Abstract
Background
Tuberculosis (TB) is the leading infectious cause of death globally, and drug-resistant TB strains pose a serious threat to controlling the global TB epidemic. The clinical features, locations, and social factors driving transmission in settings with high incidences of drug-resistant TB are poorly understood.
Methods
We measured a network of genomic links using Mycobacterium tuberculosis whole-genome sequences.
Results
Patients with 2–3 months of cough or who spent time in urban locations were more likely to be linked in the network, while patients with sputum smear–positive disease were less likely to be linked than those with smear-negative disease. Associations persisted using different thresholds to define genomic links and irrespective of assumptions about the direction of transmission.
Conclusions
Identifying factors that lead to many transmissions, including contact with urban areas, can suggest settings instrumental in transmission and indicate optimal locations and groups to target with interventions.
Keywords: drug-resistant tuberculosis, tuberculosis transmission, whole genome sequencing, transmission networks, network models
We identified factors that lead to the transmission of extensively drug-resistant tuberculosis, including contact with urban areas; these factors can suggest settings instrumental in transmission and indicate optimal locations and groups to target with interventions.
(See the Editorial Commentary by Zelner on pages 2403–4.)
Tuberculosis (TB) is the leading infectious cause of death globally, and drug-resistant TB strains pose a serious threat to controlling the global TB epidemic [1]. The majority of drug-resistant TB cases in settings with high TB incidence are due to transmission of drug-resistant strains, rather than acquisition of resistance through inadequate TB treatment [2, 3]. Interventions to reduce incidence must include efforts to reduce transmission, which will require a clear understanding of the clinical features, types of social contact, and locations driving the spread of TB [4, 5].
Transmission heterogeneity, or variation in the number of secondary infections caused by an index case, may play a critical role in shaping TB epidemiology [6–9]. On a population level, transmission heterogeneity arises as a result of 3 factors: the extent and duration of infectiousness, the susceptibility of exposed persons to infection and disease, and the rate of contact between infected and susceptible persons [10, 11]. The first 2 factors can be considered functions of the biologic features of disease; the third is related to social and behavioral patterns that define where and with whom infected persons spend time. Identifying the clinical features of index cases that cause many secondary cases, as well as the locations or types of contact driving transmission, can identify opportunities for interventions. Those control measures that either target individuals who are likely to generate many secondary cases of disease or are implemented in specific locations of high transmission risk will likely outperform measures targeting broader areas or populations [8, 12].
Extensively drug-resistant (XDR) TB is associated with poorer treatment outcomes than drug-susceptible or multidrug-resistant TB, making prevention of transmission critical. However, the factors driving XDR-TB transmission in high-incidence settings remain poorly understood. Markers of a high bacterial burden, including sputum smear status and cavitary disease on a chest X-ray, characterize “classical” infectious TB [13–15], but in high-incidence settings these clinical features alone cannot explain observed individual-level heterogeneity in TB transmission [6]. Given the role of coughing in expelling Mycobacterium tuberculosis (Mtb) from the respiratory tract and into the surrounding environment, the duration of cough may also be associated with transmission. However, there are few studies that examine cough duration as a predictor of transmission. An individual’s likelihood of transmitting TB is also influenced by behaviors that enable the person-to-person contact necessary for transmission. Although over 80% of transmissions may occur outside of the home, there is limited understanding of the locations at which transmission occurs in community settings and among non-close contacts (ie, persons who do not know each other) [16, 17].
Identifying the factors associated with transmission events is challenging, given that disease transmission is unobserved; however, the advent of bacterial whole-genome sequencing (WGS) to characterize Mtb isolates provides new opportunities to study transmission. Relative to less sensitive genotyping methods, WGS allows for improved precision to identify cases with highly similar Mtb sequences, and thereby generate hypotheses about transmission events. Such links can be used to create a network comprised of putative transmission links between cases, and analytic methods specific to networks provide a framework for studying the case characteristics associated with transmission. Sequencing may prove especially useful in settings with high incidences of TB, where the majority of transmission is expected to occur in community locations between individuals who may not know each another and would, therefore, not be identified through conventional contact investigations. Human immunodeficiency virus (HIV) genome sequences have been used to construct transmission networks and characterize transmission clusters [18, 19]; we built on this by using network models to identify the factors associated with transmission in a network based on sequencing data from TB cases [20, 21].
