Persistence and Pervasiveness

Abstract

Objectives

Although trends in opioid-related death rates in the United States have been described, the association between state-level opioid overdose death rates in early waves and substance-related overdose death rates in later waves has not been characterized. We examined the relationship between state-level opioid overdose death rates at the beginning of the crisis (1999-2004) and overdose death rates for opioids and other substances in later years.

Methods

Using 1999-2018 multiple cause of death data from the Centers for Disease Control and Prevention, we first categorized each state by quartile of baseline (1999-2004) opioid overdose death rates. By baseline opioid overdose death rates, we then compared states’ annual overdose death rates from any opioid, heroin, synthetic opioids, sedatives, stimulants/methamphetamine, and cocaine from 2005 through 2018. To test the association between baseline opioid overdose death rates and subsequent substance-related overdose death rates for all 6 substances, we estimated unadjusted and adjusted linear models controlling for annual state-level unemployment, median household income, age, sex, and race/ethnicity.

Results

Our results suggest 2 characteristics of the opioid crisis: persistence and pervasiveness. In adjusted analyses, we found that for each additional opioid overdose death per 100 000 population at baseline, states had 23.5 more opioid deaths, 4.4 more heroin deaths, 8.0 more synthetic opioid deaths, 9.2 more sedative deaths, 3.3 more stimulant deaths, and 4.6 more cocaine deaths per 100 000 population from 2005 to 2018.

Conclusion

These findings have important implications for continued surveillance to assist policy makers in deciding how to deploy resources to combat not just opioid use disorder but also polysubstance use disorder and broader problems of substance use disorder.

Substantial attention and resources have been targeted at the opioid crisis, often described as a monolithic event. However, rather than a single episode, the opioid crisis has unfolded in 4 distinct waves.^1-5 The first wave was characterized by overdose deaths related to prescription opioid use,^2,6,7 and the second wave was a shift toward heroin use, due in part to the reformulation of Oxycontin.^8-10 The increasing use of synthetic opioids, such as fentanyl, distinguished the third wave.^11-14 The fourth wave comprised increased use of methamphetamine and cocaine, often with illicit opioids,^3,11,15-17 which has meant that health care providers, policy makers, and public health practitioners have had to continuously adjust to address the changing nature of the substance use crisis.

Although trends in opioid-related death rates in the United States have been described,^1-3,16,18-20 the association between state-level opioid overdose death rates in early waves and substance-related overdose death rates in later waves have not been well characterized.^1,14,15 Such data can inform policy responses to the crisis by identifying whether states with high death rates in early waves had high death rates in later waves and whether rates of non–opioid-related overdose deaths in later waves were correlated with rates of opioid-related overdose deaths in early waves.

Methods

We used 1999-2018 multiple cause of death data from the Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research (WONDER) database to calculate state-specific overdose death rates (per 100 000 population) in 6 categories of illicit substances.²¹ Using an established approach,^20,22,23 we first identified International Classification of Diseases, 10th Revision (ICD-10)²⁴ codes that indicate a drug-poisoning death (X40-X44, X60-X64, X85, Y10-Y14). We then identified deaths attributable to each drug by using the following ICD-10 codes: (1) any opioid (T40.0-T40.4, T40.6), (2) heroin (T40.1), (3) synthetic opioids (T40.4), (4) sedatives including benzodiazepines (T42.0-T42.4, T42.6-T42.8), and stimulants with abuse potential including (5) methamphetamine (T43.6) or (6) cocaine (T40.5).

We first calculated the mean annual opioid overdose death rates for all 50 states and Washington, DC, during a baseline period of 1999-2004. Based on mean annual opioid overdose death rates in the baseline period, we categorized each state and DC into quartiles of baseline opioid overdose death rates. For each drug type, we then depicted annual overdose death rates over time by baseline opioid overdose death rate quartile to understand secular trends in overdose death rates by baseline level of opioid overdose death rate. To quantify associations between baseline opioid overdose death rates and subsequent substance-specific overdose death rates, we estimated 2 sets of linear regressions. We estimated state-level overdose death rates for each drug type from 2005 to 2018 and from 2014 to 2018. We included the latter set of years to better capture the persistence of the association between recent rates of substance-related overdose deaths relative to baseline rates of opioid overdose deaths at least 10 years before. For each regression, the independent variable of interest was the baseline (1999-2004) state opioid overdose death rate. We estimated both unadjusted regressions and regressions adjusting for year and state-level demographic characteristics based on American Community Survey data, including median household income, median age, sex, race/ethnicity (percentage non-Hispanic White, non-Hispanic Black, and Hispanic), and education (percentage ≥high school graduate).²⁵ We included data on state-level unemployment based on Bureau of Labor Statistics data.²⁶ Our adjusted models controlled for state-level demographic and economic factors that may have changed between the baseline and follow-up periods and that might independently predict overdose death rates.^27,28 We used the t statistic to determine significant results from adjusted regressions. We clustered standard errors by state and considered P < .05 to be significant.

