectory for the time-varying covariate, were used to represent the within-person effect of the time-varying covariate. This approach to separating between- and within-person effects is called “detrending” and takes into account the fact that the time-varying covariate may undergo systematic change over the study period. We assumed missing data were missing at random and confidence intervals for parameters in all models were calculated using the profile likelihood method. Pseudo-standardized regression coefficients (β) were calculated as described by Hoffman (Hoffman, 2015). All analysis was conducted in the R statistical computing environment (R Core Team, 2018), and the lme4 R package (Bates et al., 2014) for linear mixed-effects regression models.

3. Results

3.1. Sample Characteristics

As shown in Table 1, most veterans were male (85%), racial/ethnic minorities (84%), not employed (76%), not married (86%), had attended some college (61%), and were an average of 37 years old (sd = 9.6). At baseline, the average overdose risk was 3.33 (sd = 4.24), the average depression was 9.27 (sd = 6.87), the average suicidal ideation was 0.96 (sd = 1.98), the average pain severity was 5.16 (sd = 2.16), and the average pain interference was 4.53 (sd = 2.56). Deviation absolute values indicate the typical size of deviations around each subject’s quadratic growth curve; in other words, deviation absolute values describe the spread of each subject’s repeated measures around his/her own modeled growth trajectory over longitudinal follow-up. For example, overdose risk, as measured by the ORBS Total score, was about one scale point higher or lower than each subject’s model-predicted overdose risk trajectory, an indication of within-subject variability in overdose risk over time.

Table 1.

Characteristics of Participants

	Longitudinal Analysis Sample						p
	No (n=89)		Yes (n=145)		Total (n=234)
	n / mean	% / (SD)	n / mean	% / (SD)	n / mean	% / (SD)
Male†	78	88.6	120	82.8	198	85.0	.304
Age at Baseline	35.91	(10.69)	37.16	(8.80)	36.69	(9.55)	.332
Race/Ethnicity†							.057
Hispanic	15	17.0	36	24.8	51	21.9
Non-Hispanic Black	48	54.5	88	60.7	136	58.4
Non-Hispanic White	20	22.7	18	12.4	38	16.3
Non-Hispanic Other	5	5.7	3	2.1	8	3.4
Some College†	51	58.0	91	63.2	142	61.2	.512
Marital Status†							.283
Single	52	59.1	70	48.6	122	52.6
Separated/Divorced	26	29.5	51	35.4	77	33.2
Married	10	11.4	23	16.0	33	14.2
Employed†	25	28.4	30	20.8	55	23.7	.247
Opioid Use Past Year at Baseline							.588
Prescription Only	59	66.3	99	68.3	158	67.5
Heroin Only	5	5.6	6	4.1	11	4.7
Prescription and Heroin	24	27.0	40	27.6	64	27.4
Methadone Only	1	1.1	0	0.0	1	0.4
Overdose Risk Behavior Subscales at Baseline (0-30)
Adherence	5.81	7.64	5.31	7.69	5.50	7.66	.626
Alternative Administration	0.81	1.92	0.96	3.98	0.90	3.35	.734
Solitary Use	3.91	6.58	4.14	7.30	4.05	7.02	.807
Non-prescribed Use	3.84	4.06	3.99	4.99	3.93	4.65	.819
Concurrent Use	2.03	3.15	2.62	4.03	2.40	3.72	.238
Overdose Risk Behavior Scale (ORBS) Total (0-30)
Baseline	3.15	(3.55)	3.44	(4.62)	3.33	(4.24)	.614
Deviation Absolute Value	--	--	1.02	(1.01)
Patient Health Questionnaire (PHQ)-9 Severity (0-27)
Baseline	9.26	6.70	9.27	(7.00)	9.27	(6.87)	.993
Deviation Absolute Value	--	--	2.39	(1.48)
Suicidal Ideation (0-10)
Baseline	1.13	2.18	0.85	(1.84)	0.96	(1.98)	.300
Deviation Absolute Value	--	--	0.51	(0.67)
Pain Severity (0-10)
Baseline	4.95	2.14	5.29	(2.16)	5.16	(2.16)	.252
Deviation Absolute Value	--	--	0.80	(0.40)
Pain Interference (0-10)
Baseline	4.48	2.58	4.56	(2.56)	4.53	(2.56)	.808
Deviation Absolute Value	--	--	0.81	(0.43)

^†n and percent; all other variables are described with mean and (SD).

