- Email: [email protected]
Tuberculosis transmission in nontraditional settings
Tuberculosis Transmission in Nontraditional Settings A Decision-Tree Approach J. Steve Kammerer, MBA, Scott J.N. McNabb, PhD, MS, Jose E. Becerra, MD, MPH, Lisa Rosenblum, MD, MPH, Nong Shang, PhD, Michael F. Iademarco, MD, MPH, Thomas R. Navin, MD Background: Tuberculosis (TB) transmission in nontraditional settings and relationships (non-TSR) often eludes detection by conventional contact investigation and is increasingly common. The U.S.-based National Tuberculosis Genotyping and Surveillance Network collected epidemiologic data and genotyping results of Mycobacterium tuberculosis isolates from 1996 to 2000. Methods:
In 2003–2004, we determined the number and characteristics of TB patients in non-TSR that were involved in recent transmission, generated a decision tree to profile those patients, and performed a case– control study to identify predictors of being in non-TSR.
Results:
Of 10,844 culture-positive reported TB cases that were genotyped, 4724 (43.6%) M. tuberculosis isolates were clustered with at least one other isolate. Among these, 520 (11%) had epidemiologic linkages discovered during conventional contact investigation or cluster investigation and confirmed by genotyping results. The decision tree identified race/ ethnicity (non-Hispanic white or black) as having the greatest predictive ability to determine patients in non-TSR, followed by being aged 15 to 24 years and having positive or unknown HIV infection status. From the 520, 85 (16.4%) had non-TSR, and 435 (83.6%) had traditional settings and relationships (TSR). In multivariate analyses, patients in non-TSR were significantly more likely than those in TSR to be non-Hispanic white (adjusted odds ratio [aOR]⫽6.1; 95% confidence interval [CI]⫽1.7–21.1]) or to have an M. tuberculosis isolate resistant to rifampin (aOR⫽5.2; 95% CI⫽1.5–17.7).
Conclusions: Decision-tree analyses can be used to enhance both the efficiency and effectiveness of TB prevention and control activities in identifying patients in non-TSR. (Am J Prev Med 2005;28(2):201–207) © 2005 American Journal of Preventive Medicine
Background
A
lthough increasingly common, tuberculosis (TB) transmission in nontraditional settings and relationships (non-TSR) often eludes detection by conventional contact investigation.1– 4 Conventional contact investigations for TB rely on the concentric circle approach and focus mainly on the home and close family, other relatives or friends, and the workplace. This approach is infrequently extended to non-TSR.3,5,6 We defined non-TSR in this study as settings and relationships other than the home, workplace, school, common congregate settings, relatives, friends, and co-workers. Cluster investigation, defined
From the Division of Tuberculosis Elimination, National Center for HIV, STD, and TB Prevention, Centers for Disease Control and Prevention (Kammerer, McNabb, Becerra, Rosenblum, Shang, Iademarco, Navin); and independent contractor (Kammerer), Atlanta, Georgia Address correspondence and reprint requests to: J. Steve Kammerer, Epidemiology Team, Surveillance, Epidemiology, and Outbreak Investigation Branch, Division of Tuberculosis Elimination, National Center for HIV, STD, and TB Prevention, Centers for Disease Control and Prevention, 1600 Clifton Road, Mailstop E-10, Atlanta GA 30333. E-mail: [email protected].
as the epidemiologic investigation of TB patients with genotypically matched Mycobacterium tuberculosis isolates, provided additional epidemiologic information about the possible dynamics of TB transmission in the “outer rings” of the concentric circle. Recent data suggest that patients identified by cluster investigation are more likely to have acquired TB in non-TSR than those patients identified by conventional contact investigation.3,7 The National Tuberculosis Genotyping and Surveillance Network (NTGSN) was established in seven sentinel sites, including the states of Arkansas, Maryland, Massachusetts, Michigan and New Jersey, six counties in California (Alameda, Contra Costa, Marin, San Mateo, Santa Clara, and Solano), and four counties in Texas (Tarrant, Dallas, Hildalgo, and Cameron). These sites were funded from 1996 to 2000 to assess the utility of genotyping of M. tuberculosis isolates.8,9 The target population for this study included TB patients involved in recent transmission, defined as those with epidemiologic links that were confirmed by having matching genotypes. Most patients in the NTGSN did not meet these criteria, and could not be
Am J Prev Med 2005;28(2) © 2005 American Journal of Preventive Medicine • Published by Elsevier Inc.
0749-3797/05/$–see front matter doi:10.1016/j.amepre.2004.10.011
201
confirmed as being involved in recent transmission; such patients are not addressed in the study. The objective of the study was to characterize patients that were involved in recent transmission of TB and more likely to be found in non-TSR to assist TB control programs in targeting conventional contact investigations. Three different methods were used: descriptive statistics, a decision tree, and multivariate regression modeling using a generalized estimating equation (GEE). Associations between exposures and health outcomes derived from logistic regression modeling analyses often do not translate into practical guidance for use in public health practice, including the setting of conventional TB contact investigation.10 Reasons for this include the fact that several exposures may be independently associated with the health outcome, yet from a practical point of view, the public health scientist cannot discern which are more relevant and, as such, should be targeted for intervention. The decision tree approach overcomes this problem.11 Each terminal node of the decision tree provides an estimate of the yield of the health outcome under study, with respect to a given subset of patients. For example, decision-trees have been used successfully to predict tuberculin skin test (TST) results for contacts identified during conventional contact investigations.10 In 2003–2004, we determined the number and characteristics of TB patients with non-TSR in the NTGSN. The health outcome was defined as a TB patient being in non-TSR, not TB transmission, per se. Decision-tree analyses were performed to predict which subgroups of TB patients were at increased risk of being in non-TSR. We also performed a case– control study to identify predictors of TB patients in non-TSR that included multivariate regression.
Methods Population The study population and the epidemiologic and laboratory methods used in the NTGSN have been described elsewhere.12 The NTGSN collected data from 15,035 TB patients (16% of the total number reported in the United States during this time period).9 The seven NTGSN sites reported epidemiologic linkages discovered during conventional contact investigations of TB patients with isolates culture-positive for M. tuberculosis. Cluster investigations were performed when no epidemiologic linkage was discovered during conventional contact investigation and upon the recognition of a genetic cluster. A genetic cluster refers to a group of two or more TB patients within the same NTGSN site having M. tuberculosis isolates with matched genotypes. Four sites out of seven completed cluster investigations. Three of the sites (Arkansas, Maryland, and Massachusetts) completed cluster investigations from 1998 to 2000. California completed cluster investigations solely from Santa Clara County and only from 1999 to 2000. Both medical record reviews and addi-
202
tional patient interviews were completed. The study population for this investigation included those reported TB cases with epidemiologic linkages having M. tuberculosis isolates with matched genotypes.
