Statistical/mathematical analysis after discussion with all authors (in particular BW and MW): NN. Conceived and designed the experiments: EWT MJvdW MB BW NN. Performed the experiments: EWT NN. Analyzed the data: EWT NN. Wrote the paper: NN EWT.
The authors have declared that no competing interests exist.
The prognosis, specifically the case fatality and duration, of untreated tuberculosis is important as many patients are not correctly diagnosed and therefore receive inadequate or no treatment. Furthermore, duration and case fatality of tuberculosis are key parameters in interpreting epidemiological data.
To estimate the duration and case fatality of untreated pulmonary tuberculosis in HIV negative patients we reviewed studies from the pre-chemotherapy era. Untreated smear-positive tuberculosis among HIV negative individuals has a 10-year case fatality variously reported between 53% and 86%, with a weighted mean of 70%. Ten-year case fatality of culture-positive smear-negative tuberculosis was nowhere reported directly but can be indirectly estimated to be approximately 20%. The duration of tuberculosis from onset to cure or death is approximately 3 years and appears to be similar for smear-positive and smear-negative tuberculosis.
Current models of untreated tuberculosis that assume a total duration of 2 years until self-cure or death underestimate the duration of disease by about one year, but their case fatality estimates of 70% for smear-positive and 20% for culture-positive smear-negative tuberculosis appear to be satisfactory.
Before the advent of chemotherapy, tuberculosis was one of the major causes of death
in both Western
The prognosis of untreated tuberculosis is difficult to study these days as leaving
patients untreated, especially in a study setting, is unethical. As an alternative,
one could consider, as an approximation, the prognosis of multi-drug resistant (MDR)
tuberculosis treated with first line drugs. However, MDR-tuberculosis patients may
benefit to some extent from first line therapy
To estimate the duration and case fatality of untreated pulmonary tuberculosis, we
reviewed studies from the pre-chemotherapy era. For tuberculosis in HIV infected
patients there are, of course, no data from the pre-chemotherapy era. Thus the only
data that are potentially relevant are those on MDR-tuberculosis HIV positive
patients treated with (inadequate) first line tuberculosis drugs, as their prognosis
would probably be similar to that of untreated HIV positive tuberculosis. However,
it is clearly important to distinguish patients by stage of HIV disease and by
treatment (ART, type of ART, or not
We studied the duration until death or self-cure of untreated tuberculosis and 5- and 10-year survival probabilities in representative adult populations (>15 yrs of age) with pulmonary tuberculosis, identifiable as either smear-positive or smear-negative.
Not a single study has measured the duration of disease directly, as this would
require an exhaustive ascertainment of incident cases as well as a follow-up to
either death, which is easy to establish, or cure, which is more difficult to
establish, while withholding treatment, at least for some time. One thus has to
rely on indirect information to estimate duration of disease, on the assumption
that duration of disease (D) and case fatality (CF) are related to incidence
(I), prevalence (P) and mortality (M): D = P/I and
CF = M/I
We defined four types of data sources which may contribute information on the natural duration and/or outcome of disease:
Follow-up (cohort) studies. Diagnosed patients are individually followed–up over time and their mortality and morbidity experience recorded. Inevitably there is some kind of selection (bias) involved in such studies as they exclude undiagnosed patients. Patients included may be those identified through the health system, or those who attended a specific institution (e.g. sanatorium), or patients may have been identified through a tuberculosis survey. These cohort studies provide key information on CF, but do not generally provide estimates of duration of disease, as the start of the tuberculosis episode is normally unknown and cure is usually not recorded.
Prevalence and incidence studies. A comparison between prevalent and incident cases would yield the duration immediately if the population is stable, i.e. no migration. However, if incidence is measured through repeated waves of surveys (instead of recorded continuously), one has to take into account the fact that incident cases occurring in-between surveys, but who recovered or died before the next survey wave, will be missed by the study. Although such studies are ideal for estimating the duration of disease they are less suitable for estimating the CF. In order to obtain an estimate of the CF one needs either follow-up of incident cases or estimates of the frequency with which disease ends in death among those patients for whom the end of disease is observed.
Notification and mortality studies. Studies that relate notification to mortality are also relevant. While such studies may provide little information on the duration of disease they do provide data on ultimate outcome (cure versus death) as CF = M/I although one cannot be certain that all incident cases are notified nor that all deaths occur among patients ever notified.
