The impact of buprenorphine prescriber data on geospatial access to treatment in HEALing Communities Study communities, 2022

Abstract

Introduction:

The location of buprenorphine treatment providers in the United States is pivotal to the understanding of regional factors associated with prescription and uptake. We evaluated how distinct data sources of treatment providers and their associated locations contribute to the differences observed when measuring buprenorphine accessibility.

Methods:

We compared buprenorphine treatment provider data from the Drug Enforcement Administration (DEA) and data from the behavioral health treatment locator from the Substance Abuse and Mental Health Services Administration (SAMHSA) for July 2022. Both DEA and SAMHSA data, while similar in spirit, vary substantially in how and why each data set is collected. DEA registration was required by law, while SAMHSA data was an opt-in registry of provider-submitted details. Analyzing the underlying semantics of the data is important for understanding the contextual factors driving observable differences in analytical outputs. We measured accessibility using enhanced two-step floating catchment area (E2SFCA) analysis in three states participating in the HEALing Communities Study (Kentucky, Ohio, Massachusetts). Within communities, we compared decile rankings of accessibility per census tract using each data source. We linked prescribing data from Kentucky’s prescription drug monitoring program (PDMP) to measure accessibility using providers prescribing buprenorphine. We explored the significance of localized rank exchanges using neighbor set local indicators of mobility association (LIMA).

Results:

The number and rate of providers per capita differed substantially at the community level between data sources in the three states. These differences were less impactful in the spatial context of buprenorphine accessibility, which required both supply and demand in regions smaller than our intervention communities. Shifts did occur when measuring the intercommunity decile ranking of accessibility of census tracts, but LIMA results indicated that these rank exchanges were not significant.

Conclusions:

When analyzing accessibility within a community using E2SFCA analyses, either DEA or SAMHSA data sources are acceptable; linkage to Kentucky’s PDMP demonstrated that SAMHSA provider data is equally suitable to PDMP data for research studies involving spatial relationships with providers while being both significantly easier to obtain and less sensitive. When analyzing treatment provider rates per capita, results may vary substantially across these different data sources. Therefore, context must be considered when choosing an appropriate data source to use.

1. Introduction

In 2023, opioids contributed to 78,226 out of 103,793 (75.4%) reported overdose deaths (Vital Statistics Rapid Release - Provisional Drug Overdose Data, 2024). Buprenorphine is a medication used to treat opioid use disorder (OUD) by partially activating receptors in the brain that receive opioids, reducing withdrawal symptoms, cravings, and pain (Johnson et al., 2003). The Drug Addiction Treatment Act of 2000 (DATA 2000) allowed qualified physicians to apply for a waiver that would enable them to treat opioid dependency with buprenorphine products approved by the Food and Drug Administration (FDA) through December of 2022 (Cioe et al., 2020). To obtain the waiver, physicians needed to be licensed under state law, registered with the Drug Enforcement Administration (DEA), and to have completed a certification or training course. Between 2009 and 2018, buprenorphine treatment per 1,000 population increased annually from 1.97 to 4.43 which, although a positive trend, was still lower than the estimated rates for OUD (Olfson et al., 2020). National OUD estimates vary greatly according to methodology with ranges from 3.2% to 4.1% (2023 NSDUH Detailed Tables, 2024; Saha et al., 2016). Geography influences OUD prevalence; for example, in Kentucky, OUD estimates per county range from 1.3% to 17.7% (Thompson et al., 2023). In December 2022, the Mainstreaming Addiction Treatment (MAT) Act was passed to reduce federal policy barriers by removing the waiver requirements and limitations on the number of patients in treatment (Varisco et al., 2023). Despite federal efforts, 15 states, including Kentucky and Ohio, had restricted state-level policies requiring waiver training or additional counseling requirements, challenging the implementation of the MAT Act (Silwal et al., 2023). A survey of state medical, nursing, physician assistant, and osteopathic boards and Single State Agencies indicated that 17 states did not realize federal requirements had changed, which suggests shared governance across federal, state, and local levels is a barrier to consistent policy change (Silwal et al., 2023).

