Prevalence, Predictors, and Successful Treatment Outcomes of Xpert MTB/RIF–identified Rifampicin-resistant Tuberculosis in Post-conflict Eastern Democratic Republic of the Congo, 2012–2017: A Retrospective Province-Wide Cohort Study

Abstract Background Multidrug-resistant tuberculosis (MDR-TB) jeopardizes global TB control. The prevalence and predictors of Rifampicin-resistant (RR) TB, a proxy for MDR-TB, and the treatment outcomes with standard and shortened regimens have not been assessed in post-conflict regions, such as the South Kivu province in the eastern Democratic Republic of the Congo (DRC). We aimed to fill this knowledge gap and to inform the DRC National TB Program. Methods of adults and children evaluated for pulmonary TB by sputum smear microscopy and Xpert MTB/RIF (Xpert) from February 2012 to June 2017. Multivariable logistic regression, Kaplan–Meier estimates, and multivariable Cox regression were used to assess independent predictors of RR-TB and treatment failure/death. Results Of 1535 patients Xpert-positive for TB, 11% had RR-TB. Independent predictors of RR-TB were a positive sputum smear (adjusted odds ratio [aOR] 2.42, 95% confidence interval [CI] 1.63–3.59), retreatment of TB (aOR 4.92, 95% CI 2.31–10.45), and one or more prior TB episodes (aOR 1.77 per episode, 95% CI 1.01–3.10). Over 45% of RR-TB patients had no prior TB history or treatment. The median time from Xpert diagnosis to RR-TB treatment initiation was 12 days (interquartile range 3–60.2). Cures were achieved in 30/36 (83%) and 84/114 (74%) of patients on 9- vs 20/24-month MDR-TB regimens, respectively (P = .06). Predictors of treatment failure/death were the absence of directly observed therapy (DOT; adjusted hazard ratio [aHR] 2.77, 95% CI 1.2–6.66) and any serious adverse drug event (aHR 4.28, 95% CI 1.88–9.71). Conclusions Favorable RR-TB cure rates are achievable in this post-conflict setting with a high RR-TB prevalence. An expanded Xpert scale-up; the prompt initiation of shorter, safer, highly effective MDR-TB regimens; and treatment adherence support are critically needed to optimize outcomes.

Multidrug-resistant tuberculosis (MDR-TB), defined as resistance to rifampicin and isoniazid, poses a global threat to TB control efforts [1,2]. In contrast to downward trends in 30 high-burden countries, the Democratic Republic of the Congo (DRC) has had persistently high TB incidences and mortality rates [2]. In 2017, among the 81 million people in DRC, there was an estimated TB incidence of 322/100 000 and an estimated 60 000 TB-related deaths [2]. Estimated DRC MDR-TB rates were 2.2% among new cases and 9.7% in retreatment cases [2]. Limited laboratory infrastructure for both MTB cultures and drug susceptibility testing (DST) makes the determination of the actual MDR-TB burden and of treatment outcomes challenging [3,4].
The second Congo war, in the aftermath of the 1994 Rwandan genocide, caused a massive disruption of healthcare services in the eastern DRC [5]. In 2012, South Kivu was the first province of eastern DRC to roll out the Xpert MTB/ RIF assay (Xpert; Cepheid, Sunnyvale, CA), a rapid genotypic assay for rifampicin susceptibility via rpoB mutation detection [6][7][8]. Laboratory capacities for Xpert were established at 10 TB diagnostic and treatment centres (CSDTs) in South Kivu province through the Stop TB Partnership's TB REACH initiative [9]. Creswell et al reported a 5.3% prevalence of Xpert-diagnosed rifampicin-resistant TB (RR-TB) in South Kivu [10], compared to a 9.6% prevalence among smear-negative TB cases in the capital city Kinshasa, as reported by Mbonze et al [11]. Both studies focused on strategies to improve finding TB cases and on early Xpert user experiences, but did not document the risk factors for, or clinical outcomes of, RR-TB treatment.
We aimed to assess the prevalence and predictors of RR-TB, as well as treatment outcomes, to inform the strategic plan of the DRC National TB Program (NTP) and guide models development to improve MDR-TB care in low-and middle-income countries (LMICs).

