Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Nov 28.
Published in final edited form as: Int J Drug Policy. 2024 Jan 20;125:104322. doi: 10.1016/j.drugpo.2024.104322

Characterizing opioid overdose hotspots for place-based overdose prevention and treatment interventions: A geo-spatial analysis of Rhode Island, USA

Elizabeth A Samuels a,b,c,d,*, William C Goedel c, Victoria Jent e, Lauren Conkey d, Benjamin D Hallowell d, Sarah Karim d, Jennifer Koziol d, Sara Becker f, Rachel R Yorlets c,h, Roland Merchant g, Lee Ann Jordison Keeler b, Neha Reddy i, James McDonald d, Nicole Alexander-Scott d, Magdalena Cerda e, Brandon DL Marshall c
PMCID: PMC12658596  NIHMSID: NIHMS2124869  PMID: 38245914

Abstract

Objective:

Examine differences in neighborhood characteristics and services between overdose hotspot and non-hotspot neighborhoods and identify neighborhood-level population factors associated with increased overdose incidence.

Methods:

We conducted a population-based retrospective analysis of Rhode Island, USA residents who had a fatal or non-fatal overdose from 2016 to 2020 using an environmental scan and data from Rhode Island emergency medical services, State Unintentional Drug Overdose Reporting System, and the American Community Survey. We conducted a spatial scan via SaTScan to identify non-fatal and fatal overdose hotspots and compared the characteristics of hotspot and non-hotspot neighborhoods. We identified associations between census block group-level characteristics using a Besag-York-Mollié model specification with a conditional autoregressive spatial random effect.

Results:

We identified 7 non-fatal and 3 fatal overdose hotspots in Rhode Island during the study period. Hotspot neighborhoods had higher proportions of Black and Latino/a residents, renter-occupied housing, vacant housing, unemployment, and cost-burdened households. A higher proportion of hotspot neighborhoods had a religious organization, a health center, or a police station. Non-fatal overdose risk increased in a dose responsive manner with increasing proportions of residents living in poverty. There was increased relative risk of non-fatal and fatal overdoses in neighborhoods with crowded housing above the mean (RR 1.19 [95 % CI 1.05, 1.34]; RR 1.21 [95 % CI 1.18, 1.38], respectively).

Conclusion:

Neighborhoods with increased prevalence of housing instability and poverty are at highest risk of overdose. The high availability of social services in overdose hotspots presents an opportunity to work with established organizations to prevent overdose deaths.

Keywords: Overdose surveillance, Geospatial analysis, Social determinants of health

Introduction

Overdose is an ongoing public health crisis in the United States (US) (Centers for Disease Control & Prevention, 2023; Ahmad, Cisewski, Rossen, & Sutton, 2023; Kaiser Family Foundation, 2022; Rudd et al., 2016), and mortality is disproportionately increasing among people who are Black, Latino/a, or Indigenous (Friedman & Hansen, 2022), as well as among people who are unhoused (Los Angeles County Department of Public Health, Center for Health Impact Evaluation, 2021; Cawley et al., 2022; Fine et al., 2022; Baggett et al., 2013; Barocas et al., 2019). Overdose has become a leading cause of death for people under 50 years of age (Rudd et al., 2016) and a major contributor to declining life expectancy (Centers for Disease Control & Prevention, 2023).

Individual-level risk factors associated with overdose are well-documented and include use of substances without a bystander capable of providing overdose response, recurrent use after loss of tolerance, mixing of opioids with other sedating substances (e.g., benzodiazepines, alcohol), and exposure to increasingly potent opioids (specifically fentanyl analogues) and other additives (e.g., xylazine) (Olfson et al., 2018; Olfson et al., 2018).

Differences in overdose incidence over space and time, neighborhood- and population-level overdose risk factors, and the availability of community resources in areas with increased overdose incidence are less well understood. Prior research has also shown significant differences in location of a person’s residence and the location of fatal (Nesoff et al., 2020) or non-fatal overdose (Pustz et al., 2023). Previously identified neighborhood characteristics associated with increased overdose incidence include substance and opioid use disorder treatment accessibility (Bozorgi et al., 2021), presence of tobacco and liquor outlets (Bozorgi et al., 2021), low neighborhood-level annual income (Bozorgi et al., 2021; Schell et al., 2022), and low neighborhood-level educational attainment (Bozorgi et al., 2021; Schell et al., 2022). Residential instability (Schell et al., 2022), income inequality (Nandi et al., 2006), and qualities of the built environment (i.e., lack of green space, building vacancies) (Nandi et al., 2006) have also been identified as neighborhood-level factors associated with fatal overdoses.

Geographic Information Systems (GIS) analysis, or geospatial analysis, is used by public health practitioners to identify geographic hotspots of disease and detect changes in population health over space and time (Clarke, 1996; Miranda et al., 2013). Identification of overdose incidence changes over space and time and examination of neighborhood-level factors associated with overdose can be used for intervention development and place-based resource deployment to prevent overdose deaths. Prior studies have characterized overdose hotspots generally, however the differences between hotspot and non-hotspot neighborhoods, and availability of services in these areas for overdose prevention, are poorly understood.

Rhode Island has one of the highest rates of overdose mortality in the US. In 2020, the rate of opioid overdose death was 38.2 per 100,000 person years, the 11th highest opioid overdose death rate nationally (Kaiser Family Foundation, 2022; Rhode Island Department of Health, 2023). In this study, we sought to identify neighborhood-level factors associated with overdose events and examine differences in neighborhood characteristics and services availability between hotspot and non-hotspot neighborhoods in Rhode Island.

Methods

Study setting and design

Rhode Island is a state in the northeastern US, a region with one of the highest rates of US overdose deaths. We conducted a population-based, retrospective, cross-sectional study of Rhode Island residents who had a fatal or non-fatal overdose in Rhode Island between 2016 and 2020 to characterize overdose hotspots, examine differences in neighborhood characteristics between hotspot and non-hotspot neighborhoods, and identify neighborhood-level characteristics associated with increased overdose incidence. This study was conducted in accordance with STROBE guidelines for cross-sectional studies (von Elm et al., 2007).

