Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Dec 28.
Published in final edited form as: Am J Drug Alcohol Abuse. 2021 Jun 9;47(6):711–721. doi: 10.1080/00952990.2021.1929273

Fatal opioid overdoses in the U.S. declined more than reported between 2017 and 2018

Andrew Boslett 1,2, Alina Denham 1, Elaine Hill 1
PMCID: PMC9797028  NIHMSID: NIHMS1855448  PMID: 34107224

Abstract

Background:

In U.S. death records, many drug overdoses do not have classified drug involvement, which challenges surveillance of opioid overdoses across time and space.

Objective:

To estimate the 2017–2018 change in opioid overdose deaths that accounts for probable opioid involvement in unclassified drug overdose deaths.

Methods:

In this retrospective design study, data on all drug overdose decedents from 2017–2018 in the U.S. were used to calculate the year-to-year change in known opioid overdoses, predict opioid involvement in unclassified drug overdoses, and estimate the year-to-year change in corrected opioid overdoses, which include both known and predicted opioid deaths. We used the Multiple Cause of Death (MCOD) data from CDC.

Results:

We estimated that the decrease in the age-adjusted opioid overdose death rate from 2017–2018 was 7.0%. There is a striking variation across states. Age-adjusted opioid overdose death rates decreased by 9.9% in Ohio and more than 5.0% in other Appalachian states (Pennsylvania, West Virginia, Kentucky), while they increased by 6.8% in Delaware.

Conclusions:

Our models suggest that opioid overdose-related mortality declined from 2017 to 2018 at a higher rate than reported (7.0% versus than the reported 2.0%), potentially indicating that clinical efforts and federal, state, and local government policies designed to control the epidemic have been effective in most states. Our local area estimates can be used by researchers, policy-makers and public health officials to assess effectiveness of state policies and interventions in smaller jurisdictions implemented in response to the crisis.

Keywords: Drug overdose epidemic, Opioid overdoses, Machine learning

Introduction

The opioid epidemic took close to 450,000 lives in 1999–2018,(1) but 2018 saw a turning point in the fight against opioid mortality. According to the Centers for Disease Control and Prevention (CDC), the rate of fatal opioid overdoses declined by 2.0% between 2017 and 2018 (from 14.9 to 14.6 per 100,000 people).(2)

Yet, this estimate is based solely on those drug overdoses explicitly classified as opioid-involved. Many drug overdoses do not have classified drug involvement. Prior research indicates that approximately 20% of drug overdoses from 1999 to 2016 did not have classified drug involvement (14.6% in 2016).(3) Statistical extrapolations suggest that approximately 70% of these unclassified drug overdoses from 1999–2016 involve opioids, substantially increasing the reported death toll in the opioid epidemic by approximately 100,000 over this time period.(3,4) Accounting for unclassified drug overdoses involving opioids may alter the previously reported decline in fatal opioid overdoses from 2017 to 2018.

To address this gap and to inform further opioid overdose prevention programs and policies, we provide new estimates of fatal opioid overdoses in the U.S. in 2017 and 2018 and compute new estimates of changes in opioid overdose rates between 2017 and 2018. To highlight changes in specific localities, we also provide sub-national estimates of opioid overdose rates and changes over time at the state and county levels.

We expect that this year-to-year change in opioid overdoses will depend on (1) the difference in the predicted levels of opioid involvement in drug overdoses with unknown drug involvement in 2017 and 2018; and (2) the change in the overall proportion of total drug overdoses with unknown drug involvement. The second factor is not trivial; there has been a declining trend in the proportion of drug overdoses with no drug involvement since 2008.(3,4) Therefore, whether and by how much fatal opioid overdoses declined from 2017 to 2018, when unclassified drug overdoses are accounted for, remains an empirical question. This study aims to correct the reported change from 2017 to 2018 using and building on the methods that have been developed and implemented in prior research on unclassified drug overdoses.(3) We do not examine what may have caused our estimate in the change in opioid overdose-related mortality from 2017 to 2018 or why unclassified drug overdoses change over time.