Study Design, Setting, and Population
We performed a secondary analysis of data collected routinely from adults and children and evaluated for suspected pulmonary TB at 10 Xpert-equipped CSDTs serving 34 health zones of the South Kivu province from 1 February 2012 to 30 June 2017 ( Figure 1). While described as post-conflict, the eastern DRC is still considered unstable, given the continued presence of multiple active militias [5]. TB cases were found and identified either passively (by referral) or actively (via symptom screening by community health workers [CHWs]). Ziehl-Neelsen sputum smear microscopy for acid-fast bacilli (AFB) and Xpert were performed on different sputum samples. Human immunodeficiency virus (HIV) status was not documented systematically, due to out-of-stock test kits.
Patient responses to MDR-TB treatment were monitored monthly during the intensive phase and every 2 months during the continuation phase, with a sputum smear for AFB (if positive at baseline) and/or mycobacterial culture most frequently done on traditional Löwenstein-Jensen solid media. The BACTEC Mycobacteria Growth Indicator Tube liquid cultures (Becton Dickinson, Sparks, MD) were performed only occasionally, due to their relatively high cost. Monitoring for drug toxicity included baseline and follow-up clinical assessments; electrolyte, liver (if symptomatic), renal, and thyroid function tests; and visual acuity and audiometric assessments, per DRC NTP guidelines [12,14]. MDR-TB patients were not hospitalized unless medically indicated. Directly observed therapy (DOT) was conducted by a nurse at the nearest CSDT, either daily during the intensive phase or weekly during the continuation phase; on the remaining days of the week, home-based DOT was performed by a trained CHW or an assigned family member. Of note, patients were provided incentives-food parcels, for nutritional support, and transportation subsidies (US $30/per month)-to facilitate their daily or weekly attendance at CSDTs for DOT [12].

Treatment Outcomes Definitions
WHO definitions were used for MDR-TB treatment outcomes [13]. Cured indicates a patient who has completed treatment and has at least 3 consecutive, negative sputum smears for AFB or MTB cultures, with at least 30 days in between. Treatment completion indicates a patient who has completed treatment, but does not have bacteriologic confirmation. Death indicates a patient who dies for any reason during the course of treatment. Failure indicates a patient with (1) a positive sputum smear for AFB or MTB culture at 5 months or later after initiation of MDR-TB regimen, or (2) a bacteriological reversion to positive after conversion. Treatment success includes patients with cures or treatment completion, while unsuccessful treatment includes patients who died, failed treatment, or relapsed. LTFU refers to a patient lost to follow-up for 2 or more consecutive months. Of note, the WHO benchmarks for treatment successes and cures among TB patients, regardless of their drug resistance statuses, are ≥90% and >80%, respectively [2].

Statistical Analysis
Data were summarized using proportions and medians (interquartile range [IQR]) for categorical and continuous variables, respectively. Chi-square tests and Wilcoxon rank sum tests were applied for tests of association, where appropriate. Using logistic regression models, we estimated the unadjusted association between individual, baseline characteristics and RR-TB. We estimated adjusted associations by including all baseline covariates a priori in a multivariate model. Starting with a full model, we then used a backward elimination procedure, excluding predictor variables with a P value < .1, and compared the estimated reduced model adjusted odds ratios (aOR) and associated 95% confidence intervals (CIs) to the full multivariate model estimates. The final model was based on variables with statistical significance in both the full and reduced models, plus variables known to have clinical significance. Patients who were LTFU were censored at the date of last appointment. Treatment failure was summarized using Kaplan-Meier estimates, and TB treatment survival was compared between regimens using log-rank tests. Univariable and multivariable Cox regression modelling identified those predictors independently associated with the hazard of treatment success or failure. Adjusted hazard ratios (aHR) and their associated 95% CIs were calculated to summarize the strength of associations between patient characteristics and treatment failure. For each baseline characteristic, we evaluated the proportional hazards assumption by checking for parallel lines in plots of -log{-log(survival)} versus log(time) for each category. For the final multivariate models, we performed a universal test of proportional hazards based on Schoenfeld residuals. All P values that we report are exact and two-tailed; a value less than .05 was considered statistically significant. Stata software version 13.1 (Stata, College Station, TX) was used for all statistical analyses.

