Paucibacillary Tuberculosis Drives the Low Positive Predictive Value of Xpert MTB/RIF Ultra for Rifampicin Resistance Detection in Low-Prevalence Settings

Abstract

Background

Xpert MTB/RIF Ultra (Ultra) aimed to improve the specificity in identifying rifampicin-resistant tuberculosis (RR-TB), compared to Xpert MTB/RIF.

Methods

In a nationwide study in Rwanda, patients diagnosed with RR-TB by Ultra between December 2021 and January 2024 underwent repeat Ultra testing, complemented by rpoB gene sequencing and phenotypic drug-susceptibility testing (pDST), serving as reference tests.

Results

Of 129 patients initially diagnosed with RR-TB by Ultra, only 41 (32%) had concordant rifampicin results upon repeat Ultra testing. The remaining 88 patients (68%) had unconfirmed resistance on repeat Ultra. Reference testing was available for 40 (98%) of 41 confirmed cases, all verified as true RR-TB. Among 88 unconfirmed cases, reference testing was available for 61 (69%), with 7 (11%) confirmed as true RR-TB, whereas 54 (89%) were found to have rifampicin-susceptible TB. Notably, 89% of 55 patients with very low bacillary loads on their initial Ultra had false RR-TB results, a significantly higher risk of false resistance compared to other bacillary load categories combined (risk ratio: 8.20; 95% confidence interval [CI]: 3.56–18.85; P < .001). Consequently, 53% (54/101) of initial RR patients with available reference testing received unnecessary RR-TB treatment.

Conclusions

Ultra represents a valuable tool for rapid RR-TB detection; however, in low prevalence settings its low positive predictive value for RR detection is largely driven by samples with very low bacillary loads. As programs expand active case-finding and early detection of asymptomatic disease, the proportion of TB detected with very low bacillary load will increase. Diagnostic algorithms require adjustments to prevent unnecessary RR-TB treatment.

Graphical Abstract

This graphical abstract is also available at Tidbit: https://tidbitapp.io/institutional-portal/clinical-infectious-diseases/tidbits/paucibacillary-tuberculosis-drives-the-low-positive-predictive-value-of-xpert-mtb-rif-ultra-for-rifampicin-resistance-detection-in-low-prevalence-settings-cf8df8bc-b811-4a72-b3b8-bef16fdedeab/update

positive predictive value (PPV), Xpert MTB/RIF ultra, false-positive, paucibacillary TB, TB case management

Advancements in molecular diagnostics have significantly transformed tuberculosis (TB) healthcare by enhancing early detection, drug resistance identification, and case management. Rifampicin (RIF), the cornerstone of first-line anti-TB therapy, is crucial for achieving relapse-free cures [1]. Rifampicin-resistant (RR)-TB is an indicator of multidrug resistance (MDR-TB), requiring the use of second-line anti-TB treatment. Accurate tests for RR are essential for appropriate patient management. The roll-out of Xpert MTB/RIF (Cepheid, USA) has been pivotal, offering rapid and robust TB diagnosis with simultaneous RR detection [2]. However, the assay's tendency to report false RR in paucibacillary samples had raised concerns [3–7]. Its successor, Xpert MTB/RIF Ultra (Ultra), demonstrates an increased sensitivity for TB detection at a modest specificity cost, although the use of sloppy molecular beacon probes with melting temperature (Tm) analysis is expected to mitigate the issue of false-RR detection [8, 9]. Prior to its global deployment in 2017, the early validation of Ultra in enriched clinicals sputum samples suggested an excellent 99.0% specificity in detecting RR, with no false RR reported in low bacterial load samples [8]. A multicenter diagnostic accuracy study across 8 countries demonstrated that Ultra had comparable sensitivity (95%) and specificity (98%) to Xpert for detecting RR [10]. These data and subsequent performance reports led to the recommendation of Ultra and its global adoption by national TB control programs (NTPs) to enhance TB detection while offering more reliable RR results [11, 12]. However, few post-implementation studies have validated Ultra's RR performance in programmatic settings, particularly in low RR-TB prevalence or systematic screening contexts, where a low positive predicted value (PPV) is expected [13].

