The South African Tuberculosis Care Cascade: Estimated Losses and Methodological Challenges

1South Africa is one of 6 countries accounting for 60% of the global tuberculosis burden. In 2015 there were an estimated 454000 incident cases, at a rate of 834 cases per 100000 population. It has the highest burden of human immunodeficiency virus (HIV) co-infected cases globally, estimated at 258000. Multidrug-resistant tuberculosis (MDR-TB) prevalence rates have remained stable, at 2.9%, in the 2001–2002 survey, compared with 2.8% in 2012–2014, but the rate of rifampicin resistance (RIF-R) has increased, from 3.4% to 4.6%. While estimated tuberculosis incidence rates and mortality appear to be decreasing (Figure 1), the current rate of decline is too slow to meet the 2030 Sustainable Development Goals or 2035 End TB Strategy targets. Extrapolating from World Health Organization estimates, by 2030 and 2035, tuberculosis incidence rates for South Africa would need to decrease to 167 and 83 cases per 100000 population, respectively, and mortality would need to decrease to 9800 and 4900 cases, respectively.

2Reaching these ambitious targets demands a more invigorated response from the National TB Programme in South Africa. The historical focus of the National TB Programme has been on treatment success rates, which fail to reflect upstream losses contributed by individuals who do not access health services and by those for whom tuberculosis is not diagnosed, notified, and treated. A better understanding of the continuum of care (also known as the “care cascade”) can provide the evidence needed on where to target interventions to reduce attrition along the pathway to successful treatment. This approach has previously been used in South Africa to assess and address implementation challenges in the program to prevent mother-to-child HIV transmission and in antiretroviral treatment services.

3We undertook an analysis of the tuberculosis care cascade in South Africa to assess the number and proportion of individuals with tuberculosis who accessed tests, were diagnosed with tuberculosis, initiated treatment, and successfully completed treatment, to quantify losses at each step for (1) all tuberculosis cases, (2) drug-susceptible (DS) tuberculosis cases, (3) HIV-coinfected DS tuberculosis cases, and (4) RIF-R tuberculosis cases.

4South Africa has a district health system, with 52 districts among its 9 provinces. There are 3500 public primary healthcare facilities, equivalent to 1 facility per 15000 population. Only 18.4% of the population was covered by private medical aid schemes in 2013, with the balance largely dependent on the public health sector.

5The majority of tuberculosis tests (93%) and almost all treatment are provided within the public health sector in South Africa. Individuals treated in the mining sector and correctional services are reported to the South African National Department of Health. Tuberculosis diagnostic and treatment services are provided free of user costs in the public sector.

6The National Health Laboratory Services, a parastatal organization, provides tuberculosis diagnostic services, mostly through centralized laboratories. Historically, sputum microscopy formed the basis of tuberculosis diagnosis, with liquid culture (Bactec MGIT 960) available for smear-negative, HIV-coinfected new cases and culture and drug susceptibility testing (conventional or GenoType MTBDRplus, Hain Lifescience, Nehren, Germany) available for previously treated cases. Xpert MTB/RIF (hereafter, “Xpert”; Cepheid, Sunnyvale, CA) was phased in during 2011, replacing smear microscopy for all presumptive tuberculosis cases.

7Standardized first-line tuberculosis treatment is provided at a primary care level. Standardized second-line tuberculosis treatment is initiated mostly in tuberculosis hospitals, with decentralization to the primary healthcare level in some areas. All individuals receiving treatment have standard clinical records completed; data are summarized in paper-based registers at the facility level and captured into the electronic tuberculosis register (ETR) or the electronic drug-resistant tuberculosis register at the subdistrict levels. Subdistrict reports are dispatched to the district level and then to provincial and national levels.

8We estimated the number of cases at each step of the care cascade for the following: (1) all tuberculosis cases, (2) DS tuberculosis cases, (3) HIV-coinfected DS tuberculosis cases, and (4) RIF-R cases. We elected to do the analysis for the 2013 cohorts because this was the most recent year for which a cleaned, deduplicated national data set for DS tuberculosis cases was available from the ETR and for which RIF-R tuberculosis treatment outcomes were available from the electronic drug-resistant tuberculosis register.