We also estimated 2 sets of sensitivity analyses to determine the robustness of our results. First, because states may vary in their reporting of overdose death rates for specific substances (ie, opioids),^23,29 we alternatively defined our independent variable of interest to be baseline (1999-2004) overdose death rates from any drug-poisoning death (ie, without any T codes). Second, to account for the limited years in which some states had too few deaths to be reported by CDC WONDER, we used 3 alternate imputations of these values. We included a low imputation (assumed 1 death), a middle imputation (assumed 5 deaths), and a high imputation (assumed 9 deaths). In baseline analyses, we omitted observations with suppressed death counts when a state had <10 overdose deaths in a year (ranging from <0.5% for opioids to >20% for stimulants/methamphetamine).

Because we used publicly available, state-level data, we determined our analysis was not human subjects research and did not require institutional review board review. We estimated all analyses using Stata version 15.1 (StataCorp).

Results

Eleven states were in the highest quartile of baseline opioid overdose death rates, and 13 states each were in the second-highest, second-lowest, and lowest quartiles (Table 1). Across each quartile of baseline opioid overdose death rates, states had increases in each measure of opioid overdose death rates, including any opioid, heroin, and synthetic opioids (eg, fentanyl) deaths (Figure). Although the magnitude of opioid overdose death rates was larger than the overdose death rates from other substances, large increases in overdose death rates for sedatives and stimulants/methamphetamine occurred from baseline to 2018, and increases in overdose death rates for cocaine began around 2015. States in the highest quartile of baseline opioid overdose death rates not only had large increases in opioid overdose death rates from 2005 to 2018, but they also had large increases in overdose death rates for sedatives, stimulants/methamphetamine, and cocaine relative to states in the lowest quartiles of baseline opioid death rates. We found minimal crossing of death rate curves (Figure); states in the highest quartile of baseline opioid overdose death rates generally had the highest overdose death rates from opioids and other substances throughout the study period. Conversely, states with the lowest opioid overdose death rates at baseline had the lowest rates of overdose death from opioids and other substance use from 2005 to 2018.

Table 1.

States included in each quartile of baseline opioid overdose death rates, United States, 1999-2004^a

Lowest quartile	Second-lowest quartile	Second-highest quartile	Highest quartile
Alabama	Alaska	California	Arizona
Georgia	Arkansas	Colorado	District of Columbia
Indiana	Hawaii	Connecticut	Maine
Iowa	Idaho	Delaware	Maryland
Kansas	Missouri	Florida	Massachusetts
Louisiana	Montana	Illinois	Nevada
Michigan	New York	Kentucky	New Mexico
Minnesota	Ohio	New Hampshire	Oklahoma
Mississippi	Pennsylvania	New Jersey	Rhode Island
Nebraska	Tennessee	North Carolina	Utah
North Dakota	Texas	Oregon	Washington
South Carolina	Wisconsin	Vermont	West Virginia
South Dakota	Wyoming	Virginia

^aQuartiles based on 1999-2004 data from the Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research detailed mortality files.²¹ Quartiles calculated using mean 1999-2004 opioid overdose death rates by state.

Figure.

State-level overdose death rates during 1999-2018, stratified by quartile of baseline (1999-2004) opioid overdose death rates, United States. Based on 1999-2018 data from the Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research (CDC WONDER).²¹ Because the CDC WONDER database suppressed all instances in which a state had <10 deaths for a given overdose cause in a given year, we omitted these data from the calculations. Quartiles calculated using mean 1999-2004 opioid overdose death rates by state.

