Preface
Tuberculosis (TB) is a leading cause of disease and death, with 
2 billion people infected and 2 million deaths annually1. Sputum smear microscopy (SSM) has remained the cornerstone of TB diagnosis for more than a century and is a pillar of the global strategy to control the disease, although it has significant limitations. As the epidemic continues, more attention is being paid to the impact that improving existing diagnostic methods and introducing new procedures might have in resource-limited settings. We estimated the potential global impact of better diagnostic tests, to provide guidance for health-care workers, test developers, funding agencies and policymakers.
Introduction
TB is the seventh leading cause of death worldwide2, second only to human immunodeficiency virus (HIV)/acquired immunodeficiency syndrome (AIDS) among infectious diseases. In 2004, an estimated 8.9 million people developed TB and 1.7 million died. The disease is concentrated in the developing world, and 80% of all cases occur in the 22 highest-burden countries3. Although the incidence of TB has decreased in many countries over the past decade, case numbers continue to rise in much of sub-Saharan Africa, where HIV is endemic. Eastern Europe has also seen an increased burden of TB, which is associated with poor treatment outcomes due to multidrug resistance (MDR)3.
Individuals with a sputum smear that is positive for TB (ss+) are responsible for the bulk of Mycobacterium tuberculosis transmission to others, 90% of whom remain asymptomatic with quiescent or latent TB infection (LTBI). The risk of reactivation or progression from LTBI to active disease is greatest within the first 2 years after infection and subsequently declines over time but does not disappear4. Hence, there remains a large reservoir of people with LTBI. The risk of TB reactivation and rapid disease progression is greater in individuals whose immune systems have been suppressed through malnutrition, HIV co-infection or other chronic diseases (such as diabetes).
In the early 1990s, the World Health Organization (WHO) introduced the DOTS strategy as a cost-effective way to control TB and improve health5. DOTS has five essential components: SSM, directly observed treatment with standardized short-course chemotherapy (SCC), a system to deliver drugs without interruption and free of charge, standardized recording and reporting of cases, and national political commitment.
The introduction and broad adoption of the DOTS strategy in 183 countries has substantially improved TB control. The WHO set worldwide targets to diagnose 70% of new (incident) ss+ cases, to cure 85% of these cases by 2005 and to maintain or improve on this performance from 2006 onwards3. It was projected that if these goals were met, TB incidence would decline by 5–10% annually in countries without endemic HIV. However, meeting these targets, especially for case detection, has proven difficult, with only 53% of incident ss+ cases having been detected in 2004 (ref. 3). More-over, even if these targets were met, a substantial burden of disease would remain and other interventions to accelerate the decline would continue to be urgently needed5, 6, 7.
Current diagnostic tests
The DOTS strategy focuses on the passive case-finding of infectious pulmonary TB, which typically occurs when the patient presents with persistent cough4. Diagnosis is anchored in the SSM of three specially stained sputum samples4. SSM targets the most infectious cases (ss+ TB) and is highly specific in most high-prevalence settings. However, the technique has several limitations. Test sensitivity is limited by the many patients with extrapulmonary TB or ss- disease who have <10,000 bacilli per ml of sputum and are therefore missed by SSM. Moreover, the test requires the collection and laborious examination of many samples, and because of consequent delays (ranging from 2 to 7 days) many people do not return for results8. Finally, in many locations in the developing world, peripheral health facilities lack trained microscopists, properly functioning microscopes and adequate reagents9, 10, 11.
So far, other diagnostic methods have offered limited benefits in the developing world, due to the constraints on their effectiveness and the infrastructure required. Culture is the most sensitive method for detecting TB, but it can take several weeks to yield results and demands advanced technical infrastructure that is not widely accessible in many countries12. Radiographic examination can detect some ss- cases and, together with a trial of antibiotics, is the recommended procedure for such diag-nosis13. The radiographic appearance of TB is not uniform, however, and image interpretation is subject to observer error, which limits the sensitivity and, more importantly, the specificity of X-rays in the field4. A course of broad-spectrum antibiotics, which are ineffective against TB, can be administered to help determine whether a patient has some other respiratory condition, such as chronic bronchitis. Interpretation of results can be difficult, however, because the symptoms of some patients without TB will not improve14, and a substantial fraction of patients with TB might also have symptomatic relief from coughing that is caused in part by some other disease15, 16, 17.