In 2019, over 75% of counties in the United States did not have an opioid treatment program which provide medications for opioid use disorder (MOUD) including methadone and buprenorphine (Corry et al., 2022); in 2020, only 45% of counties had a facility offering any type of MOUD (Cantor et al., 2021). Despite more than 20 years of authorized buprenorphine treatment for OUD, geographic disparities exist and access to buprenorphine remains challenging for many communities (Geographic Disparities Affect Access to Buprenorphine Services for Opioid Use Disorder, 2020). In 2018, 40% of US counties did not have any waivered providers registered with the DEA and more than half of the counties in highest need of services did not have adequate capacity to treat patients (Geographic Disparities Affect Access to Buprenorphine Services for Opioid Use Disorder, 2020). A survey from 2020 indicated 30% of pharmacies reported limitations when filling buprenorphine prescriptions, with 20% not filling these prescriptions at all (Kazerouni et al., 2021). Furthermore, reports of the scarcity of waivered providers or waivered providers accepting new patients suggest both the need to better understand geographic access to buprenorphine and to specifically target problematic regions with efforts designed to increase provider participation (Andrilla et al., 2018; Beetham et al., 2019; Gimenez et al., 2024; Saunders et al., 2022). Racial and economic disparities are related to access to treatment programs and longer travel distances (Amiri et al., 2024; Hsu et al., 2024). Rural areas have poor geographic access to buprenorphine regardless of the community’s social vulnerability (Joudrey et al., 2022). Additionally, long distances to treatment resources are related to reduced retention of treatment for OUD (Villamil et al., 2024).

The HEALing (Helping to End Addiction Long-term^®) Communities Study (HCS) is a multi-site study with the overarching objective of implementing evidence-based practices (EBPs) to significantly reduce opioid-related overdose fatalities in 67 urban and rural communities across four states (Kentucky, Massachusetts, New York, and Ohio) (Walsh et al., 2020). HCS examined the effectiveness of the Communities That HEAL (CTH) intervention, which connects communities and EBPs through the Opioid-overdose Reduction Continuum of Care Approach (ORCCA) (Stopka et al., 2024; Winhusen et al., 2020). The ORCCA provides thematic menus of EBP: (1) overdose prevention education and naloxone distribution, (2) effective delivery of MOUD, and (3) safer opioid prescribing and dispensing (Winhusen et al., 2020). The EBPs from these domains have been shown to reduce opioid overdose risk while community engagement strategies can be used to improve their uptake and sustainability (Sprague Martinez et al., 2020). Buprenorphine plays a pivotal role in the second ORCCA menu, effective delivery of MOUD. This menu includes strategies for maintaining, linking, and expansion of MOUD treatment availability. Furthermore, trends in opioid overdose and MOUD are closely monitored via dashboards made available to community members engaged in the intervention (Wu et al., 2020).

The location of treatment providers is pivotal in understanding regional factors that influence potential prescribing and uptake of buprenorphine. There are multiple data sources available for the location of waivered providers. The DEA maintains an active registrant database for those receiving DATA 2000 waivers to prescribe buprenorphine for OUD in the US; registration was required to receive the waiver prior to the MAT Act of 2022. SAMHSA maintains a behavioral health treatment locator with a roster of buprenorphine practitioners (Buprenorphine Treatment Practitioner Locator, 2024); providers must actively consent to release their practice information within SAMHSA’s buprenorphine locator. These two data sources were created and are maintained for very different purposes. The DEA’s provider database contains data for anyone who acquired a waiver, although obtaining a waiver does not imply that the provider uses it in practice. The SAMHSA database includes providers electively opting in to advertise themselves as providers capable of prescribing buprenorphine, suggesting they are willing to prescribe buprenorphine to those in need. Consequently, the SAMHSA list is expected to have fewer prescribers. Given these differences, we explore how data sources for treatment providers impact the measurement of buprenorphine accessibility.

2. Methods

2.1. Data

The overall goal of the study was to compare buprenorphine accessibility using provider data available from the DEA registrant database and the SAMHSA treatment locator for June 2022. We obtained statewide DEA data through the HCS for the purpose of measuring the CTH intervention and its impact on the availability of waivered providers. To identify each buprenorphine provider in the DEA database, we used the business activity code variable that described the type and limit of the practitioner: civilian assistant physicians (MM, MN, MP, MS), civilian physician assistants (MG, MI, ML, MR), civilian nurse practitioners (MF, MH, MK, MQ), and civilian physicians (C1, C4, CB, CK). For comparison, we obtained buprenorphine data from the SAMHSA treatment locator database (Buprenorphine Treatment Practitioner Locator, 2024). For both provider databases, we restricted the location of analysis to Kentucky (KY), Massachusetts (MA), and Ohio (OH). The study protocol (Pro00038088) was approved by Advarra Inc., the HEALing Communities Study single Institutional Review Board.

