Study Objective and Design

Kopanyo means “people gathering together” in the local Tswana language in Botswana. Consistent with this name, the overarching goal of the Kopanyo Study was to use geospatial analysis and patient interviews to define TB transmission networks and locations of TB transmission in 2 districts in Botswana, a country characterized by high rates of TB and HIV. More precisely, we aimed to describe and compare the clinical and microbiological characteristics of TB patients given a diagnosis in Gaborone and Ghanzi districts; describe and compare the spatial clustering and genotype clustering of patients and strains across and within districts; and determine the factors associated with genotype clustering, spatial clustering, as well as combined genotype and spatial clustering across and within districts. The design and procedures of this population-based, prospective study have been described in detail elsewhere (8,13).

Setting

Figure 1

Figure 1. Catchment areas for population-based geospatial and molecular epidemiologic study of tuberculosis transmission dynamics, Botswana, 2012–2016 (Kopanyo Study). The 2 catchment areas are outlined in black. The neighborhoods within the Gaborone...

Botswana is an economically and politically stable sub-Saharan country that has a universal healthcare system for its citizens. We recruited participants from 2 distinct geographic areas: the capital city and surrounding suburbs, Gaborone district; and the rural district of Ghanzi (Figure 1). The study sites were purposefully selected because they were believed to represent disparate populations in Botswana in terms of demographic, environmental, epidemiologic (i.e., HIV prevalence, TB prevalence) and socioeconomic characteristics. With a population of 354,380 persons, Gaborone is the largest and most crowded urban area in the country. In 2013, at the start of this study, 17% of the general population in Gaborone was estimated to be living with HIV (14). Annual TB rates in Gaborone ranged from 440 to 470 cases/100,000 population during the 5 years before study implementation; ≈70% of TB patients were co-infected with HIV (15).

In contrast, Ghanzi is a rural district in northwestern Botswana, and most of the 44,100 persons in this district live in congregate housing; the town of Ghanzi has a population of 12,179 persons. Most of the population in the district is of San ethnicity. The San kept a traditionally hunter-gatherer lifestyle until early 1990s, when they were forced to transition to farming as a result of government-mandated modernization programs. Since then, most San live in large and crowded private freehold farms most of the year. For short periods during the year (ranging from days to several weeks), they transition through mid-size villages and the town of Ghanzi. Migration between farms and villages in the district is the norm and it is seasonal. However, because of cultural and geographic reasons there is little migration outside the district. Thus, despite major rotational migration between villages and farms, the community remains highly insular. Altogether, these unique cultural and social conditions contribute to the higher rates of TB transmission in Ghanzi. Over the past 2 decades, the TB notification rate in Ghanzi district has consistently been the highest in the country (722 cases/100,000 persons) (14,15). In 2013, the proportion of the general population in Ghanzi estimated to be living with HIV (17%) was similar to that for Gaborone; however, only 36% of TB patients are co-infected with HIV (14).

Recruitment

Participants were enrolled during August 2012–March 2016. All patients given a diagnosis of TB were eligible for enrollment. Participants were recruited from TB clinics and directly observed treatment centers in greater Gaborone (n = 24) and Ghanzi District (n = 6). Patients receiving TB treatment for >14 days before study screening, incarcerated persons, or those who did not consent were excluded from the study.

Data Sources, Measurements, and Variables

Behavioral, clinical, and demographic information were obtained by medical record abstraction and standardized interview at enrollment. Primary residential address, work place address at diagnosis, and address of social gathering venues of patients were obtained through patient interview. All addresses were verified by site visit geotagging, or through a reference layer created by manually relocating addresses in satellite imagery by using OpenStreetMap (http://www.openstreetmap.org) (16), Google Maps, and ArcGIS (Environmental System Research Institute, https://www.esri.com) online geocoding services. WGS 84 projection system latitude and longitude coordinates (with 1.1-m precision) were exported for each address. Botswana population and housing data was used to define geographic boundaries and enumerate localized populations necessary for TB incidence rates (17). We defined high-burden geographic areas if the estimated annualized TB incidence was >305 TB patients/100,000 persons, which is the estimated national TB incidence rate at the start of the study period.

HIV status was determined for all enrolled participants. Following national guidelines, we offered HIV testing to all participants who did not have documented HIV test results or had negative test results from >12 months before enrollment. Patients were asked to report the average number of alcoholic beverages consumed on the same occasion in the previous 30 days, the number of days consuming alcohol in the previous 30 days, and if they currently smoke tobacco. We defined excessive alcohol consumption as a self-report of >5 drinks on the same occasion or drinking on >5 days within the previous 30 days (18). Venues for social gathering were classified as alcohol-related (e.g., bars, liquor stores, pubs, shebeens), places of work, places of worship (e.g., churches, mosques, temples), and healthcare facilities.