Categorical variables compared with Chi-Squared and continuous variables with independent-samples t-tests.

Veterans included in the longitudinal analysis (n=145) as well as veterans with too few longitudinal observations for inclusion (n=89) were compared on categorical variables with Chi-squared and on continuous variables with independent-samples t-tests. Veterans with and without at least four longitudinal visits were not significantly different on any of the baseline characteristics described.

3.2. Growth Models

Estimates from mixed-effects regression models for change (i.e., growth models) show overdose risk behavior and each potential predictor decreased over the study period, and there was significant curvature in average trajectories for overdose risk behavior, pain severity, and pain interference (Table 2). The fixed-effects estimates indicate the average trajectory for overdose risk decreased from about 3.2 to 2.1 over the study period. The average depression severity trajectory decreased from 9.7 to 8.1, suicidal ideation decreased from 0.99 to 0.70, and both pain severity and interference decreased from 4.9 to 4.3.

Table 2.

Growth Models for Overdose Risk Behavior and Time-Varying Covariates

	ORBS Total	Depression Severity	Suicidal Ideation	Pain Severity	Pain Interference
	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)
Fixed-Effects
Intercept	2.40 (1.89, 2.91)	8.51 (7.50, 9.52)	0.74 (0.54, 0.94)	4.22 (3.88, 4.57)	3.94 (3.55, 4.32)
Linear	−15.53 (−24.15, −6.79)	−22.87 (−37.24, −8.43)	−4.27 (−7.67, −0.86)	−9.25 (−14.39, −4.06)	−7.28 (−12.73, −1.75)
Quadratic	6.83 (0.31, 13.32)	10.16 (−1.28, 21.59)	2.59 (−0.06, 5.24)	9.55 (5.33, 13.75)	6.25 (1.73, 10.72)
Variance Components†
Intercept (I)	3.06 (2.71, 3.45)	6.08 (5.39, 6.86)	1.20 (1.06, 1.36)	2.07 (1.83, 2.33)	2.34 (2.08, 2.63)
Linear (L)	45.56 (38.31, 53.53)	70.11 (58.12, 83.47)	15.32 (12.22, 18.61)	25.96 (22.00, 30.40)	27.60 (23.28, 32.43)
Quadratic (Q)	31.99 (26.33, 38.10)	52.20 (41.37, 63.72)	10.01 (7.06, 12.91)	20.60 (17.01, 24.49)	22.12 (18.23, 26.33)
I with L	0.01 (−0.19, 0.21)	0.46 (0.27, 0.62)	0.20 (−0.03, 0.42)	0.42 (0.24, 0.58)	0.35 (0.16, 0.51)
I with Q	−0.03 (−0.23, 0.18)	−0.33 (−0.53, −0.11)	−0.10 (−0.38, 0.19)	−0.54 (−0.69, −0.36)	−0.49 (−0.65, −0.31)
L with Q	−0.72 (−0.85, −0.55)	−0.42 (−0.64, −0.16)	−0.61 (−0.92, −0.27)	−0.40 (−0.59, −0.17)	−0.38 (−0.58, −0.16)
Residual	1.75 (1.69, 1.81)	3.40 (3.28, 3.51)	0.98 (0.95, 1.01)	1.10 (1.06, 1.14)	1.15 (1.12, 1.19)

^†Variance components are given as standard deviations and correlations.