This study was conducted in conjunction with the Rhode Island Department of Health (RIDOH)’s Community Overdose Engagement (CODE) program. CODE provided overdose surveillance data and support to high burden municipalities to design and implement interventions to reduce overdose death (Lasher et al., 2021). CODE was implemented from 2017 to 2023 in 3 phases. In the first two phases (2017 and 2018), all Rhode Island municipalities were encouraged to apply for funding to create and then implement an overdose prevention and response plan. In the third phase (2020–2023), RIDOH funded three municipalities with the highest burden of overdose. These communities convened multisector collaboratives and conducted neighborhood needs assessments to identify priority interventions, such as targeted harm reduction supply distribution, street and business outreach, provision of housing supports, and embedding peer recovery supports into emergency departments.

Data sources

Overdose incidence (primary study outcome) included location of nonfatal and fatal overdoses aggregated to the census block group. Non-fatal overdoses were identified through a query of the RIDOH emergency medical services (EMS) database, which includes state-wide reports from all ambulance services. EMS report their data to a central RIDOH EMS office using the ImageTrend electronic health record (EHR) platform. Data reported includes information about EMS response location (street address), visit date and time, patient name, date of birth, sex, and documentation of naloxone administration by a bystander and/or first responder. EMS runs were included in this analysis if they met RIDOH’s definition for a suspected non-fatal opioid overdose. According to this definition, an EMS run is considered to be opioid overdose-related if it meets one of the following: (1) Primary or secondary impression is overdose related AND naloxone was administered; (2) primary or secondary impression is overdose related AND terms for both naloxone AND unresponsive are in the EMS narrative; (3) Naloxone was administered AND medication response was marked as improved; (4) Terms for both naloxone and unresponsive are in the narrative AND medication response is undocumented; or (5) Naloxone was given prior to EMS arrival AND it is noted who administered it. EMS runs are excluded if it was “an interfacility transfer, the patient is determined to be biologically dead upon arrival, or the patients age 10 and younger.” (Hallowell et al., 2021; Lasher et al., 2019; Rhode Island Department of Health, 2022)

Fatal overdoses in Rhode Island were identified using the Center for Disease Control and Prevention’s State Unintentional Drug Overdose Reporting System (SUDORS) (Rhode Island Department of Health, 2023; Centers for Disease Control & Prevention, 2023), for which professional data abstractors review data from death certificates, medical examiner reports, postmortem toxicology, police reports, and hospital records to compile comprehensive data about unintentional overdose deaths including decedent age, sex, city/town of residence, race, ethnicity, location of death (street address), toxicology, and circumstances surrounding death. Overdose deaths included in SUDORS are events where: (1) the manner of death was deemed accidental or undetermined; (2) the overdose death occurred in a Rhode Island municipality; and (3) a drug is listed on the death certificate as the primary cause of death or a significant contributing factor (Centers for Disease Control & Prevention, 2023; Marshall et al., 2022).

Neighborhood characteristics were assessed using the American Community Survey (ACS) 5-year estimates 2010–2019 (United States Census Bureau, 2023) in combination an environmental scan of neighborhood resources. The ACS is an annual survey which includes census-block level data about educational attainment, employment, food security, housing, income, and transportation. ACS data included in the analysis was census block-level data about sex, race, and ethnicity of census block residents, rates of high school graduation, poverty, unemployment, receipt of public assistance, proportion of cost burdened households, housing vacancy, proportion rental units, and proportion of households with a car.

To identify neighborhood services, organizations, and characteristics to include in the environmental scan, the study principal investigator met with RIDOH CODE grant recipients. Services included in the environmental scan are listed in Supplemental Table 1. Locations of addiction treatment providers and hospitals were collected from a registry maintained by the Department of Behavioral Healthcare, Developmental Disabilities, and Hospitals (BHDDH) (Rhode Island Department of Behavioral Healthcare, Developmental Disabilities, & Hospitals, 2023). Locations of addiction treatment clinicians and programs included licensed Opioid Treatment Programs as well as office-based buprenorphine providers listed in the Substance Abuse and Mental Health Administrations (SAMHSA) buprenorphine provider database (Rhode Island Department of Behavioral Healthcare, Developmental Disabilities, & Hospitals, 2023; Substance Abuse & Mental Health Services Administration, 2023). Locations of liquor and alcohol retailers were collected from publicly available state liquor license database (Rhode Island Department of Business Regulation, 2023). Locations of housing services, police and fire, pharmacies, libraries, public transportation, religious organizations, and social services were assessed through the Rhode Island GIS Geospatial Data Hub of the University of Rhode Island (Rhode Island Geographic Information Systems, 2023), which is a publicly available database of Rhode Island GIS maps and data. Research assistants identified additional neighborhood services not previously mapped through a detailed and systematic review of local harm reduction, recovery, and social services resource guides and review of Google Maps in overdose hotspots. Information collected in the environmental scan was used to provide information to hotspot stakeholders, but was not used in comparative geospatial analyses since it was not conducted in non-hotspot neighborhoods. This study was reviewed and approved by expedited review by Institutional Review Boards at Lifespan (IRB #s 1422855–8 and 1422855–9) and RIDOH (IRB# 2021–06).

Analysis

Identifying hotspots

Hotspots are broadly defined as geographic areas in which the occurrence of disease is disproportionately higher than would be expected by chance alone (Kulldorff, 1997). We used RIDOH EMS and SUDORS data to identify hotspots for non-fatal and fatal overdose hotspots, respectively. We used a spatial scan conducted via SaTScan to identify hotspots of overdose events, taking population size into account in Rhode Island census block groups (Greene et al., 2016; Sherman et al., 2014; Stewart et al., 2017; Goedel et al., 2019). Because some individuals may have multiple overdoses for which EMS is called during the four-year study period, the unit of analysis for non-fatal overdose was EMS calls for overdose events, not person-level overdose incidence. For this analysis, we consider the cumulative burden of non-fatal or fatal overdose over time. As such, we calculated annualized cumulative incidence rates for use in SaTScan, with population sizes derived from the five-year estimates from the American Community Survey (United States Census Bureau, 2023). Using a series of circular windows inclusive of varying combinations of adjacent census block groups that included no more than 50 % of the state’s population, we compare the observed number of overdose events across census block groups with distributions randomly generated from a Poisson distribution to identify hotspots, or areas where the number of EMS runs for opioid-involved overdose or the number of opioid-involved overdose deaths is higher than would otherwise be expected by chance (Kulldorff et al., 2005). A census block group was considered to be included in a hotspot if its centroid fell within the circular window. After hotspots were identified, we conducted descriptive analyses summarizing overdose frequency, year to year changes, and decedent demographics.

