This is an observational cohort study, analyzing comprehensive 18-month longitudinal data funded by the US Centers for Disease Control and Prevention.^20,21 Patients aged ≥18 years with OUD receiving treatment from 62 participating outpatient OUD treatment facilities across 16 states were referred to this study from March 2018 through December 2019.²¹ The specific OUD treatment a patient was receiving at the treatment facility when screened for study eligibility is referred to as the index treatment. The treatment facility where patients received their index treatment is referred to as the index facility. The original study protocol only recruited patients having started the index treatment at the index facility within the first 30 days. However, due to lower-than-expected patient recruitment numbers, the protocol was expanded with Institutional Review Board (IRB) approval to also include patients who were already receiving the index treatment for more than 30 days at the time of their enrollment into the study.²¹ This study was approved by the Office of Management and Budget and an independent IRB.

Participants screened eligible underwent an informed consent process before study enrollment (Supplemental Figure 1). Five-wave questionnaires were used to collect data from the enrolled patients between March 2018 and May 2021: a baseline (ie, study enrollment) self-reported questionnaire, and 4 follow-up self-reported questionnaires at approximately 3-, 6-, 12-, and 18-months post-baseline. Patients were followed from baseline until they completed the 18-month post-baseline questionnaire, disenrolled (patients were retained in the study if they missed a follow-up questionnaire unless they rescinded their study enrollment) or the study period ended (May 31, 2021), whichever occurred first. In addition, since patients may end the index treatment or used other MOUD while on the same index treatment between the scheduled questionnaires, this study only followed the patients before the scheduled questionnaire time when patients reported stopping the index treatment or taking other MOUD while on the same index treatment, whichever came first.

The exposure in this study is index buprenorphine (oral, implantable, or injectable) or index methadone treatment since baseline. Patients reporting naltrexone or counseling-only as their index treatment were excluded due to a lack of power for statistical significance from low enrollment numbers. Patients were excluded if, at baseline, they reported (1) cessation of the index treatment, or (2) taking any MOUD other than their index treatment in the 12 months before baseline or during the period between the time index treatment began and baseline, whichever was longer (Supplemental Figure 2). “Buprenorphine” and “methadone” in subsequent text refers to the patient’s index treatment received at the index facility and not any subsequent MOUD received at other facilities.

All 5 patient questionnaires from baseline to 18-month post-baseline were used in this study and captured patient characteristics such as demographics (eg, race/ethnicity and sex), index treatment (buprenorphine or methadone), stopping the index treatment (Yes/No), and past 30-day substance use.

In addition to patient questionnaires, facility directors completed questionnaires when their index facility entered the study (ie, baseline) and again 18-months post-baseline. The baseline treatment facility questionnaire was used in this study to capture information on the facility’s primary healthcare focus, geographic region, and patient capacity. State policies related to MOUD access where the facilities were located were also included. (State policies contain state laws from all 50 states and Washington, D.C., enacted or in effect between 2018 and 2021. The laws contained demonstrate the scope of state laws that may affect access to MOUD.) Policy-related information was evaluated, created, and validated using 2 primary databases—LexisNexis and WestLaw. More information about recruitment protocols, data collection, and patient or treatment facility questionnaires has been documented elsewhere.²¹

The outcome of interest, collected throughout the 5-wave patient questionnaires, was past 30-day nonprescribed opioid use (Yes/No). Nonprescribed opioid use was defined as either nonmedical use of a prescription opioid, heroin use, or IMF use.

Covariates were chosen based on their likelihood of having an association with nonprescribed opioid use, including patient-level covariates such as demographics, comorbidities, adverse childhood experiences (ACEs), number of nonprescribed opioids used before baseline, treatment duration between the time index treatment began and baseline, treatment duration since baseline, as well as facility-level covariates such as facility region and patient capacity.^22–26 State policy covariates included were “summative ranking of state legislation to reduce prior authorization barriers to MOUD” and “high State Opioid Response (SOR)/State Targeted Response to the Opioid Crisis (STR) grant funds going to state” (specific information about aforementioned covariates are in Supplemental Table 1).