As part of the Global Health Diagnostics Forum convened by the Bill & Melinda Gates Foundation, the TB working group evaluated the potential role of better diagnostics in improving disease control in developing countries. To this end, we developed a decision-tree model to simulate the introduction of several theoretical tests for pulmonary TB to be used in patients with chronic persistent cough4. Although public-health efforts so far have generally focused on the most infectious manifestation of the disease (ss+ TB), we also examined the potential impact of a new test capable of identifying ss- TB that would detect all cases, not just those that contribute to transmission18. We also examined the effects of variable test specificity, the introduction of a rapid test (that would not be subject to the same loss of patients to follow-up), and variable patient access as determined by the test requirements and available health-system infrastructure.
We considered modelling the potential impact of tests for other TB diagnostic indications, including the detection of incipient disease in HIV negative (HIV-) and HIV+ adults with latent TB (which could make preventive treatment cost-effective), a test of cure in treated patients, detection of extrapulmonary and paediatric TB, and a screening test of high sensitivity and lower specificity to exclude the disease in suspected cases. However, detailed analysis of these alternatives was not carried out for a variety of reasons including infeasibility or low likelihood of test development for a predictor of incipient disease in HIV- adults, low predicted cost savings for a test of cure, and low predicted epidemiological/public health impact for a screening test, and for detection of extrapulmonary and paediatric TB. We also concluded that a diagnostic of incipient disease in HIV+ patients is not a priority, as they could be treated with one course of isoniazid. Preliminary results on a test for suspected MDR TB and a discussion of the other tests have been reported elsewhere19.
Methods
Analytic overview
We developed a decision-tree model to simulate the introduction of new tests to detect symptomatic adult pulmonary TB in patients presenting to health clinics with persistent cough lasting 
3 weeks. We refer to these patients as persistent coughers. The model calculates the impact of hypothetical improved diagnostic tests on treatment delivery as measured by reductions in TB-related mortality and in overtreatment of non-TB cases, compared with estimated treatment delivery through the status quo diagnostic tests. The primary health outcome in our model is adjusted lives saved. This composite measure incorporates both lives saved due to the number of additional treated cases of TB and lives saved indirectly through reductions in overtreatment20.
Photo by Sharon Farmer courtesy of the Bill & Melinda Gates Foundation
The difference between the status quo outcomes and those after the introduction of a new test is the theoretical benefit of the diagnostic. Estimated numbers of annual incremental true-positive ss+ and ss- patients and true negatives detected were translated into incremental annual TB deaths averted using the WHO estimates of case-fatality rates.
The outcomes are a function of the test characteristics of the new diagnostic (that is, sensitivity to ss+ and ss- TB, and specificity) and access to the results, which depends on the level of infrastructure required to use the diagnostic and the time required to obtain results. The changes in outcomes as we vary these parameters show which test characteristics are crucial. Because HIV infection can significantly affect the disease course and clinical manifestations of TB, we evaluated persistent coughers with HIV infection separately, but using an identical tree structure.
Access to diagnostics depends on the laboratory infrastructure required to perform them. The three categories used here are advanced infrastructure (as found in urban hospitals), the infrastructure of facilities currently performing SSM (as found in current TB clinics and hospitals) and no infrastructure (meaning that all patients have access)20.
Modelling detection of pulmonary TB in HIV+ and HIV- adults
The decision trees depict current and hypothetical new diagnostic strategies for pulmonary TB in HIV+ and HIV- adults aged 15–49 years with respiratory symptoms, in the four WHO regions with the highest number of TB deaths annually: Africa, eastern Mediterranean, southeast Asia and western Pacific. Europe and the Americas, which together account for <7% of the annual deaths from TB worldwide3 and have comparatively well- developed diagnostic infrastructures, were excluded from the model. HIV+ cohorts were modelled in two regions, Africa and southeast Asia, which together account for 93% of global HIV+ TB mortality. Results for HIV- persistent coughers in the four regions, and HIV+ coughers in the two regions, were calculated separately and then aggregated.
