RaayGunz/county-health-maternal-clean

County Health — Maternal / Infant Health (Cleaned)

A cleaned, modeling-ready county-level dataset for predicting low birthweight from social and environmental factors. Derived from County Health Rankings & Roadmaps (CHR&R) 2025.

What's in it

• —3,035 U.S. counties, 19 columns
• —Target: low_birthweight_pct (percent of live births under 2,500g)
• —Binary target: high_lbw (above/below median split)
• —Features: air pollution (PM2.5), drug overdose deaths, firearm fatalities, adult smoking, children in poverty, median household income, percent rural, uninsured percent
• —Engineered missingness flags: homicidesuppressed, drugoverdosemissing, firearmmissing, airpollutionpm25_missing
• —Log transforms: drugoverdosedeathslog, firearmfatalities_log

A full data dictionary is included as chr_data_dictionary.csv.

Cleaning summary

• —Promoted the 2025 file's stacked measure-code header; dropped US/state rollup rows
• —Unit reconciliation: four proportion columns x100 to percent scale
• —homicide_rate (53% missing, MNAR suppression in small rural counties): dropped continuous column, kept a binary suppression flag
• —drug_overdose, firearm, air_pollution: median-imputed with missingness flags
• —Dropped 8 Connecticut counties (Census replaced CT counties with planning regions; county-level values are blank)
• —Final frame certified: 0 missing values

Results (in the accompanying analysis)

• —Regression: R^2 0.48 vs. dummy floor ~0.00 (real signal; varies by region, fold R^2 0.19-0.56)
• —GaussianNB classification: accuracy 0.76 vs. dummy floor 0.50, balanced across classes (recall low 0.79 / high 0.73)

Limitations

• —Features are correlated (income/poverty/smoking cluster); this is a predictive model, not a causal one - individual effects cannot be isolated
• —~52% of variation is unexplained (healthcare access, genetics, individual behavior not included)
• —County-level associations only; not individual-level or causal claims

License & Source

Derived from County Health Rankings & Roadmaps 2025 (University of Wisconsin Population Health Institute). Source data is subject to the CHR&R Terms of Use (https://www.countyhealthrankings.org/terms-use), which prohibit commercial use - including training AI models offered as commercial products - without CHR&R's express written consent.

This cleaned/derived dataset is shared for educational and non-commercial research purposes only. Users must comply with CHR&R's original terms. For commercial use, contact info@countyhealthrankings.org.

Citation: University of Wisconsin Population Health Institute. County Health Rankings & Roadmaps 2025. countyhealthrankings.org