Bystander Presence and Response During Accidental and Undetermined Drug Overdose Deaths: Rhode Island, January 1, 2016-December 31, 2021.

With timely intervention from a bystander, drug overdose victims are more likely to survive. To characterize the frequency of bystander presence and identify overdose response barriers, we analyzed data from overdose fatalities occurring in Rhode Island from 2016 to 2021. Overall, about half (n=1,039; 48.7%) of all overdose deaths in Rhode Island had at least one bystander present. Among decedents who had at least one bystander who was unable to respond (n=338), top reasons of non-response were because they were spatially separated (64.8%), failed to recognize the signs of overdose (54.1%), or were unaware the victim was using drugs (40.2%). To promote by- stander presence and address barriers to bystander response during an overdose, intervention strategies should include efforts that reduce solitary drug use and maximize bystander efficacy, including increasing awareness on the dangers of using drugs alone, increasing the availability of naloxone, and education on recognizing signs of overdose.

Public Health Dashboards in Overdose Prevention: The Rhode Island Approach to Public Health Data Literacy, Partnerships, and Action.

As the field of public health rises to the demands of real-time surveillance and rapid data-sharing needs in a postpandemic world, it is time to examine our approaches to the dissemination and accessibility of such data. Distinct challenges exist when working to develop a shared public health language and narratives based on data. It requires that we assess our understanding of public health data literacy, revisit our approach to communication and engagement, and continuously evaluate our impact and relevance. Key stakeholders and cocreators are critical to this process and include people with lived experience, community organizations, governmental partners, and research institutions. In this viewpoint paper, we offer an instructive approach to the tools we used, assessed, and adapted across 3 unique overdose data dashboard projects in Rhode Island, United States. We are calling this model the "Rhode Island Approach to Public Health Data Literacy, Partnerships, and Action." This approach reflects the iterative lessons learned about the improvement of data dashboards through collaboration and strong partnerships across community members, state agencies, and an academic research team. We will highlight key tools and approaches that are accessible and engaging and allow developers and stakeholders to self-assess their goals for their data dashboards and evaluate engagement with these tools by their desired audiences and users.

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

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.

PROVIDENT: Development and Validation of a Machine Learning Model to Predict Neighborhood-level Overdose Risk in Rhode Island.

BACKGROUND: Drug overdose persists as a leading cause of death in the United States, but resources to address it remain limited. As a result, health authorities must consider where to allocate scarce resources within their jurisdictions. Machine learning offers a strategy to identify areas with increased future overdose risk to proactively allocate overdose prevention resources. This modeling study is embedded in a randomized trial to measure the effect of proactive resource allocation on statewide overdose rates in Rhode Island (RI). METHODS: We used statewide data from RI from 2016 to 2020 to develop an ensemble machine learning model predicting neighborhood-level fatal overdose risk. Our ensemble model integrated gradient boosting machine and super learner base models in a moving window framework to make predictions in 6-month intervals. Our performance target, developed a priori with the RI Department of Health, was to identify the 20% of RI neighborhoods containing at least 40% of statewide overdose deaths, including at least one neighborhood per municipality. The model was validated after trial launch. RESULTS: Our model selected priority neighborhoods capturing 40.2% of statewide overdose deaths during the test periods and 44.1% of statewide overdose deaths during validation periods. Our ensemble outperformed the base models during the test periods and performed comparably to the best-performing base model during the validation periods. CONCLUSIONS: We demonstrated the capacity for machine learning models to predict neighborhood-level fatal overdose risk to a degree of accuracy suitable for practitioners. Jurisdictions may consider predictive modeling as a tool to guide allocation of scarce resources.

Prior Emergency Medical Services Utilization Among People Who Had an Accidental Opioid-Involved Fatal Drug Overdose-Rhode Island, 2018-2020.

