Tuberculosis is a major cause of morbidity and mortality. The HIV epidemic has resulted in dramatic increases in notifications of tuberculosis, in particular in sub-Saharan Africa . Although a direct impact of HIV on the progression of latent tuberculosis infection to disease is well established, the more indirect impact on the tuberculosis epidemic through increased transmission from HIV-infected patients with tuberculosis is less clear [2,3]. It has been argued that the potential for those with HIV-associated tuberculosis to transmit infection to the non-HIV-infected population could result in tuberculosis spinning out of control . Indeed, the annual risk (i.e., age-adjusted prevalence) of tuberculosis infection (ARTI) in Kenya increased in districts where tuberculosis notification and HIV prevalence among patients with tuberculosis had increased , and in South Africa the incidence of tuberculosis among HIV-negative gold miners increased concurrently with that among HIV-positive miners . Trends in case notifications in Ivory Coast, South Africa and Malawi, however, have suggest limited impact of the HIV epidemic on tuberculosis incidence in the HIV-negative population [7–9]. More direct evidence for this comes from repeated tuberculin surveys in Tanzania, which were started in 1983 after the introduction of a national tuberculosis control programme following the principles later adopted by the World Health Organization as its global Directly Observed Therapy Short-Course (DOTS) strategy. Conducted until 1998, these surveys showed no increase in ARTI despite a near twofold increase in notification rates of sputum smear-positive tuberculosis [10–12].
In many parts of Tanzania however, the increase in HIV prevalence (2002 estimate 8.8% among adults) and tuberculosis notification started only in the early 1990s . Since the ARTI reflects cumulative lifetime infection risk, tuberculin surveys measure the force of infection with considerable delay . Consequently, an increase in ARTI as a result of HIV-associated transmission of Mycobacterium tuberculosis could have been missed. Here we present data from the fourth round of this survey, conducted over the period 2000–2003, and compare these with data from the three earlier rounds to assess the trend in ARTI as an indicator of the impact of the HIV epidemic on the tuberculosis epidemic in Tanzania.
All four surveys were conducted by similar methods in the 59 districts that had been selected for the first round out of a total of (then) 94 districts of mainland Tanzania . The sampling scheme had been statistically indistinguishable from a simple random sample of districts taken at national level (χ2 = 1.4; two degrees of freedom). New districts that were split off the original district over time were treated together with the parent district as a single sampling unit. In each district, schools were selected by simple random sampling. Contrary to the previous rounds, in which the same schools were selected based on a total of 1200 per district in round 1, a new selection was made in order to obtain approximately 1500 children per district.
As in the previous rounds, all eligible children in grade 1 and 2 of each sampled school were selected. Tuberculin skin testing was performed by the Mantoux method following international guidelines using 2TU of RT23 in Tween-80 (State Serum Institute, Copenhagen, Denmark), administered with 1 ml disposable syringes and disposable 26 gauge needles on the dorsal side of the left forearm . Reactions were read after 48 to 96 h. All tests were administered and read by two teams trained by the same reference nurse. The BCG status was recorded as presence of a typical scar; BCG vaccination is given routinely at birth.
Administrative and ethical clearance was obtained from the Ministry of Health. Regional and district education officers and teachers of the selected schools were informed about the purpose and procedure of the test. Parents were informed through the schoolteachers; formal informed consent was not obtained. Children with reaction sizes ≥ 16 mm were examined for tuberculosis disease. They were offered treatment in accordance with national guidelines if tuberculosis disease was diagnosed and isoniazid preventive treatment if not.
Children who were absent from school could have been so because they were orphans as result of tuberculosis (with or without HIV infection) in their parents; consequently, the survey might have missed the children most at risk of tuberculosis infection. In order to assess this potential selection bias, the teachers were asked, as of mid-2002, whether children enrolled in their grades had any deceased parent.