Rwanda's NTP has gradually replaced Xpert MTB/RIF with Ultra in all 82 testing centers [14]. We conducted a prospective analysis to assess the performance of Ultra in detecting RR by repeating Ultra on a second specimen for all detected RR-TB patients, and resolving discrepancies between initial and subsequent Ultra results by a composite reference standard comprising of rpoB gene sequencing and phenotypic drug-susceptibility testing (pDST). We further explored factors contributing to false-RR results and the low PPV.

METHODS

Design and Study Population

In Rwanda, Ultra is the first-line diagnostic tool for all presumptive TB patients in Kigali. Outside Kigali, Ultra is used to test for RR among microscopically diagnosed TB patients, and for TB detection among patients categorized as high-risk and priority groups, such as people living with human immunodeficiency virus (HIV, PLHIV), children, contacts of TB patients, and prisoners [7].

A prospective nationwide diagnostic cohort study “Innovate to reduce RR-TB in Rwanda and beyond’ included all consecutive RR-TB patients, between 15 December 2021 and 18 January 2024 (Figure 1). In this sub-analysis, we included all RR-TB patients diagnosed by Ultra testing at peripheral health facilities. We extracted Ultra data, including semi-quantitative bacillary load from routine programmatic records and Xpert connectivity system.

Figure 1.

Flowchart showing initial rifampicin-resistance testing and subsequent rifampicin-resistance confirmation analysis. Abbreviations: MDR, multidrug-resistant tuberculosis; RIF, rifampicin; RR, rifampicin resistant; RS, rifampicin susceptible; Ultra, GeneXpert Ultra.

Sample Collection, Flow, and Testing Algorithm

Sputum samples were collected in prelabelled containers and stored at 2°C–8°C until tested locally or shipped to the nearest Ultra testing site, ensuring testing within 2 days per standard practice [7]. Patients diagnosed as RR-TB by initial Ultra were referred to an MDR-TB clinic in Kabutare, where 3 additional sputa were collected before MDR-TB treatment initiation: 1 on arrival (spot sample 1), 1 overnight (sputum continuously collected in the same container; sample 2), and 1 the next day (spot sample 3). Samples were sent in a cold-chain box to the National Reference Laboratory (NRL) in Kigali. Sample 1 was used for repeat Ultra, whereas samples 2 and 3 were processed for culture on Löwenstein-Jensen (LJ) and Mycobacteria Growth Indicator Tube-960 (MGIT, Becton Dickson, USA). Mycobacterium tuberculosis complex (MTBc) identification was confirmed for positive cultures by SD MPT64TBAg kit (SD Bioline, South Korea). All patients started a short 9-month MDR treatment regimen based on the initial RR-TB result, regardless of later results.

MTBc-positive cultures underwent pDST, on LJ (40 µg/mL RIF; 1 µg/mL isoniazid [INH]), or in MGIT (0.5 µg/mL RIF; 0.1 µg/mL INH). Sanger rpoB gene sequencing was performed on residuals of sputum or processed sediment of samples 2 or 3 for all participants [15]. If tests on the sputum failed, sequencing was conducted on the corresponding culture isolate when available. Similarly, the katG and fabG-inhA genes were sequenced to assess INH susceptibility.

On a subset of samples with positive culture, whole genome sequencing (WGS) was done on genomic DNA extracted from isolates, using Illumina Hiseq (CD Genomics, USA) [7]. For WGS data analysis, non-MTBc reads were removed using Centrifuge v1.0.3. MTBseq was employed for alignment with standard settings and TB-Profiler v6.2.1 was used to establish WGS-based resistance profiles [7].

We used a composite reference standard comprising pDST, Sanger rpoB sequencing, and/or rpoB reads from WGS, with any resistant result overriding: presence of any documented rpoB mutation associated with resistance and/or resistance on pDST confirmed RR-TB [16]. In contrast, absence of RR-associated mutations and sensitivity on pDST indicated rifampicin-susceptibility (RS), thereby categorizing the initial Ultra RR result as false RR. In absence of sequencing and pDST results, the RR status was deemed inconclusive.

