CDC WONDER Mortality + Social Determinants of Health

18 Years of County-Level Deaths Joined with SDOH — 55K+ Records

Every CDC WONDER grouped mortality record from 1999 to 2020 — county × year × age group × sex × race/ethnicity × ICD-10 cause of death — joined with AHRQ's Social Determinants of Health database. Poverty rates, unemployment, uninsurance, primary care access, and a computed deprivation index at the county level. Three tables: a mortality table, an SDOH indicator table, and a pre-joined table ready for ML. From cardiovascular disease and cancer to drug overdose and suicide — the complete epidemiological picture of county-level mortality with its social context.

Sample: Public DomainPaid: Commercial LicenseCSV + ParquetAnnual Updates1999–2016ICD-10 Cause ChaptersDeprivation IndexYPLL Computed3,100+ Counties55K+ Records

Why not download it directly from CDC WONDER?

CDC WONDER is a web query tool, not a bulk data API. Here's what we handled to produce this dataset:

✓ Bulk mortality extraction — CDC WONDER caps query results and requires manual form submissions. We scripted all county × demographic × cause × year combinations and reassembled them into a single flat file.
✓ Suppression flags preserved — NCHS suppresses cells with <10 deaths. We track is_suppressed as a boolean rather than dropping or imputing those rows, preserving the statistical signal.
✓ ICD-10 cause chapters computed — raw WONDER exports use 113 selected-cause codes. We mapped all codes to readable chapter labels (Cardiovascular Diseases, Neoplasms, Drug Overdose / Poisoning, Suicide, etc.) and retained individual ICD-10 codes for fine-grained analysis.
✓ YPLL computed — Years of Potential Life Lost before age 75 calculated per row using age-group midpoints and death counts, a standard epidemiological measure of premature mortality burden.
✓ AHRQ SDOH joined — five annual vintages (2016–2020) of the AHRQ Social Determinants of Health database merged at the county level with nearest-prior-year matching for 1999–2015 records.
✓ Deprivation index computed — composite Z-score from poverty rate, unemployment rate, and uninsurance rate. Classified into quintiles (1 = least deprived, 5 = most deprived) for immediate use as a categorical feature.
✓ USDA rural/urban codes merged — RUCC 1–9 classification (Metropolitan / Micropolitan / Rural) joined at county level for rural health stratification.
✓ Race/ethnicity normalised — CDC's race codes collapsed to a consistent 5-group classification across all years (Non-Hispanic White, Black / African American, Hispanic / Latino, American Indian / Alaska Native, Asian / Pacific Islander).
✓ Pre-joined ML table — mortality and SDOH combined in a single DataFrame so you can start modeling immediately without managing a multi-table join.
✓ Parquet output — columnar storage for fast filtering across 55K+ rows without loading the full CSV.

⏱ Skip the WONDER query assembly, FIPS normalisation, SDOH merging, and deprivation index computation. Ready for analysis in minutes.

55K+

Mortality Records

3,100+

Counties

1999–2016

Years Covered

SDOH Indicators

Use Cases

Epidemiological Research & Health Disparities

Quantify mortality rate disparities by race/ethnicity, age group, and county across 18 years. Examine how cardiovascular disease, cancer, and drug overdose mortality rates diverge by deprivation quintile. Track the opioid epidemic's county-level progression through cause-specific YPLL trends from 1999 to 2020.

Social Determinants of Health Modeling

The pre-joined mortality × SDOH table lets you train models that directly predict mortality rates from poverty, unemployment, uninsurance, and healthcare access — all in a single DataFrame. No manual joins across disparate government data sources. Test whether primary care shortage counties show higher preventable mortality even after controlling for age and race.

Public Health Program Evaluation

Identify counties with persistently high age-adjusted cardiovascular mortality rates despite average or below-average deprivation scores — potential high-value intervention targets. Evaluate natural experiments: counties where major hospital closures correlate with mortality changes across ICD-10 chapters. Benchmark states on avoidable premature mortality (YPLL) by cause.

Insurance & Actuarial Risk Modeling

Build county-level mortality risk models incorporating 20 years of cause-specific rate trends and SDOH covariates. Estimate excess mortality exposure for specific age-sex-race-county cohorts. Support life insurance product pricing, long-term care underwriting, and Medicare Advantage geographic risk adjustment.

AI/ML Health Prediction Applications

Train regression models to predict county-level crude or age-adjusted mortality rates from SDOH features. Classify counties into deprivation × mortality quadrants to identify 'high-need, high-mortality' vs 'high-need, low-mortality' outliers. Build time-series forecasts of cause-specific mortality by county using 18 years of ground truth.

Policy & Advocacy — Rural Health Equity

Rural counties (RUCC 7–9) have systematically higher uninsurance rates, fewer primary care physicians, and higher premature mortality. This dataset puts all those dimensions in one file. Quantify the rural-urban mortality gap by cause and state. Support Medicaid expansion impact analyses by comparing states pre- and post-expansion on uninsurance and cause-specific mortality trends.