Data Collection The NTGSN sites recorded epidemiologic linkages discovered during conventional contact investigation and cluster investigation. We defined nontraditional settings as any setting that excludes home, school, workplace, or common congregate settings such as correctional facilities, daycare centers, emergency shelters, group quarters, acute care hospitals, nursing homes, and other long-term care facilities. Examples of nontraditional settings included bars, social clubs, “hang-outs,” and churches or temples. We defined nontraditional relationships as any relationship that excluded household and nonhousehold friend and co-worker contacts. Examples of nontraditional relationships included being in the same place at the same time such as a customer in a store with another TB patient. For some epidemiologic linkages, text describing the specific setting or relationship was recorded by the NTGSN sites. When recorded, we reviewed and categorized these as nonTSR or TSR. For each TB patient with an epidemiologic linkage, we gathered demographic, behavioral, and clinical data from the national Report of Verified Case of TB (RVCT). In this study, HIV infection status was categorized as positive (i.e., “infected”), negative, or unknown. The health outcome was defined as a TB patient in non-TSR versus TSR, regardless of whether the case was a presumed source case or a secondary case in the chain of transmission. Patients who were identified in multiple epidemiologic linkages were included in the traditional category only if all epidemiologic linkages involving the patient were traditional.
Study Design We performed a case– control study to identify predictors of TB patients in non-TSR. Cases were defined as all TB patients in non-TSR (n ⫽85) that were involved in recent transmission. Controls were defined as all TB patients in TSR (n ⫽435) that were involved in recent transmission. The decision tree analyses used the same set of TB patients.
Data Analyses Descriptive statistics and univariate and multivariate models were performed using SAS, version 8.2 (SAS Institute Inc., Cary NC, 2001). For method one (descriptive), the Pearson chi-square or two-tailed Fisher’s exact test, as appropriate, was used to assess difference in proportions between the non-TSR and TSR populations regarding demographic, clinical, and behavioral characteristics. For method two (decision tree), analyses were performed using SPSS AnswerTree, version 3.1 (SPSS, Chicago, 2001). The decision tree was built using the classification and regression tree (CART) algorithm.11 The Gini impurity measure was used to determine the splits at each level of the tree. Tree growth was not limited by the use of prepruning stopping rules; the algorithm continued until no more data were available to calculate additional splits. The final tree was determined using postprun-
American Journal of Preventive Medicine, Volume 28, Number 2
Table 1. Number and characteristics of tuberculosis patients, by transmission setting and relationship Nontraditional settings and relationships Characteristic Gender Male Female Age (years) 0–4 5–14 15–24 25–44 45–64 ⬎64 Race/ethnicity White Black Hispanic Native American Asian Foreign birth HIV infection status Positive Negative Unknown Injection drug use Noninjection drug use Excess alcohol use Homelessness Previous tuberculosis disease Sputum-smear positivity Rifampin drug resistance Medication administration Self-administered DOT only Both Abnormal chest radiograph Cavitary Provider type Health department Private Both
Traditional settings and relationships
nⴝ85
(%)
nⴝ435
(%)
p value
56 29
65.9 34.1
265 170
60.9 39.1
NS Referent
1 1 22 33 25 3
1.2 1.2 25.9 38.8 29.4 3.5
36 8 60 191 104 36
8.3 1.8 13.8 43.9 23.9 8.3
Referent Referent 0.002** NS 0.016* NS
26 53 4 0 2 12
30.6 62.4 4.7 0 2.3 14.1
73 218 87 7 49 120
16.8 50.1 20 1.6 11.3 27.6
<0.001*** <0.001*** Referent
19 34 32 9 14 33 14 5 47 3
22.4 40 37.6 10.6 16.5 38.8 16.5 5.9 55.3 3.5
43 160 232 24 57 102 40 28 224 6
9.9 36.8 53.3 5.5 13.1 23.5 9.2 6.4 51.5 1.4
0.026* Referent NS NS NS 0.003** 0.044* NS NS NS
8 58 15 79 26
9.4 68.2 17.7 92.9 30.6
70 197 136 399 145
16.1 45.3 31.3 91.7 33.3
0.015* Referent 0.001** NS NS
61 5 14
71.8 5.9 16.5
227 91 89
52.2 20.9 20.5
Referent <0.001*** NS
NS 0.009**
*p⬍0.05; **p⫽⬍0.01; ***p⫽⬍0.001 (bolded). DOT, directly observed therapy; NS, not significant.
ing in which the smallest tree that minimized the misclassification rate was chosen. We used a fivefold cross-validation scheme to assess the accuracy of the tree. The data were randomly divided into five mutually exclusive groups of equal size; five trees were created in rotation using four of five groups, and the accuracy was calculated using the remaining group. The accuracy rates for the five iterations were then averaged to generate the overall accuracy. For method three (multivariate regression), crude odds ratios (ORs) and 95% confidence intervals were calculated using a GEE regression model. GEE methods were used to control for intracluster dependence among cases in the same genotypic cluster. GEE models were also used for the multivariate analyses to calculate the adjusted ORs and 95% confidence intervals. An exchangeable correlation matrix was used for all GEE models. All levels of predictors with a p value of ⬍0.20 as well as potential confounders were included in the multivariate model. The z-statistic and corresponding p values were used to determine
interaction terms that were significantly associated with the outcome. The final model was formed using a hierarchical backward elimination strategy and the z-statistic. Each variable was included in the final model if it significantly (p ⬍0.05) increased the power of the model. Confounders were kept in the model if they changed the ORs of statistically significant terms by ⬎10% or increased the precision of the estimates.
Results Of 11,923 reported culture-positive NTGSN TB cases, 10,844 (91%) M. tuberculosis isolates were genotyped. Among the 10,844 TB patients with genotyped isolates, 4727 (43.6%) were clustered with at least one other isolate in the same NTGSN site. A total of 520 (11%) met our definition of representing recent TB transmission by having epidemiologic linkages discovered by Am J Prev Med 2005;28(2)
203
Table 2. Type and distribution of nontraditional settings and relationships Type
Number (%)
Bar Social club or scene Hang-out Church or temple Drug house or motel Crackhouse Carwash Customer Hairdresser Carpool Club Total
25 (29) 17 (20) 12 (14) 8 (10) 6 (7) 5 (6) 4 (5) 2 (2) 2 (2) 2 (2) 2 (2) 85
either conventional contact investigation or cluster investigation and also having matching genotypes. From these, 85 (16.4%) had non-TSR (cases) and 435 (83.6%) had TSR (controls). For method one (descriptive), significant differences were observed in age, race/ethnicity, birth place, HIV infection status, excess alcohol use, homeless status, use of directly observed therapy, and provider type of TB patients in non-TSR compared to those in TSR (Table 1). Bars were the most frequently reported setting of transmission for non-TSR (n ⫽25), followed by “social clubs” (n ⫽17) (Table 2). For method two (decision tree), the tree identified race/ethnicity (non-Hispanic white or black) as having the greatest predictive ability to determine patients in non-TSR, followed by age 15 to 24 years, and having positive or unknown HIV infection status (Figure 1). Two subgroups of patients were more likely to be in
non-TSR. The first included patients who were nonHispanic white or non-Hispanic black, aged 15 to 24, and either HIV-infected or status unknown (56.7% non-TSR for the subgroup vs 16.4% non-TSR for the study population). The first subgroup had a 3.5-fold increase in non-TSR over the study population, respectively. In a second subgroup, TB patients in other age groups who were non-Hispanic white or non-Hispanic black, the decision tree identified excess alcohol use to be a determinant of increased rates of non-TSR (28.3% non-TSR for the subgroup vs 16.4% non-TSR for the study population). The first subgroup, which had the highest percentage of non-TSR (56.7%), identified 17 non-TSR TB patients of 30 in the group or 20% of all non-TSR patients in the study population. The second subgroup, made up of patients who were non-Hispanic white or non-Hispanic black with excess alcohol use, but not in the 15-to-24 age group, had a lower percentage of non-TSR (28.3%), yet identified 30 of 106 in the group, or 35% of the non-TSR patients in the study population. Cross-validation of the decision tree yielded accuracy of 82%. For method three (multivariate regression), in crude analyses, TB patients in non-TSR were significantly more likely than those in TSR to be non-Hispanic white or excess alcohol users. TB patients in non-TSR were significantly less likely than those in TSR to be treated by a private provider only or to have an unknown HIV infection status (Table 3). In adjusted analyses, TB patients in non-TSR were significantly more likely than TB cases in TSR to be non-Hispanic white or have an M. tuberculosis isolate
Figure 1. A decision-tree to profile tuberculosis patients in nontraditional settings and relationships. Non-TSR, nontraditional settings and relationships.
