Characteristics of US Counties With High Opioid Overdose Mortality and Low Capacity to Deliver Medications for Opioid Use Disorder

Key Points español 中文 (chinese)

Question  What are the characteristics of US counties with high rates of opioid overdose mortality and low capacity to deliver medications for opioid use disorder?

Findings  In this cross-sectional study of data from 3142 US counties, counties in the South Atlantic, Mountain, and East North Central divisions had more than twice the odds of being at high risk for opioid overdose mortality and lacking in capacity to deliver medications for opioid use disorder. Higher density of primary care clinicians, a younger population, micropolitan status, and lower rates of unemployment were associated with lower risk of opioid overdose and lower risk of lacking in capacity to deliver medications for opioid use disorder.

Meaning  Strategies to address mortality from opioid overdose by increasing treatment for addiction should target urban counties in Appalachia, the Midwest, and the Mountain division and include efforts to increase primary care clinicians and employment opportunities.

Importance  Opioid overdose deaths in the United States continue to increase, reflecting a growing need to treat those with opioid use disorder (OUD). Little is known about counties with high rates of opioid overdose mortality but low availability of OUD treatment.

Objective  To identify characteristics of US counties with persistently high rates of opioid overdose mortality and low capacity to deliver OUD medications.

Design, Setting, and Participants  In this cross-sectional study of data from 3142 US counties from January 1, 2015, to December 31, 2017, rates of opioid overdose mortality were compared with availability in 2017 of OUD medication providers (24 851 buprenorphine-waivered clinicians [physicians, nurse practitioners, and physician assistants], 1517 opioid treatment programs [providing methadone], and 5222 health care professionals who could prescribe extended-release naltrexone). Statistical analysis was performed from April 20, 2018, to May 8, 2019.

Exposures  Demographic, workforce, lack of insurance, road density, urbanicity, opioid prescribing, and regional division county-level characteristics.

Main Outcome and Measures  The outcome variable, “opioid high-risk county,” was a binary indicator of a high (above national) rate of opioid overdose mortality with a low (below national) rate of provider availability to deliver OUD medication. Spatial logistic regression models were used to determine associations with being an opioid high-risk county.

Results  Of 3142 counties, 751 (23.9%) had high rates of opioid overdose mortality. A total of 1457 counties (46.4%), and 946 of 1328 rural counties (71.2%), lacked a publicly available OUD medication provider in 2017. In adjusted models, compared with the West North Central division, counties in the East North Central, Mountain, and South Atlantic divisions had increased odds of being opioid high-risk counties (East North Central: odds ratio [OR], 2.21; 95% CI, 1.19-4.12; Mountain: OR, 4.15; 95% CI, 1.34-12.89; and South Atlantic: OR, 2.99; 95% CI, 1.26-7.11). A 1% increase in unemployment was associated with increased odds (OR, 1.09; 95% CI, 1.03-1.15) of a county being an opioid high-risk county. Counties with an additional 10 primary care clinicians per 100 000 population had a reduced risk of being opioid high-risk counties (OR, 0.89; 95% CI, 0.85-0.93), as did counties that were micropolitan (vs metropolitan) (OR, 0.67; 95% CI, 0.50-0.90) and those that had an additional 1% of the population younger than 25 years (OR, 0.95; 95% CI, 0.92-0.98).

Conclusions and Relevance  Counties with low availability of OUD medication providers and high rates of opioid overdose mortality were less likely to be micropolitan and have lower primary care clinician density, but were more likely to be in the East North Central, South Atlantic, or Mountain division and have higher rates of unemployment. Strategies to increase medication treatment must account for these factors.