9In addition to the data sources described above, we used summary district data from the National Health Laboratory Services on tuberculosis testing and published literature to estimate the numbers of cases at each step or the gaps between successive steps. Figure 2 summarizes the conceptual approach, and Table 1 describes the detailed methods and calculations used to estimate each of the steps described below.

10We used the World Health Organization 2016 time-series analysis of tuberculosis incidence to estimate the 2013 burden for all cases and for HIV-coinfected cases. We defined the “tuberculosis burden” as the number of incident tuberculosis cases in 2013 plus 50% of the number of undetected cases from 2012, with the assumption that 50% of individuals with undetected tuberculosis had died or self-cured by the start of 2013. The rationale for this approach is addressed in the Discussion, below. The number of RIF-R cases was estimated from the South African tuberculosis drug resistance survey and subtracted from these figures to derive figures for all DS tuberculosis cases and for HIV-coinfected DS tuberculosis cases.

11The gap between the number of individuals with tuberculosis who accessed tests and the number diagnosed with tuberculosis reflects cases missed by tests with a lower sensitivity than that of culture (ie, false-negative cases). The number who accessed tests was back calculated from the number who received a diagnosis, based on laboratory data on the proportion for whom tuberculosis was diagnosed by smear, Xpert, and GenoType MTBDRplus (Central Data Warehouse, National Health Laboratory Services); estimates of test sensitivity; estimates of the proportion of smear-negative and Xpert-negative cases with a culture test; and estimates of the proportion of false-negative cases that were treated empirically.

12The number of tuberculosis cases diagnosed was back calculated from the number of cases notified and treated, based on the rate of initial loss to follow-up (ILTF; ie, individuals with bacteriologically confirmed tuberculosis who did not initiate treatment).

13We estimated the ILTF rate for DS tuberculosis cases from studies that were undertaken in South Africa and published in peer-reviewed journals during 2006–2016 (online appendix). We used the “metafor” R package and computed pooled estimates by using the DerSimonian-Laird method and 95% confidence intervals by using the Wald method. Studies differed in the testing strategies used (smear, smear and culture, centralized performance of Xpert, and decentralized performance of Xpert), as well as in the duration after tuberculosis diagnosis when ILTF was defined. We did not expect studies to share a common effect size and therefore used data from the random-effects model. We included studies and study arms that used smear/culture or Xpert performed centrally (to reflect the status quo in 2013).

14Only 2 studies reported on the association between HIV status and ILTF, and neither found a significant association. We therefore assumed the same rate of ILTF among HIV-coinfected DS tuberculosis cases.

15For RIF-R tuberculosis cases, we used the ILTF rate reported from a large, nationally representative study for the 2013 cohort.

16We used patient-level data from the ETR to estimate the number of DS tuberculosis cases (pulmonary and extrapulmonary) that were notified and for which treatment was started. Cases are recorded as being one of 3 registration types: newly registered (the patient was at this facility), moved in (the patient arrived from another facility in the district where they would have been newly registered), and transferred in (the patient arrived from another facility outside the district where they would have been newly registered). To avoid double counting, only newly registered cases would usually be included as cases notified and treated. However, we adjusted this by 4% (range, 2%–6%) to account for the net difference between patients who moved in and those who moved out at a district level that suggests that some of the patients who moved in were not previously reported as newly registered at another facility in the district.

17Over 90% of patients with tuberculosis had their HIV status recorded in ETR, and among those tested, 61.7% were infected with HIV. We applied this proportion to the adjusted number of newly registered cases to derive the number of HIV-infected individuals receiving treatment for DS tuberculosis.

18The number of RIF-R cases notified and treated was based on the National Department of Health’s report for the 2013 cohort (unpublished data, Research, Information, Monitoring, Evaluation, and Surveillance, TB Control and Management Cluster, National Department of Health). These included individuals with RIF-monoresistant, multidrug-resistant, pre–extensively drug-resistant, or extensively drug-resistant tuberculosis.

19We used ETR line data to estimate weighted success rates (ie, the proportion of individuals who were cured or completed treatment) and 95% confidence intervals for all individuals with DS tuberculosis and for HIV-infected individuals with DS tuberculosis, at a district level. South Africa does not have a unique health identifier to assist in tracking patients who move between facilities, presenting a challenge to calculating treatment outcomes. For DS tuberculosis cases, the National Department of Health addresses this by including outcomes for all patients who were newly registered or moved in and excluding outcomes for patients who moved out (as these are reported at the facility where they moved in). We address the implications of using this method in the discussion.