204
American Journal of Preventive Medicine, Volume 28, Number 2
Table 3. Predictors associated with tuberculosis patients in nontraditional settings and relationships using generalized estimating equation analysis Characteristic Gender Male Female Age (years) 0–4 5–14 15–24 25–44 45–64 ⱖ65 Race/ethnicity White Black Hispanic Asian Foreign birth HIV status Positive Negative Unknown Injection drug use Noninjection drug use Excess alcohol use Homelessness Previous tuberculosis disease Sputum-smear positive Rifampin drug resistance Medication administration Self-administration DOT only Both Abnormal chest radiograph Cavitary Provider type Health department Private Both
Crude odds ratio (95% CI)
Adjusted odds ratio (95% CI)
Referent 0.8 (0.5–1.3) Referent Referent 5.1 (0.7–36.4) 3.4 (0.5–22.2) 4.5 (0.8–26.6) 2.4 (0.2–24.1) 7.1 (2.0–25.6) 3.4 (0.9–12.3) Referent 0.8 (0.1–10.9) 0.5 (0.2–1.2)
6.1 (1.7–21.1)
1 (0.4–2.5) Referent 0.4 (0.2–0.9) 1.3 (0.4–4.8) 1.1 (0.6–2.0) 2.2 (1.1–4.4) 2.2 (0.9–5.8) 1 (0.4–2.8) 1.1 (0.7–1.7) 3.4 (0.8–14.9)
5.2 (1.5–17.7)
0.6 (0.3–1.2) Referent 0.5 (0.2–1.2) 1.1 (0.7–1.6) Referent 0.3 (0.1–0.8) 0.7 (0.3–1.7)
0.3 (0.1–0.9)
CI, confidence interval; DOT, directly observed therapy.
resistant to rifampin. TB patients in non-TSR were significantly less likely than TB cases in TSR to be treated by a private provider. No interaction terms were significant. The final model included the three significant variables, adjusted for patient age and foreignbirth status. The subgroup of patients who were non-Hispanic white and not treated by a private provider identified 24 non-TSR patients of 85 in the subgroup, or 28.2% of the non-TSR patients in the study population. Rifampin drug resistance was not considered in the subgroup owing to the small number of rifampin-resistant patients in the study population.
Discussion Among the TB patients in this study (n⫽520), non-TSR accounted for 85 (16.4%), and TSR, 435 (83.6%). The decision tree analysis identified race/ethnicity as having the greatest predictive ability to determine TB patients in non-TSR, followed by age and HIV infection status. In multivariate regression, TB patients in nonTSR were significantly more likely than those in TSR to be non-Hispanic white or have an M. tuberculosis isolate that was resistant to rifampin. Further, TB patients in non-TSR were significantly less likely than TB patients in TSR to be treated by a private provider. Other molecular epidemiologic studies of TB transmission have suggested that transmission in non-TSR is more widespread than previously thought.1,3,4,6,7,13–17 In one study, 40% of contacts in the United States were identified in the third outer ring of the concentric circle.15 In another, using data from the genotyping of M. tuberculosis isolates of TB cases in the same households, 31% had unsubstantiated epidemiologic linkages discovered during conventional contact investigation.16 This implies that TB transmission might have occurred outside the household in spite of the presence of another person with TB in that household. We also know that cluster investigation of TB patients with matched genotypes (not previously linked by conventional contact investigation) adds, at a minimum, 38% more epidemiologic linkages.7 These added epidemiologic linkages discovered by cluster investigation are more likely found in non-TSR and among larger clusters.7 Taken together, these findings suggest that TB transmission dynamics in the United States (a low-morbidity country) may be changing, and may indicate that TB transmission now occurs more frequently in non-TSR (e.g., bars, social clubs, and churches). This has also been reported in high-burden countries.17 With a mobile society, changing social dynamics, and diverse lifestyles, different TB transmission dynamics might be expected. Predictive models can enhance TB program performance. When used previously in TB prevention and control, decision tree models effectively and efficiently identified TB contacts most likely to have a positive TST result10 and TB patients not requiring isolation.18 They have also been successfully applied as an additional diagnostic tool for use in hospital emergency rooms.19 The decision tree creates a set of simple classification rules, and provides information related to the potential yield in non-TSR TB cases versus the total number of cases to be investigated from each patient subgroup that the tree identifies. The advantages of decision tree models include the ability to model nonlinear relationships and automatically capture multilevel interactions among predictors. They generate a dependent, interAm J Prev Med 2005;28(2)
205
active solution, whereas more traditional logistic models provide the independent impact of each predictor. A TB controller might apply the decision tree by first investigating patients who are non-Hispanic white or black, HIV-infected or unknown, and aged 15 to 24, and ask about bars and social clubs that they might frequent. The controller might also investigate where they congregate to share drugs. By focusing on this group, there is a ⬎50% chance of identifying non-TSR for these patients. If resources allow, the TB controller could next look at non-Hispanic whites or blacks who are aged ⱖ25 years and use alcohol excessively to identify additional cases in non-TSR. The decision tree identified a subgroup of patients in this study with a much higher percentage of non-TSR than in the multivariate regressin model, perhaps due to the decision tree’s ability to uncover multilevel interactions that are associated with an increased concentration of patients in non-TSR. Decision trees are more easily and effectively applied in public health program settings, and thus have greater utility, than multivariate regression models.10 The multivariate model and the decision tree yielded different results owing to different methodologies in arriving at a prediction. The decision tree may be valuable as a new tool for assisting TB field workers in determining which TB cases should be investigated for contacts in non-TSR. This study had several limitations, including the potential for information bias. Because conventional contact investigation occurred first when the TB case became known to public health officials, whereas cluster investigation occurred later after the availability of genotyping results from the M. tuberculosis isolate, there existed a potential information bias with respect to conventional contact investigation and cluster investigation. That is, the information used to ascertain non-TSR status and predictors of non-TSR was potentially biased by the method of data collection (conventional contact investigation vs cluster investigation), but the impact of this potential bias is unknown. A variable representing whether the case was from an epidemiologic link found during conventional contact investigation versus cluster investigation was not a confounder (data not shown). We also observed that nonwhite and nonblack subpopulations could not be adequately studied because of the small numbers. Additionally, the study did not include the duration of exposure nor the volume of air shared so as to characterize and evaluate casual TB transmission. The 520 patients included in the study represent the subset of TB patients that were involved in recent transmission, and therefore the study results cannot be generalized to a larger population of all TB cases in the United States. At least one other study found a higher percentage of cases with epidemiologic links 206
What This Study Adds . . . The availability of molecular epidemiologic methods for studying tuberculosis transmission has revealed that transmission in nontraditional settings and relationships is more common than previously thought, and conventional contact investigations often fail to detect these occurrences. This study applies decision-trees to characterize tuberculosis transmission in nontraditional settings and relationships, thus enhancing the effectiveness of conventional contact investigations. Decision-tree profiles are easily applied by public health practitioners.