INH resistance was confirmed by presence of associated mutations in katG and fabG-inhA, and/or phenotypic resistance, and/or Xpert MTB/XDR testing [16]. In absence of sequencing, pDST, and Xpert-XDR data, the INH-resistance status remained inconclusive.

Statistical Analysis

To investigate factors associated with discordance between initial and repeat Ultra tests and predictors of false RR we employed Poisson regression (RStudio version 2022.12.0). Key variables included initial Ultra bacillary load results (high, medium, low, and very low), history of TB treatment (new vs retreatment), HIV coinfection status (positive vs negative), gender, and age. Due to small observations, “high,” “medium,” and “low” bacillary loads were merged into 1 group, with the combined group coefficient reported as 1. Risk ratios with 95% confidence intervals (CIs) were calculated, with statistical significance determined by CIs excluding 1.0, alongside P-values < .05.

Ethical Consideration

The study protocol was approved by: Rwanda National Ethics Committee, Kigali, Rwanda (IRB00001497; Ref No705/RNEC/2021), Institutional Review Board of the Institute of Tropical Medicine, Antwerp, Belgium (IRB/AB/AC/157; Ref No.1525/21; 23/09/2021), and Ethics Committee of Antwerp University Hospital, Belgium (Ref No. B3002021000230; 22/11/2021). All participants provided signed informed consent.

RESULTS

Repeat Ultra Testing Could Not Confirm 68% of Initial RR Results

During the study period, 153 patients initiated RR-TB treatment throughout Rwanda (Figure 1). Among these, 129 (84%) were initially diagnosed with RR-TB by Ultra. Repeat Ultra testing showed concordance for RR-detection in only 32% (41/129) of samples, whereas RS-TB was detected in 46 (52%), 22 (25%) tested negative for MTB, 17 (19%) had trace MTB detected (thus no RIF result), and 3 (4%) had MTB detected with an indeterminate RR result.

Reference Methods Revealed False-RR Results

Of 129 patients, 101 (78%) had reference methods results, whereas such data were missing for 28 (22%) due to negative polymerase chain reaction (PCR) and/or culture results (Figure 1). Among 41 patients with confirmed RR on repeat Ultra, 40 (98%) had reference results, all confirmed as true RR-TB. Conversely, among 88 patients not confirmed as RR-TB on repeat Ultra, reference results were available for 61 (69%), identifying 7 (11%) as true RR-TB, whereas 54 (89%) were RS, indicating false RR by initial Ultra testing.

Of these 54 false-RR cases, most had multiple reference methods with concordant results (Supplementary Fig. 1). Overall, 42 (78%) had valid pDST results, all showing a RS result, including 17 with WGS wild-type rpoB sequences. None of the 38 sputum-based Sanger rpoB sequencing results showed relevant mutations: 34 had wild-type rpoB sequences and 4 showed heterogeneous mutations outside the rifampicin-resistance determining region (RRDR) not associated with resistance.

Among the 47 true-RR cases, RR was confirmed by presence of an RR-associated mutation in 45 (83%) by Sanger, and by pDST in 38 (90%), of which 15 with WGS data showed RR-associated mutations. Sanger sequencing failed to detect a rpoB mutation in the sputum of 2 patients, potentially heteroresistant below Sanger's detection limit, yet their isolates were found phenotypically resistant, and thus classified as RR. Among the 4 phenotypically RS isolates (all by MGIT), 1 had the Ser450Leu mutation, whereas 3 had the borderline Asp435Tyr mutation, which is more prone to be missed by MGIT-DST. In our context, Ultra demonstrated a PPV of 46.5% for the detection of RR, with a false discovery rate of 53% (CI: 42%–64%) [17].

Very Low Bacillary Load Drives False RR Results

Of the 129 patients with initial Ultra RR-TB detected, a higher proportion (n = 107; 83%) were male. The median age was 39 years (interquartile range [IQR]: 29–46). A minority (n = 21; 16.3%) of patients were HIV coinfected, whereas 35 (27.1%) were reported as previously treated TB patients (Table 1). One-third of the patients were diagnosed in Kigali (43/129; 33.3%) (Supplementary Table 1).