Descriptive statistics were used to summarize the patient demographics and index treatment facilities among patients taking buprenorphine or methadone, separately.

The data were hierarchical with 3 levels: level 1 included the responses within each of the baseline, 3-, 6-, 12-, and 18-month questionnaires; level 2 was the clustering of responses by patients; and level 3 was the clustering of patients by treatment facilities. A multivariable, multilevel logistic regression was conducted separately among patients taking buprenorphine or methadone to quantify the association between each additional month of treatment with buprenorphine or methadone and past 30-day nonprescribed opioid use since baseline. Each model included 2 random intercepts (patient and treatment facility) and a set of covariates described above. This approach accounts for within-patient correlations among the repeated measures from multiple waves of questionnaires and allows for variation among different patients and treatment facilities.²⁷

Each multilevel model used a binomial distribution with a logit link to estimate adjusted odds ratios (aORs) and 95% confidence intervals (CIs; Supplemental Appendix 1). Model parameters were interpreted as subject-specific estimates instead of population averages. Intraclass correlations for treatment facility were examined to evaluate the variation in the outcome accounted for by the treatment facilities for each multilevel model.²⁷

The percentage of patients completing each questionnaire was 100% at baseline, 73% at 3 months, 71% at 6 months, 52% at 12 months, and 55% at 18 months postbaseline, respectively (Supplemental Table 2). Patient relocation, the COVID-19 pandemic, waning interest in the study without asking for disenrollment, or death may account for the loss-to-follow-up.²¹ Our multilevel models addressed missing data by maximum likelihood methods, which produced approximately unbiased estimates under the missing at random assumption after accounting for the covariates in the model.^28–32

Variance inflation factors were used to assess the multi-collinearity between the covariates.³³ No multicollinearity was detected in the final models. Two-tailed tests with a significance level of .05 were defined. Statistical analyses were conducted using SAS (version 9.4; SAS Institute).

Sensitivity analyses were conducted to examine the robustness of the main results. First, we increased the time period for nonprescribed opioid use from past 30 days to past 90 days (this outcome also included nonmedical use of MOUD). Second, we reclassified patients who continued the same MOUD as the index treatment at another non-index facility as continuing their index treatment. Third, we used a last-observation-carried-forward approach (for the third and fourth sensitivity analyses, we also used the last-observation-carried-forward approach to impute missing time-varying covariates for those who were lost-to-follow-up) to impute missing outcomes for those who were lost-to-follow-up; these individuals were assigned their last measured outcomes during subsequent rounds of follow-up. Fourth, we assumed a worst-case scenario for the outcomes that individuals who were lost-to-follow-up used the nonprescription opioids in the past 30 days and remained this outcome for all subsequent rounds.³⁴

Among 1974 enrolled adult patients with OUD, 1138 patients were identified as taking buprenorphine or methadone at baseline, with 353 (31%) patients taking buprenorphine (over 96% reported taking oral buprenorphine alone) and 785 (69%) patients taking methadone (Figure 1). At baseline, among patients treated with buprenorphine, middle-aged people (mean: 39 [standard deviation, SD: 11] years), females (64%), non-Hispanic White (72%), and those diagnosed with a mental illness (71%) accounted for the majority. Among patients treated with methadone, middle-aged people (mean: 42 [SD: 12] years), males (50%), non-Hispanic White (64%), and those diagnosed with a medical disease (69%) accounted for the majority. Of both groups, 86% had ACEs and around 50% had Medicaid only insurance coverage (Table 1).

The mean last MOUD treatment dose taken by patients at baseline was 11 (SD: 7) mg for buprenorphine and 93 (SD: 46) mg for methadone. The mean duration between the initiation of index treatment and baseline was 13 (SD: 24) and 25 (SD: 44) months for patients taking buprenorphine and methadone, respectively. The mean time patients continued their index treatment since baseline was 8 (SD: 7) months for buprenorphine and 10 (SD: 7) months for methadone (Table 1).