discovered by contact investigation or cluster investigation that were confirmed by genotyping.20 One reason we had a lower percentage discovered of epidemiologic links was that cluster investigations were performed for only four of seven NTGSN sites, and for a subset of the years of the study. We also required a site to ascertain a known setting and type of relationship to be an epidemiologic link in the study. Real-time access to genotyping results became a reality in the United States with the rollout of the Centers for Disease Control and Prevention’s Tuberculosis Genotyping Program in 2004.21 Genotyping results may be used to complement contact and cluster investigation by confirming epidemiologic links as representing recent transmission; they are also useful in locating additional cases of TB in non-TSR.7 The use of decision trees to profile TB patients at increased risk of being in non-TSR, combined with other epidemiologic techniques such as social network analysis,22 will enhance both the efficiency and effectiveness of conventional contact investigation. Decision tree analyses, plus real-time access to M. tuberculosis genotyping, can be used to enhance both the efficiency and effectiveness of TB prevention and control activities in identifying patients in non-TSR. We acknowledge and are grateful to Jack Crawford, PhD, Chris Braden, MD, Ida Onorato, MD, and Barbara Schable for their roles in the design and implementation of the NTGSN. Further, we thank the participating local and state TB control offices, including the laboratory investigators that supported the NTGSN; Wendy Cronin, PhD, of the Maryland Department of Health and Mental Hygiene, and Martien Borgdorff, MD, PhD, of the KNCV tuberculosis foundation, for their technical assistance. No financial conflict of interest was reported by the authors of this paper.
American Journal of Preventive Medicine, Volume 28, Number 2
References 1. Golub JE, Cronin WA, Obasanjo OO, et al. Transmission of Mycobacterium tuberculosis through casual contact with an infectious case. Arch Intern Med 2001;161:2254 – 8. 2. Weis SE, Pogoda JM, Yang Z, et al. Transmission dynamics of tuberculosis in Tarrant County, Texas. Am J Respir Crit Care Med 2002;166:36 – 42. 3. Cronin WA, Golub JE, Lathan MJ, et al. Molecular epidemiology of tuberculosis in a low- to moderate-incidence state: are contact investigations enough? Emerg Infect Dis 2002;8:1271–9. 4. Tabet SR, Goldbaum GM, Hooton TM, et al. Restriction fragment length polymorphism analysis detecting a community-based tuberculosis outbreak among persons infected with human immunodeficiency virus. J Infec Dis 1994;169:189 –92. 5. Marks SM, Taylor, Z, Qualls NL, et al. Outcomes of contact investigations of infectious tuberculosis patients. Am J Respir Crit Care Med 2000;162:2033– 8. 6. Bock NN, Mallory JP, Mobley N, et al. Outbreak of tuberculosis associated with a floating card game in the rural south: lessons for tuberculosis contact investigations. Clin Infect Dis 1998;27:1221– 6. 7. McNabb SJN, Kammerer JS, Hickey AC, et al. The added epidemiologic value to tuberculosis prevention and control of the investigation of clustered genotypes of Mycobacterium tuberculosis isolates. Am J Epidemiol 2004;160:589-97. 8. Castro KG, Jaffe HW. Rationale and methods for the National Tuberculosis Genotyping and Surveillance Network. Emerg Infect Dis 2002;8:1188 –91. 9. Ellis BA, Crawford JT, Braden CR, et al. Molecular epidemiology of tuberculosis in a sentinel surveillance population. Emerg Infect Dis 2002;8:1197–209. 10. Gerald LB, Tang B, Bruce F, et al. A decision tree for tuberculosis contact investigation. Am J Respir Crit Care Med 2002;166:1122–7. 11. Breiman L, Freidman JH, Olshen RA, et al. Classification and regression trees. Belmont, CA: Wadsworth International Group, 1984.
12. Crawford J, Braden C, Schable B, Onorato I. National Tuberculosis Genotyping and Surveillance Network: Design and methods. Emerg Infect Dis 2002;8:1192– 6. 13. Small PM, Hopewell PC, Singh SP, et al. The epidemiology of tuberculosis in San Francisco: a population-based study using conventional and molecular methods. N Engl J Med 1994;330:1703–9. 14. Weis S. Contact investigations: how do they need to be designed for the 21st century? Am J Respir Crit Care Med 2002;166:1016 –7 (Editorial). 15. Sebek M. DNA fingerprinting and contact investigation. Int J Tuberc Lung Dis 2000;4:S45– 8. 16. Bennett DE, Onorato IM, Ellis BA, et al. DNA fingerprinting of Mycobacterium tuberculosis isolates from epidemiologically linked case pairs. Emerg Infect Dis 2002;8:1224 –9. 17. Verver S, Warren R, Munch Z, et al. Proportion of tuberculosis transmission that takes place in households in a high-incidence area. Lancet 2004;363:212– 4. 18. El-Solh A, Mylotte J, Sherif S, et al. Validity of a decision tree for predicting active pulmonary tuberculosis. Am J Respir Crit Care Med 1997;155:1711– 6. 19. Mair J, Smidt J, Lechleitner P, et al. A decision tree for the early diagnosis of acute myocardial infarction in nontraumatic chest pain patients at hospital admission. Chest 1995;108:1502–9. 20. Lambregts-van Weezenbeek C, Sebek M, van Gerven M, et al. Tuberculosis contact investigation and DNA fingerprint surveillance in the Netherlands: 6 years’ experience with nation-wide cluster feedback and cluster monitoring. Int J Tuberc Lung Dis 2003;7:S463–70. 21. Rosenblum LS, Navin TR, Crawford JT. Molecular epidemiology of tuberculosis. N Engl J Med 2003;349:2364 (Letter). 22. McElroy PD, Rothenberg RB, Varghese R, et al. A network-informed approach to investigating a tuberculosis outbreak: implications for enhancing contact investigations. Int J Tuberc Lung Dis 2003;7:S486 –93.
