Mortality during tuberculosis treatment in South Africa using an 8-year analysis of the national tuberculosis treatment register

In 2011, the South African HIV treatment eligibility criteria were expanded to allow all tuberculosis (TB) patients lifelong ART. The impact of this change on TB mortality in South Africa is not known. We evaluated mortality in all adults (≥ 15 years old) treated for drug-susceptible TB in South Africa between 2009 and 2016. Using a Cox regression model, we quantified risk factors for mortality during TB treatment and present standardised mortality ratios (SMR) stratified by year, age, sex, and HIV status. During the study period, 8.6% (219,618/2,551,058) of adults on TB treatment died. Older age, male sex, previous TB treatment and HIV infection (with or without the use of ART) were associated with increased hazard of mortality. There was a 19% reduction in hazard of mortality amongst all TB patients between 2009 and 2016 (adjusted hazard ratio: 0.81 95%CI 0.80–0.83). The highest SMR was in 15–24-year-old women, more than double that of men (42.3 in 2016). Between 2009 and 2016, the SMR for HIV-positive TB patients increased, from 9.0 to 19.6 in women, and 7.0 to 10.6 in men. In South Africa, case fatality during TB treatment is decreasing and further interventions to address specific risk factors for TB mortality are required. Young women (15–24-year-olds) with TB experience a disproportionate burden of mortality and interventions targeting this age-group are needed.

In 2018, an estimated 10 million people developed tuberculosis (TB) disease worldwide and, among adults (≥ 15 years old), 64% of cases occurred in men 1 . Despite progress towards reducing TB mortality, it remained the leading cause of death from an infectious disease globally, with an estimated 1.5 million (15%) deaths 1 . A systematic review found that case fatality during TB treatment ranged from 1.8 to 33.3%; male sex was identified as a risk factor for death 2 . Understanding mortality in adults on TB treatment presents an opportunity to plan interventions to reduce mortality amongst individuals already engaged with health services.
In sub-Saharan Africa, the dual HIV and TB epidemics have been well described. Prior to the rollout of antiretroviral therapy (ART) the observed rise in TB case fatality ratio (CFR) was linked to increasing HIV prevalence 3 . In South Africa, routine HIV testing was implemented in 2010 4 and in 2012 the ART policy expanded so that all HIV-positive TB patients were eligible for lifelong ART 5 . ART uptake in South Africa has been variable with adult women showing the greatest uptake of ART 6 .
The dramatic effect of ART on reducing mortality among HIV-positive TB patients has been documented under both trial 7 and routine programmatic conditions 8 . In low-and middle-income countries, the effect of widescale ART rollout on mortality has varied by sex, with greater reductions in mortality observed in women [9][10][11] . A recent systematic review showed a mortality reduction of 44-71% among HIV-positive TB patients with the use of ART 8 . In South Africa, modelling work has shown that while the ART rollout has contributed to a substantial number of lives saved, almost 2.2 million people died between 2000 and 2014 due to delays in the rollout 12  www.nature.com/scientificreports/ Understanding the changes in mortality during TB treatment by age and sex, in the context of the scale up of ART, may provide insights into how health programs can deliver targeted services to further reduce TB mortality in South Africa and globally. The impact of the revised TB, HIV and ART policies on mortality during TB treatment in South Africa, and how the impact has differed by sex, is not well known. Using an individual patient-based national electronic TB treatment register of drug-susceptible TB treatment we documented the changes in TB mortality compared to the general population mortality and identified risk factors for death during TB treatment.

Methods
Context. South Africa had an estimated population of 55.7 million in 2016, a 11.6% increase from 49.9 million in 2009 12 . The country offers a district health service with 52 health districts delivering public health services, across 9 provinces. In 2018, South Africa had an estimated TB incidence of 520/100,000 per year and approximately 59% of TB patients were HIV-positive 1 . Laboratory diagnosis of TB in South Africa utilised smear microscopy, culture and line probe assay until 2011, after which the GeneXpert MTB/RIF assay (Xpert; Cepheid, Sunnyvale, CA) was progressively introduced for the testing of all patients with presumptive TB.