20For DR tuberculosis cases, we used summary provincial data to estimate weighted success rates and 95% confidence intervals, based on the National Department of Health Report for the 2013 cohort (unpublished data, Research, Information, Monitoring, Evaluation and Surveillance, TB Control and Management Cluster, National Department of Health).

21We express absolute numbers and the ranges of possible values and losses at each step of the cascade in relation to the estimated burden graphically in Figures 3–6.

22We estimated the tuberculosis burden to be 532005 cases (range, 333760–764480 cases; Figure 3). Of these, 504514 individuals (range, 449505–589689 individuals) accessed tuberculosis testing, and 435483 (range, 411066–461766) were diagnosed with tuberculosis. Among these, in 372577 (range, 354162–368051) tuberculosis was notified and treated, with 279816 (range, 279688–279945) successfully completing treatment, equivalent to 53% of the estimated tuberculosis burden. About 5% of individuals (27491) did not access diagnostic services, 13% (69030) were lost during the diagnostic process, 12% (62906) did not initiate treatment, and 17% (92761) did not successfully complete treatment.

23The DS tuberculosis burden was estimated to be 507533 cases (range, 322078–720905 cases; Figure 4). Of these, 483912 individuals (range, 430534–566274 individuals) accessed tuberculosis tests, and 417277 (range, 393420–442963) were diagnosed with tuberculosis. Among these, in 361107 (range, 354162–368051) tuberculosis was notified and treated, with 274441 (range, 274333–274549) successfully completing treatment, equivalent to 54% of the estimated DS tuberculosis burden.

24We estimated the burden of HIV-coinfected DS tuberculosis to be 314491 cases (range, 179355–480855 cases; Figure 5). Of these, 305910 individuals (range, 273121–354848 individuals) accessed tuberculosis tests, and 257316 (range, 242605–273155) were diagnosed with tuberculosis. Among these, in 222678 (range, 218396–226961) tuberculosis was notified and treated, with 164804 (range, 161548–168110) successfully completing treatment, equivalent to 52% of the estimated burden of HIV-coinfected DS tuberculosis burden.

25The RIF-R tuberculosis burden was estimated to be 24472 cases (range, 11682–43575 cases; Figure 5). Of these, 20602 individuals (range, 18971–23415 individuals) accessed RIF-R tuberculosis tests, and 18206 (range 17646–18803) were diagnosed with RIF-R tuberculosis. Among these, in 11470 RIF-R tuberculosis was notified and treated, with 5375 (range, 5355–5395 cases) successfully completing treatment, equivalent to 22% of the estimated burden of RIF-R tuberculosis cases. About 16% of individuals with RIF-R tuberculosis (3871) did not access diagnostic services, 10% (2395) were lost during the diagnostic process, 25% (6095) did not initiate treatment, and 22% (5375) did not successfully complete treatment.

26Although the concept of continuity of care and attrition along the care pathway has been deliberated in tuberculosis control for many years, there are few published data that quantify outcomes. A study on the care cascade in India estimated that treatment was completed in 45% of prevalent tuberculosis cases in 2013. Our study estimates that just over half (53%) of all tuberculosis cases in South Africa in 2013 were successfully treated, with substantial losses during tuberculosis diagnosis, linkage to care, and retention in care.

27In our setting, access to health care and tuberculosis diagnostic services does not appear to be a major impediment because testing was not accessed in only 5% of estimated cases. The small magnitude of this gap may reflect the wide network of free public primary healthcare facilities with ready access to tuberculosis diagnostic services. The National Income Dynamics Study confirmed high levels of healthcare access; in the 3 years surveyed (2008, 2010, and 2012), only 12.8% of respondents had never accessed health services, and 7% with a tuberculosis symptom (ie, cough, hemoptysis, fever, severe weight loss, and/or chest pain) in the last month had not previously accessed health services. This differs from findings on the tuberculosis care cascade in India, which showed that, for more than half of the estimated 55% of cases for which treatment was not completed, diagnostic services were not accessed.

