Drug Poisoning Mortality in the United States, 1999-2018
Drug Poisoning Mortality in the United States, 1999-2018
These figures present drug poisoning deaths at the national, state, and county levels. The first two dashboards depict U.S. and state trends in age-adjusted death rates for drug poisoning beginning in 1999 by selected demographic characteristics. The third, fourth, and fifth dashboards present a series of heat maps, grids, and trend-lines of model-based county estimates for drug-poisoning mortality beginning in 2003. Hierarchical Bayesian models with spatial and temporal random effects were used to generate these county-level estimates (see Technical Notes for details).
Select a dashboard from the drop-down menu, then click on “Update Dashboard” to navigate through different graphics.
- The first dashboard shows national estimates. Use the year slider to select data years for the bar charts on the top. When using the radio buttons to select age, sex, and race and Hispanic origin, the bar charts display deaths for drug poisoning by sex or age groups, and the line chart shows national trends in death rates for selected demographic groupings.
- The second dashboard shows state estimates. The line charts describe the U.S. and state trends in age-adjusted death rates for drug poisoning. The U.S. map presents age-adjusted death rates for drug poisoning per 100,000 population by state and year, with the magnitude of the state death rates indicated by the color gradient. Click on a state in the map to display that state’s trend line in the graph.
- The third dashboard is a heat map of county estimates, showing model-based crude death rates for drug poisoning per 100,000 population by county and year. The color scale indicates the magnitude of the estimated county-level death rates in ranges. Use the arrows or the slider to select a year. Click on any state to zoom into it on the map. Click outside the state to zoom back out to the map of the U.S. Users may click on the gray “home” icon in the upper right-hand corner of the map to reset the view, if necessary.
- The fourth dashboard features a county grid showing the change in estimated drug poisoning death rates rate by year using the same color scale as the county heat map. Click on a state in the map to display the counties for that state in the grid.
- The fifth dashboard displays county-level trends in drug poisoning mortality by urban-rural classification. Select one or more states from the drop-down menu at the top to compare county-level trends in drug overdose death rates across urban-rural categories.
Download datasets in CSV format by clicking on the link for the desired dataset under “CSV Format” link. Additional file formats are available for download for each dataset at Data.CDC.Gov.
† Deaths are classified using the International Classification of Diseases, Tenth Revision (ICD–10). Drug-poisoning deaths are defined as having ICD–10 underlying cause-of-death codes X40–X44 (unintentional), X60–X64 (suicide), X85 (homicide), or Y10–Y14 (undetermined intent).
‡ Estimates are based on the National Vital Statistics System multiple cause-of-death mortality files (1). Age-adjusted death rates (deaths per 100,000 U.S. standard population for 2000) are calculated using the direct method. Populations used for computing death rates for 2011–2018 are postcensal estimates based on the 2010 U.S. census. Rates for census years are based on populations enumerated in the corresponding censuses. Rates for noncensus years before 2010 are revised using updated intercensal population estimates and may differ from rates previously published.
* Estimate does not meet standards of reliability or precision. Death rates are flagged as “Unreliable” in the chart when the rate is calculated with a numerator of 20 or less.
** Death rates for some states and years may be low due to a high number of unresolved pending cases or misclassification of ICD–10 codes for unintentional poisoning as R99, “Other ill-defined and unspecified causes of mortality” (2). For example, this issue is known to affect New Jersey in 2009 and West Virginia in 2005 and 2009 but also may affect other years and other states. Drug poisoning death rates may be underestimated in those instances.
§ Smoothed county-level crude death rates (deaths per 100,000 population) were obtained according to methods described in the Technical Notes. Briefly, hierarchical Bayesian models with spatial and temporal random effects were used to generate estimates of county-level crude death rates due to drug poisoning for each year. These annual county-level estimates “borrow strength” across counties to generate stable estimates of death rates where data are sparse due to small population size (3). Estimates for 2003-2018 may differ from previously published estimates (4–6), and should not be considered directly comparable (see Technical Notes). Previously published estimates can be found here (6-7). County boundaries are largely consistent with the vintage 2005-2007 bridged-race population file geographies, with some counties merged with adjacent counties in cases where county boundaries changed over time (8). County urban-rural classification is based on the 2013 National Center for Health Statistics (NCHS) Urban–Rural Classification Scheme for Counties (9).