Electronic TB register (ETR.Net). In South Africa, all patients were presumed to have drug-susceptible TB, unless drug susceptibility testing demonstrated resistance, and were routinely registered in a paper-based register at a TB treatment facility when initiating treatment. This register was validated by facility managers and sub-district co-ordinators before it was captured into an electronic TB register (ETR.Net), and aggregated at district, provincial and national level 13 . We obtained national and provincial ETR.Net dispatch files and undertook a process of merging, de-duplication, and data cleaning. ETR.Net includes a unique identifier and this was used to identify duplicate records that were merged before deletion to provide complete data on HIV, treatment dates and outcomes. HIV data in ETR.Net is variably completed and may include HIV status, CD4 count, and the use of cotrimoxazole preventive therapy or ART. Within ETR.Net each patient was recorded for the first time as a newly registered patient and could then move or be transferred between facilities in the country. A cohort including all newly registered TB patients was used to report on TB case finding and a reporting cohort accounting for patient movement during treatment was used for the reporting of TB treatment outcomes. For this analysis, the South African reporting cohort was used.
The Thembisa model. Estimates of population denominators, HIV prevalence, and mortality were extracted from the Thembisa project (https:// www. themb isa. org/), a publicly available mathematical model of HIV and general demographic statistics 14 . The model uses age and sex-specific mortality estimates based on an analysis of South African cause-of-death statistics and the South African National Burden of Disease study, and projects mortality from 2016 to 2030 12 . Design. This study is a retrospective cohort analysis of mortality among all adults (≥ 15 years) in South Africa routinely started on drug-susceptible TB treatment between 1 January 2009 and 31 December 2016.
Definitions. Consistent with global reporting of TB we considered adults as those ≥ 15 years old 1,15 . Table 1 includes the definitions used for variables recorded within ETR.Net 15 . For outcome definitions the standard WHO and national TB program definitions were used 15,16 . CFR was calculated as the proportion of deaths as compared to the total number of people on TB treatment. The total number of people on TB treatment included patients with unknown outcomes. Two additional outcome variables were constructed to evaluate unfavourable outcomes ('Outcome 2') and mortality among patients with known final outcomes ('Outcome 3'). Person-time was calculated as the difference between the TB treatment start date and treatment outcome date, representing the person-years on TB treatment. TB deaths refer to deaths due to any cause during the TB treatment episode and underlying cause of death was not evaluated. This is consistent with the national TB program and WHO definition 15,16 . Statistical analysis. We conducted descriptive statistics for categorical variables and stratified these by HIV status. Changes over time were evaluated by year and graphically illustrated. CFRs were calculated by age, sex and HIV-status and provide a conservative estimate of mortality as the true mortality status of TB patients who were transferred out, lost to follow up or not evaluated at the end of treatment is unknown. A Cox proportional hazards model was used to determine the unadjusted and adjusted hazard ratios (HRs) with 95% confidence intervals (CI) for predictors of death. Following univariate analyses, predictors were added incrementally observing the change in significance of the likelihood ratio test of each model and collinearity between variables was considered before developing the final adjusted model. The Thembisa model provided the expected mortality rate per demographic category for each year for the general population of South Africa 14 . The expected mortality is informed by all causes and includes TB. The product of the expected mortality and person-years in the TB reporting cohort provided the number of expected deaths in the TB cohort. The standardised mortality ratios (SMR) were calculated as the ratio between the observed TB deaths and the expected deaths within the TB cohort. The data analysis for this paper was generated using SAS software, Version 9.

