Impact Of Long-Term Buprenorphine Treatment On Adverse Health Care Outcomes In Medicaid

Abstract

The optimal, or even minimum, duration of medication treatment for opioid use disorder (OUD) needed to improve long-term outcomes has not been established empirically. As a result, health plans set potentially restrictive treatment standards to guide benefits and payment. To address this gap, we used a National Quality Forum measure for OUD medication treatment duration (180 days) to examine the impact of longer treatment on health care outcomes within a key population of Medicaid enrollees. Compared to buprenorphine discontinuation around the National Quality Forum benchmark (six to nine months), longer treatment (at least fifteen months) was associated with relative reductions in the risk of having all-cause inpatient (−52 percent) and emergency department (−26 percent) use, opioid-related hospital use (−128 percent), overdose events (−173 percent), and opioid prescriptions (−120 percent) and in the rate of prescription opioid use (−124 percent). We argue that these clinical benefits provide a rationale for policies that increase access to longer-term buprenorphine treatment, including lengthening the standards for minimum treatment duration.

The opioid epidemic is a leading public health priority with elevated overdose fatalities^1,2 and increasing medical treatment for overdose and other opioid-related events in inpatient and emergency settings.^3,4 This is in spite of growing recognition of opioid-related problems and some progress in increasing access to and delivery of opioid use disorder (OUD) treatment. Extensive evidence has established the safety and effectiveness of medications to treat OUD. Specifically, buprenorphine is associated with reductions in the risk of opioid use,^5–7 hospital-based care,^8,9 mortality,¹⁰ and treatment dropout.^5–8 Though retention rates are low,¹¹ studies suggest that the benefits of buprenorphine are stronger with longer-term treatment.^7,9,12–14

The longest clinical trial to compare different buprenorphine treatment durations showed that fourteen-week episodes are superior to seven-week episodes.⁷ Although observational research has examined outcomes associated with longer-term treatment,^9,12–14 these studies were not designed to establish a minimum duration of treatment to improve outcomes. Acknowledging the limited research on optimal treatment length,^15,16 the expert consensus recommends that medication be provided as long as patients “benefit from it and wish to continue.”¹⁵

Quality measures assess whether treatment aligns with the evidence base and are used to set standards of care. Federal and state agencies, health plans, and consumers use these measures to compare care quality and evaluate clinical performance. In the period 2010–17, health care organizations and professional associations defined new measures for minimum OUD medication treatment duration using 30-day (American Psychiatric Association), 90-day (Veterans Health Administration),¹⁷ or 180-day (National Quality Forum) benchmarks.¹⁸ The most recent measure (and the one with the longest duration), endorsed by the National Quality Forum in 2017, was based on the typical clinical trial duration of three to six months.¹⁸

From the insurer perspective, these measures also inform benefit and payment design to encourage quality improvement and contain costs as part of broader efforts to increase the value of health care. Alongside development of national quality measures, public^19,20 and private²¹ insurance policies have modified coverage for buprenorphine treatment. For example, nearly all state Medicaid programs have eliminated lifetime limits on buprenorphine treatment, in part responding to final regulations that implemented the Paul Wellstone and Pete Domenici Mental Health Parity and Addiction Equity Act of 2008.^20,22,23 However, other benefit design elements may complicate continuous buprenorphine treatment by restricting prescribing practices (for example, setting limits on buprenorphine quantity or dosing) and treatment modalities (such as requiring the use of preferred drug lists, step therapy, or psychosocial or behavioral services) or requiring administrative approval (for example, prior authorization) that increases provider burden and may lead to treatment delays or discontinuation.²⁰

As the single largest US payer of addiction treatment,²⁴ Medicaid plays a key role in ensuring access to OUD medication.^24,25 All state Medicaid programs cover buprenorphine,¹⁹ which accounts for most of the annual spending on OUD medications.²⁶ Moreover, the rate of buprenorphine prescriptions among Medicaid enrollees increased from 2010 to 2017,²⁶ which underscores the importance of buprenorphine in community-based OUD treatment and the increased investment by Medicaid to support the growing demand for care. The increased treatment rates may have partially resulted from recent changes in benefit design to improve access to buprenorphine. For example, from the period 2011–13 to 2018 seventeen states removed prior authorization requirements for buprenorphine-naloxone.^20,27 Evidence from this population is critical to identify treatment trends that may facilitate improvements in patient outcomes.