In this study, we used bacterial WGS to define plausible transmission links between XDR-TB cases in KwaZulu-Natal, South Africa. Using the resulting sequencing-based network, we identified clinical and social mixing factors associated with TB transmission. The broader goal of this study was to identify the locations, types of contact, and clinical features linked with XDR-TB transmission in South Africa, providing evidence for the development of effective interventions targeted towards individuals and settings sustaining TB spread.
METHODS
Study Design and Procedures
The Transmission of HIV-Associated XDR-TB (TRAX) study is a cross-sectional study that enrolled culture-confirmed pulmonary XDR-TB patients diagnosed from May 2011 through August 2014 in KwaZulu-Natal, South Africa [2]. The primary aim of the study was to estimate the proportion of XDR-TB cases that developed due to transmission versus acquired drug resistance resulting from inadequate treatment. Detailed methods of the TRAX study have been previously published [2]. Briefly, we identified XDR-TB cases through the referral laboratory that conducts drug-susceptibility testing for all public healthcare facilities in the province. All participants provided written informed consent.
We interviewed participants to collect demographic and clinical information, including previous TB disease and HIV status, which were confirmed using medical records. Participants completed a social network questionnaire that included information on close contacts, the location and duration of time spent in congregate locations, and hospital admissions in the 5 years prior to their XDR-TB diagnosis.
Measures
Sequencing and bioinformatics procedures have been described in detail elsewhere. Briefly, we obtained participants’ diagnostic XDR-TB isolates, re‐cultured on Löwenstein‐Jensen slants, and conducted population sweeps. Raw, paired‐end sequencing reads were generated on the Illumina (MiSeq) platform and aligned to the H37Rv reference genome (NC_000962.3). Single nucleotide polymorphisms (SNPs) were detected using standard pairwise resequencing techniques (Samtools v0.1.19) against the reference and filtered for quality, read consensus, and proximity to indels and proline-glutamic acid/proline-proline-glutamic acid gene regions. Alignment files can be found at National Center for Biotechnology Information Bioproject PRJNA476470.
We defined a directed genomic link as a pair of XDR-TB cases with 5 or fewer SNP differences between their Mtb sequences, which is a cutoff used by prior studies, and considered the case with the earliest diagnosis date as the primary case [22]. We constructed a network of directed genomic links between cases, in which each node in the network represented a diagnosed XDR-TB case and each link represented a genomic link between 2 cases. For each case, we calculated the number of genomic links in which they were the primary (index) case.
The diagnostic XDR-TB sputum sample was used to determine smear status and grade. Cough duration was reported by the patient at the time of enrollment. Chest X-ray results were abstracted from patient medical records at the time of their XDR-TB diagnosis, but 34% (116) of patients were missing X-ray results, so we included cavitary disease only in a secondary analysis.
To define urban contact, we determined whether a patient reported ever living in eThekwini (Durban), the primary urban center of the province; whether they had been admitted to a hospital there in the previous 5 years; or whether they reported spending time at a congregate location there. To measure hospital contact, we summed the total number of months patients reported spending in the hospital in the 5 years prior to diagnosis. Participants also completed a social network questionnaire, in which they reported the number of contacts with whom they spent more than 2 hours per week in the previous year. We used the number of contacts reported as a proxy measure of close contact.
Exponential Random Graph Models
Conventional statistical models assume that the characteristics of each subject are independent, which is an assumption not met in disease transmission networks. In a transmission network, the unit of interest is a transmission link, which consists of 2 cases whose attributes may be correlated. Exponential random graph models are a tool for statistically modeling the probability of a link to form between nodes (cases) in a network, accounting for the inherent correlation among attributes of cases linked through transmission events. We used exponential random graph models to express the probability that a transmission link occurred between 2 cases in the network, as a function of demographic and clinical characteristics of each case (Supplementary Equation 1).