Results
The South African reporting cohort for drug-susceptible TB included 2,551,058 adults, 55.9% men, treated between 2009 and 2016 ( Table 2). HIV testing and recording increased from 54.8% of individuals with TB having a known HIV status in 2009, to 96.7% in 2016 (Table 2 (Fig. 1). The CFR differed by age and sex: among 15-24-year-olds, 4.9% of women died compared to 3.2% of men. Among 65+ year-olds, men had a CFR of 20.5% compared to 16.0% in women over the entire period (Fig. 2 (Table 3). We explored effect modification by sex and time in stratified models and found differences in point estimates for age categories and HIV status (Supplementary Tables 2 and 3). Table 1. Definitions used in the study based on variables recorded within ETR.Net and classified according to the WHO reporting framework. ART: antiretroviral therapy; ETR.Net: Electronic tuberculosis register; ICD10: International classification of disease 10; TB: tuberculosis. a Extent of disease could only be ascertained where an ICD10 code was documented. b HIV status is typically recorded at the start of TB treatment but may be updated during TB treatment. c Outcomes are assigned by routine treating clinicians and ETR.Net includes pre-programmed algorithms to verify outcomes consistent with WHO and National Department of Health definitions.  65+ year-olds the SMR for men was 6.9 and for women was 6.7 in 2016 (  (Fig. 4).

Discussion
TB CFR during treatment in South Africa is declining but remains high, with 6.9% of adult TB patients dying in 2016. Older age, male sex, previous TB treatment history and HIV status (with or without the use of ART) were associated with increased hazard of death during TB treatment. When compared to mortality in the general population, younger patients with TB had a substantially greater SMR, the highest in young women (15-24-yearolds). The SMR for HIV-positive TB patients increased over time, with a sharper increase among HIV-positive women compared to men. www.nature.com/scientificreports/ South Africa has seen an overall decrease in the TB CFR, but stratification by HIV status and ART is crucial to understand underlying patterns. While an overall decline in the CFR to 6.9% in 2016 was documented, when disaggregated by HIV status, the CFR clearly showed an increase in HIV-positive TB patients not on ART, very little change for HIV-negative individuals, and the steepest decline in TB CFR in HIV-positive TB patients on ART. By 2016, more than 94% of HIV-positive TB patients were on ART and the increasing CFR among HIV-positive patients not on ART therefore reflected the small minority of HIV-positive patients. This group likely reflects those most disengaged with the health system and are a priority group for intervention and ART    Table 3). We documented an increased hazard of death with increasing age and among men and this has been described with older age in previous studies 2 . The increased hazard of death among 65+ year-olds compared to 15-24-yearolds may relate to comorbidity and additional reasons for death among older patients. While older patients were at greater hazard of death compared to younger patients, the SMR declined substantially with age in both men and women. The greater impact of TB on SMRs therefore occurs in the younger population who have few other causes of death. The frequency of other causes of death increased with age among both adults with TB and the general population, and this resulted in the SMR declining with age. Despite a decrease in SMR over time, young women (15-24-year-olds) who had TB, died at a rate 40 times greater than women of the same age without TB in 2016. This SMR was more than double that of men of the same age and is due to differences in both observed and expected mortality. Observed mortality (TB CFR) was 1.5 times greater in women in this age group compared to men. Annually, 250,000 women die during pregnancy in Africa with the majority of deaths due to non-obstetric causes 19 . TB and HIV are among the main infectious causes of maternal deaths 19,20 and HIV and pregnancy may    www.nature.com/scientificreports/ account for this higher CFR seen in young women. Interpersonal violence among young men in South Africa remains common and among 10-17-year-olds who died due to injury, homicide was the leading cause, mainly affecting young men 21 . This may contribute to the greater than expected mortality among men. In a study from India, the SMR for TB decreased sharply with age from 15.1 among 15-44 year-olds, 7.3 among 45-59-year-olds to 2.2 among those older than 60 years 22 but was not disaggregated by sex and did not consider HIV 22 . This is consistent with previous work in which the impact of HIV on TB mortality was greatest in younger TB patients 23 .