Policy makers are striving to mitigate adverse consequences of the opioid crisis, which caused more than 130 deaths per day in 2017.¹ The population with opioid use disorder (OUD)² continues to grow, constituting between 2.1 million and 6 million individuals in 2017.^3-5 With, at most, 20% to 40% of persons with OUD receiving treatment,^3,6,7 policies that expand access to and delivery of evidence-based treatment are critical to reducing the risk of opioid overdose.^8-10

Medications for OUD treatment (MOUDs) are the criterion standard for treating OUD. Three MOUDs—methadone hydrochloride, buprenorphine hydrochloride, and extended-release naltrexone hydrochloride—have all been shown in clinical trials to reduce opioid use and adverse health outcomes.^11-23 Methadone treatment was associated with a 53% reduction and buprenorphine treatment was associated with a 37% reduction in all-cause mortality among patients with OUD compared with those receiving no MOUD in the 12 months after nonfatal overdose.²⁴

However, evidence suggests that availability of MOUDs has been slow to expand and, in many cases, is not available.^24-26 Only opioid treatment programs (OTPs), which are closely regulated at the federal and state levels, can deliver methadone²⁷; this restriction has contributed to methadone’s short and relatively flat supply over time.^3,28 Although some OTPs also supply buprenorphine products to treat OUD,^29,30 a 2002 policy change that granted physicians in outpatient nonspecialty settings the authority to prescribe buprenorphine with training and a waiver issued by the federal Substance Abuse and Mental Health Services Agency (SAMHSA) led to greater availability of buprenorphine.^31,32 Unlike methadone and buprenorphine, both opioid agonists, newer extended-release naltrexone is an opioid antagonist³³ that can be prescribed by any licensed prescriber. However, patients must be opioid abstinent for at least 7 to 10 days prior to treatment with extended-release naltrexone, and data demonstrating this drug’s effectiveness in preventing overdose among those with a nonfatal overdose are lacking.²⁴

Because evidence supports the use of MOUDs rather than other treatment modalities to reduce opioid overdose, there is a critical need to characterize areas where the need for, and availability of, MOUD treatment providers have a deleterious mismatch. Previous studies have examined availability of methadone and buprenorphine at the state level,^3,29,34 the availability of buprenorphine alone,^29,34-39 or the supply of substance use treatment facilities treating patients with MOUDs.^26,40 Recent studies have examined the geospatial association between county-level buprenorphine and OTP supply and opioid overdose mortality,^41,42 but none has characterized counties with a high overdose burden and low capacity to deliver MOUDs. Given the high level of intrastate variability in availability of MOUD providers^25,39 and opioid overdose harms,²⁵ as well as the importance of targeting resources to counties at highest risk of the mismatch between treatment and harms, we sought to fill that literature gap. We hypothesized that high rates of opioid analgesic prescribing, location in Appalachian regions, and low density of mental health care professionals and primary care clinicians would be associated with low supply of MOUD providers and high rates of opioid overdose mortality at the county level.

We analyzed characteristics associated with low availability of MOUD providers and high rates of opioid overdose mortality using a geospatial cross-sectional analysis design that combined county-level data from January 1, 2015, to December 31, 2017, from several sources with mortality data from the Centers for Disease Control and Prevention (CDC). Institutional review board approval was not required for this study that used all publicly available data, per the University of Michigan, Office of Research Operations Manual, Part 4.⁴³ This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cross-sectional studies.⁴⁴

The availability of OTPs and buprenorphine-waivered clinicians as of September 19, 2017, was determined from the publicly available SAMHSA provider locator websites.^28,45 For OTPs, this constitutes a complete list of all 1517 OTP facilities licensed to provide methadone to treat OUD. For buprenorphine, this constitutes a complete list of 24 851 clinicians (physicians, nurse practitioners, and physician assistants) certified under federal law to prescribe buprenorphine through a waiver process^31,32 and who agreed to be listed on the website for the purpose of being identified by patients seeking buprenorphine treatment. We obtained a comprehensive list from Alkermes Inc, the manufacturer of the extended-release naltrexone product Vivitrol, of the 5222 health care professionals listed on the company’s publicly available treatment locator searchable tool who were actively prescribing the medication as of November 30, 2017.⁴⁶

For each of these 3 MOUDs, we geographically coded the addresses for all treatment providers. We then counted the number of providers per 100 000 county residents for each medication separately and across the 3 MOUDs aggregated. We used 2017 US American Community Survey (ACS) estimates of county populations as the denominator in these calculations.⁴⁷

