Rifampicin resistance and mortality in patients hospitalised with HIV-associated tuberculosis

Background Patients with HIV and drug-resistant tuberculosis (TB) are at high risk of death. Objectives We investigated the association between rifampicin-resistant TB (RR-TB) and mortality in a cohort of patients who were admitted to hospital at the time of TB diagnosis. Method Adults hospitalised at Khayelitsha Hospital and diagnosed with HIV-associated TB during admission, were enrolled between 2013 and 2016. Clinical, biochemical and microbiological data were prospectively collected and participants were followed up for 12 weeks. Results Participants with microbiologically confirmed TB (n = 482) were enrolled a median of two days (interquartile range [IQR]: 1–3 days) following admission. Fifty-three participants (11.0%) had RR-TB. Participants with rifampicin-susceptible TB (RS-TB) received appropriate treatment a median of one day (IQR: 1–2 days) following enrolment compared to three days (IQR: 1–9 days) in participants with RR-TB. Eight participants with RS-TB (1.9%) and six participants with RR-TB (11.3%) died prior to the initiation of appropriate treatment. Mortality at 12 weeks was 87/429 (20.3%) in the RS-TB group and 21/53 (39.6%) in the RR-TB group. RR-TB was a significant predictor of 12-week mortality (hazard ratio: 1.88; 95% confidence interval: 1.07–3.29; P = 0.03). Conclusion Mortality at 12 weeks in participants with RR-TB was high compared to participants with RS-TB. Delays in the initiation of appropriate treatment and poorer regimen efficacy are proposed as contributors to higher mortality in hospitalised patients with HIV and RR-TB.


Introduction
Drug-resistant tuberculosis (DR-TB) is a global health concern. 1 In South Africa (SA), 13 005 cases of rifampicin-resistant TB (RR-TB) were identified in 2019. 2,3 People living with HIV are at higher risk of acquiring DR-TB, which is associated with poorer treatment outcomes in this patient population. 2,4,5,6 The treatment of DR-TB has low success rates and high mortality due to lengthy, poorly tolerated and poorly efficacious regimens. 7,8 In 2018, only 58% of patients with DR-TB globally were successfully treated. 1 Mortality estimates from studies conducted primarily in outpatients or in TB hospitals in high-burden settings, prior to the introduction of bedaquiline-based regimens, range from 11% to 39%. 9,10,11,12,13,14 The introduction of bedaquiline-based treatment has been associated with a reduction in mortality in patients with DR-TB; however, treatment failure and mortality remain high in high-prevalence settings, with mortality still ranging from 6% to 17%. 15,16,17,18,19,20,21 As most DR-TB is managed in outpatient clinics or in TB hospitals, most research on DR-TB outcomes occurs in these settings. Patients who are hospitalised with HIV-associated TB are often severely ill and inpatient case fatality rates range from 11% to 32%. 22,23,24,25,26,27 In autopsy series, inpatient deaths occur with a median of 4-5 days following admission. 28,29,30 There are scarce data on outcomes of patients diagnosed with HIV-associated DR-TB admitted to general hospitals.
http://www.sajhivmed.org.za Open Access The outcomes of a cohort of patients admitted to hospital in Khayelitsha, SA, newly diagnosed with HIV-associated TB, were previously described. 31 In this cohort, RR-TB was present in 16.9% of participants who died and in 7.2% of survivors. 31 In this secondary analysis, we aimed to describe the association between RR-TB and mortality in patients diagnosed with HIV-associated TB while hospitalised and to identify factors associated with mortality in patients with RR-TB.

Study design and setting
Patients admitted to Khayelitsha Hospital were enrolled to a prospective cohort study between January 2013 and October 2016. The hospital is in Khayelitsha -a large township in Cape Town, SA, with high rates of HIV, TB and DR-TB. 32 Most cases of TB are managed in primary healthcare clinics; however, patients may be referred to hospital if they require inpatient care.

Participants
All patients in the emergency unit and medical wards were screened on weekdays.

Data collection and definitions
Clinical data were captured on standardised case record forms from patient interviews, hospital folders and clinical review at enrolment. The primary outcome was survival at 12 weeks. If patients could not be contacted at 12 weeks, searches of electronic records were conducted. Participants with an electronic entry indicating a clinic visit, collection of medicine or a laboratory test performed beyond 12 weeks were assumed to be alive at week 12. Participants without electronic entries at or beyond 12 weeks were classified as lost to follow-up. In this study we describe and analyse the subset of patients with microbiologically confirmed TB, defined as participants with M. tuberculosis identified by culture or Xpert MTB/RIF from any clinical sample. RR-TB was defined as rifampicin resistance on any sample using either of the genotypic tests performed at the NHLS while rifampicin-susceptible TB (RS-TB) was defined as rifampicin susceptibility on all samples.