Similarly, a study from Kenya demonstrated that the SMR among TB patients decreased with age but was higher in men than women because of a lower expected mortality among younger men and a higher observed mortality among younger women 24 . The Kenyan study, conducted between 2006 and 2008, reflected a period where women were disproportionately affected by HIV and did not have the benefit of widespread access to ART 24 . In more recent work from Kenya, the SMR among TB patients was again shown to decrease with age; the SMR among older patients (65-74-year-olds) was 7.6 in women and 7.9 in men 25 Table 4). Regardless of ART use, HIV infection was associated with an increased hazard of TB death. The greater hazard among HIV-positive patients not on ART compared to HIV-positive patients on ART clearly demonstrates the protective effect of ART. Specific concerns regarding men accessing HIV testing, ART services, retention in care, and mortality have previously been raised 6,9 . The higher rate of ART initiation in women compared to men was described from as early as 2004 with the differential between sexes increasing over time 6 . Several systematic reviews have shown that within ART programs (especially in high burden settings), HIV-positive men have a greater hazard of death compared to HIV-positive women 10,11,26 . In South Africa the effect of ART rollout on a rural community has shown more rapid declines in mortality among women than men 9 . The increasing SMR observed in HIV-positive women in this study reflects this greater reduction in general mortality among the population of HIV-positive women who do not have TB. In a South African study reviewing the effect of ART on mortality, SMRs in women were higher than men and this was attributed to the higher background mortality among men 27 . Similarly, we noted much higher SMRs in women, especially in the younger age bands. This reflects the lower number of expected deaths in this patient category (Supplementary Table 5). As South Africa continues to move towards universal access to ART, monitoring of observed and expected mortality should continue across the TB and HIV programs, to track changes over time and identify populations in greatest need of targeted interventions and support.
The importance of adequate history taking, and prioritisation of individuals previously treated for TB is highlighted by the additional risk of death among previously treated TB patients. Extrapulmonary TB was also associated with an increased hazard of death, consistent with earlier data 28,29 and may be due to more severe and disseminated forms of disease, including TB meningitis.
This study was strengthened by the size of the individual patient data set and the span of 8 years, which provided the opportunity to evaluate temporal changes. We were however careful to only make inferences where biologically plausible, consistent with scientific hypotheses and the literature, and where an effect size was of epidemiological or clinical relevance. We explored effect modification by sex and time in stratified models and while we found differences in point estimates our conclusions were unchanged. Limitations included the use of TB death to refer to death due to any cause during TB treatment, possibly overestimating the effect of TB. However, this is in accordance with international standardised definitions. A large proportion of TB patients had unknown final treatment outcomes, which may have included mortality. Previous work from Malawi has suggested up to 30% of patients lost to follow up may have died 30 . However, by not assuming patients with unknown outcomes had died we have provided conservative CFRs and future work matching individuals with unknown TB treatment outcomes to the population register to ascertain mortality has been planned. A further limitation included the ART information in the TB register. ART use did not differentiate TB patients on ART before TB; those who started ART during TB treatment; or adherence to ART. There was also no information on degree of immunosuppression. We were unable to control for baseline nutritional status, anaemia, underlying disease, or alcohol and drug use as additional risk factors for mortality.
Case fatality during treatment for drug-susceptible TB in South Africa is declining but remains high with 6.9% of TB patients dying during TB treatment in 2016. HIV co-infection, regardless of ART use, increased the hazard of death during TB treatment. We show a disproportionate burden of TB mortality in young women, with a mortality rate among 15-24-year-old women on TB treatment in 2016 more than 40 times greater than for women of the same age in the general population. While future work to prospectively evaluate TB treatment outcomes and risk factors for mortality among young women especially during pregnancy is needed, health services have a clear indication of the need to engage young women during their routine access of health services. Increased education on TB screening, diagnosis, early treatment and treatment completion among young healthy women is needed. Additional care should be offered to young women diagnosed with TB to reduce mortality.