Prevalence and mortality studies. These compare the prevalence of tuberculosis to its (annual) mortality, but do not establish the fate of individual patients. To estimate the duration of disease, however, requires knowledge of the CF of (prevalent) tuberculosis cases, as well as an assumption of a stationary epidemiological situation. For then the ratio of the mortality rate and the CF estimates the incidence rate, and one can use the fact that the prevalence equals the product of the incidence and the duration (P = I*D) to obtain the duration. Conversely, estimating the CF would require knowledge of the duration of disease in addition to the prevalence and mortality rate, as the incidence would then equal the prevalence divided by the duration, and the ratio of the mortality and incidence rate would yield the CF.
We searched PubMed including OldMedline with publications from the early decades
of the 20th century up to 17 December 2010 and EMBASE, including
references from 1900 until 1966. The search strategy is summarized in
Database | PubMed |
Old Medline |
Embase |
Period included | 1-1-1954 – 17-12-2010 | Start – 31-12-1953 | Start – 1966 |
Mesh terms included | Tuberculosis, Prognosis, Mortality | Tuberculosis, Prognosis, Mortality | Tuberculosis, Prognosis, Mortality, Survival, Fatality |
Free text included (all fields) | Tuberculosis, Prognosis, Mortality, Survival, Fatality, Untreated | Tuberculosis, Prognosis, Mortality, Survival, Fatality | |
Free text included (title/abstract only) | Course | Course | Course |
Free text included (title only) | Course | Course | Tuberculosis, Prognosis, Mortality, Survival, Fatality |
Number of references retrieved | 196 | 591 | 1093 |
Number of references minus duplicates |
196 | 537 | 827 |
*‘tuberculosis’ (either as Mesh heading or as free text) and ‘untreated’ and one of the other terms (as Mesh term or as free text) were searched for.
**‘tuberculosis’ (either as Mesh heading or as free text) and one of the other terms (as Mesh term or as free text) were searched for.
‘tuberculosis’ either as subject heading or as free text in title and ‘course’ as free text in title or abstract or one of the other terms as subject heading or as free text in title.
Occuring as duplicate either within search, with searches in other electronic databases, or with snowball sample.
All references were first screened independently by two authors (ET and NN) on title and, if no title was available, in the snowball sampling method, on reference in the text to assess whether they potentially assessed the prognosis of untreated pulmonary tuberculosis in representative adult populations. Of potentially eligible papers, if available, abstracts were subsequently assessed for eligibility using the same strategy. If no abstract was available, papers were accessed in full text. Among the identified sources we selected those that would potentially provide estimates of CF and/or duration of pulmonary tuberculosis in adults (≥15 years) by any of the four methods outlined above. Studies were included provided: a) their methodology was sound (e.g. (near-to-)complete follow-up or making use of actuarial methods), considering populations that can be considered as more or less ‘population-based’ (thus not including only specific population subgroups or pre-selecting certain categories of patients), b) they contained original data (i.e., no editorials, opinion papers, minutes; reviews were only included if the literature these referred to was not found), c) we could decide whether patients included were smear-positive or smear-negative but culture-positive; in studies where patients were described as having “open” tuberculosis or “bacillary tuberculosis” before 1930 (when culture became available) we assumed that these patients were smear-positive, d) description of the available data was sufficient to enable calculation 5- and/or 10-year survival probabilities or disease duration, and e) the study population was not treated with chemotherapy or was treated with probably or proven ineffective therapy (e.g. collapse therapy, lung resection, short duration mono-drug therapy, etc.).
Eligibility and data extracted from all eligible sources were checked and discussed by two authors (NN and ET) using the criteria described above. The data sources were reviewed and summarized with respect to their information regarding the duration and outcome of untreated tuberculosis, and CF. Discrepancies between authors with respect to extracted data were resolved by discussing the differences and independently re-reviewing the data.
There are some important limitations to studying the duration and CF of untreated tuberculosis, since many of the included studies do not meet modern research standards. For example, the case definition, the onset of disease, or the beginning of follow-up in cohort studies (onset of symptoms, sputum positivity) are often ill-defined or poorly described in older publications, and many cases included in those studies would not meet modern diagnostic standards.
A large number of studies are based on passive case finding, which inevitably entails some selection bias, as diagnosed cases may well differ from undiagnosed ones. Some studies are limited to hospitalized (sanatoria) cases and therefore presumably exclude both the mildest and the most severe cases, as some of the latter probably died before they could have been hospitalized.