Materials and Methods

Data

Our study relied on restricted access Multiple Cause of Death (MCOD) data from CDC’s National Center for Health Statistics (NCHS). These data contain detailed death and decedent information for all deaths in the U.S. We extracted all deaths that occurred in 2017–2018 with a drug poisoning underlying cause of death, using ICD-10 codes X40–44 (unintentional drug poisoning), X60–64 (self-harm, or suicidal drug poisoning), X85 (assault/homicide drug poisoning), and Y10-Y14 (drug poisoning with undetermined intent), following the CDC definition.(2)

The MCOD data include up to 20 contributing causes of death to the underlying cause of death. We identified drug involvement for all drug overdoses with at least one record axis contributing cause of death indicating involvement of a specific drug. Opioid overdoses were defined as drug overdoses containing at least one ICD-10 code indicating presence of opioids: T40.0 (opium), T40.1 (heroin), T40.2 (other opioids), T40.3 (methadone), T40.4 (other synthetic narcotics), and T40.6 (other and unspecified narcotics). Non-opioid overdoses were defined as drug overdoses with at least one of ICD-10 codes T36–39, T40.5, T40.7–9, and T41–50.8 and no ICD-10 code indicating presence of opioids. We create a binary variable equal to 1 if the drug overdose involved opioids and 0 otherwise, based on the occurrence of a drug specified in the record. All drug overdoses with an ICD-10 code T50.9 as one of the contributing causes of death, without any of the codes used to identify specific drug involvement, were defined as unclassified drug overdoses. This code indicates “poisoning by, adverse effect of and overdosing of other and unspecified drugs, medicaments and biological substances.”(5)

In addition to derived drug overdose-related characteristics, the MCOD data files include decedent demographic and social characteristics, including age, sex, race, ethnicity, marital status, education level (i.e., less than high school degree, high school degree, some college, bachelor’s degree or more, or unknown), and death characteristics, including day of week and month of death, and death location (i.e., inpatient hospital, outpatient hospital, dead-on-arrival to hospital, home, nursing home and unknown).

We extracted the 100 most commonly occurring contributing causes of death by year in 2017 and 2018. We excluded those contributing causes of death that were involved in the specification of drug involvement (i.e., ICD-10 codes T36–50), as well as those that were used to identify drug overdoses (i.e., ICD-10 codes X40–44, X60–64, X85, Y10-Y14). For each contributing cause of death, we created a binary variable equal to 1 if the corresponding ICD-10 code was listed as a contributing cause in any of the record axis variables, 0 otherwise.

We used CDC’s Surveillance, Epidemiology, and End Results (SEER) U.S. Population Data – 2017–2018 to calculate opioid overdose rates per 100,000 by geographical area.(6)

Statistical Analyses

First, we calculated changes in classified (known) opioid overdoses from 2017 to 2018.

Because unclassified drug overdoses may in fact involve opioids, the corrected change in opioid overdoses depends on the proportion of unclassified drug overdoses that are predicted to have involved opioids. Before predicting opioid involvement in unclassified drug overdoses, we created a heat-map of the change in opioid overdose rates between 2017 and 2018 that would occur under a range of proportions of opioid involvement in unclassified drug overdoses. This was done on a 0.05 increment basis from 0 (no opioid involvement) to 1 (all overdoses were opioid-involved). This method provides a range of possible year-to-year changes in opioid overdoses from 2017 to 2018.

To predict opioid involvement in unclassified drug overdoses in 2017 and 2018, we used logistic regression and random forests, a commonly used machine learning technique (7). We modeled opioid involvement as a function of decedent characteristics and binary variables indicating the presence of the most common contributing causes of death. We used these estimated models to generate estimates of the probability of opioid involvement for each unclassified drug overdose. We then summed these decedent-level estimates, along with the known number of opioid overdoses, to generate new estimates of the number of opioid overdoses by year. Finally, we calculated the change in the estimated number of opioid overdoses on a year-to-year basis from 2017 to 2018. In addition to changes in the number of opioid overdoses by year, we also calculated nationwide, state- and county-specific changes in age-adjusted opioid overdose rates using 2000 U.S. standard population data for age-adjustment from SEER.(8)

