Geo-Epidemiologic and Molecular Characterization to Identify Social, Cultural, and Economic Factors Where Targeted Tuberculosis Control Activities Can Reduce Incidence in Maryland, 2004–2010

Abstract

Objectives

Tuberculosis (TB) disproportionately affects members of socioeconomically disadvantaged and minority populations in the U.S. We describe the geospatial distribution of TB cases in Maryland, identify areas at high risk for TB, and compare the geospatial clustering of cases with genotype clustering and demographic, socioeconomic, and TB risk-factor information.

Methods

Addresses of culture-positive, genotyped TB cases reported to the Maryland Department of Health and Mental Hygiene from January 1, 2004, to December 31, 2010, were geocoded and aggregated to census tracts. Geospatial clusters with higher-than-expected case numbers were identified using Poisson spatial cluster analysis. Case distribution and geospatial clustering information were compared with (1) genotype clustering (spoligotypes and 12-locus MIRU-VNTR), (2) individual-level risk and demographic data, and (3) census tract-level demographic and socioeconomic data.

Results

We genotoyped 1,384 (98%) isolates from 1,409 culture-positive TB cases. Two geospatial clusters were found: one in Baltimore City and one in Montgomery and Prince George's counties. Cases in these geospatial clusters were equally or less likely to share genotypes than cases outside the geospatial clusters. The two geospatial clusters had poverty and crowding in common but differed significantly by risk populations and behaviors.

Conclusions

Genotyping results indicated that recent transmission did not explain most geospatial clustering, suggesting that geospatial clustering is largely mitigated by social determinants. Analyses combining geospatial, genotyping, and epidemiologic data can help characterize populations most at risk for TB and inform the design of targeted interventions.

In the United States, tuberculosis (TB) incidence is highly reflective of health disparities in the underlying population. TB infection and disease are most common among people of low socioeconomic status and education, and in minority (particularly foreign-born, black, and Hispanic) populations. High-risk groups for TB also include homeless, drug and alcohol abusing, HIV-positive, and incarcerated populations.^1,2

Maryland reported 220 new TB cases in 2010 (3.8 cases per 100,000 population) and ranked 13th among the 50 states in cases per capita.³ The continued growth of the foreign-born population in Maryland (up 28% from 2000 to 2006⁴), the state's high rate of human immunodeficiency virus (HIV)/acquired immunodeficiency syndrome (AIDS) cases (19.9 AIDS diagnoses per 100,000 population vs. 11.2 diagnoses per 100,000 population nationally in 2009⁵), and the high rates of homelessness and illegal drug use in Baltimore impact local TB rates and distribution. Characterizing the epidemiologic and geospatial distribution of TB cases in Maryland with regard to these risk factors is important both to identify geographic areas in which the populations are at highest risk for TB and to better target local TB prevention activities in collaboration with community-focused efforts to reduce health disparities in high-risk populations.

Molecular epidemiology has become an important tool in TB research and program planning. Use of IS6110-based restriction fragment length polymorphism (RFLP), spoligotyping, and mycobacterial interspersed repetitive units-variable number of tandem repeat (MIRU-VNTR) assays has helped to characterize genetic variation in Mycobacterium tuberculosis, the organism that causes TB. Individuals infected with organisms having identical or similar genotype patterns are defined as being genotypically “clustered.” Because strains from a recent common ancestor are similar genetically, clustering may indicate groups of cases that were recently exposed to each other or a common source.^6,7

Geographic information systems (GISs) have similarly become important tools in TB research and program planning. GISs can be used to map disease rates, define populations at risk, detect clusters of disease located together in space, and map social and environmental risk factors.^8–12

Several research groups have combined TB genotyping and geospatial data to characterize the disease's distribution and transmission within communities.^13–17 Geospatial clusters identify populations at increased risk of disease due to geographic proximity, which may correspond to recent transmission from one or more individuals to their contacts, reactivation of latent TB infection in an aggregate of individuals infected elsewhere, or both. Because genotype clustering is used as a rough proxy for recent transmission, geospatial clusters in which cases are genotypically clustered may indicate locations where recent transmission is occurring.