Data were recorded on standard forms and double entered in EpiInfo version 6.0 (Centers for Disease Control and Prevention, Atlanta, Georgia, USA); discrepancies were checked against the raw data. Analyses were carried out in Stata version 8 (Stata Corp., College Station Texas, USA). In the primary analysis, the prevalence of tuberculosis infection was estimated from skin test reactions in children with no BCG scar using the ‘mirror method’. This method is based on the observation that the distribution of tuberculin reactions reflects two underlying distributions: that of reactions to tuberculosis infection and that of non-specific reactions [16,17]. Assuming that tuberculin reactions among infected children are normally distributed with the mean at 17 mm, and that no non-specific reactions are larger than 17 mm, the prevalence of tuberculosis infection can be inferred from the proportion of children with reaction sizes of 17 mm and larger . The prevalence of infection was, therefore, calculated as the number of children with reactions sized 17 mm, plus twice the number of children with reactions sized 18 mm or larger, divided by the total number of children tested and read [16,17]. In a concurrent survey using the same tuberculin among 540 non-HIV-infected, smear-positive patients with tuberculosis in six hospitals throughout the country, the non-zero reactions showed a normal distribution with a mean of 16.8 mm (National Tuberculosis and Leprosy Programme, unpublished data). The mirror method based on a distribution mean of 17 mm was thus considered to provide the best estimate of the infection prevalence.
The ARTI, defined as the proportion of the population that is infected or reinfected during a calendar year, was calculated as  1 − (1 − prevalence)1/mean age.
The analyses were restricted to children aged 6–14 years, based on age at last birthday plus 0.5 years. Over-all and region-specific prevalence and ARTI were calculated as the arithmetic mean over all districts with confidence intervals (CI) based on the corresponding standard errors. Results of this survey round were compared with results of the three previous rounds by reanalysing the original data files using identical methods and definitions for estimation of infection prevalence. Trend estimates were obtained by exponential least-squares fit of the ARTI estimates for each survey round. The association between trends in ARTI and HIV prevalence was assessed by comparing region-specific regression coefficients for the ARTI with region-specific prevalence of HIV infection among notified sputum smear-positive tuberculosis cases, obtained from surveys in the period 1994–1998 .
Sensitivity analyses compared various methods for establishing the prevalence of tuberculosis infection from the distribution of reaction sizes. These included the mirror method assuming that the means of the distribution among infected children was around 15 and 19 mm, and simple cut-offs for a positive tuberculin test at 10 and 15 mm. In addition, analyses were repeated for children with a BCG scar.
From February 2000 through December 2003, 723 schools were surveyed in 59 districts in all 20 regions of mainland Tanzania. In total, 130 201 children aged 6–14 years were enrolled (Table 1). This was 47% above the projected number of 88 500 children. Over-enrolment was most pronounced (172–256%) in the regions surveyed in the period mid-2001 to mid-2002 (i.e., after abolition of primary school fees). Of 130 201 enrolled children, 12 020 (9.2%) were absent at the time of testing, and an additional 20 877 (16.0%) at the time of test reading. There were 667 (0.5%) children excluded because of test errors, and 411 because the time between testing and reading had been more than 96 h or unknown (0.3%). Thus, 96 226 (73.9%) children were included in the analyses, 85 424 (88.8%) with and 10 239 (10.6%) without a BCG scar. For 563 children (0.6%), the presence of a BCG scar was unknown. The district-specific proportions of enrolled children that were included ranged from 53.7 to 93.1% (median, 75.3). Of 96 226 included children, 49 203 were boys (51.1%). Ages ranged from 6 to 14 years and were normally distributed (mean, 8.9; SD, 1.7).
A reaction size of 0 mm was observed in 50 070 children (52.0%), including 6038 (58.8%) without a BCG scar. The distributions of non-zero reactions among children with and without a BCG scar were similar except that a slightly larger proportion of children with a BCG scar had reaction sizes in the range 8–16 mm. Among children without a BCG scar, the proportion of reaction sizes of ≥ 17 mm was consistently lower than those observed in the three previous survey rounds, despite a higher proportion of reaction sizes of 2–12 mm (Fig. 1).
Among children without a BCG scar, the prevalence of infection was 6.1% (95% CI, 5.1–7.2), and the corresponding ARTI was 0.68% (95% CI, 0.55–0.81) (Table 1). The annual decline in ARTI was 3.6% between the first and the second survey rounds, 0.2% between the second and the third and 4.9% between the third and the fourth, for an average of 2.7% per year (P < 0.001). Among children with a BCG scar, the ARTI was 0.67% (95% CI, 0.60–0.74) in the fourth round. It declined strongly between the first and second survey and thereafter followed a similar trend as that observed among children without a BCG scar.
Over the period 1982–2003, the notification rate of all forms of tuberculosis increased from 62 to 180/100 000 (average annual increase 6.0%; Fig. 2). The notification rate of new sputum smear-positive tuberculosis increased from 30/100 000 in 1981 to a peak of 71/100 000 in 1998, and remained between 67 and 71/100 000 thereafter (average annual increase 4.4%).