We based our model selection on out-of-sample accuracy measures between alternative models on two margins: (1) by statistical method, including logistic regression and random forests; and (2) by sets of independent variables, differentiated by the number of contributing causes of death included (i.e., none; the top-10, top-20, top-50, and top-100 most common contributing causes of death in drug overdoses). All models control for a set of decedent characteristics, including binary indicators of age, sex, race, ethnicity, marital status, education level, day of week and month of death, and death location. We selected the top-N most common contributing causes of death based on counts of occurrences of each contributing cause in all drug overdoses in each year. These contributing causes of death were included in the models as binary variables (1 if an ICD-10 code was recorded as a contributing cause of death in any of the record axis variables, 0 otherwise). In Tables A1 and A2 in the Supplementary Information, we provide the top-100 contributing causes of death in drug overdoses in 2017 and 2018. We estimated these models on an annual basis in order to allow the effects of variables included in the models to vary across time. We generated a random split of the observation set into a training set (80%), which we used to parameterize the models, and a test set (20%), which we used to measure out-of-sample predictive accuracy. For each year’s model, we excluded all decedent characteristics and contributing causes of death in which there is no variation within the training group.

We assessed the accuracy of our models using multiple out-of-sample classification accuracy measures, including total predictive accuracy, receiver operating curves,(9) and Matthews correlation coefficient.(10) Additionally, we tested the performance of these alternative models in the presence of observations balanced between opioid-involved and non-opioid-involved drug overdoses. More details are included in the Supplementary Information.

To calculate the number of drug overdoses in each year, we summed all predictions of the probability of opioid involvement from our statistical models by geographical area and year, based on the recent literature in predicting opioid involvement in unclassified drug overdoses.(3,4)

We conducted all statistical analysis in the R programming language. For statistical analyses, we relied on base functionality of R but relied on the tidyverse collection of packages for data visualization. We used the randomForest package for random forest estimation.

Results

Descriptive statistics and trends

There were 47,600 classified opioid overdose deaths in 2017 and 46,802 in 2018, a 1.7% decrease. The percent of unclassified drug overdoses declined from 12.3% (8,663) in 2017 to 8.0% (5,396) in 2018. The scale of this decline makes intertemporal comparisons in the numbers of classified opioid or non-opioid overdoses challenging without information on the potential degree of opioid involvement in unclassified drug overdoses.

In Figure A2 in the Supplementary Information, we show a heat-map of year-to-year changes in opioid overdoses from 2017 to 2018 under a number of alternative scenarios of actual opioid involvement in unclassified drug overdoses. We show a square at the point where the proportion of opioid involvement in unclassified drug overdoses is the same as that found in drug overdoses with known drug involvement (i.e., 0.773 and 0.755 in 2017 and 2018, respectively). With these respective proportions, the number of opioid overdoses declined by 6.7% between 2017 and 2018. In other words, if one estimated opioid involvement in unclassified drug overdoses by simply assuming that the proportion of opioid overdoses among unclassified overdoses is the same as the proportion of known opioid overdoses among classified overdoses, the estimated change from 2017 to 2018 would be 6.7%.

Corrected estimates of opioid overdoses in 2017 and 2018

Figure 2 shows new, corrected estimates of the total number of opioid overdoses for 2017 and 2018 against classified opioid overdoses and total drug overdoses. We estimate that there were 6,720 additional opioid overdoses in 2017 and 3,964 additional opioid overdoses in 2018 that were previously unclassified. Together with classified opioid overdoses, these estimates imply that there were 54,320 and 50,766 opioid overdoses in 2017 and 2018, respectively. These new estimates correspond to a 6.5% decline in the number of opioid overdoses from 2017 to 2018.

Figure 2:

Figure 2:

Open in a new tab

Trends in known opioid overdoses and corrected opioid overdoses, 1999–2018.

Notes: In this figure, we display total drug overdoses, classified opioid overdoses, and our corrected estimates of opioid overdoses, which include both classified opioid overdoses and probabilistic estimates of the number of opioid overdoses within drug overdoses with no classified drug involvement.

These corrected estimates are from logistic regression models of opioid involvement as functions of decedent characteristics and binary indicators of the top-50 most frequently occurring contributing causes of death. Our choice of preferred specification is determined by a number of factors described in the Supplementary Information.

