Spatial Analysis of Drug Poisoning Deaths and Access to

Substance-use Disorder Treatment in the United States

Yelena Ogneva-Himmelberger

Department of International Development, Community and Environment, Clark University,

950 Main St., Worcester, MA, U.S.A.

Keywords: Space-time Analysis, Drug Poisoning Deaths, Medication-assisted Treatment, Opioid, Hot Spot Analysis.

Abstract: Mortality rates from drug overdose have increased exponentially throughout the US for the past 30 years.

Age-adjusted death rates from drug poisoning for 1999-2016 were analyzed at the county level using space-

time cube and hot spot analysis, and a composite index of patient access to substance-use disorder treatment

and services per each county has been calculated. More than two-thirds of all US counties have been classified

as hot spots. Combining mortality hot spots with the accessibility index highlights 81 counties with high

disease burden and low access to treatment providers. These areas deserve special attention as state and local

government and public health organizations seek new prevention and intervention strategies to address the

opioid epidemic.

1 INTRODUCTION

Substance-use disorder epidemic continues to rage

through the United States, and the mortality rates

from drug overdose have been increasing

exponentially over more than 30 years (Jalal et al.,

2018). According to the Center for Disease Control,

the age-adjusted rate of drug overdose deaths in the

country was more than three times higher in 2016

than in 1999 (Hedegaard et al., 2017), with

considerable variation in mortality rates across the

United States (Rossen et al., 2013). While scientific

evidence suggests that medication-assisted treatment

(MAT) is effective in treating substance-use

disorders, only ten percent of people with this

disorder receive any type of specialty treatment (U.S.

Department of Health and Human Services, 2016).

Three medications - methadone, buprenorphine, and

naltrexone – are safe and effective in treating

substance use disorder and opioid addiction, but their

availability at treatment facilities in the U.S. is still

limited (Jones et al., 2018). A recent survey of

facilities providing addiction treatment services

revealed that “61% of counties in the U.S. did not

have any treatment programs that offered at least one

MAT drug” (amfAR, 2018).

While several studies used GIS to analyze spatial

patterns of the opioid epidemic (Jalal et al., 2018;

Jones et al., 2018; Rossen et al., 2013; Stewart et al.,

2017), only one study used GIS and clustering to

examine spatial access to treatment and services,

focusing on buprenorphine provider availability

(Jones et al., 2018). In this research, spatial and

temporal patterns of drug poisoning death rates are

compared with patterns of access to facilities with

MAT, with the goal of identifying priority areas for

improving access to MAT. Specifically, this paper

addresses the following two research questions:

Where are the hot spots of drug poisoning death rates?

How are resources for treatment and recovery

distributed within these hot spots? The unit of

analysis is a United States county, a political and

administrative division within a state.

2 DATA

To understand where the drug overdose epidemic is

the most pronounced, age-adjusted estimates of drug

poisoning deaths per 100,000 per county were

obtained from the amfAR Opioid and Health

Indicator Database (https://opioid.amfar.org/). These

estimates are based on deaths resulting from the

following underlying causes: unintentional

poisoning, intentional/suicidal poisoning; homicidal

poisoning; and poisoning from undermined intent

Ogneva-Himmelberger, Y.

Spatial Analysis of Drug Poisoning Deaths and Access to Substance-use Disorder Treatment in the United States.

DOI: 10.5220/0007828703150321

In Proceedings of the 5th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2019), pages 315-321

ISBN: 978-989-758-371-1

315

(Rossen et al., 2017). Annual estimates of death rates

covered period from 1999 to 2016, allowing for a

spatio-temporal analysis and identification of hot

spots.

To characterize availability and access to

treatment and services, the study used the following

indicators about medical facilities and providers:

– Number of Facilities Providing Substance Abuse

Services;

– Distance to Nearest Substance Abuse Facility

providing MAT;

– Number of Facilities Providing at least one, at

least two, or all three medications used in the

treatment and Accepting Medicaid (three

indicators);

– Number of Providers Licensed to Administer

Buprenorphine.

Higher number of facilities and shorter distance to the

nearest substance abuse facility providing MAT

means better availability and access to treatment and

services.