2.2. Analysis

Data cleaning and geocoding:

We converted the raw dataset into geographical information systems (GIS) point feature shapefiles through several iterations of data cleaning and geocoding using ArcGIS Pro 3.1. We assessed the quality of the data contents (e.g., duplicates, bad addresses) and how adequately the data described the reality of the buprenorphine prescribing. Both datasets contained provider addresses in the format of line 1, line 2, city, state, and zip code. We removed duplicates from each data source where the name and location were repeated. We normalized address line fields to ensure street addresses were being captured instead of other fields such as business names. We geocoded all addresses to determinate longitude and latitude coordinates for each record. The SAMHSA data also contained precalculated geographic coordinates, but the agreement between the coordinates and the address fields was modest for some providers, suggesting that the address and coordinates were out of sync. Therefore, we geocoded the address location provided in the dataset to obtain the point feature shapefiles required for GIS analysis.

Comparison of DEA and SAMHSA data using GIS methods:

We use four methods of comparison to ensure robust assessment of potential differences that exist between accessibility indices derived from the DEA and SAMHSA datasets. These methods reflect accessibility metrics that are commonly utilized in published studies (Luo & Qi, 2009; Mitchell et al., 2022; Rey, 2016; Wang, 2000).

1). Buprenorphine provider rates:

We started with a baseline evaluation of buprenorphine providers per 100,000 individuals aged 18 years or older at the county level for Kentucky and Ohio and at the municipality level for Massachusetts, based on intervention site selection. We used the 2020 National Center for Health Statistics single-race resident population estimates as the denominator for the county-level analysis in Ohio and Kentucky (Friede et al., 1993); we used population estimates from the American Community Survey for municipalities in Massachusetts (US Census Bureau, 2022). After the calculation of rates, we created maps to visualize the difference between the DEA and SAMHSA data as rate ratios for each state.

2). Enhanced two-step floating catchment area (E2SFCA) index:

Floating catchment area (FCA) models attempt to balance the interaction between supply, demand, and distance (Wang, 2000); the enhanced two-step floating catchment adds a distance decay weight to reflect that nearer resources are generally considered more accessible and that some resources beyond a threshold are unreachable (Luo & Qi, 2009). When calculating buprenorphine accessibility with E2SFCA analyses, there are multiple key components: the location and supply of waivered providers eligible to prescribe buprenorphine, the demand location of the population using weighted-centroids, and the average commute time within a community. The E2SFCA model is represented as:

$$A_i^G={\sum }_{j=1}^n\frac{S_j{d}_{ij}^{-\beta }}{{\sum }_{k=1}^m{P}_k{d}_{kj}^{-\beta }}$$

where, ${{\mathrm{A}}_{\mathrm{i}}}^{\mathrm{G}}$ is the indicator of accessibility at the population location $\mathrm{i},\mathrm{S}$ is the supply of services at location $\mathrm{j},\mathrm{P}$ is the demand (population), $\mathrm{d}$ is the distance, and $\mathrm{\beta }$ is the friction coefficient of distance (Cliff et al., 1974). Accessibility is calculated within communities using census tracts to better understand geographic variation. Population is used as a proxy for demand due to actual OUD estimates being unavailable for census tracts; OUD estimates would represent the actual demand, but these are unavailable for areas smaller than the county level. We previously applied E2SFCA analyses to measure buprenorphine accessibility in HCS communities (Shrestha et al., 2023). Adult population totals and average commute times were taken from the American Community Survey five-year estimates from 2016–2020 (24.3 minutes for KY, 30.2 minutes for MA, 23.5 minutes for Ohio). E2SFCA has also been used to measure the accessibility of buprenorphine in regions outside of HCS (Mitchell et al., 2022). We ran the E2SFCA method for HCS communities and varied our provider data between SAMHSA and DEA data sources; we summarized how the two results differed by comparing decile rankings of accessibility.

3). Accessibility Decile Rankings:

The accessibility score of a census tract is useful only in the relative context of comparing it to other census tracts and their E2SFCA scores (e.g., lower scores imply lower accessibility and higher scores imply higher accessibility). For this reason, we cannot directly compare the E2SFCA scores generated from two separate runs using different data sources. Instead, we use E2SFCA scores to rank the accessibility of buprenorphine and collapse communities with similar scores by computing the decile to which their accessibility score belongs. In doing so, we can observe how the decile of a community’s rank changes when varying the data source.