Using our new estimates of the number of opioid overdoses, we calculated that the age-adjusted opioid overdose rate declined by 7.0% from 16.94 to 15.75 per 100,000 people between 2017 and 2018. This decline is substantially larger than the 2.2% decline in age-adjusted opioid overdose rates using only classified opioid overdoses (14.83 and 14.51 per 100,000 in 2017 and 2018, respectively).

Area-level estimates of trends in opioid overdose rates between 2017 and 2018

Figure 3 displays differences in state-level, age-adjusted opioid overdose rates in 2018, between known opioid overdoses and our new, corrected estimates of opioid overdoses. Louisiana, Pennsylvania, Alabama, Arkansas, Indiana, and Florida show the largest absolute differences between known and corrected opioid overdose rates. This same group holds for differences in state-level opioid overdose rates in 2017, along with Delaware and New Jersey (Figure A11 in the Supplementary Information).

Figure 3:

Figure 3:

Open in a new tab

Differences in state opioid overdose rates between those generated using only known opioid overdoses versus corrected estimates of total opioid overdoses, 2018.

Notes: In this figure, we provide estimates of the opioid overdose rate by state in 2018. In light red (data point on the left), we show the opioid overdose rate as calculated using only classified opioid overdoses. In dark red (data point on the right), we show the opioid overdose rate as calculated using both classified opioid overdoses and our estimates of opioid involvement in unidentified drug overdoses. We calculate opioid overdose rates per 100,000 people using state population totals from the Centers for Disease Control and Prevention’s (CDC) Surveillance, Epidemiology, and End Results (SEER) U.S. Population Data – 1999–2018.(6)

Figure 4 shows estimates of the change in state-level, age-adjusted opioid overdose rates from 2017 to 2018 using only our new, corrected estimates of total opioid overdoses. Most states have experienced declines in opioid overdose rates from 2017 to 2018. The largest declines were found in states that have had the highest opioid overdose rates, including West Virginia, Pennsylvania, and Ohio, as well as the District of Columbia. Some states have experienced increased opioid overdose rates from 2017 to 2018. These include Delaware, New Jersey, and Maryland, which now have some of the highest opioid overdose rates in the U.S.

Figure 4:

Figure 4:

Open in a new tab

Changes in state opioid overdose rates using corrected opioid overdoses, 2017– 2018.

Notes: In this figure, we display new estimates of the changes in opioid overdose rates from 2017 to 2018 by state. These estimates are based on our corrected estimates of the total number of opioid overdoses by state in each year. We calculate age-adjusted opioid overdose rates per 100,000 people using state population totals from the Centers for Disease Control and Prevention’s (CDC) Surveillance, Epidemiology, and End Results (SEER) U.S. Population Data – 1999–2018.(6)

In Figure 5, we show county-level changes in opioid overdose rates from 2017 to 2018 using our corrected estimates of total opioid overdoses. We focus on those counties with at least 500,000 people in 2017–2018. In this figure, we show the top-20 and bottom-20 counties ranked by change in opioid overdose rates between the two years. The top three counties with growth in opioid overdose rates between 2017 and 2018 include Baltimore City, MD; Greenville County, SC; and St. Louis County, MO. New Jersey counties make up five of the other top-20 counties with growing opioid overdose rates. Counties with declining opioid overdose rates between 2017 and 2018 include nine in the Appalachian states of Kentucky, Ohio, Pennsylvania, and West Virginia, including the top-3 with declining rates (Montgomery County, OH; Allegheny County, PA; and Summit County, OH). For all counties in the U.S., we show differences in opioid overdose rates by county from 2017–2018 in Figure A12 in the Supplementary Information.

Figure 5:

Figure 5:

Open in a new tab

Change in county opioid overdose rates using corrected opioid overdoses, 2017– 2018.