Combining these two approaches has shown varying results in different settings. Some researchers first characterized genotype clustering and then determined whether genotypically clustered cases were also geospatially clustered.^14–16 In some studies, cases in genotype clusters were likely to live near each other,¹⁶ while in others, they were not.¹⁴ Conversely, other researchers first characterized geospatial clustering and then determined whether geospatially clustered cases were also genotypically clustered^13,17 to identify sites of ongoing transmission. Differences in the findings of these studies indicate that different cities, regions, and states have distinct patterns of transmission and incidence, highlighting the utility of performing this type of analysis in different geographic locations.

The purpose of our study was to characterize the geospatial distribution of incident TB cases reported to the Maryland Department of Health and Mental Hygiene (DHMH) from 2004 to 2010, and to determine to what extent geospatial clustering overlaps with genotype clustering and with TB risk factors. Three research questions were addressed:

1. Is there a pattern in the geospatial distribution of TB cases indicative of population-based social health inequalities and non-constant TB risk?

2. Can the geospatial clustering of TB cases be explained by the presence of genotype clusters?

3. Can the geospatial clustering of TB cases be explained by individual- and group-level demographic, socioeconomic, and risk-factor characteristics?

METHODS

Data sources

The DHMH's TB database contains demographic, clinical, risk-factor, and genotyping information for all reported TB cases in Maryland. Specimens from culture-positive cases were genotyped by two polymerase chain reaction methods: spoligotyping and 12-locus MIRU-VNTR; for case isolates obtained after 2008, 24-locus MIRU was also performed. Genotype clusters were defined as groups of two or more cases with identical spoligotypes and 12-locus MIRU-VNTR copy numbers. This analysis included all culture-positive, genotyped TB cases reported to the DHMH from January 1, 2004, to December 31, 2010.

Individual-level information extracted from the DHMH database included year diagnosed; age; sex; race/ethnicity; foreign birth; disease site (pulmonary vs. extrapulmonary); injecting/non-injecting drug use, alcohol abuse, and homelessness within the past year; HIV status; and long-term care and correctional facility residence at the time of diagnosis.

We linked census tract-level data¹⁸ including Topologically Integrated Geographic Encoding and Referencing (known as “TIGER”) shapefiles of the 2010 census tract boundaries, population counts from the 2010 U.S. Census, and demographic and socioeconomic data from the 2006–2010 American Community Survey, by census tract identification number. Demographic and socioeconomic data included median annual household income, percentage of families and people whose income in the past 12 months was below the federal poverty level, percentage of the population who were unemployed, percentage of the population aged 25 years and older who were high school graduates, percentage of households with ≤1 occupant per room, percentage of the population born outside the U.S., and race/ethnicity.

Geospatial analyses

We geocoded patient addresses using the Juice Analytics Excel Geocoding Tool,¹⁹ imported geocoded addresses into ArcGIS^® version 10.0,²⁰ and joined them to the shapefiles to determine the number of cases and disease incidence distribution per census tract.

We identified geospatial clusters of TB cases based on excess census tract incidence rates in SaTScan^™ version 9.1.1²¹ using the discrete Poisson model for purely spatial clusters, which is based on Kuldorff's scan statistic.²² In this model, “expected” case counts are determined based on the Poisson assumption that cases are randomly spatially distributed. We uploaded case and population counts per census tract to SaTScan along with coordinates specifying each census tract's geographic center and set the maximum spatial cluster size to a radius of 40 kilometers (km).

Statistical and epidemiologic analyses

We linked genotype clustering status to geospatial clustering information to compare rates of genotype clustering within and outside geospatial clusters. We also linked geospatial clustering information to individual-level and census tract-level data to compare individual-level parameters of cases that were geospatially clustered only, genotypically clustered only, geospatially and genotypically clustered, and non-clustered; and to compare census tract-level parameters of census tracts that were geospatially clustered and non-geospatially clustered. We created choropleth maps of the census tract-level parameters.

