Factors Associated With Initial Treatment Choice, Engagement, and Discontinuation for Patients With Opioid Use Disorder

Nearly 47,000 opioid-involved overdose deaths occurred in the United States in 2018 (1). Addressing the opioid crisis involves treating the approximately two million Americans who currently have opioid use disorder (2). Yet, according to the National Survey on Drug Use and Health, in 2018 less than one-third of people with opioid use disorder received treatment in any location for substance use in the past year. Pharmacotherapy with medications for opioid use disorder approved by the U.S. Food and Drug Administration (i.e., methadone, buprenorphine, and naltrexone) is effective (3). However, these pharmacotherapies are underutilized from a clinical perspective (4). The American Society of Addiction Medicine recommends that patients with opioid use disorder be treated with a combination of pharmacotherapy and psychosocial therapy (3). However, psychosocial therapy for opioid use disorder is also underused from a clinical perspective (5).

Empirical evidence on factors associated with the initiation of treatment for patients with opioid use disorder is limited. Evidence is particularly lacking on factors associated with patient initiation of treatment among different types of treatment (i.e., pharmacotherapy, psychosocial treatment, or both). Longer treatment of patients with opioid use disorder is associated with improved outcomes (6). Yet, little empirical evidence exists on factors associated with the duration of treatment for patients with opioid use disorder.

In this analysis, we used the OptumLabs Data Warehouse (OLDW), a large, national, deidentified database of commercial or Medicare Advantage (MA) plan enrollees, to identify patient and area factors associated with initial choice of treatment for opioid use disorder as well as patient engagement with the treatment and its duration. Findings from this analysis can be used to identify patients potentially at risk for not initiating treatment or underusing treatment and to provide them with additional resources and support.

Methods

In this retrospective study, we used deidentified administrative claims data with linked county-level Area Health Resources File data from OLDW. The database contains longitudinal health information on enrollees and patients, representing a mixture of ages, race-ethnicities, and geographical regions across the United States. The claims data in OLDW include medical, behavioral health, and pharmacy claims and enrollment records for commercial and MA enrollees. Because this study involved analysis of preexisting, deidentified data, it was not subject to institutional review board review.

Data Selection

We retrieved retrospective data for enrollees with opioid use disorder who had claims for health care services between July 1, 2010, and April 1, 2019. Opioid use disorder was defined as having at least one medical claim with a diagnosis for opioid use, abuse, dependence, poisoning, or adverse effects (in any diagnosis position) or as having evidence of pharmacotherapy. Enrollees included in the study on the basis of receipt of pharmacotherapy had to have at least one claim for administration of buprenorphine or methadone, one pharmacy fill for a buprenorphine prescription, or one fill for naltrexone prescription or medical claim for injectable naltrexone with a medical claim for opioid use disorder (in any diagnosis position) within 90 days of the naltrexone claim. Our definition of pharmacotherapy excluded formulations of buprenorphine for pain treatment and included only claims for methadone administration by a licensed program or for substance abuse treatment.

All enrollees with opioid use disorder were required to be continuously enrolled in a commercial or MA plan with medical, pharmacy, and behavioral health coverage for at least 6 months before the first opioid use disorder or pharmacotherapy claim and for at least 3 months after treatment start or the first opioid use disorder claim if the enrollee did not receive treatment. Moreover, study inclusion criteria required the enrollee to have no claims for opioid use disorder or pharmacotherapy in the 6 months before their first opioid use disorder or pharmacotherapy claim; to be age ≥18 years; and to have complete age, gender, geographic region, and health plan membership information. Data for enrollees with evidence of pregnancy, cancer, ≥90 days of long-term care, or hospice or palliative care were excluded from the study. A patient’s index date was set as the earliest claim for opioid use disorder or pharmacotherapy during the identification period that met all of the sampling criteria. If a patient’s index claim occurred during an inpatient stay, the index date was set to the admission date for that stay.

Initial treatment status and type were assessed within 30 days of a patient’s index date. On the basis of whether treatment occurred and the first type of treatment observed, patients’ initial treatment type was categorized into one of four groups: no treatment, pharmacotherapy only, psychosocial treatment only, and both pharmacotherapy and psychosocial treatment. Patients with an initial claim for one of the types of treatment (e.g., psychosocial) and a claim within the following 2 weeks for the other (e.g., pharmacotherapy) were coded as “both.” Pharmacotherapy was assessed on the basis of pharmacy and medical claims for buprenorphine, naltrexone, or methadone (for a list of codes used to identify psychosocial treatment, see an online supplement to this article). Detoxification encounters were not regarded as treatment unless they were inseparable from the treatment types identified (7).

Initial treatment duration was calculated as the number of days between initial treatment start and the end date of the initial treatment episode and included both dates. A gap in either psychosocial treatment or pharmacotherapy during which no other treatment was in effect or initiated signaled the end of the initial treatment episode (for our definition of gaps in treatment, see the online supplement). Finally, we distinguished between those who initiated treatment but did not continue and those who engaged in treatment. To be classified as engaged, patients had to have an initial treatment episode of at least 30 days with two or more claims for either pharmacotherapy or psychosocial treatment (5).

Statistical Analysis

We conducted all analyses with SAS, version 7.13. We applied a multinomial logistic regression model to the data for all patients who met the study inclusion criteria, including those who did not initiate treatment, to estimate the association between patient clinical, demographic, and area characteristics and treatment type. Separate results are provided for each of the treatment categories (pharmacotherapy only, psychosocial treatment only, and both pharmacotherapy and psychosocial treatment), with odds ratios indicating the characteristics associated with each treatment type relative to no treatment.

We then assessed treatment duration by using a two-part modeling approach. First, we ran a logistic regression model (LOGISTIC procedure) to estimate the odds of treatment engagement. That analysis was limited to patients who initiated treatment. It generated odds ratios that estimated the association between patient clinical, demographic, and area characteristics and patient engagement with treatment relative to no engagement with treatment. Second, we used unadjusted Kaplan-Meier curves and a Cox proportional hazards regression model to estimate the hazard ratios for treatment discontinuation (PHREG procedure). This analysis was limited to patients who engaged in treatment. Because there may have been unmeasured patient factors that led to longer treatment retention, requiring patient data to cross a clinically meaningful threshold to be included in the duration analysis had the advantage of partially addressing the potential threat to validity from endogeneity. This analysis yielded hazard ratios that estimated the association between patient clinical, demographic, and area characteristics and patient discontinuation of treatment. (For how the proportionality assumption was evaluated, see online supplement.)

We followed the literature and controlled for patient clinical, demographic, and area characteristics that are known to be associated with treatment type for opioid use disorder (6, 8, 9). All models adjusted for patient age, sex, and insurance type. All models also adjusted for patient Census region, county mental health professional shortage area status, county rates of poverty and higher education completion, and the proportion of Whites in the county population. Models also adjusted for baseline opioid use (including morphine milligram equivalent [MME] and use of multiple opioid types), overlapping opioid and benzodiazepine use, indicators of pain, and infections from possible injection drug use (for codes used to identify infections from possible injection drug use, see online supplement). We included variables for medical comorbid conditions on the basis of select Chronic Conditions Data Warehouse condition categories from the Centers for Medicare and Medicaid Services (10). These covariates included attention-deficit hyperactivity disorder (ADHD) and conduct disorder, anxiety disorder, blood disorder, circulatory system disorder, endocrine system disorder, eye disorder, fibromyalgia, kidney disease, liver disease, mood disorder, musculoskeletal disorder, nervous system disorder, psychotic disorder, and substance use disorder (other than opioid use disorder). The treatment engagement and duration models also included initial treatment type.

Results

In total, 204,225 patients met inclusion criteria; their demographic and baseline clinical characteristics are shown in Table 1. Only 35.0% (N=71,571) of the patients initiated treatment within 30 days (for treatment initiation by study inclusion criteria, see online supplement). Patients who started treatment were more likely to be male, be younger than 55 years, have commercial coverage, and live in the Northeast and Midwest Census regions. Patients who initiated treatment were less likely to have overlapping opioid prescriptions and overlapping opioid and benzodiazepine prescriptions than those who did not initiate treatment; they were also less likely to have several general medical comorbid conditions, including pain, circulatory or endocrine system disorder, fibromyalgia, musculoskeletal disorder, kidney disease, or a nervous or respiratory system disorder. Among the patients who did initiate treatment, psychosocial treatment without pharmacotherapy was the most common initial treatment regimen (70.1%, N=50,178), followed by pharmacotherapy only (15.9%, N=11,378) and both pharmacotherapy and psychosocial treatment (14.0%, N=10,015).

Associations between patient clinical, demographic, and area characteristics and each type of treatment initiated are shown in Table 2. Separate results are provided for each of the treatment categories (pharmacotherapy only, psychosocial treatment only, and both pharmacotherapy and psychosocial treatment) relative to no treatment.

