Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Oct 1.
Published in final edited form as: Drug Alcohol Depend. 2023 Aug 23;251:110947. doi: 10.1016/j.drugalcdep.2023.110947

Applied Risk Mapping and Spatial Analysis of Address-Level Decedent Data to Inform Opioid Overdose Interventions: The Massachusetts HEALing Communities Study

Jennifer Pustz 1, Sumeeta Srinivasan 2, Shikhar Shrestha 1, Marc R Larochelle 3, Alexander Y Walley 3, Jeffrey H Samet 3,4, Hermik Babakhanlou-Chase 5, Jane F Carpenter 6, Thomas J Stopka 1
PMCID: PMC10587829  NIHMSID: NIHMS1929269  PMID: 37666091

Abstract

Background:

Death certificate data provide powerful and sobering records of the opioid overdose crisis. In Massachusetts, where address-level decedent data are publicly available upon request, mapping and spatial analysis of fatal overdoses can provide valuable insights to inform prevention interventions. We describe how we used this approach to support a community-level intervention to reduce opioid-involved overdose mortality.

Methods:

We developed a method to clean and code decedent data that substituted injury locations (the likely location of fatal overdoses) for deaths recorded in hospitals. After geomasking for greater privacy protection, we created maps to visualize the spatial distribution of decedent residence addresses, alone and juxtaposed with drive and walk-time distances to opioid treatment programs (OTPs), and place of death by overdose address. We used spatial statistical analyses to identify locations with significant clusters of overdoses.

Results:

In the 8 intervention communities, 785 individuals died from opioid-involved overdoses between 2017 and 2020. We found that 19.7% of fatal overdoses were recorded in hospitals, 50.2% occurred at the decedent’s residence, and 30.1% at another location. We identified overdose hotspots in study communities. By juxtaposing decedent residence data with drive- and walk-time analyses, we highlighted actionable spatial gaps in access to OTP treatment.

Conclusion:

To better understand local fatal opioid overdose risk environments and inform the development of community-level prevention interventions, we used publicly available address-level decedent data to conduct nuanced spatial analyses. Our approach can be replicated in other jurisdictions to inform overdose prevention responses.

Introduction

Drug overdose deaths in the United States increased by 30% between 2019 and 2020, driven in part by the presence of fentanyl in illicit opioid and stimulant supplies.1,2 Overdose death increases have expanded geographically, beginning with sharp surges in Appalachia and the Northeast in 2014, now including all regions of the US regardless of population density.3,4 In Massachusetts, more than 2,000 fatal opioid-involved overdoses have occurred each year since 2016. The opioid-involved overdose mortality rate reached a new high in 2022 (33.5 per 100,000 population), the third consecutive year of increases, following three years of decline to 28.8 per 100,000 in 2019.5 While the overdose rate among non-Hispanic White residents (33.3 per 100,000) declined slightly in 2022, rates continued their upward trend to the highest rates recorded among non-Hispanic Black (51.7 per 100,000), Hispanic (45.5 per 100,000), and non-Hispanic Asian/Pacific Islanders (3.9 per 100,000).6,7 Given the constantly changing nature of opioid-involved overdose mortality patterns, monitoring the spatial distribution of fatal overdoses may help identify high risk locations and inform targeted public health and clinical interventions that aim to decrease opioid overdose rates over time.8

Death certificate data provide powerful and sobering records of lives lost to opioid overdose. Spatial epidemiologists routinely use these data as part of surveillance activities;9,10 however, there are few published examples that use address-level descriptive mapping to monitor overdose morbidity and mortality. Many spatially-oriented studies start with individual-level data from death certificates that are used for point-pattern analyses, cluster analyses or spatial scans, for calculating variables such as distance, or aggregated to another spatial unit, such as a census tract or block group, for mapping and analysis within a geographic information system (GIS), all of which involve robust methods and provide valuable insight.11–15 To date, no studies of opioid-involved overdose mortality have employed point-level microdata to describe opioid-involved overdose risk environments, due in part to privacy concerns.13 Address-level data must be sensitively used, especially when representing the lives of those lost to stigmatized diseases; however, they are invaluable for targeting and monitoring interventions at the local scale.

In 2019, the HEALing Communities Study (HCS) launched with support from the National Institutes of Health (NIH) and the Substance Abuse and Mental Health Services Administration (SAMHSA). This multisite, parallel-group, cluster randomized waitlist-controlled trial involves 67 communities in four states (Kentucky, Massachusetts, New York, and Ohio). The goal of HCS is to test the Communities That Heal (CTH) intervention, of which one component is to engage community-based coalitions in data-driven decision making in the selection of evidence-based strategies to reduce opioid overdose deaths.16,17 The first group of active intervention sites were studied between January 2020 and June 2022. In Massachusetts, eight communities received the intervention during this first “wave,” six urban municipalities and two rural “clusters” consisting of two towns each. Working closely with community-based teams and HCS researchers, we gained insight into the most impactful ways of utilizing the strengths of vital statistics data, while remaining respectful of the people who have lost loved ones to opioid-involved overdose.

In the context of HCS, death certificate data serve as the measure for the primary outcome—fatal opioid-involved overdose.18 In Massachusetts, these data are publicly available at the address-level upon request from the Registry of Vital Records and Statistics (RVRS) and are considered non-human subjects data. Although address-level vital statistics data are not available in all states due to differing regulations, Massachusetts is an instructive case study for the use of address-level spatial analysis to monitor opioid-involved overdose mortality and inform selection and targeting of local prevention efforts.

As part of HCS’s data-driven approach, our group developed a comprehensive process to clean, geocode, geomask, map, and spatially analyze decedent data to better understand and respond to the risk landscape in intervention communities. The goals of this study were twofold. The first aim was to characterize the opioid-involved overdose risk landscape within the initial eight HCS intervention communities in Massachusetts by mapping overdose deaths at the address-level. Specifically, we developed a protocol for substituting injury address (a series of fields in the death certificate that identifies location of accidental or intentional injury, in this case, overdose) for recorded deaths at hospitals to identify locations of fatal overdose more accurately. Our second aim was to describe the strengths, challenges, and lessons learned from analyzing vital statistics data and access to treatment to characterize the overdose risk landscape and monitor changes over the course of the study.

