Faculty Bibliography

BACKGROUND AND AIMS/OBJECTIVE:In light of the accelerating drug overdose epidemic in North America, new strategies are needed to identify communities most at risk to prioritize geographically the existing public health resources (e.g. street outreach, naloxone distribution efforts). We aimed to develop PROVIDENT (Preventing Overdose using Information and Data from the Environment), a machine learning-based forecasting tool to predict future overdose deaths at the census block group (i.e. neighbourhood) level. DESIGN/METHODS:Randomized, population-based, community intervention trial. SETTING/METHODS:Rhode Island, USA. PARTICIPANTS/METHODS:All people who reside in Rhode Island during the study period may contribute data to either the model or the trial outcomes. INTERVENTION/METHODS:Each of the state's 39 municipalities will be randomized to the intervention (PROVIDENT) or comparator condition. An interactive, web-based tool will be developed to visualize the PROVIDENT model predictions. Municipalities assigned to the treatment arm will receive neighbourhood risk predictions from the PROVIDENT model, and state agencies and community-based organizations will direct resources to neighbourhoods identified as high risk. Municipalities assigned to the control arm will continue to receive surveillance information and overdose prevention resources, but they will not receive neighbourhood risk predictions. MEASUREMENTS/METHODS:The primary outcome is the municipal-level rate of fatal and non-fatal drug overdoses. Fatal overdoses will be defined as unintentional drug-related death; non-fatal overdoses will be defined as an emergency department visit for a suspected overdose reported through the state's syndromic surveillance system. Intervention efficacy will be assessed using Poisson or negative binomial regression to estimate incidence rate ratios comparing fatal and non-fatal overdose rates in treatment vs. control municipalities. COMMENTS/CONCLUSIONS:The findings will inform the utility of predictive modelling as a tool to improve public health decision-making and inform resource allocation to communities that should be prioritized for prevention, treatment, recovery and overdose rescue services.

In the United States, combined stimulant/opioid overdose mortality has risen dramatically over the last decade. These increases may particularly affect non-Hispanic Black and Hispanic populations. We used death certificate data from the US National Center for Health Statistics (2007-2019) to compare state-level trends in overdose mortality due to opioids in combination with 1) cocaine and 2) methamphetamine and other stimulants (MOS) across racial/ethnic groups (non-Hispanic White, non-Hispanic Black, Hispanic, and non-Hispanic Asian American/Pacific Islander). To avoid unstable estimates from small samples, we employed principles of small area estimation and a Bayesian hierarchical model, enabling information-sharing across groups. Black Americans experienced severe and worsening mortality due to opioids in combination with both cocaine and MOS, particularly in eastern states. Cocaine/opioid mortality increased 575% among Black people versus 184% in White people (Black, 0.60 to 4.05 per 100,000; White, 0.49 to 1.39 per 100,000). MOS/opioid mortality rose 16,200% in Black people versus 3,200% in White people (Black, 0.01 to 1.63 per 100,000; White, 0.09 to 2.97 per 100,000). Cocaine/opioid overdose mortality rose sharply among Hispanic and Asian Americans. State-group heterogeneity highlighted the importance of data disaggregation and methods to address small sample sizes. Research to understand the drivers of these trends and expanded efforts to address them are needed, particularly in minoritized groups.

Predictors of opioid overdose death in neighborhoods are important to identify, both to understand characteristics of high-risk areas and to prioritize limited prevention and intervention resources. Machine learning methods could serve as a valuable tool for identifying neighborhood-level predictors. We examined statewide data on opioid overdose death from Rhode Island (log-transformed rates for 2016-2019) and 203 covariates from the American Community Survey for 742 US Census block groups. The analysis included a least absolute shrinkage and selection operator (LASSO) algorithm followed by variable importance rankings from a random forest algorithm. We employed double cross-validation, with 10 folds in the inner loop to train the model and 4 outer folds to assess predictive performance. The ranked variables included a range of dimensions of socioeconomic status, including education, income and wealth, residential stability, race/ethnicity, social isolation, and occupational status. The R2 value of the model on testing data was 0.17. While many predictors of overdose death were in established domains (education, income, occupation), we also identified novel domains (residential stability, racial/ethnic distribution, and social isolation). Predictive modeling with machine learning can identify new neighborhood-level predictors of overdose in the continually evolving opioid epidemic and anticipate the neighborhoods at high risk of overdose mortality.