We extracted county-level rates of opioid overdose mortality from January 1, 2015, to December 31, 2017, from the CDC WONDER (Wide-ranging Online Data for Epidemiologic Research) Multiple Cause of Death database.^48,49 This database is based on US resident death certificates, which are coded into the National Vital Statistics System by states or the CDC’s National Center for Health Statistics. We searched for both intentional and unintentional underlying causes of death (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes X40-X45, X60-X65, and Y10-Y15) where any type of opioid was coded for injury and poisoning (codes T40.0-T40.4 and T40.6). For the denominator, we used county populations aggregated for 2015-2017, to create county-level death rates per 100 000 residents per year. These death rates are depicted graphically in 1322 counties (1820 were suppressed) in Figure 1.

We used the national opioid overdose mortality rate of 12.5 per 100 000 population from 2015-2017 as a threshold to divide counties with high and low opioid overdose death rates. We took the county MOUD provider availability rates across all 3 medication types (OTPs, buprenorphine-waivered clinicians, and extended-release naltrexone prescribers) and used the national rate of 9.7 providers per 100 000 residents as a threshold to divide counties into high and low MOUD provider capacity groups. We then defined opioid high-risk counties as counties with a low capacity for MOUD providers and a high rate of opioid overdose mortality. We analogously created county-level risk indicators for each MOUD separately (eAppendix in the Supplement).

The county-level risk indicators consisted of demographic characteristics, density of primary care physicians (PCPs) and mental health clinicians, proportion uninsured, road density, urbanicity, opioid prescription rate, percentage voting democratic in the 2016 presidential election, and geographic regional division. We also conducted a sensitivity analysis that excluded clinician (PCP and mental health clinician) density, given that this factor is partially encompassed in the opioid high-risk county measure. County-level demographic characteristics—including age, race/ethnicity, unemployment, and educational level—were all taken from the ACS.⁵⁰ County-level density of PCPs and mental health clinicians (counts per 100 000 population) were obtained from the Robert Wood Johnson Foundation County Health Rankings and Roadmaps 2016 data.⁵¹ County-level rates of uninsured individuals were drawn from the 2016 Small Area Health Insurance Estimates using the ACS.⁵² We obtained data on road mileage from the US Geological Survey National Geospatial Technical Operations Center’s 2006 National Transportation Data set and created a traversability variable by dividing road mileage by county land area from the ACS.⁵³ County-level urbanicity was categorized as rural, micropolitan, and metropolitan using the National Center for Health Statistics Urban-Rural Classification Scheme for Counties.⁵⁴ The number of opioid (analgesic) prescriptions dispensed per 100 persons in 2016 was drawn from the CDC.⁵⁵ The political partisan measure was drawn from the Massachusetts Institute of Technology Election Data and Science Lab data.⁵⁶ Geographic regional divisions were defined using the US Census Bureau’s 9-category categorization scheme (eFigure 1 in the Supplement).⁵⁷

Statistical analysis was performed from April 20, 2018, to May 8, 2019. First, we described and mapped counties lacking any available MOUD provider (across all 3 medications) and by county-level urbanicity categories and the 9 census-based divisions. Next, we mapped the spatial distribution of the county MOUD provider density along a continuum and overlaid rates of opioid overdose mortality on top (eFigure 2 in the Supplement). Then we mapped the 4 county types (high and low MOUD provider rates by high and low rates of opioid overdose mortality) across the United States. We also created these spatial distribution maps disaggregated by MOUD type (eFigures 3-5 in the Supplement). Finally, we contrasted opioid high-risk counties with non–high-risk counties using 2-sample comparison tests on county-level demographics, density of PCPs and mental health care clinicians, county size and road density, urbanicity, opioid prescribing, percentage democratic vote, and regional division.