Differences between overdose hotspot and non-hotspot neighborhoods

We compared differences in neighborhood social determinants of health and services availability between hotspot and non-hotspot neighborhoods. Patterns of service availability within overdose hotspot clusters were determined by examining locations of treatment, harm reduction, medical, and social services within each hotspot. Neighborhood services included in the analysis were identified through the aforementioned state and federal databases and Rhode Island GIS data hub. Using these data, we compared services availability between hotspot and non-hotspot neighborhoods.

Neighborhood factors associated with increased overdose incidence

Finally, we fit regression models to identify associations between census block group level social determinants of health data on education, employment, housing, income from the ACS and rates of non-fatal and, separately, fatal overdose. We modelled the expected five-year aggregated counts of non-fatal drug overdose EMS runs and fatal drug overdose at the census block level using a Besag-York-Mollié model specification with a conditional autoregressive (CAR) spatial random effect (Besag et al., 1991). Fixed effects included percentage of crowded housing above or below the mean and quartiles of percent of families living in poverty, vacant housing, owner occupied housing, unemployment, high school graduates or equivalent, families using public assistance, and housing cost burden. A Queen first-order adjacency spatial matrix was used to define neighbors and assign weights. The models accounted for the lack of independence of fatal and non-fatal overdoses in spatially contiguous census block groups and avoids bias due to small area differences. The Poisson-specified CAR model includes corrections for overdispersion, comparable to a negative binomial specification. We estimated posterior means and 95 % credible intervals for all variables and covariates using the integrated nested Laplace approximation method implemented with the R-INLA package in R (Rue et al., 2009). Exponential estimates are interpreted as relative risks (RR).

Results

Overdose hotspots

From 2016 to 2020, there were 1116 fatal overdoses (approximately 105 fatal overdoses/100,000 people) and 7679 EMS runs for non-fatal overdose incidents (approximately 725 nonfatal overdose incidents/100,000 people) (Table 1). Most EMS runs for overdose in overdose hotspots were for individuals 25–44 years of age (58.1 %, 1659/2852) and male (66.2 %, 1889/2852, see Supplemental Table 2). Over three quarters of deaths in overdose hotspots were among people who were male (75.7 %, 334/441), most decedents were White (83.0 %, 366/441), and one fifth were Latino/a (20.6 %, 91/441) (Supplemental Table 2).

Table 1.

Characteristics of neighborhoods which are nonfatal and fatal overdose hotspots and nonhotspots.

Neighborhood Characteristics Non-fatal Overdoses Fatal Overdoses
Hotspot Non-Hotspot Hotspot Non-Hotspot
Populationa (n) 174,770 881,841 218,882 837,729
Overdose eventsb 2937 4742 441 683
Per 100,000 336.1 107.6 39.6 16.3
% higher than by chance 140 % – 88 % –
Sex,% male 49.7 % 48.4 % 49.0 % 48.5 %
Race
 Black (%) 11.6 % 4.5 % 12.8 % 3.9 %
 White (%) 40.5 % 79.4 % 37.5 % 81.0 %
Ethnicity,% Latino/a 40.3 % 10.4 % 41.0 % 8.8 %
Education,% high school graduates 77.3 % 91.0 % 80.1 % 90.9 %
Population living at or below federal poverty line,% 24.3 % 10.0 % 23.7 % 9.4 %
Unemployment rate,% 8.5 % 5.2 % 9.4 % 4.9 %
Population receiving public assistance,% 8.6 % 3.1 % 9.4 % 2.7 %
 Cost-burdened households,% 46.2 % 43.8 % 45.9 % 43.8 %
 Housing vacancy,% housing units 14.2 % 12.0 % 15.1 % 11.6 %
 Rented units,% 66.4 % 46.1 % 64.4 % 33.1 %
 Households without a car,% 19.3 % 7.5 % 18.5 % 7.2 %
Open in a new tab
a

Number of people.

b

Count of overdose incidents.

The geospatial distribution and spatial analyses of non-fatal opioid overdoses identified 7 hotspot areas for non-fatal overdoses in Providence and Kent Counties (Supplemental Figure 1). Fifteen percent of Rhode Islanders live in these areas, but these hotspots account for 35 % of all EMS overdose runs. Per capita, this is 140 % greater EMS runs for overdose than expected.

The spatial scan of fatal opioid overdoses identified 3 hotspot areas in Rhode Island over the study period, all within Providence County (Supplemental Figure 1). Nineteen percent of Rhode Islanders live in these hotspots, where 51 % of overdose deaths occur. These hotspots experience 88 % more overdose deaths than expected.

Differences between overdose hotspot and non-hotspot neighborhoods

Residents in non-fatal and fatal overdose hotspot neighborhoods had higher proportions of Black and Latino/a residents, residents who did not complete high school, and renter-occupied housing (Table 1). Non-fatal and fatal overdose hotspot neighborhoods also had higher proportions of residents who were unemployed and receiving public assistance (Table 1). The proportion of residents living in poverty and without a car in overdose hotspots was more than double those in non-hotspots. Vacant housing and cost-burdened households were also higher in overdose hotspot neighborhoods (Table 1).

Comparing fatal overdose hotspot and non-hotspot neighborhoods, a higher proportion of fatal overdose hotspot neighborhoods had a religious organization (80.7% vs 65.9 %, Table 2) and a federally qualified health center (7.7% vs 3.0 %, Table 2). Non-hotspot neighborhoods were more likely to have a civic or social organization (69.2 % non-hotspot vs 53.0 % hotspot, Table 2). Hotspot neighborhoods were not any more or less likely to have a retail pharmacy, a fire station, a hospital, a library, a beer, wine, or liquor store, a hospital, or an opioid treatment program or DATA-waivered practitioner (Table 2). Similar patterns were observed when comparing EMS hotspot and non-hotspot neighborhoods (Table 3).