Using the 180-day National Quality Forum measure as a reference point for minimum treatment duration, we compared clinical outcomes of Medicaid enrollees with continuous treatment (for at least fifteen months) to outcomes of those who discontinued buprenorphine after six to nine months. We hypothesized that continuous treatment would be associated with lower risk of adverse all-cause and opioid-specific health care outcomes.

Study Data And Methods

Data Source

We analyzed insurance claims for 2013–17 from the MarketScan Multi-State Medicaid Database, which includes enrollment information and comprehensive health services and prescription drug data that are collected longitudinally for millions of enrollees.

Study Population

The sample included adults with continuous Medicaid enrollment from six months (180 days) before buprenorphine initiation through six months of follow-up who were ages 18–64 at treatment initiation. Using the National Quality Forum measure to define minimum treatment duration,¹⁸ we identified adults with treatment episodes of at least 180 days. We excluded people who were eligible for Medicare or whose behavioral health or prescription data were otherwise not captured in the database. People with gaps between prescriptions that resulted in their having a supply of buprenorphine for fewer than 80 percent of the days during their treatment episode (that is, for fewer than 144 days out of 180) were excluded because this threshold for the proportion of days covered is a quality measure used by the Centers for Medicare and Medicaid Services for other chronic conditions with outcomes that depend on treatment continuity.²⁸

The comparison group (whose members discontinued treatment) included people who discontinued buprenorphine 180–270 days after treatment initiation and were observed for 180 days of follow-up beyond discontinuation. The treatment group (whose members continued treatment) included people with buprenorphine episodes of at least 450 days, to ensure that all of the group’s members were treated throughout the follow-up period (360–450 days). Online appendix exhibit A1 shows the study design and sample definition.²⁹

We identified buprenorphine treatment using National Drug Codes from prescription claims. The index treatment date was defined by an initial claim that followed at least sixty days with no buprenorphine prescriptions. Treatment duration was measured using the number of days’ supply on buprenorphine claims. Similar to the case in prior studies, discontinuation was defined as going more than thirty days without a buprenorphine supply,³⁰ based on evidence that risk of adverse opioid-related outcomes is highest in the first four weeks after discontinuation.¹⁰ The discontinuation date was defined conservatively as the date of the final prescription. Following the model of clinical trials, we employed an intent-to-treat analysis that attributed outcomes to buprenorphine even if enrollees switched to another medication after discontinuation.

Outcome Measures

We assessed five main outcomes in thirty-day intervals during treatment and follow-up: all-cause inpatient use, all-cause emergency department use, opioid-related hospital (inpatient and emergency department) use, opioid and nonopioid drug overdose events, and opioid prescriptions. Many patients who initiate buprenorphine, including one-third³¹ to one-half₃₀ of Medicaid samples, also use other prescription opioids.³² Concurrent prescription opioid use, especially extensive use, could signal treatment problems such as lack of care coordination or inadequate dosage to control cravings and withdrawal symptoms. Therefore, we evaluated a subsample with any opioid prescriptions and their matched pairs (see the “Statistical Analysis” section) for the rate of prescription opioid use, measured as the number of days’ supply in each thirty-day period.

Inpatient services were measured using case identification codes. Emergency department services were measured using category of service codes. Opioid-related hospital services were defined according to methods from the Agency for Healthcare Research and Quality,³³ using primary and secondary diagnosis codes. Opioid and nonopioid drug overdose events were defined according to a Centers for Disease Control and Prevention guide,³⁴ using primary and secondary diagnosis codes. Prescription opioids were identified using National Drug Codes, excluding antitussives and injection-administered opioids.