Our primary goal was to determine characteristics of opioid high-risk counties while addressing residual correlation arising from the spatially indexed nature of the outcome, and suppressed counties whose exact rate of opioid overdose mortality was unknown. To jointly address these issues we used a modification of logistic regression that models the residual spatial trends and incorporates weights based on the estimated likelihood that a suppressed county was actually an opioid high-risk county (eAppendix in the Supplement). We considered P < .05 to be statistically significant and used 2-sided tests. All models were adjusted for covariates enumerated above and fit using R, version 3.5.1 (R Project for Statistical Computing).

eTable 1 in the Supplement shows the breakdown of 1457 of 3142 counties (46.4%) lacking any publicly listed MOUD treatment provider in late 2017, by geographic divisions and by urbanicity. Most counties in many divisions lacked any publicly available MOUD treatment provider, including in the West South Central (Arkansas, Louisiana, Oklahoma, and Texas) and West North Central (Iowa, Nebraska, Kansas, North Dakota, South Dakota, Minnesota, and Missouri) divisions. A total of 946 of 1328 rural counties (71.2%) lacked any publicly listed MOUD provider (Figure 2).

Of 3142 US counties, 751 (23.9%) had high rates of opioid overdose mortality. Table 1 shows unadjusted contrasts between opioid high-risk counties and non–high-risk counties. Figure 3 depicts the spatial distribution of opioid high-risk counties and the other 3 categories designated as non–high-risk (low rates of MOUD providers and low rates of opioid overdose mortality; high rates of MOUD providers and low rates of opioid overdose mortality; and high rates of MOUD providers and high rates of opioid overdose mortality). We identified 412 opioid high-risk counties and 1485 non–high-risk counties. Owing to suppression of data on opioid overdose mortality, 1245 counties had inadequate data to be initially categorized by risk status for unadjusted results. In unadjusted analyses, opioid high-risk counties had greater proportions of the population that were white, unemployed, and lacking a high school education; these counties also had a lower proportion of the population younger than 25 years. Opioid high-risk counties also had lower concentrations of PCPs and mental health care clinicians per 100 000 persons, a higher rate of opioid prescriptions per 100 persons, and a lower percentage democratic vote in the 2016 presidential election. In addition, opioid high-risk counties were more likely than non–high-risk counties to be overrepresented in the South Atlantic and East North Central divisions.

Maps depicting the spatial distribution of opioid high-risk and non–high-risk counties with respect to each MOUD type independently are displayed in eFigures 3-5 in the Supplement. Generally, the geographic clusters of high-risk counties were similar across MOUD provider types, although risk distributions in the buprenorphine-waivered clinician maps (eFigure 3 in the Supplement) closely resemble the distributions shown in Figure 3 across all 3 medications. Slightly more high-risk counties emerged in the OTP and extended-release naltrexone clinician prescriber maps (eFigure 4 and eFigure 5 in the Supplement), albeit in similar regional divisions that house opioid high-risk counties in the buprenorphine and combined MOUD maps.

Due to missing covariates (primarily counts of mental health clinicians and PCPs), high-risk probabilities could be estimated for only 831 of the 1245 initially uncategorized counties (eFigure 6 in the Supplement). Table 2 includes the spatial logistic regression results for characteristics associated with opioid high-risk counties, adjusted for county-level demographic, workforce, lack of insurance, road density, urbanicity, opioid prescribing, political partisanship, and regional division characteristics. Relative to the West North Central division, counties in the East North Central, Mountain, and South Atlantic divisions had increased odds of being opioid high-risk counties (East North Central: odds ratio [OR], 2.21; 95% CI, 1.19-4.12; Mountain: OR, 4.15; 95% CI, 1.34-12.89; and South Atlantic: OR, 2.99; 95% CI, 1.26-7.11). A 1% increase in unemployment was associated with an increased odds (OR, 1.09; 95% CI, 1.03-1.15) of a county being an opioid high-risk county. Counties that were micropolitan (vs metropolitan) had a reduced risk of being an opioid high-risk county (OR, 0.67; 95% CI, 0.50-0.90), as did counties with an additional 10 primary care clinicians per 100 000 population (OR, 0.89; 95% CI, 0.85-0.93) and those with an additional 1% of the population younger than 25 years (OR, 0.95; 95% CI, 0.92-0.98). An additional opioid prescription per 100 persons also was associated with marginally increased odds of opioid high-risk status (OR, 1.04; 95% CI, 1.00-1.07). In sensitivity results that excluded PCP and mental health clinician risk indicators, other risk indicators were substantively similar as in the main results. Opioid prescriptions were no longer a significant risk indicator, an additional percentage democratic vote corresponded to reduced odds (OR, 0.12; 95% CI, 0.03-0.44) of a county being an opioid high-risk county and an additional mile of road per square mile corresponded to reduced odds (OR, 0.95; 95% CI, 0.90-1.00) of a county being an opioid high-risk county (eTable 2 in the Supplement).