Stata^® 11²³ was used for all statistical analyses. We compared categorical variables using two-sample t-tests for continuous variables, Pearson's Chi-square test for dichotomous variables with expected cell values of >5, and Fischer's exact Chi-square test for dichotomous variables with expected cell values of ≤5. For all analyses, statistical significance was defined as p<0.05.

RESULTS

Geospatial analyses

From January 2004 to December 2010, 1,836 cases of TB were reported to DHMH. Of these cases, 427 (23%) were culture-negative and 25 (1%) were culture-positive but had no available genotyping results, leaving a total of 1,384 (75% of all cases and 98% of culture-positive cases) culture-positive, genotyped cases. Culture-negative cases were significantly more likely than culture-positive cases to be diagnosed in 2004, female, U.S.-born, younger (i.e., the mean age for culture-negative vs. culture-positive cases was 36.5 and 44.8 years, respectively), of unknown HIV status, and residents of a correctional facility; live outside Montgomery County; and have extrapulmonary TB. Culture-positive cases without genotyping results were significantly more likely than culture-positive cases with genotyping results to be diagnosed in a year other than 2010 and to have extrapulmonary TB (Table 1).

Table 1.

Characteristics of TB cases by culture and genotyping status: Maryland, 2004–2010

^aStatistically signficant difference (p<0.05) between culture-positive and culture-negative cases

^bStatistically significant difference (p<0.05) between culture-positive, genotyped cases and culture positive, non-genotyped cases

^cIncludes Anne Arundel, Baltimore, and Howard counties

^dIncludes Allegany, Calvert, Caroline, Carroll, Cecil, Charles, Dorchester, Frederick, Garrett, Harford, Kent, Queen Anne's, St. Mary's, Somerset, Talbot, Washington, Wicomico, and Worchester counties

^eOther race was predominantly Asian (n=502/505).

TB = tuberculosis

SD = standard deviation

HIV = human immunodeficiency virus

A choropleth map of the TB case rates by census tract (Figure 1) shows that the highest TB rates were seen in and around Baltimore City; between Baltimore and Washington, D.C.; in Montgomery and Prince George's counties; and on the Eastern Shore. The distribution of TB cases by number of census tracts in all of Maryland was similar to a Poisson distribution, validating our use of the Poisson model for the geospatial cluster analysis (these data have been reported elsewhere²⁴). The median incidence was 3.8 cases per 100,000 person-years for census tracts in Baltimore City (range: 0.0–29.2), 5.1 cases per 100,000 person-years for census tracts in Montgomery County (range: 0.0–33.6), and 3.5 cases per 100,000 person-years for census tracts in Prince George's County (range: 0.0–41.1).

Figure 1.

Geospatial clusters of TB cases and case rates: Maryland, 2004–2010^a

^aGeospatial clusters were defined using the SaTScan discrete Poisson model for purely spatial clusters, with the maximum spatial cluster size set to a radius of 40 kilometers.

TB = tuberculosis

Many census tract-level characteristics, including poverty and foreign birth, showed geographic distributions similar to the distribution of TB incidences. Figures 2 and 3 show poverty rates and rates of foreign-born population by census tract, respectively, with the high TB case-rate census tracts (>10 cases per 100,000 person-years) highlighted. In the greater Baltimore area, Montgomery and Prince George's counties, and the southern portion of the Eastern Shore, high TB incidence generally coincided with high unemployment rates and poverty levels, and low median incomes and education levels. Census tracts in western Maryland also showed these socioeconomic characteristics, but had low TB incidences. Statewide, census tracts with higher TB incidences also tended to have higher rates of crowding and foreign-born, Hispanic, black, and other nonwhite populations than census tracts with lower TB case rates. However, some census tracts with high TB rates did not follow all these patterns. Also, some census tracts exhibited these demographic characteristics but did not have high TB rates, most notably in southeastern Baltimore City and parts of Prince George's, Montgomery, Howard, and Baltimore counties.²⁴

Figure 2.