An additional methodological problem constitutes the way cases have been
classified in old studies. Using the distinction of pulmonary tuberculosis into
sputum smear-positive (smear-positive) and sputum smear-negative
(smear-negative) cases, the most common classification used today, we must
assume (highly unrealistically) that the sensitivity and specificity of direct
smear has not changed. Especially the diagnosis of smear-negative cases is
problematic as culture using the Löwenstein-Jensen (L-J) medium did not
become available until the 1930s
A further methodological pitfall is that by combining different estimates one
makes the implicit, and untested, assumption that the natural history of
tuberculosis does not differ significantly among countries and periods. However,
the risk of infection with
Data were extracted into Excel sheets and survival probabilities re-calculated
and provided with accompanying 95% Greenwood confidence intervals using
the original paper's life table's information. Where insufficient
details were available to recalculate survival probabilities, estimates as
calculated by the studies' authors were taken. Duration of active pulmonary
tuberculosis disease from diagnosis till death or cure could be assessed from
two studies with a very different study design
Because of the above-described methodological problems with combining the results of such diverse studies, we did not attempt to do a formal meta-analysis here.
Using the methods described above we identified a wide range of studies on the
prognosis of tuberculosis in the absence of chemotherapy (
Flowchart schematically showing inclusion and exclusion of papers. Those
marked with a * were excluded either because they were referred to
at places in the text that did not discuss duration of tuberculosis,
tuberculosis mortality, case fatality, life tables or natural history,
or because the title indicated that the paper was not about one of these
topics; ** for two of these, data were included to the extent
mentioned by Berg
The sources we considered relevant to the natural (pre-chemotherapy) history of
tuberculosis are listed in
Study | Design | Country | Type of Subjects | Period patients identified | N |
Hartley |
Cohort | UK | Cases treated at Brompton Hospital with open tuberculosis | 1905–1914 | 3,326 |
Sinding-Larsen |
Cohort | Denmark | Sanatorium patients with open tuberculosis | 1907–1931 | 1,114 |
Trail and Stockman |
Cohort | UK | Sanatorium patients with bacillary and abacillary tuberculosis | 1911–1928 | 2,625 |
Backer |
Cohort | Norway | Dispensary material of patients with bacillary and abacillary tuberculosis | 1911–1930 | 2,312 |
Krebs |
Cohort | Switzerland | Sanatorium patients with open and closed tuberculosis | 1912–1927 | 1,787 |
Tattersall |
Cohort | UK | Dispensary material from smear-positive patients | 1914–1940 | 1,192 |
Magnusson |
Cohort | Iceland | Sanatorium patients with open and closed tuberculosis | 1916–1935 | 792 examined, 379 with open and 413 with closed tuberculosis |
Rutledge and Crouch |
Cohort | USA | Discharged sanatorium patients with bacillary and abacillary tuberculosis | Not stated, prior to 1919 | 1,654 |
Münchbach |
Cohort | Germany | Sanatorium patients, with open tuberculosis | 1920–1927 | 3,966 |
Braeuning and Neisen |
Cohort | Poland (then Germany) | Dispensary material of bacillary/open tuberculosis patients | 1920–1921, 1927 | 951 |
Griep |
Retrospective cohort | The Netherlands | Notified cases with open tuberculosis | 1920–1938 | 1,846 |
Baart de la Faille |
Cohort | The Netherlands | Sanatorium patients, with open and closed tuberculosis | 1922–1935 | 3,615 (1,131 smear-positive at least once; 534 smear-positive at discharge) |
Buhl and Nyboe |
Cohort | Denmark | Notified cases with bacillary tuberculosis | (here) 1925–1929 | 314 |
Lindhardt |
Cohort | Denmark | Notified cases | 1925–1934 | 5,432 smear-positive cases |
Berg |
Cohort(s) | Sweden | All diagnosed open tuberculosis patients | 1928-1934 |
2,042 |
Thompson |
Cohort | UK | All diagnosed smear-positive patients | 1928–1938 | 406 |
National Tuberculosis Institute (NTI), Bangalore
|
Successive waves of surveys, prevalence and incidence | India | Active case-finding, smear-positive and/or culture-positive tuberculosis | 1961–1968 | 166,140 examined, 627 with tuberculosis |
Pamra |
Successive waves of surveys, prevalence and incidence | India | Active case-finding, smear-positive and/or culture-positive tuberculosis | 1962–1970 | 21,344–24,808 |
Drolet |
Notification and mortality | USA and UK | Notified cases with pulmonary tuberculosis (not further specified) | 1915–1935 | 299,244 (parts of USA), 323,870 (UK) |
Braeuning |
Notification and mortality | Poland (then Germany) | Notified cases with open pulmonary tuberculosis and deaths from tuberculosis | 1925–1929 | 264,500 (annual average) |
Framingham Com-munity Health & Tuberculosis
Demon-stration |
Community study; prevalence and mortality | USA | Community active and passive case finding of tuberculosis (not specified) | 1916–1925 | Not precisely given |
*Abbreviations used in this table: UK, United Kingdom; USA, United States of America; culture-positive, Löwenstein-Jensen medium culture-positive.