Am J Prev Med 2005;28(2)
207
In 2003–2004, we determined the number and characteristics of TB patients in non-TSR that were involved in recent transmission, generated a decision tree to profile those patients, and performed a case– control study to identify predictors of being in non-TSR.
Results:
Of 10,844 culture-positive reported TB cases that were genotyped, 4724 (43.6%) M. tuberculosis isolates were clustered with at least one other isolate. Among these, 520 (11%) had epidemiologic linkages discovered during conventional contact investigation or cluster investigation and confirmed by genotyping results. The decision tree identified race/ ethnicity (non-Hispanic white or black) as having the greatest predictive ability to determine patients in non-TSR, followed by being aged 15 to 24 years and having positive or unknown HIV infection status. From the 520, 85 (16.4%) had non-TSR, and 435 (83.6%) had traditional settings and relationships (TSR). In multivariate analyses, patients in non-TSR were significantly more likely than those in TSR to be non-Hispanic white (adjusted odds ratio [aOR]⫽6.1; 95% confidence interval [CI]⫽1.7–21.1]) or to have an M. tuberculosis isolate resistant to rifampin (aOR⫽5.2; 95% CI⫽1.5–17.7).
Conclusions: Decision-tree analyses can be used to enhance both the efficiency and effectiveness of TB prevention and control activities in identifying patients in non-TSR. (Am J Prev Med 2005;28(2):201–207) © 2005 American Journal of Preventive Medicine
Background
A
lthough increasingly common, tuberculosis (TB) transmission in nontraditional settings and relationships (non-TSR) often eludes detection by conventional contact investigation.1– 4 Conventional contact investigations for TB rely on the concentric circle approach and focus mainly on the home and close family, other relatives or friends, and the workplace. This approach is infrequently extended to non-TSR.3,5,6 We defined non-TSR in this study as settings and relationships other than the home, workplace, school, common congregate settings, relatives, friends, and co-workers. Cluster investigation, defined
From the Division of Tuberculosis Elimination, National Center for HIV, STD, and TB Prevention, Centers for Disease Control and Prevention (Kammerer, McNabb, Becerra, Rosenblum, Shang, Iademarco, Navin); and independent contractor (Kammerer), Atlanta, Georgia Address correspondence and reprint requests to: J. Steve Kammerer, Epidemiology Team, Surveillance, Epidemiology, and Outbreak Investigation Branch, Division of Tuberculosis Elimination, National Center for HIV, STD, and TB Prevention, Centers for Disease Control and Prevention, 1600 Clifton Road, Mailstop E-10, Atlanta GA 30333. E-mail: [email protected].
as the epidemiologic investigation of TB patients with genotypically matched Mycobacterium tuberculosis isolates, provided additional epidemiologic information about the possible dynamics of TB transmission in the “outer rings” of the concentric circle. Recent data suggest that patients identified by cluster investigation are more likely to have acquired TB in non-TSR than those patients identified by conventional contact investigation.3,7 The National Tuberculosis Genotyping and Surveillance Network (NTGSN) was established in seven sentinel sites, including the states of Arkansas, Maryland, Massachusetts, Michigan and New Jersey, six counties in California (Alameda, Contra Costa, Marin, San Mateo, Santa Clara, and Solano), and four counties in Texas (Tarrant, Dallas, Hildalgo, and Cameron). These sites were funded from 1996 to 2000 to assess the utility of genotyping of M. tuberculosis isolates.8,9 The target population for this study included TB patients involved in recent transmission, defined as those with epidemiologic links that were confirmed by having matching genotypes. Most patients in the NTGSN did not meet these criteria, and could not be
Am J Prev Med 2005;28(2) © 2005 American Journal of Preventive Medicine • Published by Elsevier Inc.
0749-3797/05/$–see front matter doi:10.1016/j.amepre.2004.10.011
201
confirmed as being involved in recent transmission; such patients are not addressed in the study. The objective of the study was to characterize patients that were involved in recent transmission of TB and more likely to be found in non-TSR to assist TB control programs in targeting conventional contact investigations. Three different methods were used: descriptive statistics, a decision tree, and multivariate regression modeling using a generalized estimating equation (GEE). Associations between exposures and health outcomes derived from logistic regression modeling analyses often do not translate into practical guidance for use in public health practice, including the setting of conventional TB contact investigation.10 Reasons for this include the fact that several exposures may be independently associated with the health outcome, yet from a practical point of view, the public health scientist cannot discern which are more relevant and, as such, should be targeted for intervention. The decision tree approach overcomes this problem.11 Each terminal node of the decision tree provides an estimate of the yield of the health outcome under study, with respect to a given subset of patients. For example, decision-trees have been used successfully to predict tuberculin skin test (TST) results for contacts identified during conventional contact investigations.10 In 2003–2004, we determined the number and characteristics of TB patients with non-TSR in the NTGSN. The health outcome was defined as a TB patient being in non-TSR, not TB transmission, per se. Decision-tree analyses were performed to predict which subgroups of TB patients were at increased risk of being in non-TSR. We also performed a case– control study to identify predictors of TB patients in non-TSR that included multivariate regression.
Methods Population The study population and the epidemiologic and laboratory methods used in the NTGSN have been described elsewhere.12 The NTGSN collected data from 15,035 TB patients (16% of the total number reported in the United States during this time period).9 The seven NTGSN sites reported epidemiologic linkages discovered during conventional contact investigations of TB patients with isolates culture-positive for M. tuberculosis. Cluster investigations were performed when no epidemiologic linkage was discovered during conventional contact investigation and upon the recognition of a genetic cluster. A genetic cluster refers to a group of two or more TB patients within the same NTGSN site having M. tuberculosis isolates with matched genotypes. Four sites out of seven completed cluster investigations. Three of the sites (Arkansas, Maryland, and Massachusetts) completed cluster investigations from 1998 to 2000. California completed cluster investigations solely from Santa Clara County and only from 1999 to 2000. Both medical record reviews and addi-
202
tional patient interviews were completed. The study population for this investigation included those reported TB cases with epidemiologic linkages having M. tuberculosis isolates with matched genotypes.