Table 2.

Proportion of census block groups within and outside of fatal overdose hotspots with certain amenities and services.

Hotspot areas Non-hotspot areas
Beer, wine, and liquor stores 4.8 % (39/181) 20.1 % (164/634)
Civic or social organizations 53.0 % (96/181) 69.2 % (439/634)
DATA-waivered provider 14.9 % (27/181) 12.3 % (78/634)
Federally qualified health center 7.7 % (14/181) 3.0 % (19/634)
Fire stations 9.4 % (17/181) 17.0 % (108/634)
Hospitals 7.7 % (14/181) 7.7 % (49/634)
Libraries 10.5 % (19/181) 13.7 % (87/634)
Police stations 6.6 % (12/181) 8.4 % (53/634)
Religious organizations 80.7 % (146/181) 65.9 % (418/634)
Retail pharmacies 22.7 % (41/181) 20.8 % (132/634)
Open in a new tab

Table 3.

Proportion of census block groups within and outside of nonfatal overdose hotspots with certain amenities and services.

Hotspot areas Non-hotspot areas
Beer, wine, and liquor stores 21.3 % (30/141) 25.7 % (173/674)
Civic or social organizations 57.4 % (81/141) 67.4 % (454/674)
DATA-waivered provider 16.3 % (23/141) 12.2 % (82/674)
Federally qualified health center 9.9 % (14/141) 2.8 % (19/674)
Fire stations 10.6 % (15/141) 16.3 % (110/674)
Hospitals 12.1 % (17/141) 6.8 % (46/674)
Libraries 9.9 % (14/141) 13.6 % (92/674)
Opioid treatment programs 3.5 % (5/141) 1.8 % (12/674)
Police stations 8.5 % (12/141) 7.9 % (53/674)
Religious organizations 79.4 % (112/141) 67.1 % (452/674)
Retail pharmacies 24.1 % (34/141) 20.6 % (139/674)
Open in a new tab

Neighborhood factors associated with increased overdose incidence

In the regression models, we found an increased relative risk of non-fatal overdoses in neighborhoods with crowded housing above the mean (RR 1.19 [95 % CI 1.05, 1.34]). We observed a dose-response of the relative risk of non-fatal overdoses with increasing proportions of residents living in poverty and the proportion of families receiving public assistance (Table 4). Finally, risk of non-fatal overdoses was lower in neighborhoods with the proportion of high school graduates above the 75th percentile (RR 0.6 [95 % CI 0.51, 0.75]).

Table 4.

Neighborhood characteristics and non-fatal and fatal overdose incidence at the census block level, 2016–2020.

Variable Non-fatal Overdoses* RR (95 % CI) Fatal Overdoses** SMR (95 % CI)
Crowded housing
Below mean (1.90 %) ref ref
Above mean 1.19 (1.05, 1.34) 1.21 (1.18, 1.38)
Proportion population living at or below federal poverty line
Below 25th percentile ref ref
25th to 50th percentile 1.23 (1.04, 1.46) 1.16 (0.91, 1.48)
50th to 75th percentile 1.27 (1.05, 1.54) 1.09 (0.83, 1.43)
Above 75th percentile 1.30 (1.04, 1.63) 1.17 (0.87, 1.59)
Proportion vacant housing
Below 25th percentile ref ref
25th to 50th percentile 1.07 (0.91, 1.27) 1.14 (0.92, 1.42)
50th to 75th percentile 1.17 (0.99, 1.37) 1.15 (0.92, 1.43)
Above 75th percentile 1.00 (0.85, 1.18) 1.12 (0.90, 1.39)
Proportion owner occupied housing
Below 25th percentile ref ref
25th to 50th percentile 0.78 (0.69, 0.92) 0.98 (0.80, 1.20)
50th to 75th percentile 0.81 (0.66, 0.91) 0.77 (0.59, 1.00)
Above 75th percentile 0.68 (0.54, 0.85) 0.85 (0.63, 1.15)
Proportion unemployment
Below 25th percentile ref ref
25th to 50th percentile 0.97 (0.83, 1.14) 1.10 (0.88, 1.38)
50th to 75th percentile 0.97 (0.83, 1.14) 1.12 (0.90, 1.39)
Above 75th percentile 1.10 (0.93, 1.29) 1.24 (1.00, 1.54)
Proportion high school graduate or equivalent completed
Below 25th percentile ref ref
25th to 50th percentile 0.92 (0.79, 1.09) 1.04 (0.84, 1.28)
50th to 75th percentile 0.85 (0.71, 1.01) 0.97 (0.77, 1.22)
Above 75th percentile 0.62 (0.51, 0.75) 0.87 (0.67, 1.14)
Proportion families receiving public assistance
Below 25th percentile ref ref
25th to 50th percentile 1.10 (0.93, 1.30) 1.11 (0.87, 1.41)
50th to 75th percentile 1.26 (1.04, 1.51) 1.21 (0.94, 1.58)
Above 75th percentile 1.60 (1.28, 2.02) 1.67 (1.24, 2.27)
Proportion cost burdened households
Below 25th percentile ref ref
25th to 50th percentile 1.00 (0.84, 1.18) 1.29 (1.02, 1.63)
50th to 75th percentile 0.97 (0.82, 1.16) 1.19 (0.93, 1.51)
Above 75th percentile 0.94 (0.79, 1.12) 1.21 (0.96, 1.54)
Open in a new tab
*

(WAIC 4292.10, DIC 4400.19).

**

(WAIC 2445.57, DIC 2442.45).

Relative risk of fatal overdose was higher in neighborhoods with crowded housing above the mean (RR 1.21 [95 % CI 1.18, 1.38], Table 4), at the highest percentile of families receiving public assistance (RR 1.67 [95 % CI 1.24, 2.27]), and in cost-burdened neighborhoods at the 25–50th percentile (RR 1.29 [95 % CI 1.02, 1.63], Table 4).