Materials and Methods

Data sources.

We obtained address-level data for all opioid-involved overdose decedents whose fatal overdose was recorded in a first wave Massachusetts intervention community (N=8) from January 2017 to December 2020 from the RVRS. These deaths were defined as opioid-related overdose as an underlying cause-of-death (ICD-10 codes X40-X44, X60-X64, X85, Y10-Y14) that involved opioids alone or with other drugs (ICD-10 codes T.40.0-T40.4, or T40.6).18 Variables of interest included addresses of residence, injury, and place of recorded death for all decedents who experienced a fatal opioid-involved overdose. Characteristics of decedents included sex, age, race and ethnicity, veteran status, marital status, educational attainment, and categorical and text fields describing places of injury and death.

We compiled additional spatially-oriented data including: locations of acute care hospitals downloaded from MassGIS19 and addresses of licensed opioid treatment programs (OTPs) provided by the Bureau of Substance Addiction Services (BSAS); and street network data available from MassGIS and in ArcGIS Pro 2.8 (Esri, Redlands, CA).

This study protocol (Pro00038088) was approved by Advarra Inc., the HCS single Institutional Review Board (sIRB).

Data preparation and management.

We first geocoded decedent addresses for place of residence, injury location, and the location where the opioid-involved death was recorded. We achieved geocoding match rates in the range of 90–95%, within acceptable standards.20 In geocoding, we match addresses to a reference database to obtain longitude (X) and latitude (Y) measures that ultimately allow us to place events on a map like pushpins. For individuals who died at home, we confirmed that their death address matched the residence address, and reclassified observations as necessary. We reclassified deaths recorded at hospitals, to the location identified as the injury address, if available. If an individual’s death was recorded at an “other” location, we confirmed that their death address did not match their home address [Figure 1]. We created categorical variables for the text fields associated with the other category, “injury place” and “death place,” based on a methodology described in Siegler, et al. to characterize other locations as residence of other people, hotels, outside, indoor public places, among others.21 We retained hospital deaths that we were not able to recode in order to acknowledge and quantify the impact of opioid-involved overdose on local hospitals, to acknowledge all decedents in these communities, and in the case of descriptive maps, their absence might lead to underestimation of risk.

Figure 1. Data cleaning and recoding workflow.

Figure 1.

Open in a new tab

Prior to geocoding and mapping death certificate data, we used a series of cleaning and recoding steps to: 1) identify cases in which the overdose death had been recorded in a hospital and an injury address had also been recorded and 2) confirm that fatal overdoses recorded at decedent’s homes or other locations were coded correctly by comparing residence and death addresses.

To address concerns about sensitivity of individual-level decedent addresses in maps that would be shared outside of the research team, we developed a geomasking method based on Hampton, et al and Zandbergen.22–24 Briefly, we added random bimodal gaussian perturbation to the addresses within a “donut” with an inner radius no less than 25 meters and an outer radius a minimum of 100 meters from the original location, taking into consideration the underlying population density. We selected these parameters based on the concept of spatial k-anonymity (the probability that the true address could be discovered) and the number of overdoses recorded in the HCS communities.24 After geomasking, we joined individual-level characteristics to the geomasked points using a unique ID.

Descriptive statistics and risk maps.

We compiled descriptive statistics for opioid-involved overdoses for each intervention community [Figure 2A] including: residence, injury, and death addresses; sex; race and ethnicity; marital status; education; categorical place of death; and categorical place of injury. We designed a standardized set of templates for four categories of maps in collaboration with HCS study staff and community partners: decedent race and ethnicity by place of residence, place of fatal overdose of HCS community residents only, place of fatal overdose within HCS communities (residents and non-residents), and decedent race and ethnicity by place of residence juxtaposed with 15-minute walk/drive time buffers to OTPs. The first set depicted the spatial distribution of decedent residences by race and ethnicity in Wave 1 intervention communities. Initial maps focused on four categories used by HCS: non-Hispanic Black, non-Hispanic White, Hispanic, and other race and ethnicity [Figure 2B]. In urban study communities with more diverse populations, we responded to requests from community teams for more specific race and ethnicity categories to better understand the diversity of the Hispanic (e.g., Puerto Rican, Dominican) and non-Hispanic Black decedents (e.g., Cape Verdean, Haitian, African American) in local communities [Figure 2C].

Figure 2. HEALing Communities Wave 1 intervention communities, Massachusetts and opioid-involved overdose decedents in Brockton, Massachusetts, by race and ethnicity, 2017–2020.

Figure 2.

Open in a new tab

A) The first wave of the HEALing Communities Study included eight intervention communities. In Massachusetts, urban intervention communities are identified at the municipality level, rural sites are made up of two or more adjacent or proximate towns clustered for the purpose of the intervention. B) We created maps for each intervention community to visualize the spatial distribution of four categories of race and ethnicity of overdose decedents using geomasked points for residence addresses. C) After a significant increase in overdose fatalities among people of color was observed in Brockton in 2019 and 2020, study staff requested a map that provided more detail about racial and ethnic background, which we mapped using the certified race field in the vital statistics dataset. Note that populations with fewer than 10 overdoses are not depicted in the maps due to spatial anonymity concerns, but the symbology and case counts are included in the legend to illustrate our methods.

We developed a second series of risk maps to depict geomasked addresses symbolized by fatal overdose location (e.g., decedent residence, hospital, and other locations). For the first set in the series, we used a symbology that provides two pieces of information for each point in addition to the approximate, geomasked location. The point is located near the decedent’s residence and the symbol identifies the type of location where their fatal overdose occurred [Figure 3A]. Red houses denote decedents’ residences, squares with a dot in the center denote hospitals, and colored shapes denote other locations like outdoors or indoor public places. We used the same symbology in the second set in this series but placed the points near fatal overdose locations. In early versions of these maps, we used the recorded death address as the point location [Figure 3B]. After recognizing that the injury location, when available, was a more accurate indicator of where the overdose occurred in the case of hospital-recorded fatalities, we began substituting injury locations for deaths recorded at hospitals, when that information was available [Figure 3C], and made the data recoding process described in Figure 1 part of our standard data preparation procedure.