We constructed parsimonious models, which excluded terms that did not change main effect estimates by more than 10% when removed from the model, and fully specified models, which included additional terms for the year of study enrollment and the Mtb strain, both of which may be artifactually related to the number of genomic links per case.
In addition to the primary predictors of interest, we defined a priori and included potential confounders in models: sex, age, and HIV status. The code for all models is available at https://github.com/kbratnelson/tb-ergms.
Alternate Models and Sensitivity Analyses
Given uncertainty regarding the SNP threshold for defining a direct transmission event, we fit models to networks constructed using 2 additional SNP thresholds to define genomic links: 1 more stringent (≤3 SNPs) and 1 less stringent (≤10 SNPs). We also tested a definition for genomic links that combined WGS with restriction fragment length polymorphism typing (RFLP) results, since this method can capture regions of the genome that sequencing does not. In addition to defining networks using directed genomic links, we tested associations in “undirected” networks, in which we did not assume a direction of transmission.
RESULTS
The Transmission of Human Immunodeficiency Virus–Associated Extensively Drug-resistant Tuberculosis Study Cohort
Between 2011 and 2014, we screened a convenience sample of 521 (51%) of 1027 culture-confirmed XDR-TB patients diagnosed in KwaZulu-Natal, and enrolled 404 (78% of screened). Among the 404 participants, 234 (58%) were female, with a median age of 34 years (interquartile range 28–43). There were 311 (77%) participants living with HIV. Mycobacterium tuberculosis isolates from 344 (85%) participants passed all sequencing quality filters and were available for analysis (Table 1).
Table 1.
The Transmission of Human Immunodeficiency Virus–associated Extensively Drug-resistant Tuberculosis Study Cohort Characteristics
| Demographics | |
| Female | 202 (59) |
| Age, median (IQR) | 34 (29–43) |
| 0–15 | 12 (3) |
| 16–34 | 171 (50) |
| 35–54 | 134 (39) |
| ≥55 | 27 (8) |
| Monthly household income | |
| <Rand (R)500 | 120 (35) |
| R500–R2500 | 153 (44) |
| >R2500 | 71 (21) |
| Clinical characteristics | |
| Current or former smoker | 35 (10) |
| Diabetes | 22 (6) |
| HIV positive | 266 (77) |
| Receiving antiretroviral therapy | 204 (77) |
| CD4 cell count, median (IQR) | 240 (111–425) |
| Virologic suppression (<400 copies/mL) | 134 (50) |
| Cough | |
| Patients with cough | 284 (83) |
| Median duration of cough (IQR) | 8 (4–12) |
| Sputum smear–negative for acid-fast bacilli | 109 (32) |
| Scanty-positive | 37 (11) |
| Smear-positive, Grade 1 | 59 (17) |
| Smear-positive, Grade 2 | 51 (15) |
| Smear-positive, Grade 3+ | 88 (26) |
| Chest X-ray available | 228 (66) |
| Cavities on chest X-ray | 60 (26) |
| Previous treatment for multidrug-resistant tuberculosis | 105 (31) |
| Social mixing characteristics | |
| Number of reported contacts, median (IQR) | 7 (4–10) |
| Number of months in hospital, median (IQR) | 3 (2–5) |
| Previous stay at urban hospital | 175 (51) |
| Current or previous urban residence | 39 (11) |
| Reported visiting urban congregate setting | 38 (11) |
Data are presented as n (%) unless otherwise indicated. n = 344. Abbreviations: HIV, human immunodeficiency virus; IQR, interquartile range.
Genomic Links
Among 344 cases comprising the sequencing-based network (threshold ≤5 SNPs), there were a total of 740 genomic links. There were 125 cases (36%) that had at least 1 link. Among those cases with links, the number of links ranged from 1 to 28; 38 cases (30%) had between 1 and 5 links, and 18 cases (14%) had more than 10 links (Figure 1). At a threshold of ≤10 SNPs, there were 181 (53%) cases with at least 1 link; at ≤3SNPs, there were 116 (34%) cases with at least 1 link (Supplementary Table 1).
Figure 1.