OBJECTIVE: To help understand whether decreased emergency medical services (EMS) utilization due to the COVID-19 pandemic contributed to increased accidental fatal drug overdoses, we characterized recent EMS utilization history among people who had an accidental opioid-involved fatal drug overdose in Rhode Island. METHODS: We identified accidental opioid-involved fatal drug overdoses among Rhode Island residents that occurred from January 1, 2018, through December 31, 2020. We linked decedents by name and date of birth to the Rhode Island EMS Information System to obtain EMS utilization history. RESULTS: Among 763 people who had an accidental opioid-involved fatal overdose, 51% had any EMS run and 16% had any opioid overdose-related EMS run in the 2 years before death. Non-Hispanic White decedents were significantly more likely than decedents of other races and ethnicities to have any EMS run (P < .001) and any opioid overdose-related EMS run (P = .05) in the 2 years before death. Despite a 31% increase in fatal overdoses from 2019 through 2020, corresponding with the onset of the COVID-19 pandemic, EMS utilization in the prior 2 years, prior 180 days, or prior 90 days did not vary by time frame of death. CONCLUSION: In Rhode Island, decreased EMS utilization because of the COVID-19 pandemic was not a driving force behind the increase in overdose fatalities observed in 2020. However, with half of people who had an accidental opioid-involved fatal drug overdose having an EMS run in the 2 years before death, emergency care is a potential opportunity to link people to health care and social services.

Association between comorbid chronic pain or prior hospitalization for mental illness and substance use treatment among a cohort at high risk of opioid overdose.

INTRODUCTION: Chronic pain and serious mental illness increase risk of opioid use, and opioid use can exacerbate both conditions. Substance use disorder (SUD) treatment can be lifesaving, but chronic pain and serious mental illness may make recovery challenging. We evaluated the association between current chronic pain and prior hospitalization for mental illness and 90-day SUD treatment engagement, among emergency department (ED) patients at high risk of opioid overdose. METHODS: We conducted a cohort analysis of 648 ED patients enrolled in a randomized controlled trial in Rhode Island. We linked baseline study data on chronic pain and prior hospitalization for mental illness to statewide administrative data on state-licensed treatment programs (including methadone) and buprenorphine treatment via prescription. We defined treatment engagement as initiation of a state-licensed treatment program, transfer between state-licensed programs/providers, or a buprenorphine prescription (re-)fill. We used modified Poisson models to estimate the association between each baseline comorbidity and treatment engagement within 90 days following the ED visit, adjusted for a priori potential confounders. In an exploratory analysis, models were stratified by baseline treatment status. RESULTS: The mean age of participants was 37 years; 439 (68 %) were male, and 446 (69 %) had been recently unhoused. Overall, 278 participants (43 %) engaged in treatment within 90 days of the ED visit. Participants with prior hospitalization for mental illness were more likely to engage in treatment than those without (adjusted risk ratio [ARR] = 1.24, 95 % confidence interval [CI] = 1.01-1.53), although this association was only among those already accessing treatment at baseline (ARR = 1.58, 95 % CI = 1.10-2.27). Chronic pain was not associated with 90-day treatment engagement overall (ARR = 1.12, 95 % CI = 0.91-1.38) or within baseline treatment subgroups. CONCLUSIONS: Among ED patients at high risk of opioid overdose and accessing treatment at baseline, those with prior hospitalization for mental illness (but not chronic pain) were more likely to engage in treatment following the ED visit, which may reflect disproportionate initiation of additional treatment programs, transfer between programs/providers, or ongoing buprenorphine treatment. Touchpoints within the medical system should be leveraged to ensure that everyone, including those with serious mental illness, can access high-quality SUD treatment at the desired intensity level.

Differences in Substance Use and Harm Reduction Practices by Race and Ethnicity: Rhode Island Harm Reduction Surveillance System, 2021-2022.

CONTEXT: In the United States, minority populations are disproportionately affected by the overdose epidemic, have higher mortality rates, and unequal access to harm reduction and treatment services. OBJECTIVE: This analysis aims to better understand harm reduction utilization and substance use patterns among minority populations to improve overdose outreach and prevention initiatives in Rhode Island. DESIGN: The present analysis used data from the Harm Reduction Surveillance System from January 2021 to December 2022 (N = 393). Chi-square tests and multivariable regression models were used to investigate differences in substance use behaviors by race and ethnicity. SETTING: Rhode Island. PARTICIPANTS: Participants include individuals who self-reported the use of illicit drugs, currently reside in Rhode Island, and were older than 18 years. MAIN OUTCOME MEASURES: Methods of drug use and uptake of harm reduction practices. RESULTS: Among survey participants, 41% were non-Hispanic White, 57% were aged 25 to 44 years, 62% identified as male, and 95% had health insurance coverage. Most participants reported smoking as their method of drug use (90%) and harm reduction practices were underutilized by all race and ethnicity groups. Fewer non-Hispanic Black participants reported carrying naloxone compared to the other race and ethnicity groups. Non-Hispanic Black and Hispanic participants were significantly less likely to inject drugs compared with non-Hispanic White participants (adjusted odds ratio [AOR] = 0.14; 95% confidence interval [CI], 0.04-0.45) (AOR = 0.40; 95% CI, 0.18-0.90). CONCLUSIONS: Smoking was the most common self-reported method of substance administration for all participants, whereas injection was more prevalent among non-Hispanic White participants. There is a continued need for minority-led and culturally informed harm reduction and treatment services for minority populations.