As of March 2017, private for-profit organizations

operated 60 percent of facilities with opioid treatment

programs certified by the Substance Abuse and

Mental Health Services Administration

(Substance Abuse and Mental Health Services

Administration, 2017), meaning that some parts of the

country lack affordable options for any treatment

(U.S. Department of Health and Human Services,

2016). To include the affordability aspect, this

research analyzes facilities that provide MAT and

also accept Medicaid, a government health care

program. It is important to differentiate between the

facilities that provide only one, or two, or all three

medications because providing multiple options for

MAT increases chances of a successful treatment. A

separate variable on buprenorphine providers is also

included, because buprenorphine has several

advantages over the other two medications, including

the option of receiving weekly or monthly

prescriptions in the general office setting (Jones et al.,

2018). Physicians, physician assistants, and nurse

practitioners who have received specific training and

obtained a waiver to prescribe buprenorphine for

treatment of opioid use disorder are separate from

providers working at the substance abuse facilities, so

it is important to include them in the analysis.

All indicators mentioned above were downloaded

in a tabular format from the amfAR Opioid and

Health Indicator Database. To map these indicators,

GIS layer of county boundaries was downloaded from

the U.S. Census Bureau (https://www.census.gov/

geo/maps-data/data/cbf/cbf_counties.html), and

indicator tables were joined to GIS layer using county

FIPS codes. There are 3142 counties in the United

States. Table 1 provides details about data used in this

study.

3 METHODS

To identify spatio-temporal hot spots of drug

poisoning mortality, first a space-time cube was

created in ArcGIS Pro (ESRI, 2019). Space-time cube

approaches were previously applied to analyze

spatio-temporal patterns of traffic accidents (Cheng et

al., 2019; Rahman et al.,, 2018), crimes (Bunting et

al., 2018), and anthrax epidemics in livestock

(Abdrakhmanov et al., 2017), but haven’t yet been

applied to drug overdose mortality.

Table 1: Indicators used in the study.

Description Year

Age-adjusted Drug Poisoning Deaths per 100,000 (Modeled) 1999-2016

Number of facilities that provide substance abuse services (per 100,000) 2017

Average geographic distance in miles to travel to a substance use disorder treatment facility providing at least one

form of MAT

2017

Number of substance abuse treatment facilities offering all three MAT services (Buprenorphine, Methadone,

Naltrexone) and accepting Medicaid (per 100,000)

2017

Facilities providing at least two of the three forms of MAT and accepting Medicaid (per 100,000) 2017

Facilities providing at least one form of MAT and accepting Medicaid (per 100,000) 2017

Number of healthcare providers licensed to administer buprenorphine (per 100,000) 2018

HGIS 2019 - Special Session on GIS in Healthcare

316

Space-time cube is a collection of spatial units (in

this case, counties) layered vertically according to

time. The bottom layer of the cube corresponds to

1999, the earliest year in the dataset, and the top layer

of the cube corresponds to 2016, the latest year. Thus,

a particular county at a given year is referred to as a

bin within the space-time cube. Following this,

Emerging Hot spot analysis tool in ArcGIS Pro was

applied to identify hot spots and cold spots of

mortality. A hot/cold spot has mortality that is

significantly higher/lower at a given time than the

mean mortality for the entire space-time cube. This

tool uses the Getis-Ord Gi* statistic and Mann-

Kendall test (Getis and Ord, 1992) and categorizes

hot/cold spots into several categories based on their

temporal trends: new, consecutive, intensifying,

persistent, diminishing, sporadic, oscillating, and

historic (ESRI, 2019). To be considered a hot/cold

spot, each bin is evaluated in the context of its spatial

and temporal neighbors. Selection of the

neighborhood parameter can influence the results, so

the tool was applied multiple times with different

neighborhood parameters to explore variation in the

outcome.

To identify different levels of availability and

access to treatment and services, a composite index of

six indicators was created using the following

procedure. First, indicators measuring the number of

facilities or providers per county, were recalculated

per 100,000 persons. It was necessary to normalize by

population, because some counties are much more

populated than others. Second, all six indicators were

standardized to a range of 0-100 to be comparable,

using the following formula:

inew

= (X

ioriginal

– X

imin

) /(X

imax

– X

imin

) *100

Where X

inew

is the standardized value of an indicator

i for each county, ranging from 0 to 100;

ioriginal

is the original value of an indicator i for a

county;

imin

is the minimum value of an indicator i for the

entire country;

imax

is the maximum value of an indicator i for the

entire country

Third, these standardized indicators were aggregated

(using averaging) into a composite index of access to

treatment and services (from here on referred to as

“access index” for brevity). Averaging approach was

used as the aggregation method because it weighs all

indicators equally, is intuitive, and allows

maintaining the same scale (0-100).