Statistical analyses
Data were analysed using R version 4.02 (R Core Team, Vienna, Austria).
Median values with interquartile ranges were used as measures of central tendency and dispersion. Categorical variables were described using counts and proportions and were compared using the Fisher's exact test. Continuous variables were compared between study groups using the Mann-Whitney U test.
To investigate the association of RR-TB with 12-week mortality, we hypothesised that RR-TB could cause mortality through two mechanisms: (1) delayed initiation of effective therapy and (2) lower efficacy of DR-TB therapy in preventing early mortality. We therefore treated receipt of appropriate anti-TB therapy (i.e. rifampicin-based therapy for RS-TB, any DR-TB therapy for RR-TB and an individualised regimen for extremely drug-resistant [XDR] TB) as a time-dependent variable in a Cox proportional hazards model, where hazard of mortality after start of appropriate therapy was compared to hazard of mortality prior to initiation of appropriate therapy, capturing effect of delay to appropriate treatment initiation. This allowed assessment of the association between RR-TB and mortality after adjusting for an effect of RR-TB on delay to appropriate treatment initiation.
We further hypothesised that measures of disease severity at baseline (hypoxia defined as peripheral oxygen saturation < 94% on room air, serum creatinine, sodium, protein gap [PG] [defined as the difference between total protein and albumin], haemoglobin, CD4 count, HIV viral load, Glasgow coma scale [GCS] < 15 and weight) could confound TB treatment status and mortality (with patients presenting more unwell being initiated on anti-TB therapy more urgently). Finally, we considered that patient factors associated with both RR-TB and mortality, including age and sex, and the above markers of baseline disease severity could confound the relationship between RR-TB and mortality. Both these sets of variables were therefore included as covariates in the model.
Continuous predictor variables were log-transformed to resolve highly skewed distributions, and because proportional or multiplicative changes in the values of these variables were thought to be more biologically meaningful. Peripheral oxygen saturation was dichotomised to reflect a non-linear relationship between oxygen saturation and partial pressure of oxygen as parsimoniously as possible.
The hypothesised causal structure is summarised in a directed acyclic graph ( Figure 1).
Observations on variables included in Cox regression were all > 95% complete except for weight which was missing in 8% of participants. All missing values were considered missing completely at random or missing at random after adjusting for other observed covariates and single-imputed for model fitting using classification and regression trees implemented with the mice package in R.
Participants lost to follow-up were censored on their last day of contact with health services. Kaplan-Meier plots were used to estimate survival over the study period, stratified by rifampicin-sensitivity status.

Ethical considerations
This study was approved by the University of Cape Town Human Research Ethics Committee (reference number 057/2013). Participants provided written informed consent where possible. The process involved in enrolling patients who did not have the capacity to provide informed consent has been described previously.  Table 2.

Time to treatment initiation and treatment status
Participants in the RS-TB group initiated appropriate treatment (rifampicin-based therapy for RS-TB, any DR-TB therapy for RR-TB and an individualised regimen for XDR-TB) a median of 1 day following enrolment (IQR: 1-2 days) compared to three days in the RR-TB group (IQR: 1-9 days), P < 0.001. Eight (1.9%) participants in the RS-TB group were not initiated on appropriate treatment during their admission, six of whom died. Six (11.3%) participants in the RR-TB group died prior to the initiation of appropriate treatment ( Figure 2). Four of the six participants died within three days of enrolment. In the two remaining participants, RR-TB was confirmed on blood culture -the results of which were only available after their deaths.

Mortality
Mortality at 12 weeks was 87/429 (20.3%) in the RS-TB group and 21/53 (39.6%) in the RR-TB group (P = 0.008). The Kaplan-Meier curve in Figure 3 demonstrates the relationship between rifampicin susceptibility and time-to-death.