Sociodemographic And Clinical Covariates

Patients were matched on baseline sociodemographic and clinical characteristics measured in the 180-day period before buprenorphine initiation. Sociodemographic characteristics were identified using enrollment data on sex, age, race/ethnicity, and insurance plan (capitation versus fee-for-service). Clinical characteristics included behavioral health and medical comorbidities (measured using primary and secondary diagnosis codes) and health services use.

Mental health diagnoses included depression, anxiety, schizophrenia or bipolar disorder, and any other mental illness. Substance use disorders included diagnoses for opioids, alcohol, cannabis, tobacco, cocaine, sedatives, and any other drug. Medical indicators included hepatitis C and the Agency for Healthcare Research and Quality’s Elixhauser Comorbidity Index, a weighted composite measure of overall medical burden modified to exclude duplicative diagnoses.³⁵ The final comorbidity index score accounted for twenty-five medical conditions (for example, HIV, hypertension, diabetes, cancer, and organ disease) that are associated with hospital-based resource use and mortality.³⁵

Health services included any opioid or non-opioid drug overdose treatment, total prescription opioid supply, and any prescription psychotropics (antidepressants, antipsychotics, stimulants, mood stabilizers, benzodiazepines, or sedatives). Because associations between buprenorphine treatment characteristics and discontinuation or relapse have been reported, we included indicators for initial dose³⁶ and proportion of days covered, a measure of treatment continuity.³⁷

Statistical Analysis

We implemented a one-to-one nearest neighbor propensity score match without replacement to identify continuously treated individuals who were similar to individuals who discontinued buprenorphine on all of the baseline characteristics described above. The 180-day follow-up period for continuously treated individuals was defined by the follow-up periods for matched pairs who discontinued buprenorphine.

To assess baseline comparability between study groups before and after matching, we calculated descriptive statistics (see below) as well as measures of covariate balance (standardized difference in means) to show the bias reduction achieved by matching. Appendix exhibit A2 shows results of the propensity score matching.²⁹

The main analyses were conducted among the propensity score–matched sample. Appendix exhibit A3 shows sample selection criteria and sample size.²⁹ We assessed the population-averaged association between treatment duration and each outcome within a difference-in-differences framework, using multivariable generalized estimating equation models to account for correlation within individuals over time and calculating variances using a first-order autoregressive parameter. Analyses were conducted at the person-month level, defined as thirty-day intervals during the treatment and follow-up periods. We used logistic regression to estimate the five main outcomes and Poisson regression to estimate the rate of prescription opioid use, controlling for all baseline characteristics. Models included an interaction term for study group and study period to compare changes from treatment to follow-up between individuals with continuous treatment and those who discontinued buprenorphine.

We calculated predicted probabilities for all difference-in-differences estimates and separately by study group to show the adjusted differences in outcomes between treatment and follow-up (see below). The relative change in each outcome was calculated as the difference-in-differences estimate divided by the observed rate during treatment for the group with continuous buprenorphine, expressed as a percentage.

In this difference-in-differences framework, the group with continuous treatment plausibly represented a counterfactual group to the one whose members discontinued buprenorphine if the groups had parallel trends before follow-up.³⁸ We assessed the parallel trends assumption using multivariable generalized estimating equation models for each outcome and determined that the study groups did not have significantly different trends during treatment and, therefore, could reasonably be expected to have had similar outcomes during follow-up if individuals who discontinued buprenorphine had instead remained on treatment.³⁹ Appendix exhibit A4 shows the results of the parallel trends analyses.²⁹ We also conducted sensitivity analyses to assess the robustness of findings to unmeasured confounding.⁴⁰ Appendix exhibit A5 shows the results of the sensitivity analyses.²⁹

Limitations

Several limitations should be noted. First, findings from Medicaid claims might not be generalizable to individuals with no insurance or private coverage. In addition, though the data are from a large multistate sample, the results might not be generalizable to all Medicaid enrollees. The sample likely represents relatively stable patients, since they had continuous enrollment and were among the minority of adults who reached six months of buprenorphine treatment.^11,31,36 Nonetheless, clinical considerations might influence buprenorphine discontinuation around this time (six to nine months). Also, we could not observe fatal overdoses in claims data. Since people who discontinue buprenorphine are at increased risk of overdose fatality,¹⁰ we might have underestimated the benefits associated with continuous treatment.