Figure 3. Opioid-involved overdose decedents, Salem, Massachusetts, 2018 – 2020.

Figure 3.

Open in a new tab

A) In this depiction of geomasked decedent residence addresses in Salem, Massachusetts, (N=49), each point provides two pieces of information. The geomasked point represents each decedent’s home address and the symbol identifies the type of place where the person experienced fatal opioid-involved overdose. Maps B) and C) illustrate the value of substituting injury addresses for death addresses when deaths were recorded at hospitals. Map B) depicts death addresses symbolized by location type for all decedents (residents and non-residents) with a Salem death address (N=98). Sixty-six of these deaths were recorded at the local hospital. Map C) depicts fatal overdose locations of decedents with a Salem death address by type, after substituting injury locations when known for deaths recorded at hospitals. These substitutions provide a more accurate picture of where individuals overdosed prior to transport to the hospital, including many whose overdose occurred in adjacent communities.

Spatial analyses.

In addition to developing descriptive maps, we conducted spatial analyses of opioid-involved overdose data using the original point locations and injury addresses for hospital deaths, when known. We calculated Kernel density estimates on the community level to create “heat maps” to visualize density of decedents’ residences and fatal overdose locations per square mile [Figures 4A–4B].25,26 We conducted optimized hotspot cluster analyses and optimized outlier analyses to identify statistically significant clusters of home and death locations in communities where enough points (more than 30) were available [Figures 4C–4E].27

Figure 4. Kernel density estimates, hotspot and outlier analyses for Brockton, Massachusetts, 2018–2020.

Figure 4.

Open in a new tab

Each map includes locations of treatment and harm reduction services. A) Kernel density or heat maps created using residence addresses tended to depict slightly different density patterns than those created using B) fatal overdose locations, the latter often being somewhat skewed by the high number of deaths recorded at hospitals, even once known injury locations are substituted for a portion of the hospital deaths. Conducting an optimized hotspot cluster analysis using the same two datasets, C) we observe a large cluster of statistically significant decedent residence addresses in the center of Brockton, while D) the size of significant hotspots is much smaller for fatal overdose locations, and likely influenced by the hospital in eastern Brockton. The optimized outlier analysis E) was significant only in the case of residence addresses, where we observe that decedent residence locations in the center of Brockton are the most consistent hotspots, with the edges of this region having low counts adjacent to higher counts. This analysis also highlights small areas of high counts within areas of lower overdose decedent residence in the west side of Brockton.

Finally, we created drive- and walk-time maps to depict travel times to OTPs. Although we mapped locations of all treatment and harm reduction assets for intervention communities, our spatial analyses focused on access to OTPs to inform decision-making among HCS community coalitions as they considered needs for and the geographic targeting of this treatment type. OTPs are typically located outside of residential neighborhoods, requiring considerable planning for transportation to these sites, typically on a daily basis. By mapping and highlighting potential placement of new OTPs, and overlaying residence addresses of opioid-involved overdose decedents, we were able to illustrate how the new site could improve access to this important treatment option. We created drive-time maps with 5, 10, and 15-minute buffers and, walk-time maps with, 5, 10, 15, and 20-minute buffers, or displayed buffers depicting 15-minute drive- or walk-times [Figure 5]. We added the decedent home address layer symbolized according to race and ethnicity to the travel time maps to depict the proximities between decedents’ places of residence and the nearest OTP.

Figure 5. Drive and walk times to the nearest opioid treatment program (OTP) juxtaposed decedent residence locations by race and ethnicity, Plymouth and Holyoke, Massachusetts, 2018–2020.

Figure 5.

Open in a new tab

A) In Plymouth, only one OTP was nearby, in a neighboring town at the start of the HCS intervention, which was further than a fifteen-minute drive for potential patients. B) A planned HCS-funded OTP for north Plymouth would improve access within a fifteen-minute drive; however, walk times within fifteen minutes will still be limited. C) Drive time analyses for Holyoke revealed a significant improvement in access to methadone treatment after a new HCS-funded OTP opened in downtown (northeast) Holyoke, where more opioid-involved deaths had occurred in prior years; D) walk time analyses further highlighted improved access for those without cars and the addition of bus stops/routes suggest other convenient modes of access. Note that populations with fewer than 10 overdoses are not depicted in the maps due to spatial anonymity concerns, but the symbology and case counts are included in the legend to illustrate our methods.

We used Stata 16 (College Station, TX) to clean and recode data and calculate descriptive statistics. We used ArcMap 10.8.1 and ArcGIS Pro 2.8 (ESRI, Redlands, CA) to geocode addresses, create risk maps, and conduct spatial and network analyses. We used R (Vienna, Austria) to geomask address points.

Community informed mapping and spatial analytics.

The Massachusetts HCS GIS team met monthly via videoconference with HCS community teams, and the study’s public health and addiction medicine experts to review and discuss maps and spatial analyses. During preparation for the intervention, specifically Phase 1 (Getting Started)28, we focused on creating descriptive maps of decedent residence, death, and injury addresses using historic data from 2015–2017 to establish recent spatial patterns of fatal opioid-involved overdose and to develop a visual language for descriptive maps. As the study progressed and community data managers and engagement staff became acquainted with the available data and how it could be presented, we refined the cartography and variables depicted in the maps.

These meetings helped us verify local findings, revise maps and spatial analyses, and deepen our understanding of each HCS community’s unique risk landscape. In particular, these meetings helped the GIS staff step back from the depiction of points on a map to be mindful of the fact that these points represent individual lives lost and the impact of that loss on family members, friends, and countless others, a point we make before every presentation of our maps. In addition, we developed the geomasking method previously described in response to concerns about protecting the privacy of decedents’ families (all data presented here have been geomasked using this method).