"I Don't Go Overboard": Perceptions of Overdose Risk and Risk Reduction Strategies among People Who Use Drugs in Rhode Island.

Background: North America experiencing an unprecedented overdose epidemic, with data estimating almost 110,000 overdose deaths occurring in 2022 in the United States (US). To address fatal overdoses in the US, community organizations and local health departments in some jurisdictions have expanded community distribution of naloxone, and increased overdose prevention education, and other harm reduction supplies and services (e.g., fentanyl test strips, drug checking programs) to reduce harm for people who use drugs (PWUD). Objectives: Understanding how PWUD manage overdose risk within the context of these expanded services is important for ensuring public health services are meeting their needs. Semi-structured qualitative interviews were conducted with 25 PWUD who were accessing harm reduction services in Rhode Island. Data were imported into NVivo where they were coded and analyzed thematically. Results: Our findings demonstrate the complexity of managing overdose risk in the context of a fentanyl drug supply. While most participants were concerned about overdosing, they sought to manage overdose risk through their own harm reduction practices (e.g., testing their drugs, going slow) and drug purchasing dynamics, even when using alone. Conclusions: Study findings point to the need to implement and scale-up community-level interventions to better support PWUD within the context of the current US overdose crisis.

Racial and Ethnic Disparities in Accidental and Undetermined Drug Overdose Deaths - Rhode Island, 2016-2021.

BACKGROUND: Over the last 21 years, drug overdose deaths have increased nationwide, particularly in Rhode Island (RI). The goal of this work is to identify populations that are disproportionately impacted by accidental drug overdose deaths to guide future intervention efforts.  Methods: We obtained data on accidental and undetermined drug overdose deaths from the RI State Unintentional Drug Overdose Reporting System (SUDORS) database from 2016 to 2021. We compared basic demographic factors stratified by decedent race and ethnicity. Chi Square and Fisher's exact tests were used to determine differences in characteristics across racial groups. FINDINGS: Across all race and ethnic groups, cocaine-involved (64.5%) fatal overdoses were higher among Black non-Hispanics. White, non-Hispanics (31.4%) were more likely to receive treatment for substance use disorder (SUD) when compared to minority groups. CONCLUSION: To help address the gaps in treatment for SUD among minority groups, culturally responsive intervention efforts should prioritize connecting minority groups to treatment.

Correlates of healthcare-seeking behavior for acute gastroenteritis-United States, October 1, 2016 -September 30, 2017.

BACKGROUND: In the United States, public health surveillance systems often underestimate the burden of acute gastroenteritis (AGE) because they only identify disease among those who interact with the healthcare system. OBJECTIVE: To identify factors associated with healthcare-seeking behavior among individuals experiencing community-acquired AGE. METHODS: From October 2016 -September 2017, we conducted a weekly, age-stratified, random sample of Kaiser Permanente Northwest members located in northwest Oregon and southwest Washington, United States. Individuals who completed the online survey and experienced AGE were included in the analysis. Univariate and multivariable logistic regressions were performed to identify predictors of healthcare-seeking behavior. RESULTS: Of the 3,894 survey respondents, 395 experienced an AGE episode and were eligible for analysis, of whom, 82 (21%) sought care for their AGE episode. In the final multivariable model, individuals with a concurrent fever (odds ratio [OR]: 4.76, 95% confidence interval [95% CI]: 2.48-9.13), increased diarrhea duration (≥6 days vs 1-4 days, OR: 4.22, 95% CI: 1.78-10.03), or increased vomiting duration (≥3 days vs 1 days, OR: 2.97, 95% CI: 1.22-7.26), were significantly more likely to seek healthcare. In the adjusted model, no sociodemographic or chronic disease variables were associated with healthcare-seeking behavior. CONCLUSION: These findings suggest that individuals with a short duration of AGE and those without concurrent fever are underrepresented in healthcare facility-based surveillance systems.