Discussion
In this study we describe 482 patients with HIV-associated, microbiologically confirmed TB, diagnosed while admitted to a general hospital. We found a high prevalence of RR-TB in the study cohort and a high 12-week mortality in participants with RR-TB. Rifampicin-resistant TB was significantly associated with mortality, even when controlling for markers of disease severity and delay to treatment initiation.
The mortality rate we describe is significantly higher than described in other studies of patients with HIV-associated RR-TB. Unlike our study, these studies included participants both with and without HIV, longer follow-up periods and participants in predominantly outpatient or TB hospital settings. Prospective cohorts of outpatients with RR-TB, with follow-up periods ranging 6-24 months, describe mortality rates of 9% -31%, with HIV prevalence 38% -85%. 34,35,36,37 Retrospective cohorts of patients with RR-TB in TB hospitals describe mortality rates of 18% -29%, with HIV prevalence 52% -100%. 38,39,40 Retrospective studies of outpatients describe mortality rates of 13% -21%, with HIV prevalence 70% -100%. 10,12,13 In our cohort, 29% of participants with RR-TB who died did so prior to the initiation of a DR-TB regimen. This suggests that a substantial proportion of acutely ill patients with DR-TB die as inpatients, before a diagnosis of DR-TB can be confirmed, before DR-TB treatment regimens can be initiated and before they are registered on the national TB programme. These patients would contribute to the recognised gap  http://www.sajhivmed.org.za Open Access between numbers of DR-TB diagnoses and numbers registered in the DR-TB programme. 3 The authors hypothesised that the high mortality rates they observed could be driven by two, not necessarily mutually exclusive, mechanisms: delays to initiation of appropriate treatment for RR-TB and less effective therapies for RR-TB compared to RS-TB.
In this study, time to initiation of appropriate treatment was shorter than described in previous studies, 9,41 likely reflecting access to rapid diagnostics, multiple different samples taken for TB testing, and regular monitoring available in hospital. Although the difference in time to treatment initiation between the RS-TB and RR-TB groups was statistically significant, the difference is of unknown clinical significance. Several factors may have contributed to delays in appropriate treatment initiation in the RR-TB group including delays in specimen collection (patients in the RR-TB group may have been sicker at baseline and thus sputum sample acquisition for microbiological testing may have been more difficult), in awaiting the results of TB culture and drug susceptibility testing and in initiating appropriate treatment once these became available. Previous studies have failed to demonstrate a reduction in mortality in RR-TB with a reduction in time to treatment initiation. 9,42 In this study analysis, the effect of being on appropriate TB therapy on mortality during the follow-up period was unclear. This implies that the association between RR-TB and 12-week mortality observed in this cohort was not mediated by diagnostic delay. Despite the marginal difference in time to treatment initiation, the 12-week mortality in the RR-TB group was much higher. When   adjusting for TB treatment status as a time-dependent variable, RR-TB remained significantly associated with 12week mortality, suggesting that additional factors related to RR-TB, other than time to treatment initiation, were important. These additional factors could include residual confounding, but the finding is also consistent with lower efficacy of DR-TB regimens during the study period. Interestingly, the Kaplan-Meier survival curves for participants with RR-TB versus RS-TB ( Figure 3) diverge at day 21 of follow-up, suggesting early factors, including delays to treatment initiation, are less important contributors to differences in mortality than later factors, such as poorer efficacy of DR-TB treatment regimens. This study preceded the introduction of bedaquiline-based regimens for patients with RR-TB and participants in our study received RR-TB regimens containing injectable second-line drugs which are poorly tolerated and which have poor efficacy. 8 The introduction of bedaquiline-based regimens has resulted in promising improvements in survival in DR-TB; 17 however, prospective data in hospitalised patients at high risk of early mortality are lacking. This study suggests that efficacy of evolving DR-TB regimens in seriously ill patients with HIV-associated TB is an important topic for future research.
This study has several limitations. Although our cohort was relatively large, the population of participants with RR-TB was small. The authors did not objectively measure adherence to TB treatment once participants had been discharged, potentially introducing uncertainty into their measurement of the association between treatment status and mortality. Due to the observational design of this study, the authors were unable to eliminate residual confounding when measuring the association between RR-TB and mortality and they cannot therefore make conclusions as to the causal nature of this relationship. The strengths of this study include its extensive TB testing and prospective follow-up, with vital status at 12 weeks identified for 98% of participants. This study also describes a unique, under-studied population of hospitalised patients with HIV-associated RR-TB and CD4 count < 350 cell/μL.

Conclusion
This study describes a high 12-week mortality in patients admitted to hospital at the time of TB diagnosis with HIVassociated RR-TB. We suggest that delays in treatment initiation and poorly efficacious treatment regimens, prior to the introduction of bedaquiline as standard of care, are possible contributors to mortality in this population. Hospitalised patients with DR-TB represent an under-studied group with a high risk of early mortality. Research into the contributors to mortality in this population and improved diagnostic and therapeutic strategies are required.

Data availability
The data sets generated and analysed during the current study are available from the corresponding author, R.S., on reasonable request.