Ethics Approvals
This study was approved by the Health Research Ethics Committees of Stellenbosch University, Cape Town, South Africa (reference number S15/03/059) and the Université Catholique de Bukavu's Referral Provincial Hospital, Bukavu, South Kivu Province, DRC (reference number 25.042/946/ Staff/LK/HPGRB/BKV/014). Figure 2 displays the patient flow for all 10 CSDTs during the study period: 16 488 patients with presumptive pulmonary TB were screened by Xpert, capturing over 90% of those diagnosed with TB during the study period from these CSDTs. Their age range was 1-98 years, their median age was 35 years (IQR 24-52), and 55.2% were male. Ultimately, 1535 (9.3%) patients were confirmed by Xpert or culture to have TB. The median time from a positive Xpert result to TB treatment initiation was significantly longer for RR-TB vs RS-TB patients (12. Table 1).

DISCUSSION
We observed a high prevalence (11.1%) of RR-TB among nearly 1600 adults and children with Xpert-positive MTB in the South Kivu province. Independent predictors of RR-TB were sputum smear positivity and retreatment TB, and 46.5% of RR-TB patients had no prior TB episode. High proportions of RR-TB patients on 9-(83%) and 20/24-month (74%) MDR-TB regimens were cured (P = .06). Independent predictors of treatment failure/death while on an RR-TB regimen were the lack of DOT and the occurrence of any SAE.
Our data are consistent with those of Mbonze et al, who found a high prevalence (9.6%) of RR-TB among 991 samples from Kinshasa [11]. Creswell et al, however, in their TB REACH report analyzing data from three South Kivu sites, found that only 5.3% (n = 567) of samples that tested positive by Xpert were RR-TB [10]. Factors accounting for the differences between our study and that of Creswell include our larger sample size and coverage area, the longer observation time, and the greater number of retreatment cases tested.
Patients with RR-TB were more likely to have positive sputum smears for AFB and to be retreatment TB cases. The latter situation could be a consequence of the prior lack of DST, due to limited laboratory capacities and leading to empiric RS-TB treatment. Inadequate treatment leads to progressive TB and the presence of persistently sputum smear-positive individuals in the community, with increased TB transmission as strains have more opportunities to mitigate fitness costs through compensatory mutations [15,16]. We also found that even one prior TB treatment episode was associated with a roughly 80% increase in the odds of acquiring RR-TB, which is in agreement with several other reports from LMICs [17][18][19].
Importantly, we documented the evidence for transmitted RR-TB, since over 40% of those with RR-TB had no prior TB episode (Table 1). This is in line with published empiric and modelling data suggesting that, in high TB burden settings, many cases of RR-TB result from the transmission of drug-resistant strains of TB, rather than from acquired resistance [2,20,21]. Therefore, improving the diagnosis, prevention, and treatment strategies of both RS-and RR-TB cases, including access to genotypic DST and shorter regimens, is critical [22].
Of concern, the median time from a positive Xpert result to treatment initiation was delayed to nearly 60 days for more than 25% of patients with RR-TB, due to out-of-stock drugs and supply chain issues of second-line drugs (SLDs), as they need to be transported from the capital city of Kinshasa to the remote CSDTs. To maximally benefit from Xpert testing, these important logistical and structural challenges must be addressed by the DRC's NTP and stakeholders.
We did not find any significant association between HIV infection and RR-TB prevalence, treatment failure, or mortality; however, our study had limited statistical power for such associations, most likely due to the low HIV prevalence (<5%) in the South Kivu province and/or the 18.8% of RR-TB patients not tested for HIV. While there are conflicting reports about associations between HIV and MDR-TB prevalence [23][24][25], the negative impact of HIV on MDR mortality is well established [26]. Indeed, in a meta-analysis including 2725 HIVinfected adults and children with MDR-TB, Isaakidis et al found that only 57% were treated successfully, and the mortality among HIV-infected adults was 4 times higher than that of HIV-negative populations (38% vs 11%), while mortality among children was twice as high (11.5% vs 6%) [26].