Buprenorphine Dose and Time to Discontinuation Among Patients With Opioid Use Disorder in the Era of Fentanyl.

Importance: Buprenorphine treatment for opioid use disorder (OUD) has more than doubled since 2009. However, current US Food and Drug Administration buprenorphine dosing guidelines are based on studies among people using heroin, prior to the emergence of fentanyl in the illicit drug supply. Objective: To estimate the association between buprenorphine dose and time to treatment discontinuation during a period of widespread fentanyl availability. Design, Setting, and Participants: This retrospective cohort study used statewide Rhode Island Prescription Drug Monitoring Program data. Participants were Rhode Island residents initiating buprenorphine treatment for OUD between October 1, 2016, and September 30, 2020. Data analysis was performed from December 9, 2022, to August 10, 2023. Exposure: Daily dose of buprenorphine (16 mg and 24 mg) defined starting on the day of initiation based on total quantity and days' supply dispensed. Patients were censored on any dose change. Main Outcomes and Measures: Buprenorphine treatment discontinuation in the 180 days following initiation, defined as a gap in treatment of more than 27 days based on prescription fill dates and days' supply. Kaplan-Meier and Cox regression survival analyses were conducted to estimate the association between buprenorphine dose and time to treatment discontinuation, controlling for potential informative censoring and measured potential confounders. Results: Among 6499 patients initiating buprenorphine treatment for OUD, most were aged 25 to 44 years (57%; n = 3682), were male (61%; n = 3950), and had private (47%; n = 3025) or Medicaid (33%; n = 2153) insurance. More than half of patients were prescribed a daily dose of interest at initiation (16 mg: 50%; n = 3264; 24 mg: 10%; n = 668). In Kaplan-Meier analyses, 58% of patients discontinued buprenorphine treatment within 180 days (16 mg: 59% vs 24 mg: 53%; log-rank test P = .005). In Cox regression analyses, patients prescribed a dose of 16 mg had a greater risk of treatment discontinuation than those prescribed 24 mg (adjusted hazard ratio, 1.20; 95% CI, 1.06-1.37). Conclusions and Relevance: In this cohort study of patients initiating buprenorphine treatment from 2016 to 2020, patients prescribed a 24 mg dose of buprenorphine remained in treatment longer than those prescribed 16 mg. The value of higher buprenorphine doses than currently recommended needs to be considered for improving retention in treatment.

The impact of polysubstance use patterns on engagement of substance use disorder treatment among emergency department patients at high risk of opioid overdose.

Background: Substance use patterns are diverse, and multiple substances are often involved in fatal and nonfatal overdoses. Additionally, polysubstance use is associated with greater difficulty accessing and remaining in substance use disorder (SUD) treatment. The aim of this study was to identify substance use patterns and determine their association with SUD treatment engagement among emergency department (ED) patients at risk of opioid overdose. Methods: This was a sub-analysis of a randomized controlled trial comparing two behavioral interventions for individuals at two EDs in Rhode Island from 2018 to 2021. Past six-month substance use frequency for eight substances plus injection drug use was self-reported at trial enrollment, and SUD treatment engagement within 90 days after enrollment was obtained using administrative data linkages. Latent class analysis identified substance use patterns and multivariable log-binomial models estimated the association with SUD treatment engagement. Results: Among 607 participants, there were four substance use patterns: 1) low reported use (n = 295), 2) frequent injection and heroin use (n = 131), 3) high frequency broad polysubstance use (n = 62), and 4) low frequency broad polysubstance use (n = 119). Compared to participants with the low reported use pattern, those with the frequent injection and heroin pattern had a greater likelihood of SUD treatment engagement (adjusted risk ratio = 1.28; 95% confidence interval = 1.02-1.61). Conclusions: Distinct and meaningful polysubstance use patterns showed differential SUD treatment engagement after ED discharge. Nuanced relationships between substance use patterns and treatment highlight the necessity for tailored harm reduction, treatment, and recovery services.