Table 2 presents associations between the characteristics of patients who initiated treatment with pharmacotherapy only relative to those who did not initiate treatment. Those who were male or in the youngest age category had higher odds of initiating pharmacotherapy only compared with those who were female or older than 34 years, respectively. Those who filled moderate- or high-MME opioid prescriptions at baseline also had higher odds of initiating pharmacotherapy only than those who filled low-MME or no opioid prescriptions. Some behavioral health comorbid conditions were associated with higher odds of initiation of pharmacotherapy only (i.e., anxiety, mood, or substance use disorder). Patients with psychotic disorder had lower odds of initiating pharmacotherapy only compared with those who did not have that condition in the baseline period. In addition to baseline pain, some comorbid general health conditions were associated with lower odds of initiation of pharmacotherapy only (e.g., fibromyalgia; blood disorder; and circulatory, endocrine, musculoskeletal, nervous, or respiratory system disorders).

Table 2 also presents associations between the characteristics of patients who initiated psychosocial treatment only relative to those who did not initiate treatment. Patients who were male or in the youngest age group also had higher odds of initiating psychosocial treatment only compared with those who were female or older than 34, respectively. Patients living in a mental health shortage area had lower odds of initiating psychosocial treatment only than patients who did not live in a mental health shortage area. In addition to comorbid behavioral health conditions that were also associated with initiation of pharmacotherapy only (i.e., anxiety, mood, or substance use disorder), ADHD, conduct disorder, and psychotic disorder were associated with higher odds of initiating psychosocial treatment only. Many of the same general health comorbid conditions, such as fibromyalgia; blood disorder; and circulatory, endocrine, musculoskeletal, or respiratory system disorders, that were associated with lower odds of initiation of psychosocial treatment only were also associated with lower odds of initiation of pharmacotherapy only, as was baseline pain.

In addition, Table 2 presents associations between the characteristics of patients who initiated both treatment types relative to those who did not initiate treatment. Being male or in the youngest age category was associated with higher odds of initiating both pharmacotherapy and psychosocial treatment compared with those who were female or older than 34, respectively. Being in a mental health shortage area was associated with lower odds of initiating both types of treatment. Several comorbid behavioral health conditions, such as anxiety, mood, or substance use disorder, were associated with higher odds of initiating both types of treatment. Several comorbid general health conditions, including fibromyalgia; blood or eye disorder; kidney disease; and circulatory, endocrine, musculoskeletal, nervous, or respiratory system disorders, were associated with lower odds of initiating both types of treatment, as was baseline pain.

Of the patients who initiated treatment, 69.1% (N=49,476) also subsequently engaged in treatment. Table 3 shows the results of an analysis of factors associated with treatment engagement. Men were more likely to engage in treatment than women, and patients living in the Northeast were more likely to engage than those in the South. Older patients had lower odds of engaging in treatment, and this outcome was particularly the case for patients age ≥65 years. Certain comorbid behavioral health conditions (i.e., anxiety, mood, psychotic, or substance use disorder) were associated with higher odds of treatment engagement. Individuals who had evidence of infections from possible injection drug use at baseline were also more likely to engage in treatment than those who had no such baseline infections. In contrast, having overlapping opioid prescriptions and several comorbid general health conditions (e.g., endocrine or musculoskeletal system disorder, pain, or fibromyalgia) were associated with lower odds of treatment engagement. We found that patients who initiated treatment with pharmacotherapy only or with both pharmacotherapy and psychosocial treatment were more likely to engage in treatment than were patients who started treatment with psychosocial therapy only. In fact, patients who initiated both pharmacotherapy and psychosocial treatment had seven times the odds of engaging with treatment compared with those who initiated psychosocial therapy only.

Among those who engaged in treatment, patients who initiated psychosocial therapy only, pharmacotherapy only, and a combination of pharmacotherapy and psychosocial therapy had a median Kaplan-Meier–adjusted treatment duration of 36, 169, and 144 days, respectively. Table 4 presents the results of the Cox proportional hazards model, where the outcome was discontinuation of treatment among patients who engaged in treatment (the Kaplan-Meier curves are shown in the online supplement). Patients who initiated treatment with pharmacotherapy only or a combination of pharmacotherapy and psychosocial treatment were more likely to engage in treatment and also had a lower hazard of discontinuing treatment compared with patients who initiated psychosocial treatment only.

Men had a higher hazard of discontinuing treatment than women. Among the different age categories, patients who were ages 18–34 had the highest hazard of discontinuing treatment. Patients with high MME opioid use or infections from injection drug use during the baseline period had a higher hazard of discontinuing treatment, compared with patients without such use or infections. Patients with several behavioral health comorbid conditions had a lower hazard of discontinuing treatment (i.e., ADHD and conduct, anxiety, or mood disorder), compared with patients without these conditions. However, patients with substance use disorder had a higher hazard of discontinuing treatment than those without this disorder. Patients with circulatory system disorder had a higher hazard of discontinuing treatment than those without this disorder.