Discussion

We used state administrative databases to identify 7 non-fatal overdose hotspots and 3 fatal overdose hotspots in Rhode Island, US between 2016 and 2020; we also evaluated neighborhood social determinants of health and characteristics of the built environment associated with increased overdose incidence. Other Departments of Health and the Centers for Disease Control and Prevention have previously recognized the potential of EMS data for population health surveillance (Kukafka & Yasnoff, 2007; Kukafka et al., 2007; Birkhead et al., 2015; Friedman et al., 2013; McVeigh et al., 2013; Paul et al., 2015; Newton-Dame et al., 2016; Lavrač et al., 2007; Buckeridge, 2007; Politi et al., 2014; Oyana et al., 2017), but there is unrealized potential for using a combination of EMS and other opioid overdose data sources to identify opioid overdose hotspots on a statewide scale to direct resource allocation. Building on unique and innovative partnerships between researchers, RIDOH, and overdose hotspot community-based organizations, this study utilized multiple data sources for precise opioid overdose hotspot identification and characterization.

Our study results not only help identify where resources should be allocated, but also inform intervention design. We found some important differences and similarities between fatal and nonfatal hotspots, which is important for the design and implementation overdose prevention interventions. Consistent with prior research (Bozorgi et al., 2021; Schell et al., 2022; Nandi et al., 2006), our findings indicate that poverty alleviation is essential to address the overdose crisis and to reduce the disproportionate burden of overdose in high-poverty neighborhoods. Multiple prior studies have identified low educational attainment and low income as associated with increased neighborhood-level overdoses (Bozorgi et al., 2021; Schell et al., 2022). Our study showed increased overdose burden in neighborhoods with higher rates of poverty, increased housing crowding, lower educational attainment, and lower proportion of owner-occupied housing. Over the last two decades, overdose has become a leading cause of death among people who are unhoused (Los Angeles County Department of Public Health, Center for Health Impact Evaluation, 2021; Cawley et al., 2022; Fine et al., 2022; Baggett et al., 2013; Barocas et al., 2019). In 2020 and 2021, some cities had more than a 50 % increase in deaths among people who are unhoused, largely driven by overdose (Los Angeles County Department of Public Health, Center for Health Impact Evaluation, 2021; Cawley et al., 2022). While harm reduction and addiction treatment services are of utmost importance in preventing overdose deaths, taken together these studies suggest that to successfully prevent overdoses communities and state agencies must invest in poverty alleviation initiatives, particularly housing-related initiatives. These include eviction relief and prevention, affordable and supportive housing, housing first, employment training and opportunities, and programs to support completion of high school (or equivalent) or additional job or skills training.

Second, the availability of existing services in overdose hotspots presents an opportunity to work with established organizations and infrastructure within communities to prevent overdose deaths. While prior work and our study have shown increased overdose incidence in areas with higher poverty, we found that there are significant resources available in these neighborhoods which could be partners and sites for overdose prevention initiatives. We found that a higher proportion of overdose hotspots had religious organizations, federally qualified health centers, and police stations, while a higher proportion of non-hotspot neighborhoods had civic and social services. In contrast to prior studies, we did not find an association with presence of liquor stores (Fine et al., 2022) nor any differences in the availability of retail pharmacies, opioid treatment programs, libraries, hospitals, or fire stations. Existing organizations in overdose hotspots could be key partners in overdose response efforts. Such efforts include overdose education and naloxone distribution, stigma reduction, services referral, and providing information about and/or provision of evidence-based harm reduction services and addiction treatment. Federally qualified health centers are obvious sites for provision of harm reduction services and addiction treatment, but there is also research which has demonstrated the feasibility of providing addiction treatment in religious organizations, specifically in Black churches (Jordan et al., 2021). This is a crucial area for further work to inform intervention design and address racial and economic inequities in treatment access, engagement, and outcomes.

This study has several limitations. First, while EMS calls for non-fatal overdose and SUDORS data for overdose deaths represents the most comprehensive datasets available to measure overdose incidence, it does not include non-fatal overdoses for which EMS was not called (e.g., overdoses reversed by bystanders, overdoses for which friends and/or family transported patients to the emergency department). As a result, this analysis underestimates true overdose incidence in Rhode Island. Second, availability of addiction treatment and harm reduction services access is more complex than we were able to measure. Locations of addiction treatment providers throughout Rhode Island included information about opioid treatment programs, addiction specialist services licensed by the state, and prescribers on SAMHSA’s DATA 2000 database. However, these databases do not contain important information about treatment access including hours of operation, types(s) of insurance accepted, types of services provided, and appointment availability. Furthermore, the SAMHSA list of buprenorphine prescribers is opt-in, not updated regularly, and does not include information about whether clinicians are actively prescribing or accepting patients. Harm reduction services and resources may also not be sufficiently accounted for in the study model. Mobile syringe services, street outreach routes and times, and locations of NaloxBoxes, publicly available naloxone for emergency use, are not included in our model. However, while not in the analytic model, these are examples of interventions which this type of data analyses could be used to inform implementation. Third, while we examined geographic location of hotspots, we did not examine changes in hotspot locations over time. General geographic locations of overdose hotspots in Rhode Island have been fairly stable over time (Clarke, 1996; Rhode Island Department of Health, 2023, von Elm et al., 2007, Hallowell et al., 2021); however, novel approaches may be needed to rapidly identify emerging hotspots for prompt resource provision. Fourth, findings from this study are observational and at the neighborhood level, may not be generalizable to other areas, and should not be interpreted as causal or as representing individual overdose risk factors. Moreover, method replicability depends on data availability, which may limit generalizability of our approach in areas with more restricted data sources. Finally, multiple factors at the neighborhood and individual levels contribute to overdose risk. One factor not measured in this analysis is the increasing potency of the drug supply, both opioid and stimulant, and overdose toxicology. The introduction of fentanyl analogues into the heroin, counterfeit pill, and stimulant drug supplies is a major factor in increased overdose fatalities and the geographic distribution of these analogues was not measured in the analysis.