4). Local indicator of mobility association (LIMA):

We explored the significance of localized rank exchanges using neighbor set LIMA, which considers the degree of concordance or discordance in the neighbor set of local spatial contexts (Rey, 2016). LIMA was originally developed for the study of regional income mobility, where spatial concordance can be used to identify concordance or discordance of income mobility considering the spatial context (Rey, 2016). This methodology allowed us to analyze the significance of observable shifts in accessibility rankings using the two different provider data sources, which is important because the accessibility indices are relative scores that are only directly comparable to each other in a single data source.

Comparison of DEA and SAMHSA data using data linkage to prescription drug monitoring programs:

Both the DEA registrant database and SAMHSA treatment locator contain potential prescribers of buprenorphine. Kentucky linked the provider data to the state’s prescription drug monitoring program, KASPER (Kentucky All Schedule Prescription Electronic Reporting) to determine actual prescribers. KASPER does not release a list of provider locations, so linkage to a provider database containing address data, such as the DEA or SAMHSA data, is necessary for our accessibility analyses. Other states were unable to perform this linkage due to the data being unavailable at the time of our analyses. We directly linked the PDMP data to the DEA data using DEA numbers available in each data set; SAMHSA did not have DEA numbers, so we first linked SAMHSA to DEA data using fuzzy matching on name and location. Our PDMP data did not have prescriber locations; the locations specified in the DEA and SAMHSA databases were used for analysis.

3. Results

3.1. Data Cleaning

We removed duplicate records from both the SAMHSA data (2.2% in KY, 4.2% in MA, 5.9% in OH) and DEA data (0.01% in KY, 0.09% in MA, 0.06% in OH) where name and location were repeated. In Kentucky, address line 1 sufficiently described most of the records (85.2% for SAMHSA and 92.5% for DEA); address line 2 helped recover an additional 13.9% of SAMHSA records and an additional 7.4% of DEA records. The most common issue with address line 1 and address line 2 is that strings other than a street address were included, such as business names. For KY, adding address line 2 when needed yielded 99.1% (SAMHSA) and 99.9% (DEA) of the addresses that were recoverable for geocoding; less than 1% of the addresses did not contain a street number or were PO boxes, which either limit geocoding precision or prevent geocoding. OH and MA had similar rates above 99%.

3.2. Buprenorphine Provider Rates

Figure 1 shows the ratio of the DEA to SAMHSA provider rates per state to indicate which communities had more or fewer providers in the DEA data than the SAMHSA data. These findings demonstrated that the differences observed are state-wide and that regional differences may exist community to community.

Figure 1:

Ratio of providers in DEA registrant data to SAMHSA treatment locator data is colored according to magnitude. Communities that had fewer providers in the DEA data than the SAMHSA data are colored dark to light blue; communities that had more providers in the DEA data are colored white to dark red). Some communities did not have a provider listed in the SAMHSA data, so a ratio was not able to be calculated; these communities are depicted with hatch marks.

Roughly one-third of Kentucky counties had a DEA to SAMHSA provider ratio of one or less; 11 counties (9%) had DEA to SAMHSA provider ratios greater than two. Only 66% of the SAMHSA providers in Kentucky matched records in the DEA data. Figure 1 shows that roughly twenty percent of Ohio’s 88 counties had a DEA to SAMHSA provider ratio of one or less; 10 counties (11%) had DEA to SAMHSA provider ratios greater than two. In Massachusetts the DEA to SAMHSA provider ratio was much higher in Boston and its adjoining suburbs as compared to the rest of the state.

3.3. Enhanced two-step floating catchment area (E2SFCA) index

The E2SFCA results are visualized in Figure 2, where column A is SAMHSA and column B is DEA; the darker sections of the map indicate areas of higher buprenorphine accessibility. E2SFCA yields a numerical score, where smaller values represent lower accessibility, and higher values represent greater accessibility.

Figure 2:

E2SFCA results for buprenorphine accessibility across census tracts in KY, MA, and OH, 2022. Column A results relied on SAMHSA data and column B results relied on DEA data. The darker red sections of the map, with higher scores, indicate areas of higher buprenorphine-waivered provider accessibility. Lower scores, represented by yellow shading, indicate lower buprenorphine provider accessibility.