Trends in Initiate Pediatric Opioid Prescriptions in Rhode Island: 2017-2021.

OBJECTIVE: To analyze recent trends in initiate pediatric opioid prescriptions dispensed in Rhode Island. METHODS: All Rhode Island residents aged 0-17 years with an initiate opioid prescription dispensed between January 1, 2017 and December 31, 2021 were obtained from the Rhode Island Prescription Drug Monitoring Program. Analyses were conducted to investigate trends related to patient demographics, prescription characteristics, diagnosis codes, and prescriber type. RESULTS: From 2017-2021, there was a decrease in the number of unique pediatric patients dispensed an initiate prescription, the number of initiate pediatric opioid prescriptions, and the initiate prescription dosage. Initiate opioid prescriptions were primarily related to dental-related diagnoses, and dentists and oral and maxillofacial (OMF) surgeons comprised the largest category of prescriber type. CONCLUSION: Initiate pediatric opioid prescriptions have decreased in Rhode Island in recent years. However, there remain opportunities to educate prescribers on reducing opioid exposure to vulnerable populations, including the use of alternate analgesics.

Examination of the Accuracy of Existing Overdose Surveillance Systems.

Importance: Health departments have used a variety of methods for overdose surveillance, and the Centers for Disease Control and Prevention (CDC) is implementing a standardized case definition to improve overdose surveillance nationally. The comparative accuracy of the CDC opioid overdose case definition vs existing state opioid overdose surveillance systems is unknown. Objective: To evaluate the accuracy of the CDC opioid overdose case definition and existing Rhode Island Department of Health (RIDOH) state opioid overdose surveillance system. Design, Setting, and Participants: This cross-sectional study of ED opioid overdose visits was conducted at 2 EDs in Providence, Rhode Island, at the state's largest health system from January to May 2021. Electronic health records (EHRs) were reviewed for opioid overdoses identified by the CDC case definition and opioid overdoses reported to the RIDOH state surveillance system. Included patients were those at study EDs whose visit met the CDC case definition, was reported to the state surveillance system, or both. True overdose cases were confirmed by EHR review using a standard case definition; 61 of 460 EHRs (13.3%) were double reviewed to estimate classification accuracy. Data were analyzed from January through May 2021. Main Outcome and Measure: Accurate identification of an opioid overdose was assessed by estimating the positive predictive value of the CDC case definition and state surveillance system using results from the EHR review. Results: Among 460 ED visits that met the CDC opioid overdose case definition, were reported to the RIDOH opioid overdose surveillance system, or both (mean [SD] age, 39.7 [13.5] years; 313 males [68.0%]; 61 Black [13.3%], 308 White [67.0%], and 91 other race [19.8%]; and 97 Hispanic or Latinx [21.1%] among each patient visit), 359 visits (78.0%) were true opioid overdoses. For these visits, the CDC case definition and RIDOH surveillance system agreed that 169 visits (36.7%) were opioid overdoses. Of 318 visits meeting the CDC opioid overdose case definition, 289 visits (90.8%; 95% CI, 87.2%-93.8%) were true opioid overdoses. Of 311 visits reported to the RIDOH surveillance system, 235 visits (75.6%; 95% CI, 70.4%-80.2%) were true opioid overdoses. Conclusions and Relevance: This cross-sectional study found that the CDC opioid overdose case definition more often identified true opioid overdoses compared with the Rhode Island overdose surveillance system. This finding suggests that using the CDC case definition for opioid overdose surveillance may be associated with improved data efficiency and uniformity.

Benzodiazepine Dose Intensity among Patients Concurrently Prescribed Buprenorphine in Rhode Island.