Public health implications

Geospatial analysis of overdose hotspots has important public health applications to select optimal locations for resource provision, monitor changes in overdose incidence, and ensure that harm reduction, addiction treatment, and overdose prevention interventions achieve maximum public health benefit. In Rhode Island, RIDOH’s CODE initiative has leveraged geospatial analysis for overdose prevention through the state’s Health Equity Zones (local collaborations funded by RIDOH to address health equity and population health problems and improve public health) (Rhode Island Department of Health, 2023) and community organizations to design and implement place-based, community-led programs in overdose hotspots. Future work should seek to establish the effectiveness of these place-based, community-led programs; if shown to be effective, the CODE program could serve as a model for other states to rapidly deploy resources to curb overdose deaths.

Ongoing work in predictive analytics is key to evaluate the feasibility of using state-level data to predict areas of increased overdose incidence and prevent overdose death (Marshall et al., 2022). To reduce overdose deaths, our study findings support directing resources toward poverty alleviation initiatives. Overdose hotspots in Rhode Island have an existing infrastructure of civic and social services which, if given sufficient resources, organizational support, and timely data, may be able to reduce overdose deaths through expansion of evidence-based interventions.

Supplementary Material

Supplementary Materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.drugpo.2024.104322.

Acknowledgements

The authors would like to thank the RIDOH Health Equity Zones and CODE grant recipients and participants for their participation, collaboration, and ongoing work and dedication to preventing overdose deaths in their local communities. We would also like to thank Rachael Scagos, MPH for her support of this study.

Funding

This study was funded by an Advance Clinical and Translational Research Mentored Research Award at Brown University which is supported by a National Institute of General Medical Sciences IDeA-CTR grant (U54GM115677) and the National Institute on Drug Abuse (R01-DA046620). Dr. Samuels is also partially supported by the Centers for Disease Control and Prevention (R01CE003516). Dr. Marshall and Dr. Goedel’s contributions were partially supported by the National Institute of General Medical Sciences (P20-GM125507) and the National Institute on Drug Abuse (R01-DA046620). Dr. Becker’s contribution was partially supported by the National Institute on Drug Abuse (P50DA054072, U2CDA057717).

Footnotes

CRediT authorship contribution statement

Elizabeth A. Samuels: Writing – review & editing, Writing – original draft, Visualization, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. William C. Goedel: Writing – review & editing, Writing – original draft, Visualization, Methodology, Formal analysis, Data curation, Conceptualization. Victoria Jent: Writing – review & editing, Writing – original draft, Methodology, Formal analysis. Lauren Conkey: Writing – review & editing, Project administration, Conceptualization. Benjamin D. Hallowell: Writing – review & editing, Data curation. Sarah Karim: Formal analysis, Data curation. Jennifer Koziol: Writing – review & editing, Project administration, Conceptualization. Sara Becker: Writing – review & editing, Conceptualization. Rachel R. Yorlets: Writing – review & editing, Investigation, Data curation. Roland Merchant: Writing – review & editing, Conceptualization. Lee Ann Jordison Keeler: Writing – review & editing, Project administration. Neha Reddy: Writing – review & editing, Data curation. James McDonald: Writing – review & editing, Conceptualization. Nicole Alexander-Scott: Writing – review & editing, Conceptualization. Magdalena Cerda: Writing – review & editing, Supervision, Funding acquisition. Brandon D.L. Marshall: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization.

Declaration of competing interest

None.