Kentucky’s spatial accessibility results identified in Figure 2 were stable regardless of which provider data set was used for the supply component of E2SFCA. The differences in provider numbers summarized in Figure 1 did not appear to impact the results of our E2SFCA calculations. Kentucky has consistent areas of high accessibility in urban areas and in rural parts of eastern Kentucky. Similar to Kentucky, Ohio’s spatial accessibility was comparable across data sets, and both sets revealed substantial disparities in buprenorphine access across the state. The EF2SCA results demonstrated that counties in the central and more rural parts of Ohio tend to have poorer access to buprenorphine providers compared to urban areas such as Cuyahoga, Hamilton, and Franklin counties. Of note, these counties contain Ohio’s largest cities: Cleveland, Cincinnati, and Columbus. For Massachusetts, Figure 2 suggested that the high-access census tracts were also in Boston and its neighboring suburbs, but other high-access locations existed along the Interstate 91 corridor and in western Massachusetts in areas such as Springfield and Pittsfield.

3.4. Accessibility Decile Rankings

We quantified the shifts in accessibility scores in Table 1 which shows a census tract decile ranking using DEA (horizontal ranking) and SAMHSA (vertical ranking); the diagonal cell is highlighted to show the percentage of rankings that agree between the two data sources. Using the first row as an example, 76.5% of communities in the lowest decile using DEA data did not exchange rank when using SAMHSA data; 21.0% went from the 1^st decile to the 2^nd decile; 0.62% went from the 1^st decile to the 3^rd decile; surprisingly, 1.85% went from the lowest decile to the highest decile.

Table 1:

Percent of decile (1–10) rank exchanges of accessibility for census tracts using DEA (horizontal decile) and SAMHSA (vertical decile) in Massachusetts (MA), Kentucky (KY), and Ohio (OH). The diagonal (in bold) represents percent in agreement, which is typically the largest percentage; exceptions are noted in italics where the largest percentage of records for this decile were the result of a rank exchange.

State		1	2	3	4	5	6	7	8	9	10
MA	1	76.54	20.99	0.62	0	0	0	0	0	0	1.85
	2	16.15	50.31	26.09	6.21	0.62	0	0	0	0.62	0
	3	4.35	18.63	48.45	22.98	4.35	1.24	0	0	0	0
	4	2.47	6.79	17.9	38.89	28.4	4.32	0	0	0	1.23
	5	0.62	1.86	4.97	24.84	34.16	26.09	7.45	0	0	0
	6	0.00	0.62	1.24	4.97	14.91	36.02	37.27	4.35	0	0.62
	7	0.00	0.62	0.62	1.85	17.28	12.35	33.33	33.95	0	0
	8	0.00	0	0	0	0	19.25	7.45	50.31	22.98	0
	9	0.00	0	0	0.62	0	0	14.29	8.07	66.46	10.56
	10	0.00	0	0	0	0	0.62	0.62	3.09	9.88	85.8
KY		1	2	3	4	5	6	7	8	9	10
	1	65	18.8	6.1	3.6	1.8	0.9	1.8	1.8	0	0
	2	26.8	41.8	25.9	4.3	0.9	0.3	0	0	0	0
	3	6.4	29.8	37.4	21.3	4	0.6	0.3	0.3	0	0
	4	1.5	7.6	22.3	38.4	21	5.8	1.2	1.8	0.3	0
	5	0	1.5	5.2	20.1	39.2	22.8	8.8	1.8	0.6	0
	6	0	0.3	1.8	9.1	20.4	35.4	23.8	7.3	1.8	0
	7	0	0	0.9	1.5	10.1	23.2	28	33.2	3	0
	8	0.3	0	0	0.3	1.8	7.3	26.4	28.6	34.3	0.9
	9	0	0	0.3	1.2	0.6	3.4	7.9	21	41.2	24.4
	10	0	0	0.3	0	0.3	0.3	1.5	4.3	18.5	74.8
OH		1	2	3	4	5	6	7	8	9	10
	1	78.16	18.99	2.22	0	0.63	0	0	0	0	0
	2	16.14	52.53	23.1	6.01	1.9	0.32	0	0	0	0
	3	3.48	20.25	41.46	25.32	8.86	0.63	0	0	0	0
	4	0.95	6.03	23.49	34.29	29.52	5.4	0.32	0	0	0
	5	0	1.58	7.59	26.27	30.7	27.85	6.01	0	0	0
	6	0	0.32	1.27	4.11	20.25	33.86	25.95	12.66	1.58	0
	7	1.27	0	0.63	1.27	3.17	13.97	29.21	39.05	11.11	0.32
	8	0	0.32	0.32	1.9	1.9	10.76	13.61	26.9	38.29	6.01
	9	0	0	0	0	3.16	5.7	13.92	12.34	23.73	41.14
	10	0	0	0	0.63	0	1.58	10.76	9.18	25.32	52.53

Exchanging ranks becomes more frequent toward the middle deciles, since the lowest deciles can only improve, and the highest deciles can only worsen. The diagonal represents the percentage of communities not exchanging ranks which is typically the largest portion of records per decile. We highlight those infrequent occurrences with italics, where perfect agreement is not the highest value in the row. Overall, very low percentages of communities shifted by more than one decile when comparing results across DEA and SAMHSA datasets. When comparing states, Table 1 shows that most communities in all three states did not change ranks, but some rank exchanges did occur. Rank exchanges were more common in Ohio compared to the other states, but exchanges were still largely limited to nearby deciles. Specifically for Ohio, 39.05%, 38.29% and 41.14% of communities in the 7th, 8th and 9th deciles using DEA data changed rank when using the SAMSHA data.