Results

Descriptive statistics and risk maps:

Between 2017–2020, at least 785 individuals died from opioid overdose in the eight Wave 1 intervention communities [Table 1, Figure 2A]. Nearly three-quarters of decedents were male (72.2%) and 71.8% were non-Hispanic White. We observed a decrease in the percent of opioid-involved overdose decedents who were non-Hispanic White from a high in 2017 (76.6%) to a low in 2020 (65.1%) and increases in the percent of decedents who were non-Hispanic Black (from 8.4% in 2017 to 14.9% in 2020) and Hispanic (13.5% in 2017 to 16.9% in 2020). The majority of decedents (63.1%) had never married. The mean age of overdose decedents in Wave 1 communities during the study period was 42.6 years (sd=12.1), with a low of 39 in 2017 and high of 43 in 2020. Based on the original recorded place of death, 34.5 percent of overdoses were recorded at a hospital, 47.6 at home, and 17.6 at other locations. However, after reclassifying deaths recorded at hospitals with recorded injury addresses, we found that 19.7 percent of fatal overdoses were recorded in hospitals, 50.2 percent at home, and 30.1 percent in other locations. Our additional subcategorization of “other” locations revealed that the majority (141/232, 60.8%) of these overdoses occurred at another person’s residence, with outdoors being the second most common (54/232, 23.3%), followed by indoor public places (31/232, 13.4%), and institutional residences (6/232, 2.6%).

Table 1.

Demographic characteristics of opioid-involved overdose decedents in eight Wave 1 intervention communities of the HEALing Communities Study, Massachusetts, 2017–2020 (N=785).

2017 2018 2019 2020 Total
N 214 194 182 195 785
Sex
Male 158 (73.8) 134 (69.1) 133 (73.1) 142 (72.8) 567 (72.2)
Female 56 (26.2) 60 (30.9) 49 (26.9) 53 (27.2) 218 (27.8)
Race and Ethnicity
NH White 164 (76.6) 146 (75.3) 127 (69.8) 127 (65.1) 564 (71.8)
NH Black 18 (8.4) 18 (9.3) 16 (8.8) 29 (14.9) 81 (10.3)
Hispanic 29 (13.5) 25 (12.9) 36 (19.8) 33 (16.9) 123 (15.7)
Other 3 (1.4) 5 (2.3) 3 (1.6) 6 (3.1) 17 (2.2)
Marital status
Never married 147 (68.7) 111 (57.2) 120 (65.9) 117 (60.0) 495 (63.1)
Married 22 (10.3) 39 (20.1) 25 (13.7) 26 (13.3) 112 (14.3)
Separated/Divorced/Widowed 45 (21.0 39 (20.1) 35 (19.2) 47 (24.1) 166 (21.1)
Unknown 0 5 (2.6) 2 (1.1) 5 (2.6) 12 (1.5)
Veteran
Yes 11 (5.1) 16 (8.2) 3 (1.6) 3 (1.5) 33 (4.2)
No/Unknown 203 (94.9) 178 (91.8) 179 (98.3) 192 (98.5) 752 (95.8)
Age
mean (sd) 40.1 (11.7) 43.1 (12.3) 42.3 (11.9) 44.1 (12.5) 42.6 (12.1)
Median 39 42 40.5 43 41
IQR 32, 51 33,51 33,53 35, 52 33,52
Recorded Place of Death
Hospital 87 (40.6) 70 (36.1) 54 (29.7) _ 60 (30.8) 271 (34.5)
Decedenťs residence 87 (40.6) 98 (50.5) 88 (48.3) 101 (51.8) 374 (47.6)
Hospice 0 0 0 1 (0.5) 1 (0.1)
Nursing home 1 (0.5) 0 0 0 1 (0.1)
Asst living/rest home 0 0 0 0 0
Other 39 (18.2) 26 (13.4) 40 (21.9) 33 (16.9) 138 (17.6)
Recorded Injury location
No 119 (55.6) 110 (56.7) 105 (57.7) 108 (55.4) 442 (56.3)
Yes 95 (44.4) 84 (43.3) 77 (42.3) 87 (44.6) 343 (43.7)
Recoded Place of Fatal Overdose
Hospital, no recorded injury location 51 (23.8) 41 (21.1) 31 (17.0) 32 (16.4) 155 (19.7)
Decedent’s residence 89 (65.4) 108 (55.7) 88 (48.4) 109 (56.0) 394 (50.2)
Other location 74 (34.6) 45 (23.2) 63 (34.6) 54 (27.7) 232 (30.1)
Recoded “other” locations of fatal overdose
Other person’s residence1 44 (59.4) 24 (53.3) 43 (71.6) 30 (56.6) 141 (60.8)
Institutional residences (of other people)2 2 (2.7) 2 (4.4) 1 (16) 1 (18) 6 (2.6)
Outside3 18 (24.3) 12 (26.7) 13 (21.7) 11 (20.7) 54 (23.3)
Public indoor4 10 (13.5) 7 (15.5) 3 (5.0) 11 (20.7) 31 (13.4)
Open in a new tab

NH = Non-Hispanic; sd = standard deviation; IQR = Interquartile Range

1

Other person’s residence: Independent apartments, houses, public housing.

2

Institutional residences (of other people): Homeless shelters, single-room occupancies (SROs), supportive housing, nursing homes, assisted living, drug treatment facilities, and sober homes.

3

Public indoor: Bars, restaurants, hotels, public bathrooms, offices, building lobbies, elevators, and stairways.

4

Outside: Parks, streets, cars.

