Rural vs. Metropolitan Health Disparities Across Texas Counties

CDC has reported higher age-adjusted death rates from leading causes, including heart disease, in nonmetropolitan than in metropolitan areas of the United States,^[3] and higher lung cancer incidence in nonmetropolitan counties.^[4] Texas spans 254 counties, from counties in metropolitan areas of more than a million people to counties with a few dozen residents, so it is a useful place to ask whether the national pattern appears in current county data.

This report uses USDA’s county-level metropolitan and nonmetropolitan categories. They are not the same as “urban” and “rural”: USDA notes that most counties, metropolitan or not, contain both urban and rural populations, and that metropolitan areas include substantial rural territory.^[2] The word “rural” in the title follows common usage; the analysis itself compares metropolitan and nonmetropolitan counties.

Vellora’s earlier report Geographic Variation in Cardiovascular Health Across Texas Counties compared the two groups on eight cardiovascular measures. This study reuses some of the same verified data and reproduces those comparisons exactly where they overlap, and adds an observed cancer outcome, a prespecified primary family, effect sizes in each measure’s own units with confidence intervals, an analysis of missing data and sensitivity analyses.

Research question: How do selected health outcomes and risk factors differ between metropolitan and nonmetropolitan Texas counties?

2.1 Groups

Counties were classified with the USDA Economic Research Service Rural-Urban Continuum Codes 2023 (RUCC),^[1] matched on five-digit FIPS codes: codes 1–3 metropolitan and 4–9 nonmetropolitan. The stored codes were checked against a fresh download of the USDA file on 2026-10-23: all 254 Texas rows matched, giving 86 metropolitan and 168 nonmetropolitan counties (Table 1).

2.2 Indicators

Eight indicators were selected for relevance, county coverage and data quality (Table 2). The primary family contains the two observed rates, counted from death certificates and the cancer registry. The six PLACES measures form a secondary family because they are small-area model estimates whose inputs include county demographics, so part of any difference between the groups may come from the model itself.^[9] Stroke prevalence and other PLACES and cancer measures were not selected, to keep the comparison focused. Crude and age-adjusted values are never combined.

Age adjustment. The death rate and the cancer rate are adjusted to the 2000 U.S. standard population (adults 35 and over for deaths, all ages for cancer).^{[5, 7]} PLACES age-adjusted estimates cover adults 18 and over; CDC’s PLACES documentation does not name their standard population.^{[9, 10]} Each indicator is compared only with itself across the two groups, never with another indicator.

2.3 Statistical analysis

The analysis followed a plan committed to the project’s version history before any group difference was computed (see Reproducibility). Differences are always nonmetropolitan minus metropolitan. For each indicator and group we report the number of counties with a published value, the number suppressed, the median, interquartile range (IQR), mean, standard deviation and range.

• Difference in medians: the nonmetropolitan median minus the metropolitan median.
• Hodges–Lehmann shift: the median of every difference between a nonmetropolitan and a metropolitan county, a robust estimate of how far one distribution sits from the other, with a distribution-free 95% confidence interval.^{[12, 13]}
• Probability of superiority: the chance that a randomly chosen nonmetropolitan county has a higher value than a randomly chosen metropolitan county (0.5 means no tendency either way), and the equivalent rank-biserial correlation (2 × probability − 1).
• Mann–Whitney U test, two-sided, with tie and continuity corrections.^[11] p-values are adjusted with the Benjamini–Hochberg procedure separately within the primary family (2 tests) and the secondary family (6 tests).^[14] Both raw and adjusted values are reported.

A small p-value says a difference is unlikely to be due to chance alone, not that it is large or important; a non-significant result does not show that the groups are the same. Counties are unweighted. A population-weighted analysis was not performed: the values are age-adjusted rates and model-based prevalence estimates, and weighting them by 2020 population would not produce a valid pooled rate without age-specific counts, which are not in the verified data.

2.4 Missing data and exploratory analyses

Suppressed values were excluded and listed with the source’s reason; nothing was imputed or replaced with zero. For each indicator with suppressed values, Fisher’s exact test compares how often values are suppressed in each group. Three exploratory analyses (unadjusted p-values, not used for the conclusions) were planned: E1 repeats the comparison with crude rather than age-adjusted PLACES prevalence; E2 correlates each indicator with the RUCC code from 1 to 9 (Spearman); E3 compares nonmetropolitan counties adjacent to a metropolitan area (codes 4, 6, 8) with nonadjacent ones (5, 7, 9).

3.1 Counties and sample sizes

The 86 metropolitan counties held 90% of Texans in 2020 (26,249,929 people; median county 92,260.5), and the 168 nonmetropolitan counties 10% (2,895,576; median county 11,480.5). Every indicator except lung cancer incidence was published for all counties but Loving County (nonmetropolitan). Table 3 gives each indicator’s own sample sizes.

3.2 Primary family: observed rates

Heart disease deaths. The median nonmetropolitan county’s death rate was 415.1 per 100,000, against 361.2 per 100,000 for metropolitan counties. The Hodges–Lehmann shift was +45.4 per 100,000 (95% CI +25.7 to +65.0), and a randomly chosen nonmetropolitan county had the higher rate with probability 0.67 (Mann–Whitney p < 0.0001; adjusted p < 0.0001).