3.5. Local indicator of mobility association (LIMA)

LIMA results support that numerous rankings of buprenorphine provider accessibility were exchanged amongst the census tracts of communities across a variety of deciles (Figure 2, Table 1). LIMA was used to determine if these rank exchanges are spatially relevant. In significant neighbor set LIMA results, red areas represent spatial discordance and imply significant rank exchanges occurred among neighboring census tracts (Figure 3). Blue areas represent concordance where significant rank exchanges were not observed among a census tract’s neighbors. White areas represent areas with insignificant discordance and concordance. For all three states, we observed wide concordance for the ranks between the accessibility index derived from SAMHSA and DEA datasets. However, there were regions of discordance or neighborhood rank exchanges in Kentucky in urban areas such as Louisville and the more densely populated regions of the state. Figure 3 suggests that locations of spatial discordance in Kentucky included suburban/rural in Owen, Gallatin, Shelby, Greenup, Butler, and Marshall counties. In Ohio, we observed discordance in Paulding, Henry, Shelby, and Medina counties. These locations appear to have very different access depending on the source of data. This is also observed in comparing the E2SFCA maps in Figure 1.

Figure 3:

Buprenorphine-waivered provider accessibility ranking comparisons based on SAMHSA vs. DEA data in Kentucky, Massachusetts, and Ohio census tracts, 2022. Neighbor set LIMA, where blue areas represent spatial concordance (no or limited rank exchanges) and red areas represent rank discordance (rank exchanges with neighbors).

Figure 3 reported very few areas per state where discordance was observable within a community’s neighbor set. There are cases where communities in the upper deciles using one data source are downgraded to the lower deciles using the alternate data source, but most of the disagreements are simply shifts to the next closest decile. Observable differences are expected when utilizing different input data when calculating spatial accessibility; our results demonstrate these differences are not significant.

3.6. Data Linkage to Prescription Drug Monitoring Programs

Kentucky linked the DEA provider data to KASPER (Kentucky All Schedule Prescription Electronic Reporting) to determine how many of the DEA providers had corresponding buprenorphine prescriptions; this linkage is summarized in Table 2, which delineates those who were authorized to prescribe and those who prescribe in practice. Among Kentucky DEA providers, 971 (88%) of the providers were matched to prescriptions in the PDMP data, but only 48% of those had prescribed buprenorphine as MOUD. Because we studied specific communities as part of the CTH intervention, the attribution of providers to communities was required; it is worth noting that requiring the patient’s residence location match the provider’s community reduced the number of matches from 48% to 39%. Kentucky also linked the data to SAMHSA providers (Supplemental Table 1). Far fewer providers were linked from the SAMHSA data compared to the DEA data, but the percentage of providers from SAMHSA actively prescribing buprenorphine in KY’s PDMP was 67%, which is greater than the DEA where only 48% were prescribing buprenorphine. It is worth noting that four of the counties with suppressed values for provider counts in the supplemental table were able to link 100% of their providers to the KASPER data; two of these counties had 100% of their providers having prescribed buprenorphine. The accessibility scores using linked PDMP data demonstrated similar trends in rank exchanges as seen in Table 1.

Table 2:

Buprenorphine-waivered DEA providers and prescription rates after linkage to Kentucky’s prescription drug monitoring program (PDMP), KASPER (Kentucky All Schedule Prescription Electronic Reporting) Program for HEALing Communities Study counties in Kentucky, 2022. Rates were calculated for providers having any prescription (Any Rx) and for those having specifically buprenorphine (Bup. Rx).