Adjusted regression results specific to each MOUD provider type are presented in eTables 3-5 in the Supplement. The factors for high-risk counties with respect to buprenorphine-waivered clinician availability are similar to those presented in Table 2. For low OTP provider availability paired with high rates of opioid overdose mortality, the East North Central division was no longer associated with greater risk and New England and the Mid-Atlantic divisions were associated with reduced risk. For the extended-release naltrexone models, the East South Central division had greater risk, whereas percentage democratic vote was associated with reduced risk.

We analyzed characteristics of counties that exhibited both historically high rates of opioid overdose mortality and low MOUD provider availability as of late 2017. Robust MOUD delivered via qualified providers in counties exhibiting these characteristics could realistically reduce opioid overdose deaths in vulnerable populations by as much as 40% to 60%.²⁴ In terms of MOUD provider availability alone, we found that 46.4% of all counties and 71.2% of rural counties still lacked an OTP, buprenorphine-waivered clinician, or extended-release naltrexone–prescribing clinician identifiable through public listings. When we also considered treatment need, opioid high-risk counties tended to be overrepresented in the East North Central, South Atlantic, and Mountain divisions and had higher unemployment; micropolitan status was associated with lower risk of being an opioid high-risk county.

Prior work has examined the availability of MOUDs in various ways. Several studies have demonstrated that at the national and state levels, OTP availability has remained steady from 2003 to 2012.^3,37 At the same time, numbers of Drug Addiction Treatment Act of 2000–waivered physicians capable of providing buprenorphine have significantly increased from 2003 to 2016.^3,34,37,39 Increasing numbers of buprenorphine-waivered physicians are associated with higher rates of opioid overdose at the state level, suggesting that perhaps supply has been responsive to demand specific to this medication.^29,34 Other state characteristics positively associated with rates of buprenorphine-waivered physicians in the literature include being in the Northeast region, the proportion of the population covered by Medicaid, the supply of OTPs, and the supply of substance use disorder treatment programs.²⁹ Although shortages of buprenorphine-waivered clinicians have decreased over time, rurality is a persistent risk factor of shortage areas.^35,37,39 In terms of specialty treatment facilities, past research has found shortages in the Southeast, Southwest, and Northeast regions.^26,40 Geospatial analyses have found the greatest mismatch between OUD treatment programs and opioid overdose mortality in counties in Ohio, the District of Columbia, and West Virginia, and limited buprenorphine provider access relative to opioid overdose mortality throughout much of the Midwest and South.^41,42

To our knowledge, this study is the first to present a picture of OUD treatment capacity across all 3 MOUDs and to compare this availability with recent historical need at the county level. These geospatial results indicate the specific types of counties where resources should be targeted to have greatest potential of increasing treatment and reducing overdose mortality. Given the characteristics of opioid high-risk counties that emerged from our main results, MOUD provider resources should be targeted to nonmicropolitan areas and in the East North Central (eg, Ohio, Michigan, Indiana, and Illinois), South Atlantic (eg, Virginia, West Virginia, the District of Columbia, Maryland, and Florida), and Mountain (eg, New Mexico, Arizona, Utah, and Nevada) divisions. Strategies to increase numbers of PCPs and other clinicians capable of and willing to provide MOUDs in these areas may be protective against a county persistently being high risk.⁸ Other innovative strategies to overcome workforce and geographic barriers—such as telemedicine, engagement of nonphysician prescribers in treatment, addressing stigma, providing peer-to-peer clinician support as in the Project ECHO (Extension for Community Healthcare Outcomes) model,⁵⁸ providing hub and spoke models of OUD treatment along the continuum of care, expanding Medicaid to address health care access among low-income and unemployed individuals, and dispelling myths—are also likely needed.^8,59-61