Figure 1a presents the status quo tree for adult pulmonary TB. In the model, persistent coughers in the status quo can proceed according to one of the three main diagnostic pathways: SSM, a clinical/radiologic algorithm (that is, an antibiotic trial, X-ray or both), or no testing or treatment. In the status quo tree, persistent coughers with positive smears who return for test results are treated with DOTS SCC. Patients with TB are considered ss+ in the model if a series of smears in a high-quality laboratory would have detected them. Each persistent cougher is assumed to receive at most one full set of diagnostic tests. Persistent coughers not returning for diagnostic test results are considered untreated with respect to the mortality consequences. Those with ss- results undergo follow-up testing using X-rays, a trial of broad-spectrum antibiotics for presumed bacterial infection or both. If clinically indicated, they are then given DOTS SCC. Persistent coughers evaluated exclusively with a clinical/radiologic algorithm follow the same diagnostic pathway as those with an ss- result, but are assumed to continue with non-DOTS care and to receive TB treatment that is slightly less effective than DOTS SCC.
Figure 1: Tree for adults presenting with persistent cough.
(a) Status quo tree for adult pulmonary TB. (b) Diagnostic pathway with a hypothetical new test.; AFB, acid-fast bacillus; FP, false positive; ss-, sputum smear negative; ss+, sputum smear positive; TB, tuberculosis; TP+, true positive ss+, TP- true positive ss-, FN; false negative.
High resolution image and legend (102K)Figure 1b represents the diagnostic pathway with a hypothetical new test. The fraction of patients that can be evaluated depends on the infrastructure requirements of the test, which in turn determine access20. Persistent coughers with or without access to the new test are distributed in the three status quo groups. For any given hypothetical percentage access to the new test, we assume that it starts with the patients currently tested with smears, spreads to those receiving non-sputum-based diagnostics and finally reaches those in locations with no current diagnosis. This hierarchical access assumption is discussed in more detail elsewhere20.
To isolate the effect of diagnostic improvements, we assume that DOTS SCC is not expanded to locations more peripheral than the microscopy centre, so those who do not start with SSM in the status quo still receive less-effective TB treatment in the hypothetical model. In each group, persistent coughers are given the new test only if it leads to lower mortality than the status quo test. The hypothetical new tests are described in terms of five parameters: first, the percentage of persistent coughers undergoing testing that is lost to follow-up before results are communicated, which is assumed to be zero for a rapid test; second, the sensitivity to ss+ TB; third, the sensitivity to pauci-bacillary ss- TB (<10,000 bacilli per ml of sputum); fourth, the specificity; and fifth, the proportion of persistent coughers in the region who could have access to the test, as determined by the health-system infrastructure.
In both trees, we assume there is no sorting into test branches by symptom severity, so the probability of being ss+ given a persistent cough is the same on all test branches. We consider deaths only from TB or TB treatment: the death rate for true negatives is therefore zero, and for HIV+ individuals the case-fatality rate represents the incremental deaths due to TB.
Model parameters
The model parameters for two of the regions are given in Table 1 and discussed briefly below.
![Table 1 - Tuberculosis model base case values and ranges
|[ast]|](/common/images/table_thumb.gif)
The population size and number of incident ss+ and ss- adult TB cases for each region are derived from the WHO estimates21. The proportion of persistent coughers who are initially tested with SSM is calculated by increasing the WHO estimates of the proportion of ss+ TB cases treated with DOTS SCC by the estimated percentage of ss+ cases lost to follow-up8, 22, 23 or false-negative diagnosis9, 11 despite starting with SSM. The proportion of persistent coughers who receive no test is calculated from the WHO estimates of the percentage of TB cases that are untreated by subtracting the estimated false negatives. The remaining persistent coughers receive a non-sputum test.