Data Collection The NTGSN sites recorded epidemiologic linkages discovered during conventional contact investigation and cluster investigation. We defined nontraditional settings as any setting that excludes home, school, workplace, or common congregate settings such as correctional facilities, daycare centers, emergency shelters, group quarters, acute care hospitals, nursing homes, and other long-term care facilities. Examples of nontraditional settings included bars, social clubs, “hang-outs,” and churches or temples. We defined nontraditional relationships as any relationship that excluded household and nonhousehold friend and co-worker contacts. Examples of nontraditional relationships included being in the same place at the same time such as a customer in a store with another TB patient. For some epidemiologic linkages, text describing the specific setting or relationship was recorded by the NTGSN sites. When recorded, we reviewed and categorized these as nonTSR or TSR. For each TB patient with an epidemiologic linkage, we gathered demographic, behavioral, and clinical data from the national Report of Verified Case of TB (RVCT). In this study, HIV infection status was categorized as positive (i.e., “infected”), negative, or unknown. The health outcome was defined as a TB patient in non-TSR versus TSR, regardless of whether the case was a presumed source case or a secondary case in the chain of transmission. Patients who were identified in multiple epidemiologic linkages were included in the traditional category only if all epidemiologic linkages involving the patient were traditional.
Study Design We performed a case– control study to identify predictors of TB patients in non-TSR. Cases were defined as all TB patients in non-TSR (n ⫽85) that were involved in recent transmission. Controls were defined as all TB patients in TSR (n ⫽435) that were involved in recent transmission. The decision tree analyses used the same set of TB patients.
Data Analyses Descriptive statistics and univariate and multivariate models were performed using SAS, version 8.2 (SAS Institute Inc., Cary NC, 2001). For method one (descriptive), the Pearson chi-square or two-tailed Fisher’s exact test, as appropriate, was used to assess difference in proportions between the non-TSR and TSR populations regarding demographic, clinical, and behavioral characteristics. For method two (decision tree), analyses were performed using SPSS AnswerTree, version 3.1 (SPSS, Chicago, 2001). The decision tree was built using the classification and regression tree (CART) algorithm.11 The Gini impurity measure was used to determine the splits at each level of the tree. Tree growth was not limited by the use of prepruning stopping rules; the algorithm continued until no more data were available to calculate additional splits. The final tree was determined using postprun-
American Journal of Preventive Medicine, Volume 28, Number 2
Table 1. Number and characteristics of tuberculosis patients, by transmission setting and relationship Nontraditional settings and relationships Characteristic Gender Male Female Age (years) 0–4 5–14 15–24 25–44 45–64 ⬎64 Race/ethnicity White Black Hispanic Native American Asian Foreign birth HIV infection status Positive Negative Unknown Injection drug use Noninjection drug use Excess alcohol use Homelessness Previous tuberculosis disease Sputum-smear positivity Rifampin drug resistance Medication administration Self-administered DOT only Both Abnormal chest radiograph Cavitary Provider type Health department Private Both
Traditional settings and relationships
nⴝ85
(%)
nⴝ435
(%)
p value
56 29
65.9 34.1
265 170
60.9 39.1
NS Referent
1 1 22 33 25 3
1.2 1.2 25.9 38.8 29.4 3.5
36 8 60 191 104 36
8.3 1.8 13.8 43.9 23.9 8.3
Referent Referent 0.002** NS 0.016* NS
26 53 4 0 2 12
30.6 62.4 4.7 0 2.3 14.1
73 218 87 7 49 120
16.8 50.1 20 1.6 11.3 27.6
<0.001*** <0.001*** Referent
19 34 32 9 14 33 14 5 47 3
22.4 40 37.6 10.6 16.5 38.8 16.5 5.9 55.3 3.5
43 160 232 24 57 102 40 28 224 6
9.9 36.8 53.3 5.5 13.1 23.5 9.2 6.4 51.5 1.4
0.026* Referent NS NS NS 0.003** 0.044* NS NS NS
8 58 15 79 26
9.4 68.2 17.7 92.9 30.6
70 197 136 399 145
16.1 45.3 31.3 91.7 33.3
0.015* Referent 0.001** NS NS
61 5 14
71.8 5.9 16.5
227 91 89
52.2 20.9 20.5
Referent <0.001*** NS
NS 0.009**
*p⬍0.05; **p⫽⬍0.01; ***p⫽⬍0.001 (bolded). DOT, directly observed therapy; NS, not significant.
ing in which the smallest tree that minimized the misclassification rate was chosen. We used a fivefold cross-validation scheme to assess the accuracy of the tree. The data were randomly divided into five mutually exclusive groups of equal size; five trees were created in rotation using four of five groups, and the accuracy was calculated using the remaining group. The accuracy rates for the five iterations were then averaged to generate the overall accuracy. For method three (multivariate regression), crude odds ratios (ORs) and 95% confidence intervals were calculated using a GEE regression model. GEE methods were used to control for intracluster dependence among cases in the same genotypic cluster. GEE models were also used for the multivariate analyses to calculate the adjusted ORs and 95% confidence intervals. An exchangeable correlation matrix was used for all GEE models. All levels of predictors with a p value of ⬍0.20 as well as potential confounders were included in the multivariate model. The z-statistic and corresponding p values were used to determine
interaction terms that were significantly associated with the outcome. The final model was formed using a hierarchical backward elimination strategy and the z-statistic. Each variable was included in the final model if it significantly (p ⬍0.05) increased the power of the model. Confounders were kept in the model if they changed the ORs of statistically significant terms by ⬎10% or increased the precision of the estimates.
Results Of 11,923 reported culture-positive NTGSN TB cases, 10,844 (91%) M. tuberculosis isolates were genotyped. Among the 10,844 TB patients with genotyped isolates, 4727 (43.6%) were clustered with at least one other isolate in the same NTGSN site. A total of 520 (11%) met our definition of representing recent TB transmission by having epidemiologic linkages discovered by Am J Prev Med 2005;28(2)
203
Table 2. Type and distribution of nontraditional settings and relationships Type
Number (%)
Bar Social club or scene Hang-out Church or temple Drug house or motel Crackhouse Carwash Customer Hairdresser Carpool Club Total
25 (29) 17 (20) 12 (14) 8 (10) 6 (7) 5 (6) 4 (5) 2 (2) 2 (2) 2 (2) 2 (2) 85
either conventional contact investigation or cluster investigation and also having matching genotypes. From these, 85 (16.4%) had non-TSR (cases) and 435 (83.6%) had TSR (controls). For method one (descriptive), significant differences were observed in age, race/ethnicity, birth place, HIV infection status, excess alcohol use, homeless status, use of directly observed therapy, and provider type of TB patients in non-TSR compared to those in TSR (Table 1). Bars were the most frequently reported setting of transmission for non-TSR (n ⫽25), followed by “social clubs” (n ⫽17) (Table 2). For method two (decision tree), the tree identified race/ethnicity (non-Hispanic white or black) as having the greatest predictive ability to determine patients in non-TSR, followed by age 15 to 24 years, and having positive or unknown HIV infection status (Figure 1). Two subgroups of patients were more likely to be in
non-TSR. The first included patients who were nonHispanic white or non-Hispanic black, aged 15 to 24, and either HIV-infected or status unknown (56.7% non-TSR for the subgroup vs 16.4% non-TSR for the study population). The first subgroup had a 3.5-fold increase in non-TSR over the study population, respectively. In a second subgroup, TB patients in other age groups who were non-Hispanic white or non-Hispanic black, the decision tree identified excess alcohol use to be a determinant of increased rates of non-TSR (28.3% non-TSR for the subgroup vs 16.4% non-TSR for the study population). The first subgroup, which had the highest percentage of non-TSR (56.7%), identified 17 non-TSR TB patients of 30 in the group or 20% of all non-TSR patients in the study population. The second subgroup, made up of patients who were non-Hispanic white or non-Hispanic black with excess alcohol use, but not in the 15-to-24 age group, had a lower percentage of non-TSR (28.3%), yet identified 30 of 106 in the group, or 35% of the non-TSR patients in the study population. Cross-validation of the decision tree yielded accuracy of 82%. For method three (multivariate regression), in crude analyses, TB patients in non-TSR were significantly more likely than those in TSR to be non-Hispanic white or excess alcohol users. TB patients in non-TSR were significantly less likely than those in TSR to be treated by a private provider only or to have an unknown HIV infection status (Table 3). In adjusted analyses, TB patients in non-TSR were significantly more likely than TB cases in TSR to be non-Hispanic white or have an M. tuberculosis isolate
Figure 1. A decision-tree to profile tuberculosis patients in nontraditional settings and relationships. Non-TSR, nontraditional settings and relationships.