Notes: In this figure, we display new estimates of the change in opioid overdose rates from 2017 to 2018 by county. These estimates are based on our corrected estimates of the total number of opioid overdoses by state in each year. We show the top-20 and bottom-20 counties in change in unadjusted opioid overdose rates between the two years. We calculate unadjusted opioid overdose rates per 100,000 people using county population totals from the Centers for Disease Control and Prevention’s (CDC) Surveillance, Epidemiology, and End Results (SEER) U.S. Population Data – 1999–2018.(6)

Discussion

This is the first study that corrects opioid mortality counts in 2018 by taking into account unclassified drug overdose death with likely opioid involvement and thus provides corrected estimates of the decline in opioid mortality from 2017 to 2018. We estimated that there was a 6.5% decrease in the number of fatal opioid overdoses and a 7.0% decrease in age-adjusted opioid overdose rates from 2017 to 2018, the latter being substantially larger than the reported 2.0% decline in the rate.(2) Prior work on correcting opioid overdose death estimates has shown that corrected year-to-year growth rate in opioid overdoses was higher than reported in 1999–2013 and lower than reported in 2014–2015.(11) Although our methods and those in prior literature differ to some degree, our study contributes to the understanding of how corrected opioid overdose deaths have trended, in relation to CDC reports.

Prior studies have also shown that unclassified drug overdose rates vary substantially across states, (12) and as expected, our state-level estimates of the corrected 2017–2018 change in opioid overdose rates deviate non-trivially from the reported state-level changes in certain states.(2) Among the thirty-nine states on which CDC reports 2017–2018 changes in opioid overdose death rates, our estimates are the most different for Indiana, Florida, Colorado and Kentucky. In all four states, our corrected estimate of the change indicates a better improvement (i.e., larger decrease in the opioid overdose rate) than reported by the CDC.(2) In addition, our estimates suggest slightly better improvements in Mississippi and Alaska as well.

A decline in opioid overdose deaths may imply that policies and interventions designed to stem the opioid overdose epidemic may have been, in some combination or another, to some degree, effective. Our finding that the decline is larger than reported indicates success at the national level and, necessarily, at least in some states and local jurisdictions. Because a multitude of public health policies and interventions have been implemented over the years, with great variability among states and smaller jurisdictions, it is difficult to credit any one policy for this decline. Nationally, the 2016 CDC guidelines for prescribing opioids for chronic pain(13) were associated with accelerated reductions in high-dose opioid dispensing,(14) which may have impacted opioid mortality. Most policies, however, have been implemented by states.(15–20)

Specifically for states, we found evidence of downward trends in opioid overdose rates in Appalachia, a region with historically high rates of opioid prescribing(21,22) and one of the epicenters of the opioid overdose epidemic.(23,24) At the same time, we observed that certain areas along the northeastern seaboard of the U.S. have experienced increased opioid overdose rates between 2017 and 2018.

This variation in changes in opioid overdose levels over time likely reflects variation in the degree and character of implemented policies and ongoing interventions to fight the opioid overdose epidemic. States have responded with a variety of policies and laws, including Prescription Drug Monitoring Programs, naloxone access laws, and Good Samaritan laws, among others. In addition to state-level interventions, there are local public health responses to the epidemic,(25,26) such as syringe exchange programs, jail-based programs, and community-based naloxone distribution efforts. Further, how state laws are implemented can vary across smaller areas within states. For example, there is within-state, county variation in actual levels of naloxone dispensing from retail pharmacies,(27) in addition to state differences in naloxone laws. Further, variation in county-level changes in opioid overdoses from 2017 to 2018 is also likely related to availability of medication treatment for opioid use disorder. Recent work shows that there is substantial county variation in treatment availability, including many counties with no medication-assisted treatment provider.(28) Our state- and county-level findings are relevant for guiding public health policies and interventions to the areas that need them most. Future research in this area can use our estimates of state-by-state changes in opioid overdose rates from 2017 to 2018, as well as in earlier years, using the same methods.(3)

There have been declining rates of unclassified drug overdoses since 2008, and recent efforts by the federal government may reduce these rates even further.(29) Yet, the changing proportion of unclassified drug overdoses over time may compromise studies examining the effectiveness of policies and interventions designed to address the epidemic. These studies often rely on historical spatio-temporal variation in fatal opioid overdoses,(30–34) and it is important that researchers make realistic corrections for non-random variation in the proportion of unclassified drug overdoses. This study provides further support for methods incorporating the suite of contributing causes of death in opioid surveillance in drug overdoses(3) by showing evidence of the method’s robustness to subsets of the data balanced between opioid and non-opioid overdoses while highlighting the potential diminishing returns beyond a certain number of contributing causes of death. These methods can be used to correct opioid mortality counts in specific areas and in future years, to more accurately evaluate the effectiveness of public health interventions to decrease opioid deaths.