Second, although indicators for insurance type were included in our analyses, specific benefits information for buprenorphine coverage was not available. States were deidentified for confidentiality, which meant that we were not able to assess state plan information. Considering the variability across state programs, these characteristics could represent important barriers to longer-term treatment that warrant future research.

Third, our results could have been biased by residual confounding by unmeasured variables associated with both treatment duration and clinical outcomes. While outcome rates observed during treatment can differ between study groups without biasing the results, differences could raise concerns about unmeasured confounding.³⁹ To address this concern, propensity score matching was used to minimize potential inaccuracy in the estimates.³⁹ Furthermore, sensitivity analyses to determine the robustness of results⁴⁰ indicated that unmeasured confounding was unlikely to explain away the associations between buprenorphine treatment and the outcomes.

Study Results

The sample included 4,433 adults who discontinued buprenorphine 180–270 days after treatment initiation (exhibit 1). Before propensity score matching, 14,277 adults had had continuous treatment for at least 450 days. Before matching, study groups were significantly different in terms of baseline sociodemographic characteristics and many clinical characteristics. Compared to the group that discontinued treatment, the continuously treated group was older, on average, and a greater proportion of it was female or white or had fee-for-service insurance. Before matching, study groups were similar in terms of OUD diagnosis and prescription opioid use, but the continuously treated group had lower rates of overdose, hepatitis C, and nonopioid substance use disorders, as well as higher initial buprenorphine dosage and proportion of days covered, on average. After matching, no significant differences remained between individuals with continuous treatment and individuals who discontinued buprenorphine.

Exhibit 1:

Baseline sample characteristics before and after propensity score matching in the study of buprenorphine use

	Discontinued treatment (n = 4,433)	Continued treatment
		Matched (n = 4,433)	Unmatched (n = 14,277)
Sociodemographic characteristics
Sex (%)
Male	36.1	34.5	32.7****
Female	63.9	65.5	67.3****
Mean age, years	34.2	34.3	35.1****
Race/ethnicity (%)
White	92.6	93.1	93.5**
Black	3.8	3.7	3.1**
Hispanic	1.1	0.8	1.0
Other	2.6	2.5	2.4
Insurance (%)
Fee-for-service	26.3	27.0	28.5***
Capitation	73.7	73.0	71.5***
Clinical characteristics
Mental health diagnosis (%)
Depression	21.6	21.3	20.0**
Anxiety	26.2	27.0	25.5
Schizophrenia or bipolar disorder	11.1	10.6	9.8**
Other	4.2	4.3	4.1
Substance use disorder (%)
Opioids	64.6	65.1	64.2
Alcohol	6.7	6.9	5.5***
Cannabis	5.7	6.0	4.5****
Tobacco	29.7	28.9	26.6****
Cocaine	3.3	3.5	2.6**
Sedatives	2.9	2.9	2.0***
Other	10.3	10.4	11.3**
Medical comorbidity
Mean Elixhauser Comorbidity Index	0.8	0.8	0.7*
Hepatitis C (%)	10.0	9.5	8.3****
Overdose event (%)
Opioid overdose	3.0	3.3	1.7****
Nonopioid drug overdose	3.5	3.7	2.5****
Prescription drugs (%)
Antidepressants	36.7	36.7	35.6
Antipsychotics	12.0	11.6	10.9*
Stimulants	5.5	5.6	5.8
Mood stabilizers	22.3	21.7	21.8
Benzodiazepines	20.4	20.4	19.9
Sedatives	3.8	3.9	3.8
Mean opioid supply (days)	15.0	14.9	15.6
Buprenorphine treatment
Mean initial dose	7.9	7.9	8.0****
Mean proportion of days covered	0.9	0.9	1.0****

SOURCE Authors’ analysis of data for 2013–17 from the MarketScan Multi-State Medicaid database.

NOTE Significance was determined from pairwise t-tests.