204
American Journal of Preventive Medicine, Volume 28, Number 2
Table 3. Predictors associated with tuberculosis patients in nontraditional settings and relationships using generalized estimating equation analysis Characteristic Gender Male Female Age (years) 0–4 5–14 15–24 25–44 45–64 ⱖ65 Race/ethnicity White Black Hispanic Asian Foreign birth HIV status Positive Negative Unknown Injection drug use Noninjection drug use Excess alcohol use Homelessness Previous tuberculosis disease Sputum-smear positive Rifampin drug resistance Medication administration Self-administration DOT only Both Abnormal chest radiograph Cavitary Provider type Health department Private Both
Crude odds ratio (95% CI)
Adjusted odds ratio (95% CI)
Referent 0.8 (0.5–1.3) Referent Referent 5.1 (0.7–36.4) 3.4 (0.5–22.2) 4.5 (0.8–26.6) 2.4 (0.2–24.1) 7.1 (2.0–25.6) 3.4 (0.9–12.3) Referent 0.8 (0.1–10.9) 0.5 (0.2–1.2)
6.1 (1.7–21.1)
1 (0.4–2.5) Referent 0.4 (0.2–0.9) 1.3 (0.4–4.8) 1.1 (0.6–2.0) 2.2 (1.1–4.4) 2.2 (0.9–5.8) 1 (0.4–2.8) 1.1 (0.7–1.7) 3.4 (0.8–14.9)
5.2 (1.5–17.7)
0.6 (0.3–1.2) Referent 0.5 (0.2–1.2) 1.1 (0.7–1.6) Referent 0.3 (0.1–0.8) 0.7 (0.3–1.7)
0.3 (0.1–0.9)
CI, confidence interval; DOT, directly observed therapy.
resistant to rifampin. TB patients in non-TSR were significantly less likely than TB cases in TSR to be treated by a private provider. No interaction terms were significant. The final model included the three significant variables, adjusted for patient age and foreignbirth status. The subgroup of patients who were non-Hispanic white and not treated by a private provider identified 24 non-TSR patients of 85 in the subgroup, or 28.2% of the non-TSR patients in the study population. Rifampin drug resistance was not considered in the subgroup owing to the small number of rifampin-resistant patients in the study population.
Discussion Among the TB patients in this study (n⫽520), non-TSR accounted for 85 (16.4%), and TSR, 435 (83.6%). The decision tree analysis identified race/ethnicity as having the greatest predictive ability to determine TB patients in non-TSR, followed by age and HIV infection status. In multivariate regression, TB patients in nonTSR were significantly more likely than those in TSR to be non-Hispanic white or have an M. tuberculosis isolate that was resistant to rifampin. Further, TB patients in non-TSR were significantly less likely than TB patients in TSR to be treated by a private provider. Other molecular epidemiologic studies of TB transmission have suggested that transmission in non-TSR is more widespread than previously thought.1,3,4,6,7,13–17 In one study, 40% of contacts in the United States were identified in the third outer ring of the concentric circle.15 In another, using data from the genotyping of M. tuberculosis isolates of TB cases in the same households, 31% had unsubstantiated epidemiologic linkages discovered during conventional contact investigation.16 This implies that TB transmission might have occurred outside the household in spite of the presence of another person with TB in that household. We also know that cluster investigation of TB patients with matched genotypes (not previously linked by conventional contact investigation) adds, at a minimum, 38% more epidemiologic linkages.7 These added epidemiologic linkages discovered by cluster investigation are more likely found in non-TSR and among larger clusters.7 Taken together, these findings suggest that TB transmission dynamics in the United States (a low-morbidity country) may be changing, and may indicate that TB transmission now occurs more frequently in non-TSR (e.g., bars, social clubs, and churches). This has also been reported in high-burden countries.17 With a mobile society, changing social dynamics, and diverse lifestyles, different TB transmission dynamics might be expected. Predictive models can enhance TB program performance. When used previously in TB prevention and control, decision tree models effectively and efficiently identified TB contacts most likely to have a positive TST result10 and TB patients not requiring isolation.18 They have also been successfully applied as an additional diagnostic tool for use in hospital emergency rooms.19 The decision tree creates a set of simple classification rules, and provides information related to the potential yield in non-TSR TB cases versus the total number of cases to be investigated from each patient subgroup that the tree identifies. The advantages of decision tree models include the ability to model nonlinear relationships and automatically capture multilevel interactions among predictors. They generate a dependent, interAm J Prev Med 2005;28(2)
205
active solution, whereas more traditional logistic models provide the independent impact of each predictor. A TB controller might apply the decision tree by first investigating patients who are non-Hispanic white or black, HIV-infected or unknown, and aged 15 to 24, and ask about bars and social clubs that they might frequent. The controller might also investigate where they congregate to share drugs. By focusing on this group, there is a ⬎50% chance of identifying non-TSR for these patients. If resources allow, the TB controller could next look at non-Hispanic whites or blacks who are aged ⱖ25 years and use alcohol excessively to identify additional cases in non-TSR. The decision tree identified a subgroup of patients in this study with a much higher percentage of non-TSR than in the multivariate regressin model, perhaps due to the decision tree’s ability to uncover multilevel interactions that are associated with an increased concentration of patients in non-TSR. Decision trees are more easily and effectively applied in public health program settings, and thus have greater utility, than multivariate regression models.10 The multivariate model and the decision tree yielded different results owing to different methodologies in arriving at a prediction. The decision tree may be valuable as a new tool for assisting TB field workers in determining which TB cases should be investigated for contacts in non-TSR. This study had several limitations, including the potential for information bias. Because conventional contact investigation occurred first when the TB case became known to public health officials, whereas cluster investigation occurred later after the availability of genotyping results from the M. tuberculosis isolate, there existed a potential information bias with respect to conventional contact investigation and cluster investigation. That is, the information used to ascertain non-TSR status and predictors of non-TSR was potentially biased by the method of data collection (conventional contact investigation vs cluster investigation), but the impact of this potential bias is unknown. A variable representing whether the case was from an epidemiologic link found during conventional contact investigation versus cluster investigation was not a confounder (data not shown). We also observed that nonwhite and nonblack subpopulations could not be adequately studied because of the small numbers. Additionally, the study did not include the duration of exposure nor the volume of air shared so as to characterize and evaluate casual TB transmission. The 520 patients included in the study represent the subset of TB patients that were involved in recent transmission, and therefore the study results cannot be generalized to a larger population of all TB cases in the United States. At least one other study found a higher percentage of cases with epidemiologic links 206
What This Study Adds . . . The availability of molecular epidemiologic methods for studying tuberculosis transmission has revealed that transmission in nontraditional settings and relationships is more common than previously thought, and conventional contact investigations often fail to detect these occurrences. This study applies decision-trees to characterize tuberculosis transmission in nontraditional settings and relationships, thus enhancing the effectiveness of conventional contact investigations. Decision-tree profiles are easily applied by public health practitioners.