Background: The concomitant use of buprenorphine and benzodiazepines has been linked to patient fatalities, with greater risk occurring with higher doses of benzodiazepines. We assessed benzodiazepine dose intensity among patients who were concurrently prescribed buprenorphine, as compared with patients prescribed benzodiazepines who were not receiving buprenorphine. Methods: We conducted a cross-sectional analysis of adult patients who received at least a 30-day supply of benzodiazepines during 2018, using data from the Rhode Island (RI) Prescription Drug Monitoring Program. Mean daily diazepam milligram equivalents (DME) were calculated overall and according to patient sex, age group, payment type, and RI county. Multivariable logistic regression analyses were conducted to assess the odds of higher-dose benzodiazepine utilization among patients with concurrent use of buprenorphine, as compared with patients not prescribed buprenorphine, adjusting for patient demographics. Results: Compared to patients prescribed benzodiazepines who were not receiving buprenorphine, those with concurrent buprenorphine utilization had a significantly higher mean DME/day (19.22, 95% CI: 18.70-19.74; vs 10.94, 95% CI: 10.93-10.95; p < 0.001). Patients who were prescribed benzodiazepines with concurrent utilization of buprenorphine also had a comparatively higher odds of a DME/day ≥15 (aOR: 2.86, 95% CI: 2.63-3.10), ≥20 DME/day (aOR: 2.98, 95% CI: 2.75-3.24), and ≥25 DME/day (aOR: 2.99, 95% CI: 2.65-3.18). Conclusion: Compared to patients prescribed benzodiazepines for at least 30 days who were not receiving buprenorphine, patients concurrently utilizing benzodiazepines and buprenorphine had more than twice the odds of higher dose benzodiazepine utilization. Future studies are needed to assess the relationship between benzodiazepine dose intensity, overdose outcomes, and treatment retention among patients receiving buprenorphine.

Receipt of Prescription Psychostimulants and Stimulant-Involved Fatal Overdose: A Population-Based Case-Control Study.

Background: Rates of psychostimulant use, misuse, and hospitalization have increased markedly over the past decade. The objective of this study was to estimate the association between receipt of a psychostimulant prescription in the past year and fatal, unintentional psychostimulant-involved overdose. Methods: We conducted a population-based case-control study using linked, state-level databases from the Rhode Island Department of Health. Cases were defined as Rhode Island residents who experienced a fatal, unintentional drug overdose involving a psychostimulant, and controls included non-psychostimulant involved fatal overdoses occurring between May 1, 2017 and May 31, 2020 The primary exposure of interest was receipt of a psychostimulant prescription within 12 months prior to death, ascertained through linkage to the state's prescription drug monitoring program. Conditional logistic regression was used to estimate unadjusted and adjusted odds ratios. Results: Of 894 eligible overdose fatalities, the majority were white/non-Hispanic (72%), mean age was 43 years, and most resided in Providence County (69%). A total of 39 (4%) involved a psychostimulant. After adjusting for year of death and matching by sex, age, and county of residence, cases had 4.1 (95% confidence interval: 1.6, 10.6) times the odds of receiving a prescription stimulant in the past year compared to controls. Conclusions: Our findings suggest that there is a strong, positive association between prescription psychostimulant receipt and psychostimulant-involved fatal overdose. In response to an evolving polysubstance use landscape, current harm reductions measures, including naloxone training, fentanyl test strip distribution, and overdose education, should be expanded to include patients who receive psychostimulant prescriptions.

Translating Predictive Analytics for Public Health Practice: A Case Study of Overdose Prevention in Rhode Island.

Prior applications of machine learning to population health have relied on conventional model assessment criteria, limiting the utility of models as decision support tools for public health practitioners. To facilitate practitioners' use of machine learning as a decision support tool for area-level intervention, we developed and applied 4 practice-based predictive model evaluation criteria (implementation capacity, preventive potential, health equity, and jurisdictional practicalities). We used a case study of overdose prevention in Rhode Island to illustrate how these criteria could inform public health practice and health equity promotion. We used Rhode Island overdose mortality records from January 2016-June 2020 (n = 1,408) and neighborhood-level US Census data. We employed 2 disparate machine learning models, Gaussian process and random forest, to illustrate the comparative utility of our criteria to guide interventions. Our models predicted 7.5%-36.4% of overdose deaths during the test period, illustrating the preventive potential of overdose interventions assuming 5%-20% statewide implementation capacities for neighborhood-level resource deployment. We describe the health equity implications of use of predictive modeling to guide interventions along the lines of urbanicity, racial/ethnic composition, and poverty. We then discuss considerations to complement predictive model evaluation criteria and inform the prevention and mitigation of spatially dynamic public health problems across the breadth of practice. This article is part of a Special Collection on Mental Health.