**as reported by Berg
only the years of which least biased data (according to Berg's
Smear-negative tuberculosis was defined as growth of mycobacteria on Malachite-green culture whereas no bacilli were identified in the patient's sputum.
Data re-analyzed by Fürth
Depending on survey wave (first survey had 21,344 participants, fourth and last had 24,808 participants).
1. Berg's study
2. Sinding-Larsen
3. Backer
4. Krebs
5. Tattersall
6. Magnusson
7. Rutledge and Crouch
8. Münchbach
9. Braeuning and Neisen
10. Griep
11. Baart de la Faille
12. Buhl and Nyboe
13. Lindhardt
14. Thompson
The study of the National Tuberculosis Institute, Bangalore, India (NTI)
Unfortunately, the reporting of both studies leaves much to be desired. For
example, prognosis (death or cure) is not presented broken down by Z-N
status (i.e. for smear-positive and smear-negative separately). Moreover,
Pamra and colleagues did not give any information about treatment of
tuberculosis
Drolet
The Framingham Community Health and Demonstration project
Direct estimates are available from cohort studies.
Study | Number of participants under observation | 5-year survival (95% CI) | 10-year survival (95% CI) |
|
|||
Hartley |
3326 | 58% (56%–60%) | - |
Sinding-Larsen |
1114 | 57% (54%–60%) | 47% (44%–50%) |
Trail & Stockman |
2615 | 50% (48%–52%) | 34% (32%–36%) |
Backer |
2312 | 35% (33%–37%) | 21% (19%–23%) |
Fürth |
996 | 30% (27%–33%) | 19% (17%–22%) |
Tattersall |
1082 | Not reported | 23% (21%–26%) |
Magnusson |
379 | 37% (33%–43%) |
27% (23%–32%) |
Rutledge & Crouch |
511 | 39% (35%–43%) | - |
Münchbach |
3966 | 50% (48%–52%) | - |
Braeuning & Neisen |
607 | 25% (22%–29%) | 18% (15%–21%) |
Griep; smear-positive |
975 | 51% (48%–54%) | 34% (31%–37%) |
Baart de la Faille |
534 | 38% (34%–42%) | 29% (25%–33%) |
Buhl & Nyboe |
314 | 45% (39%–51%) | 34% (29%–40%) |
Lindhardt; only smear-positive |
11,797 | 43% (42%–44%) | - |
Berg |
2042 | 42% (40%–44%) | 29% (27%–31%) |
Thompson; only smear-positive |
406 | 27% (23%–32%) | 14% (11%–18%) |
|
|||
Fürth |
469 | 88% (85%–91%) | 78% (74%–82%) |
Magnusson |
413 | 92% (89%–94%) |
85% (81%–88%) |
Rutledge & Crouch |
185 | 86% (80%–91%) | - |
Baart de la Faille |
597 | 85% (82%–88%) | 75% (71%–78%) |
Baart de la Faille |
2484 | 90% (89%–91%) | - |
*As reported by Berg
In this re-analysis, 1464 of the total of 1787 tuberculosis patients were included, for part of whom 5- and 10-year survival rates could be calculated;
Based on 975 cases diagnosed between 1920 and 1930;
These are 534 patients who were smear-positive at the time of discharge from sanatorium and also originally diagnosed as smear-positive;
We only used the period (notified cases between 1928 and 1934) for which the author considered his material to be least biased;
These 597 patients were once smear-positive but had become smear-negative at the time of discharge from sanatorium;
4- instead of 5-year survival;
These are 2484 patients who were consistently smear-negative but it is unclear how many were culture-positive.