Baseline overdose death rates for 1999-2004 for each substance ranged from 0.55 per 100 000 population per year for stimulants/methamphetamine to 4.07 per 100 000 population per year for any opioid (Table 2). Baseline opioid overdose death rates were associated with significantly higher subsequent overdose death rates for each drug type. For example, adjusted estimates showed that a 1-point increase in the baseline opioid overdose death rate per 100 000 population was associated with 23.5 more opioid overdose deaths per 100 000 population from 2005 to 2018 (ie, 1.68 more deaths per 100 000 population per year during 14 years). Similarly, a 1-point increase in the baseline opioid overdose death rate per 100 000 population was associated with 4.4 more heroin deaths (0.31 more deaths per year), 8.0 more synthetic opioid deaths (0.57 more deaths per year), 9.2 more sedative deaths (0.66 more deaths per year), 3.3 more stimulant/methamphetamine deaths (0.24 more deaths per year), and 4.6 more cocaine deaths (0.33 more deaths per year) per 100 000 population from 2005 to 2018. Results were similar in unadjusted and adjusted analyses.

Table 2.

Estimated associations between baseline (1999-2004) opioid overdose death rates and state-level overdose death rates from 2005 to 2018, by substance, United States^a

Substance	Baseline (1999-2004) deaths per 100 000 population	Estimated association between annualized change in overdose death rate (2005-2018) and baseline opioid overdose death rate (1999-2004) per 100 000 population
		Unadjusted	Adjustedb
Opioid (any)	4.07	1.46 (0.22) [<.001]	1.68 (0.17) [<.001]
Heroin	1.03	0.27 (0.10) [.01]	0.31 (0.11) [.01]
Synthetic opioids	0.58	0.44 (0.20) [.03]	0.57 (0.12) [<.001]
Sedatives	1.10	0.49 (0.19) [.01]	0.66 (0.15) [<.001]
Stimulants/methamphetamine	0.55	0.26 (0.07) [<.001]	0.24 (0.05) [<.001]
Cocaine	1.78	0.32 (0.10) [.002]	0.33 (0.10) [.002]

^aAll values are coefficient (standard error [SE]) [P value] unless otherwise indicated. P value based on t statistic from regression coefficient estimates; P < .05 considered significant. Mortality data based on 1999-2018 Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research (CDC WONDER) detailed mortality files.²¹ Adjusted estimates include data from the US Census Bureau’s American Community Survey²⁵ and the Bureau of Labor Statistics’ local unemployment statistics.²⁶ Because the CDC WONDER database suppressed all instances in which a state had <10 deaths for a given overdose cause in a given year, we omitted these data from the calculations.

^bAdjusted regressions control for year, state-level unemployment, median household income, median age, percentage female, percentage high school graduate, percentage non-Hispanic White, percentage non-Hispanic Black, and percentage Hispanic. State-level data on median household income and education were not yet available for 2018; as such, they were assigned 2017 values. (Dropping 2018 had little impact on estimates.) All regressions cluster SEs at the state level.

Baseline opioid overdose death rates were significantly associated with overdose death rates for each drug type more than a decade later, based on results using overdose deaths restricted to 2014-2018. In adjusted models, a 1-point increase in baseline opioid overdose deaths per 100 000 population was associated with 1.99 more opioid overdose deaths, 0.36 more heroin deaths, 1.03 more synthetic opioid deaths, 0.72 more sedative deaths, 0.42 more stimulant/methamphetamine deaths, and 0.28 more cocaine deaths per 100 000 population per year (Table 3).

Table 3.

Adjusted associations between baseline (1999-2004) opioid overdose death rates and subsequent state-level overdose death rates (2014-2018), by substance and alternate time periods, United States^a

Substance	Baseline (1999-2004) overdose death rate per 100 000 population	Estimated association between annualized change in overdose death rates (2014-2018) and baseline opioid overdose death rates (1999-2004) per 100 000 population
		Unadjusted	Adjustedb
Opioid (any)	4.07	1.96 (0.46) [<.001]	1.99 (0.33) [<.001]
Heroin	1.03	0.55 (0.18) [.003]	0.36 (0.17) [.04]
Synthetic opioids	0.58	1.01 (0.44) [.03]	1.03 (0.24) [<.001]
Sedatives	1.10	0.59 (0.23) [.01]	0.72 (0.21) [.001]
Stimulants/ methamphetamine	0.55	0.37 (0.15) [.02]	0.42 (0.06) [<.001]
Cocaine	1.78	0.39 (0.19) [.04]	0.28 (0.10) [.01]

^aAll values are coefficient (standard error [SE]) [P value] unless otherwise indicated. P value based on t statistic from regression coefficient estimates; P < .05 considered significant. Mortality data based on 1999-2018 Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research (CDC WONDER) detailed mortality files.²¹ Adjusted estimates include data from the US Census Bureau’s American Community Survey²⁵ and the Bureau of Labor Statistics’ local unemployment statistics.²⁶ Because the CDC WONDER database suppressed all instances in which a state had <10 deaths for a given overdose cause in a given year, we omitted these data from the calculations.