The variance components in Table 2 describe how individual veterans deviate from the average trajectory for overdose risk and each time-varying covariate. Because an orthogonal polynomial transformation was applied to time, it is useful to visualize individual differences in trajectories of overdose risk and time-varying covariates to get a sense of deviations around the average trajectory. Figure 1 shows quadratic growth models for overdose risk behavior and each potential predictor of overdose risk behavior. Trajectories for individual veterans are plotted in black with alpha transparency. When there is little overlap, the trajectory for an individual veteran appears as light grey, and when many veterans have a similar trajectory, those trajectories appear as black (e.g., many veterans with consistently low suicidal ideation). Average trajectories are shown in bold blue, with 95% confidence intervals in lighter blue.

Figure 1.

Estimated quadratic growth trajectories for individual veterans (light grey with alpha transparency) and average trajectories (black) with 95% confidence intervals (dashed lines).

3.3. Between-Person Associations

Estimates of between-person associations between time-varying covariates and overdose risk behavior in mixed-effects regression models were consistently positive (Table 3). Figure 2a shows expected changes in overdose risk behavior for deviations from the sample average of the time-varying predictor around the middle of the follow-up period. For depression severity, veterans whose depression growth trajectory at the average of study time was one unit higher were expected to have overdose risk behavior approximately one-quarter of a point higher on the ORBS total score (B = 0.24, t(148) = 7.21, p < .001; β = .48). For suicidal ideation, veterans whose growth trajectory at the average of study time was one unit higher were expected to have overdose risk behavior eight-tenths of a point higher (B = 0.80, t(151) = 4.27, p < .001; β = .30). Veterans whose pain severity growth trajectory was one unit higher at the average of study time were expected to have higher overdose risk behavior (B=0.61, t(147) = 5.81, p < .001; β = .42). Similarly, veterans with higher pain interference at the average of study time were expected to have higher overdose risk behavior (B=0.54, t(147) = 5.77, p < .001; β = .41).

Table 3.

Associations Between Time-Varying Covariates and Overdose Risk Behavior

	Depression	Suicidal	Pain	Pain
	Severity	Ideation	Severity	Interference
	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)	Estimate (95% CI)
Fixed-Effects
Intercept	2.31 (1.88, 2.75)	2.34 (1.86, 2.82)	2.31 (1.85, 2.77)	2.28 (1.82, 2.74)
Linear	−16.01 (−24.70, −7.21)	−15.51 (−24.19, −6.71)	−16.01 (−24.69, −7.22)	−15.82 (−24.51, −7.02)
Quadratic	(−0.13, 6.04 12.18)	6.28 (−0.07, 12.60)	4.24 (−2.02, 10.48)	4.51 (−1.65, 10.65)
Time-varying Covariate Average	0.24 (0.17, 0.30)	0.80 (0.43, 1.17)	0.61 (0.40, 0.82)	0.54 (0.35, 0.73)
Time-varying Covariate Deviation	0.11 (0.09, 0.14)	0.29 (0.21, 0.37)	0.30 (0.23, 0.37)	0.36 (0.30, 0.43)
Variance Components†
Intercept (I)	2.60 (2.29, 2.93)	2.89 (2.55, 3.25)	2.76 (2.44, 3.11)	2.77 (2.44, 3.11)
Linear (L)	46.27 (39.06, 54.22)	46.12 (38.87, 54.10)	46.23 (39.00, 54.19)	46.38 (39.18, 54.31)
Quadratic (Q)	29.67 (24.18, 35.54)	30.96 (25.36, 36.99)	30.23 (24.71, 36.14)	29.74 (24.28, 35.59)
I with L	−0.03 (−0.23, 0.17)	−0.01 (−0.20, 0.19)	−0.08 (−0.27, 0.12)	−0.07 (−0.26, 0.13)
I with Q	0.03 (−0.19, 0.24)	0.02 (−0.19, 0.23)	0.10 (−0.12, 0.31)	0.11 (−0.10, 0.32)
L with Q	−0.70 (−0.84, −0.52)	−0.72 (−0.85, −0.55)	−0.73 (−0.86, −0.56)	−0.72 (−0.85, −0.54)
Residual	1.71 (1.65, 1.77)	1.73 (1.67, 1.79)	1.72 (1.66, 1.78)	1.70 (1.64, 1.76)

^†Variance components are given as standard deviations and correlations.