The number of patients presenting with persistent cough is estimated as the number of diagnostic sputum-smear evaluations divided by the proportion of incident ss+ cases tested with SSM21. Estimates of the number of persistent coughers examined by region are derived from a survey of TB laboratory-testing facilities12. The proportion of TB cases among those presenting with persistent cough is then calculated by simple division, and is similar to the results from surveys of patients24. This proportion is inversely related to incidence, because persistent cough is more evenly distributed around the world than is TB.
The recommended SSM procedure requires three samples, and loss to follow-up includes those who fail to return to provide all of the samples or to obtain the results8. The proportions for this and the other status quo diagnostic parameters are taken from the literature (Table 1).
The case-fatality outcomes of treatment depend on whether individuals have HIV, and on the nine combinations of TB status (ss+ TB, ss- TB or no TB) and treatment (DOTS SCC, non-DOTS TB treatment or no treatment)21.
Limits on the harm of unnecessary treatment were inferred from the fact that treatment based on diagnosis by SSM is recommended over treating all persistent coughers, by the opportunity costs of unnecessary treatment when resources are limited and by a survey of experts19, 20.
Results
To validate the model, we compared our status quo mortality and treatment results with the related WHO estimates. Our calculated numbers of deaths of adults with pulmonary TB by region, assuming the status quo, varied from 70% of the total WHO-estimated deaths (which include children and extrapulmonary TB) in the eastern Mediterranean region to 81% in Africa. The calculated status quo detection rates were 66% for adult ss+ cases (with 51% starting with SSM and the other 15% being diagnosed by other methods) and 45% for adult pulmonary ss- cases. These rates make the modelled status quo slightly better than the current WHO detection estimates, which reduces the estimated gains of new tests.
Table 2 shows the results of improved diagnostic testing for adults presenting with persistent cough. The tests shown in Table 2 are split into three sections according to the health-system infrastructure/access categories. The lowest section of Table 2 represents decreased infrastructure requirements and, hence, increased accessibility. Access to TB clinics varies from 44% in the eastern Mediterranean region to 66% in the western Pacific region. Within each section of Table 2, the rows represent hypothetical tests characterized by four parameters: sensitivity to detect ss+ cases, sensitivity to detect ss- cases, specificity and percentage loss to follow-up (assumed to be zero with rapid tests). To set the upper limits on the impact of improvements in each parameter, we might report gains from extreme hypothetical test characteristics, such as 100% sensitivity, 100% specificity or 100% access, even though such performance levels can only be approached and not achieved.
The last two numerical columns in Table 2 represent outcomes relative to the status quo (Table 2, test 2), which is SSM offered at TB clinics and hospitals, with our estimated parameters for current SSM. For example, test 3 in Table 2, which is similar to the current SSM except that sensitivity for ss+ is improved to 100%, would save 46,000 additional adjusted lives annually over the status quo. This gain includes only current cases, and is adjusted by a small penalty for the harm of treatment due to overtreatment of false positives. These 46,000 averted deaths, divided by the annual worldwide TB deaths of 1.75 million in 2003, represent 2.6% of the global TB deaths.
The maximum annual gain from better diagnosis and treatment of patients presenting with persistent cough is 625,000 adjusted lives saved or 36% of the global TB deaths (Table 2, test 14). The lives saved correspond to treating 2.02 million additional cases with TB, and avoiding the current 2.7 million TB treatments of persistent coughers who do not have TB (data not shown). The net 680,000 fewer treatments than in the status quo are valued as 34,000 additional lives saved, and this small adjustment for the harm of treatment means that comparisons of adjusted lives saved with annual unadjusted mortality from disease are not exact. About 85% of these gains are among HIV- patients, with most of the HIV+ gains being seen among patients in Africa.