Across the four survey rounds, the ARTI showed a declining trend in 17 of the 20 regions. Comparison by region of the regression coefficients for the ARTI with the prevalence of HIV infection among notified smear-positive tuberculosis cases over the period 1994–1998 showed no clear pattern (Fig. 3; P = 0.575). All eight regions with an estimated HIV prevalence among patients with tuberculosis of ≥ 40% showed a declining trend in ARTI. The median interval between the HIV prevalence survey and the fourth round of the tuberculin survey was 5 years (range, 3–8).
With the renewed sampling of schools for the fourth survey round, 606 schools selected in round 1 were replaced by 630 new schools, while 93 schools in 43 districts remained in the sample. In analyses restricted to these 93 schools, the ARTI declined from 0.93% in the third to 0.61% in the fourth survey round (P = 0.123). Among all children, the ARTI declined over this period from 1.01% to 0.65% (P = 0.003).
The ARTI declined over time irrespective of the method by which the prevalence of infection was established, except when all reactions of ≥ 10 mm were assumed to indicate infection (Fig. 4). In the fourth round, no significant differences in ARTI were observed between children with and without a BCG scar if the prevalence of infection was estimated by the mirror method around 15, 17 or 19 mm, but there was a significant but limited difference if prevalence was estimated as all reaction sizes ≥ 15 mm (0.70% versus 0.59%; P = 0.026).
In the fourth survey round, 5558 of 48 108 children (11.6%) were reported to have at least one deceased parent. This was the mother for 4348 (9.0%), the father for 2105 (4.4%) and both parents for 895 (1.9%). The probability of tuberculosis infection was increased for children where the mother [odds ratio (OR), 1.85], the father (OR, 2.12) or both parents (OR, 2.51) had died (all three comparisons, P < 0.001). The probability that an enrolled child was tested and the test read was also increased if the mother (OR, 1.35; P < 0.001), the father (OR, 1.16; P = 0.025) or both parents had died (OR, 1.47; P < 0.001).
In an analysis by district included in the survey, the proportion of children tested and read was not associated with the notification rate of smear-positive tuberculosis cases averaged over the preceding 10 years (data not shown).
Since the early 1980s, the notification rate of tuberculosis in mainland Tanzania has increased threefold, and that of new sputum smear-positive tuberculosis twofold. Despite this, we found no increase, and even a significant decrease, in ARTI among schoolchildren. This pattern was consistent across the 20 regions, of which only three showed an increase in ARTI.
Between the periods 1991–1993 and 1994–1998, the prevalence of HIV infection among smear-positive tuberculosis cases increased from 28 to 40% . A study done in 2000–2003 among smear-positive patients in six hospitals throughout the country showed an HIV prevalence of 46% (National Tuberculosis and Leprosy Programme, unpublished data). Of all smear-positive tuberculosis among HIV-infected patients, 86% was estimated to be directly attributable to HIV coinfection . The increase in notification of smear-positive tuberculosis has, therefore, been mainly attributable to an increase in notification of HIV-coinfected patients.
On the one hand, we found no association between the direction and rate of change in region-specific ARTI and the prevalence of HIV infection among patients with tuberculosis in the period 1994–1998 (i.e., the period over which the ARTI was estimated in the latest survey round). On the other hand, the observed trend of the ARTI was not only consistent with a constant proportional decline but also with a decline over the period 1985–1990, followed by a near-stable ARTI between 1990 and 1995, and a continuation of the earlier decline from 1995 onwards. These periods coincided with low, increasing and stabilized notification rates of HIV-associated smear-positive tuberculosis, respectively. This suggests that there was some impact of HIV-associated tuberculosis on transmission but that this was limited to the phase during which tuberculosis incidences increased, and occurred against a background of a declining ARTI.
Tuberculin surveys have been subject to criticism [18,19]. Methodological difficulties in their interpretation include the influence of infection with environmental mycobacteria, the effect of BCG vaccination, whether the unvaccinated children assessed are representative, misclassification of vaccination status, and various types of selection bias [17–19].
The observed distribution of reaction sizes did not allow us to distinguish whether reactions were due to tuberculosis infections or to infections with environmental mycrobacteria. In this situation, estimates of the ARTI are strongly affected by the method by which the prevalence of tuberculosis infection is established . However, estimates of the trend in ARTI over time are fairly robust to the method used . We observed the same declining trend with various methods for establishing the prevalence of infection. The only exception was when we applied a simple cut-off at 10 mm, a method with limited specificity for detecting tuberculosis infection with this tuberculin in settings with high prevalence of environmental infections .