Limitations

There are limitations to this study. First, our approach assumes that there are no unobserved factors differentiating unclassified drug overdoses from those with classified drug involvement. This allows us to apply our models parameterized using drug overdoses with classified drugs to generate predictions of opioid involvement in unclassified drug overdoses. We believe that our assumption is reasonable, as the percent of unclassified drug overdoses varied substantially between states. This variation is likely due to county and state variation in policies governing death examination and reporting.(35–37) If this is true, then it’s unlikely that the probability of the inclusion of T50.9 for an overdose is due to unobserved factors associated with the overdose or decedent, beyond those explicitly included as controls. With that said, there is a possibility that unclassified drug overdoses do not occur at random, which may compromise some of our predictions. Second, our approach assumes that opioid involvement is affected by contributing causes of death and not vice versa. Third, we assume that the classification of opioid and non-opioid involvement, as indicated in MCOD data from the NCHS, is correct. It is possible that some classified non-opioid drug overdoses may have actually involved opioid drugs. We cannot speculate how often this occurred, but this limitation would affect all studies examining opioid overdoses separate from non-opioid overdoses. Finally, we do not use more advanced machine learning models, which could improve accuracy of prediction of opioid involvement in drug overdoses (e.g., neural networks).

Conclusions

We estimated that the age-adjusted rate of opioid overdoses declined by 7.0% from 2017 to 2018, using recently developed methods in predicting opioid involvement in unclassified drug overdoses. This is a substantially larger estimate than the CDC estimate of 2.0% based on classified opioid overdoses. Our new estimates of the national decline in opioid overdoses from 2017 to 2018, as well as new estimates of state-level changes, can help policy-makers and public health officials understand the changing dynamics of the epidemic, as well as help steer resources to areas that are still struggling to curtail the epidemic.

Supplementary Material

Online Appendix

Figure 1:

Figure 1:

Open in a new tab

Trends in the proportion of drug overdoses without a classified causal drug, 1999–2018.

Notes: In this figure, we display the annual percent of total drug overdoses without an identified drug from 1999 to 2018. These drug overdoses featured the use of ICD-10 code T50.9 with no other contributing causes of death indicating a specific drug involved in the drug overdose.

Funding

This work was supported by the National Institutes of Health under Grant Number: DP5OD021338.

Footnotes

Disclosures

The authors report no relevant disclosures.

Data availability statement

The data used in this study are Multiple Cause of Death (MCOD) data from CDC’s National Center for Health Statistics (NCHS). All scripts used to generate results are available from the authors’ websites.