Sequencing-based network, ≤5 single nucleotide polymorphisms (SNPs). The degree distribution of the sequencing-based network has links defined by genomic links of ≤5 SNPs. The sequencing-based network is inset. In both visuals, cases with a lower network degree are in yellow, and those with a higher degree are in red.
Clinical Infectiousness Measures
In our primary model, which adjusted for age, sex, and HIV status, participants reporting 2 or 3 months of cough were more likely to be linked than those reporting no cough. The odds of a genomic link among patients with 2 or 3 months of a cough were 2.68 times higher (95% confidence interval [CI] 2.19–3.27) and 2.31 times higher (95% CI 1.91–2.80) than those with no cough, respectively (Table 2). This trend did not continue in the highest category of cough duration (>4 months), but this group was small (n = 33).
Table 2.
Adjusted Associations Between Clinical Features and Being Linked in the Directed Network by ≤5 Single Nucleotide Polymorphisms
| Clinical Feature | n (%) | Odds Ratio | 95% CI | P |
|---|---|---|---|---|
| Cough duration | ||||
| No cough reported | 128 (37) | Ref | ||
| 1 month | 60 (17) | 0.54 | (.39–.74) | <.01 |
| 2 months | 51 (14) | 2.68 | (2.19–3.27) | <.01 |
| 3 months | 72 (21) | 2.31 | (1.91–2.80) | <.01 |
| ≥4 months | 33 (10) | 1.06 | (.78–1.45) | .70 |
| Smear status | ||||
| Smear negative | 109 (32) | Ref | ||
| Smear positive, scanty positive | 37 (11) | 0.98 | (.77–1.25) | .88 |
| Smear positive, Grade 1 | 59 (17) | 0.65 | (.52–.83) | <.01 |
| Smear positive, Grade 2 | 51 (15) | 0.73 | (.58–.91) | <.01 |
| Smear positive, Grade 3+ | 88 (26) | 0.58 | (.47–.72) | <.01 |
The model also includes terms for HIV status, sex, and age. Abbreviations: CI, confidence interval; HIV, human immunodeficiency virus; Ref, reference group for odds ratio calculation.
In the same model, smear-positive cases were less likely to be linked than smear-negative cases, irrespective of smear grade. Cases with the highest smear grade of 3+ were the least likely to be linked (odds ratio [OR] 0.58, 95% CI 0.47–0.72; Table3); this association was similar in models excluding cough (Supplementary Table 2). When considered as a dichotomous variable (smear-negative and smear-positive), smear-positive cases were 0.65 times as likely to have a link as smear-negative cases (95% CI 0.55–0.76; Supplementary Table 3). In models including only patients with available chest X-ray results, cavitary disease was associated with a higher likelihood of genomic links (Supplementary Table 4).
Table 3.
Adjusted Associations Between Social Mixing Measures and Being Linked in the Directed Network by ≤5 Single Nucleotide Polymorphisms
| n (%) | Odds Ratio | 95% CI | P | |
|---|---|---|---|---|
| Contact with urban areas | ||||
| 0 urban locations | 149 (43) | Ref | ... | ... |
| 1 urban location | 147 (43) | 2.57 | (2.17–3.05) | <.01 |
| ≥2 urban locations | 48 (20) | 1.73 | (1.33–2.25) | <.01 |
| Duration in hospital | ||||
| 0–2 months | 113 (33) | Ref | ... | ... |
| 3–5 months | 81 (24) | 0.82 | (.69–.98) | .03 |
| >5 months | 59 (17) | 0.38 | (.29–.50) | <.01 |
| Named contacts | ||||
| 0–4 contacts | 108 (31) | Ref | ... | ... |
| 5–10 contacts | 144 (42) | 1.17 | (.98–1.41) | .08 |
| >10 contacts | 85 (25) | 1.44 | (1.18–1.76) | <.01 |
Model also includes terms for HIV status, sex, and age, smear status and grade, and cough duration. Abbreviations: CI, confidence interval; HIV, human immunodeficiency virus.
In networks defined using a more stringent SNP threshold (≤3 SNPs) and in which we did not assume a direction of transmission, the direction and magnitude of associations with infectiousness were generally similar (Supplementary Table 5). In full models, including terms for Mtb strain and year of study enrollment, results were similar to those from the primary model (Supplementary Table 6).
