Pushing the boundaries of prediction to address the opioid crisis

In The Lancet Public Health, Charles Marks and colleagues1Marks C Abramovitz D Donelly CA et al.Identifying countries at risk of high overdose mortality burden during the emerging fentanyl epidemic in the USA: a predictive statistical modelling study.Lancet Pub Health. 2021; (published online June 9.)https://doi.org/10.1016/S2468-2667(21)00080-3Summary Full Text Full Text PDF Scopus (5) Google Scholar applied a statistical modelling approach to predict county-level overdose deaths between 2013 and 2018 in the USA using previous-year measures of health-care availability, drug markets, socioeconomic characteristics, and geographical spread of overdose events. To evaluate the performance of their negative binomial model, the authors ranked counties in terms of their predicted number of overdose deaths and examined how many counties were predicted to be in the top decile of overdose deaths. These rankings were compared with those predicted by a simple change rate (referred to as the benchmark) based on overdose deaths from the previous 2 years. Overall, the authors found that the model predicted 42–57% of the counties in the top decile for overdose deaths, an improvement over the 29–43% of counties identified by the benchmark alone. Notably, the model was far more capable of predicting counties that would newly join the top decile in a given year (22–26% of counties) than the benchmark was (3–5%). There are growing calls for timely surveillance strategies that can facilitate public health responses to the opioid crisis in the USA. Researchers in some states, such as Michigan, have developed dashboards to rapidly disseminate data on county-level overdoses and overdose deaths to public health officials.2Goldstick J Ballesteros A Flannagan C Roche J Schmidt C Cunningham RM Michigan system for opioid overdose surveillance.Inj Prev. 2021; (published online Jan 4.)http://dx.doi.org/10.1136/injuryprev-2020-043882Crossref PubMed Scopus (3) Google Scholar Despite these and other models, there is a shortage of predictive strategies that can identify high-risk jurisdictions in advance and inform targeted resource delivery. Marks and colleagues' findings show the limits of using within-county surveillance to inform predictions. The most consistently used predictor across bootstrapped iterations (81% of simulations) was geographical proximity to overdoses in other counties, referred to as overdose gravity, highlighting the geographical clustering of fatal overdose events. This finding aligns with broader calls to understand and react to contextual factors driving overdose deaths, including the role of regional drug supply and structural determinants of health.3Mars SG Rosenblum D Ciccarone D Illicit fentanyls in the opioid street market: desired or imposed?.Addiction. 2019; 114: 774-780Crossref PubMed Scopus (66) Google Scholar, 4El-Bassel N Shoptaw S Goodman-Meza D Ono H Addressing long overdue social and structural determinants of the opioid epidemic.Drug Alcohol Depend. 2021; 222108679Crossref PubMed Scopus (12) Google Scholar Marks and colleagues' approach represents a meaningful advancement in predictive modelling over simple jurisdictional change rates. However, there is much room to improve predictions, evidenced by the ability of the model to identify only 57% of counties in the top decile for overdose deaths. The authors note that they could not look at the escalation of specific drug-involved overdose deaths due to inconsistent reporting of specified overdose deaths, the severity of which has been reported previously.5Lowder E Ray B Huynh P Ballew A Watson DP Identifying unreported opioid deaths through toxicology data and vital records linkage: case study in Marion County, Indiana, 2011–2016.Am J Public Health. 2018; 108: 1682-1687Crossref PubMed Scopus (16) Google Scholar Opioids are increasingly used in combination with other substances.6Cicero TJ Ellis MS Kasper ZA Polysubstance use: a broader understanding of substance use during the opioid crisis.Am J Public Health. 2019; 110: 244-250Crossref PubMed Scopus (89) Google Scholar Predictive strategies must account for polysubstance use in opioid-involved overdose deaths and related supply-side drivers of substance availability, which are thought to be key considerations in the role of fentanyl and its analogs in overdose deaths.3Mars SG Rosenblum D Ciccarone D Illicit fentanyls in the opioid street market: desired or imposed?.Addiction. 2019; 114: 774-780Crossref PubMed Scopus (66) Google Scholar Policy changes can be time-intensive and difficult to measure, but capturing these changes—particularly in national datasets—might be crucial to improved predictions. The role of policy and broader historical context is particularly important in the context of the COVID-19 pandemic. Current evidence suggests the scale of the pandemic and lockdown procedures might have exacerbated overdose events.7Rosenbaum J Lucas N Zandrow G et al.Impact of a shelter-in-place order during the COVID-19 pandemic on the incidence of opioid overdoses.Am J Emerg Med. 2021; 41: 51-54Summary Full Text Full Text PDF PubMed Scopus (11) Google Scholar By contrast, increasingly accessible treatment during COVID-19 could have the potential to reduce overdose deaths.8Haley DF Saitz R The opioid epidemic during the COVID-19 pandemic.JAMA. 2020; 324: 1615-1617Crossref PubMed Scopus (65) Google Scholar Community availability of naloxone and interventions at the first responder or emergency department level might also reduce fatal overdose events, in addition to treatment availability broadly (eg, buprenorphine and other opioid antagonists). The authors note unavailable county-level treatment indicators as a limitation of their modelling approach. Public health interventions to address opioid use and its associated harms operate at multiple levels,9Alexandridis AA Doe-Simkins M Scott G A case for experiential expertise in opioid overdose surveillance.Am J Public Health. 2020; 110: 505-507Crossref PubMed Scopus (1) Google Scholar and there is emerging evidence that even broad policy reform can improve population-level opioid outcomes.10Ansari B Tote KM Rosenberg ES Martin EG A rapid review of the impact of systems-level policies and interventions on population-level outcomes related to the opioid epidemic, United States and Canada, 2014–2018.Public Health Rep. 2020; 135: 100S-127SCrossref PubMed Scopus (10) Google Scholar As public health and public policy responses to the opioid crisis evolve, comprehensive and timely documentation of these initiatives would enable more rigorous research by public health researchers. Marks and colleagues' modelling study provides another piece of evidence that the opioid crisis—perhaps now more accurately termed the overdose crisis—is evolving rapidly. There is little doubt that this evolution will be exacerbated by the changing social climate of the COVID-19 pandemic. Yet, the margin of error for predicting overdose deaths suggests there is much work to be done in improving predictions. Critically, advancing research on predictive modelling requires sufficient data infrastructure for timely reporting of overdose-related events and other contextual factors (eg, policy changes, treatment availability, and supply-side measures) that affect these events. It is promising that the accuracy of Marks and colleagues' model improved in 2017 and 2018 when predicting counties in the top decile. Their approach, and accompanying dashboard, shows that it is possible to push the boundaries of risk prediction and underscores the need for further efforts of this kind. I declare no competing interests. Identifying counties at risk of high overdose mortality burden during the emerging fentanyl epidemic in the USA: a predictive statistical modelling studyOur model shows that a regression approach can effectively predict county-level overdose death rates and serve as a risk assessment tool to identify future high mortality counties throughout an emerging drug use epidemic. Full-Text PDF Open Access

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.