Community	Providers in DEA	With Any Rx in PDMP	With Any Rx (Same County) in PDMP	With Bup. Rx in PDMP	With Bup. Rx (Same County) in PDMP
Bourbon	6	1–5*	1–5*	1–5*	1–5*
Boyd	45	35 (78%)	33 (73%)	25 (56%)	23 (51%)
Boyle	28	27 (96%)	23 (82%)	15 (54%)	8 (29%)
Campbell	14	13 (93%)	13 (93%)	1–5*	1–5*
Carter	10	9 (90%)	8 (80%)	7 (70%)	6 (60%)
Clark	19	17 (89%)	17 (89%)	14 (74%)	13 (68%)
Fayette	389	349 (90%)	319 (82%)	195 (50%)	157 (40%)
Floyd	35	32 (91%)	30 (86%)	21(60%)	19 (54%)
Franklin	18	18 (100%)	14 (78%)	11(61%)	7 (39%)
Greenup	6	6 (100%)	1–5*	1–5*	1–5*
Jefferson	360	316 (88%)	309 (86%)	140 (39%)	128 (36%)
Jessamine	10	6 (60%)	6 (60%)	6 (60%)	6 (60%)
Kenton	76	61 (80%)	56 (74%)	31 (41%)	24 (32%)
Knox	40	38 (95%)	35 (88%)	22 (55%)	16 (40%)
Madison	39	36 (92%)	35 (90%)	25 (64%)	23 (59%)
Mason	4	1–5*	1–5*	0 (0%)	0 (0%)
Any HCS County	1099	971 (88%)	909 (83%)	524 (48%)	439 (39%)

^*Five or fewer providers; suppressed to preserve privacy.

4. Discussion

Access to buprenorphine is critical in reducing the risk of opioid-related overdose. While there has been significant research on access to buprenorphine and methadone, the underlying data for provider locations has varied across studies. Using incomplete or inaccurate buprenorphine provider databases may lead to poor estimates of access. In our study, we compared the two most used data sources for buprenorphine providers – DEA waivered provider data and SAMHSA treatment locator data.

Our findings indicate that DEA data had a significantly higher number of buprenorphine providers listed in the data compared to SAMHSA data. There were substantial differences between provider data sources with respect to provider ratios at the community level, where as high as a 10 to 1 ratio of DEA providers compared to SAMHSA providers was observed. These differences can be easily explained since the SAMHSA buprenorphine locator is opt-in and voluntarily self-reported by physicians who obtained data waivers, whereas registration with the DEA was legally required at the time of the study. To be part of the SAMHSA database, providers are required to share with SAMHSA their name, DEA number, state license number, and contact information, which includes full address, phone, fax, and email; additionally, the provider must submit updates to SAMHSA when anything changes, such as moving locations which may imply that data collection and accuracy issues exist as a natural consequence. Given these differences, analyses that directly dependent upon the rate per capita of providers must take these contextual differences into consideration when reporting results. Furthermore, the ratio of DEA to SAMHSA providers varied per county for all states (Figure 1), suggesting that overlap between the two data sources exists, but county assignment per provider may be inconsistent.

Spatial analyses go beyond the community level and into smaller subregions. We used the E2SFCA method to calculate buprenorphine accessibility per census tract within a community. E2SFCA requires supply (providers), demand (population), and a drive-time metric for reasonable reachability. Although we observed significant differences in the rates of providers per data set at the community level, these differences did not substantially impact the measurement of buprenorphine accessibility at the census tract level. Furthermore, we did not observe any significant differences in accessibility deciles after linking to Kentucky’s PDMP data. Only about half of the providers in the Kentucky DEA data had actually prescribed buprenorphine, which is a notable gap but largely consistent with observations from other studies focused on providers actively prescribing buprenorphine (Lanham et al., 2022; Meyerson et al., 2024). Complementing this, SAMHSA had fewer providers overall, but 88% were prescribing buprenorphine (Supplemental Table 1). Despite these gaps, our spatial accessibility results identified in Figure 2 were stable regardless of which provider data was used for the supply component of E2SFCA.

Areas of low accessibility were generally still low and high areas were generally still high when using any data source. For both data sources, we found that all states had consistent areas of high accessibility in urban areas and areas of lower accessibility in rural areas, which is consistent with other published findings (Joudrey et al., 2022; Mitchell et al., 2022). Rural regions of eastern Kentucky had areas of high accessibility, which we previously observed in a related study using DEA data as the source of provider locations (Shrestha et al., 2023). The rural parts of eastern Kentucky that have high accessibility are likely due to efforts aimed to address the region’s long history in the opioid crisis (Luu et al., 2019). We confirmed that ranks of buprenorphine accessibility are exchanged among census tracts across deciles, which is a natural consequence of using different data sources for provider locations. However, movement was generally limited to neighboring ranks because these entries contained the largest percentage of overlap. The LIMA analysis revealed a high degree of spatial concordance and no significant shifts in accessibility rankings using the two provider data sets, indicating lower exchange mobility (Figure 3).