^*p < 0.10

^**p < 0.05

^***p < 0.01

^****p < 0.001

Health Care Outcomes

The group with continuous treatment generally showed reductions for each outcome, while the group that discontinued buprenorphine showed increases for each outcome in the follow-up period (26,598 person-months) compared to the treatment period (34,907 person-months). Appendix exhibit A6 shows the observed rates of health services use.²⁹

Exhibit 2 shows adjusted estimates of the effect of treatment duration on outcomes from difference-in-differences models. Compared to people who discontinued buprenorphine, those with continuous treatment had significantly lower probability of all-cause inpatient use, all-cause emergency use, opioid-related hospital use, all overdose events, and prescription opioid use in the follow-up period than in the treatment period. For the group with continuous treatment in the follow-up period, these changes translate into large relative decreases in the thirty-day risk of overdose events (−173 percent), opioid-related hospital use (−128 percent), and prescription opioids (−120 percent) and smaller but potentially meaningful decreases in all-cause inpatient (−52 percent) and emergency department use (−26 percent) (data not shown).

Exhibit 2: Adjusted differences between buprenorphine treatment and follow-up periods in the risk of each main outcome.

SOURCE Authors’ analysis of data for 2013–17 from the MarketScan Multi-State Medicaid Database.

NOTES Predicted probabilities were calculated from separate difference-in-differences generalized estimating equation logistic models that estimated each outcome. All models controlled for baseline sociodemographic characteristics (sex, age, race/ethnicity, and insurance); comorbid medical (modified Elixhauser Comorbidity Index and hepatitis C), mental health (depression, anxiety, schizophrenia or bipolar disorder, and other mental illness), and substance use (opioids, alcohol, cannabis, tobacco, cocaine, sedatives, and other drugs) diagnoses; health services (opioid overdose; nonopioid drug overdose; number of days with prescription opioid supply; any use of prescription antidepressants, antipsychotics, stimulants, mood stabilizers, sedatives, or benzodiazepines); and buprenorphine treatment characteristics (initial dose and proportion of days covered). The values of the difference-in-differences results are represented by the vertical lengths of the bars (discontinued and continued treatment portions combined). The outcomes were as follows: all-cause inpatient use, −0.9 percentage points; all-cause emergency use, −2.4 percentage points; opioid-related hospital use, −1.2 percentage points; all overdose events, −0.5 percentage points; and prescription opioids, −4.6 percentage points. All results were significant (p < 0.001). The error bars indicate 95% confidence intervals.

Rates Of Prescription Opioid Use

The group with continuous treatment showed lower rates of prescription opioid use, while the group that discontinued buprenorphine showed higher rates in the follow-up period (13,482 person-months) compared to the treatment period (17,699 person-months). Appendix exhibit A6 shows the observed rates of prescription opioid use.²⁹ Both groups showed steady reductions in the rate of prescription opioid use during buprenorphine treatment (exhibit 3). However, those who discontinued buprenorphine increased their use to a rate similar to that in the period of buprenorphine initiation, while those with continuous treatment maintained consistently low rates throughout the follow-up period.

Exhibit 3: Percent of days with a prescription opioid supply out of the days in each 30-day period, for prescription opioid users and matched pairs.

SOURCE Authors’ analysis of data for 2013–17 from the MarketScan Multi-State Medicaid Database.

NOTE The rates were calculated among the subsample with any prescription opioid use during the study period and their matched pairs (4,494 people in all) for each thirty-day treatment or follow-up period.

Compared to individuals who discontinued buprenorphine, individuals with continuous treatment had a significantly lower rate of prescription opioid use (−3.6 percentage points) in the follow-up period than in the treatment period. Appendix exhibit A6 shows the difference-in-differences estimates.²⁹ This change translates into a large relative decrease in the rate of prescription opioid use (−124 percent) for the group with continuous treatment in the follow-up period.