Sputa Collection and Laboratory Methods

At least 1 expectorated sputum sample was obtained from each enrolled patient. Patients unable to produce enough sputum or high-quality sputum underwent inhaled nebulized hypertonic saline solution induction. Sputa were decontaminated by using the N-acetyl-L-cysteine and NaOH method with a final concentration of 1% NaOH, and then inoculated into 1 Mycobacterial Growth Indicator Tube (MGIT; Becton Dickinson, https://www.bd.com). MGIT cultures were incubated at 35°C–37°C in the MGIT960 instrument (Becton Dickinson) for <6 weeks. MGIT cultures scored as positive were examined by microscopy and Ziehl-Neelsen staining to identify acid-fast bacilli. SD. The Bioline TB Ag MPT64 Rapid Test (Abbott, https://www.globalpointofcare.abbott/en/product-details/sd-bioline-tb-ag-mpt64-rapid.html) was used to identify the M. tuberculosis complex. Cultures positive for acid-fast bacilli but with negative TB Ag MPT64 results were classified as nontuberculous mycobacteria. Cultures with evidence of both Mycobacterium species and other potential contaminating species were redecontaminated by using the standard method described above. Drug susceptibility testing (DST) for first-line anti-TB drugs was performed by using MGIT DST. Susceptibility was set at 0.1 μg/mL for isoniazid and 1.0 μg/mL for rifampin. We used DST with Lowenstein-Jensen medium in instances for which MGIT DST results were not available.

M. tuberculosis Genotyping

The first culture isolate per patient was genotyped (Genoscreen, https://www.genoscreen.fr) by using 24-locus mycobacterial interspersed repetitive units–variable number of tandem repeats (MIRU-VNTR) and standardized methods (19). MIRU-VNTR results with >1 copy number at >1 loci (i.e., double alleles) as seen with mixtures of different clonal subpopulations, or with missing or indeterminate copy number at any locus, were considered noninterpretable for cluster assignment and were excluded from analysis (20). Two or more patient isolates that had valid, complete, and matching MIRU results were classified as a genotype cluster.

Localized Transmission Events

We used SaTScan (https://www.satscan.org) to identify geographic areas with a larger-than-expected rate of unique genotype clusters. We also used data for all other culture-positive TB patients reported during the study as the background rate (20–22). In brief, all individual MIRU-VNTR results were assigned to the corresponding geocoordinates of the patient’s residence. Each unique MIRU-VNTR result was then scanned separately, applying a purely spatial analysis, in which the number of events in an area was assumed to be Poisson distributed to generate circular zones of various sizes up to a maximum radius of 50 km. A log-likelihood ratio was calculated for each zone in comparison with all possible zones, with the maximum likelihood ratio representing the zone most likely to identify statistically significant spatial concentrations for each MIRU-VNTR result. Thus, by definition, localized transmission was characterized by genotypic and spatial clustering. A Monte Carlo simulation with 9,999 repetitions was used to determine the distribution of the scan statistic under the null hypothesis of spatial randomness; significant spatial clusters were chosen by using an α of p<0.05. No duplicative case counting occurred. The purpose of the spatial scan was to characterize each patient (based on residence) for a dichotomous outcome: member of a localized transmission event or not.

Statistical Methods

Annualized TB incidence per 100,000 persons and 95% CIs, assuming a Poisson distribution, were calculated for local geographies on the basis of the number of cases enrolled from each geography divided by the population estimates for each geography annualized to the duration of the study period. Estimates were assigned to the geographic centroid in ArcGIS. Isopleth cartographic images were produced by using a raster layer interpolated with inverse distance weighting (21). Differences in proportions between behavioral, clinical, and demographic variables by geographic location were assessed by using the χ² test. Multivariable logistic regression analysis was conducted to assess the association of involvement in a localized transmission event (coded as a binary yes/no variable) and select variables by using adjusted odds ratios (aORs) that were significant at the 95% CIs. All variables statistically associated with the main outcome in bivariable analyses (p<0.1) were included in the multivariate model.

Ethics

This study was approved by the Centers for Disease Control and Prevention Institutional Review Board; the Health Research and Development Committee, Ministry of Health and Wellness, Botswana; and the University of Pennsylvania Institutional Review Boards. Participants provided written informed consent.

Patient Characteristics

Figure 2

Figure 2. Flowchart of study enrollment for population-based geospatial and molecular epidemiologic study of TB transmission dynamics, Botswana, 2012–2016. TB, tuberculosis.