Works Cited

  • 1.Centers for Disease Control and Prevention. Opioid Data Analysis and Resources. [Internet]. 2020. Available from: Available from: https://www.cdc.gov/drugoverdose/data/analysis.html
  • 2.Wilson N Drug and Opioid-Involved Overdose Deaths — United States, 2017–2018. MMWR Morb Mortal Wkly Rep [Internet]. 2020. [cited 2020 Aug 23];69. Available from: https://www.cdc.gov/mmwr/volumes/69/wr/mm6911a4.htm [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Boslett AJ, Denham A, Hill EL. Using contributing causes of death improves prediction of opioid involvement in unclassified drug overdoses in US death records. Addiction [Internet]. 2020. [cited 2020 May 29]; Available from: https://onlinelibrary.wiley.com/doi/abs/10.1111/add.14943 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Ruhm CJ. Corrected US opioid-involved drug poisoning deaths and mortality rates, 1999–2015. Addiction. 2018;113(7):1339–44. [DOI] [PubMed] [Google Scholar]
  • 5.U.S. Department of Health & Human Services. Centers for Disease Control and Prevention. ICD-10-CM Tool [Internet]. 2020. [cited 2020 Jul 2]. Available from: https://icd10cmtool.cdc.gov/?fy=FY2020
  • 6.Centers for Disease Control and Prevention, National Cancer Institute. U.S. Population Data - SEER Population Data [Internet]. SEER. [cited 2020 Aug 12]. Available from: https://seer.cancer.gov/popdata/index.html
  • 7.Boulesteix A-L, Janitza S, Kruppa J, König IR. Overview of random forest methodology and practical guidance with emphasis on computational biology and bioinformatics. WIREs Data Min Knowl Discov. 2012;2(6):493–507. [Google Scholar]
  • 8.U.S. Centers for Disease Control and Prevention. National Cancer Institute. Standard Populations (Millions) for Age-Adjustment - SEER Population Datasets [Internet]. SEER. [cited 2020 Aug 23]. Available from: https://seer.cancer.gov/stdpopulations/index.html
  • 9.Streiner DL, Cairney J. What’s under the ROC? An Introduction to Receiver Operating Characteristics Curves. Can J Psychiatry. 2007. Feb 1;52(2):121–8. [DOI] [PubMed] [Google Scholar]
  • 10.Chicco D, Jurman G. The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation. BMC Genomics. 2020. Jan 2;21(1):6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Ruhm C Deaths of Despair or Drug Problems? [Internet]. Cambridge, MA: National Bureau of Economic Research; 2018. Jan [cited 2019 Jan 26]. Report No.: w24188. Available from: http://www.nber.org/papers/w24188.pdf [Google Scholar]
  • 12.Boslett AJ, Denham A, Hill EL, Adams MCB. Unclassified drug overdose deaths in the opioid crisis: emerging patterns of inequity. J Am Med Inform Assoc. 2019. Aug 1;26(8–9):767–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Dowell D, Haegerich TM, Chou R. CDC Guideline for Prescribing Opioids for Chronic Pain — United States, 2016. MMWR Recomm Rep [Internet]. 2016. [cited 2020 Aug 23];65. Available from: https://www.cdc.gov/mmwr/volumes/65/rr/rr6501e1.htm [DOI] [PubMed] [Google Scholar]
  • 14.Bohnert ASB, Guy GP, Losby JL. Opioid Prescribing in the United States Before and After the Centers for Disease Control and Prevention’s 2016 Opioid Guideline. Ann Intern Med. 2018. Sep 18;169(6):367–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Naumann RB, Durrance CP, Ranapurwala SI, Austin AE, Proescholdbell S, Childs R, et al. Impact of a community-based naloxone distribution program on opioid overdose death rates. Drug Alcohol Depend. 2019. 01;204:107536. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Abouk R, Pacula RL, Powell D. Association Between State Laws Facilitating Pharmacy Distribution of Naloxone and Risk of Fatal Overdose. JAMA Intern Med. 2019. 01;179(6):805–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Moallef S, Hayashi K. The effectiveness of drug-related Good Samaritan laws: A review of the literature. Int J Drug Policy. 2020. May 25;102773. [DOI] [PubMed] [Google Scholar]
  • 18.McClellan C, Lambdin BH, Ali MM, Mutter R, Davis CS, Wheeler E, et al. Opioid-overdose laws association with opioid use and overdose mortality. Addict Behav. 2018;86:90–5. [DOI] [PubMed] [Google Scholar]