28However, there is substantial uncertainty about the burden of disease estimate, and the magnitude of this gap in South Africa may well be underestimated, as indicated by the wide estimate range; at the upper level of the range, the gap could be as high as 34%. Estimating the annual burden of disease based on either prevalence or incidence data presents challenges. South Africa has not had a tuberculosis prevalence survey to date (the first survey is currently underway). Although tuberculosis prevalence surveys offer the best direct measure of disease burden at a point in time, estimating the annual disease burden from this would require data on the duration of disease for both HIV-infected individuals and HIV-uninfected individuals, which cannot be determined accurately. In the Indian study, the use of point prevalence data and the uncertainty in determining the 1-year period prevalence were identified as limitations.

29Direct estimates of tuberculosis incidence are generally not financially or logistically feasible, requiring prospective cohort assessments of several hundred thousand individuals over the period of a year. Indirect estimates are therefore used; for South Africa, these were based on case notification data and expert opinion on case detection gaps (ie, underdiagnosis and/or underreporting).

30Incidence estimates, however, fail to account for undetected cases from prior periods that were ongoing in 2013 and that contribute to the burden of cases to be identified and treated. Our method for estimating the tuberculosis burden (calculated as the number of incident cases plus 50% of the number of undetected cases from the previous year) is based on the Styblo rule that cases remain infectious for an average of 2 years, with case-fatality rates of 50%. This may well be an underestimate—a systematic review of untreated pulmonary tuberculosis in HIV-uninfected individuals estimated a 3-year duration to self-cure or death. On the other hand, this is balanced by HIV-coinfected individuals (62% in South Africa), who are likely to become symptomatic faster, leading to earlier detection and treatment.

31About 13% of all cases were lost between tuberculosis testing and diagnosis, partly because of the failure to comply with diagnostic algorithms. Despite South Africa being the biggest consumer of Xpert cartridges (procuring more than half of the 7.5 million cartridges procured globally to mid-2014) and the national roll-out of Xpert, <60% of individuals with presumptive tuberculosis received an Xpert test. The remainder were tested by less sensitive smear microscopy. Poor compliance with the diagnostic algorithm extended to follow-on performance of culture (recommended for all HIV-infected individuals with initial negative test results). A national study undertaken during the scale-up of Xpert found that only 14% of Xpert-negative and 32% of smear-negative, HIV-infected individuals had culture performed. These lapses have major implications for case detection. For example, testing 80% of presumptive tuberculosis cases with an Xpert test and 80% of Xpert-negative, HIV-infected cases with culture would reduce the number of tuberculosis cases missed at this step by almost 32000. Addressing this gap requires both routine HIV testing of presumptive tuberculosis cases and monitoring adherence to testing algorithms. Reflex laboratory testing algorithms, in which Xpert is automatically used as the first diagnostic test among presumptive tuberculosis cases and cultures are undertaken for all Xpert-negative, HIV-infected cases, could also be considered.

32An estimated 12% of all cases were diagnosed but not notified and treated. These 62906 cases with ILTF exceeded the total number of tuberculosis cases notified in 2015 in 5 of South Africa’s neighboring countries combined (ie, Botswana, Namibia, Lesotho, Swaziland, and Zimbabwe). In the Indian study, the 7% with ILTF equated to 212068 cases. In South Africa, the loss at this step was substantially higher among RIF-R tuberculosis cases (25%) than among DS tuberculosis cases (11%). Diagnostic delay has been reported as one of the factors contributing to ILTF, and improved adherence to algorithms could potentially reduce delay and thus ILTF. Fragmented data systems between laboratories and health facilities contribute to poor linkage to care, and system integration can improve this. Health system failures, such as poor recording of patients’ contact details, results not being available when patients return to the health facility, and poor follow-up of patients who do not return for test results, should be prioritized. In addition, perceptions of poor quality of services, including long waiting times and disrespectful staff attitudes, which have been well documented in the public health sector and contribute to ILTF, need to be addressed. These infectious cases contribute to ongoing tuberculosis transmission and to high mortality rates. Having invested scarce financial resources in diagnosing tuberculosis, urgent efforts are required to close this gap, to avoid this fruitless expenditure.