Increased access and introduction of rapid tests lead to health gains
Gains from a hypothetical new test increase roughly in proportion with access to testing. For example, a rapid, 100% sensitive and 100% specific test (Table 2, test 1) would save 105,000 lives annually if available only to those with access to advanced infrastructure, but would save 359,000 lives if available at existing TB clinics (Table 2, test 8) and 625,000 if accessible to everyone (Table 2, test 14). Rapid tests increase access to results by avoiding the 10–15% loss to follow-up during the diagnostic process. Therefore, the introduction of a rapid test that is otherwise similar to SSM would save 95,000 lives annually if available at TB clinics (Table 2, test 4) and 263,000 - 133,000 = 130,000 more lives than SSM if accessible everywhere (Table 2, tests 9 and 10). Reduction in loss to follow-up with existing tests similar to SSM could also be achieved by on-the-spot reporting of smear results or by implementing a tracking system for defaulters. Therefore, operational modifications could save up to 95,000 lives annually.
High sensitivity and specificity are important
We varied test-performance characteristics (sensitivity in diagnosing ss+ and ss- cases, and specificity) in relation to access and loss-to-follow-up parameters, to determine the potential health gains from a new diagnostic. Improving the sensitivity of SSM in existing locations would provide only modest gains: 46,000 lives could be saved if field performance could be brought up to that of top-quality laboratories (either through training or development of an easier-to-use method; Table 2, test 3), and 60,000 lives could be saved if a new test (with performance similar to current SSM) is 85% sensitive for paucibacillary (ss-) cases as well (Table 2, test 5). A rapid test with the same performance as test 5 in Table 2 could save an additional 100,000–130,000 lives, depending on how widely it is used (Table 2, test 6 or 12); for example, 130,000 (392,000 - 263,000) if accessible everywhere. Even an extremely slow test that is 100% sensitive and 100% specific (as achieved by a reliable system for transporting specimens for culture and returning results) would have a major impact, saving 146,000 lives if offered in current TB clinics (Table 2, test 7) and 334,000 if accessible to all (Table 2, test 13).
After evaluating potential outcomes, we concluded that important gains could be achieved by a test that is 85% sensitive in detecting both ss+ and ss- cases, 97% sensitive, rapid (no loss to follow-up) and available to all (Table 2, test 12). Such a test could save 392,000 lives, which is a reduction of more than 22% of the annual global TB deaths.
Other sensitivity and specificity trade-offs affect health outcomes
Figure 2a shows how sensitivity and specificity affect the gains over the status quo for tests that are rapid and accessible to all persistent coughers (no infrastructure requirements). Figure 2a gives isocurves for the number of adjusted lives saved for combinations of specificity (vertical axis) and sensitivity (horizontal axis) of a new diagnostic test for all symptomatic pulmonary TB. The isocurve for 100,000 lives saved extends roughly from 60% sensitivity and 100% specificity (Fig. 2a, near the upper left-hand corner) to 90% sensitivity and 50% specificity (Fig. 2a, near the lower right-hand corner). These gains are due to improved access, as the current smear test is substantially more accurate. Therefore, any rapid universal tests for which sensitivity (%) + 0.6 
specificity (%) = 120% could save 100,000 lives annually worldwide. To the upper right of the 100,000 line in Fig. 2a, the isocurves are parallel and relatively close together because the new test is better for all persistent coughers, and the gains relate simply to the number of correct diagnoses. To the lower left, Fig. 2a shows that there is little gain for less accurate tests that cannot even replace X-rays and trials of antibiotics.
Figure 2: Sensitivity and specificity tradeoffs in new tests.
(a) Annual lives saved (in thousands) worldwide with a rapid test with no infrastructure requirements. (b) Annual lives saved worldwide with a rapid test that could be performed in tuberculosis clinics.
High resolution image and legend (60K)
Figure 2b similarly shows the gains for rapid tests that require the infrastructure available at TB clinics. There are essentially no gains for tests with 70% sensitivity. The values are 50% smaller than those in Fig. 2a, with the curve for saving 100,000 lives annually approximated by the following: sensitivity (%) + 0.6 specificity (%) = 140%. The value for a universally accessible test that does not detect ss- cases is similar to that in Fig. 2a, but the values on the isocurves are 20% smaller (data not shown).