Table 1.

Associations Between Various Factors and the Non-confirmation of Initial Rifampicin-resistant Results on Repeat Ultra Testing, Highlighting Adjusted Odds Ratios, Confidence Intervals, and Statistical Significance

Variables	Patients (n = 129)	RR-TB Result Not Confirmed on Repeat Ultra	Univariable
			Risk Ratio (95% CI)	P value
HIV status, n (%)
Negative	108 (84%)	76 (70%)	1	1
Positive	21 (16%)	12 (57%)	0.81 (.55–1.20)	.30
Age, M (IQR)
Median	39 (29–46)	39.5 (28–51.25)	1.00 (1.00–1.01)	.28
TB history, n (%)
Re-treatment	35 (27%)	20 (57%)	1	1
New	94 (73%)	68 (72%)	0.79 (.58–1.08)	.14
Gender, n (%)
Female	22 (17%)	13 (59%)	1	1
Male	107 (83%)	75 (70%)	1.19 (.82–1.71)	.36
Initial Ultra semi-quantitative bacillary load, n (%)
High	22 (17%)	0 (0%)	1	1
Medium	9 (7%)	1 (11%)	1.22 (.13–11.85)	.86
Low	16 (12%)	6 (38%)	4.81 (1.15–20.18)	.031
Very low	82 (64%)	81 (99%)	10.46 (2.79–39.26)	<.001

Abbreviations: CI, confidence interval; HIV, human immunodeficiency virus; IQR, interquartile range; M, median; RR-TB, rifampicin resistance tuberculosis.

Discordant paired Ultra results (initial vs repeat) were significantly more likely in patients with low or very low bacillary loads compared to high loads (respectively, RR 4.81; 95% CI: 1.15–20.18; P = .031, and RR 10.46; 95% CI: 2.79–39.29; P < .001; Table 1).

A majority (91%) of 54 false-RR patients had a very low bacillary load by initial Ultra testing, whereas the remaining 5 (9%) had a low bacillary load (Table 2). Conversely, two-thirds of the 47 true-RR patients had a high or medium bacillary load by initial Ultra testing, whereas 10 (21%) had low and 6 (13%) had very low bacillary loads. Samples with a very low bacillary load had a significantly higher risk ratio of false-RR results on initial Ultra compared to samples from all other bacillary load categories combined (risk ratio 8.20; 95% CI: 3.56–18.85; P < .001). No significant association was found between false RR and history of previous TB treatment, HIV status, gender, or age. The geographical distribution by diagnostic center suggests that false RR results were dispersed, with no clear link to specific peripheral test sites or Xpert devices used (Supplementary Table 1, Supplementary Fig 2).

Table 2.

Predictors Contributing to False Rifampicin Resistance on Initial Ultra Assay for Patients With Confirmed Rifampicin Status Through Reference Methods, Highlighting Adjusted Odds Ratios, Confidence Intervals, and Statistical Significance

Variables	Patients (n = 101)	False Rifampicin Resistance, n (%)	Univariable
			Risk Ratio (95% CI)	P value
HIV status, n (%)
Negative	84 (83%)	46 (55%)	1	1
Positive	17 (17%)	8 (47%)	0.86 (.50–1.48)	.58
Age, M (IQR)
Median	38 (28–45)	37 (28–53.75)	1.00 (1.00–1.01)	.57
TB history, n (%)
New	76 (75%)	43 (57%)	1	1
Re-treatment	25 (25%)	11 (44%)	0.78 (.48–1.26)	.31
Gender, n (%)
Female	18 (18%)	7 (39%)	1	1
Male	83 (82%)	47 (57%)	1.47 (.79–2.67)	.23
Initial Ultra semi-quantitative bacillary load, n (%)
High	22 (22%)	0 (0%)	1	1
Medium	9 (9%)	0 (0%)
Low	15 (15%)	5 (33%)
Very low	55 (54%)	49 (89%)	8.20 (3.56–18.85)	<.001

Abbreviations: CI, confidence interval; HIV, human immunodeficiency virus; IQR, interquartile range; M, median.