Social Mixing Measures
In our primary model, cases reporting activity in 1 or more urban locations (residential, health care, or other congregate) were more likely to be linked than those who did not (OR for 1 location, 2.57, 95% CI 2.17–3.05; OR for 2 or more locations, 1.73, 95% CI 1.33–2.25; Table 3). When we deconstructed the variable describing contact with urban locations to determine whether there was a specific component responsible for this association, we found that reporting a stay in an urban hospital was most strongly associated with being linked (OR 2.65, 95% CI 1.60–4.39). However, cases who spent prolonged periods of time in the hospital (regardless of whether the hospital was in an urban location) were less likely to be linked in the network than those spending ≤2 months (OR for 3–5 months in hospital, 0.82, 95% CI 0.69–0.98; OR for 5 or more months, 0.38, 95% CI 0.29–0.50). The number of named contacts was moderately associated with being linked in the network: cases who reported 5–10 or >10 contacts were more likely to be linked than cases who reported ≤4 named contacts (Table 3). Notably, associations with infectiousness measures persisted and were similar in magnitude in models including all social mixing variables. In models that also included terms for Mtb strain and year enrolled, associations were similar in direction and magnitude to results from the primary model (Supplementary Table 7).
In networks with alternative thresholds for genomic links, the directions of association were consistent with the primary model, though association strength was model-dependent (Supplementary Table 8).
DISCUSSION
We constructed a network of genomic links using Mtb WGS and found that XDR-TB participants reporting 2–3 months of cough or contact with urban locations were more likely to be linked. Contrary to the notion that smear positivity is a marker of infectiousness, we found that smear-positive cases were less likely to be linked than smear-negative cases. These associations persisted in networks using different thresholds to define genomic links.
Aside from the negative association with smear status, these findings are largely consistent with prior studies of TB transmission [13]. Cough duration and cavitary disease have previously been associated with genotypic linkages in both drug-susceptible and multidrug-resistant TB [15, 23, 24]. However, these studies employed conventional genotyping methods; our study is the first to our knowledge to use WGS, which has higher specificity to detect transmission links. Although the trend between cough and genomic links did not persist in the group with the longest duration of cough symptoms; this may be due to the small size of this group or alternative etiologies of prolonged cough, or may reflect true behavioral changes among cases with severe symptoms, resulting in reduced contact rates and fewer transmission events.
There are several potential explanations for the unexpected finding that smear-positive cases were less likely to be linked than smear-negative cases. Patients with smear-negative disease may have less severe or clinically atypical disease, leading to delays in diagnoses, longer infectious periods, and, thus, more opportunities for transmission. If smear positivity does indeed indicate more severe disease, these cases may be more likely to die earlier, due to unsuccessful treatment, or be more likely to be hospitalized, reducing opportunities for community transmission. Indeed, we found that the median time spent in hospital was longer among smear-positive than smear-negative cases (3 vs 2 months, respectively). Lastly, we only considered a single sputum result at the time of diagnosis, which may fail to represent a patient's smear status over the course of TB disease.
Participants who spent time in urban locations prior to diagnosis were more likely to be linked in the network. Urban areas tend to have higher incidences of TB and often provide ideal conditions for disease spread, as person-to-person contact rates may be higher than in rural areas. Moreover, urban areas of high TB incidence have been hypothesized to drive disease incidence in wider geographic areas, and our findings support previous findings from this same cohort suggesting that rural-urban migration may be driving the transmission of XDR-TB in KwaZulu-Natal [25]. We found that the number of months spent in the hospital prior to diagnosis was negatively associated with being linked in the network. Health-care facilities and other institutional settings are often considered to be amplifiers of transmission; however, individuals who spent time in the hospital during their infectious periods may have had lower effective contact rates than those who spent their infectious periods living, working, and socializing in their communities [26].
We found a weak, positive association between the number of contacts named by participants and being linked. We hypothesized that reporting more contacts would reflect a higher level of engagement in person-to-person contact (either through social activities, employment, school, or home life) and, thus, more opportunities for transmission. However, this trend did not persist among those reporting the highest numbers of contacts. This may reflect a threshold above which additional contact does not necessarily lead to additional transmission events—a notion supported by previous modeling studies—or may indicate that our measure of close contact is a poor proxy for the extent of true close contact during individuals’ infectious periods [6].