Lung cancer incidence. Among counties with a published rate, the medians were 52.9 (nonmetropolitan, n = 125) and 50.8 (metropolitan, n = 78) per 100,000. The shift was +3.5 per 100,000 (95% CI −0.9 to +7.9); the interval includes zero and the difference was not statistically significant (adjusted p = 0.1340). This does not show that the groups have the same incidence: see 3.4 on suppression.

3.3 Secondary family: CDC PLACES estimates

All six PLACES age-adjusted estimates were higher in nonmetropolitan counties, with confidence intervals for the shift above zero and adjusted p ≤ 0.0033 (Table 4, Figure 2). The largest shifts in percentage points were for no leisure-time physical activity (+2.7 percentage points (95% CI +1.5 to +3.9)) and current smoking (+1.8 percentage points (95% CI +1.2 to +2.3)); by probability of superiority the clearest separations were coronary heart disease (0.73) and current smoking (0.73).

3.4 Missing data

Lung cancer incidence is suppressed when a county had fewer than 16 cases over 2019–2023. That affected 43 of 168 nonmetropolitan counties (26%) and 8 of 86 metropolitan counties (9%), a difference unlikely to be due to chance (Fisher’s exact p = 0.0026). Suppressed counties are small: median 2020 population 2,547 (largest 7,694), against 24,435 for counties with a rate. The nonmetropolitan lung cancer comparison therefore describes mainly the larger nonmetropolitan counties, and its result could differ if the smallest counties could be included. For the other seven indicators only Loving County (population 64) was suppressed.

3.5 Exploratory analyses

E1, crude prevalence. With crude rather than age-adjusted PLACES prevalence, every shift remained positive and most were larger (for example coronary heart disease +1.5 against +0.5 percentage points), consistent with nonmetropolitan counties having older adult populations; this is why age-adjusted values are the main analysis.

E2, gradient. Every indicator was positively correlated with the RUCC code (ρ 0.11 to 0.31), so values tended to rise from the most metropolitan to the least populated, least connected counties; the correlation for diabetes was not statistically significant.

E3, adjacency. Among nonmetropolitan counties, those adjacent to a metropolitan area did not differ clearly from nonadjacent ones on any indicator (all p > 0.05; heart disease deaths shift −0.3, 95% CI −23.9 to +22.1 per 100,000).

In current Texas data, nonmetropolitan counties tended to have higher heart disease death rates and higher modelled prevalence of coronary heart disease, high blood pressure, diabetes, obesity, smoking and physical inactivity than metropolitan counties. The direction of the heart disease finding matches national CDC reports.^[3] Because the death rate is age-standardized, the difference is not simply a result of nonmetropolitan counties having older residents, but this study cannot say what does explain it: income, education, insurance, occupation, access to care and many other factors differ between the groups and are not analysed here.

The lung cancer comparison did not show a clear difference, unlike national findings of higher incidence in nonmetropolitan counties.^[4] It should be read as inconclusive rather than as evidence of no difference: the confidence interval is wide, and a quarter of nonmetropolitan counties had no published rate, so the comparison mostly describes the larger nonmetropolitan counties. The PLACES differences are consistent with the death-rate finding but are weaker evidence on their own, because the estimates come from a model that uses county demographics.

The differences are between distributions of counties. A probability of superiority of 0.67 for heart disease deaths means that the groups overlap substantially: many nonmetropolitan counties have lower rates than many metropolitan ones. Within the nonmetropolitan group, adjacency to a metropolitan area made no clear difference.

Every number in this report, its PDF and its figures is read from one results file, generated by the analysis code from Vellora’s verified copies of the source files and following the analysis plan, which was committed before any group difference was computed. The analysis is deterministic. An independent Python implementation using SciPy and NumPy recomputed every group size, descriptive statistic, Hodges–Lehmann interval, Mann–Whitney test, adjusted p-value, Fisher test and exploratory result, and matched the results file. Where the measures overlap, the group medians and Mann–Whitney tests also match the earlier cardiovascular report exactly. Related: Smoking Prevalence and Lung Cancer Incidence Across Texas Counties.

• Analysis plan (Markdown)Indicators, families, methods and sensitivity analyses, committed before any group difference was computed
• Results (JSON)Every statistic reported here, the suppressed counties with their reasons, and the per-county values in Figure 1
• Analysis code (TypeScript)The analysis that produced the results file
• Independent check (Python)Recomputes every statistic with SciPy and NumPy
• PLACES inputs (JSON)Vellora’s verified copy of CDC PLACES 2025 age-adjusted estimates; crude values are in /data/places-tx.v2025.1.json
• Heart disease mortality input (JSON)Vellora’s verified copy of CDC county heart disease death rates
• Lung cancer input (JSON)Vellora’s verified copy of State Cancer Profiles county incidence
• County classification input (JSON)USDA ERS RUCC 2023 codes and 2020 Census populations

To reproduce: regenerate the results with npm run research:rural and check them with uvx --with scipy --with numpy python scripts/verify-rural-study.py. The input files record the source releases and download dates; the sources may revise their data.