Sensitivity analyses
In order to gain a better understanding of how the parameters of the model influence the outcomes, we carried out a series of one-way and two-way sensitivity analyses. Figure 3 illustrates the effect of a 0.01 increase in many parameters on the gains in adjusted lives saved over the status quo in southeast Asia, which is the region with the largest burden of disease. Figure 3 shows the change in gains from a baseline test that is rapid, accessible to everyone, and has 85% sensitivity to both ss+ and ss- TB with 97% specificity (Table 2, test 12). With these baseline parameters, such a test would save 123,000 lives over the status quo in southeast Asia. The parameters are ranked in order of impact. For example, a 1% increase in the incidence of ss+ TB among persistent coughers means that the new test would save 6,000 additional lives to give a total of 129,000 lives saved. The characteristics of the new test are highly influential, as are the case-fatality rates: higher fatality rates of untreated cases make the new diagnostic more efficacious, whereas higher fatality rates of properly-treated cases make it less so (Fig. 3, bars pointing to the left). The health harm associated with each treatment increases the gains, because the new diagnostic test reduces the number of treatments. If the status quo tests — SSM, X-rays and trials of antibiotics — are more effective than assumed, the gains are reduced. All the variables associated with the follow-up of negative smears and non-sputum diagnosis have little effect on the results, and are therefore omitted. Figure 3 can also be used to set an upper limit on the impact of larger changes in the parameters. For example, a 0.01 reduction in the return rate for test results reduces the gains by 4,000; therefore, if the return rate for the new diagnostic test is 95% (5% less than the baseline 100%), the lives saved are 103,000 (123,000 - 5 4,000).
Figure 3: Impact of a 0.01 change in parameters on lives gained (in thousands) in southeast Asia.
DOTS, directly observed short course chemotherapy; non-DOTS, less-effective TB treatment; ss-, sputum smear negative; ss+, sputum smear positive; TB, tuberculosis.
High resolution image and legend (72K)
Table 3 includes the results of the Monte-Carlo test to compute the impact of parameter uncertainties on the gains from the test. In the third column of Table 1, we give the upper and lower limits used for each parameter in the analysis for the southeast Asian HIV- population. The characteristics for the new diagnostic are held fixed, but the status quo test characteristics are varied within those limits. As shown in Table 3, the standard deviation of the gains is relatively large and increases, but at a slower rate than the gains from increasing access and performance in the new test. In most cases, the standard deviation is between 14,000 and 40,000 lives, and the coefficient of variation (defined as the standard deviation divided by the mean) varies from 2 to 0.25.
Discussion
This modelling exercise shows theoretically that improved diagnostic tests for adults presenting with persistent cough in the four highest-burden WHO regions could have a notable impact on TB outcomes, reducing annual mortality by up to 625,000 lives or 36% of the total. This result assumes an ideal test with 100% access to a rapid diagnostic with 100% sensitivity and 100% specificity.
Although 625,000 deaths averted annually is a significant number, one might wonder why an ideal test would not reduce annual mortality by >36%. There are several explanations. First, patients not included in the model — including children, individuals with extrapulmonary disease, and patients in Europe and the Americas — account for 24% of annual deaths. Second, an additional 15% represent the incremental deaths due to cases assumed to receive less-effective non-DOTS TB treatment. Third, even if all symptomatic TB cases were started on DOTS treatment at the time of diagnosis, the assumed average case-fatality rate of 8% accounts for the remaining 25% of annual deaths.
If we consider the impact of improved but less-than-perfect new tests, we see that annual deaths from TB are so numerous that even substantial gains from improved diagnostics might represent only a small fraction of the current total. For example, improvements in the sensitivity of SSM in existing laboratories so that 100% of ss+ patients are detected will save 46,000 lives, which is an impressive number in absolute terms but only 3% of the current annual death toll. Ensuring 100% compliance with the diagnostic strategy of three smears could save an additional 95,000 lives. Larger gains require either increased access or much better test performance.