^bAdjusted regressions controlled for year, state-level unemployment, median household income, median age, percentage female, percentage high school graduate, percentage non-Hispanic White, percentage non-Hispanic Black, and percentage Hispanic. State-level data on median household income and education were not yet available in the 2018 data; as such, they were assigned 2017 values. (Dropping 2018 had little impact on estimates.) All regressions cluster SEs at the state level.

We estimated 2 sets of sensitivity analyses. The first analysis used overall overdose death rates at baseline as the independent variable of interest. Results were similar to our baseline estimates, albeit slightly lower. The second analysis used alternate approaches for dealing with suppressed values. The results were also similar to our baseline estimates using each of the different approaches to handling suppressed values. (Supplemental tables available from authors upon request.)

Discussion

Our results highlight 2 important features of the opioid crisis. First, although all states had increases in the rates of opioid and other drug overdose deaths, states that had higher opioid overdose death rates at baseline (1999-2004) had significantly greater increases in opioid overdose death rates than states with lower baseline opioid overdose death rates. This finding suggests that although state-level opioid overdose death rates diverged during the study period, states with high overdose death rates at baseline had persistently elevated opioid overdose death rates in later years of the crisis relative to states with low baseline overdose death rates.

Second, in recent years, there has been an increase in the overdose death rate in the United States from sedatives, stimulants/methamphetamine, and cocaine. Compared with states that had smaller increases in rates of overdose death, states with greater increases in overdose death rates from sedatives, stimulants/methamphetamine, and cocaine also had higher baseline opioid overdose death rates. After adjusting for sociodemographic and state-level differences, baseline opioid overdose death rates in 1999-2004 were significantly associated with future opioid- and non–opioid-related overdose death rates. States that had high rates of opioid overdose deaths at baseline had high rates of overdose deaths from multiple substances in later periods; we found little upward or downward shifting of death rate quartiles among states. This finding highlights the need to expand monitoring and treatment systems beyond opioids, or even individual substances, to understand and prevent substance-related overdose deaths that stem from an increasingly complex group of substances. An important area of future research is to examine individual- and community-level factors that are associated with a shifting pattern of substance use–related deaths.

Limitations

This study had several limitations. First, we were unable to control for person-level characteristics because we were not able to link individual-level characteristics to death records of those who died of an overdose. However, we did adjust for state-level factors. Second, we were limited by the inability to disentangle the role of synthetic opioids in non–opioid-related overdose deaths. For example, CDC WONDER data do not allow for the identification of deaths caused by a substance but not synthetic opioids. However, we estimated additional models controlling for the concurrent synthetic opioid overdose death rate (available upon request). We found that all estimates were smaller in magnitude relative to our baseline adjusted estimates but remained positive and significant (with the one exception being the cocaine estimate, which was no longer significant); our adjusted estimates for 2005-2018 were 12% lower than our baseline adjusted estimates for sedatives, 6% lower for stimulants/methamphetamine, and 58% lower for cocaine. Finally, although our aim was to document the evolving patterns of overdose death rates and the association between baseline opioid overdose death rates and subsequent opioid and other substance-related overdose death rates, our results are primarily descriptive. As such, we could not make causal inferences.

Conclusion

Without effective, large-scale interventions, trends in overdose death rates may persist for decades. A narrow focus on individual substances is likely to be ineffective given the strong associations between opioid- and non–opioid-related overdose deaths. It is critical that states and local municipalities continue to collect surveillance data to understand changing patterns of substance use disorders, including both polysubstance use and independent spikes in non–opioid-related substance use and deaths.

ORCID iDs

Joel E. Segel, PhD https://orcid.org/0000-0001-8937-0531

Tyler N.A. Winkelman, MD, MSc https://orcid.org/0000-0002-9581-5223