Percentage of families and people whose income in the past 12 months was below the federal poverty level: Maryland census tracts, American Community Survey, 2006–2010^a

^aHigh-case-rate census tracts, in which the TB case rate was >10 per 100,000 person-years during the study period, are highlighted.

TB = tuberculosis

Figure 3.

Percentage of the population that is foreign-born: Maryland census tracts, American Community Survey, 2006–2010^a

^aHigh-case-rate census tracts, in which the TB case rate was >10 per 100,000 person-years during the study period, are highlighted.

TB = tuberculosis

Identification of geospatial clusters

We found two statistically significant geospatial clusters, shown in Figure 1, overlaying the census tract TB incidences. The first geospatial cluster (designated the Montgomery/Prince George's Cluster) encompassed 176 census tracts in Montgomery and Prince George's counties, with a radius of 12.7 km and a population of 751,533. The number of expected cases in the geospatial cluster was 180.2, and the number of observed cases was 473.0 (p<0.001; median cases per census tract: 2; range: 0–12). The second cluster (designated the Baltimore Cluster) encompassed 96 census tracts in Baltimore City, with a radius of 4.8 km and a population of 245,120. The number of expected cases was 58.8, and the number of observed cases was 127 (p<0.001; median cases per census tract: 1; range: 0–5).

Compared with cases not in a geospatial cluster, cases in the Montgomery/Prince George's Cluster were more likely to be black/African American, Hispanic/Latino, foreign-born, younger (i.e., the mean age for cases in the Montgomery/Prince George's Cluster vs. cases not in a geospatial cluster was 40.7 vs. 46.8 years), and HIV-positive, but less likely to be diagnosed in 2010 or to report drug or excess alcohol use. Census tracts in the Montgomery/Prince George's Cluster had higher rates of unemployment, poverty, and crowding (>1 occupant per room); lower education levels; and, consistent with individual-level data, greater proportions of foreign-born and Hispanic/Latino populations compared with census tracts not in a geospatial cluster (Table 2).

Table 2.

Characteristics of TB cases and census tracts by geospatial clustering: Maryland, 2004–2010

^aStatistically significant difference between cases/census tracts in the Baltimore Cluster and cases/census tracts not in a geospatial cluster

^bStatistically significant difference between cases/census tracts in the Montgomery/Prince George's Cluster and cases/census tracts not in a geospatial cluster

TB = tuberculosis

SD = standard deviation

HIV = human immunodeficiency virus

Compared with cases not in a geospatial cluster, cases in the Baltimore Cluster were more likely to be diagnosed in 2004 or 2005, male, black/African American, U.S.-born, HIV-positive, and homeless; to reside in correctional and long-term care facilities; to report drug and excess alcohol use; and to have pulmonary TB. Census tracts in the Baltimore Cluster had higher rates of unemployment and poverty, lower median incomes, higher rates of crowding, and lower education levels compared with census tracts not in a geospatial cluster; however, in contrast with the Montgomery/Prince George's Cluster, these census tracts had smaller proportions of foreign-born and larger proportions of black residents compared with census tracts not in a geospatial cluster (Table 2).

Comparing geospatial and genotype clusters

Overall, 595 of the 1,384 culture-positive, genotyped cases (43%) were in 137 genotype clusters (median cases per cluster: 2; range: 2–36). Genotypically clustered cases were more likely to be black/African American (p=0.002), U.S.-born (p<0.001), HIV-positive (p=0.04), and homeless (p<0.001); to report injection drug use (p=0.02), non-injection drug use (p<0.001), and excess alcohol use (p=0.003); and to have pulmonary TB (p=0.01) compared with non-genotypically clustered cases (data not shown).²⁴