This study has several limitations. First, we enrolled 40% of all diagnosed XDR-TB cases in KwaZulu-Natal during this time period; therefore, there are missing cases and genomic links in this network. If missing cases are intermediaries in the transmission chain between the sampled cases we found to be linked, we might observe larger genomic differences between sampled cases than those that represent true transmission events. However, reducing the SNP threshold showed similar results to our primary analysis. Although participants enrolled in the study were demographically similar to all diagnosed XDR-TB patients, the extent to which enrolled cases are representative of unsampled, undiagnosed cases is not clear. In a modeling study examining the potential for biased sampling in this study, we found that while cases with many transmission links may have been undersampled, our models do not suggest biased sampling by patient attributes like smear status [27]. Second, although we also conducted our analyses using a network in which we did not assume directionality and found largely similar results, we still cannot distinguish between individual-level factors that increase the risk of infection from those that increase the risk of transmission. Lastly, these transmission patterns may be different in settings with low TB incidence or low HIV prevalence, and further research should aim to characterize local TB transmission patterns in other contexts.
Identifying the individual-level factors driving TB transmission can inform the development of prevention strategies targeting specific groups, to achieve maximal reductions in transmission. Our study suggests that individuals who spend time in urban settings and those with extended symptom durations may cause many secondary TB cases, a finding which should be examined more thoroughly in future studies. Interventions targeting urban areas and those with high contact rates within them may be indicated.
Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Notes
Acknowledgments. The authors thank the study team at the University of KwaZulu-Natal for their tireless efforts in data collection, record abstraction, participant recruitment, and interviews, and the participants and their families who consented to participate in this study.
Disclaimer. The findings and conclusions in this manuscript are those of the authors and do not necessarily represent the official position of the funding agencies. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Financial support. This work was supported by the US National Institute of Allergy and Infectious Diseases, National Institutes of Health (grant numbers R01AI089349, R01AI087465, R01AI138646, and K24AI114444 to N. R. G.; R01AI114304 to J. C. M. B.; and K23AI134182 to S. C. A.); Emory (grant number CFAR P30AI050409 to Curran); Einstein (grant number CFAR P30AI124414 to Goldstein); and Einstein/Montefiore (grant number ICTR UL1 TR001073 to Shamoon).
Potential conflicts of interest. B. A. L. has received personal fees from Takeda Pharmaceutical, the Centers for Disease Control and Prevention Foundation, and Hall Booth Smith, Professional Corporation, outside the submitted work. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.
References
- 1. World Health Organization. Global tuberculosis report. Geneva, Switzerland: World Health Organization, 2017. [Google Scholar]
- 2. Shah NS, Auld SC, Brust JC, et al. Transmission of extensively drug-resistant tuberculosis in South Africa. N Engl J Med 2017; 376:243–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Yang C, Luo T, Shen X, et al. Transmission of multidrug-resistant Mycobacterium tuberculosis in Shanghai, China: a retrospective observational study using whole-genome sequencing and epidemiological investigation. Lancet Infect Dis 2017; 17:275–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Auld SC, Kasmar AG, Dowdy DW, et al. Research roadmap for tuberculosis transmission science: where do we go from here and how will we know when we’re there? J Infect Dis 2017; 216:662–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. The paradigm shift: Global plan to end TB 2016–2020: Stop TB Partnership. Copenhagen, Denmark: United Nations Office for Project Services (UNOPS), 2015. [Google Scholar]