Maps of decedent race and ethnicity clearly illustrate that in most intervention communities, decedents were predominantly non-Hispanic White. More racially and ethnically diverse communities displayed greater variation in race and ethnicity among those who died from opioid overdose [Figure 2B]. Upon request from community-engaged staff, we created additional maps that further categorized non-Hispanic Black and Hispanic decedents into more detailed groups reflecting ethnicity or country of origin. For example, in Brockton, study staff observed a significant increase in overdoses among the Black and Hispanic communities, which led to mapping the community based not only on ethnicity but nativity, to identify patterns among immigrant populations [Figure 2C]. Although we found that Brockton’s overdose decedents were a racially and ethnically heterogenous and dispersed population, over the course of the study period we observed an increase in overdose among Cape Verdean immigrants in 2020.

Using the symbology we developed and described above, we created three variations of the maps visualizing the location of fatal opioid-involved overdose before settling on a final approach. The first located the geomasked point representing decedents’ home addresses with symbols that identified the type of place where overdoses occurred (home, hospital, other location) [Figure 3A]. This approach allowed us to identify spatial patterns that might be present with regard to the geographic location of a decedent’s home and the type of place where they experienced their overdose. Although we did not observe any significant patterns, these maps did highlight cases in which deaths within one community were recorded at multiple hospitals, as in Salem, where deaths were recorded at the local North Shore Medical Center and as far away as hospitals in Boston. The second iteration of this map depicted the point near the recorded place of death and used the same set of symbols [Figure 3B]. Across all intervention communities, we observed that the majority of overdoses occurred at home and that using the hospital address as the place of fatal overdose obscured the true spatial distribution of overdose locations. We sought to rectify this issue in the third variation of these maps, which used the reclassified place of death categories based on replacing injury locations when available for hospital deaths [Figure 3C]. In some intervention communities, the difference between versions two and three were small, while in others, we observed patterns that significantly enhanced our understanding of where overdoses occurred. In Salem, for example, we found that a number of fatal overdoses that were initially recorded at Salem at North Shore Medical Center actually occurred in adjacent communities at other people’s homes. The third iteration became our standard approach from that point on.

Spatial analyses.

Kernel density estimates (heat maps) highlighted variation in densities of decedents place of residence compared to the location where the opioid-involved overdose took place. For example, in Brockton, we observed a larger geographic area with a high density of fatal overdoses linked to residence addresses compared to that of the heatmap for injury locations [Figure 4A–B]. The fatal overdose location density map does use injury locations when available for deaths recorded in hospitals, yet still indicates visual bias when hospitals are used as the fatal overdose location. Because so many deaths were recorded there, hospitals appear as high-density locations.

We used optimized hotspot cluster analyses to identify areas that not only had a high rates of decedent residences or fatal overdose locations, but that also had clusters of higher than average counts that were statistically significant [Figure 4C–D]. The analyses of residence addresses identified two areas of interest: a large hotspot in central Brockton and a coldspot, where decedent residence counts were lower than average, in the western part of the city. However, when we conducted the same analysis using the overdose locations, the hotspots were observed in two areas east of downtown Brockton near the OTP and hospital. An optimized outlier analysis further enhanced understanding of the hot and coldspots associated with decedent residences [Figure 4E].27 Based on these results, we might conclude that decedent residence locations in the center of Brockton are the most consistent hotspots, while areas at the edge of this cluster have significantly lower counts. Further investigation of the outliers might lead to helpful intervention approaches. Collectively, this series of spatial analyses provided a data-driven approach to identify areas within Brockton where statistically significant cluster of decedents lived and experienced fatal overdoses in order to help target placement of appropriate resources.

To further illustrate gaps between overdose burden and access to services, we used network analysis tools to depict drive- and walk-times to OTPs [Figure 5]. In some communities, we found very stark deficits in access to OTPs. For example, in Plymouth, although a considerable number of opioid-involved overdose decedents live in the north side of the city, the nearest OTP was more than a 15-minute drive [Figure 5A]. Community awareness of this gap led to the opening of a new OTP in this section of Plymouth, which has the potential to have a strong impact on improving access to treatment for people who use opioids who live nearby [Figure 5B]. We observed a similar spatial inequity in Holyoke, where the nearest OTP was located in the southeast corner of the city, far from the downtown area in the northeast where the majority of decedents lived and overdose fatalities occurred. As in Plymouth, community partners recognized this gap and opened a new OTP in the downtown area and our maps confirmed the greatly improved access whether by car or on foot [Figure 5C–D].

Discussion

We used address-level death certificate data to depict the spatial distribution of opioid-involved overdose fatalities by place of death and race and ethnicity, as well as significant overdose clusters in Massachusetts HCS Wave 1 communities between 2017 and 2020. Address-level analyses highlighted the fact that the majority of these overdoses occurred in decedents’ homes compared to all other location types. We also found that injury data, when available, provided valuable information about the actual location of fatal overdoses, despite the fact that many of these opioid-involved deaths were recorded at a hospital. Finally, our maps of drive- and walk-time buffers juxtaposed with the locations of decedent home addresses highlighted significant gaps in services (i.e., treatment deserts), which have since been addressed as part of HCS interventions.

Although prior research on increasing opioid overdose deaths has employed information from address-level decedent data focused on place of death (at or away from home) and demographic characteristics (e.g., race and ethnicity), to our knowledge, few have examined these fatalities descriptively at the address level.29 Mack et. al. used place of death information from vital statistics in their investigation of illicit drug use, which was primarily focused on comparing metropolitan and nonmetropolitan areas, and like our study, found that death at home was most common.30 Siegler et. al.’s analysis of place-based analysis of opioid-involved overdose deaths in New York City between 2005–2010 identified demographic and geographic factors associated with overdose at or away from home.21 Similarly, Somerville et. al. found in their use of data extracted from Medical Examiner charts that the majority of overdose deaths in three southeastern Massachusetts counties occurred at home or another private residence. Our work builds upon these studies, which were based on aggregated data, to provide nuanced, address-level, local perspectives through GIS mapping and spatial analysis.