In the fourth survey round, BCG vaccination had minimal effect on reaction sizes that are most relevant for determining the prevalence of tuberculosis infection (i.e., 15 mm and above). This is consistent with the observation from several studies that BCG vaccination given at birth, as happens in Tanzania, does not affect the proportion of tuberculin reactions ≥ 10 mm after the first year of life, and that these reactions may even be 0 mm . The effect of BCG vaccination on the distribution of reaction sizes seen in the earlier surveys could be related to vaccination campaigns among schoolchildren; these campaigns were abolished in the 1990s.
With increasing BCG coverage, the proportion of unvaccinated children was now only 11%. Our best estimate of the ARTI among unvaccinated children in this survey was identical to that among vaccinated children, and in most of the sensitivity analyses differences in ARTI between vaccinated and unvaccinated children were not significant or were small. This suggests that the unvaccinated children largely represented all children in our survey with regard to infection risk. A variable proportion of vaccinated children do not develop a typical BCG scar, and in several studies scar formation did not correlate with the tuberculin response . Therefore, there may have been vaccinated children among those without a BCG scar, and we may have overestimated the ARTI among them. This may have biased out trend estimates. However, since the proportion of children without a BCG scar declined over time, such bias would result in a rising rather than a declining trend in ARTI, thus reinforcing our findings. Misclassification of vaccination status, however, may have affected the proportion of children with intermediate size reactions (2–9 mm), in particular in round 4 in which only 11% had no BCG scar.
We found the same decline in ARTI when the analyses were restricted to schools that had been selected in both the third and the fourth round. This indicates that the reselection of schools for the fourth round neither introduced nor removed selection bias. Of all children who were enrolled in school, 25% were excluded from the analysis because they were absent at the time of testing and/or reading. The proportion of children excluded varied considerably by district. There was no association between this proportion and the average notification rate of smear-positive tuberculosis cases over the period during which these children were at risk, suggesting that this variation did not introduce substantial selection bias. We postulated that children whose parent(s) had died were more likely to be absent from school, while they also had higher risk of tuberculosis infection because of socioeconomic status or tuberculosis in the deceased parent. We found indeed a higher ARTI among children with deceased parents. However, children with one or two deceased parents had a higher rather than a lower probability of being included in the analysis, thus reinforcing our finding of a decreasing ARTI.
Finally, children who were enrolled in school may have been less at risk of M. tuberculosis infection that those who were not (e.g., reflecting differences in household income). This possibility could not be excluded. However, school enrolment increased substantially between the third and fourth survey. The net school enrolment ratio averaged 56% during the third survey round, and 74% during the fourth because of the abolition of school fees in 2001 . This increase is likely to have had most effect in school enrolment of the poorest children. Underestimation of the ARTI owing to non-enrolment of poor children, therefore, is expected to have become less during the fourth survey, the associated bias being an increase rather than the observed decrease in ARTI.
The lack of impact of the increase in HIV-associated tuberculosis on the ARTI has several possible explanations. HIV-infected patients with tuberculosis may transmit M. tuberculosis less easily. Household contact studies suggest that those with tuberculosis who are also HIV positive produce less secondary infections than those without HIV infection [24–29]. However, several other studies, including from Africa, found no such difference , and the observed differences may reflect differences in duration of infectiousness of the source case rather than differences in transmission potential . Perhaps more importantly, HIV-infected patients with tuberculosis may, on average, be infectious for a shorter period than non-HIV-infected patients. HIV-infected patients with tuberculosis have higher mortality . In addition, they may present for diagnosis and treatment earlier than non-HIV-infected patients because of more rapid disease progression . This would primarily apply when tuberculosis services are widely accessible and treatment is effective. The Tanzanian National Tuberculosis and Leprosy Programme is recognized as a well-organized and managed DOTS programme . The DOTS coverage is nationwide and tuberculosis services are increasingly integrated into the general health services. The estimate for the smear-positive case detection rate in 2003 was only 43%, but its reliability is not easily verified because the estimation method may not apply when the prevalence of HIV is high . Moreover, the Tanzanian National Tuberculosis and Leprosy Programme reported treatment success rates of around 80% since 1984 . In Kenya, where the trend in ARTI over the first half of the 1990s did correlate with the trend in HIV prevalence among patients with tuberculosis, the treatment regimens that were used until 1993–1995 had substantially lower cure rates . It may be that a well performing DOTS programme is a key factor in containing the increased transmission of tuberculosis owing to HIV coinfection. That this was not observed among South African gold miners despite active case finding and directly observed therapy could be because of conditions specific to that setting, such as crowding and silicosis, rather than to insufficient effectiveness of the DOTS approach .