Recognizing that treatment with various MOUDs must be personalized based on patient factors, we also disaggregated county risk status by type of MOUD. Risk factors for buprenorphine were largely similar to those in analyses across all 3 MOUDS, suggesting that buprenorphine—the most available MOUD by far—was driving many of our main findings. Risk factors associated with buprenorphine were also consistent with prior studies of buprenorphine; namely, those that show risks of a shortage of buprenorphine to be associated with areas outside the northeast, particularly in the southeast.^29,34,37 Our results differ in that they do not identify rurality as a risk factor, likely because of the more recent time frame of our mortality data, during which deaths from overdose of illicit opioids in urban settings were prevalent.

In the OTP model, greater availability of OTPs in the mid-Atlantic and New England divisions made counties in these regions less likely to be categorized as opioid high-risk counties, despite their high concentration of opioid overdose deaths from 2015 to 2017 (Figure 2). The New England and the mid-Atlantic divisions had higher availability of treatment providers across all 3 MOUDs, and some areas within them (eg, Massachusetts) have exhibited declines in rates of opioid overdose mortality starting in 2017.⁶² For high-risk status associated with extended-release naltrexone, the East South Central division (eg, Kentucky and Tennessee) served as an additional risk factor.

Disaggregation by medication type can help to inform intervention strategies. For example, areas with greater OTP capacity may be good candidates to employ hub and spoke models. Areas without OTP capacity (eg, the Mountain division) might instead consider enhanced telemedicine. According to all models, treatment options should be targeted in counties with high rates of unemployment, and potentially paired with initiatives to help individuals with OUD or in recovery to find employment. The “deaths of despair” hypothesis advanced by Case and Deaton,⁶³ suggesting that structural determinants are key to opioid-associated harms, supports such approaches.⁶⁴

Our study has several limitations. First, we used publicly available treatment locator data to assess the availability of MOUD providers from a patient’s or typical clinician’s viewpoint when identifying treatment options. Some buprenorphine-waivered clinicians do not consent to being on the buprenorphine list, making our shortage statistics overestimated,³⁹ and we could be missing some MOUD providers not otherwise located by SAMHSA or Alkermes Inc. Alternatively, some buprenorphine-waivered providers who agree to be included on the buprenorphine list do not actively prescribe buprenorphine, potentially making our shortage statistics underestimated. Second, the Alkermes Inc list includes clinicians actively prescribing extended-release naltrexone, but does not differentiate whether this prescribing is for OUD or alcohol use disorder. Third, MOUD provider lists indicate availability, rather than volume of actual patient treatment. Most clinicians who prescribe buprenorphine prescribe the medication to a median monthly panel of 13 patients, many times even less^65,66; thus, this measure may be an optimistic view of availability.