A recent paper focusing on the “journey to overdose” used residence and death address information from the Medical Examiner to conduct spatial social network analysis to investigate geographically discordant overdose deaths, in which the decedent’s residence address differed from their death address.31 The authors found that tracts on the overdose end of the journey were more likely to be impoverished, have lower educational attainment, higher housing insecurity, and part of Hispanic communities, compared to census tracts of decedents’ home addresses. This application of point-level data suggests one way that our future analyses might continue to evolve.

We also learned that some of the power of individual-level decedent data lies in the injury address and injury place text fields, which have been only recently addressed in published studies on opioid overdose deaths.29,32 In cases when the recorded place of death was a hospital, injury information provided the location of the actual opioid-involved overdose. A recent paper by Banks and colleagues similarly used injury information to identify fatal overdose locations, which may have helped to more accurately detect a statistically significant difference in location of death for Black decedents in the post-fentanyl period in St. Louis.32 Although people who have experienced overdose may ultimately die in the hospital after transport, it is unlikely that the overdose occurred there. Access to this additional location data allows for a more accurate representation of the risk environment for opioid-involved overdose, resulting in more effective placement of overdose prevention and OUD treatment intervention resources, such as overdose emergency boxes with naloxone and more effective placement of OUD intervention and treatment resources.

Mapping the race and ethnicity of overdose decedents provided a valuable visual summary of the impact of opioid overdose deaths on communities of color. Our observations of the increase in opioid-involved mortality among non-Hispanic and Hispanic people is consistent with changes in overdose rates among these populations reported by the State of Massachusetts and in peer-reviewed literature.1,6,32–35 Although cross-sectional studies do not allow us to make definitive statements about the drivers of these changes, recent studies have suggested that opioid-involved mortality increases among communities of color may be related to the more recent proliferation of fentanyl in the illicit stimulant market.2,32,36 In recent reporting on opioid-involved overdose mortality rates, the Massachusetts Department of Public Health (MDPH), noted that the presence of cocaine and amphetamines have been increasing steadily in the toxicology screens of opioid-involved OD decedents since 2014 and 2016, respectively.36 Non-Hispanic Black people were also less likely to be prescribed opioid analgesics during the early wave of the opioid overdose epidemic due to structural racism; as the epidemic shifted to fentanyl, communities of color have been at greater risk for opioid-involved overdose.37 Visual representations of overdose patterns among people of color within the context of their community’s geography allow local coalitions and interventionists to see at a granular level where vulnerable populations live relative to local resources or high-risk locations.

The drive- and walk-time maps have proven to be especially powerful tools in helping highlight gaps in services. Although coalitions in Plymouth and Holyoke were aware of gaps in access to OTPs in their communities, our maps presented visually persuasive evidence that the locations they selected for new services would substantially close those gaps. By adding the layer displaying decedents’ race and ethnicity, we provided information that may promote culturally sensitive communication, outreach and services. Regional bus information in the maps further highlights an important mode of transport that may be overlooked. We added bus stop and route information to the walk-time maps, which was especially valuable in Holyoke, a smaller city in Western Massachusetts well outside of the urban transit hub of the nearby metropolis of Springfield. While this group of maps focuses on OTP locations, our group has conducted other analyses focusing on access to buprenorphine.38 Although other spatial analysis methods, such as Enhanced 2-Stage Floating Catchment Area (E2SFCA) can address treatment accessibility using both supply and demand variables,39,40 we found that as a preliminary step, community coalitions found drive- and walk-time maps valuable and more intuitive than E2SFCA maps we shared. No one map is able to tell every story. A suite of maps employing a variety of descriptive and analytical approaches is ideal; our experience was that some types were more illustrative of the local opioid-involved overdose landscape in some communities than others. The methods we describe here represent the most effective examples of our iterative process, which developed over the course of the study’s first wave and was fully employed in the second wave, which started in July 2022.

Despite their power to more fully represent the population of opioid-involved overdose decedents, vital statistics have limitations. The most notable limitations are that they describe only fatal overdoses, while non-fatal overdose data can provide actionable data about individuals who are still living, and data lags of at least six months prevent access to the most timely data.41 Vital statistics are also subject to a variety of biases related to toxicology, intent, and misclassification of race and ethnicity.9,42 Finally, injury information in our datasets was not complete (approximately 43% of decedents had injury information recorded) and we discerned no pattern to explain its presence or absence.29 Despite these limitations, vital statistics data, especially when available at the address-level, offer the opportunity to depict the overdose risk environment with relative precision. These data are also among the few sources that include information about race and ethnicity at the individual level, thus providing the potential to study opioid-involved overdose outcomes with a health equity lens.

Conclusion

This study describes a methodology that yielded keen insights into the use of vital statistics data to better understand and respond to the opioid-involved overdose deaths in communities. Our approach provides a blueprint for future work with spatially-oriented decedent data and a guide to presenting these data on the address-level to effectively mitigate the risks in local environments.

Highlights.

  • Sensitive use of address-level data aids hyper-local understanding of opioid OD.

  • Injury data, when available, clarifies locations of ODs recorded at hospitals.

  • Mapping decedent home addresses and drive/walk times highlights treatment deserts.

Acknowledgements:

This study protocol (Pro00038088) was approved by Advarra Inc., the HEALing Communities Study single Institutional Review Board. This research was supported by the National Institutes of Health and the Substance Abuse and Mental Health Services Administration through the NIH HEAL (Helping to End Addiction Long-termSM) Initiative under award number UM1DA049412 (ClinicalTrials.gov Identifier: NCT04111939). We wish to acknowledge the participation of the HEALing Communities Study communities, community coalitions, and Community Advisory Boards and state government officials who partnered with us on this study. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the Substance Abuse and Mental Health Services Administration or the NIH HEAL InitiativeSM

Role of Funding Source

This research was supported by the National Institutes of Health and the Substance Abuse and Mental Health Services Administration through the NIH HEAL (Helping to End Addiction Long-termSM) Initiative under award number UM1DA049412 (ClinicalTrials.gov Identifier: NCT04111939). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the Substance Abuse and Mental Health Services Administration or the NIH HEAL InitiativeSM