Among index facilities treating patients with buprenorphine, 44% focused primarily on substance use services, 19% offered both mental health and substance use services, and 16% focused on general health services. Nearly three-fourths (74%) of these facilities were private nonprofit organizations and two-thirds (66%) provided trauma-informed care. In total, 65% of the index facilities were in the South.

Among index facilities treating patients with methadone, 93% focused primarily on substance use services, 54% were private for-profit organizations, and 54% provided-trauma informed care. One-third (33%) of the index facilities were located in the West. The mean reported distance for patients traveling to their index facility was 10 (SD: 13) and 8 (SD: 11) miles for buprenorphine and methadone, respectively (Table 2).

After patient- and facility-level factors were controlled for, there appears to be an association between each additional month of MOUD treatment since baseline and decreased odds of past 30-day nonprescribed opioid use. Each additional month of MOUD treatment was associated with a 25% decrease in the odds of past 30-day nonprescribed opioid use for patients treated with buprenorphine (aOR = 0.75; 95% CI = 0.68-0.83) and a 17% decrease in the odds of past 30-day nonprescribed opioid use for patients treated with methadone (aOR = 0.83; 95% CI = 0.79-0.87; Table 3).

From baseline to 18-month follow-up, the cumulative percent of patients who reported discontinuation of their index treatment was 24% and 17% for patients taking buprenorphine and methadone, respectively.

Four sensitivity analyses corroborated our findings in the main analysis, indicating that past 30-day nonprescribed opioid use appeared to decrease with each additional treatment month for both MOUD treatments. The range for decreased odds of past 30-day nonprescribed opioid use with each additional treatment month appeared to be 15% to 30% for patients treated with buprenorphine, and 7% to 23% for patients treated with methadone across 4 sensitivity analyses (Supplemental Tables 3–6).

For both buprenorphine and methadone groups, factors significantly associated with decreased odds (95% CI) of past 30-day nonprescribed opioid use included: Each additional month of MOUD treatment prior to baseline (aOR = 0.93 [0.89-0.97]; aOR = 0.97 [0.97-0.98]), having high social support (aOR = 0.48 [0.26-0.88]; aOR = 0.44 [0.28-0.70]) versus moderate social support, and compliance with MOUD treatment regimen (aOR = 0.29 [0.14-0.60]; aOR = 0.42 [0.24-0.74]). In addition, among individuals in the buprenorphine treatment group, being employed (aOR = 0.51 [0.27-0.95]), attending a self-help group (aOR = 0.45 [0.24-0.83]), and residing in a state receiving higher grant funding to combat the opioid crisis (aOR = 0.37 [0.16-0.84]) were associated with decreased odds of nonprescribed opioid use (Table 3).

Factors significantly associated with increased odds (95% CI) of past 30-day nonprescribed opioid use for individuals treated with either buprenorphine or methadone included: completing questionnaires during the COVID-19 pandemic (aOR = 9.29 [2.96-29.17]; aOR = 3.19 [1.74-5.86]) and a self-reported adverse reaction to MOUD (aOR = 3.07 [1.11-8.48]; aOR = 2.51 [1.01-6.22]). In addition, among those treated with buprenorphine, patients having suicidal thoughts in the past year (aOR = 2.46 [1.22-4.96]) were associated with increased odds of non-prescribed opioid use. While among individuals treated with methadone, those reporting nonmedical use of at least 2 types of nonprescribed opioids (aOR = 6.03 [1.39-26.15]) before baseline were associated with increased odds of nonprescribed opioid use (Table 3).

Among patients undergoing outpatient OUD treatment with buprenorphine or methadone up to 18 months following baseline, those who continued treatment for each additional month appeared to be more likely to report 17% to 25% decreased odds of past 30-day nonprescribed opioid use including IMF. Continued efforts at the patient, clinician, health system, and practice levels are needed to improve MOUD treatment retention. This work also underscores several key factors associated with nonprescribed opioid use that can help inform future prevention efforts to prevent nonprescribed opioid use. Our findings can be used by clinicians, patients, and their family members to improve MOUD retention.