References

  1. Baggett TP, Hwang SW, O’Connell JJ, Porneala BC, Stringfellow EJ, Orav EJ, et al. (2013). Mortality among homeless adults in Boston. JAMA Internal Medicine, 173(3), 189. 10.1001/jamainternmed.2013.1604 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Barocas JA, Wang J, Marshall BDL, LaRochelle MR, Bettano A, Bernson D, et al. (2019). Sociodemographic factors and social determinants associated with toxicology confirmed polysubstance opioid-related deaths. Drug and Alcohol Dependence, 200, 59–63. 10.1016/j.drugalcdep.2019.03.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Besag J, York J, & Mollié A (1991). Bayesian image restoration, with two applications in spatial statistics. Annals of the Institute of Statistical Mathematics, 43(1), 1–20. 10.1007/BF00116466 [DOI] [Google Scholar]
  4. Birkhead GS, Klompas M, & Shah NR (2015). Uses of electronic health records for public health surveillance to advance public health. Annual Review of Public Health, 36(1), 345–359. 10.1146/annurev-publhealth-031914-122747 [DOI] [PubMed] [Google Scholar]
  5. Bozorgi P, Porter DE, Eberth JM, Eidson JP, & Karami A (2021). The leading neighborhood-level predictors of drug overdose: A mixed machine learning and spatial approach. Drug and Alcohol Dependence, 229, Article 109143. 10.1016/j.drugalcdep.2021.109143 [DOI] [PubMed] [Google Scholar]
  6. Buckeridge DL (2007). Outbreak detection through automated surveillance: A review of the determinants of detection. Journal of Biomedical Informatics, 40(4), 370–379. 10.1016/j.jbi.2006.09.003 [DOI] [PubMed] [Google Scholar]
  7. Cawley C, Kanzaria HK, Zevin B, Doran KM, Kushel M, & Raven MC (2022). Mortality among people experiencing homelessness in San Francisco during the COVID-19 pandemic. JAMA Network Open, 5(3), Article e221870. 10.1001/jamanetworkopen.2022.1870 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Ahmad F, Cisewski J, Rossen LM, Sutton P Provisional drug overdose death counts.; 2023. https://www.cdc.gov/nchs/nvss/vsrr/drug-overdose-data.htm. Accessed September 27, 2023. [Google Scholar]
  9. Centers for Disease Control and Prevention, National Center for Health Statistics. Life expectancy in the U.S. dropped for the second year in a row in 2021. Published August 31, 2022. https://www.cdc.gov/nchs/pressroom/nchs_press_releases/2022/20220831.htm. Accessed September 27, 2023.
  10. Centers for Disease Control and Prevention. (2023b). National center for injury prevention and control. CDC’s State Unintentional Drug Overdose Reporting System (SUDORS). Published August 23 https://www.cdc.gov/drugoverdose/fatal/sudors.html Accessed September 27, 2023. [Google Scholar]
  11. Clarke K (1996). On epidemiology and geographic information systems: A review and discussion of future directions. Emerging Infectious Diseases, 2(2), 85–92. 10.3201/eid0202.960202 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Fine DR, Dickins KA, Adams LD, De Las Nueces D, Weinstock K, Wright J, et al. (2022). Drug overdose mortality among people experiencing homelessness, 2003 to 2018. JAMA Network Open, 5(1), Article e2142676. 10.1001/jamanetworkopen.2021.42676 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Friedman DJ, Parrish RG, & Ross DA (2013). Electronic health records and US public health: Current realities and future promise. American Journal of Public Health, 103(9), 1560–1567. 10.2105/AJPH.2013.301220 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Friedman JR, & Hansen H (2022). Evaluation of increases in drug overdose mortality rates in the US by race and ethnicity before and during the COVID-19 pandemic. JAMA Psychiatry, 79(4), 379. 10.1001/jamapsychiatry.2022.0004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Goedel WC, Green TC, Viner-Brown S, Rich JD, & Marshall BDL (2019). Increased overdose mortality during the first week of the month: Revisiting the “check effect” through a spatial lens. Drug and Alcohol Dependence, 197, 49–55. 10.1016/j.drugalcdep.2018.12.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Greene SK, Peterson ER, Kapell D, Fine AD, & Kulldorff M (2016). Daily reportable disease spatiotemporal cluster detection, New York City, New York, USA, 2014–2015. Emerging Infectious Diseases, 22(10), 1808–1812. 10.3201/eid2210.160097 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hallowell BD, Chambers LC, Rhodes J, Basta M, Viner-Brown S, & Lasher L (2021). Using emergency medical services data to monitor nonfatal opioid overdoses in real time. Public Health Reports, 136(1_suppl), 40S–46S. 10.1177/00333549211018989 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Jordan A, Babuscio T, Nich C, & Carroll KM (2021). A feasibility study providing substance use treatment in the Black church. Journal of Substance Abuse Treatment, 124, Article 108218. 10.1016/j.jsat.2020.108218. Epub 2020 Dec 3. [DOI] [PubMed] [Google Scholar]
  19. Kaiser Family Foundation. (2022). Opioid overdose death rates and all drug overdose death rates per 100,000 population (Age-Adjusted). Kaiser Family Foundation. [Google Scholar]
  20. Kukafka R, Ancker JS, Chan C, Chelico J, Khan S, Mortoti S, et al. (2007). Redesigning electronic health record systems to support public health. Journal of Biomedical Informatics, 40(4), 398–409. 10.1016/j.jbi.2007.07.001 [DOI] [PubMed] [Google Scholar]
  21. Kukafka R, & Yasnoff WA (2007). Public health informatics. Journal of Biomedical Informatics, 40(4), 365–369. 10.1016/j.jbi.2007.07.005 [DOI] [PubMed] [Google Scholar]
  22. Kulldorff M (1997). A spatial scan statistic. Communications in Statistics: Theory and Methods, 26(6), 1481–1496. 10.1080/03610929708831995 [DOI] [Google Scholar]
  23. Kulldorff M, Heffernan R, Hartman J, Assunção R, & Mostashari F (2005). A space–time permutation scan statistic for disease outbreak detection. PLoS Medicine, 2(3), e59. 10.1371/journal.pmed.0020059 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lasher L, Hallowell BD, Chambers LC, Koziol J, McDonald J, Elmaleh R, et al. (2021). Using timely overdose data to address a spike in nonfatal overdoses and inform a coordinated community-level response in Rhode Island, 2019. Public Health Reports, 136(1_suppl), 24S–30S. 10.1177/00333549211012407 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lasher L, Rhodes J, & Viner-Brown S (2019). Identification and description of non-fatal opioid overdoses using Rhode Island EMS Data, 2016–2018. Rhode Island Medical Journal (2013), 102(2), 41–45. [PubMed] [Google Scholar]
  26. Lavrač N, Bohanec M, Pur A, Cestnik B, Debeljak M, & Kobler A (2007). Data mining and visualization for decision support and modeling of public health-care resources. Journal of Biomedical Informatics, 40(4), 438–447. 10.1016/j.jbi.2006.10.003 [DOI] [PubMed] [Google Scholar]
  27. Marshall BDL, Alexander-Scott N, Yedinak JL, Hallowell BD, Goedel WC, Allen B, … Cerdá M (2022). Preventing overdose using information and data from the environment (PROVIDENT): Protocol for a randomized, population-based, community intervention trial. Addiction, 117(4), 1152–1162. 10.1111/add.15731 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Los Angeles County Department of Public Health, Center for Health Impact Evaluation. Recent trends in mortality rates and causes of death among people experiencing homelessness in Los Angeles county. Published January 2021. http://publichealth.lacounty.gov/chie/reports/HomelessMortality2020_CHIEBrief_Final.pdf. Accessed September 27, 2023. [Google Scholar]