We documented similar programmatic treatment success with 9-and 20/24-month MDR-TB regimens. These results are in line with those reported from observational studies in Bangladesh and sub-Saharan Africa [27][28][29][30][31][32][33], as well as the interim results of the standardised treatment regimen of anti-TB drugs for patients with MDR-TB (STREAM) Stage 1 trial, which showed that both the shorter and longer MDR-TB regimens achieved success in roughly 80% of participants [34,35]. The documented 0.0% MTB resistance to Mfx and Km in our setting may account, in part, for this favorable outcome. Also, our data, similar to those of others, show that SAEs are associated with unsuccessful MDR-TB outcomes [36]. Indeed, prevalent SAEs are associated with the toxicity of second-line injectable drugs: that is, renal failure or ototoxicity associated with Km. In the STREAM Stage 2 trial, second-line injectable drugs were replaced by a fully oral, bedaquiline (BDQ)-containing regimen, one of the two new (along with delamanid [DLM]) potent MDR-TB drugs [37,38] that are now being prescribed increasingly in LMICs following compassionate pilot access programs [39]. In August 2018, the WHO recommended that initiation of the shorter regimen be conditional to: (1) availability of DST, to exclude resistance to at least Mfx and Km, and (2) close  monitoring of the patient's safety, as well as their treatment response [13,35]. Switching non-responding patients (or those experiencing SAEs) to alternative SLDs and/or new regimens needs to be based on the new regrouping of MDR-TB drugs, ideally all oral, including BDQ or DLM [35]. Finally, we found that treatment under DOT was significantly associated with a decreased hazard of death/failure among RR-TB patients, as also reported in the meta-analysis by Ahuja et al [40]. However, it is likely that DOT is facilitated by socioeconomic incentives that, in turn, minimize LTFU and optimize patient outcomes, as also reported by Charles et al in post-earthquake Port-au-Prince, Haiti [41]. Of note, a meta-analysis by Richterman et al showed that patients receiving cash incentives during treatment for active RS-TB experienced improved clinical outcomes [42], but an assessment of the efficacy and cost-effectiveness of socioeconomic incentives for MDR-TB patients is needed.
Our data have important clinical/public health implications and can inform models of MDR-TB care in LMICs, including: (1) the scale-up of Xpert and genotype DST to identify earlier cases of MDR-TB among new and retreatment cases; (2) investments in socioeconomic incentives and training CHWs to deliver home-based DOT; (3) the integration of TB/HIV testing and treatment services; and (4) the need for household-based MDR-TB prevention strategies. However, confirmation of the impact of some of the above interventions will require prospective, interventional studies and cost-effectiveness analyses.
Our study has several strengths: (1) a detailed analysis of a large sample of Xpert-diagnosed TB cases in adults and children recruited from multiple TB diagnostic and treatment centers across the South Kivu province; (2) cohort data from a post-conflict setting, with multiple barriers to health care providing both feasibility data and MDR-TB treatment outcomes data relevant to TB clinicians and NTP implementers, as well Our study also has several limitations: (1) the retrospective, cohort study design is subject to unmeasured confounding factors that were not adjusted for and to the possibility of recall bias or data capture errors, leading to the misclassification of information on prior TB episodes/treatment; (2) extended genotypic DST of other key first-line drugs or SLDs was not routinely affordable/available in this setting; (3) a limited number of RR-TB patients were treated with the 9-month MDR-TB treatment regimen; and (4) the unknown HIV status of 18.8% of RR-TB patients, due to the unavailability of HIV test kits. Despite these shortcomings, our study provides important, new, and significant findings about RR-TB epidemiology and treatment outcomes in a challenging setting.

CONCLUSIONS
RR-TB was present in 11.1% of patients with confirmed TB in post-conflict DRC and was more frequent among patients with a positive sputum smear and retreatment TB. More than 45% of those with RR-TB had no prior TB episode, which highlights an urgent need for novel and effective household MDR-TB prevention strategies. Finally, we demonstrate that successful MDR-TB treatment is achievable in approximately 80% of patients in this setting. Scaling up the use of Xpert; the timelier initiation of shorter, safer, highly potent MDR-TB regimens; and consistent treatment adherence support are needed to optimize treatment outcomes.  results, writing of the manuscript, and revisions. R. M. helped with the data analysis, drafting of the manuscript, and revisions. Z. M. K. advised on the clinical and programmatic aspects of the manuscript. All authors reviewed the paper for additional interpretation, editing, and revisions and approved the final version.