The HIV epidemic has dramatic effects on tuberculosis morbidity and mortality in sub-Saharan Africa. Nevertheless, our findings suggest that in the presence of a strong control programme the overall population-level effect of the HIV epidemic on tuberculosis transmission is limited.
We acknowledge the Tanzania National Tuberculin Team (P. Mahimbo, J. Mnape, G. Mbura, M. Kiondo, M. Massay, A. Songa, C. Mushi, N. Mbogo, S. Sakalani, L. Ndamugoba and I. Kazema) for data collection; J. Goodluck and C. Chipaga for data entry; and E. Nkiligi for data management.
Sponsorship: The study was financed by the Ministry of Health of the United Republic of Tanzania. F. Cobelens, C. Verhage, N. Kalisvaart and M. Borgdorff received financial support from the Dutch Ministry of Foreign Affairs. The Madurodam Foundation (The Hague, the Netherlands) donated a vehicle.
1. Corbett EL, Watt CJ, Walker N, Maher D, Williams BG, Raviglione MC, et al
. The growing burden of tuberculosis: global trends and interactions with the HIV epidemic. Arch Intern Med 2003; 163:1009–1021.
2. Frieden TR, Sterling TR, Munsiff SS, Watt CJ, Dye C. Tuberculosis. Lancet 2003; 362:887–889.
3. Borgdorff MW, Corbett EL, DeCock KM, Trendston T, Hayes RJ, Chaisson RE, et al
. Stable incidence rates of tuberculosis (TB) among human immunodeficiency virus (HIV)-negative South African gold miners during a decade of epidemic HIV-associated TB. J Infect Dis 2003; 188:1156–1163.
4. De Cock KM, Chaisson RE. Will DOTS do it? A reappraisal of tuberculosis control in countries with high rates of HIV infection. Int J Tuberc Lung Dis 1999; 3:457–465.
5. Odhiambo JA, Borgdorff MW, Kiambih FM, Kibuga DK, Kwamanga DO, Ng‘ang’a L, et al
. Tuberculosis and the HIV epidemic: increasing annual risk of tuberculous infection in Kenya, 1986-1996. Am J Public Health 1999; 89:1078–1082.
6. Sonnenberg P, Glynn JR, Fielding K, Murray J, Godfrey-Faussett P, Shearer S. HIV and pulmonary tuberculosis: the impact goes beyond those infected with HIV. AIDS 2004; 18:657–662.
7. Richards SB, St Louis ME, Nieburg P, Coulibaly IM, Coulibaly D, Abouya L, et al
. Impact of the HIV epidemic on trends in tuberculosis in Abidjan, Cote d'Ivoire. Tuber Lung Dis 1995; 76:11–16.
8. Corbett EL, Charalambous S, Fielding K, Clayton T, Hayes RJ, Chaisson RE, et al
. Stable incidence rates of tuberculosis (TB) among human immunodeficiency virus (HIV)-negative South African gold miners during a decade of epidemic HIV-associated TB. J Infect Dis 2003; 188:1156–1163.
9. Glynn JR, Crampin AC, Ngwira BM, Mwaungulu FD, Mwafilirwa DT, Floyd S, et al
. Trends in tuberculosis and the influence of HIV infection in northern Malawi, 1988–2001. AIDS 2004; 18:1459–1463.
10. Chum HJ. Ten years of the National Tuberculosis/Leprosy Programme in Tanzania. Bull Int Union Tuberc Lung Dis 1989; 64:34–36.
11. Range N, Ipuge YA, O'Brien RJ, Egwaga SM, Mfinanga SG, Chonde TM, et al
. Trend in HIV prevalence among tuberculosis patients in Tanzania, 1991–1998. Int J Tuberc Lung Dis 2001; 5:405–412.
12. Tanzania Tuberculin Survey Collaboration. Tuberculosis control in the era of the HIV epidemic: risk of tuberculosis infection in Tanzania, 1983–1998.Int J Tuberc Lung Dis
13. UNAIDS/WHO Working Group on Global HIV/AIDS and STI Surveillance. Epidemiological fact sheets. United Republic of Tanzania. Update 2004
. Geneva: UNAIDS; 2004.
