The National Health and Nutrition Examination Survey (NHANES) has tested for Mycobacterium tuberculosis infection three times: in 1971–1972, 1999–2000, and 2011–2012. Based on tuberculin skin test results, the estimated national prevalence of latent tuberculosis infection (LTBI) among adults was 11%–18% in 1971–1972 but has remained ≤6% in subsequent NHANES cycles (1–4). A single 2-year NHANES cycle is designed to produce accurate and stable estimates for conditions of ≥10% prevalence in the noninstitutionalized civilian U.S. population (5–7), suggesting that NHANES might no longer be as nationally representative for LTBI as it is for more common health conditions. Approximately 30 counties were selected for each 2-year cycle (5). We wished to examine whether persons in selected counties might have been systematically more or less likely to have a positive tuberculin skin test than their counterparts in the approximately 3,100 counties that were not selected for NHANES participation.

We created a non-NHANES dataset with demographic profiles and tuberculosis data for all 3,143 U.S. county equivalents (Table 1). The U.S. Census Bureau and Department of Agriculture websites provided each county’s population size and racial/ethnic composition, rural vs. urban classification, and poverty prevalence for 1970 through 2013. The National Tuberculosis Surveillance System provided annual tuberculosis disease incidence, with the U.S. Census Bureau’s Current Population Survey providing county population denominators.

We also used genotyping results to derive an estimate of longstanding LTBI prevalence for each county. Briefly, this simple back-calculation method assumed that TB cases not attributed to recent transmission (i.e., based on genotyping results) instead arose from preexisting LTBI. Then a 0.1% annual risk of reactivation was used to derive an estimated number of county residents with untreated LTBI (8). This county-level LTBI estimation method has not been validated; the only potential comparison in the literature is based on 1958‒1965 data (9). However, median estimated LTBI prevalence among the U.S.-born was 0.7% (lower and upper quartiles: 0.4%, 1.3%) and among the non-U.S.-born was 13.1% (lower and upper quartiles: 8.8%, 18.5%), both of which were similar to previous national NHANES-based estimates (1, 3, 4).

Because of disclosure risk, county of residence is not included in the NHANES public-use datasets and cannot be released (10). Following Research Data Center procedures (https://www.cdc.gov/rdc/), the National Center for Health Statistics merged masked NHANES 1971–1972, 1999–2000, and 2011–2012 data with our non-NHANES county dataset and allowed us to conduct this geographic analysis without access to county identifiers.

The demographics of the counties selected for NHANES participation in 1971–1972, 1999–2000, and 2011–2012 were similar across time, except that the counties selected for NHANES 2011–2012 had higher proportions of residents living in poverty, consistent with the recent national trend (Table 1). We compared the selected to nonselected counties within strata of counties of similar population size, racial/ethnic composition, rural vs. urban classification, and poverty prevalence. Within those strata, mean tuberculosis disease incidence and estimated LTBI prevalence were similar (P > 0.05 for each test of equivalence) when comparing the selected to the nonselected counties in NHANES 1971–1972, 1999–2000, and 2011–2012. In 90% of the counties selected for NHANES 2011–2012, the unweighted prevalence of a tuberculin skin test result of ≥10 mm of induration was within 1% of the genotyping-derived LTBI prevalence estimate for that county (8).

Because of strict confidentially protections that include the geographic locations of NHANES participants, the results of this masked analysis cannot be shown in more detail (10). However, our findings reinforce confidence in national LTBI prevalence estimates based on past NHANES cycles (1–4). We found no evidence that the selected counties had different tuberculosis disease incidence or different LTBI prevalence from the counties not selected for NHANES participation in 1971–1972, 1999–2000, and 2011–2012.

Despite these reassuring findings, both the low prevalence and geographic heterogeneity of this condition in the United States suggest that incorporating tuberculosis components (e.g., interferon-gamma release assays) into future NHANES cycles for >2 consecutive years would help achieve more stable population estimates (4–7). Our findings also imply that genotyping-derived estimates of county-level LTBI prevalence that do not rely on NHANES data could continue to prove useful in the future (8).