Discussion

In this study population, slightly more than one-third of patients initiated treatment within 30 days. Most of the patients who initiated treatment did not use pharmacotherapy, and nearly one-third of patients who started treatment did not engage in treatment. The median treatment duration was longest for patients receiving pharmacotherapy only, followed by those receiving both pharmacotherapy and psychosocial treatment, and then those receiving psychosocial treatment only; however, the median duration for all treatment types was less than the 12 months recommended by the American Society of Addiction Medicine.

Relative to initiating psychosocial treatment only, initiating pharmacotherapy only or initiating both pharmacotherapy and psychosocial treatment was associated with higher odds of engaging in treatment and a lower hazard of discontinuing treatment. Among those who engaged in treatment, the median Kaplan-Meier–adjusted duration was longer for patients who initiated pharmacotherapy only or both pharmacotherapy and psychosocial therapy. Although pharmacotherapy alone and combined pharmacotherapy and psychosocial treatment were associated with higher odds of engagement and longer treatment duration, they were the least common starting treatment regimens, representing only 29.9% (N=21,393) of those who initiated treatment. In addition to relatively few patients initiating treatment, many began with psychosocial treatment only, which was associated with lower engagement and shorter treatment duration. Longer treatment is associated with improved patient outcomes and is a benefit of initiating treatment with evidence-based pharmacotherapy (6). The difference in the median duration of treatment observed for patients receiving pharmacotherapy only compared with those receiving both treatment types may be an important topic for further study.

Several patient characteristics, including behavioral and general health comorbid conditions, were associated with the decision to initiate treatment and the type of treatment initiated. Providers and payers may wish to provide additional support to patients whose characteristics are associated with lower odds of treatment initiation or initiation of treatment with psychosocial treatment only. Those characteristics include conditions that may be associated with pain (e.g., baseline pain, fibromyalgia, and musculoskeletal system disorder). Older age and comorbid conditions associated with older age were also associated with lower odds of initiation of evidence-based pharmacotherapy treatment. Older age can be associated with increased pain, and the experience of pain may make patients hesitant to pursue treatment for opioid use disorder because of the relief from pain that opioids can provide.

Patients with certain behavioral health comorbid conditions, including anxiety and mood disorders, had higher odds of initiating and engaging in treatment; they also had a lower hazard of discontinuing treatment. Diagnosis with these conditions may suggest previous contact with the behavioral health treatment system, which, in turn, may indicate the importance of a connection with the treatment system in initiating, engaging, and remaining in treatment for opioid use disorder. Although patients with baseline substance use disorder had higher odds of initiating and engaging in treatment, they were more likely to discontinue treatment. This finding may reflect the struggle of remaining in treatment that patients with more severe substance use face. Patients with infections due to possible injection drug use in the baseline period also had a greater hazard of discontinuing treatment, indicating the challenges of remaining in treatment that patients with more severe opioid use disorder may encounter.

This study contributes to the literature by identifying patient characteristics associated with initiation of treatment for opioid use disorder, the type of treatment initiated, and engagement in and duration of treatment. A strength of the study was inclusion of a large number of patients from across the United States from commercial and MA plans. Its limitations included the lack of clinical detail in claims data and the possibility that unmeasured characteristics, including the severity of conditions, were associated with treatment initiation, engagement, and duration. However, this limitation is common in retrospective studies that are based on claims data. It is possible that some patients classified as having opioid use disorder in this study, such as those with adverse event codes, were not candidates for treatment. In this study, we used relatively wide inclusion criteria in view of the underdiagnosis of opioid use disorder (11) and because patients who experience opioid-related adverse events may benefit from opioid use disorder treatment (12). The results of our study may not generalize to people with types of insurance not included in the analysis, including Medicaid, or to those without insurance.

Conclusions

In this study, we identified several patient characteristics associated with initiation of opioid use disorder treatment and the type of treatment started. Providers and plans may wish to provide extra support to patients whose characteristics are associated with not initiating treatment or initiating treatment without pharmacotherapy. Our results show that patients who initiate pharmacotherapy only or pharmacotherapy and psychosocial therapy are more likely to engage in treatment and are less likely to discontinue treatment than patients who initiate psychosocial treatment only. Patients who remain in treatment longer have better outcomes (6), and the association of initial treatment type with treatment engagement and longer treatment duration may further support the use of evidence-based pharmacotherapy and the combination of pharmacotherapy and psychosocial treatment for opioid use disorder.