Drug overdose deaths are a major public health issue in the United States.
In this CDC/NCHS dataset, overdose death rates appear to increase dramatically leading up to 2020 and beyond.

Goal: Compare overdose death rates in two time windows:

• Pre-pandemic: 2012–2019
• Post-pandemic: 2019–2022

The dataset provides drug overdose death rates (per 100,000 resident population) for the United States by:

• Drug type (all overdoses, synthetic opioids, heroin)
• Population grouping (Total, Sex, Age group, Race and Hispanic origin)

Each row is an estimate of deaths for a specific year, drug category, and population group.

• SUBTOPIC: the drug category (“All drug overdose deaths”)
• GROUP: how the population is split (Total, Sex, Age group)
• SUBGROUP: the specific category inside the group (Male, Female, 35–44 years etc.)
• ESTIMATE_TYPE: whether the rate is age-adjusted or crude
• TIME_PERIOD: year
• ESTIMATE: the rate value (deaths per 100,000)
• STANDARD_ERROR, ESTIMATE_LCI, ESTIMATE_UCI: uncertainty (standard error and 95% CI)
• FLAG: “Warning label” that the estimate may be unreliable (unstable estimate, suppressed value, etc.)
• SUBGROUP_ORDER: recommended order for age groups (helps keep age groups in the right order)

1. Did the death rate for overdoses increase overall in 2019–2022 compared to 2012–2019?
2. Which drugs show as the most used after 2019?
3. Do trends differ by sex?
4. Which age groups have the highest overdose rates post-pandemic?

missing_est <- sum(is.na(raw$ESTIMATE))
flagged <- sum(!is.na(raw$FLAG))

df <- data.frame(
  Observations = num_rows,
  Attributes = num_cols,
  'Beginning Year' = min_year,
  'Latest Year' = max_year,
  'Missing Values' = missing_est,
  'Flagged Observations' = flagged)

verify_kable <- knitr::kable(df, caption = "Data Set Overview")

clean <- raw %>%
  mutate(TIME_PERIOD = as.integer(TIME_PERIOD),
         ESTIMATE = as.numeric(ESTIMATE),
         period = if_else(TIME_PERIOD <= 2019,
                          "Pre-Pandemic",
                          "Post-Pandemic")) %>%
  filter(TIME_PERIOD %in% years) %>%
  filter(!is.na(ESTIMATE)) %>%
  filter(is.na(FLAG))

dup_rows <- clean %>%
  count(SUBTOPIC, ESTIMATE_TYPE, TIME_PERIOD, GROUP, SUBGROUP) %>%
  filter(n > 1) %>%
  nrow()

overall <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
         GROUP == "Total",
         SUBTOPIC == "All drug overdose deaths") %>%
  arrange(TIME_PERIOD)

overall_period <- overall %>%
  group_by(period) %>%
  summarize(mean_rate = mean(ESTIMATE), .groups = "drop")

p_period <- ggplot(overall_period, aes(x = period, y = mean_rate, fill = period)) +
  geom_col() +
  coord_flip() +
  scale_fill_manual(values = c(
    "Pre-Pandemic" = "blue",
    "Post-Pandemic" = "red")) +
  labs(title = "Mean age-adjusted Overdose Rate by Period",
       x = "",
       y = "Mean deaths per 100,000",
       fill = "")

p_period_widget <- ggplotly(p_period, height = 320)

model1 <- lm(ESTIMATE ~ TIME_PERIOD, data = overall)

coef_tbl <- summary(model1)$coefficients
coef_kable <- knitr::kable(coef_tbl, digits = 3, caption = "Regression Coefficients (Overall Trend)")

overall$fitted <- fitted(model1)
overall$resid <- resid(model1)

p_resid <- ggplot(overall, aes(x = fitted, y = resid)) +
  geom_point() +
  geom_hline(yintercept = 0) +
  labs(title = "",
       x = "Fitted Values",
       y = "Residuals")

p_resid_widget <- ggplotly(p_resid, height = 320)