discovered by contact investigation or cluster investigation that were confirmed by genotyping.20 One reason we had a lower percentage discovered of epidemiologic links was that cluster investigations were performed for only four of seven NTGSN sites, and for a subset of the years of the study. We also required a site to ascertain a known setting and type of relationship to be an epidemiologic link in the study. Real-time access to genotyping results became a reality in the United States with the rollout of the Centers for Disease Control and Prevention’s Tuberculosis Genotyping Program in 2004.21 Genotyping results may be used to complement contact and cluster investigation by confirming epidemiologic links as representing recent transmission; they are also useful in locating additional cases of TB in non-TSR.7 The use of decision trees to profile TB patients at increased risk of being in non-TSR, combined with other epidemiologic techniques such as social network analysis,22 will enhance both the efficiency and effectiveness of conventional contact investigation. Decision tree analyses, plus real-time access to M. tuberculosis genotyping, can be used to enhance both the efficiency and effectiveness of TB prevention and control activities in identifying patients in non-TSR. We acknowledge and are grateful to Jack Crawford, PhD, Chris Braden, MD, Ida Onorato, MD, and Barbara Schable for their roles in the design and implementation of the NTGSN. Further, we thank the participating local and state TB control offices, including the laboratory investigators that supported the NTGSN; Wendy Cronin, PhD, of the Maryland Department of Health and Mental Hygiene, and Martien Borgdorff, MD, PhD, of the KNCV tuberculosis foundation, for their technical assistance. No financial conflict of interest was reported by the authors of this paper.
American Journal of Preventive Medicine, Volume 28, Number 2
References 1. Golub JE, Cronin WA, Obasanjo OO, et al. Transmission of Mycobacterium tuberculosis through casual contact with an infectious case. Arch Intern Med 2001;161:2254 – 8. 2. Weis SE, Pogoda JM, Yang Z, et al. Transmission dynamics of tuberculosis in Tarrant County, Texas. Am J Respir Crit Care Med 2002;166:36 – 42. 3. Cronin WA, Golub JE, Lathan MJ, et al. Molecular epidemiology of tuberculosis in a low- to moderate-incidence state: are contact investigations enough? Emerg Infect Dis 2002;8:1271–9. 4. Tabet SR, Goldbaum GM, Hooton TM, et al. Restriction fragment length polymorphism analysis detecting a community-based tuberculosis outbreak among persons infected with human immunodeficiency virus. J Infec Dis 1994;169:189 –92. 5. Marks SM, Taylor, Z, Qualls NL, et al. Outcomes of contact investigations of infectious tuberculosis patients. Am J Respir Crit Care Med 2000;162:2033– 8. 6. Bock NN, Mallory JP, Mobley N, et al. Outbreak of tuberculosis associated with a floating card game in the rural south: lessons for tuberculosis contact investigations. Clin Infect Dis 1998;27:1221– 6. 7. McNabb SJN, Kammerer JS, Hickey AC, et al. The added epidemiologic value to tuberculosis prevention and control of the investigation of clustered genotypes of Mycobacterium tuberculosis isolates. Am J Epidemiol 2004;160:589-97. 8. Castro KG, Jaffe HW. Rationale and methods for the National Tuberculosis Genotyping and Surveillance Network. Emerg Infect Dis 2002;8:1188 –91. 9. Ellis BA, Crawford JT, Braden CR, et al. Molecular epidemiology of tuberculosis in a sentinel surveillance population. Emerg Infect Dis 2002;8:1197–209. 10. Gerald LB, Tang B, Bruce F, et al. A decision tree for tuberculosis contact investigation. Am J Respir Crit Care Med 2002;166:1122–7. 11. Breiman L, Freidman JH, Olshen RA, et al. Classification and regression trees. Belmont, CA: Wadsworth International Group, 1984.
12. Crawford J, Braden C, Schable B, Onorato I. National Tuberculosis Genotyping and Surveillance Network: Design and methods. Emerg Infect Dis 2002;8:1192– 6. 13. Small PM, Hopewell PC, Singh SP, et al. The epidemiology of tuberculosis in San Francisco: a population-based study using conventional and molecular methods. N Engl J Med 1994;330:1703–9. 14. Weis S. Contact investigations: how do they need to be designed for the 21st century? Am J Respir Crit Care Med 2002;166:1016 –7 (Editorial). 15. Sebek M. DNA fingerprinting and contact investigation. Int J Tuberc Lung Dis 2000;4:S45– 8. 16. Bennett DE, Onorato IM, Ellis BA, et al. DNA fingerprinting of Mycobacterium tuberculosis isolates from epidemiologically linked case pairs. Emerg Infect Dis 2002;8:1224 –9. 17. Verver S, Warren R, Munch Z, et al. Proportion of tuberculosis transmission that takes place in households in a high-incidence area. Lancet 2004;363:212– 4. 18. El-Solh A, Mylotte J, Sherif S, et al. Validity of a decision tree for predicting active pulmonary tuberculosis. Am J Respir Crit Care Med 1997;155:1711– 6. 19. Mair J, Smidt J, Lechleitner P, et al. A decision tree for the early diagnosis of acute myocardial infarction in nontraumatic chest pain patients at hospital admission. Chest 1995;108:1502–9. 20. Lambregts-van Weezenbeek C, Sebek M, van Gerven M, et al. Tuberculosis contact investigation and DNA fingerprint surveillance in the Netherlands: 6 years’ experience with nation-wide cluster feedback and cluster monitoring. Int J Tuberc Lung Dis 2003;7:S463–70. 21. Rosenblum LS, Navin TR, Crawford JT. Molecular epidemiology of tuberculosis. N Engl J Med 2003;349:2364 (Letter). 22. McElroy PD, Rothenberg RB, Varghese R, et al. A network-informed approach to investigating a tuberculosis outbreak: implications for enhancing contact investigations. Int J Tuberc Lung Dis 2003;7:S486 –93.
Am J Prev Med 2005;28(2)
207