Footnotes

Declaration of Competing Interest

The authors have no conflicts of interest to disclose.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

REFERNCES

  • 1.Kariisa M, Davis N, Kumar S, et al. Vital Signs: Drug Overdose Deaths, by Selected Sociodemographic and Social Determinants of Health Characteristics — 25 States and the District of Columbia, 2019–2020. MMWR Morb Mortal Wkly Rep. 2022;71. doi: 10.15585/mmwr.mm7129e2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Jenkins RA. The fourth wave of the US opioid epidemic and its implications for the rural US: A federal perspective. Prev Med. 2021;152:106541. doi: 10.1016/j.ypmed.2021.106541 [DOI] [PubMed] [Google Scholar]
  • 3.Kiang MV, Basu S, Chen J, Alexander MJ. Assessment of Changes in the Geographical Distribution of Opioid-Related Mortality Across the United States by Opioid Type, 1999–2016. JAMA Netw Open. 2019;2(2):e190040. doi: 10.1001/jamanetworkopen.2019.0040 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Wilt GE, Lewis BE, Adams EE. A Spatial Exploration of Changes in Drug Overdose Mortality in the United States, 2000–2016. Prev Chronic Dis. 2019;16:180405. doi: 10.5888/pcd16.180405 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Massachusetts Department of Public Health. Data Brief: Opioid-Related Overdose Deaths among Massachusetts Residents.; 2023. Accessed July 3, 2023. https://www.mass.gov/doc/opioid-related-overdose-deaths-among-ma-residents-june-2023/download
  • 6.Larochelle MR, Slavova S, Root ED, et al. Disparities in Opioid Overdose Death Trends by Race/Ethnicity, 2018–2019, From the HEALing Communities Study. Am J Public Health. Published online September 9, 2021:e1–e4. doi: 10.2105/AJPH.2021.306431 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Massachusetts Department of Public Health. Opioid-Related Overdose Deaths, All Intents, MA Residents – Demographic Data Highlights.; 2023. Accessed July 3, 2023. https://www.mass.gov/doc/opioid-related-overdose-deaths-demographics-june-2023/download
  • 8.Chandler RK, Villani J, Clarke T, McCance-Katz EF, Volkow ND. Addressing opioid overdose deaths: The vision for the HEALing communities study. Drug Alcohol Depend. 2020;217:108329. doi: 10.1016/j.drugalcdep.2020.108329 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Dasgupta N, Funk MJ, Brownstein JS. Comparing Unintentional Opioid Poisoning Mortality in Metropolitan and Non-Metropolitan Counties, United States, 1999–2003. In: Thomas YF, Richardson D, Cheung I, eds. Geography and Drug Addiction. Springer; Netherlands; 2008:175–192. doi: 10.1007/978-1-4020-8509-3_11 [DOI] [Google Scholar]
  • 10.Jalal H, Buchanich JM, Roberts MS, Balmert LC, Zhang K, Burke DS. Changing dynamics of the drug overdose epidemic in the United States from 1979 through 2016. Science. 2018;361(6408):eaau1184. doi: 10.1126/science.aau1184 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Delcher C, Anthony N, Mir M. Xylazine-involved fatal overdoses and localized geographic clustering: Cook County, IL, 2019–2022. Drug Alcohol Depend. Published online June 16, 2023:110833. doi: 10.1016/j.drugalcdep.2023.110833 [DOI] [PubMed] [Google Scholar]
  • 12.Marotta PL, Hunt T, Gilbert L, Wu E, Goddard-Eckrich D, El-Bassel N. Assessing Spatial Relationships between Prescription Drugs, Race, and Overdose in New York State from 2013 to 2015. J Psychoactive Drugs. 2019;51(4):360–370. doi: 10.1080/02791072.2019.1599472 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Sauer J, Stewart K. Geographic information science and the United States opioid overdose crisis: A scoping review of methods, scales, and application areas. Soc Sci Med. 2023;317:115525. doi: 10.1016/j.socscimed.2022.115525 [DOI] [PubMed] [Google Scholar]
  • 14.Nesoff ED, Branas CC, Martins SS. The Geographic Distribution of Fentanyl-Involved Overdose Deaths in Cook County, Illinois. Am J Public Health. 2020;110(1):98–105. doi: 10.2105/AJPH.2019.305368 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Stopka TJ, Amaravadi H, Kaplan AR, et al. Opioid overdose deaths and potentially inappropriate opioid prescribing practices (PIP): A spatial epidemiological study. Int J Drug Policy. 2019;68:37–45. doi: 10.1016/j.drugpo.2019.03.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Young AM, Brown JL, Hunt T, et al. Protocol for community-driven selection of strategies to implement evidence-based practices to reduce opioid overdoses in the HEALing Communities Study: a trial to evaluate a community-engaged intervention in Kentucky, Massachusetts, New York and Ohio. BMJ Open. 2022;12(9):e059328. doi: 10.1136/bmjopen-2021-059328 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Winhusen T, Walley A, Fanucchi LC, et al. The Opioid-overdose Reduction Continuum of Care Approach (ORCCA): Evidence-based practices in the HEALing Communities Study. Drug Alcohol Depend. 2020;217:108325. doi: 10.1016/j.drugalcdep.2020.108325 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Slavova S, LaRochelle MR, Root ED, et al. Operationalizing and selecting outcome measures for the HEALing Communities Study. Drug Alcohol Depend. 2020;217:108328. doi: 10.1016/j.drugalcdep.2020.108328 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Massachusetts Bureau of Geographic Information. MassGIS. MassGIS. Accessed November 30, 2020. https://www.mass.gov/orgs/massgis-bureau-of-geographic-information
  • 20.Zandbergen PA. Geocoding Quality and Implications for Spatial Analysis. Geogr Compass. 2009;3(2):647–680. doi: 10.1111/j.1749-8198.2008.00205.x [DOI] [Google Scholar]