For Total (age-adjusted), this dataset includes opioid-related categories plus the overall overdose category.
A negative percent change means the average rate decreased from pre-pandemic to post-pandemic.

drug_change <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
         GROUP == "Total") %>%
  group_by(SUBTOPIC, period) %>%
  summarize(mean_rate = mean(ESTIMATE), .groups = "drop") %>%
  pivot_wider(names_from = period, values_from = mean_rate) %>%
  mutate(abs_change = `Post-Pandemic` - `Pre-Pandemic`,
         pct_change = abs_change / `Pre-Pandemic` * 100) %>%
  arrange(desc(pct_change))

key_drugs <- c(
  "All drug overdose deaths",
  "Drug overdose deaths involving any opioid",
  "Drug overdose deaths involving synthetic opioids other than methadone",
  "Drug overdose deaths involving heroin")

drug_key_trend <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
         GROUP == "Total",
         SUBTOPIC %in% key_drugs) %>%
  group_by(SUBTOPIC, TIME_PERIOD) %>%
  summarize(rate = mean(ESTIMATE), .groups = "drop")

sex_trend <- clean %>%
  filter(
    ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
    GROUP == "Sex",
    SUBTOPIC == "All drug overdose deaths"
  ) %>%
  arrange(TIME_PERIOD)

For age groups, I used crude rates since “age-adjusted” doesn’t make sense within a single age band.

age_trend <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, crude",
         GROUP == "Age group",
         SUBTOPIC == "All drug overdose deaths")

age_levels <- age_trend %>%
  distinct(SUBGROUP, SUBGROUP_ORDER) %>%
  arrange(SUBGROUP_ORDER) %>%
  pull(SUBGROUP)

age_summary <- age_trend %>%
  group_by(SUBGROUP, period) %>%
  summarize(mean_rate = mean(ESTIMATE), .groups = "drop") %>%
  mutate(SUBGROUP = factor(SUBGROUP, levels = age_levels))

Comparing:

• x = Year
  
• y = All overdose rate (age-adjusted, Total)
  
• z = Synthetic opioid rate (age-adjusted, Total)

all_rate <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
         GROUP == "Total",SUBTOPIC == "All drug overdose deaths") %>%
  select(TIME_PERIOD, all_overdose = ESTIMATE)

syn_rate <- clean %>%
  filter(ESTIMATE_TYPE == "Deaths per 100,000 resident population, age adjusted",
         GROUP == "Total",
         SUBTOPIC == "Drug overdose deaths involving synthetic opioids other than methadone") %>%
  select(TIME_PERIOD, synthetic_opioids = ESTIMATE)

trend3d <- inner_join(all_rate, syn_rate, by = "TIME_PERIOD") %>% arrange(TIME_PERIOD)

• The overall age-adjusted overdose death rate is higher post 2019 and dramatically so up to and after 2020.
• The biggest increase stems from synthetic opioids other than methadone use.
• The data shows that males had an increase use of these drugs when compared to females. However, both groups showed an increase in drug use.
• Post-pandemic crude overdose rates are highest among working-age adults,especially in the 35–44 age range.

• This is a descriptive pre/post pandemic comparison.
• The dataset reports rates per 100,000, not raw death counts.
• Flagged estimates were removed programmatically to reduce instability in small subgroups.
• There is much more data pre-pandemic vs post-pandemic.

National Center for Health Statistics. “DQS Drug overdose death rates, by drug type, sex, age, race, and Hispanic origin: United States from CDC WONDER.” Centers for Disease Control and Prevention, data.cdc.gov, Accessed 19 Apr. 2026. https://data.cdc.gov/National-Center-for-Health-Statistics/DQS-Drug-overdose-death-rates-by-drug-type-sex-age/rdjz-vn2n/about_data