14. Styblo K. Epidemiology of tuberculosis. Selected papers
, Vol. 24. The Hague: KNCV; 1991.
15. Arnadottir T, Rieder HL, Trebucq A, Waaler HT. Guidelines for conducting tuberculin skin test surveys in high prevalence countries. Tuberc Lung Dis 1996; 77(Suppl 1):1–19.
16. Bleiker MA, Sutherland I, Styblo K, ten Dam HG, Misljenovic O. Guidelines for estimating the risks of tuberculous infection from tuberculin test results in a representative sample of children. Bull Int Union Tuberc Lung Dis 1989; 64:7–12.
17. Rieder HL. Methodological issues in the estimation of the tuberculosis problem from tuberculin surveys. Tuberc Lung Dis 1995; 76:114–121.
18. Fine PE, Bruce J, Ponnighaus JM, Nkhosa P, Harawa A, Vynnycky E. Tuberculin sensitivity: conversions and reversions in a rural African population. Int J Tuberc Lung Dis 1999; 3:962–975.
19. Enarson DA. Measuring tuberculosis: lessons from Afghanistan. Int J Tuberc Lung Dis 2004; 8:1041–1042.
20. Nagelkerke NJ, Borgdorff MW, Kim SJ. Logistic discrimination of mixtures of M. tuberculosis
and non-specific tuberculin reactions. Stat Med 2001; 20:1113–1124.
21. Berkel GM, Cobelens FGJ, De Vries G, Draayer-Jansen I, Borgdorff MW. Tuberculin skin test: estimation of positive and negative predictive values from routine data. Int J Tuberc Lung Dis 2005; 9:310–316.
22. Menzies D. What does tuberculin reactivity after bacille Calmette–Guerin vaccination tell us? Clin Infect Dis 2000; 31(Suppl 3):71–74.
23. Tanzania Ministry of Education and Culture. Basic Statistics in Education, 1999–2003. National Data
. Dar es Salaam: Ministry of Education and Culture; 2003.
24. Elliott AM, Hayes RJ, Halwiindi B, Luo N, Tembo G, Pobee JO, et al
. The impact of HIV on infectiousness of pulmonary tuberculosis: a community study in Zambia. AIDS 1993; 7:981–987.
25. Cauthen GM, Dooley SW, Onorato IM, Ihle WW, Burr JM, Bigler WJ, et al
. Transmission of Mycobacterium tuberculosis
from tuberculosis patients with HIV infection or AIDS. Am J Epidemiol 1996; 144:69–77.
26. Espinal MA, Perez EN, Baez J, Henriquez L, Fernandez K, Lopez M, et al
. Infectiousness of Mycobacterium tuberculosis
in HIV-1-infected patients with tuberculosis: a prospective study. Lancet 2000; 355:275–280.
27. Cruciani M, Malena M, Bosco O, Gatti G, Serpelloni G. The impact of human immunodeficiency virus type 1 on infectiousness of tuberculosis: a meta-analysis. Clin Infect Dis 2001; 33:1922–1930.
28. Carvalho AC, DeRiemer K, Nunes ZB, Martins M, Comelli M, Marinoni A, et al
. Transmission of Mycobacterium tuberculosis
to contacts of HIV-infected tuberculosis patients. Am J Respir Crit Care Med 2001; 164:2166–2171.
29. Mohammad Z, Naing NN, Salleh R, Ahmad N, Hamzah WM, Mahmud R. A preliminary study of the influence of HIV infection in the transmission of tuberculosis. SE Asian J Trop Med Public Health 2002; 33:92–98.
30. Pai M, McCulloch M, Colford JM Jr. Meta-analysis of the impact of HIV on the infectiousness of tuberculosis: methodological concerns. Clin Infect Dis 2002; 34:1285–1287.
31. Pablos-Mendez A, Sterling TR, Frieden TR. The relationship between delayed or incomplete treatment and all-cause mortality in patients with tuberculosis. JAMA 1996; 276:1223–1228.
32. Corbett EL, Charalambous S, Moloi VM, Fielding K, Grant AD, Dye C, et al
. Human immunodeficiency virus and the prevalence of undiagnosed tuberculosis in African gold miners. Am J Respir Crit Care Med 2004; 170:673–679.
33. World Health Organization Stop TB. Global Tuberculosis Control: Surveillance, Planning, Financing
. Geneva: World Health Organization; 2005.