Of the 784 cases that did not fall into one of the two geospatial clusters, 311 (40%) belonged to 85 genotype clusters (median cases per cluster: 2; range: 2–19). In the Montgomery/Prince George's Cluster, only 135 of 473 (29%) cases were genotypically clustered with at least one other case in the geospatial cluster (p<0.001 compared with cases not in a geospatial cluster). These cases fell into 41 genotype clusters (median cases per cluster: 2; range: 2–14). In contrast, 56 of 127 (44%) cases in the Baltimore Cluster were genotypically clustered with at least one other case in the geospatial cluster (p=0.89 compared with cases not in a geospatial cluster). These cases fell into 13 genotype clusters (median cases per cluster: 2; range: 2–26). Although most genotype clustering fell within or outside the two geospatial clusters, there was some genotype clustering across geospatial clusters. Thirty-five (4%) cases not in a geospatial cluster were genotypically clustered with cases in one of the two geospatial clusters. Thirty-five (7%) cases in the Montgomery/Prince George's Cluster and 23 (18%) cases in the Baltimore Cluster were genotypically clustered with cases in the other geospatial cluster or cases not in a geospatial cluster (data not shown).

For cases not in a geospatial cluster, genotypically clustered cases were more likely to be diagnosed in 2009, black/African American, U.S.-born, HIV-positive, and homeless; to report non-injection drug and excess alcohol use; and to have pulmonary TB compared with non-genotypically clustered cases (Table 3). In the Montgomery/Prince George's Cluster, characteristics of genotypically clustered cases did not differ significantly from those of non-genotypically clustered cases. In the Baltimore Cluster, genotypically clustered cases were more likely than non-genotypically clustered cases to be male, non-Hispanic, U.S.-born, and homeless, and to report drug use.

Table 3.

Characteristics of TB cases by genotype clustering and geospatial clustering: Maryland, 2004–2010

^aStatistically significantly different from cases not in a geospatial cluster that do not share a genotype with another case not in a geospatial cluster

^bStatistically significantly different from cases in the Baltimore Cluster that do not share a genotype with another case in the Baltimore Cluster

TB = tuberculosis

SD = standard deviation

HIV = human immunodeficiency virus

DISCUSSION

The results of these analyses help to characterize the epidemiologic and geospatial distribution of TB in Maryland and identify locations where TB affects minority groups, the poor, and other disenfranchised populations, recognizing the social health inequalities in the state. A better understanding of the geography and the epidemiologic factors related to the incidence and spread of TB can help to inform and target interventions to reduce its burden throughout Maryland.

As shown in Figure 1, the census tracts with the highest TB rates were located in and around Baltimore City; along the I-95 corridor between Baltimore and Washington, D.C.; in Montgomery and Prince George's counties; and on the Eastern Shore. These census tracts had different levels of demographic and socioeconomic variables, depending on their geographic location, indicating that high TB incidences are being driven by different disparities in different geographic areas. This finding highlights the importance of characterizing these risk groups separately. In particular, the two geospatial clusters in Baltimore City and Montgomery/Prince George's counties had different genotyping, demographic, risk-factor, and socioeconomic characteristics contributing to the TB burden. Local programs must focus on the highest-risk census tracts and on factors pertinent to their coverage areas.

Cases in the Montgomery/Prince George's Cluster were significantly less likely to be genotypically clustered than were non-geospatially clustered cases, correlating with the fact that these cases were almost exclusively foreign-born. Presumably, many of these cases acquired TB in their home countries and it was reactivated after they moved to Maryland, as opposed to acquiring it in Maryland recently. Cases in this geospatial cluster that were genotypically clustered were not significantly different from those that were not, further supporting this hypothesis of little recent transmission. Although our data showed that residents of census tracts in this geospatial cluster were largely nonwhite or Hispanic and clearly poor and less educated, TB risks such as drug or alcohol abuse and incarceration were rare.

Cases in the Baltimore Cluster and cases not in a geospatial cluster were equally likely to be genotypically clustered. Thus, recent transmission does not entirely explain the high case rate in Baltimore; however, there is more evidence for recent transmission in this geospatial cluster than in the Montgomery/Prince George's Cluster. Genotype clustering, and, hence, recent transmission, occurred most frequently among U.S.-born, homeless, drug-using populations in Baltimore. Other risk factors associated with a high incidence of TB in the Baltimore Cluster were a predominantly African American population, high rates of HIV and alcohol abuse, and, as in the Montgomery/Prince George's Cluster, poverty and crowding.