Our study has several implications for the monitoring of buprenorphine accessibility. In our related work, we used DEA provider data to report accessibility of buprenorphine as part of the Communities That HEAL Intervention implemented by the HEALing Communities Study (Shrestha et al., 2023). Our findings suggest that for spatial accessibility studies both DEA and SAMHSA data are suitable and capable of producing similar results. These results may be used to identify areas of low buprenorphine accessibility which in turn can be targeted by evidence-based practices designed to increase the number of providers willing to prescribe buprenorphine. Furthermore, the SAMHSA provider data is publicly available and can be downloaded directly from the treatment locator website; the DEA registrant data were obtained using a Freedom of Information Act request. Furthermore, PDMP data is highly sensitive due to patient privacy considerations, and Kentucky is one of only 16 states in 2022 that reported allowing record linkage for research purposes (Lee et al., 2022). The findings of this paper demonstrate that these lists are not perfect but remain adequate for spatial studies despite their natural limitations. Given the overhead of obtaining DEA data or the potential impossibility of obtaining linked PDMP data for research, the SAMHSA provider data may provide both suitable and expedient data for research studies examining spatial relationships involving providers.

Due to changes in the DATA 2000 waiver requirements, the specifics of DEA registrant data are less important because buprenorphine does not require specific waivers or business codes for prescribing. Consequently, anyone in the DEA database is potentially a relevant prescriber and the database now simply provides provider locations. Furthermore, the landscape of treatment locators are changing; SAMHSA has implemented findtreatment.gov which was authorized by the 21st Century Cures Act and contains location information for resources related to substance use, mental health, healthcare centers, buprenorphine practitioners, and opioid treatment programs (FindTreatment.Gov, 2024). Despite longstanding accuracy concerns, treatment search tools and directories are useful for helping those in need find treatment and highlights the need for a reliable, national effort to manage and validate data (Riley et al., 2024).

Our study is limited by the embedded accuracy of provider addresses in the DEA registrant data and SAMHSA locator data. In order to attribute a provider to a specific community, we geocoded the address listed, but providers may work in multiple regions, may move between regions, work via telehealth, or may substantially change roles and no longer be in a position to prescribe buprenorphine. Prior studies have reported varying degrees of accuracy when using SAMHSA data (Barenie et al., 2022; Lanham et al., 2022; Meyerson et al., 2024). Since OUD estimates are not available at the census tract, our E2SFCA calculations used adult population estimates for the demand component, which may not accurately reflect demand. Our E2SFCA implementation considered edge effects of neighboring communities within the same state, but there may still be underestimation occurring near state borders since we did not have provider data for neighboring states. Furthermore, we did not look at differences in race, ethnicity, and rurality to understand the impact of regional disparities on accessibility. Our study focused on buprenorphine and did not take into account the spatial impact of opioid treatment programs, which primarily administer methadone and methadone availability also suffers from accessibility issues (Iloglu et al., 2021; Joudrey et al., 2020). Buprenorphine for MOUD may be administered in an emergency department or office setting and is captured through KASPER. We also note that the three states analyzed may not be representative of other locations.

5. Conclusion

Our results indicate that spatial studies, such as measuring buprenorphine accessibility via E2SFCA, are not heavily impacted by the differing raw rates of providers included in the reference data since the volume of providers is only one component of the E2SFCA algorithm. The spatial distribution of providers relative to other parts of a community heavily influences accessibility. Shifts in ranks of accessibility by decile do occur, but typically they are limited to neighboring ranks. Our work supports that both context and application are paramount when choosing provider data.

Highlights.

• Provider data from the Drug Enforcement Administration (DEA) and the Substance Abuse and Mental Health Services Administration (SAMHSA) are substantially different in number and rate per capita due to differences in how and why each data set is created and maintained.

• Despite substantial differences, both DEA and SAMHSA provider data yield similar results when used to calculate spatial access to buprenorphine using an enhanced two-step catchment area (E2SFCA) analysis.

• Shifts did occur when measuring intercommunity decile ranking of buprenorphine accessibility, but these changes were not significant and are typically limited to neighboring deciles.

• Buprenorphine accessibility calculated using prescription drug monitoring program data yielded similar results to both DEA and SAMHSA provider data, which suggests less sensitive, easier to obtain data sources such as SAMHSA are equally suitable for measuring accessibility.