Figure 2.

Estimated a) between-person and b) within-person associations of time-varying covariates with ORBS total score (black) and 95% confidence intervals (dashed lines).

3.4. Within-Person Associations

Estimates of within-person associations between time-varying covariates and overdose risk behavior in mixed-effects regression models, like between-person associations, were consistently positive (Table 3). Figure 2b shows expected changes in overdose risk behavior for deviations from an individual’s growth trajectory in a time-varying predictor. An increase of one unit of depression severity above an individual’s expected value based on quadratic growth in depression was expected to increase overdose risk behavior by approximately one-tenth of a point (B = 0.11, t(1796) = 9.51, p < .001; β = .21). An increase of one unit above an individual’s expected value based on quadratic growth in suicidal ideation was expected to increase overdose risk behavior by approximately one-third of a point (B = 0.29, t(1795) = 6.93, p < .001; β = .15). A higher deviation from an individual’s expected value based on quadratic growth in pain severity also was expected to increase overdose risk behavior (B = 0.30, t(1783) = 8.39, p < .001; β = .19). Similarly, a higher deviation from an individual’s expected value based on quadratic growth in pain interference was expected to increase overdose risk behavior (B = 0.36, t(1784) = 10.71, p < .001; β = .24).

4. Discussion

This analysis focused on overdose risk behaviors over time and their relation to pain and mental health challenges. Veterans whose depression, suicidal ideation, pain interference, or pain severity was substantially higher than the sample average in the middle of the follow-up period (between-person association) had the equivalent of an additional day or more of overdose risk behaviors in the past month. Also, when veterans had depression, suicidal ideation, pain interference or pain severity that was substantially higher than their own usual level (within-person association), this too was associated with the equivalent of an additional day or more of overdose risk behaviors in the past month. Among a population already at high risk of opioid-related OD, the addition of even a single day of risk within a month (from a sample Total ORBS mean of 3.44) entails an appreciable increase in the hazard of OD mortality. Results point to the potential for OD risk to emerge as a complex biopsychosocial process involving maladaptive coping to challenging physical and psychosocial events and circumstances experienced by veterans in changing degrees over time.

Findings indicate that physical pain severity and interference, depression, and suicidal ideation are associated with higher OD risk behaviors, suggesting a need for ongoing assessment and treatment among opioid-using veterans in accessible and welcoming community settings, especially for those veterans who do not utilize the VHA. Physical pain and mental health concerns can be entangled. For example, long-term chronic pain can potentially exacerbate depressive symptoms and, ultimately, suicidal ideation, or severe depression may aggravate physical pain (Baria et al., 2018). As indicated by research, social factors can also exacerbate physical and mental health concerns and can serve to increase overdose risk for some who are using opioids to manage various forms of pain or to manage opioid dependency (Baria et al., 2018; Bennett et al., 2017a; Bennett et al., 2019; Gatchel, 2004). Incorporating discussion of these topics into clinical interviews with veterans who use opioids may have the potential to destigmatize opioid use disorder. Self-medicating behaviors related to depression or chronic pain, while fundamentally unsafe, may provide a more culturally sensitive and clinically appropriate platform for discussions of self-care and risk reduction than abuse and addiction. Non-judgmental discussions with veterans and among veterans with shared experiences in comforting settings hold potential to increase awareness and educate other veterans on evidence-based non-pharmacological forms of care and healing including mindfulness, cognitive behavioral therapy, and eye movement desensitization and reprocessing (EMDR).