A total of 5,515 patients were given a diagnosis of TB during the study period; 4,331 (79%) were enrolled (Figure 2). Primary residence was geocoded and validated for 3,736 (86%) participants. Of these participants, 2,659 (71%) resided in Gaborone District, 723 (19%) resided in Ghanzi District, and 354 (10%) resided in other locations outside Gaborone or Ghanzi Districts. There were no significant differences in proportions with regards to sex (p = 0.116) and having valid MIRU results (p = 0.555) between the 3 locations (Table 1). However, there were differences in proportions for Gaborone, Ghanzi District, and other locations, respectively, regarding age (p<0.001), residence in a high-burden neighborhood (14% vs. 73% vs. 0%; p<0.001), excessive alcohol use (25% vs. 32% vs. 14%; p<0.001), current smoking (20% vs. 43% vs. 13%; p<0.001), unemployment (30% vs. 67% vs. 58%; p<0.001) history of incarceration (5% vs. 8% vs. 5%; p = 0.007), positive HIV status (66% vs. 37% vs. 64%; p<0.001), culture-positive TB (68% vs. 78% vs. 48%; p<0.001), and MDR TB at baseline (1% vs. 2 vs. 6%; p<0.001) (Table 1).

Geographic Distribution of TB

The estimated annualized TB incidence for the overall study population was 306 TB patients/100,000 persons (95% CI 228–339) (Table 2). The incidence rate varied considerably by geography, ranging from 66 (95% CI 44–99) TB patients/100,000 persons in the suburban areas of Gaborone to 1,140 (95% 836–1,556) TB patients/100,000 persons in remote, rural villages of the Ghanzi District. The degree of heterogeneity was more pronounced in Gaborone than in Ghanzi District. For example, a 7.3-fold difference in annualized incidence was found between the highest (location A) and lowest (location J) burden areas in Gaborone. This difference was 1.3 fold in Ghanzi District (highest location W and lowest location KU). In this context, we observed that certain neighborhoods contributed disproportionally to the district-level burden of TB. Some locations that had the lowest TB prevalence had the highest number and proportion of patients co-infected with HIV (Table 1).

Patient Isolate Characteristics

A total of 2,462 (56%) patients had >1 positive culture result; 2,162 (88%) were classified as M. tuberculosis complex, whereas 300 (12%) were classified as nontuberculous mycobacteria and were excluded. MIRU-VNTR results were available for 2,137 patient isolates. We excluded 213 (10%) patient isolates that had incomplete or noninterpretable genotyping results; this exclusion was described elsewhere (23,24). Thus, 1,924 patients were included in phylogenetic analysis. Among these patients, 128 had no residential geocoordinates, which resulted in 1,796 patients available for localized transmission analysis (Figure 2).

Localized Transmission

Figure 3

Figure 3. Annualized estimated TB incidence with the most probable localized transmission events superimposed, Botswana, 2012–2016. A) Gaborone; B) Ghanzi District. Shown are areas with most probable clusters identified by using spatial...

A total of 780 (43%) patients were members of localized transmission events. Among these patients, 537 (69%) resided in Gaborone, 241 (31%) resided in the Ghanzi District, and 2 (0.3%) resided elsewhere. Localized transmission was independently associated with younger age (<15 years of age, aOR 2.20, 95% CI 1.67–4.15; 16–24 years of age, aOR 1.41, 95% CI 1.05–1.89), residing in a high-burden neighborhood (aOR 2.75, 95 CI% 2.21–3.41), being a current smoker (aOR 1.71, 95% CI 1.38–2.11), and being unemployed (aOR 1.31, 95 CI% 1.08–1.59) (Table 3; Appendix Table). Patients who had a known HIV-positive status had lower odds of being a member of localized transmission (aOR 0.71, 95CI % 0.58–0.85). When we superimposed the SaTScan results over the interpolated with inverse distance-weighted TB incidence estimates, the spatial concentration for localized transmission coincided with higher TB incidence rates (Figure 3).

Self-Reported Social Gathering Venues

The proportion matched by TB genotype was significantly larger in Ghanzi District (80%) than in Gaborone (64%) and other locations (4%; p<0.001) (Table 4). A total of 494 (22%) patients resided at the same address as another patient; among these, 29% matched by TB genotype. The proportion matched by TB genotype was significantly larger in the Ghanzi District (50%) than in Gaborone (18%; p<0.001). A total of 605 (32%) patients reported the same place of worship as another patient; among these, 11% matched by TB genotype. The proportion matched by TB genotype was significantly larger in Ghanzi District (24%) than in Gaborone (9%; p = 0.002). A total of 585 (30%) reported the same alcohol-related venue as another patient; among these, 28% matched by TB genotype. The proportion matched by TB genotype was significantly larger in Ghanzi District (57%) than in Gaborone (16%; p<0.001).