ABSTRACT/UNASSIGNED:Chronic pain is associated with mental and physical health difficulties and is prevalent among veterans. Cannabis has been put forth as a treatment for chronic pain, and changes in laws, attitudes, and use patterns have occurred over the last two decades. Differences in prevalence of non-medical cannabis use and cannabis use disorder (CUD) were examined across two groups: veterans/non-veterans and those reporting/not reporting recent pain. Data from the National Epidemiologic Survey on Alcohol and Related Conditions-III (2012-2013; n=36,309) were analyzed using logistic regression. Prevalence Differences (PD) for three cannabis outcomes: (1) past-year non-medical cannabis use, (2) frequent (≥3 times a week) non-medical use, and (3) DSM-5 CUD were estimated for those reporting recent moderate-severe pain (veterans/non-veterans), and veterans reporting/not reporting recent pain. Difference in differences were calculated to investigate prevalence differences on outcomes associated with residence in a state with medical cannabis laws (MCLs). Associations between physical and mental health and cannabis variables were tested. Results indicated that the prevalence of recent pain was greater among veterans (PD=7.25%, 95% CI [4.90, 9.60]). Among veterans, the prevalence of frequent cannabis use was greater among those with pain (PD=1.92%, 98% CI [0.21, 3.63]), and, among veterans residing in a state with MCLs, the prevalence of CUD was greater among those reporting recent pain (PD=3.88%, 98% CI [0.36, 7.39]). Findings failed to support the hypothesis that cannabis use improves mental or physical health for veterans with pain. Providers treating veterans with pain in MCL states should monitor such patients closely for CUD.

Over the past decade, US adolescents' depressive symptoms have increased, and changing religious beliefs and service attendance may be contributing factors. We examined the contribution of religious factors to depressive symptoms among 417,540 US adolescents (grades: 8, 10, 12), years:1991-2019, in survey-weighted logistic regressions. Among adolescents who felt religion was personally important, those who never attended services had 2.23 times higher odds of reporting depressive symptoms compared to peers attending weekly. Among adolescents who did not feel that religion was important, the pattern was reversed. Among adolescents, concordance between importance of religion and religious service attendance may lower risk of depressive symptoms. Overall, we estimate that depressive symptom trends would be 28.2% lower if religious factors had remained at 1991 levels.

The opioid overdose crisis is driven by an intersecting set of social, structural, and economic forces. Simulation models offer a tool to help us understand and address this complex, dynamic, and nonlinear social phenomenon. We conducted a systematic review of the literature on simulation models of opioid use and overdose up to September 2019. We extracted modeling types, target populations, interventions, and findings. Further, we created a database of model parameters used for model calibration, and evaluated study transparency and reproducibility. Of the 1,398 articles screened, we identified 88 eligible articles. The most frequent types of models were compartmental (36%), Markov (20%), system dynamics (16%), and Agent-Based models (16%). Over a third evaluated intervention cost-effectiveness (40%), and another third (39%) focused on treatment and harm reduction services for people with opioid use disorder (OUD). More than half (61%) discussed calibrating their models to empirical data, and 31% discussed validation approaches used in their modeling process. From the 63 studies that provided model parameters, we extracted the data sources on opioid use, OUD, OUD treatment, cessation/relapse, emergency medical services, and mortality parameters. This database offers a tool that future modelers can use to identify potential model inputs and evaluate comparability of their models to prior work. Future applications of simulation models to this field should actively tackle key methodological challenges, including the potential for bias in the choice of parameter inputs, investment in model calibration and validation, and transparency in the assumptions and mechanics of simulation models to facilitate reproducibility.

Agent-based modeling and G-computation can both be used to estimate impacts of intervening on complex systems. We explored each modeling approach within an applied example: interventions to reduce posttraumatic stress disorder (PTSD). We used data from a cohort of 2,282 adults representative of the adult population of the New York City metropolitan area from 2002-2006, of whom 16.3% developed PTSD over their lifetimes. We built four models: G-computation, an agent-based model with no between-agent interactions, an agent-based model with violent interaction dynamics, and an agent-based model with neighborhood dynamics. Three interventions were tested: reducing violent victimization by 1) 37.2% (real-world reduction), 2) 100%, and 3) supplementing the income of 20% of lower-income participants. The G-computation model estimated population-level PTSD risk reductions of 0.12% (95% CI: -0.16, 0.29), 0.28% (95% CI: -0.30, 0.70), and 1.55% (95% CI: 0.40, 2.12), respectively. The agent-based model with no interactions replicated the findings from G-computation. Introduction of interaction dynamics modestly decreased estimated intervention effects (income supplement risk reduction dropped to 1.47%), whereas introduction of neighborhood dynamics modestly increased effectiveness (income supplement risk reduction increased to 1.58%). As compared with G-computation, agent-based modeling permitted deeper exploration of complex systems dynamics at the cost of further assumptions.