Veterans Health Administration (VHA) has taken robust steps to stem opioid-related OD among veterans who utilize their services, including the implementation and expansion of the overdose education and naloxone distribution (OEND). VHA has developed programs across the country to monitor patients at high risk for OD (Oliva, Elizabeth M et al., 2017; Oliva, Elizabeth M. et al., 2017). VHA has also implemented services to help veterans deal with pain, mental health co-morbidities and social inclusion. Considering only half of US military veterans are enrolled in VHA, and fewer utilize VHA as patients (Bagalman, 2014), it is critical to expand the scope of services to assist veterans who do not utilize VHA with access to low-threshold services. The OEND model is based on the pioneering work of many community-based naloxone distribution programs which have proliferated across the nation (Wheeler et al., 2015). These low-threshold programs often provide additional services including mental health, legal, employment and social services, increase motivation to engage in such services through motivational and cognitive-behavioral based treatments (Britton et al., 2012; Karlin et al., 2012) and provide an opportunity to reduce overdose risk behaviors themselves (Bohnert et al., 2016; Coffin et al., 2017).

These services hold great potential to augment and complement the services offered by VHA. Study findings suggest that overdose risk reduction may require more holistic understanding of individual health care and substance abuse treatment needs. “Wrap-around” interventions (O’Toole et al., 2003; Pringle et al., 2002; Rivers, 1998; Smelson et al., 2013) and forms of outreach that coordinate different healthcare and social service modalities through an active dialogue among practitioners about the unique needs of individuals provide an opportunity to intervene. This approach has been reinvigorated through advances in the patient-centered medical home model (Ferrante et al., 2010; Jackson et al., 2013; Klein, 2011; Nelson et al., 2014; Nutting et al., 2009; Rosland et al., 2013; True et al., 2013), an approach to health care that involves both dialogue among care and service providers and the central role of a patient care coordinator (Perlin et al., 2004; Pham et al., 2009), which may involve peers who share critical life experiences with veterans (Ashbury et al., 1998; Barber et al., 2008; Norris et al., 2006).

Veterans at risk may be more receptive to peer-based approaches to addressing substance use/mental health, housing, and employment concerns (Chinman et al., 2017; Chinman et al., 2015; Hebert et al., 2008). Furthermore, peer navigation can be an effective and empowering extension of wraparound services and the patient-centric home model (Webel et al., 2010). Such a veteran-led health outreach model can be a means to more broadly deliver OEND and other services to veterans who do not utilize VHA and/or are disconnected from other services throughout the community. This model could also help connect veterans to programs that assist with non-pharmacological management of physical and mental pain as well as social challenges.

4.1. Limitations

This study is limited by the nature of self-reported data, which may understate the extent of socially undesirable behavior or suffer from issues related to recall. There are other fixed and time-varying variables potentially relevant to overdose risk behavior that were not considered in our analysis (e.g., social factors). It also should be noted that the study addresses overdose risk behaviors, not fatal and non-fatal overdoses. The relatively small, and regionally defined sample is noteworthy for its disproportionate number of urban, disadvantaged, unstably housed, and predominately minority veterans. This limits generalization to the broader population of veterans in the US, although following and collecting monthly data from a non-VA urban sample of veterans who use opioids is also a strength of the work. Another potential limitation is selection bias due to missing follow-up data. However, in Table 1, we compared veterans included in the longitudinal analysis sample (n=145) with veterans who had too few assessments overtime (n=89) on all variables summarized. None of the variables in Table 1 was significantly different among those who did, or did not complete at least four longitudinal assessments. Nevertheless, veterans excluded from analysis or contributing fewer observations over time due to missing follow-ups may be different in ways we could not measure or detect.

5. Conclusions

US military veterans endure myriad biopsychosocial challenges elevating risk for opioid-related overdose. When pain severity, pain interference, suicidal ideation and depression were higher than usual relative to each individual’s quadratic growth trajectory, overdose risk behavior was higher. Likewise, when these health issues were less of a problem than usual, overdose risk behavior was lower. Understanding military veterans’ health challenges in combination over time may inform the development of tailored, holistic interventions that could have a broad reach and impact.

Highlights:

• Depression was positively related to opioid-related overdose risk over time.

• Suicidal Ideation was positively related to overdose risk over time.

• Pain severity and interference were positively related to overdose risk over time.