To answer the second research question (“How

are resources for treatment and recovery distributed

within hot spots?”), the access index and the hot spots

were overlaid and examined using various “select by

attribute” queries in GIS. The level of access to

treatment and services for counties that fall inside hot

spots was further evaluated state by state.

4 RESULTS

Space-time cube consisted of 3142 locations

(counties) and 18 time slices resulting in 56,556

space-time bins. Emerging Hot Spot Analysis tool

was used several times, with different configurations

of neighborhood (contiguity with edges and corners;

contiguity with edges only; eight nearest neighbors)

to evaluate sensitivity. Resulting spatio-temporal

patterns of hot/cold spots were very similar, with

minor variations. Results reported below are based on

the contiguity with edges and corners because this

conceptualization of neighborhood was used in a

previous study of the drug overdose epidemic in the

United States (Jalal et al, 2018).

Figure 1 and Table 2 show distribution of hot and

cold spots over the 18-year period. More than two-

thirds of the counties (68%) experienced hot spot

trends; 30% of counties did not show any pattern, and

only 2% of counties experienced cold spot trends. Of

the counties with hot spots, 50% were an “oscillating”

hot spot, 37% - a “consecutive” hot spot, 11% - a

“new” hot spot, and 1.5% - an “intensifying” hot spot.

While significant hot spots are present in every state,

there is considerable variation in the extent of the hot

spots within each state (Map 1). For example, some

states have only a few counties in hot spots (Dakotas,

Nebraska, Kansas, Iowa, New York), while other

states are entirely covered with hot spots

(Washington, California, Nevada, Utah, New

Mexico, Michigan, Oklahoma, Florida, Tennessee,

West Virginia, Vermont, New Hampshire, Maine,

Massachusetts). There was a small number of

counties with cold spots, clustered in North Dakota,

South Dakota and Nebraska. Even though these areas

show rates lower than the national mean, they are in

diminishing and historic cold spot categories,

meaning that the rates are increasing and areas are

becoming less cold over time

Spatial Analysis of Drug Poisoning Deaths and Access to Substance-use Disorder Treatment in the United States

317

Figure 1: Hot/cold spots of Age-adjusted Drug Poisoning Deaths per 100,000.

Table 2: Number of counties in Hot/cold spots of Age-adjusted Drug Poisoning Deaths per 100,000 (1999-2016).

Type Hot spot Cold spot Description (ESRI, 2019)

New 237 0

A location that is a statistically significant hot spot for the final time step and has never been

a statistically significant hot spot before.

Consecutive 794 0

A location with a single uninterrupted run of statistically significant hot spot bins in the final

time-step intervals. The location has never been a statistically significant hot spot prior to the

final hot spot run and less than ninety percent of all bins are statistically significant hot spots.

Intensifying 32 0

A location that has been a statistically significant hot spot for ninety percent of the time-step

intervals, including the final time step. In addition, the intensity of clustering of high counts

in each time step is increasing overall and that increase is statistically significant.

Oscillating 1080 0

A statistically significant hot spot for the final time-step interval that has a history of also

being a statistically significant cold spot during a prior time step. Less than ninety percent of

the time-step intervals have been statistically significant hot spots.

Diminishing 0 43

A location that has been a statistically significant cold spot for ninety percent of the time-step

intervals, including the final time step. In addition, the intensity of clustering of low counts in

each time step is decreasing overall and that decrease is statistically significant.

Historical 0 28

The most recent time is not cold, but at least ninety percent of the time-step intervals have

een statistically significant cold spots.

Figure 2: Composite index of access to substance-use disorder treatment and services.

HGIS 2019 - Special Session on GIS in Healthcare

318

The access index ranges from 2.8 to 57.8,

representing the worst/the best availability and access

to treatment and services respectively. The mean

value of the access index for the entire country is 17.4,

and the median – 16.9. To facilitate the analysis,

composite index values were mapped using standard

deviation classification – three below the national

mean and three above the national mean, each class

corresponding to one standard deviation (Figure 2).