  29. McVeigh K, Newton-Dame R, Perlman S, Chernov C, Thorpe L, Singer J, Green C Developing an electronic health record-based population health surveillance system.; 2013. https://www.nyc.gov/assets/doh/downloads/pdf/data/nyc-macro-report.pdf. Accessed September 27, 2023.
  30. Miranda ML, Ferranti J, Strauss B, Neelon B, & Califf RM (2013). Geographic health information systems: A platform to support the ‘triple aim’. Health Affairs, 32 (9), 1608–1615. 10.1377/hlthaff.2012.1199 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Nandi A, Galea S, Ahern J, Bucciarelli A, Vlahov D, & Tardiff K (2006). What explains the association between neighborhood-level income inequality and the risk of fatal overdose in New York City? Social Science & Medicine (1982), 63(3), 662–674. 10.1016/j.socscimed.2006.02.001 [DOI] [PubMed] [Google Scholar]
  32. Nesoff ED, Branas CC, & Martins SS (2020). The geographic distribution of fentanyl-involved overdose deaths in cook county, Illinois. American Journal of Public Health, 110(1), 98–105. 10.2105/AJPH.2019.305368. Epub 2019 Nov 14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Newton-Dame R, McVeigh KH, Schreibstein L, Perlman S, Lurie-Moroni L, Jacobson L, et al. (2016). Design of the New York City macroscope: Innovations in population health surveillance using electronic health records. eGEMs (Generating Evidence & Methods to Improve Patient Outcomes), 4(1), 26. 10.13063/2327-9214.1265 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Olfson M, Crystal S, Wall M, Wang S, Liu SM, & Blanco C (2018b). Causes of death after nonfatal opioid overdose. JAMA Psychiatry, 75(8), 820. 10.1001/jamapsychiatry.2018.1471 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Olfson M, Wall M, Wang S, Crystal S, & Blanco C (2018a). Risks of fatal opioid overdose during the first year following nonfatal overdose. Drug and Alcohol Dependence, 190, 112–119. 10.1016/j.drugalcdep.2018.06.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Oyana TJ, Podila P, Wesley JM, Lomnicki S, & Cormier S (2017). Spatiotemporal patterns of childhood asthma hospitalization and utilization in Memphis Metropolitan Area from 2005 to 2015. Journal of Asthma, 54(8), 842–855. 10.1080/02770903.2016.1277537 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Paul MM, Greene CM, Newton-Dame R, Thorpe LE, Perlman SE, McVeigh KH, et al. (2015). The state of population health surveillance using electronic health records: A narrative review. Population Health Management, 18(3), 209–216. 10.1089/pop.2014.0093 [DOI] [PubMed] [Google Scholar]
  38. Politi L, Codish S, Sagy I, & Fink L (2014). Use patterns of health information exchange through a multidimensional lens: Conceptual framework and empirical validation. Journal of Biomedical Informatics, 52, 212–221. 10.1016/j.jbi.2014.07.003 [DOI] [PubMed] [Google Scholar]
  39. Rhode Island Department of Behavioral Healthcare, Developmental Disabilities, & Hospitals. (2023). Licensed substance abuse treatment providers. Rhode Island Department of Behavioral Healthcare. Developmental Disabilities, & Hospitals. Published 2023. https://bhddh.ri.gov/substance-useaddiction/licensed-treatment-providers-0 Accessed September 27, 2023 [Google Scholar]
  40. Rhode Island Department of Business Regulation. (2023). Liquor enforcement and compliance. Rhode Island Department of Business Regulation Division of Commercial Licensing. Published https://dbr.ri.gov/real-estate-and-commercial-licensing/liquor Accessed September 27, 2023. [Google Scholar]
  41. Rhode Island Department of Health. (2022). Rhode Island Department of Health: Drug overdose surveillance data hub: Opioid overdose-related emergency department visits. Rhode Island Health Department of Health Map Portal. Published https://ridoh-drug-overdose-surveillance-eddashboard-rihealth.hub.arcgis.com Accessed September 27, 2023. [Google Scholar]
  42. Pustz J, Srinivasan S, Shrestha S, Larochelle MR, Walley AY, Samet JH, Babakhanlou-Chase H, Carpenter JF, Stopka TJ (2023). Applied risk mapping and spatial analysis of address-level decedent data to inform opioid overdose interventions: The Massachusetts healing communities study. Drug and Alcohol Dependence, 251, 110947.doi: 10.1016/j.drugalcdep.2023.110947. Epub 2023 Aug 23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Rhode Island Department of Health. Rhode Island’s health equity zone (HEZ) initiative. Published August 28, 2023. https://health.ri.gov/programs/detail.php?pgm_id=1108. Accessed September 27, 2023.
  44. Rhode Island Department of Health. (2023a). Rhode Island Department of Health: Drug overdose surveillance data hub: Overdose fatalities. Rhode Island Health Department of Health Map Portal. Published 2022 https://ridoh-drug-overdose-surveillance-fatalities-rihealth.hub.arcgis.com Accessed September 27. [Google Scholar]
  45. Rudd RA, Seth P, David F, & Scholl L (2016). Increases in drug and opioid-involved overdose deaths — United States, 2010–2015. MMWR Morbidity and Mortality Weekly Report, 65(5051), 1445–1452. 10.15585/mmwr.mm655051e1 [DOI] [PubMed] [Google Scholar]
  46. Rue H, Martino S, & Chopin N (2009). Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society. Series B, Statistical Methodology, 71(2), 319–392. 10.1111/j.1467-9868.2008.00700.x [DOI] [Google Scholar]
  47. Schell RC, Allen B, Goedel WC, Hallowell BD, Scagos R, Li Y, et al. (2022). Identifying predictors of opioid overdose death at a neighborhood level with machine learning. American Journal of Epidemiology, 191(3), 526–533. 10.1093/aje/kwab279 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Sherman RL, Henry KA, Tannenbaum SL, Feaster DJ, Kobetz E, & Lee DJ (2014). Applying spatial analysis tools in public health: An example using SaTScan to detect geographic targets for colorectal cancer screening interventions. Preventing Chronic Disease, 11, Article 130264. 10.5888/pcd11.130264 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Stewart K, Cao Y, Hsu MH, Artigiani E, & Wish E (2017). Geospatial analysis of drug poisoning deaths involving heroin in the USA, 2000–2014. Journal of Urban Health, 94(4), 572–586. 10.1007/s11524-017-0177-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Rhode Island Geographic Information System. Rhode Island maps and data: Geospatial data hub. Published 2023. https://www.rigis.org. Accessed September 27, 2023.
  51. Substance Abuse and Mental Health Services Administration. Buprenorphine Practitioner Locator. Published 2023. https://www.samhsa.gov/medication-assisted-treatment/find-treatment/treatment-practitioner-locator?field_bup_state_value=47. Accessed September 27, 2023.
  52. United States Census Bureau. American community survey. Published August 16, 2023. https://www.census.gov/programs-surveys/acs/data.html. Accessed September 27, 2023.
  53. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. (2007). The strengthening the reporting of observational studies in epidemiology (STROBE) statement: Guidelines for reporting observational studies. Lancet, 370(9596), 1453–1457. 10.1016/S0140-6736(07)61602-X [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Materials
NIHMS2124869-supplement-Supplementary_Materials.docx (331.1KB, docx)

RESOURCES