Association of Buprenorphine-Waivered Physician Supply With Buprenorphine Treatment Use and Prescription Opioid Use in Medicaid Enrollees

Key Points Question Is the availability of buprenorphine-waivered physicians associated with buprenorphine treatment use and prescription opioid use among Medicaid enrollees? Findings In this economic evaluation study using the all-capture Medicaid prescription data between 2011 and 2016, a 10% increase in the number of buprenorphine-waivered physicians was associated with an approximately 10% increase in the Medicaid-covered buprenorphine prescribing rate and a 1.2% reduction in the opioid prescribing rate. Meaning Expanding capacity for buprenorphine treatment holds the potential to improve access to opioid addiction treatment, which may further reduce prescription opioid use and slow the ongoing opioid epidemic in the United States.

This supplementary material has been provided by the authors to give readers additional information about their work.

eAppendix 1. Identification of Buprenorphine for Medication-Assisted Treatment of Opioid Addiction
We used the National Drug Code (NDC) numbers linking to the FDA Orange Book to identify the following sets of Buprenorphine Hydrochloride formulations (including Buprenorphine-Naloxone Hydrochloride formulations) for medication-assisted treatment (MAT) of opioid addiction: (i) Suboxone® sublingual tablets and films, (ii) Subutex® sublingual tablets, (iii) Zubsolv® sublingual tablets, (iv) Bunavail® buccal films, and (v) the generic equivalents of (i), (ii), (iii) and (iv). These medications are FDA-approved for MAT. Any use of them for the treatment of pain or depression is considered an off-label, unapproved use.
We excluded the following sets of Buprenorphine Hydrochloride formulations generally prescribed for pain management: (i) Buprenex® injectable, (ii) Butrans® transdermal patches, (iii) Belbuca® buccal films, and (iv) the generic equivalents of (i), (ii) and (iii). These medications cannot be prescribed for MAT, even by a buprenorphine-waived physician.
We also excluded methadone and naltrexone, the other two FDA-approved medications for MAT, due to their unique funding streams and multiple purposes in treatment: first, methadone and naltrexone are prescribed exclusively in opioid treatment programs and mostly funded by state and local substance abuse treatment agencies, which cannot be fully captured by the study data and may not be directly affected by Medicaid expansions; second, the small segment of the methadone and naltrexone that the study data do capture may not be relevant to opioid addiction treatment, since a considerable proportion of methadone is prescribed for pain management and the majority of naltrexone is for alcohol addiction treatment. Buprenorphine Use s,t = α 1 + α 2 #100-Waived Physician s,t + α 3 #30-Waived Physician s,t + α 4 X 1 s,t + ρ s + τ t + ε s,t Opioid Use s,t = α 1 + α 2 #100-Waived Physician s,t + α 3 #30-Waived Physician s,t + α 4 X 1 s,t + ρ s + τ t + ε s,t Prescription Opioid Use s,t = β 1 + β 2 Buprenorphine Use s,t + β 3 X 1 s,t + ρ s + τ t + ρs×t + ε s,t 5 OLS Estimation vs. TSLS-IV Estimation: In the simple OLS regression, we used the state and quarter fixed effects as well as the group-specific linear trends to isolate the within-state variations in prescription opioid use over time, thereby identifying the effect of buprenorphine use on prescription opioid use. Nonetheless, reverse causality and omitted-variable bias may lead to underestimation in the OLS estimates. First, an increase in prescription opioid use may cause more problems of opioid addiction, thus generating higher potential needs for buprenorphine use. Simply replacing or instrumenting contemporary value of buprenorphine use with its lagged values is problematic if the error terms are correlated over time. This reverse causality from prescription opioid use to buprenorphine use may result in downward-biased OLS estimates. Second, there may be unobserved time-varying factors correlated both with buprenorphine use and with prescription opioid use. Some important unobservables are changes in the composition of state Medicaid enrollees and their underlying risk for addiction, as well as fluctuations in the market price and availability of opioids due to varied intensity of opioid crackdowns and enforcement of prescription regulations. If the omitted variables largely affected both buprenorphine use and prescription opioid use in the same direction, the OLS estimates would be biased towards the null. To address the endogeneity of buprenorphine use with regard to prescription opioid use, we used the number of buprenorphine-waived physicians to isolate the potentially exogenous variation in buprenorphine use. The two instrumental variables in our main analyses are the numbers of 100-patient-waived physicians and 30-patient-waived physicians per 1,000,000 State residents (100-Waived Physicians,t & 30-Waived Physician s,t). In the sensitivity analyses, we used the total number of buprenorphine-waived physicians. (i) per capita number of office-based primary care physicians: primary care specialties include general practice, general internal medicine, and family medicine, but exclude pediatrics; Data Source: Health Resources and Services Administration (HRSA). Area Health Resources Files (AHRF): https://datawarehouse.hrsa.gov/topics/ahrf.aspx; (ii) per capita number of office-based psychiatrists (including addiction psychiatrists); Data Source: Health Resources and Services Administration (HRSA). Area Health Resources Files (AHRF): https://datawarehouse.hrsa.gov/topics/ahrf.aspx; (iii) poverty rate: calculated for the civilian noninstitutionalized population based on household income, household size, and household composition, relative to a set of dollar value thresholds called the "federal poverty level (FPL)"; institutionalized persons, those in military group quarters, and those living in college dormitories, and unrelated children under the age of 15 excluded from the numerator and denominator when calculating the poverty rate; Data Source: Health Resources and Services Administration (HRSA). Area Health Resources Files (AHRF): https://datawarehouse.hrsa.gov/topics/ahrf.aspx; (iv) unemployment rate: calculated as the number of unemployed persons divided by the number of persons in the labor force (aged 16 and above); the numerator and denominator excluding the institutionalized persons or those without employment who are not seeking employment; Data Source: Health Resources and Services Administration (HRSA ρs×t: group-specific linear trends at the Census division level accounting for the unobserved division-wide confounding factors that evolve over time at a constant rate; All models were population-weighted and were state-clustered to correct for the within-state serial correlation in the error terms; We excluded D.C. and Rhode Island, as well as one or several quarters of observations in Arizona, Illinois, Kansas, Kentucky, Louisiana, Mississippi, New Jersey, New Mexico, and Oregon, from the study data due to the inconsistency in state data reporting. Regarding the data quality, although the Medicaid SDUD is very timely and useful data, it is prone to unusual patterns in state reporting due to the close linkage to the Medicaid drug rebate program. It is not unusual for states to revise and resubmit data for this system, with retroactive changes sometimes occurring several years after the end of a quarter. Therefore, we have checked with the CMS staff who are responsible for maintenance of the files about the potential data quality concerns and plotted out the state trends in the fee-for-service (FFS) and managed care data separately to check for any unusual, suspicious patterns during the study period. We have also compared the total FFS drug spending against the amounts reported on Form CMS-64, prior to rebates, to see if the orders of magnitude are similar as another quality check. One potential concern, as the reviewer noted, is that prescription drugs paid through Medicaid managed care organizations (MCO) were excluded from the required quarterly data reporting/rebate collection until March 23, 2010, when the ACA expanded the reporting/collection requirement to all the Medicaid MCO carve-ins. By the end of 2011 2nd quarter, the majority of the 22 States using a carve-in approach had collected all the required data and performed data verification checks. A few states revised/resubmitted the complete data later. D.C. is the only state that may still have incompleteness in its first three quarters of 2011 data. Therefore, we also excluded D.C. and started our study period from 2011 onward. Another potential data quality issue concerns the new adult eligibility group under the ACA Medicaid expansion. At the beginning of 2014, CMS noticed some confusion and inconsistency in state reporting associated with the new adult eligibility group and the increased federal matching rates available to this group. CMS provided states with timely reviews, clarifying information, and technical assistance. The 2014 onward data should have acceptable quality for analysis.