On this map, the darker the color the further away

from the mean is the value: dark red corresponds to

the lowest index value, and the dark green – to the

highest index value. The map shows that access to

treatment and services is distributed unevenly

throughout the country. Northeastern states and

selected counties within West Virginia, Kentucky,

Ohio, Wisconsin, Colorado, New Mexico, and

Oregon have the best access. The lowest access is

observed in the Plains region, especially Montana,

and in Texas and Nevada.

To explore access to treatment and services within

the hot spots only, the access index and hot spots were

overlaid, and three classes with access index values

below the national mean were mapped separately

(Figure 3). This map reveals that 55% of the hot spot

counties have access index below the national mean.

Of particular importance are 81 counties that fall

within two lowest access categories: 72 counties that

have very low access (i.e. their access index is

between one and two standard deviations below the

national mean), and nine counties that have extremely

low access (i.e. their access index is below two

standard deviations from the national mean). These

81 counties account for 3% of total U.S. population

and are primarily located in Texas (26 counties),

Montana (13 counties), Alaska (11 counties),

Oklahoma (six counties), Nevada (six counties) and

New Mexico (five counties). A smaller number of

counties in this category are located in Washington,

Wyoming, Utah, Arkansas, South Dakota, Michigan,

Kansas, Iowa and Idaho.

5 DISCUSSION

This study of data from 3142 counties found a

significant geographic variation in the concentration

of the drug poisoning deaths, with the majority of

counties (68%) falling inside a hot spot – an area

where the death rates are statistically significantly

higher than the national average. Among all types of

hot spots, new and intensifying hot spots are of

particular concern. The new hot spots are counties

that were never a hot spot in previous years, and

became a hot spot in 2016. These new hot spots are

often located at the fringes of previously existing hot

spots and symbolize a current frontier or opioid

epidemic. The majority of the new hot spots are

located in the South – in Texas, Georgia, Arkansas,

Louisiana, but also in Illinois and Virginia.

Figure 3: Composite index in hot spot counties. Red color shows hot spots with the lowest access to treatment and services;

orange color shows hot spots with second lowest access. Light brown color shows hot spots with access within one standard

deviation below the mean. Hatched green shows hot spots with access above the national mean.

Spatial Analysis of Drug Poisoning Deaths and Access to Substance-use Disorder Treatment in the United States

319

The intensifying hot spots are counties that

experienced increasing intensity of clustering of high

mortality rates in each time step. There are two

distinct areas of intensifying hot spots – a big cluster

of contiguous 30 counties in the Appalachia region

(Kentucky, West Virginia, Virginia) and a small

cluster of two counties in north-central New Mexico.

This study aggregated six indicators into one

composite index of availability and access to

treatment and services, instead of analyzing access-

related data separately. The averaging approach used

here is an intuitive and easy to understand, especially

when relative importance of contributing variables is

unknown. One limitation of this approach is that it

potentially compensates low scores in one variable

with high scores in another variable, thus masking a

more nuanced distribution and interaction between

the variables. Future research will consider other

ways of creating an access index.

The final index map shows that the availability of

treatment and services varies widely. When it is

overlaid with the hot spot map, some alarming

patterns of high drug overdose deaths and low

availability of treatment become evident. The study

identified 81 hot spot counties that have extremely

low access to treatment and services. In these

counties, the average distance to the closest facility

with MAT is 90 miles (minimum = 52 mi, maximum

= 415 mi). Sixty-five of these counties have no

facilities providing substance abuse services. The

remaining 16 counties have 29 such facilities, and

only one of them provides MAT with one medication.

Only seven of 81 counties have Buprenorphine

providers (total of 28 providers). These areas in

Texas, Montana, Alaska, Oklahoma, Nevada and

New Mexico need immediate attention for the local

and state public health organizations.

6 CONCLUSION

This study used a novel approach to analyze opioid

overdose death rates concurrently through space and

time, by creating a space-time cube and identifying

hot spots using GIS. Resulting hot spot maps provide

a comprehensive assessment of the geographical

patterns of death rates from drug overdose. This study

also illustrates how a composite indicator can

facilitate the assessment of accessibility and

availability of treatment and services. Combining

mortality hot spots with accessibility index spotlights

areas with high disease burden and low availability of

treatment centers and providers. These areas deserve

special attention as state and local government and

public health organizations seek new prevention and

intervention strategies to address the opioid epidemic.