In studies that reported on this (particularly Berg
Nevertheless, although mortality rates decline, long-term survivorship (of 10
years or more) is much poorer (a 10-year CF of 70% or more) than
5-year survival showing that tuberculosis can be a very long-lasting,
chronic disease. Taking the crude unweighted average of all studies one
arrives at a 5-year case fatality of 58% and a 10-year case fatality
of 73% for open (smear-positive) tuberculosis. Taking an average
weighted by sample size these numbers are 55% and 72%
respectively. Of course, these mortality data are somewhat distorted by
mortality from other causes, as most studies do not record cause of death,
and all-cause mortality rates may have been somewhat higher in the
pre-antibiotic era than they are now. On the basis of the above data,
especially the studies by Berg
A single, aggregate, CF for all smear positive patients is only justified if
in most studies the differences in mortality between the sexes and age
groups are rather small. This appears to be the case for sex, but higher
ages appear to have somewhat poorer prognosis. For example, in Berg's
study (providing the most detailed data), age- and sex specific 10-year
mortality rates were 66% for men aged 15–29 years, 70%
for men aged between 30 and 49 years, and 94% for men of 50 and
older. For women, these rates were 70%, 69%, and 92%
respectively
Braeuning
Drolet
The CF of pulmonary tuberculosis, smear-positive and/or culture-positive, can
also be estimated from the NTI study
There is a better approach to estimating the CF from the NTI data. Ultimately, all tuberculosis patients will either die or get cured. If the ratio of the mortality rate to the cure rate is independent of disease duration, then one can simply look at the ratio of the number of deaths to number of patients cured over a fixed period of follow-up. This assumption seems to be supported by their data (their Fig. 2), as the cured-to-death ratio among the cohort of tuberculosis patients discovered at survey 1 seems to remain about equal at 27.8/30.2, to 23.6/20.0 to 17.2/15.0 in the intervals between survey 1 and 2, between survey 2 and 3, and between survey 3 and 4 respectively. Thus, the prognosis (outcome) of the participants' disease (death or cure) did not seem to depend on the time they had already suffered from tuberculosis. Nevertheless, an exception may have to be made for incident tuberculosis patients who appear to fare somewhat better than prevalent cases, with a cured-to-death ratio of 44/24.
The study reports a total of 428 (often overlapping) individuals alive with tuberculosis at the beginning of any of the 1.5-year intervals. During the subsequent 1.5 years, a total of 89 died and 132 were cured, suggesting an (ultimate) tuberculosis mortality of 40.2%. However, this is not entirely correct as the paper reports 7% mortality among “cured cases” (most presumably dying from tuberculosis) and a “relapse rate” of 10%. We thus subtract 17% from 132 giving 109.5 and add 7% of 132 to 89 giving 98. Thus if the fate of prevalent cases would equal that of incident cases, 47% would ultimately die.
The assumption, as stated above, that there is a constant death-to-cure ratio
may not be entirely true, as among follow-up
Unfortunately, it is not possible to estimate the CF for smear-positive and
smear-negative tuberculosis separately from the data provided. If we accept
an ultimate mortality of smear-positive tuberculosis of 70% (based on
the studies presented elsewhere in this paper) then assuming that 50%
of cases are smear-positive (of all
Category | CFR (%) | Region to which CFR is applied |
|
||
smear-positive untreated | 70% | Global |
smear-negative untreated | 20% | Global |
|
||
smear-positive untreated | 83% | Global |
smear-negative untreated | 74% | Global |
*WHO: World Health Organization; CFR: case fatality rate.
The duration of disease is the time from onset of disease till cure or death. For
tuberculosis, it is not possible to measure exactly when it started, as patients
may remain asymptomatic or have very mild symptoms shortly after getting the
disease. Moreover, of the two possible end points, cure is hard to measure, as
relapses are common
Duration of disease can be estimated indirectly from the ratio of prevalence
to mortality. The Framingham Community Health and Tuberculosis Demonstration
The duration of disease in the pre-chemotherapy era was only studied
prospectively in one other study,
As waves of surveys in the NTI study were 1.5 years apart (even 2 years for
the interval between wave 3 and 4)
There is almost no reliable information regarding the relative duration of
smear-positive and smear-negative tuberculosis disease. A study from South
India
In our study we combined available information on untreated tuberculosis to estimate its case fatality and duration of disease. We found only few studies from the pre-chemotherapy era that allow for estimation of CFs and duration of disease of smear-positive tuberculosis. Given the limited information available and assuming that a 10-year CF will closely approximate lifetime CF, we conclude that (lifetime) CF in untreated smear-positive tuberculosis among HIV negative individuals is approximately 70% and about the same for both sexes. Mortality seems to be approximately independent of age until the age of 50 years after which it increases, perhaps due to concomitant complicating diseases such as diabetes or cancer and a greater mortality from other causes. However, this age effect would only be important in (patient) populations with a dramatically different age structure than the ones used in this review. For most high burden countries this is not the case.