OBJECTIVE:Availability of medications for opioid use disorder (MOUD) remains sparse. To date, there has been no national, state-by-state comparison of patient MOUD utilization relative to treatment availability and burden of overdose deaths. We aimed to quantify, for each state, the number of MOUD patients relative to (1) office-based buprenorphine providers and opioid treatment programs (OTPs) and (2) overdose deaths. METHODS:We conducted a spatial analysis of patients receiving MOUD from OTPs or buprenorphine providers in March 2017 across all 50 states and Washington, DC. For each state, we calculated the number of patients receiving MOUD from OTPs and buprenorphine prescriptions, relative to available OTPs and buprenorphine providers; as well as ratios of number of patients receiving MOUD relative to overdose deaths. RESULTS:In March 2017, 942,368 patients attended an OTP (410,288) or received a buprenorphine prescription (486,318). Patient to OTP ratio was highest in West Virginia, Delaware, Washington, DC, New Jersey, New Hampshire, Connecticut and Ohio, ranging from 91 to 193 patients per OTP in the first quintile to 430 to 648 in the fifth. Patient to buprenorphine provider ratio was highest in Kentucky and West Virginia, ranging from 3 to 7 patients per provider in the first quintile to 19 to 28 in the fifth. Median MOUD patients per overdose death was 21 (IQR:14.9-28.2). Of high overdose states, Washington, DC, New Jersey, and Ohio had the smallest number of patients on MOUD relative to deaths. CONCLUSIONS:High patient volume relative to treatment availability in overdose-burdened areas may indicate strain on MOUD providers and OTPs. Promoting greater utilization while expanding MOUD providers and programs is critical.

BACKGROUND:Firearm violence remains a persistent public health threat. Comparing the impact of targeted high-risk versus population-based approaches to prevention may point to efficient and efficacious interventions. We used agent-based modeling to conduct a hypothetical experiment contrasting the impact of high-risk (disqualification) and population-based (price increase) approaches on firearm homicide in New York City (NYC). METHODS:We simulated 800,000 agents reflecting a 15% sample of the adult population of NYC. Three groups were considered and disqualified from all firearm ownership for five years, grouped based on prevalence: low prevalence (psychiatric hospitalization, alcohol-related misdemeanor and felony convictions, 0.23%); moderate prevalence (drug misdemeanor convictions, domestic violence restraining orders, 1.03%); and high prevalence (all other felony/misdemeanor convictions, 2.30%). Population-level firearm ownership was impacted by increasing the price of firearms, assuming 1% price elasticity. RESULTS:In this hypothetical scenario, to reduce firearm homicide by 5% in NYC, 25% of the moderate prevalence group, or 12% of the high prevalence group needed to be effectively disqualified; even when all of the low prevalence group was disqualified, homicide did not decrease by 5%. An 18% increase in price similarly reduced firearm homicide by 5.37% (95% CI 4.43-6.31%). Firearm homicide declined monotonically as the proportion of disqualified individuals increased and/or price increased. A combined intervention that both increased price and effectively disqualified "high-risk" groups achieved approximately double the reduction in homicide as any one intervention alone. Increasing illegal firearm ownership by 20%, a hypothetical response to price increases, did not meaningfully change results. CONCLUSION:A key takeaway of our study is that adopting high-risk versus population-based approaches should not be an "either-or" question. When individual risk is variable and diffuse in the population, "high-risk approaches" to firearm violence need to focus on relatively prevalent groups and be highly efficacious in disarming people at elevated risk to achieve meaningful reductions in firearm homicide, though countering issues of social justice and stigma should be carefully considered. Similar reductions can be achieved with population-based approaches, such as price increases, albeit with fewer such countering issues.