  • 19.Nechuta SJ, Tyndall BD, Mukhopadhyay S, McPheeters ML. Sociodemographic factors, prescription history and opioid overdose deaths: a statewide analysis using linked PDMP and mortality data. Drug Alcohol Depend. 2018. 01;190:62–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Puac-Polanco V, Chihuri S, Fink DS, Cerdá M, Keyes KM, Li G. Prescription Drug Monitoring Programs and Prescription Opioid-Related Outcomes in the United States. Epidemiol Rev. 2020. Apr 3; [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.McDonald DC, Carlson K, Izrael D. Geographic Variation in Opioid Prescribing in the U.S. J Pain. 2012. Oct 1;13(10):988–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Rolheiser LA, Cordes J, Subramanian S v. Opioid Prescribing Rates by Congressional Districts, United States, 2016. Am J Public Health. 2018. Jul 19;108(9):1214–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Rudd RA, Aleshire N, Zibbell JE, Gladden RM. Increases in Drug and Opioid Overdose Deaths — United States, 2000–2014. Morb Mortal Wkly Rep. 2016;64(50 & 51):1378–82. [DOI] [PubMed] [Google Scholar]
  • 24.Scholl L, Seth P, Kariisa M, Wilson N, Baldwin G. Drug and Opioid-Involved Overdose Deaths — United States, 2013–2017. Morb Mortal Wkly Rep. 2018. Jan 4;67(51–52):1419–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Johnson Q Case Study: County-Level Responses to the Opioid Crisis in Northern Kentucky. J Law Med Ethics J Am Soc Law Med Ethics. 2018. Jun;46(2):382–6. [DOI] [PubMed] [Google Scholar]
  • 26.Castillo T Harm Reduction Strategies for the Opiod Crisis. N C Med J. 2018. Jun;79(3):192–4. [DOI] [PubMed] [Google Scholar]
  • 27.Guy GP, Haegerich TM, Evans ME, Losby JL, Young R, Jones CM. Vital Signs: Pharmacy-Based Naloxone Dispensing - United States, 2012–2018. MMWR Morb Mortal Wkly Rep. 2019. Aug 9;68(31):679–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Haffajee RL, Lin LA, Bohnert ASB, Goldstick JE. Characteristics of US Counties With High Opioid Overdose Mortality and Low Capacity to Deliver Medications for Opioid Use Disorder. JAMA Netw Open. 2019. 05;2(6):e196373. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Centers for Disease Control and Prevention Injury Center. Enhanced State Opioid Overdose Surveillance [Internet]. 2019. [cited 2020 Jul 8]. Available from: https://www.cdc.gov/drugoverdose/foa/state-opioid-mm.html
  • 30.Powell D, Pacula RL, Taylor E. How increasing medical access to opioids contributes to the opioid epidemic: Evidence from Medicare Part D. J Health Econ. 2020. May 1;71:102286. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Alpert A, Powell D, Pacula RL. Supply-Side Drug Policy in the Presence of Substitutes: Evidence from the Introduction of Abuse-Deterrent Opioids. Am Econ J Econ Policy. 2018. Nov;10(4):1–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Li G, Brady JE, Lang BH, Giglio J, Wunsch H, DiMaggio C. Prescription drug monitoring and drug overdose mortality. Inj Epidemiol. 2014. Apr 24;1(1):9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Venkataramani AS, Bair EF, O’Brien RL, Tsai AC. Association Between Automotive Assembly Plant Closures and Opioid Overdose Mortality in the United States: A Difference-in-Differences Analysis. JAMA Intern Med [Internet]. 2019. Dec 30 [cited 2020 Jan 14]; Available from: https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2757788 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Dowell D, Zhang K, Noonan RK, Hockenberry JM. Mandatory Provider Review And Pain Clinic Laws Reduce The Amounts Of Opioids Prescribed And Overdose Death Rates. Health Aff (Millwood). 2016. Oct 1;35(10):1876–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Warner M, Paulozzi LJ, Nolte KB, Davis GG, Nelson LS. State Variation in Certifying Manner of Death and Drugs Involved in Drug Intoxication Deaths. Acad Forensic Pathol. 2013. Jun 1;3(2):231–7. [Google Scholar]
  • 36.Fischtein D, Cina SJ. Errors on Death Certificates Requiring Amendments: The Broward County Experience. Am J Forensic Med Pathol. 2011. Jun;32(2):146–8. [DOI] [PubMed] [Google Scholar]
  • 37.Slavova S, Delcher C, Buchanich JM, Bunn TL, Goldberger BA, Costich JF. Methodological Complexities in Quantifying Rates of Fatal Opioid-Related Overdose. Curr Epidemiol Rep. 2019. Jun 1;6(2):263–74. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Online Appendix
NIHMS1855448-supplement-Online_Appendix.docx (19KB, docx)

Data Availability Statement

The data used in this study are Multiple Cause of Death (MCOD) data from CDC’s National Center for Health Statistics (NCHS). All scripts used to generate results are available from the authors’ websites.

RESOURCES