User Guide
Social Health ATLAS brings county-level social determinants and health outcomes into one workspace for data discovery, mapping, association screening, and publication-ready reporting.
Available to everyone
Indicators, data-source documentation, and the User Guide are available without an account.
Account required
Sign in to create, save, reopen, edit, and rerun Map or ExWAS tasks.
SDoH indicators describe social, economic, environmental, and community conditions that can be examined as potential determinants or exposures. Outcome indicators are health endpoints used to describe or model a health result.
A source can contain several indicators, and an indicator can vary by year, geographic coverage, unit, population, or available dimensions. The source card and indicator details are the record of those distinctions.
County-level patterns and ExWAS estimates are associations. They do not establish that one factor caused an outcome, and they should not be interpreted as an individual-level effect.
Check population definitions, time alignment, data completeness, and source limitations before sharing a conclusion. Cite the underlying provider and release shown in the data source card.
Using each dashboard page
Each page answers a different part of a data or analysis question.
Search and filter the catalog to narrow measures by source, category, type, coverage, or year. Open a result to confirm its definition, unit, availability, and latest county-level distribution before using it elsewhere.
Open IndicatorsStart with the large SDoH and Outcome groups to focus the overview. Within a source card, review provider, release, citation, methodology, coverage, and limitations; use its indicator search to find measures in that source.
Open Data SourcesSelect an indicator and an available year, then apply state or demographic filters when the data provides them. Use Compare for two side-by-side maps; inspect a county by selecting it on the map and use the comparison data table for paired county values.
Open Map TasksCreate a named screening task, choose an outcome and predictor years, check data quality, and then run the two-phase analysis. Completed results and report outputs are saved to the signed-in account for later review and editing.
Open ExWAS TasksSigned-in users can save map configurations and complete ExWAS analyses. Reopening an ExWAS analysis restores its result tables, diagnostics, and report without rerunning; edit its setup and rerun to update the same record, or save the revision as a new analysis. Older setup-only records remain reproducible.
Open Saved TasksMap and ExWAS workspaces are shown only after sign-in. This keeps task creation, autosave, saved results, editing, reruns, and exports attached to the correct account. Public catalog and documentation pages remain available when signed out.
Sign inExWAS: screening associations
An exposure-wide association study screens many candidate SDoH indicators against one health outcome. It is designed to surface associations worth investigating further, not to make causal claims.
The default configuration uses the available default outcome, exposure libraries, recommended demographic covariates, 80% coverage threshold, Python backend, the latest available data for each predictor, complete-case analysis, and state adjustment.
The workflow is 1. Set up analysis, 2. Check data quality, then 3. Run ExWAS. Select Check data after reviewing the setup; ExWAS remains unavailable until the current check passes. Changing a setting requires a new check. The selected settings are shown in the resulting study report.
In Libraries mode, choose one data-year rule for the whole screen. In Variables mode, set a year for each selected exposure or covariate. Each selected variable shows its available data years; an individual choice is used directly for that variable. Then adjust coverage requirements, missing-data and imputation controls, predictor transformation, collinearity pruning, state adjustment, exposure sources, and covariates. Data Check and the report show the actual years used for every selected override.
Changes can affect which counties and variables are eligible, so compare the readiness summary and report before comparing runs.
The two-phase design is adapted from Hu et al. (2021), Science of the Total Environment 768:144832. The study report distinguishes the published R 3.6 count-model design from this app's all-Python outcome-specific approximations.
| Outcome type | Model support | Effect shown |
|---|---|---|
| Count | Poisson or negative-binomial, based on dispersion | Rate ratio / multiplicative rate ratio |
| Rate, continuous, index, or currency | Ordinary least squares (OLS) | Coefficient on the outcome scale |
| Percentage or proportion | Fractional logit | Odds ratio |
| Binary or two-level categorical | Logistic regression | Odds ratio |
| Ordered categorical / ordinal | Ordered logistic regression | Cumulative odds ratio |
| Nominal, multiclass categorical | Not currently supported; the run is blocked clearly | No estimate produced |
Readiness: begin with coverage, eligibility, and warnings. They explain what data reached the model.
Phase 1: the screen uses a 50/50 discovery and replication split. A retained hit meets Benjamini–Hochberg FDR q < .05 in both halves.
Phase 2: adjusted estimates are fit jointly for Phase 1 hits. Only exposures that remain p < .05 are labeled final retained findings; use “All Phase 1 hits” to inspect those that did not remain significant. Compare direction, interval, and effect scale—not just the p-value.
Outcome values are not imputed. The primary analysis uses complete cases; optional MICE-style imputation is a single-imputation sensitivity analysis, and all warnings are documented in the study report.
The pipeline aligns county observations across sources and applies its temporal rules before modeling. Review the report's source diagnostics and alignment details when measures come from different releases or years.
A null screen or no replicated hits is a valid analysis result, not necessarily a software failure. It means no variables met the configured retention rule.
Exports and reports
Export controls preserve the visible context of a dashboard view for sharing, reporting, or further analysis.
Quality checks and troubleshooting
These checks resolve the most common reasons a map or analysis does not look as expected.
No map values: try another available year or source, clear a restrictive filter, and check the indicator's county coverage.
ExWAS cannot start: run Check data for the current setup, then review any blocking messages for outcome type, coverage, and compatible data. Nominal multiclass outcomes are intentionally not modeled.
Results differ between runs: compare saved settings, source releases, filters, eligible counties, transformation, and missing-data choices in the study report.