REFERENCES

Abdrakhmanov, S., Mukhanbetkaliyev, Y., Korennoy, F.,

Karatayev, B., Mukhanbetkaliyeva, A., and

Abdrakhmanova, A. (2017). Spatio-temporal analysis

and visualisation of the anthrax epidemic situation in

livestock in Kazakhstan over the period 1933-2016.

Geospatial health, 12.

amfAR, t. F. f. A. R. (2018). Toward an Effective Strategy

to Combat, HIV, Hepatitis C and the Opioid Epidemic:

Recommendations for Policy Makers. Retrieved from

Washington, DC.

Bunting, R. J., Chang, O. Y., Cowen, C., Hankins, R.,

Langston, S., Warner, A., Roy, S. S. (2018). Spatial

Patterns of Larceny and Aggravated Assault in Miami–

Dade County, 2007–2015. The Professional

Geographer, 70(1), 34-46. doi:10.1080/00330124.

2017.1310622

Cheng, Z., Zu, Z., and Lu, J. (2019). Traffic Crash

Evolution Characteristic Analysis and Spatiotemporal

Hotspot Identification of Urban Road Intersections.

Sustainability, 11(1):160.

ESRI. (2019). ArcGIS Pro Help. Retrieved from

https://pro.arcgis.com/en/pro-app/help/main/welcome-

to-the-arcgis-pro-app-help.htm

Getis, A., and Ord, J. K. (1992). The Analysis of Spatial

Association by Use of Distance Statistics.

Geographical Analysis, 24(3), 189-206.

doi:doi:10.1111/j.1538-4632.1992.tb00261.x

Hedegaard H., Warner, M., and Miniño AM. (2017). Drug

overdose deaths in the United States, 1999–2016. CHS

Data Brief, no 294. Hyattsville, MD: National Center

for Health Statistics.

Jalal, H., Buchanich, J. M., Roberts, M. S., Balmert, L. C.,

Zhang, K., and Burke, D. S. (2018). Changing

dynamics of the drug overdose epidemic in the United

States from 1979 through 2016. Science, 361(6408),

eaau1184. doi:10.1126/science.aau1184

Jones, C. W., Christman, Z., Smith, C. M., Safferman, M.

R., Salzman, M., Baston, K., and Haroz, R. (2018).

Comparison between buprenorphine provider

availability and opioid deaths among US counties. J

Subst Abuse Treat, 93, 19-25.

doi:10.1016/j.jsat.2018.07.008

Rahman, M. K., Crawford, T., and Schmidlin, T. W. (2018).

Spatio-temporal analysis of road traffic accident fatality

in Bangladesh integrating newspaper accounts and

gridded population data. GeoJournal, 83(4), 645-661.

doi:10.1007/s10708-017-9791-x

Rossen, L. M., Bastian, B., Warner, M., Khan, D., and

Chong, Y. (2017). Drug poisoning mortality: United

States, 1999–2016. Downloaded from National Center

for Health Statistics https://www.cdc.gov/nchs/data-

visualization/drug-poisoning-mortality/

HGIS 2019 - Special Session on GIS in Healthcare

320

Rossen, L. M., Khan, D., and Warner, M. (2013). Trends

and geographic patterns in drug-poisoning death rates

in the U.S., 1999-2009. Am J Prev Med, 45(6), e19-25.

doi:10.1016/j.amepre.2013.07.012

Stewart, K., Cao, Y., Hsu, M. H., Artigiani, E., and Wish,

E. (2017). Geospatial Analysis of Drug Poisoning

Deaths Involving Heroin in the USA, 2000-2014. J

Urban Health, 94(4), 572-586. doi:10.1007/s11524-

017-0177-7

Substance Abuse and Mental Health Services

Administration (2017). National Survey of Substance

Abuse Treatment Services (N-SSATS). Data on

Substance Abuse Treatment Facilities.

U.S. Department of Health and Human Services, O. o. t. S.

G. (2016). Facing Addiction in America: The Surgeon

General’s Report on Alcohol, Drugs, and Health.

Retrieved from https://addiction.surgeongeneral.gov/

sites/default/files/report-highlights.pdf.

Spatial Analysis of Drug Poisoning Deaths and Access to Substance-use Disorder Treatment in the United States

321