Schema — Mortality Table

cdc_wonder_mortality.csv / .parquet — one row per county × year × age × sex × race × cause combination

Field	Type	Description
record_id	int	Unique row identifier
county_fips	string	5-digit county FIPS code (e.g. 06037 = Los Angeles)
state_fips	string	2-digit state FIPS code
county_name	string	County name
state_name	string	State name
year	int	Year (1999–2016)
age_group_label	string	Age group: <1, 1-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84, 85+
age_group_numeric_start	int	Numeric start of age group for sorting
sex	string	Male / Female
race_ethnicity	string	Race/ethnicity: Non-Hispanic White, Black / African American, Hispanic / Latino, American Indian / Alaska Native, Asian / Pacific Islander
cause_icd10	string	ICD-10 underlying cause of death code
cause_label	string	Human-readable cause of death label
cause_chapter	string	ICD-10 chapter grouping (computed): Cardiovascular Diseases, Neoplasms, Respiratory Diseases, etc.
deaths	float	Number of deaths (null where suppressed — <10 deaths per NCHS policy)
is_suppressed	bool	True where death count suppressed for privacy (<10 deaths in cell)
population	float	County population denominator for the year/demographic group
crude_rate	float	Crude mortality rate per 100,000 population
crude_rate_95ci_lower	float	Lower bound of 95% confidence interval for crude rate
crude_rate_95ci_upper	float	Upper bound of 95% confidence interval for crude rate
mortality_tier	string	Very Low / Low / Moderate / High / Very High — quintile classification within cause × year (computed)
years_potential_life_lost	float	YPLL before age 75: deaths × (75 − midpoint age). Quantifies premature mortality burden (computed)

Schema — SDOH Table

cdc_wonder_sdoh.csv / .parquet — one row per county × year (AHRQ SDOH vintages 2016–2020)

Field	Type	Description
sdoh_record_id	int	Unique SDOH row identifier
county_fips	string	5-digit county FIPS code (join key to mortality table)
state_fips	string	2-digit state FIPS code
state_name	string	State name
year	int	SDOH data year (2016–2020, from AHRQ)
poverty_pct	float	% of county population below 185% of the federal poverty level (ACS)
median_household_income	float	Median household income in dollars (ACS)
unemployment_rate	float	Annual county unemployment rate — % (BLS)
uninsured_pct	float	% of county population without health insurance (ACS)
primary_care_per_100k	float	Primary care physicians per 100,000 population (AHRF)
college_pct	float	% of adults with a bachelor's degree or higher (ACS)
housing_cost_burdened_pct	float	% of households spending 30%+ of income on housing (ACS)
food_environment_index	float	USDA food environment index (0–10; higher = better food access)
air_quality_index_avg	float	Average annual Air Quality Index for the county (EPA AQS)
rural_urban_code	string	USDA Rural-Urban Continuum Code 1–9 (1 = large metro, 9 = completely rural)
rural_urban_label	string	Metropolitan / Micropolitan / Rural label (computed from RUCC)
deprivation_index	float	Composite deprivation Z-score: mean of poverty, unemployment, uninsured Z-scores (computed)
deprivation_quintile	int	1 (least deprived) to 5 (most deprived) quintile classification (computed)

The pre-joined table (cdc_wonder_mortality_sdoh_joined.csv / .parquet) combines all mortality columns with SDOH indicators on county_fips, using nearest-prior-year SDOH matching. Use this table to start ML workflows without managing the join yourself.

Quick Start

import pandas as pd

# Load mortality table
mort = pd.read_parquet("cdc_wonder_mortality.parquet")

# Top-10 counties by cardiovascular YPLL (non-suppressed rows only)
cardio = mort[
    (mort.cause_chapter == "Cardiovascular Diseases") &
    (~mort.is_suppressed)
]
top_ypll = (
    cardio.groupby(["county_name", "state_name"])["years_potential_life_lost"]
    .sum()
    .sort_values(ascending=False)
    .head(10)
)
print(top_ypll)

# Load pre-joined table for ML
joined = pd.read_parquet("cdc_wonder_mortality_sdoh_joined.parquet")

# Predict crude mortality rate from deprivation quintile + rural/urban
features = joined[joined.cause_chapter == "Drug Overdose / Poisoning"][[
    "deprivation_quintile", "rural_urban_code",
    "poverty_pct", "uninsured_pct", "crude_rate"
]].dropna()

# Deprivation quintile 5 (most deprived) drug overdose rate vs quintile 1
q5 = features[features.deprivation_quintile == 5]["crude_rate"].mean()
q1 = features[features.deprivation_quintile == 1]["crude_rate"].mean()
print(f"Drug overdose rate ratio Q5/Q1: {q5/q1:.2f}x")

Pairs Well With

FDA FAERS Drug Adverse Events

Join county drug overdose mortality with FAERS adverse event reports by drug class. Validate county-level overdose trends against reported adverse event volume.

CFPB Consumer Financial Complaints

Correlate financial distress (complaint rates by product) with county deprivation index and cardiovascular or mental health mortality rates.

OSHA Workplace Injury & Illness

Layer occupational injury rates with county-level SDOH and mortality to build a holistic county health risk profile for ESG and insurance applications.

Pricing

Sample

Free

1,000 rows (CSV) + full schema documentation

License: Public Domain

Download Sample

Complete

$129

55K+ county x year mortality records, 1999-2016, CSV + Parquet. All 50 states + DC, 3,100+ counties.

License: Commercial License

Buy Complete Dataset

Annual

$249/yr

Complete dataset + annual updates as CDC refreshes mortality data

License: Commercial License

Subscribe Annual

Data Provenance

Mortality: CDC WONDER Compressed Mortality File, Underlying Cause of Death 1999–2016 (ICD-10). Source: National Center for Health Statistics (NCHS), CDC. Public domain — U.S. government work.
SDOH indicators: AHRQ Social Determinants of Health Database, vintages 2016–2020. Includes ACS (Census Bureau), BLS, AHRF (HRSA), USDA, HUD, EPA sources. All public domain.
Rural/urban classification: USDA Economic Research Service Rural-Urban Continuum Codes 2013. Public domain.
ClarityStorm processing: ICD-10 chapter mapping, YPLL computation, deprivation index, SDOH join, suppression flag preservation, and Parquet conversion. Pipeline source code included in purchase.
License: Free sample is public domain. ClarityStorm Commercial Data License covers paid tiers — internal use, no redistribution or resale of raw data.