Census tract incidences in Maryland varied from 0 to 40 per 100,000 person-years, indicating that there were significant differences in TB rates not only across the state, but even within local jurisdictions, where they reflect significant socioeconomic and health-related population-based disparities. Cases in geospatial clusters were not more likely to be genotypically clustered than cases outside geospatial clusters, suggesting that although recent transmission was occurring, differences in TB rates were primarily driven by patterns of social determinants that put people at high risk for TB infection and progression to active disease. This finding suggests that interventions must address underlying TB risk factors; contact and outbreak investigations alone are not sufficient. An understanding of varying social determinants of TB in different parts of the state can help target high-risk populations when designing and implementing interventions at the local level.

It would be informative to explore in detail reasons why some census tracts with similar high-risk characteristics (e.g., low socioeconomic status and large minority populations) do not have high TB case rates and vice versa. Furthermore, the use of more discriminating genotyping techniques (such as 24-locus MIRU-VNTR or RFLP assays for all case isolates and/or the use of data characterizing epidemiologic links between cases) could help clarify the extent to which genotype clusters represent recent transmission.

This study was primarily ecological in nature and, thus, is not indicative of causal relationships between demographic/socioeconomic/risk-factor variables and TB rates. We chose a descriptive analytical approach to identify areas at risk for the purposes of this study for more locally intensive exploration and intervention. Future studies could use regression analyses to characterize causal relationships.

This analytical approach can be adopted in other settings with minimal training in the use of GIS software and in accessing Census data, yet it characterizes the unequal distribution of TB in Maryland, can lead to improved monitoring and surveillance, and will inform targeted intervention planning that varies by census tract. Because of discrepant risks of TB across the state, particularly in Montgomery County, a field project is underway within high-risk census tracts to further explain why TB rates are unexpectedly high, given the lack of usual risk factors for TB, and to create specific interventions targeted at those areas by working with local community action groups, faith-based organizations, and local ethnic social groups, which we anticipate may vary by census tract.

Limitations

This study was subject to several limitations. One limitation was the lack of systematically reported individual-level data on additional TB risk factors during the study period, including smoking, diabetes, other chronic diseases, and immunosuppressive therapy. Recent changes (2009) in reporting requirements will assist with better defining high-risk populations in future analyses. Another limitation was the reliance on cases' home addresses at diagnosis. Cases might have transmitted or acquired TB at other locations, including workplaces, recreational spaces, or prior residences. In addition, this study did not account for other factors that may indirectly affect TB incidence, such as social determinants of treatment adherence and response. However, Maryland has been a national leader in TB treatment completion indices, and directly observed therapy is used for more than 90% of Maryland TB cases.³ Finally, the study included only culture-positive TB cases, as isolates cannot be genotyped for culture-negative cases, so results do not reflect the total burden of TB in Maryland. However, because culture-negative cases are generally not considered infectious,²⁵ their usefulness in characterizing areas of recent transmission is limited.

CONCLUSIONS

In Maryland from 2004 to 2010, two distinct geospatial clusters of TB cases were identified—one in Baltimore City and the other in Montgomery and Prince George's counties. The TB cases and census tracts that make up these geospatial clusters had distinct demographic, socioeconomic, and risk-factor characteristics that differed from characteristics of the state at large. These TB clusters show a clear distribution of social health inequality. The differences between the findings of our study and those from similar, previous studies in terms of geographic distribution of TB and in patterns of risk factors and recent transmission^13–17 highlight the utility of conducting such analyses in different geographic locations; the TB burden in every geographic area is unique due to differences in living conditions that change transmission patterns as well as differences in the underlying populations and their risk factors for disease. This analysis can inform programmatic efforts targeting selected sections of the state to reduce existing social health inequalities. As TB disease rates decline nationally, focusing these efforts on identified local risk areas may be crucial.