  • 21.Siegler A, Tuazon E, Bradley O’Brien D, Paone D. Unintentional opioid overdose deaths in New York City, 2005–2010: A place-based approach to reduce risk. Int J Drug Policy. 2014;25(3):569–574. doi: 10.1016/j.drugpo.2013.10.015 [DOI] [PubMed] [Google Scholar]
  • 22.Allshouse WB, Fitch MK, Hampton KH, et al. Geomasking sensitive health data and privacy protection: an evaluation using an E911 database. Geocarto Int. 2010;25(6):443–452. doi: 10.1080/10106049.2010.496496 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Hampton KH, Fitch MK, Allshouse WB, et al. Mapping Health Data: Improved Privacy Protection With Donut Method Geomasking. Am J Epidemiol. 2010;172(9):1062–1069. doi: 10.1093/aje/kwq248 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Zandbergen PA. Ensuring Confidentiality of Geocoded Health Data: Assessing Geographic Masking Strategies for Individual-Level Data. Adv Med. 2014;2014. doi: 10.1155/2014/567049 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.ESRI. How Kernel Density Works. Accessed July 27, 2023. https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/how-kernel-density-works.htm [Google Scholar]
  • 26.De Smith MJ, Goodchild MF, Longley P. Geospatial Analysis: A Comprehensive Guide to Principles, Techniques and Software Tools.; 2015. https://spatialanalysisonline.com/HTML/index.html [Google Scholar]
  • 27.ESRI. Optimized Outlier Analysis (Spatial Statistics)—ArcGIS Pro | Documentation. Accessed July 3, 2023. https://pro.arcgis.com/en/pro-app/2.9/tool-reference/spatial-statistics/optimizedoutlieranalysis.htm [Google Scholar]
  • 28.Sprague Martinez L, Rapkin BD, Young A, et al. Community engagement to implement evidence-based practices in the HEALing communities study. Drug Alcohol Depend. 2020;217:108326. doi: 10.1016/j.drugalcdep.2020.108326 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Pustz J, Srinivasan S, Larochelle MR, Walley AY, Stopka TJ. Relationships between places of residence, injury, and death: Spatial and statistical analysis of fatal opioid overdoses across Massachusetts. Spat Spatio-Temporal Epidemiol Published online October 10, 2022:100541. doi: 10.1016/j.sste.2022.100541 [DOI] [PubMed] [Google Scholar]
  • 30.Mack KA . Illicit Drug Use, Illicit Drug Use Disorders, and Drug Overdose Deaths in Metropolitan and Nonmetropolitan Areas — United States. MMWR Surveill Summ. 2017;66. doi: 10.15585/mmwr.ss6619a1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Forati A, Ghose R, Mohebbi F, Mantsch JR. The journey to overdose: Using spatial social network analysis as a novel framework to study geographic discordance in overdose deaths. Drug Alcohol Depend. 2023;245:109827. doi: 10.1016/j.drugalcdep.2023.109827 [DOI] [PubMed] [Google Scholar]
  • 32.Banks DE, Scroggins S, Paschke ME, et al. Examining Increasing Racial Inequities in Opioid Overdose Deaths: a Spatiotemporal Analysis of Black and White Decedents in St. Louis, Missouri, 2011–2021. J Urban Health. Published online May 23, 2023. doi: 10.1007/s11524-023-00736-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Massachusetts Department of Public Health. Opioid-Related Overdose Deaths, All Intents, MA Residents--Demographic Data Highlights.; 2021. https://www.mass.gov/doc/opioid-related-overdose-deaths-demographics-november-2021/download
  • 34.Abdalla SM, Galea S. Invited Commentary: Toward a Better Understanding of Disparities in Overdose Mortality. Am J Epidemiol. 2022;191(7). doi: 10.1093/aje/kwac053 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Black JC, Rockhill KM, Dart RC, Iwanicki J. Clustering patterns in polysubstance mortality in the United States in 2017: a multiple correspondence analysis of death certificate data. Ann Epidemiol. Published online April 2, 2022. doi: 10.1016/j.annepidem.2022.03.011 [DOI] [PubMed] [Google Scholar]
  • 36.Stopka TJ, Larochelle MR, Li X, et al. Opioid-related mortality: Dynamic temporal and spatial trends by drug type and demographic subpopulations, Massachusetts, 2005–2021. Drug Alcohol Depend. 2023;246:109836. doi: 10.1016/j.drugalcdep.2023.109836 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Hoopsick RA, Homish GG, Leonard KE. Differences in Opioid Overdose Mortality Rates Among Middle-Aged Adults by Race/Ethnicity and Sex, 1999–2018. Public Health Rep. 2021;136(2):192–200. doi:DOI: 10.1177/0033354920968806 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Shrestha S, Lindstrom MR, Harris D, et al. Spatial access to buprenorphine-waivered prescribers in the HEALing communities study: Enhanced 2-step floating catchment area analyses in Massachusetts, Ohio, and Kentucky. J Subst Use Addict Treat. 2023;150:209077. doi: 10.1016/j.josat.2023.209077 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Cao Y, Stewart K, Wish E, Artigiani E, Sorg MH. Determining spatial access to opioid use disorder treatment and emergency medical services in New Hampshire. J Subst Abuse Treat. 2019;101:55–66. doi: 10.1016/j.jsat.2019.03.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Forati AM, Ghose R, Mantsch JR. Examining Opioid Overdose Deaths across Communities Defined by Racial Composition: a Multiscale Geographically Weighted Regression Approach. J Urban Health. Published online July 6, 2021. doi: 10.1007/s11524-021-00554-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Warner M, Hedegaard H. Identifying Opioid Overdose Deaths Using Vital Statistics Data. Am J Public Health. 2018;108(12):1587–1589. doi: 10.2105/AJPH.2018.304781 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Zalla LC, Martin CL, Edwards JK, Gartner DR, Noppert GA. A Geography of Risk: Structural Racism and Coronavirus Disease 2019 Mortality in the United States. Am J Epidemiol. 2021;190(8):1439–1446. doi: 10.1093/aje/kwab059 [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES