Association between Discontinuing Chronic Opioid Therapy and Newly Diagnosed Substance Use Disorders, Accidents, Self-Inflicted Injuries and Drug Overdoses within the Prescribers’ Health Care System: A Retrospective Cohort Study

Abstract

Background and aim:

Prescribers are commonly confronted with discontinuing opioid therapy among patients prescribed chronic opioid therapy (COT). This study aimed to measure the association between discontinuing COT and diagnoses of substance use disorders (SUDs) and opioid-related adverse outcomes (AOs).

Design:

Retrospective cohort study

Setting:

U.S. Veterans Healthcare Administration

Participants:

Veterans with chronic pain on COT who discontinued opioid therapy were compared with those continuing COT using data from fiscal years 2009–2015.

Measurements:

Newly diagnosed substance use disorders (SUD composite; individual types: opioid, non-opioid drug, and alcohol use disorders) and opioid-related adverse outcomes (AO composite; individual types: accidents resulting in wounds/injuries, opioid-related accidents/overdoses, alcohol and non-opioid medication-related accidents/overdoses, self-inflicted injuries, and violence-related injuries) were evaluated. Primary analyses were conducted using 1:1 matching of discontinuers with those continuing COT based on propensity score and index date (±180-day window). Sensitivity analyses were conducted using logistic regressions with stabilized inverse probability of treatment weighting (SIPTW) and instrumental variable (IV) models.

Findings:

A total of 15,695 (75.4%) and 17,337 (76.6%) discontinuers were matched with those continuing COT among the cohorts testing SUD and AO development respectively. In the primary propensity score matched analyses, the composite SUD outcome was not different between discontinuers and those continuing COT (odds ratio (OR)=0.932, 95% confidence interval (CI): 0.850, 1.022). The composite AO outcome was lower among discontinuers (OR=0.660, 95%CI: 0.623, 0.699) compared with those continuing COT. SIPTW analyses found lower SUD (OR=0.789, 95%CI: 0.743, 0.837), and AO (OR=0.660, 95%CI: 0.623, 0.699) rates among discontinuers. IV models found mixed and sometimes contradictory results.

Conclusions:

Discontinuing patients from chronic opioid therapy appears to be associated with decreased diagnoses for opioid-related adverse outcomes. The association with substance use disorders appears to be inconclusive.

Introduction

Due to rising opioid overdose rates, national and state-level policies are being implemented to curb opioid prescribing. On a national level, the Centers for Disease Control and Prevention, in 2016, issued opioid prescribing guidelines for primary care settings that emphasize other means of managing chronic pain. (1) States are also increasing the documentation requirements for initiating and continuing patients on chronic opioid therapy (COT). (2,3) Some insurance companies and statewide policy efforts are limiting the number of days’ supply of initial opioid prescriptions. (4,5) Collectively, these initiatives influence prescribers to limit COT prescribing including discontinuing opioids among current COT patients.

Similar initiatives are being undertaken in the VHA. Due to the drastic rise in opioid prescribing in the past decade, the VHA launched the Opioid Safety Initiative (OSI) in 2013. (6) The OSI developed and disseminated many educational materials for opioid use and tapering as well as provided a tool for identifying patients on COT. (6) The VHA also created the Stratification Tool for Opioid Risk Mitigation (STORM), which is a predictive model for identifying Veterans at high risk for events related to opioid overdose and suicide. (7)

The literature on discontinuing opioid therapy among chronic pain patients on COT is limited. Studies of patients with COT discontinuation have noted high rates of suicidal ideation, self-directed violence, (8) substance use disorder (SUD)-related adverse events, (9), and death due to overdose and suicide. (10) However, many of these studies are limited to one location or a limited number of patients. Therefore, it is not clear if discontinuing COT among chronic pain patients poses an increase or reduction in likelihood of engagement in care for various SUDs and other potentially opioid-related adverse outcomes (AOs).

The purpose of this study is to evaluate the association between discontinuing COT, compared to continuing COT, and diagnoses of SUDs and AOs within a national Veteran cohort with chronic, non-cancer pain (CNCP) who have newly initiated COT. We hypothesized that newly diagnosed SUDs [opioid use disorder (OUD), non-opioid drug use disorder (DUD), alcohol use disorder (AUD)] and AOs [accidents resulting in wounds or injuries, self-inflicted injuries, opioid-related accidents and overdoses, alcohol and non-opioid drug-related accidents and overdoses, and violence-related injuries] would be significantly lower among those discontinuing opioid therapy than among those continuing COT. Hypothesized reductions in diagnoses may reflect a combination of reduced occurrence of new SUDs and AO’s and/or reduced visits within the health care system that was prescribing the opioids thereby decreasing recognition of these diagnoses.

Methods

Data Source

Inpatient, outpatient, demographic, and outpatient pharmacy files from the Corporate Data Warehouse (CDW) of the Veterans Health Administration (VHA) were used from the fiscal years of 2009–2015. The study was approved by the Central Arkansas Veterans Healthcare System Institutional Review Board. A protocol outlining these study aims and an overview of the methods were pre-specified in our grant application and Institutional Review Board application; however, the protocol was not made available in a publicly accessible site prior to study execution and these findings should be considered exploratory.

Study Design and Subjects

This was a retrospective cohort study of Veterans with CNCP prescribed COT. Veterans with a diagnosis of at least one CNCP condition (arthritis, back pain, neck pain, neuropathic pain, or headache/migraine) and on COT were identified. (11) The VHA Drug Class Code CN101 (Opioid Analgesics) was used to identify opioids, and COT was defined as receiving at least a 90 days’ supply of non-parenteral opioids within any 180-day period with no more than a 30 day gap in supply. (12)

Main Independent Variable

After the qualifying 180-day period in which Veterans were first determined to be on COT, Veterans were followed for an additional 180-day period to identify those Veterans that either continued COT or discontinued opioid therapy. Those continuing COT were required to meet the original COT definition in the second 180-day period. Those discontinuing opioid therapy were those Veterans who did not have any opioid prescription fills in the second 180-day period. (12) The index date was defined as the first day of the second 180-day block. The index date had to occur between October 1, 2009 and October 1, 2014 to ensure data availability in the year before and after the index date for baseline covariate and outcome assessment respectively. We used VHA CDW data from October 1, 2008 to September 30, 2015 (i.e., fiscal years 2009–2015). See eFigure 1 for a visual representation of the cohort.

eFigure 1:

Study Design and Time Frame

Exclusion Criteria

This study focused on adults newly initiating COT with reliable opioid prescription data who regularly sought care either within the VHA system or paid for by the VHA and who did not have a history of cancer or terminal illnesses. We focused this study on patient newly initiating COT to reliably ascertain the duration of COT and to limit unmeasurable confounding that may be introduced by including patients with varying lengths of prior COT. Ten exclusion criteria were implemented based on the CDW records observed in the 12 months prior to the index date (unless noted otherwise): (1) index date before 10/1/2009 or after 10/1/2014 to ensure a year of data before and after the index date for baseline covariates and follow-up, (2) ≤18 years of age at the index date as VHA does not allow research on those <18, (3) diagnosis for cancer (except for non-melanoma skin cancer) as we are evaluating opioid therapy for CNCP, (4) lacking at least 2 visits at least 30 days apart to any VHA facility or facility outside VHA where the care was paid for by the VHA to better ensure capture of care should it have been received, (5) more visits with providers outside the VHA than with VHA providers, as measured using the fee-for-service files for care obtained outside VHA, to also ensure capture of data on care received, (6) receipt of hospice/palliative care as our focus is on CNCP, (7) death in the 180-day period after the index date to ensure discontinuation of opioid therapy was not due to death, (8) potentially erroneous opioid prescription records [unable to calculate morphine milligram equivalents (MME), average daily dose above 1000 MMEs, or prescription quantity greater than 1000 units] or injectable opioids (unable to calculate MME) in the 180 days before the index date, (9) fewer than 2 pain scores in the 180 days prior to the index date with at least 1 of the pain scores being either on or within 90 days prior to the index date to adjust for pain characteristics, specifically changes in pain scores as this is likely an important confounder to discontinuing COT and AO/SUD development, and (10) switched to intermittent opioid therapy, defined as having received an opioid prescription in the 180-day period after the index date that did not meet the COT definition, as this is a comparison of discontinuers to those continuing COT. After implementing these exclusion criteria (Figure 1), we created two cohorts to assess newly diagnosed SUDs and AOs each respectively. With the cohort to assess newly diagnosed SUDs, we excluded patients with a SUD diagnosis or had a visit to an Opioid Replacement Therapy clinic (i.e., methadone clinic) in the 12 months prior to the index date. With the cohort to assess newly diagnosed AOs, we excluded patients with an AO diagnosis in the 12 months prior to the index date. These additional criteria and the creation of two cohorts were implemented to ensure that the analysis could detect new onset of SUD and AO after discontinuation among those at high risk for the outcomes (e.g., opioid overdoses among those patients with OUD).

Figure 1.

Derivation of the study cohorts

Study Outcomes

Opioid-Related Adverse Outcomes

Study outcomes were evaluated during the 12-month period after the index date. AOs were based on definitions by Seal et al. (13) using International Classification of Diseases, 9^th Revision, Clinical Modifications (ICD-9-CM codes) for accidents resulting in wounds/injuries, opioid-related accidents and overdoses, alcohol and non-opioid, drug-related accidents and overdoses, self-inflicted injuries, and violence-related injuries. AOs were assessed as a composite measure, as in the patient received at least one ICD-9-CM code for one of the five AO types in the 12-month period after the index date, and then individually for each of the five AOs.

Substance Use Disorders

SUDs were assessed similarly to AOs, first as a composite measure, then individually for each of three types of SUDs. The types of SUDs were OUD, DUD, and AUD. Definitions for each SUD type were derived from ICD-9-CM definitions from the VHA Northeast Program Evaluation Center (NEPEC; eTable 1). (14) DUDs included use disorders for stimulants, marijuana, benzodiazepines, and other non-opioid psychoactive substances. Veterans could be classified as having diagnoses for more than one category of SUDs or AOs (e.g. have both OUD and DUD or have both opioid-related accidents and overdoses and self-inflicted injuries).

eTable 1.

ICD-9-CM^✠ Definitions for Substance Use Disorders

Substance Use Disorder	ICD-9-CM Codes
OUD ±	304.00, 304.01, 304.02, 304.03, 304.70, 304.71, 304.72, 304.73, 305.50, 305.51, 305.52, 305.53
DUD ⨂	292.0, 292.11, 292.12, 292.2, 292.81, 292.82, 292.83, 292.84, 292.85, 292.89, 292.9, 304.10, 304.11, 304.12, 304.13, 304.20, 304.21, 304.22, 304.23, 304.30, 304.31, 304.32, 304.33, 304.40, 304.41, 304.42, 304.43, 304.50, 304.51, 304.52, 304.53, 304.60, 304.61, 304.62, 304.63, 304.80, 304.81, 304.82, 304.83, 304.90, 304.91, 304.92, 304.93, 305.20, 305.21, 305.22, 305.23, 305.30, 305.31, 305.32, 305.33, 305.40, 305.41, 305.42, 305.43, 305.60, 305.61, 305.62, 305.63, 305.70, 305.71, 305.72, 305.73, 305.80, 305.81, 305.82, 305.83, 305.90, 305.91, 305.92, 305.93, 648.30, 648.31, 648.32, 648.33, 648.34, 760.75, 779.5, V65.42
AUD ¥	291.0, 291.1, 291.2, 291.20, 291.3, 291.4, 291.5, 291.81, 291.82, 291.89, 291.9, 303.00, 303.01, 303.02, 303.03, 303.90, 303.91, 303.92, 303.93, 305.00, 305.01, 305.02, 305.03

^✠ICD-9-CM=International Classification of Diseases, Ninth Revision, Clinical Modification

^±OUD=Opioid Use Disorder

^⨂DUD=Non-Opioid Drug Use Disorder

^¥AUD=Alcohol Use Disorder

Covariates

Baseline covariates were assessed in the 12 months before the index date. Baseline covariates were based on existing literature denoting their association with development of SUDs or AOs. (15–20) Demographic covariates included age, race, marital status, sex, and geographic region. (21) Medical covariates included the enhanced Charlson comorbidity score, (22) diagnoses of mental health conditions (schizophrenia, major depressive disorder, post-traumatic stress disorder, anxiety disorders, bipolar disorder, multiple mental health conditions), and diagnoses for CNCP conditions (listed above). Medication classes that aid in treating pain or increase the risk of AOs when combined with opioids (benzodiazepines, hypnotics/other non-benzodiazepine sedatives, skeletal muscle relaxants, antidepressants, other non-opioid analgesics) were also identified using VHA Drug Class Codes. These medication classes were characterized as any use in the 12 months before the index date. Opioid medication characteristics for the first 180 days of COT were also evaluated including schedule of opioids used (CII-CV), duration of action (long-acting, short-acting), average morphine equivalent daily dose, and mean days’ covered. Using VHA stop codes, health care visits (physical therapy, pain clinic, chiropractic care, medicine/primary care, and mental health visits) were characterized in two ways: 1) any visit in the 12-month period prior to the index date, and 2) the number of days with each healthcare visit type. VHA stop codes are 3 digit, standardized codes used to characterize VHA outpatient clinics, and thereby, the type of service administered to the patient. (23) Using the vital sign files, pain scores were also characterized from the first 180-day period of COT as average pain score, first pain score in the 180-day period of COT, last pain score before the index date, and pain score change (from initial to last pain score). The vital sign files are databases containing the vital signs recorded for each patient, which include readings such as pain scores, blood pressure and weight.

Statistical Analysis

Using a 1:1 greedy matching algorithm without replacement, Veterans discontinuing opioid therapy were matched to Veterans who continued COT on both the propensity score and index date (within ± 180 days of each other). (24,25) Propensity scores were estimated using a logistic regression to predict the likelihood of being in the discontinued (vs. continued) group based on covariates listed above (except the counts of each type of healthcare visit). Standardized differences were used to assess the balance of covariates between comparison groups before and after matching. Standardized differences less than 10% were indicative of well-balanced covariates. (26) Logistic regression models were then estimated among the propensity score matched sample. Only the dummy variable for opioid status (continuing COT vs. discontinuing) and the counts of each type of healthcare visit in the 12 months prior to the index date, which were not included in the propensity score estimation, were included. We matched patients on whether or not they used each type of healthcare service (e.g., physical therapy), but included the counts of each type of healthcare visit in the logistic regression models to provide additional adjustment. Logistic models were estimated for the two composite outcomes (i.e., SUDs, AOs) in addition to the individual types. Since our goal was to identify safety signals, no multiple testing adjustment was made.

Analyses were conducted using SAS Enterprise Guide 7.1 using a two-sided significance level of 0.05.

Sensitivity Analysis

To explore the robustness of the primary analyses, two alternative analytic approaches were conducted. First, a variant of the original propensity score approach was used; stabilized inverse probability of treatment weighting (SIPTW). Instead of matching on the propensity score, the SIPTW was calculated, and the weights were integrated into the logistic regression model. The SIPTW was calculated using publicly available SAS code. (27) Veterans in non-overlapping regions of the propensity score distribution were excluded from the analytical sample; however, SIPTW analyses use almost the entire cohort to better ensure generalizability of the results.

Second, instrumental variable (IV) models were estimated to account for potential unobserved confounding. (28–30) A detailed description of the IV analyses can be found in Technical Appendix 1.

Another sensitivity analysis was conducted including Veterans who did not have pain scores in the 180 days prior to the index date (described as exclusion 9 in the Exclusion Criteria section above). We conducted both propensity score matched and SIPTW analyses among this modified cohort.

Results

Sample Derivation and Characteristics

A total of 135,597 and 151,462 Veterans were retained in the cohorts assessing newly diagnosed SUDs and AOs following the implementation of the inclusion and exclusion criteria, 114,779 continuing COT and 20,818 discontinuing opioid therapy in the cohort assessing SUDs and 128,816 continuing COT and 22,646 discontinuing opioid therapy in the cohort assessing AOs (Figure 1). For both unmatched cohorts, approximately two thirds were white, 90% male, 40–50% between the ages of 50 and 64, and approximately 70% from urban areas (Tables 1 and 2). Arthritis and back or neck pain were the most common pain conditions. Approximately half of Veterans in each group had one or more mental health diagnoses. Over 70% in each of the groups used non-opioid analgesics.

Table 1.

Baseline Demographic Characteristics of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy among the Unmatched, Matched, and SIPTW^¥ Cohorts Testing Potentially Opioid-Related Adverse Outcomes

	Cohorts to Test Potentially Opioid-Related Adverse Outcomes
	Unmatched Cohort (N=151,462)			Matched Cohort of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy (N=34,674)			SIPTW¥ Cohort of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy (N=151,313)
	Continued Chronic Opioid Therapy (N=128,816)	Discontinued Opioid Therapy (N=22,646)	Abs Std Diff (%)✠	Continued Chronic Opioid Therapy (N=17,337)	Discontinued Opioid Therapy (N=17,337)	Abs Std Diff (%)✠	Continued Chronic Opioid Therapy (N=128,685)	Discontinued Opioid Therapy (N=22,628)	Abs Std Diff (%)✠£
	N (Col %)	N (Col %)		N (Col %)	N (Col %)		N (Col %)	N (Col %)
Race
White	90,233 (70.05)	14,911 (65.84)	9.0	11,521 (66.45)	11,533 (66.52)	0.2	90,124 (70.03)	14,899 (65.84)	0.9
Black	22,076 (17.14)	4231 (18.68)	4.0	3,256 (18.78)	3,240 (18.69)	0.2	22,067 (17.15)	4,228 (18.68)	0.1
Multiracial	4,336 (3.37)	746 (3.29)	0.4	581 (3.35)	596 (3.44)	0.5	4,334 (3.37)	746 (3.30)	0.1
Other	8,100 (6.29)	1,993 (8.80)	9.5	1,395 (8.05)	1,394 (8.04)	0.0	8,096 (6.29)	1,990 (8.79)	2.4
Unknown	4,071 (3.16)	765 (3.38)	1.2	584 (3.37)	574 (3.31)	0.3	4,064 (3.16)	765 (3.38)	0.7
Age
Mean (SD)	56.20 (13.04)	57.39 (14.95)		57.45 (14.12)	57.40 (14.33)		56.20 (13.05)	57.40 (14.94)
18–30	6,293 (4.89)	1,477 (6.52)	7.1	979 (5.65)	978 (5.64)	0.0	6,291 (4.89)	1,468 (6.49)	1.1
31–49	26,777 (20.79)	4,538 (20.04)	1.9	3,472 (20.03)	3,404 (19.63)	1.0	26,760 (20.79)	4,536 (20.05)	1.2
50–64	66,132 (51.34)	9,440 (41.69)	19.4	7,590 (43.78)	7,817 (45.09)	2.6	66,037 (51.32)	9,436 (41.70)	1.1
≥65	29,614 (22.99)	7,191 (31.75)	19.8	5,296 (30.55)	5,138 (29.64)	2.0	29,597 (23.00)	7,188 (31.77)	2.9
Gender
Male	117,991 (91.60)	20,425 (90.19)	4.9	15,734 (90.75)	15,739 (90.78)	0.1	117,870 (91.60)	20,410 (90.20)	0.2
Marital Status
Married	61,960 (48.10)	11,240 (49.63)	3.1	8,518 (49.13)	8,534 (49.22)	0.2	61,885 (48.09)	11,233 (49.64)	1.5
Rural-Urban Commuting Area
Urban	89,811 (69.72)	16,126 (71.21)	3.3	12,284 (70.85)	12,291 (70.89)	0.1	89,723 (69.72)	16,112 (71.20)	1.9
Large Rural	18,634 (14.47)	2,940 (12.98)	4.3	2,330 (13.44)	2,281 (13.16)	0.8	18,615 (14.47)	2,938 (12.98)	2.3
Isolated Small Rural	18,096 (14.05)	2,856 (12.61)	4.2	2,220 (12.80)	2,281 (13.16)	1.1	18,073 (14.04)	2,856 (12.62)	0.4
Missing	2,275 (1.77)	724 (3.20)	9.2	503 (2.90)	484 (2.79)	0.7	2,274 (1.77)	722 (3.19)	1.6
Enhanced Charlson Comorbidity Index
Mean (SD)	2.29 (1.85)	2.24 (1.90)		2.30 (1.90)	2.31 (1.91)		2.29 (1.85)	2.24 (1.90)
0	17,270 (13.41)	3,462 (15.29)	5.4	2,447 (14.11)	2,431 (14.02)	0.3	17,256 (13.41)	3,455 (15.27)	1.6
1	33,737 (26.19)	6,062 (26.77)	1.3	4,512 (26.03)	4,537 (26.17)	0.3	33,715 (26.20)	6,055 (26.76)	0.0
2	30,773 (23.89)	5,079 (22.43)	3.5	3,993 (23.03)	3,948 (22.77)	0.6	30,738 (23.89)	5,078 (22.44)	1.2
3	20,423 (15.85)	3,416 (15.08)	2.1	2,663 (15.36)	2,667 (15.38)	0.1	20,394 (15.85)	3,415 (15.09)	0.1
4	11,833 (9.19)	2,040 (9.01)	0.6	1,611 (9.29)	1,636 (9.44)	0.5	11,825 (9.19)	2,040 (9.02)	0.5
5	6,637 (5.15)	1,166 (5.15)	0.0	953 (5.50)	952 (5.49)	0.0	6,629 (5.15)	1,166 (5.15)	1.0
≥6	8,143 (6.32)	1,421 (6.27)	0.2	1,158 (6.68)	1,166 (6.73)	0.2	8,128 (6.32)	1,419 (6.27)	2.8
Pain Condition
Back and/or Neck Pain Only	20,461 (15.88)	3,087 (13.63)	6.4	2,418 (13.95)	2,342 (13.51)	1.3	20,440 (15.88)	3,083 (13.62)	0.6
Arthritis Only	24,202 (18.79)	5,061 (22.35)	8.8	3,758 (21.68)	3,742 (21.58)	0.2	24,194 (18.80)	5,059 (22.36)	0.9
Headaches Only	1,029 (0.80)	230 (1.02)	2.3	158 (0.91)	152 (0.88)	0.4	1,029 (0.80)	229 (1.01)	0.5
Neuropathic Pain Only	1,964 (1.52)	336 (1.48)	0.3	268 (1.55)	266 (1.53)	0.1	1,953 (1.52)	335 (1.48)	0.6
Arthritis and Back and/or Neck Pain Only	37,985 (29.49)	6,520 (28.79)	1.5	5,015 (28.93)	5,088 (29.35)	0.9	37,927 (29.47)	6,516 (28.80)	0.9
Arthritis, Back and/or Neck Pain, and Headaches Only	7,534 (5.85)	1,281 (5.66)	0.8	953 (5.50)	981 (5.66)	0.7	7,525 (5.85)	1,281 (5.66)	1.7
Neuropathic Pain and One or More Others	28,319 (21.98)	4,770 (21.06)	2.2	3,756 (21.66)	3,763 (21.71)	0.1	28,300 (21.99)	4,767 (21.07)	0.7
All Tracer Pain Conditions	1,644 (1.28)	258 (1.14)	1.3	208 (1.20)	205 (1.18)	0.2	1,643 (1.28)	257 (1.14)	0.2
Other Multiple Pain Conditions	5,678 (4.41)	1,103 (4.87)	2.2	803 (4.63)	798 (4.60)	0.1	5,674 (4.41)	1,101 (4.87)	1.2
Other Medication Use
Antidepressant Use	70,131 (54.44)	11,036 (48.73)	11.4	8,777 (50.63)	8,810 (50.82)	0.4	70,051 (54.44)	11,022 (48.71)	0.7
Skeletal Muscle Relaxant Use	48,950 (38.00)	7,979 (35.23)	5.7	6,303 (36.36)	6,322 (36.47)	0.2	48,900 (38.00)	7,973 (35.24)	0.4
Benzodiazepine Use	33,363 (25.90)	4,785 (21.13)	11.3	3,976 (22.93)	3,922 (22.62)	0.7	33,308 (25.88)	4,779 (21.12)	0.4
Other Non-Opioid Analgesic Use	90,793 (70.48)	16,603 (73.32)	6.3	12,610 (72.73)	12,733 (73.44)	1.6	90,745 (70.52)	16,589 (73.31)	0.5
Hypnotics and Non-Benzodiazepine Sedative Use	19,843 (15.40)	2,946 (13.01)	6.9	2,442 (14.09)	2,379 (13.72)	1.1	19,823 (15.40)	2,943 (13.01)	0.6
Mental Health Conditions
No Mental Health Conditions	60,244 (46.77)	11,422 (50.44)	7.4	8,460 (48.80)	8,510 (49.09)	0.6	60,195 (46.78)	11,418 (50.46)	0.8
Schizophrenia	1,032 (0.80)	192 (0.85)	0.5	160 (0.92)	148 (0.85)	0.7	1,031 (0.80)	191 (0.84)	2.1
Major Depressive Disorder	18,059 (14.02)	2,768 (12.22)	5.3	2,251 (12.98)	2,234 (12.89)	0.3	18,042 (14.02)	2,767 (12.23)	3.0
Post-Traumatic Stress Disorder	7,244 (5.62)	1,333 (5.89)	1.1	1,047 (6.04)	1,015 (5.85)	0.8	7,235 (5.62)	1,331 (5.88)	1.2
Bipolar Disorder	1,528 (1.19)	251 (1.11)	0.7	210 (1.21)	194 (1.12)	0.9	1,514 (1.18)	249 (1.10)	1.0
Anxiety Disorders	6,784 (5.27)	1,111 (4.91)	1.6	872 (5.03)	865 (4.99)	0.2	6,772 (5.26)	1,111 (4.91)	0.6
Multiple Mental Health Conditions	33,925 (26.34)	5,569 (24.59)	4.0	4,337 (25.02)	4,371 (25.21)	0.5	33,896 (26.34)	5,561 (24.58)	1.0
Substance Use Disorders in the 12 Months before Index Date
Composite Substance Use Disorder	23,638 (18.35)	4,005 (17.69)	1.7	3,116 (17.97)	3,175 (18.31)	0.9	23,612 (18.35)	3,992 (17.64)	0.6
Opioid Use Disorder (OUD)	2,755 (2.14)	561 (2.48)	2.3	396 (2.28)	404 (2.33)	0.3	2,746 (2.13)	549 (2.43)	1.5
Other, Non-Opioid Drug Use Disorder (DUD)	11,941 (9.27)	2,142 (9.46)	0.7	1,644 (9.48)	1,653 (9.53)	0.2	11,925 (9.27)	2,132 (9.42)	2.2
Alcohol Use Disorder (AUD)	16,668 (12.94)	2,754 (12.16)	2.4	2,208 (12.74)	2,210 (12.75)	0.0	16,660 (12.95)	2,750 (12.15)	0.1
Potentially Opioid-Related Adverse Outcomes in the 12 Months before Index Date
Composite Potentially Opioid-Related Adverse Outcomes	--	--	--	--	--	--	--	--	--
Accidents resulting in Wounds or Injuries	--	--	--	--	--	--	--	--	--
Opioid-Related Accidents and Overdoses	--	--	--	--	--	--	--	--	--
Non-Opioid, Drug-Related Accidents and Overdoses	--	--	--	--	--	--	--	--	--
Self-Inflicted Injuries	--	--	--	--	--	--	--	--	--
Violence-Related Injuries	--	--	--	--	--	--	--	--	--
Percent with Each of the Following Visit Types in the 12 Months before Index Date
Physical Therapy	42,398 (32.91)	7,949 (35.10)	4.6	5,999 (34.60)	6,051 (34.90)	0.6	42,381 (32.93)	7,942 (35.10)	2.9
Pain Clinic	18,243 (14.16)	2,771 (12.24)	5.7	2,116 (12.21)	2,240 (12.92)	2.2	18,213 (14.15)	2,767 (12.23)	2.6
Chiropractic Care	1,973 (1.53)	368 (1.63)	0.8	301 (1.74)	290 (1.67)	0.5	1,972 (1.53)	368 (1.63)	0.1
Medicine and Primary Care	128,711 (99.92)	22,606 (99.82)	2.7	17,322 (99.91)	17,308 (99.83)	2.3	128,580 (99.92)	22,588 (99.82)	0.0
Mental Health Care	67,564 (52.45)	11,100 (49.02)	6.9	8,749 (50.46)	8,681 (50.07)	0.8	67,485 (52.44)	11,085 (48.99)	0.4
Duration of Action of Opioid Use in 180 Days before Index Date
Long-Acting Only	2,532 (1.97)	199 (0.88)	9.2	166 (0.96)	171 (0.99)	0.3	2,476 (1.92)	197 (0.87)	3.4
Short-Acting Only	114,043 (88.53)	21,328 (94.18)	20.2	16,239 (93.67)	16,201 (93.45)	0.9	114,014 (88.660)	21,316 (94.20)	7.2
Combination of Long and Short-Acting	12,241 (9.50)	1,119 (4.94)	17.7	932 (5.38)	965 (5.57)	0.8	12,195 (9.48)	1,115 (4.93)	6.2
Schedule of Opioid Use in 180 Days before Index Date
Schedule II Only	73,467 (57.03)	8,672 (38.29)	38.2	7,360 (42.45)	7,536 (43.47)	2.1	73,353 (57.00)	8,667 (38.30)	4.6
Schedule III Only	3,484 (2.70)	985 (4.35)	8.9	702 (4.05)	723 (4.17)	0.6	3,482 (2.71)	985 (4.35)	0.9
Schedule IV Only	24,529 (19.04)	8,047 (35.53)	37.7	5,428 (31.31)	5,096 (29.39)	4.2	24,526 (19.06)	8,038 (35.52)	7.0
Schedule V Only	0 (0.00)	0 (0.00)	0.0	0 (0.00)	0 (0.00)	0.0	0 (0.00)	0 (0.00)	0.0
Use of Multiple Schedules	27,336 (21.22)	4,942 (21.82)	1.5	3,847 (22.19)	3,982 (22.97)	1.9	27,324 (21.23)	4,938 (21.82)	2.0
	Mean (SD)	Mean (SD)		Mean (SD)	Mean (SD)
Average Total Days of Opioid Supply
180 Days before Index Date	151.75 (22.92)	113.52 (23.53)	164.6	121.52 (24.07)	120.03 (23.11)	6.3	151.73 (22.92)	113.53 (23.53)	1.1
180 Days after Index Date	150.60 (25.17)	0.00 (0.00)	--	134.73 (26.61)	0.00 (0.00)	--	150.58 (25.17)	0.00 (0.00)	--
Average Morphine Equivalent Daily Dose
180 Days before Index Date	31.83 (40.70)	21.88 (24.26)	29.7	23.66 (24.24)	23.27 (26.29)	1.5	31.51 (38.19)	21.86 (24.13)	9.8
180 Days after Index Date	36.81 (45.70)	0.00 (0.00)	--	26.89 (28.32)	0.00 (0.00)	--	36.51 (43.84)	0.00 (0.00)	--
Pain Characteristics in 180 Days before Index Date
First Pain Score	5.01 (3.25)	4.74 (3.33)	8.2	4.81 (3.30)	4.78 (3.34)	0.7	5.01 (3.25)	4.74 (3.33)	4.6
Last Pain Score	4.27 (3.28)	3.45 (3.28)	25.1	3.73 (3.30)	3.67 (3.31)	1.9	4.27 (3.28)	3.45 (3.28)	1.3
Pain Score Average	4.54 (2.40)	4.06 (2.42)	20.1	4.18 (2.44)	4.19 (2.42)	0.7	4.54 (2.40)	4.06 (2.42)	3.4
Change from First to Last Pain Score	−0.75 (3.80)	−1.30 (3.95)	14.3	−1.07 (3.81)	−1.11 (3.96)	1.0	−0.75 (3.80)	−1.30 (3.95)	2.9
Service Visits in the 12 Months before Index Date Conditional on Use of the Visit Type ‡
Physical Therapy	3.83 (5.80)	4.41 (6.00)	--	4.00 (6.29)	4.43 (6.20)	--	3.83 (5.80)	4.41 (6.00)	--
Pain Clinic	3.26 (3.35)	3.06 (3.15)	--	3.27 (3.48)	3.07 (3.09)	--	3.26 (3.35)	3.06 (3.15)	--
Chiropractic Care	4.25 (4.74)	3.96 (3.87)	--	4.06 (4.74)	4.01 (4.00)	--	4.25 (4.74)	3.96 (3.87)	--
Medicine and Primary Care	10.06 (7.22)	9.72 (7.39)	--	9.83 (7.30)	9.92 (7.49)	--	10.06 (7.22)	9.72 (7.39)	--
Mental Health Care	10.65 (16.37)	10.71 (17.05)	--	11.15 (17.05)	10.70 (16.99)	--	10.65 (16.37)	10.69 (17.04)	--

^¥SIPTW = Stabilized Inverse Probability of Treatment Weighting

^✠Abs Std Diff (%) = Absolute Standardized Differences in Percentage Form

^‡Absolute Standardized Differences not calculated as the count of services visits in the 12 months before index date are conditional on any use of these types of services and were included as covariates in the final models

^£Absolute standardized differences calculated among the SIPTW sample are calculated after weighting the sample

Table 2.

Baseline Demographic Characteristics of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy among the Unmatched, Matched, and SIPTW^¥ Cohorts Testing Substance Use Disorders

	Cohorts to Test Substance Use Disorders
	Unmatched Cohort (N=135,597)			Matched Cohort of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy (N=31,390)			SIPTW¥ Cohort of Veterans Continuing Chronic Opioid Therapy and Discontinuing Opioid Therapy (N=135,501)
	Continued Chronic Opioid Therapy (N=114,779)	Discontinued Opioid Therapy (N=20,818)	Abs Std Diff (%)✠	Continued Chronic Opioid Therapy (N=15,695)	Discontinued Opioid Therapy (N=15,695)	Abs Std Diff (%)✠	Continued Chronic Opioid Therapy (N=114,701)	Discontinued Opioid Therapy (N=20,800)	Abs Std Diff (%)✠£
	N (Col %)	N (Col %)		N (Col %)	N (Col %)		N (Col %)	N (Col %)
Race
White	82,581 (71.95)	14,015 (67.32)	10.1	10,809 (68.87)	10,720 (68.30)	1.2	82,513 (71.94)	14,005 (67.33)	1.2
Black	17,490 (15.24)	3,571 (17.15)	5.2	2,646 (16.86)	2,672 (17.02)	0.4	17,485 (15.24)	3,566 (17.14)	0.2
Multiracial	3,629 (3.16)	667 (15.53)	0.2	469 (2.99)	517 (3.29)	1.8	3,628 (3.16)	666 (3.20)	0.7
Other	7,284 (6.35)	1,841 (8.84)	9.4	1,253 (7.98)	1,248 (7.95)	0.1	7,283 (6.35)	1,839 (8.84)	2.8
Unknown	3,795 (3.31)	724 (3.48)	1.0	518 (3.30)	538 (3.43)	0.7	3,792 (3.31)	724 (3.48)	0.6
Age
Mean (SD)	56.75 (13.51)	58.50 (15.25)		58.40 (14.49)	58.48 (14.62)		56.75 (13.52)	58.50 (15.24)
18–30	5,650 (4.92)	1,257 (6.04)	4.9	828 (5.28)	817 (5.21)	0.3	5,649 (4.92)	1,252 (6.02)	1.7
31–49	23,801 (20.74)	4,029 (19.35)	3.5	3,051 (19.44)	2,965 (18.89)	1.4	23,786 (20.74)	4,026 (19.36)	1.8
50–64	55,832 (48.64)	8,100 (38.91)	19.7	6,446 (41.07)	6,669 (42.49)	2.9	55,782 (48.63)	8,097 (38.93)	0.4
≥65	29,496 (25.70)	7,432 (35.70)	21.8	5,370 (34.21)	5,244 (33.41)	1.7	29,484 (25.71)	7,425 (35.70)	2.0
Gender
Male	104,066 (90.67)	18,581 (89.25)	4.7	14,079 (89.70)	14,116 (89.94)	0.8	103,999 (90.67)	18,566 (89.26)	0.5
Marital Status
Married	59,099 (51.49)	11,053 (53.09)	3.2	8,312 (52.96)	8,308 (52.93)	0.1	59,064 (51.49)	11,045 (53.10)	1.8
Rural-Urban Commuting Area
Urban	79,083 (68.90)	14,675 (70.49)	3.5	11,000 (70.09)	11,025 (70.25)	0.4	79,027 (68.90)	14,662 (70.49)	2.2
Large Rural	17,039 (14.85)	2,743 (13.18)	4.8	2,176 (13.86)	2,098 (13.37)	1.5	17,030 (14.85)	2,742 (13.18)	2.7
Isolated Small Rural	16,488 (14.36)	2,705 (12.99)	4.0	2,033 (12.95)	2,112 (13.46)	1.5	16,475 (14.36)	2,703 (13.00)	0.6
Missing	2,169 (1.89)	695 (3.34)	9.1	486 (3.10)	460 (2.93)	1.0	2,169 (1.89)	693 (3.33)	1.8
Enhanced Charlson Comorbidity Index
Mean (SD)	2.31 (1.90)	2.29 (1.96)		2.37 (1.97)	2.37 (1.96)		2.31 (1.90)	2.29 (1.96)
0	15,922 (13.87)	3,237 (15.55)	4.7	2,225 (14.18)	2,207 (14.06)	0.3	15,905 (13.87)	3,231 (15.53)	0.7
1	29,593 (25.78)	5,397 (25.92)	0.3	3,993 (25.44)	3,982 (25.37)	0.2	29,581 (25.79)	5,391 (25.92)	0.4
2	26,692 (23.26)	4,525 (21.74)	3.6	3,467 (22.09)	3,474 (22.13)	0.1	26,676 (23.26)	4,521 (21.74)	2.3
3	17,948 (15.64)	3,153 (15.15)	1.4	2,404 (15.32)	2,425 (15.45)	0.4	17,929 (15.63)	3,153 (15.16)	0.8
4	10,714 (9.33)	1,912 (9.18)	0.5	1,503 (9.58)	1,525 (9.72)	0.5	10,710 (9.34)	1,912 (9.19)	0.5
5	6,156 (5.36)	1,124 (5.40)	0.2	927 (5.91)	914 (5.82)	0.4	6,149 (5.36)	1,124 (5.40)	0.9
≥6	7,754 (6.76)	1,470 (7.06)	1.2	1,176 (7.49)	1,168 (7.44)	0.2	7,751 (6.76)	1,468 (7.06)	2.6
Pain Condition
Back and/or Neck Pain Only	17,603 (15.34)	2,632 (12.64)	7.8	2,026 (12.91)	1,995 (12.71)	0.6	17,590 (15.34)	2,630 (12.64)	1.0
Arthritis Only	21,687 (18.89)	4,723 (22.69)	9.4	3,363 (21.43)	3,433 (21.87)	1.1	21,681 (18.90)	4,720 (22.69)	0.2
Headaches Only	910 (0.79)	206 (0.99)	2.1	126 (0.80)	139 (0.89)	0.9	910 (0.79)	205 (0.99)	0.8
Neuropathic Pain Only	1,803 (1.57)	322 (1.55)	0.2	249 (1.59)	256 (1.63)	0.4	1,792 (1.56)	321 (1.54)	0.6
Arthritis and Back and/or Neck Pain Only	33,393 (29.09)	5,916 (28.42)	1.5	4,488 (28.60)	4,515 (28.77)	0.4	33,366 (29.09)	5,914 (28.43)	1.8
Arthritis, Back and/or Neck Pain, and Headaches Only	6,895 (6.01)	1,202 (5.77)	1.0	924 (5.89)	896 (5.71)	0.8	6,889 (6.01)	1,201 (5.77)	0.8
Neuropathic Pain and One or More Others	25,761 (22.44)	4,572 (21.96)	1.2	3,565 (22.71)	3,566 (22.72)	0.0	25,750 (22.45)	4,570 (21.97)	1.0
All Tracer Pain Conditions	1,600 (1.39)	253 (1.22)	1.6	213 (1.36)	199 (1.27)	0.8	1,598 (1.39)	250 (1.20)	1.8
Other Multiple Pain Conditions	5,127 (4.47)	992 (4.77)	1.4	741 (4.72)	696 (4.43)	1.4	5,125 (4.47)	989 (4.75)	2.7
Other Medication Use
Antidepressant Use	59,649 (51.97)	9,494 (45.60)	12.8	7,534 (48.00)	7,511 (47.86)	0.3	59,613 (51.97)	9,484 (45.60)	0.6
Skeletal Muscle Relaxant Use	42,982 (37.45)	7,095 (34.08)	7.0	5,535 (35.27)	5,557 (35.41)	0.3	42,931 (37.43)	7,090 (34.09)	0.6
Benzodiazepine Use	29,972 (26.11)	4,339 (20.84)	12.5	3,539 (22.55)	3,510 (22.36)	0.4	29,951 (26.11)	4,336 (20.85)	1.7
Other Non-Opioid Analgesic Use	80,007 (69.71)	15,133 (72.69)	6.6	11,355 (72.35)	11,399 (72.63)	0.6	79,984 (69.73)	15,118 (72.68)	0.0
Hypnotics and Non-Benzodiazepine Sedative Use	17,175 (14.96)	2,547 (12.23)	8.0	2,109 (13.44)	2,047 (13.04)	1.2	17,168 (14.97)	2,547 (12.25)	0.3
Mental Health Conditions
No Mental Health Conditions	57,958 (50.50)	11,421 (54.86)	8.8	8,305 (52.91)	8,373 (53.35)	0.9	57,937 (50.51)	11,411 (54.86)	1.4
Schizophrenia	767 (0.67)	160 (0.77)	1.2	118 (0.75)	120 (0.76)	0.2	766 (0.67)	159 (0.76)	1.4
Major Depressive Disorder	15,392 (13.41)	2,418 (11.61)	5.4	1,942 (12.37)	1,932 (12.31)	0.2	15,388 (13.42)	2,416 (11.62)	2.6
Post-Traumatic Stress Disorder	6,270 (5.46)	1,191 (5.72)	1.1	891 (5.68)	900 (5.73)	0.3	6,266 (5.46)	1,191 (5.73)	0.1
Bipolar Disorder	1,141 (0.99)	191 (0.92)	0.8	137 (0.87)	147 (0.94)	0.7	1,111 (0.97)	190 (0.91)	4.8
Anxiety Disorders	6,098 (5.31)	995 (4.78)	2.4	801 (5.10)	756 (4.82)	1.3	6,096 (5.31)	995 (4.78)	0.0
Multiple Mental Health Conditions	27,153 (23.66)	4,442 (21.34)	5.6	3,501 (22.31)	3,467 (22.09)	0.5	27,137 (23.66)	4,438 (21.34)	1.5
Substance Use Disorders in the 12 Months before Index Date
Composite Substance Use Disorder	--	--	--	--	--	--	--	--	--
Opioid Use Disorder (OUD)	--	--	--	--	--	--	--	--	--
Other, Non-Opioid Drug Use Disorder (DUD)	--	--	--	--	--	--	--	--	--
Alcohol Use Disorder (AUD)	--	--	--	--	--	--	--	--	--
Potentially Opioid-Related Adverse Outcomes in the 12 Months before Index Date
Composite Potentially Opioid-Related Adverse Outcomes	9,614 (8.38)	2,180 (10.47)	7.2	1,562 (9.95)	1,636 (10.42)	1.6	9,610 (8.38)	2,175 (10.46)	1.6
Accidents resulting in Wounds or Injuries	8,278 (7.21)	1,950 (9.37)	7.8	1,403 (8.94)	1,452 (9.25)	1.1	8,275 (7.21)	1,945 (9.35)	1.0
Opioid-Related Accidents and Overdoses	106 (0.09)	24 (0.12)	0.7	19 (0.12)	17 (0.11)	0.4	106 (0.09)	24 (0.12)	1.2
Non-Opioid, Drug-Related Accidents and Overdoses	206 (0.18)	41 (0.20)	0.4	28 (0.18)	32 (0.20)	0.6	206 (0.18)	41 (0.20)	0.9
Self-Inflicted Injuries	1,249 (1.09)	204 (0.98)	1.1	163 (1.04)	164 (1.04)	0.1	1,249 (1.09)	204 (0.98)	2.0
Violence-Related Injuries	207 (0.18)	34 (0.16)	0.4	23 (0.15)	27 (0.17)	0.6	206 (0.18)	34 (0.16)	0.2
Percent with Each of the Following Visit Types in the 12 Months before Index Date
Physical Therapy	37,758 (32.90)	7,501 (36.03)	6.6	5,567 (35.47)	5,566 (35.46)	0.0	37,743 (32.91)	7,490 (36.01)	3.8
Pain Clinic	15,875 (13.83)	2,409 (11.57)	6.8	1,825 (11.63)	1,915 (12.20)	1.8	15,858 (13.83)	2,405 (11.56)	2.4
Chiropractic Care	1,738 (1.51)	337 (1.62)	0.8	255 (1.62)	256 (1.63)	0.1	1,738 (1.52)	336 (1.62)	0.3
Medicine and Primary Care	114,668 (99.90)	20,771 (99.77)	3.2	15,680 (99.90)	15,664 (99.80)	2.7	114,590 (99.90)	20,756 (99.79)	0.6
Mental Health Care	54,087 (47.12)	8,942 (42.95)	8.4	6,975 (44.44)	6,882 (43.85)	1.2	54,032 (47.11)	8,934 (42.95)	0.1
Duration of Action of Opioid Use in 180 Days before Index Date
Long-Acting Only	2,075 (1.81)	162 (0.78)	9.1	139 (0.89)	141 (0.90)	0.1	2,054 (1.79)	160 (0.77)	3.0
Short-Acting Only	101,873 (88.76)	19,649 (94.38)	20.4	14,716 (93.76)	14,706 (93.70)	0.3	101,847 (88.79)	19,637 (94.41)	6.2
Combination of Long and Short-Acting	10,831 (9.44)	1,007 (4.84)	17.9	840 (5.35)	848 (5.40)	0.2	10,800 (9.42)	1,003 (4.82)	5.3
Schedule of Opioid Use in 180 Days before Index Date
Schedule II Only	66,307 (57.77)	8,037 (38.61)	39.1	6,744 (42.97)	6,915 (44.06)	2.2	66,248 (57.76)	8,035 (38.63)	3.7
Schedule III Only	3,147 (2.74)	930 (4.47)	9.3	643 (4.10)	675 (4.30)	1.0	3,145 (2.74)	927 (4.46)	0.5
Schedule IV Only	21,603 (18.82)	7,262 (34.88)	36.9	4,802 (30.60)	4,535 (28.89)	3.7	21,597 (18.83)	7,255 (34.88)	7.1
Schedule V Only	0 (0.00)	0 (0.00)	0.0	0 (0.00)	0 (0.00)	0.0	0 (0.00)	0 (0.00)	0.0
Use of Multiple Schedules	23,722 (20.67)	4,589 (22.04)	3.4	3,506 (22.34)	3,570 (22.75)	1.0	23,711 (20.67)	4,583 (22.03)	3.1
	Mean (SD)	Mean (SD)		Mean (SD)	Mean (SD)
Average Total Days of Opioid Supply
180 Days before Index Date	151.16 (23.07)	112.83 (23.23)	165.6	121.07 (23.91)	119.50 (22.93)	6.7	151.15 (23.07)	112.85 (23.23)	0.8
180 Days after Index Date	149.92 (25.28)	0.00 (0.00)	--	134.19 (26.74)	0.00 (0.00)	--	149.91 (25.28)	0.00 (0.00)	--
Average Morphine Equivalent Daily Dose
180 Days before Index Date	31.46 (39.74)	21.24 (22.00)	31.8	22.95 (21.85)	22.56 (23.83)	1.7	31.26 (38.04)	21.24 (21.99)	7.5
180 Days after Index Date	36.24 (44.84)	0.00 (0.00)	--	25.84 (25.54)	0.00 (0.00)	--	36.05 (43.51)	0.00 (0.00)	--
Pain Characteristics in 180 Days before Index Date
First Pain Score	4.93 (3.26)	4.66 (3.34)	8.2	4.71 (3.29)	4.70 (3.35)	0.4	4.93 (3.26)	4.66 (3.34)	5.0
Last Pain Score	4.20 (3.27)	3.31 (3.24)	27.4	3.60 (3.26)	3.56 (3.28)	1.2	4.20 (3.27)	3.31 (3.24)	0.1
Pain Score Average	4.47 (2.40)	3.95 (2.40)	21.8	4.07 (2.42)	4.09 (2.41)	1.0	4.47 (2.40)	3.95 (2.40)	3.1
Change from First to Last Pain Score	−0.72 (3.79)	−1.35 (3.95)	16.1	−1.11 (3.82)	−1.14 (3.95)	0.7	−0.73 (3.79)	−1.34 (3.95)	4.2
Service Visits in the 12 Months before Index Date Conditional on Use of the Visit Type ‡
Physical Therapy	3.95 (6.22)	4.68 (6.46)	--	4.08 (6.20)	4.67 (6.55)	--	3.95 (6.22)	4.68 (6.46)	--
Pain Clinic	3.22 (3.28)	3.06 (3.02)	--	3.27 (3.41)	3.10 (3.06)	--	3.22 (3.28)	3.06 (3.03)	--
Chiropractic Care	4.38 (5.04)	3.91 (3.80)	--	4.35 (4.36)	4.02 (3.94)	--	4.38 (5.04)	3.90 (3.81)	--
Medicine and Primary Care	10.07 (7.25)	9.80 (7.47)	--	9.98 (7.50)	9.97 (7.50)	--	10.07 (7.25)	9.80 (7.47)	--
Mental Health Care	8.21 (11.49)	8.18 (12.37)	--	8.49 (11.64)	8.21 (12.60)	--	8.21 (11.50)	8.18 (12.38)	--

^¥SIPTW = Stabilized Inverse Probability of Treatment Weighting

^✠Abs Std Diff (%) = Absolute Standardized Differences in Percentage Form

^‡Absolute Standardized Differences not calculated as the count of services visits in the 12 months before index date are conditional on any use of these types of services and were included as covariates in the final models

^£Absolute standardized differences calculated among the SIPTW sample are calculated after weighting the sample

Prior to matching, age, race, rural/urban status, opioid schedule (e.g., use of Schedule II opioids only), and average pain scores in the baseline period were different between those who continued COT compared to those who discontinued opioids. Mean days of opioid supply in the 180 days prior to the index date (cohort testing SUDs: 151.16 days for Veterans continuing COT; 112.83 for Veterans discontinuing opioids; cohort testing AOs: 151.75 days for Veterans continuing COT; 113.52 for Veterans discontinuing opioids), use of combinations of long- and short-acting opioids (cohort testing SUDs: 9.44% for Veterans continuing COT; 4.84% for Veterans discontinuing opioids; cohort testing AOs: 9.50% of Veterans continuing COT; 4.94% of Veterans discontinuing opioids), and average daily morphine equivalent dose (cohort testing SUDs: 31.46 average daily MME for Veterans continuing COT; 21.24 average daily MME for Veterans discontinuing opioid therapy; cohort testing AOs: 31.83 average daily MME for Veterans continuing COT; 21.88 average daily MME for Veterans discontinuing opioid therapy) were also higher for those continuing COT prior to matching. In the 180 days after the index date, the mean days of opioid supply remained similar at 149.92 and 150.60 among Veterans continuing COT in the cohorts testing SUDs and AOs respectively and 0.00 for those discontinuing opioid use. After matching, 15,695 and 17,337 Veterans continuing COT matched to Veterans discontinuing opioid therapy in the cohorts testing SUDs and AOs (cohort testing SUDs: 13.7% of Veterans continuing COT and 75.4% of Veterans discontinuing opioid therapy; cohort testing AOs: 13.5% of Veterans continuing COT and 76.6% of Veterans discontinuing opioid therapy). The baseline covariates were well balanced after propensity score matching (all standardized differences <10%).

Substance Use Disorders and Adverse Outcomes: Veterans Continuing COT vs. Veterans Discontinuing Opioid Therapy

The composite rate of SUDs was not different between Veterans discontinuing opioid therapy and Veterans continuing COT in the matched sample (eFigure 2, Table 3; OR=0.932, 95%CI: 0.850, 1.022). For individual SUDs, the odds of DUD and AUD did not differ between the two groups, but the odds of OUD was lower in Veterans discontinuing opioid therapy than in those continuing COT (OR=0.681, 95%CI: 0.535, 0.867). The rate of composite AOs was lower for Veterans discontinuing opioid therapy than for those continuing COT (eFigure 3, Table 4; OR=0.703, 95%CI: 0.648, 0.763). The lower rate of composite AO was driven by differences in accidents resulting in wounds/injuries (OR=0.610, 95%CI: 0.553, 0.672), opioid-related accidents and overdoses (OR=0.499, 95%CI: 0.286, 0.871), and violence-related injuries (OR=0.410, 95%CI: 0.229, 0.735); none of the other individual types of AOs were different between the two groups.

eFigure 2.

SUD* Treatment in VHA comparing Veterans Discontinuing Opioid Therapy to Veterans Continuing Chronic Opioid Therapy with PS Match^¥ and SIPTW^Δ✠ Samples

*SUD=Substance Use Disorder

^ΔSIPTW=Stabilized Inverse Probability of Treatment Weighting

^✠Referent: Continued Chronic Opioid Users; OUD=Opioid Use Disorder; DUD=Non-Opioid, Drug Use Disorder; AUD=Alcohol Use Disorder

^¥ PS Match=Propensity Score Matched

Table 3.

SUD Treatment in VHA comparing Veterans Continuing COT to Veterans Discontinuing Opioid Therapy among the Unmatched, Matched, and SIPTW Samples

	Unmatched Sample of Veterans Continuing COT‡ and Discontinuing Opioid Therapy		Matched Sample of Veterans Continuing COT‡ and Veterans Discontinuing Opioid Therapy		Matched Sample (Discontinuing Opioid Therapy vs. Continuing COT‡) Odds Ratio (OR) and Confidence Interval (CI)✠*			SIPTW⨂ Sample of Veterans Continuing COT‡ and Veterans Discontinuing Opioid Therapy∞		SIPTW⨂ Sample (Discontinuing Opioid Therapy vs. Continuing COT‡) Odds Ratio (OR) and Confidence Interval (CI)✠*Ω
	N=135,597		N=31,390					N=135,501
	Discontinuing Opioid Therapy N=20,818 N (%)	Continuing COT‡ N=114,779 N (%)	Discontinuing Opioid Therapy N=15,695 N (%)	Continuing COT‡ N=15,695 N (%)	OR	Lower 95% CI	Upper 95% CI	Discontinuing Opioid Therapy N=20,800 N (%)	Continuing COT‡ N=114,701 N (%)	OR	Lower 95% CI	Upper 95% CI
Composite SUDs	1185 (5.7)	9337 (8.1)	944 (6.0)	1008 (6.4)	0.932	0.850	1.022	1,184 (5.7)	9,327 (8.1)	0.789	0.743	0.837
OUD±	132 (0.6)	1851 (1.6)	113 (0.7)	165 (1.1)	0.681	0.535	0.867	132 (0.6)	1,846 (1.6)	0.739	0.644	0.847
DUD⨂	596 (2.9)	5159 (4.5)	477 (3.0)	522 (3.3)	0.909	0.801	1.032	595 (2.9)	5,153 (4.5)	0.723	0.666	0.785
AUD¥	762 (3.7)	5006 (4.4)	608 (3.9)	584 (3.7)	1.042	0.927	1.170	761 (3.7)	5,003 (4.4)	0.862	0.798	0.930

^✠Referent: Veterans Continuing Chronic Opioid Therapy

^*Logistic Regression Models among the matched sample include counts of healthcare service visits as covariates.

^‡COT=Chronic Opioid Therapy

^±OUD=Opioid Use Disorder

^⨂DUD=Non-Opioid Drug Use Disorder

^¥AUD=Alcohol Use Disorder

^⨂SIPTW=Stabilized Inverse Probability of Treatment Weighting

^∞This sample applies criteria to exclude those patients in non-overlapping regions of the propensity score distribution but does not apply the SIPTW

^ΩThe odds ratios are derived after excluding those patients in non-overlapping regions of the propensity score distribution and applying the SIPTW

eFigure 3.

AO^⨂ Treatment in VHA comparing Veterans Discontinuing Opioid Therapy to Veterans Continuing Chronic Opioid Therapy with PS Match^¥ and SIPTW^Δ✠ Samples

^⨂AO=Adverse Outcome

^ΔSIPTW=Stabilized Inverse Probability of Treatment Weighting

^✠Referent: Continued Chronic Opioid Users

^¥ PS Match=Propensity Score Matched

Table 4.

AO Treatment in VHA comparing Veterans Continuing COT to Veterans Discontinuing Opioid Therapy among the Unmatched, Matched, and SIPTW Samples

	Unmatched Sample of Veterans Continuing COT‡, Switching to Intermittent Opioid Therapy, and Discontinuing Opioid Therapy		Matched Sample of Veterans Continuing COT‡ and Veterans Discontinuing Opioid Therapy		Matched Sample (Continuing COT‡ vs Discontinuing Opioid Therapy) Odds Ratio (OR) and Confidence Interval (CI)✠*			SIPTW⨂ Sample of Veterans Continuing COT‡ and Veterans Discontinuing Opioid Therapy∞		SIPTW⨂ Sample (Continuing COT‡ vs Discontinuing Opioid Therapy) Odds Ratio (OR) and Confidence Interval (CI)✠* Ω
	N=151,462		N=34,674					N=151,313
	Continuing COT‡ N=128,816 N (%)	Discontinuing Opioid Therapy N=22,646 N (%)	Continuing COT‡ N=17,337 N (%)	Discontinuing Opioid Therapy N=17,337 N (%)	OR	Lower 95% CI	Upper 95% CI	Continuing COT‡ N=128,685 N (%)	Discontinuing Opioid Therapy N=22,628 N (%)	OR	Lower 95% CI	Upper 95% CI
Composite AOs	11528 (9.0)	1412 (6.2)	1517 (8.8)	1098 (6.3)	0.703	0.648	0.763	11,515 (9.0)	1,410 (6.2)	0.660	0.623	0.699
Wounds and Injuries	8050 (6.3)	928 (4.1)	1114 (6.4)	700 (4.0)	0.610	0.553	0.672	8,042 (6.3)	928 (4.1)	0.610	0.569	0.655
Opioid-Related	378 (0.3)	22 (0.1)	37 (0.2)	19 (0.1)	0.499	0.286	0.871	377 (0.3)	22 (0.1)	0.638	0.463	0.879
Alcohol and Non-Opioid Medication Related	2360 (1.8)	305 (1.4)	248 (1.4)	258 (1.5)	1.041	0.873	1.241	2,356 (1.8)	304 (1.3)	0.850	0.758	0.953
Self-Inflicted Injuries	2177 (1.7)	325 (1.4)	255 (1.5)	257 (1.5)	1.016	0.852	1.211	2,174 (1.7)	325 (1.4)	0.811	0.718	0.915
Violence-Related Injuries	206 (0.2)	24 (0.1)	39 (0.2)	16 (0.1)	0.410	0.229	0.735	206 (0.2)	23 (0.1)	0.527	0.338	0.823

^✠Referent: Veterans Continuing Chronic Opioid Therapy

^*Logistic Regression Models among the matched sample include counts of healthcare service visits as covariates.

^‡COT=Chronic Opioid Therapy

^⨂SIPTW=Stabilized Inverse Probability of Treatment Weighting

^∞This sample applies criteria to exclude those patients in non-overlapping regions of the propensity score distribution but does not apply the SIPTW

^ΩThe odds ratios are derived after excluding those patients in non-overlapping regions of the propensity score distribution and applying the SIPTW

Sensitivity Analyses

SIPTW Logistic Regressions

A total of 135,501 and 151,313 Veterans were retained in the SIPTW cohorts testing SUDs and AOs respectively after excluding Veterans in non-overlapping regions of the propensity score distribution, 114,701 Veterans continuing COT and 20,800 Veterans discontinuing opioid therapy in the cohort testing SUDs and 128,685 Veterans continuing COT and 22,628 Veterans discontinuing opioid therapy in the cohort testing AOs. In SIPTW analyses, which retained close to the full sample, the point estimates were similar to the propensity score matched point estimates; however, the confidence intervals were narrower. Veterans discontinuing opioid therapy had lower odds of the composite SUD outcome (eFigure 2, Table 3; OR=0.789, 95%CI: 0.743, 0.837) and all individual SUD types (OUD: OR=0.689, 95%CI: 0.550, 0.863, DUD: OR=0.838, 95%CI: 0.741, 0.947, AUD: OR=0.862, 95%CI: 0.798, 0.930) compared to Veterans continuing COT. Also consistent with the primary analysis, Veterans discontinuing opioids had lower odds of the composite AO outcome (eFigure 3, Table 4; OR=0.660, 95%CI: 0.623, 0.699) than those continuing COT. All of the individual AO types were lower among Veterans discontinuing opioid therapy (accidents resulting in wounds/injuries: OR=0.610, 95%CI: 0.569, 0.655, opioid-related accidents and overdoses: OR=0.638, 95%CI: 0.463, 0.879, alcohol and non-opioid medication related accidents and overdoses: OR=0.850, 95%CI: 0.758, 0.953, self-inflicted injuries: OR=0.811, 95%CI: 0.718, 0.915, and violence-related injuries: OR=0.527, 95%CI: 0.338, 0.823).

IV Analyses

The entire unmatched sample was used for the IV analyses (cohort to test SUDs: N=135,597, cohort to test AOs: N=151,462). Most of the covariates were well balanced between Veterans getting care at VHA stations above and below the median discontinuation rate (cohort to test SUDs: 14.9%; cohort to test AOs: 14.5%) but differed on several important covariates (eTable 2). Veterans at VHA stations with higher rates of discontinuation were more likely to be non-white, more likely to be prescribed schedule IV opioids, less likely to be from isolated rural areas, were prescribed 5 fewer days of opioids, and had lower pain score averages and last pain score readings before the index date as compared to those receiving care at VHA stations with lower rates of discontinuation.

eTable 2.

Demographic Characteristics of Veterans by Instrumental Variable: Percent of Veterans Discontinuing Opioid Therapy

	Cohort to Test Substance Use Disorders N=135,597			Cohort to Test Potentially Opioid-Related Adverse Outcomes N=151,462
	Associated with Bottom 50% of Locations for Discontinuation (N=78,245)	Associated with Top 50% of Locations for Discontinuation (N=57,352)	Abs Std Diff (%)	Associated with Bottom 50% of Locations for Discontinuation (N=81,787)	Associated with Top 50% of Locations for Discontinuation (N=69,675)	Abs Std Diff (%)
	N (Column %)	N (Column %)		N (Column %)	N (Column %)
Race
White	58124 (74.28)	38472 (67.08)	15.88	59141 (72.31)	46003 (66.03)	13.64
Black	11842 (15.13)	9219 (16.07)	2.59	13770 (16.84)	12537 (17.99)	3.05
Multiracial	2275 (2.91)	2021 (3.52)	3.49	2563 (3.13)	2519 (3.62)	2.67
Other	3372 (4.31)	5753 (10.03)	22.31	3684 (4.50)	6409 (9.20)	18.66
Unknown	2632 (3.36)	1887 (3.29)	0.41	2629 (3.21)	2207 (3.17)	0.27
Age
Mean and Standard Deviation	56.65 ± 13.69	57.52 ± 13.95		56.07 ± 13.25	56.73 ± 13.46
18–30	4086 (5.22)	2821 (4.92)	1.38	4292 (5.25)	3478 (4.99)	1.16
31–49	16557 (21.16)	11273 (19.66)	3.73	17371 (21.24)	13944 (20.01)	3.03
50–64	37113 (47.43)	26819 (46.76)	1.34	40898 (50.01)	34674 (49.77)	0.48
≥65	20489 (26.19)	16439 (28.66)	5.56	19226 (23.51)	17579 (25.23)	4.01
Gender
Male	71189 (90.98)	51458 (89.72)	4.27	75116 (91.84)	63300 (90.85)	3.53
Marital Status
Married	40711 (52.03)	29441 (51.33)	1.39	39716 (48.56)	33484 (48.06)	1.01
Rural-Urban Commuting Area
Urban	52444 (67.03)	41314 (72.04)	10.90	55772 (68.19)	50165 (72.00)	8.32
Large Rural	12755 (16.30)	7027 (12.25)	11.59	12585 (15.39)	8989 (12.90)	7.14
Isolated Small Rural	12526 (16.01)	6667 (11.62)	12.73	12891 (15.76)	8061 (11.57)	12.23
Missing	520 (0.66)	2344 (4.09)	22.62	539 (0.66)	2460 (3.53)	20.15
Enhanced Charlson Comorbidity Index
Mean and Standard Deviation	2.28 ± 1.88	2.35 ± 1.94		2.25 ± 1.83	2.32 ± 1.90
0	11160 (14.26)	7999 (13.95)	0.91	11266 (13.77)	9466 (13.59)	0.55
1	20460 (26.15)	14530 (25.33)	1.86	21801 (26.66)	17998 (25.83)	1.87
2	18008 (23.01)	13209 (23.03)	0.04	19408 (23.73)	16444 (23.60)	0.30
3	12234 (15.64)	8867 (15.46)	0.48	12982 (15.87)	10857 (15.58)	0.80
4	7233 (9.24)	5393 (9.40)	0.55	7361 (9.00)	6512 (9.35)	1.20
5	4076 (5.21)	3204 (5.59)	1.67	4087 (5.00)	3716 (5.33)	1.52
≥6	5074 (6.48)	4150 (7.24)	2.97	4882 (5.97)	4682 (6.72)	3.08
Pain Condition
Back and/or Neck Pain Only	11844 (15.14)	8391 (14.63)	1.42	12854 (15.72)	10694 (15.35)	1.02
Arthritis Only	15246 (19.48)	11164 (19.47)	0.05	15800 (19.32)	13463 (19.32)	0.01
Headaches Only	628 (0.80)	488 (0.85)	0.53	646 (0.79)	613 (0.88)	0.99
Neuropathic Pain Only	1213 (1.55)	912 (1.59)	0.32	1216 (1.49)	1084 (1.56)	0.56
Arthritis and Back and/or Neck Pain Only	22623 (28.91)	16686 (29.09)	0.40	24031 (29.38)	20474 (29.39)	0.01
Arthritis, Back and/or Neck Pain, and Headaches Only	4720 (6.03)	3377 (5.89)	0.61	4793 (5.86)	4022 (5.77)	0.38
Neuropathic Pain and One or More Others	17325 (22.14)	13008 (22.68)	1.29	17702 (21.64)	15387 (22.08)	1.06
All Tracer Pain Conditions	1058 (1.35)	795 (1.39)	0.29	1041 (1.27)	861 (1.24)	0.33
Other Multiple Pain Conditions	3588 (4.59)	2531 (4.41)	0.83	3704 (4.53)	3077 (4.42)	0.54
Other Medication Use
Antidepressant Use	40596 (51.88)	28547 (49.78)	4.22	44247 (54.10)	36920 (52.99)	2.23
Skeletal Muscle Relaxant Use	29785 (38.07)	20292 (35.38)	5.57	31629 (38.67)	25300 (36.31)	4.88
Benzodiazepine Use	19849 (25.37)	14462 (25.22)	0.35	20583 (25.17)	17565 (25.21)	0.10
Other Non-Opioid Analgesic Use	54414 (69.54)	40726 (71.01)	3.21	57539 (70.35)	49857 (71.56)	2.65
Hypnotics and Non-Benzodiazepine Sedative Use	11512 (14.71)	8210 (14.32)	1.13	12411 (15.17)	10378 (14.89)	0.78
Mental Health Conditions
No Mental Health Conditions	39764 (50.82)	29615 (51.64)	1.64	38763 (47.40)	32903 (47.22)	0.34
Schizophrenia	494 (0.63)	433 (0.75)	1.49	607 (0.74)	617 (0.89)	1.60
Major Depressive Disorder	10316 (13.18)	7494 (13.07)	0.35	11294 (13.81)	9533 (13.68)	0.37
Post-Traumatic Stress Disorder	4281 (5.47)	3180 (5.54)	0.32	4561 (5.58)	4016 (5.76)	0.81
Bipolar Disorder	785 (1.00)	547 (0.95)	0.50	952 (1.16)	827 (1.19)	0.21
Anxiety Disorders	4140 (5.29)	2953 (5.15)	0.64	4271 (5.22)	3624 (5.20)	0.09
Multiple Mental Health Conditions	18465 (23.60)	13130 (22.89)	1.67	21339 (26.09)	18155 (26.06)	0.08
Percent with Each of the Following Visit Types in the 12 Months before Index Date
Physical Therapy	25219 (32.23)	20040 (34.94)	5.74	26464 (32.36)	23883 (34.28)	4.08
Pain Clinic	9788 (12.51)	8496 (14.81)	6.71	10522 (12.87)	10492 (15.06)	6.33
Chiropractic Care	1202 (1.54)	873 (1.52)	0.11	1277 (1.56)	1064 (1.53)	0.28
Medicine and Primary Care	78172 (99.91)	57267 (99.85)	1.58	81723 (99.92)	69594 (99.88)	1.22
Mental Health Care	36178 (46.24)	26851 (46.82)	1.16	41937 (51.28)	36727 (52.71)	2.87
Duration of Action of Opioid Use in 180 Days before Index Date
Long-Acting Only	1318 (1.68)	919 (1.60)	0.65	1552 (1.90)	1179 (1.69)	1.55
Short-Acting Only	70152 (89.66)	51370 (89.57)	0.29	73072 (89.34)	62299 (89.41)	0.23
Combination of Long and Short-Acting	6775 (8.66)	5063 (8.83)	0.60	7163 (8.76)	6197 (8.89)	0.48
Schedule of Opioid Use in 180 Days before Index Date
Schedule II Only	47831 (61.13)	26513 (46.23)	30.22	50156 (61.33)	31983 (45.90)	31.30
Schedule III Only	2154 (2.75)	1923 (3.35)	3.49	2238 (2.74)	2231 (3.20)	2.74
Schedule IV Only	13003 (16.62)	15862 (27.66)	26.83	13319 (16.28)	19257 (27.64)	27.69
Schedule V Only	0 (0.00)	0 (0.00)	0.00	0 (0.00)	0 (0.00)	0.00
Use of Multiple Schedules	15257 (19.50)	13054 (22.76)	8.00	16074 (19.65)	16204 (23.26)	8.79
				Mean (SD)	Mean (SD)
Average Morphine Equivalent Daily Dose
180 Days before Index Date	30.63 (38.15)	28.88 (37.16)	4.67	31.29 (39.80)	29.24 (37.68)	5.30
Average Total Days of Opioid Supply
180 Days before Index Date	146.36 (27.38)	140.94 (29.37)	19.11	147.32 (27.09)	142.00 (29.16)	18.93
Pain Characteristics in 180 Days before Dosage Change Index Date
First Pain Score	4.92 (3.25)	4.85 (3.29)	2.05	5.05 (3.24)	4.89 (3.29)	4.93
Last Pain Score	4.13 (3.27)	3.98 (3.29)	4.48	4.25 (3.27)	4.03 (3.31)	6.54
Pain Score Average	4.44 (2.40)	4.32 (2.41)	4.93	4.56 (2.39)	4.36 (2.41)	8.43
Change in Pain Score	-0.79 (3.80)	-0.87 (3.85)	2.09	-0.80 (3.81)	-0.86 (3.86)	1.41
Service Visits in the 12 Months before Dosage Change Index Date Conditional on Use of the Visit Type
Physical Therapy	3.76 (5.98)	4.46 (6.58)	--	3.69 (5.58)	4.18 (6.10)	--
Pain Clinic	2.83 (2.85)	3.63 (3.60)	--	2.89 (2.93)	3.59 (3.64)	--
Chiropractic Care	4.37 (5.08)	4.21 (4.55)	--	4.14 (4.64)	4.28 (4.57)	--
Medicine and Primary Care	9.77 (7.08)	10.38 (7.53)	--	9.72 (7.10)	10.35 (7.40)	--
Mental Health Care	8.04 (11.53)	8.43 (11.73)	--	10.22 (15.87)	11.15 (17.10)	--

The Wald estimator for percentage of Veterans discontinuing opioid therapy per VHA station was significant for SUDs (β=0.095, 95%CI: 0.051, 0.140) and insignificant for AOs (β= 0.20, 95%CI: -0.026, 0.066) indicating that discontinuation may increase the risk for SUD diagnoses but have no effect on AO diagnoses. The 2SLS models, which incorporate covariates, found an increased risk of receiving a new SUD diagnosis with discontinuation (β=0.272, 95%CI: 0.182, 0.362) and a decreased risk of receiving a new AO diagnosis (β= −0.088, 95%CI: −0.173, −0.003). All of the individual types of SUDs were higher among discontinuers (OUD: β=0.087, 95%CI: 0.047, 0.126; DUD: β=0.173, 95%CI: 0.106, 0.240; AUD: β=0.131, 95%CI: 0.065, 0.198). Of the individual types of AOs, accidents resulting in wounds/injuries (β= −0.076, 95%CI: −0.148, −0.004) and self-inflicted injuries (β= −0.048, 95%CI: −0.087, −0.009) were lower among discontinuers while all of the other individual types were not different between the two groups. The biprobit models found the risk of receiving a new SUD diagnosis increased with discontinuation (β=0.166, 95%CI: 0.050, 0.283) while the risk of receiving a new AO diagnosis was lower with discontinuation (β=-0.134, 95%CI: -0.246, -0.022). Bivariate probit models for the individual types of SUDs and AOs were insignificant except for DUD (β=0.183, 95%CI: 0.040, 0.326), AUD (β=0.164, 95%CI: 0.028, 0.301), and accidents resulting in wounds and injuries (β= −0.202, 95%CI: −0.330, −0.075). The Durbin and Wu-Hausman tests were significant for composite SUDs (p<0.001) but insignificant for composite AOs (p=0.1408) indicating the existence of endogeneity in testing SUDs and a lack of endogeneity in testing AOs. Therefore, the use of an IV model may be necessary in testing SUDs and unnecessary in testing AOs. The F-statistics were large (F-statistic for percentage of Veterans discontinuing opioid therapy per VHA station: cohort testing AOs= 444.8; cohort testing SUDs= 387.68) indicating that discontinuation rates by VHA parent station was a strong IV.

Sensitivity Analysis including Veterans without Pain Scores in the 180 Days prior to the Index Date

By including Veterans who did not have pain scores recorded in the 180 days prior to the index date a total of 233,279 and 258,828 Veterans were retained in the cohorts assessing newly diagnosed SUDs and AOs following the implementation of the other inclusion and exclusion criteria, 192,351 continuing COT and 40,928 discontinuing opioid therapy in the cohort assessing SUD diagnosis and 214,328 continuing COT and 44,500 discontinuing opioid therapy in the cohort assessing AO diagnosis. After matching, 30,060 and 33,010 Veterans continuing COT matched to Veterans discontinuing opioid therapy in the cohorts testing SUDs and AOs. A total of 233,154 and 258,011 Veterans were retained in the SIPTW cohorts testing SUDs and AOs respectively after excluding Veterans in non-overlapping regions of the propensity score distribution, 192,234 Veterans continuing COT and 40,920 Veterans discontinuing opioid therapy in the cohort testing SUDs and 213,514 Veterans continuing COT and 44,497 Veterans discontinuing opioid therapy in the cohort testing AOs. The baseline covariates were well balanced after propensity score matching and weighting (all standardized differences <10%).

The matched analyses including Veterans without pain scores in the 180 days prior to the index date found that all SUDs were lower among discontinuers as compared to Veterans continuing COT (Composite SUDs: OR=0.867, 95%CI: 0.809, 0.928; OUD: OR=0.763, 95%CI: 0.634, 0.918; DUD: OR=0.855, 95%CI: 0.777, 0.941) except for AUD (OR=0.920, 95%CI: 0.846, 1.001). The SIPTW analyses found similar results. However, all SUDs were significantly lower for discontinuers as compared to Veterans continuing COT (Composite SUDs: OR=0.792, 95%CI: 0.757, 0.828; OUD: OR=0.704, 95%CI: 0.631, 0.785; DUD: OR=0.758, 95%CI: 0.712, 0.806, AUD: OR=0.858, 95%CI: 0.811, 0.908).

All types of AOs were lower among discontinuers as compared to Veterans continuing COT (Composite AOs: OR=0.696, 95%CI: 0.652, 0.743; accidents resulting in wounds/injuries: OR=0.594, 95%CI: 0.549, 0.642, opioid-related accidents and overdoses: OR=0.514, 95%CI: 0.330, 0.800, violence-related injuries: OR=0.559, 95%CI: 0.357, 0.874) except for alcohol and non-opioid medication related accidents and overdoses (OR=1.022, 95%CI: 0.890, 1.173) and self-inflicted injuries (OR=0.980, 95%CI: 0.852, 1.127). The SIPTW analyses found similar results. Although, like the SIPTW analyses for SUDs, all AOs were significantly lower for discontinuers as compared to Veterans continuing COT (Composite AOs: OR=0.636, 95%CI: 0.608, 0.666; accidents resulting in wounds/injuries: OR=0.556, 95%CI: 0.525, 0.589; opioid-related accidents and overdoses: OR=0.560, 95%CI: 0.424, 0.739; alcohol and non-opioid medication related accidents and overdoses: OR=0.858, 95%CI: 0.785, 0.939; self-inflicted injuries: OR=0.814, 95%CI: 0.739, 0.897; violence-related injuries: OR=0.606, 95%CI: 0.433, 0.847).

Discussion

This study assessed the association between discontinuing COT, after receiving COT for six months, and newly diagnosed adverse outcomes that are potentially related to opioid use in a Veteran population with chronic pain and found that approximately 15% completely discontinued opioid therapy in the ensuing six-month period. To evaluate the potential unintended effects associated with discontinuing opioid therapy within the first year of COT initiation, we evaluated newly diagnosed outcomes that might be potential risks influenced by changes in opioid therapy including multiple types of SUDs: OUD, DUD, and AUD as well as many types of AOs that are potentially opioid-related: accidents resulting in wounds or injuries, opioid-related accidents and overdoses, alcohol and non-opioid drug-related accidents and overdoses, self-inflicted injuries, and violence-related injuries.

None of our findings, using several analytic approaches, suggest that discontinuation of opioid therapy was associated with increases in new diagnoses for AOs such as overdoses or self-inflicted injuries within VHA. On the contrary, our primary analysis showed discontinuation was associated with decreases in accidents resulting in wounds and injuries, opioid-related accidents and overdoses, and violence-related injuries; the SIPTW analyses showed decreases in composite AOs and all individual AO types among discontinuers. Sensitivity analyses including Veterans who did not have pain scores in the 180 days prior to the index date found similar results. IV analyses also showed decreased risk for new diagnoses of composite AOs and accidents resulting in wounds and injuries with discontinuing opioids. In absolute terms, the potential reduction in new diagnoses for opioid related harms is significant; completely discontinuing opioid therapy was associated with a decreased risk of new diagnoses for the composite measure of AOs by approximately 30% compared to a similar group of persons continuing COT (6.3% vs 8.8%). However, it is possible that some of the lower rates of AOs could also be explained by less frequent monitoring of Veterans who discontinued opioids and in turn may be less likely to have the potential opioid adverse outcomes detected, particularly for outcomes like self-harm which could be detected during routine clinical encounters. Veterans that discontinued opioid therapy had on average 19.24 (SD=21.08) unique days of visits in the 365 days of follow up as compared to 25.94 (SD=22.71) observed in the cohort that continued COT.

Our findings and methods differ from those of other studies in the literature. Self-directed violence was found to be high among Veterans who discontinued COT (12%; n=59) (8); however, this study did not include a control group of Veterans who continued opioids. Another study among patients enrolled in an opioid registry found an increased risk of overdose death among patients who discontinued COT. However, this study was small and only within one primary care clinic; therefore, it is difficult to ascertain the representativeness of the cohort. (31) A study more similar to ours was recently published by Oliva et al. They used a national cohort of Veterans and found discontinuing opioid treatment was associated with an increased risk of death from overdose or suicide across all types of opioid users and more pronounced among persons taking opioids more chronically. (10) While our study and Oliva et al. both used a national cohort of Veterans, our studies differed in important ways. First, the Oliva et al. study used a Veteran population who had received at least one outpatient opioid prescription whereas our study only included persons with CNCP who were newly initiated on COT. Second, our outcome definitions differed. The Oliva et al. study used an outcome of death due to overdose or suicide while our study evaluated new diagnoses for overdose and suicide recorded by ICD-9-CM codes within VHA inpatient and outpatient visit files. Our data, which looked at the risk over a one-year period, found no increases in new diagnoses for AOs, including overdoses, within VHA, the setting in which the patients were prescribed opioids. Most of our analyses point to a decreased risk in treatment for a composite measure of AOs, which was driven, almost exclusively, by accidents resulting in wounds and injuries [1,114 Veterans continuing COT (6.4%) vs. 700 Veterans discontinuing opioid therapy (4.0%)]. Lastly, Oliva et al. used a Cox non-proportional hazards model with the main independent variable of interest being the interaction between the date of stopping opioid treatment and length of opioid treatment. Our analyses did not examine the temporal nature of risks, and it is possible that the AOs we studied could have increased immediately after discontinuation and later offset by subsequent reductions in risks in the months after discontinuation. Overall, the findings from the Oliva et al. study and this study are not directly comparable as the Oliva et al. study assessed death due to overdose/suicide and our study relied on diagnoses recorded at the VHA or providers that bill the VHA which may miss SUD or AO events not captured in VHA data (i.e., not in a VHA facility or paid for by VHA) and fatal overdoses in which no medical care was sought. To mitigate missed events for care received outside the VHA, we excluded Veterans who were not regular users of VHA care. We also excluded 4,201 Veterans that died from any cause in the 180 days after the index date to ensure proper assignment as having discontinued opioid therapy and an unknown fraction may have died of an overdose. Cause of death were unavailable when this study extracted data.

The association of discontinuing chronic opioid therapy and subsequent SUD is unclear based on our findings. Our primary analysis using propensity score matching found no statistically significant difference in new diagnoses for composite SUD nor DUD or AUD but did find that the odds of new diagnoses for OUD decreased by about 32%. The SIPTW models were consistent with the primary analysis in terms of point estimates except for AUD, but the relative reductions in new diagnoses for composite SUDs and each of the other individual SUD subtypes among discontinuers were statistically significant, likely because of the larger sample size. Unlike the primary analysis, the SIPTW models found lower rates for newly diagnosed AUD (OR=0.862), which was of a different direction from the primary analysis despite the primary analysis being null (OR=1.042). The IV models found a slight but significant increased risk of overall SUD, DUD, and AUD development with discontinuing opioid therapy.

Unfortunately, we found the analytic approach used had a material impact on our findings on the effect of discontinuation on SUD development. The propensity-score matched analysis resulted in well-balanced comparison groups on all factors but significantly reduced the sample size, particularly in the comparison group that continued COT. Fortunately, both the primary analysis based on propensity score matching, and the SIPTW analyses point to similar results for all SUD outcomes studied except AUD but differ in their precision where the SIPTW, with much larger sample sizes, yielded more significant findings. Though we controlled for a range of factors that could confound the relationship between discontinuation and diagnosis of SUDs within VHA including pain scores and changes in pain scores, neither of these approaches account for factors not observed in the data, such as the reason for opioid discontinuation which could be due to aberrant behaviors, improvements in the underlying pain condition, or other factors which not recorded. The IV approaches potentially account for unobserved confounding but relies on the restriction exclusion assumption that the IV is unrelated to outcomes, in this case SUDs in VHA center regions, except through influencing treatment choice. This may not be viable given the variation in SUD throughout the U.S. The propensity score approaches consistently pointed to an association between discontinuation and lower new diagnosis rates for OUD but no differences were found in the IV model. The impact of discontinuation on DUD and AUD are suggestive of a potential decrease in diagnoses of these SUDs but decreases were only observed in the SIPTW. Taken together, discontinuation of COT appeared to not increase new diagnoses for SUDs and potentially reduced the risk of being newly diagnosed with OUD. Additional studies are warranted to explore these associations further.

We are not aware of other studies that have directly examined the association between discontinuing opioids and subsequent newly diagnosed SUDs. A series of three studies have examined the reasons for clinician-initiated discontinuation of opioids and the subsequent SUD diagnosis using 600 Veterans who were discontinued from COT, 300 of which had SUDs and 300 were matched controls. A high percentage of discontinuers (28%) were discontinued due to positive urine drug tests (UDTs) for alcohol, cannabis, or other illicit or non-prescription controlled substances. (32) Veterans with SUD diagnoses were also more likely to discontinue COT because of aberrant behaviors (81% vs 68%), with a substantial driver being alcohol abuse, (33) and patients with UDTs positive for alcohol or illicit substances were more likely to be discontinued due to the positive UDTs as compared to patients who tested positive for non-prescribed prescription medications (adjusted OR = 13.10). (34) These studies illustrate that aberrant drug use behaviors are a frequent reason why clinicians may discontinue COT.

Limitations

Several limitations exist with this study. First, findings from these analyses in Veterans may not be generalizable to the civilian population due to their military background and because Veterans are mostly white males. Second, history bias may potentially be problematic for this study. History bias becomes a factor when the effects of relevant external events during study progression are not equal between the groups. (35) In regards to opioid therapy, many policies within and outside the VHA may influence transitional status over time. (1,36–39) However, since Veterans were matched based on the index date in the primary analysis, the effects of any external opioid policies should be balanced across the groups compared in our primary analyses, but the timing of these external effects were not directly accounted for in our sensitivity analyses. Third, using VHA data from CDW does not allow for obtaining information on opioid fills outside of the VHA. It is possible and likely that some Veterans sought opioid medications outside of the VHA system, particularly after being forced to discontinue COT. A recent study has found that 32% of Veterans on COT received concurrent non-VHA opioid prescriptions. (40,41) To mitigate receipt of opioids outside the VHA, our study required Veterans to have more visits to the VHA system than fee-for-service visits. Fourth, reasons for discontinuation are not known. For many patients, it is likely that discontinuation is due to improvements in the disease processes that are leading to pain, or use of other therapies such as surgical interventions that may not be gathered from the electronic medical record/measured in this study. It is also likely, that opioid misuse or a SUD was detected prompting discontinuation. Indeed, a recent study of Veterans evaluating the reasons for discontinuation of COT found that 85% were discontinued because of a clinician decision, not a patient decision. Of those 85%, 75% were discontinued because of opioid-related aberrant behaviors.(33) This creates potential temporal ambiguity where it is unclear if the development of a SUD or AO was the result or the cause of opioid discontinuation; however, this is unlikely to explain the largely null findings or reduced risk associated with discontinuation observed in this study when evaluating SUD. We attempted to account for this by excluding persons that experienced a AO or SUD in the baseline period, however, since the six-month window in which discontinuation or continued COT overlapped with the first six months of the outcome window, reverse causality cannot be ruled out and our findings should be interpreted as associations. Fifth, our exclusion criteria, which defined the final cohort for our analyses, were somewhat selective. For example, we excluded Veterans who died in the 180-day period after the index date, Veterans who switched to intermittent opioid therapy, and those with previous AOs if assessing AOs. We excluded Veterans who died in the 180-day period after the index date because we wanted to ensure appropriate classification of the patients (i.e., discontinuation of opioid therapy and not discontinuation due to death). We excluded those switching to intermittent opioid therapy to isolate the effects of true discontinuation, and we excluded those with a previous AO in assessing subsequent AOs to better ensure that an AO in the outcome period was a true, new AO and not an ICD-9-CM carryover from the baseline period. Sixth, the identification of AOs and SUDs were obtained only through medical record documentation from the VHA CDW; therefore, AOs and SUDs occurring outside of the VHA system or from care not paid for by the VHA are not captured in these data and could cause the observed outcome rates to be lower than actual outcome rates. Seventh, ICD-9-CM codes are intended to capture data for billing purposes, not research; therefore, the appropriate capture of diagnoses via ICD-9-CM codes may be subjective based on the clinician’s personal practice. Eighth, this study looked at opioid discontinuation after patients initiated one 180-day period of COT. Therefore, outcomes could differ by length of COT and the results may be different for Veterans who discontinue opioids after longer durations of COT. Ninth, it is plausible that geographic variation in AO and SUD rates could cause prescribers to discontinue opioids. If this is the case, the IV analysis results may not be valid as then the IV does not only effect the treatment but also the outcome. Therefore, the results of the IV analyses should be interpreted with caution. Tenth, continuation of COT is associated with more daily contacts to the health system, and therefore more opportunities to detect study outcomes, in particular SUDs. As stated previously, the number of unique days of visits in our cohorts for testing AOs were higher among those Veterans who continued COT (25.94, SD=22.71) as compared to those who discontinued (19.24, SD=21.08). The same was true for our cohorts testing SUDs (continued COT: 24.97, SD=21.08; discontinuers: 18.47, SD=19.12). Eleventh, our analyses did not include cause of death as we did not have death certificate data available to us when the data were extracted. This could underestimate the number of patients who experienced an overdose if the patient was not seen in a healthcare setting. So, we did a scenario analysis with our SIPTW cohort for testing AOs. A total of 147 patients died from any cause in the time period of 180 to 360 days after the index date. Increasing the number of opioid overdoses from 22 to 32 among the SIPTW cohort for testing AOs would switch the SIPTW model to a null funding. Therefore, 10 of the 147 deaths would need to be opioid-related to make the findings null, which is a 6.8% rate. Per the CDC, the death rate due to opioid overdoses for the general U.S. population ranged from 6.5 to 9 per 100,000 (0.0065–0.009%) over the years of the study (2009 to 2015). (42)

Conclusions

Among our cohort of non-cancer patients with chronic pain that initiate COT, our findings suggest that new diagnoses for opioid-related adverse outcomes, such as overdose, self-inflicted injuries, accidents resulting in wounds/injuries, may be lower among Veterans discontinuing opioid therapy after initially receiving COT for six months when compared to those continuing COT. The association between discontinuing opioid therapy and newly diagnosed SUDs is less clear; however, some of our findings suggest that those that discontinue COT may have lower rates of newly diagnosed OUD in VHA compared to those continuing COT. Further research is needed to better understand the effects of discontinuing COT.