For culture-positive smear-negative tuberculosis, lifetime CF is probably slightly over 20%, although this could only be estimated indirectly and with uncertain precision.
The duration of tuberculosis from onset to cure or death is approximately 3 years and appears to be grossly similar for smear-positive and smear-negative tuberculosis.
Because of the expected heterogeneity between studies with respect to study design and population, study period, duration and intensity of follow-up, definition of pulmonary tuberculosis (‘open’/‘closed’, bacillary/abacillary, smear-positive/smear-negative), etc., we did not do a formal meta-analysis. Additional heterogeneity among studies may also exist in patient selection and diagnostic procedures, for example the number of sputum samples analyzed and how these were obtained (e.g. induced or spontaneous). However, these data were hardly ever reported in the included studies.
Despite the fact that (HIV negative) tuberculosis has for centuries been a major cause of mortality, the number of studies on its natural history is surprisingly low.
This contrasts sharply with, for example, HIV for which detailed information on its natural history became available within decades of the discovery of the virus. Long term follow-up studies of HIV patients in carefully monitored cohorts have generated this information. In contrast, follow-up of most tuberculosis patients is nowadays usually limited to the duration of their treatment.
Another limitation is our serious lack of knowledge on the prognosis of extra-pulmonary and smear-negative pulmonary tuberculosis as most data on the natural history are available for patients who tested sputum smear-positive. No reliable prospective data on smear-negative culture-positive pulmonary patients are available and their long term survival can only be estimated indirectly and thus with great uncertainty. These patients form currently the group most likely to receive no or inadequate treatment, and may well account for large proportion of tuberculosis deaths. The prognosis of untreated extra-pulmonary patients - a very heterogeneous group that also includes most tuberculosis in children - is even more uncertain, and insufficient data were identified to include it in our review.
An important limitation of using electronic databases going back in time is that
these do not include abstracts and searches therefore may miss potentially
eligible papers. We have tried to obviate this by including quite general search
terms (see
We therefore supplemented our search strategy with snowball sampling. A
limitation of this approach is that it depends, perhaps heavily so, on its
starting point. We choose dr. Rieder's book
Quite some of the identified potentially eligible papers were not available to us. In theory, this may have influenced the outcome of our review. However, we were able to identify papers appearing in a variety of journals, text books and published as reports (‘grey literature’) and did not find any evidence for a correlation between the type of source and the quality of the data. Therefore, we expect no important ‘availability bias’ correlated with prognosis of untreated tuberculosis.
Another limitation of our review is that most of the included studies on CF were on predominantly Caucasian populations whereas most untreated patients currently are of different ethnicity. This is probably mainly due to the fact that evaluating the natural history of tuberculosis requires long term follow-up which has proven to be difficult, especially in resource constrained settings.
A key limitation is that we had to restrict our review to HIV-negative patients,
as explained in the
As regards the duration of disease, findings from these studies
While pre-chemotherapy data appeared to be a useful source of data for the prognosis of untreated tuberculosis, inevitably questions remain. Particularly, the impact of risk factors other than (variably defined) smear status was hard to explore systematically. Perhaps, long-term follow-up of patients with inadequately treated MDR or XDR tuberculosis may fill some of the gaps in our knowledge. Such follow-up may also fill other gaps in our knowledge such as the frequency of transitions between smear-positive and smear-negative tuberculosis and the prognosis and duration of HIV-positive tuberculosis.
We thank dr. Ana Bierrenbach, dr. Philippe Glaziou and dr. Ikushi Onozaki from the World Health Organization, and dr. Masja Straetemans from KNCV Tuberculosis Foundation as well as participants to workshops on the revision of the tuberculosis estimates for their useful comments on our data analysis and on the manuscript.