County-level estimates were generated using Hierarchical Bayesian models with spatial and temporal random effects using the INLA package for R (3,10–11). These models borrow strength over time and across neighboring counties to produce stable estimates of drug overdose death rates by county and year. Because the Hierarchical Bayesian models were more computationally intensive than prior models (4–5), models including additional years (e.g., 1999-2002) or age-specific terms to compute age-adjusted values were not able to be implemented.
Annual county-level drug overdose deaths were modeled as a function of:
- An overall intercept;
- A fixed effect for year;
- A spatial random effect, which accounts for clustering of drug overdose death rates;
- A non-spatial random effect, which accounts for any residual county-level variation;
- A temporal random effect, which accounts for nonlinearities over time by allowing the value in any given year to depend on the value in a prior year, plus an error term;
- A space-time interaction term (i.e., a county- and year-specific random effect), accounting for any residual spatiotemporal variation.
Drug overdose death counts at the county-level are not normally distributed, as they are highly zero-inflated and right-skewed. To account for this, models used a zero-inflated binomial distribution. Models were also fit using zero-inflated Poisson and zero-inflated negative binomial distributions. Models were compared using the Deviance Information Criterion (DIC) where lower values are preferred (12), and indicated that the zero-inflated binomial models offered the best fit.
Posterior predicted median death rates were obtained from the best fitting model. These model-based estimates are provided to show which counties have higher or lower drug overdose death rates, and how these rates have changed over time. Model-based estimates may over- or under-estimate true drug overdose death rates, and may not match drug overdose death rates obtained from CDC WONDER. Bayesian credible intervals are provided, which show a range of values within which there is a 95% probability that the true drug overdose death rate will fall, based on the observed death rates and the model. The updated estimates may differ from the previous model-based estimates for several reasons, and should not be considered directly comparable.
NCHS, National Vital Statistics System, mortality data (see http://www.cdc.gov/nchs/deaths.htm).
- National Center for Health Statistics. National Vital Statistics System: Mortality data. Available from: http://www.cdc.gov/nchs/deaths.htm.
- CDC. CDC Wonder: Underlying cause of death 1999–2016. Available from: http://wonder.cdc.gov/wonder/help/ucd.html.
- Khan D, Rossen LM, Hedegaard H, Warner M. A Bayesian spatial and temporal modeling approach to mapping geographic variation in mortality rates for subnational areas with R-INLA. J Data Sci. 2018;18: 147-182.
- Rossen LM, Khan D, Warner M. Trends and geographic patterns in drug-poisoning death rates in the U.S., 1999–2009. Am J Prev Med 45(6): 19-25. 2013.
- Rossen LM, Khan D, Warner M. Hot spots in mortality from drug poisoning in the United States, 2007–2009. Health Place 26:14–20. 2014.
- Rossen LM, Bastian B, Warner M, and Khan D. NCHS – Drug Poisoning Mortality by County: United States, 1999-2016. Available from: https://data.cdc.gov/NCHS/NCHS-Drug-Poisoning-Mortality-by-County-United-Sta/p56q-jrxg.
- Rossen LM, Bastian B, Warner M, and Khan D. NCHS – Drug Poisoning Mortality by County: United States, 1999-2017. Available from: https://data.cdc.gov/NCHS/NCHS-Drug-Poisoning-Mortality-by-County-United-Sta/rpvx-m2md.
- National Center for Health Statistics. County geography changes: 1990–present. Available from: https://www.cdc.gov/nchs/data/nvss/bridged_race/County-Geography-Changes-1990-present.pdfpdf icon.
- Ingram DD, Franco SJ. 2013 NCHS urban–rural classification scheme for counties. National Center for Health Statistics. Vital Health Stat 2(166). 2014.
- Bivand RS, Rubio-Gomez V, Rue H. Spatial data analysis with R-INLA with some extensions. J Stat Softw. 2015;63(20).(294):1-8.
- Rue H, Martino S, Chopin N. Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. J R Stat Soc Series B Stat Methodol. 2009;71:319-392.
- Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A. Bayesian measures of model complexity and fit (with discussion). J R Stat Soc Series B Stat Methodol. 2002;64(4):583-639.
Rossen LM, Bastian B, Warner M, Khan D, Chong Y. Drug poisoning mortality: United States, 1999–2018. National Center for Health Statistics. 2020.
Designed by B Bastian, L Rossen, JM Keralis, and Y Chong: CDC/National Center for Health Statistics.