Access to results can be improved if tests require less infrastructure (thereby allowing more people to start the diagnostic process) or are faster and so have reduced loss to follow-up before the end of the diagnostic process. If the user requirements of a rapid test (with no loss to follow-up) with the same performance of SSM were so minimal that every persistent cougher could be tested, 15% of annual TB deaths could be averted. These gains from rapid tests might be overstated; in the model, loss to follow-up with a rapid test is assumed to be 10–15% better than with the status quo SSM, and those individuals lost to follow-up are assumed to be untreated. People who begin treatment after the rapid test but do not return for sputum results might be less likely than average to complete therapy. Also, even if these individuals are lost initially in the status quo model, some will eventually return and ultimately be treated.
Due to the limited sensitivity of SSM, many ss- persistent coughers are treated empirically for TB. One mechanism to increase the number of correct treatments is to improve the sensitivity of the tests. The model shows that notable gains are possible from making slow but highly sensitive and specific tests widely accessible.
The results of our model indicate that the number of TB deaths can be reduced through improved use of existing technologies and multiple new diagnostic approaches. Indeed, all of the parameters tested — performance, speed and access — are important for achieving gains in TB outcomes, which can be realized by implementing improvements in any of these areas. Moreover, these solutions are synergistic: the benefits gained from implementing multiple solutions exceed the sum of the incremental gains achieved from individual solutions. For example, access is particularly important in determining the benefits of a diagnostic, but the value of increased access depends on how beneficial it is for patients to receive the test. Therefore, although increased access and improved performance might need to be traded off against each other when designing new diagnostic tools, a greater impact can be created by bringing together a combination of improvements. Accordingly, a rapid and universally accessible test that is not affected by HIV status, with a sensitivity of 85% for ss+ and ss- cases, and a specificity of 97% (Table 2, test 12), was estimated to save 392,000 adjusted lives annually, or 22% of the global TB deaths. If the test were available only to people visiting TB clinics or hospitals, the impact would be reduced by 50%. Therefore, although access is critical, either test provides a feasible target for developers with an enormous impact.
There are several limitations to our analysis. Our primary interest was to compare the effects of different dimensions of diagnostic test performance on potential reductions in disease burden. Assuming that the actual gains throughout the lifetime of the new test (in terms of case fatalities and transmission) would be roughly proportional to the immediate gains, we used simple static decision analytic models to highlight the impact of the diagnostic encounter on these gains. This simplified the analysis; however, the limitations of our modelling decision should be noted25, 26. Future changes in the infrastructure, accessibility and success of TB treatment, and in the epidemiology of TB and other factors affecting health, might render some characteristics more important than is reflected in these initial annual estimates. Because our models are based on the WHO regional averages, adjustments in diagnostic protocols for particular local circumstances were not modelled.
We did not consider costs, and assumed that individuals with geographic access to a test would receive it, and that treatment would be available for those diagnosed with the disease. Although improved diagnosis and, hence, improved results from treatment should increase care-seeking behaviour, to the extent that these optimistic assumptions are not met, the actual gains will be less than the potential gains. Until a new diagnostic is fully implemented, the gains will be delayed, so trade-offs between more demanding test characteristics and the time taken to achieve full implementation will have to be made. Even if the model represents the diagnostic episode adequately, data were scarce or lacking for several key parameters, so we used expert judgment to populate the model. To deal with these uncertainties, we presented sensitivity analyses of how the results vary with each parameter and with the whole range of uncertainties.
In conclusion, this modelling exercise shows that death from adult pulmonary TB can be significantly reduced through the application of various new diagnostic solutions. Improvements in test performance, access and speed are important and synergistic. The model provides a framework for quantifying the gains from any proposed incremental improvements. The most important gains, considering tests that seem feasible to develop, would come from a rapid and widely available test with 85% or more sensitivity (for both ss+ and ss- cases) and 97% specificity. We summarize our key findings in Box 1.
With political will and international collaboration, reducing deaths through the improved use of existing tools (enhanced training) and simplification of the current diagnostic process (same-day results and/or patient-tracking systems) can be undertaken immediately. For the future, the development and appropriate use of better diagnostic tests could have a huge impact on TB control.