Discussion

Continuous buprenorphine treatment beyond six to nine months was associated with reduced risk of adverse health care outcomes among Medicaid enrollees. The greatest decreases were found for overdose events, opioid-related hospital use, and prescription opioids. Reductions in any prescription opioids as well as the rate of prescription opioid use suggest that discontinuation was associated with a subsequent return to prescription opioids for some patients and an increase in their use for others—which could indicate a lack of coordinated care. These findings are consistent with prior research demonstrating that minimal or no opioid use is associated with treatment retention^41,42 and that most patients who discontinue buprenorphine return to opioid use within a month.⁴³ The overall pattern of results indicates that longer-term buprenorphine treatment more effectively protects patients from adverse clinical outcomes.

These protective effects extend the findings from clinical trials that examined shorter treatment periods (less than six months)⁷ and those from observational studies that suggested greater benefits with longer-term treatment without assessing minimum treatment duration.^9,12–14 The lack of research on minimum treatment duration may be related to challenges involved in identifying adequate samples, since a minority of patients have been successfully retained on buprenorphine for at least six months.^31,36,44 The high prevalence of short treatment periods underscores the challenges of retaining patients in community-based buprenorphine treatment and the need to develop and implement models of care that support transitions from initiation to maintenance treatment—including ways to address administrative and structural barriers.⁴⁵

In 2003, Massachusetts implemented a collaborative care model for office-based opioid treatment with buprenorphine, demonstrating its effectiveness in primary care settings.⁴⁶ The Massachusetts model was expanded to community health centers in 2007 to reach underserved and disadvantaged populations,⁴⁷ including a large proportion of Medicaid enrollees.⁴⁸ The model explicitly included strategies to maintain patient engagement and uninterrupted access to medication, and in 2013, 67 percent of patients were retained in treatment after one year.⁴⁷ The success of this program can be attributed in large part to addressing health system barriers to buprenorphine treatment such as care fragmentation, large provider caseloads, and staff and other resource shortages.⁴⁹ Additionally, the model relied on a Medicaid capitated payment, which allowed for adaptation in the range of services provided to ensure coverage of administrative costs. Prospective payment structures have also supported innovative delivery models in federally qualified health centers in Missouri and Washington,²⁰ but more work is needed to address insurance and payment factors that impede OUD treatment across settings.

As was the case with the Massachusetts model, policy strategies to overcome barriers to treatment continuity may be more effective if they incorporate state-specific considerations. Because state identifiers were not included in our data, we were unable to assess whether outcomes varied by state. However, this information—together with data on benefit design—could reveal promising targets for policy reform. For example, utilization management techniques that restrict access to care to curb spending may result in unintended consequences. Studies in Massachusetts have demonstrated that prior authorization policies are unlikely to result in significant savings⁵⁰ and that this type of cost containment strategy could lead to increased risks of relapse and death.⁵¹ Considering that OUD medication treatment is also associated with substantial reductions in nonaddiction health care costs,⁵² future studies should examine the extent to which improving access to longer-term treatment accrues savings by reducing adverse health care outcomes.

We focused on a sample with broad access to care, since all state Medicaid programs include coverage for buprenorphine.¹⁹ Yet nearly all programs also impose restrictions on buprenorphine coverage through quantity or dosing limits; preferred drug lists; and requirements for prior authorization, step therapy, or psychosocial/behavioral treatment.^19,20 Nonetheless, access to buprenorphine has expanded in recent years, with decreases in the number of Medicaid programs imposing lifetime treatment limits or requiring prior authorization or psychosocial/behavioral therapy.^20,27 However, most states require prior authorization for buprenorphine (forty states) or buprenorphine-naloxone (thirty-one states).^20,27 Many policies authorize buprenorphine treatment for periods of three or six months, but some require authorization more frequently (for example, every thirty-four days in Indiana),²⁰ which highlights potential opportunities to reduce barriers to continuous treatment.

Conclusion

The lower risk of adverse clinical outcomes observed among patients retained in buprenorphine treatment for over a year, compared to similar patients who discontinued treatment after a maintenance phase of six to nine months, underscores the importance of extended treatment. Policies that increase long-term access to buprenorphine treatment, such as lengthening benchmarks and standards for minimum treatment duration, could help address the ongoing opioid epidemic. Future research should examine whether these protective effects extend to even longer treatment duration over multiple years.