/excess_drug_overdoses

Reproducible code for our paper about excess drug poisonings in California in 2020

Primary LanguageHTML

Excess fatal drug overdoses in California during the COVID-19 pandemic by race/ethnicity, educational attainment, and region: A population-based study

This is reproducible code for our paper, “Excess fatal drug overdoses in California during the COVID-19 pandemic by race/ethnicity, educational attainment, and region: A population-based study” in The Lancet Regional Health - Americas. The full citation is:

Kiang MV, Acosta RJ, Chen Y-H, Matthay EC, Tsai AC, Basu S, Glymour MM, Bibbins-Domingo K, Humphreys K, Arthur KN, “Sociodemographic and geographic disparities in excess fatal drug overdoses during the COVID-19 pandemic in California: a population-based study.” The Lancet Regional Health - Americas. 2022 Jul;11:100237. doi: 10.1016/j.lana.2022.100237. Epub 2022 Mar 19. PubMed PMID: 35342895; PubMed Central PMCID: PMC8934030.

Abstract

Background. The coronavirus disease 2019 (COVID-19) pandemic is co-occurring with a drug addiction and overdose crisis.

Methods. We fit overdispersed Poisson models to estimate the excess fatal drug overdoses (i.e., deaths greater than expected), using data on all deaths in California from 2016 through 2020.

Findings. Between January 5, 2020 and December 26, 2020, there were 8,605 fatal drug overdoses—a 44% increase over the same period one year prior. We estimated 2,084 (95% CI: 1,925 to 2,243) fatal drug overdoses were excess deaths, representing 5·28 (4·88 to 5·68) excess fatal drug overdoses per 100,000 population. Excess fatal drug overdoses were driven by opioids, especially synthetic opioids. The non-Hispanic Black and Other non-Hispanic populations were disproportionately affected with 10·1 and 13·26 excess fatal drug overdoses per 100,000 population, respectively, compared to 5·99 per 100,000 population in the non-Hispanic white population. There was a steep, nonlinear educational gradient with the highest rate among those with only a high school degree. There was a strong spatial patterning with the highest levels of excess mortality in the southernmost region and consistent decreases at more northern latitudes (7·73 vs 1·96 per 100,000).

Interpretation. Fatal drug overdoses disproportionately increased in 2020 among structurally marginalized populations and showed a strong geographic gradient. Local, tailored public health interventions are urgently needed to reduce growing inequities in overdose deaths.

Issues

Please submit issues via Github or via email.

Important note about reproducibility

In accordance with our data use agreement with the California Department of Public Health - Vital Statistics, we cannot share decedent-level data required to fully reproduce our results. To run the code from start to finish, you must request restricted-access data from CDPH. (See Requirements.) When allowed by our DUA, we share aggregated data on this repo and the numeric representation of all figures (./output).

Requirements

Restricted-access death certificate level data

You can request the decedent-level microdata from CDPH using the Vital Stats website.

Software

All analyses are conducted using R, which can be downloaded via CRAN. We also recommend the use of RStudio when running R, which will allow users to take advantage of renv for dependency management.

One supplemental analysis requires using the Joinpoint Regression Program, which can be downloaded from the National Cancer Institute.

Analysis pipeline

  • code: The scripts in ./code are designed to be run in order. The header of all analytic files provide a brief description. Some files are quite long so it is suggested you use RStudio’s Document Outline feature (CMD + SHIFT + O on macOS).
  • config.yml: Contains all the configuration information including the starting and ending dates for the predictions as well as the location of the raw and processed files on the secure compute environment.
  • data: All data after it has been processed and aggregated.
  • data_private: Data that cannot be publicly shared (not on Github).
  • data_raw: Publicly available raw data from the US Census.
  • joinpoint: Contains files necessary to replicate our sensitivity analysis using joinpoint regression.
  • output: The numerical representation of all figures in csv format.
  • plots: All figures in the manuscript and supplement.
  • renv: Files necessary to load the same R packages we used in our analysis.
  • rmds: Rmarkdown files such as supplemental tables.

Session information

See ./session_info.txt for more reproducibility information.

Authors (alphabetical)