33The historical focus of the tuberculosis program on treatment success rates among new, smear-positive cases (which accounted for only one third of tuberculosis cases) has resulted in insufficient attention to outcomes for all cases. Although treatment success rates for new, smear positive cases exceed 80%, treatment success rates for all cases are substantially lower. One in every four patients initiating tuberculosis treatment did not successfully complete treatment, equivalent to the loss of 92761 individuals during treatment. It is likely that this figure is an underestimate because the method used in calculating treatment success assumes that all patients who move out of health facilities move in to facilities elsewhere. Net differences between these figures in several districts suggest that this not the case. The use of a unique patient identifier would enable improved monitoring and quantification of this loss.

34Health system performance, including integration of tuberculosis and HIV services and strong district and facility leadership and management, are predictors of high treatment success rates. Poor patient knowledge, lack of empowerment among patients to engage appropriately with health providers, and high costs associated with attending daily for direct observed therapy have been found to contribute to nonadherence and need to be addressed. mHealth and other interventions to improve adherence to tuberculosis treatment need to be evaluated to reduce this significant gap.

35Treatment outcomes were substantially poorer among RIF-R tuberculosis cases than among DS tuberculosis cases, with just over half the patients with RIF-R tuberculosis successfully treated. South Africa is pursuing new policy options for interventions that have been shown to improve MDR-TB treatment outcomes, including decentralized models of MDR-TB care and the shortened 9-month treatment regimen.

36Although the outcome in the RIF-R tuberculosis care cascade was substantially worse than in the DS tuberculosis care cascade, the small number of cases results in minimal impact on the cascade for all tuberculosis cases. Despite the small numbers, RIF-R tuberculosis cases draw disproportionately on financial resources, with about 50% of national tuberculosis expenditures allocated to treat the 7% of patients with RIF-R tuberculosis in 2014. Unless efforts are made to reduce leakages in the RIF-R tuberculosis care cascade to help reduce the high levels of primary transmission, this situation could be exacerbated in the future.

37Overall outcomes were similar in the DS tuberculosis cascade and the HIV-coinfected DS tuberculosis cascade, suggesting parity in outcomes between individuals with and those without HIV infection. This may reflect national efforts to increase access to antiretroviral treatment.

38Although the tuberculosis care cascade enumerates losses at each step, it does not reflect the delays that occur between successive steps. Two systematic reviews suggest significant time delays. The first of these reported overall delays of 25–185 days, with delays from symptom onset to the first healthcare visit of 5–162 days (average, 29 days) for patients and delays from the first healthcare visit to diagnosis of 2–87 days (average, 25 days) for health systems. The second review reported overall delays of 21–136 days, with patient delays ranging from 7 to 69 days and health system delays ranging from 2 to 120 days. The patient pathway analysis, a complementary approach that seeks to address the bottlenecks and time delays that contribute to ongoing tuberculosis transmission would add value to the current approach.

39Despite relatively advanced data systems in South Africa, the absence of a unique health identifier and poor data integration makes it difficult to track individuals along the care cascade from tuberculosis testing to diagnosis, treatment initiation, and completion. We have thus estimated the number of cases at each step, and the methods we used have limitations, as already discussed. The use of a unique health identifier would enable a true cohort approach to reporting on the tuberculosis care cascade and would provide a more accurate reflection of attrition at each step.

40The national findings on the tuberculosis care cascades may hide significant variations that occur at provincial, district, and facility levels. Improved data systems capable of providing this information at a more granular level would allow appropriate geographic targeting of interventions to areas where they are needed most.

41In conclusion, despite the vast majority of tuberculosis cases engaging the public health system in South Africa, just over half the estimated cases were successfully treated in 2013. Health system failures at every level, from poor adherence to tuberculosis testing algorithms to poor linkage and retention in care, contribute. The analysis suggests that no single intervention will help achieve the End TB Strategy goals and that high-impact interventions are required at multiple points to improve overall outcomes.

42Research and development can play a role in reducing losses, such as through an effective triage test, more-sensitive diagnostic tests, and shorter/improved treatment regimens. However, many of the factors contributing to losses in the tuberculosis care cascades reflect poor implementation of existing policies and protocols. We currently have the tools available to significantly close these gaps. The tuberculosis care cascade offers a simple, visual way to illustrate where losses occur in the care continuum and, with improved data integration, can be used to routinely track programmatic efforts to close gaps in tuberculosis diagnosis, treatment initiation, and successful treatment completion.

43Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.