- 6. McCreesh N, White RG. An explanation for the low proportion of tuberculosis that results from transmission between household and known social contacts. Sci Rep 2018; 8:5382 Available at: https://www.nature.com/articles/s41598-018-23797-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Paull SH, Song S, McClure KM, Sackett LC, Kilpatrick AM, Johnson PT. From superspreaders to disease hotspots: linking transmission across hosts and space. Front Ecol Environ 2012; 10:75–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Lloyd-Smith JO, Schreiber SJ, Kopp PE, Getz WM. Superspreading and the effect of individual variation on disease emergence. Nature 2005; 438:355–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Li LM, Grassly NC, Fraser C. Quantifying transmission heterogeneity using both pathogen phylogenies and incidence time series. Mol Biol Evol 2017; 34:2982–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. VanderWaal KL, Ezenwa VO. Heterogeneity in pathogen transmission: mechanisms and methodology. Funct Ecol 2016; 30:1606–22. [Google Scholar]
- 11. Dowdy DW, Azman AS, Kendall EA, Mathema B. Transforming the fight against tuberculosis: targeting catalysts of transmission. Clin Infect Dis 2014; 59:1123–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Dowdy DW, Golub JE, Chaisson RE, Saraceni V. Heterogeneity in tuberculosis transmission and the role of geographic hotspots in propagating epidemics. Proc Natl Acad Sci USA 2012; 109:9557–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Mathema B, Andrews JR, Cohen T, et al. Drivers of tuberculosis transmission. J Infect Dis 2017; 216:644–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Behr MA, Warren SA, Salamon H, et al. Transmission of Mycobacterium tuberculosis from patients smear-negative for acid-fast bacilli. Lancet 1999; 353:444–9. [DOI] [PubMed] [Google Scholar]
- 15. Rodwell TC, Kapasi AJ, Barnes RF, Moser KS. Factors associated with genotype clustering of Mycobacterium tuberculosis isolates in an ethnically diverse region of southern California, United States. Infect Genet Evol 2012; 12:1917–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Johnstone-Robertson SP, Mark D, Morrow C, et al. Social mixing patterns within a South African township community: implications for respiratory disease transmission and control. Am J Epidemiol 2011; 174:1246–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Glynn JR, Guerra-Assunção JA, Houben RM, et al. Whole genome sequencing shows a low proportion of tuberculosis disease is attributable to known close contacts in rural Malawi. PLOS One 2015; 10:e0132840. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Oster AM, Wertheim JO, Hernandez AL, Ocfemia MC, Saduvala N, Hall HI. Using molecular HIV surveillance data to understand transmission between subpopulations in the United States. J Acquir Immune Defic Syndr 2015; 70:444–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Morgan E, Oster AM, Townsell S, Peace D, Benbow N, Schneider JA. HIV-1 infection and transmission networks of younger people in Chicago, Illinois, 2005–2011. Public Health Rep 2017; 132:48–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Klovdahl AS, Graviss EA, Yaganehdoost A, et al. Networks and tuberculosis: an undetected community outbreak involving public places. Social Science & Medicine (1982) 2001; 52:681–94. [DOI] [PubMed] [Google Scholar]
- 21. Oster AM, France AM, Mermin J. Molecular epidemiology and the transformation of HIV prevention. JAMA 2018; 319:1657–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Hatherell HA, Colijn C, Stagg HR, Jackson C, Winter JR, Abubakar I. Interpreting whole genome sequencing for investigating tuberculosis transmission: a systematic review. BMC Med 2016; 14:21 Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4804562/ [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Turner RD, Bothamley GH. Cough and the transmission of tuberculosis. J Infect Dis 2015; 211:1367–72. [DOI] [PubMed] [Google Scholar]
- 24. Turner RD, Hooper R, Birring SS, Bothamley Graham H. Daily cough frequency in tuberculosis is associated with rates of household infection. San Fransisco, CA: American Thoracic Society Conference, 2016. [Google Scholar]
- 25. Nelson KN, Shah NS, Mathema B, et al. Spatial patterns of extensively drug-resistant tuberculosis transmission in KwaZulu-Natal, South Africa. J Infect Dis 2018; 218:1964–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Basu S, Stuckler D, McKee M. Addressing institutional amplifiers in the dynamics and control of tuberculosis epidemics. Am J Trop Med Hyg 2011; 84:30–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Nelson KN, Gandhi NR, Mathema B, et al. Modeling missing cases and transmission links in networks of extensively drug-resistant tuberculosis in KwaZulu-Natal, South Africa. bioRxiv 2019: 655969. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.