eAppendix 3. Further Explanation of Mediation Analysis Using Two-Stage Least Squares-Instrumental Variable (TSLS-IV) Model
TSLS-IV Stage I: Buprenorphine Use s,t = α 1 + α 2 #100-Waived Physician s,t + α 3 #30-Waived Physician s,t + α 4 X 1 s,t + ρ s + τ t + ε s,t TSLS-IV Stage II: Prescription Opioid Use s,t = β 1 + β 2 Buprenorphine Use s,t + β 3 X 1 s,t + ρ s + τ t + ε s,t The two-stage least squares model instrumental variable (TSLS-IV) analysis was used to explored the extent to which buprenorphine treatment use serves as one of the key pathways from expanding the availability of buprenorphine-waived physicians to reducing prescription opioid use. The first stage of the TSLS-IV analysis treated the availability of buprenorphinewaived physicians as the instrument to estimate the variation in buprenorphine treatment use. In the second stage, this quasiexperimental source of variation in buprenorphine treatment use was then treated as the independent variable of interest to estimate the effect on prescription opioid use.
The TSLS-IV estimates can be generalized from the policies aimed at expanding the availability of buprenorphine-waived physicians to a broader range of capacity expansion policies (e.g., allowing nurse practitioners and physician assistants to provide buprenorphine treatment, removing the existing restrictions on Medicaid coverage for buprenorphine treatment).
A key assumption behind the TSLS-IV analysis is that a valid instrument should affect the outcome only through the endogenous independent variable of interest. Conceptually, there may exist other pathways from buprenorphine-waived physician availability to prescription opioid use. For instance, when deciding about waiver application, physicians may consider the overall opioid prescribing rates to gauge the potential treatment needs. However, physicians were unlikely to be able to anticipate the future changes in opioid prescribing rates for Medicaid enrollees in particular, and they were unlikely to decide on apply for the waiver in anticipation of reductions in opioid prescribing rates. Another possible pathway, one may argue, lies in that the required training for buprenorphine-waived physicians may prompt the physicians to become more cautious in opioid prescribing for their pain patients. Nonetheless, the relative small numbers of buprenorphine-waived physicians and pain patients cared by these physicians may not suffice to change the overall patterns of opioid prescribing practices. Furthermore, our TSLS-IV estimates suggest that 83 percent of the reduction in opioid prescribing rate associated with expanding buprenorphine-waived physician availability was through the pathway of increasing buprenorphine prescribing rate. The two instruments allowed for an overidentification test of the exclusion restrictions (Appendix Tables A8 Sargan-Hansen J statistics of the joint overidentification test), which also lends weight to the validity of our instruments. 3.9% 0.07** (0.0 0.9 0.10** (0.0

$ All Opioid Prescriptions
- Yes Notes: † p<0.10, *p<0.05, **p<0.01, ***p<0.001; Marginal effects associated with one more buprenorphine-waived physician; Standard errors in parentheses clustered at the state level; Significant percentage changes associated with 10 percent increase in the number of buprenorphine-waived physicians (i.e., approximately two more 100-patient-waived physicians per 1,000,000 residents or five more 30-patient-waived physicians per 1,000,000 residents based on the 2011-2016 average numbers); Shown in Table 1

Group-Specific Linear Trends
N Y Notes: p-values in parentheses estimated from the joint overidentification test of the numbers of 100-patient-waived physicians and 30-patient-waived physicians per 1,000,000 residents; the Sargan-Hansen J statistics lending statistical support to the exogeneity of both instruments across the board.