Fourth, our outcome variable of risk was subjectively determined, and treatment shortage or high-risk areas could be otherwise defined. Findings in studies specific to buprenorphine and OTP treatment that have used alternative definitions of this concept have been reasonably consistent with ours.^29,34,40-42 Moreover, there could be some lag time in treatment availability responsive to overdose mortality, so the counties defined as opioid high-risk counties may no longer be high-risk in real time. Fifth, county-level opioid overdose death reporting drawn from national vital statistics data may include some measurement error, likely undercounting, that could be differential across counties.^67,68 However, these counts are more reliable to infer causes of death than are other sources (eg, toxicology reports), are the best source of comparative death data available, have been previously analyzed at the county level, and are subject to analogous measurement error as occurs at the state level.^69-71 Moreover, our outcome measure compared opioid overdose mortality with MOUD provider availability along continuums, rather than in absolute terms, which may have reduced the potential for biased conclusions. Sixth, we included many county-level covariates in our models, but it is possible there were unmeasured confounders not easily measured at the county level (eg, local county drug enforcement) that could be associated with opioid high-risk status. Seventh, we estimated high opioid overdose mortality probabilities in counties in which these counts were suppressed to add information to our adjusted models. Our findings were generally consistent between adjusted and unadjusted models, lending confidence that our weighting techniques were reasonable. Although we were able to impute information for most initially uncharacterized counties (831 of 1245 [66.7%]), adjusted results could have been biased by the omission of data from 414 predominantly rural counties.

Policy makers in the United States increasingly recognize the inadequacy of MOUD treatment availability and its potential to significantly reduce overdose mortality, as evidenced by congressional activity emphasizing and substantial resources allocated for addiction treatment.^72,73 This study provides new information to assist in identifying opioid high-risk counties and developing strategies to target resources. For instance, through the SAMHSA Substance Abuse Prevention and Treatment Block Grants to states and SUPPORT (Substance Use-Disorder Prevention that Promotes Opioid Recovery and Treatment) for Patients and Communities Act treatment augmentation provisions, prioritizing fund allocation and clinician workforce augmentation efforts around MOUD in nonmicropolitan counties, including in many Appalachian and Mountain regions, could be particularly effective in reducing opioid-related risks. In addition, focusing MOUD augmentation efforts in areas with fewer PCPs and higher unemployment rates would likely be an efficient use of resources to mitigate opioid harms. Although overall buprenorphine-waivered clinicians and funds for OUD treatment to states have increased in recent years, to have the largest effect on the opioid crisis these resources need to be funneled to local county areas with the greatest unmet need, together with new models of care to reach people with OUD.

Accepted for Publication: May 11, 2019.

Published: June 28, 2019. doi:10.1001/jamanetworkopen.2019.6373

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2019 Haffajee RL et al. JAMA Network Open.

Corresponding Author: Rebecca L. Haffajee, JD, PhD, MPH, Department of Health Management and Policy, University of Michigan School of Public Health, 1415 Washington Heights, Ann Arbor, MI 48109 (haffajee@umich.edu).

Author Contributions: Drs Haffajee and Goldstick had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: All authors.

Acquisition, analysis, or interpretation of data: Haffajee, Goldstick.

Drafting of the manuscript: Haffajee, Goldstick.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Haffajee, Goldstick.

Administrative, technical, or material support: Lin, Goldstick.

Supervision: Haffajee, Goldstick.

Conflict of Interest Disclosures: None reported.

Funding/Support: Dr Haffajee was supported by grant KL2TR002241 from the National Center for Advancing Translational Sciences of the National Institutes of Health. Drs Haffajee, Bohnert, and Goldstick were supported by grant 3R49CE002099-05S1 from the Centers for Disease Control and Prevention for the University of Michigan Injury Prevention Center and grant U01CE002780 from the Centers for Disease Control and Prevention. Dr Lin was supported in part by a Career Development Award (CDA 18-008) from the US Department of Veterans Affairs Health Services Research & Development Service.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Wanqi Ouyang, MS, and Lino Sanchez, MS, School for Environment & Sustainability and School of Natural Resources and Environment, University of Michigan, provided geospatial mapping assistance. Matthew Myers, MPH, Injury Prevention Center, Department of Emergency Medicine, University of Michigan Medical School, provided data management assistance. Ms Ouyang and Mr Sanchez were paid at an hourly rate for their work on this project. Mr Myers is a salaried employee who works for Dr Haffajee and provided assistance in that capacity for this study. We thank the University of Michigan Behavioral Health Workforce Center and the National Council for facilitating access to the list of prescribers of extended-release naltrexone from Alkermes Inc (free of charge).