Bacillary Load Among RR-TB Versus Non-RR-TB on Ultra

During the study period, 53 096 Ultra results were captured in the nationwide laboratory connectivity system. In total, 2979 (6%) had a positive MTB result with bacillary load above “Trace,” yielding an RIF result for 2.951 (99%) cases (Table 3). Bacillary load distribution differed significantly between initial RR cases and cases with RR not detected (P < .001), with a very low bacillary load being disproportionately higher among RR-TB compared to RS-TB cases.

Table 3.

Semi-quantitative Bacillary Load Distribution Among Samples With an “MTB Detected” Result Above “Trace” by Xpert MTB/RIF Ultra From December 2021 to January 2024, Captured in the Xpert Connectivity System

Bacillary load, n (%)	MTB Detected
	Rifampicin Resistance Detected	Rifampicin Resistance Not Detected
High	22 (17%)	1071 (38%)
Medium	9 (7%)	609 (22%)
Low	16 (12%)	675 (24%)
Very low	82 (64%)	467 (16%)
Total	129 (100%)	2822 (100%)

Abbreviation: MTB, Mycobacterium tuberculosis.

INH Susceptibility Testing May Aid in Ruling Out False RR in Rwanda

Among 47 patients with true RR-TB, 46 (98%) had concomitant INH resistance, including the 6 with a very low bacillary load. Only 1 (2%) exhibited susceptibility to INH (Supplementary Fig. 3). Conversely, among the 54 false RR-TB patients, 51 (94%) had INH-susceptibility results of whom 49 (96%) were INH susceptible, including 45 with very low bacillary load. Only 2 patients with false RR had INH-monoresistant TB. The PPV of INH in ruling out a false-RR diagnosis among patients with very low bacillary loads was 85% (CI: 42%–98%) with a negative predictive value (NPV) of 100% (CI: 92%–100%).

DISCUSSION

This population-based study showed a high false RR rate, driven by samples with very low bacterial loads (89%), mirroring the similarly low PPV seen with Xpert MTB/RIF [7]. In a low-RR-prevalence setting like Rwanda (1.3% in new cases), the expected PPV for Ultra is around 48%, similar to the PPV of 46.5% found in this study [13]. Importantly, these results highlight that the low PPV is mainly due to very low bacillary load samples, underscoring the need for diagnostic algorithms that account for bacillary load to improve RR-TB detection with Ultra. Although previous studies reported no false-RR results with Ultra, there are key differences with our study, including a larger RR-TB sample size, prospective design with fresh samples and specific focus on assessing the test's PPV for RR-TB detection [18, 19].

In this analysis, false-RR diagnosis resulted in unnecessary MDR-TB treatment for at least 54 patients. Even though patients had relapse-free cure, unnecessary MDR-TB treatment results in social, economic, physical, and emotional harms for affected patients, families and communities [20]. Moreover, it misallocates scarce healthcare resources. In Rwanda, the shorter MDR/RR-TB treatment regimen was implemented in 2014, yet treatment remains costly (∼12 000 USD per patient). In 2021, the World Health Organization (WHO) lowered the prevalence threshold for community screening with molecular tools from 1% to 0.5%, greatly expanding Ultra's use case [21]. The deployment of Ultra in increasingly low prevalence populations will increase the proportion of “very low” samples and hence, increase the risk of false RR. Increasing the threshold for RR calling to exclude cases with very low bacillary loads, similar to MTB “Trace,” where RIF susceptibility remains indeterminate, could be a relatively quick fix in the algorithm. This would urge NTPs to interpret Ultra RR-TB cautiously in patients with very low bacillary loads.

Updated WHO guidelines recommend repeat Ultra testing for patients at low RR risk, such as new TB patients with no RR-TB contact, while considering the initial Ultra result definitive for high-risk patients, such as those with prior TB treatment [13]. However, routine repeat testing is labor-intensive, costly, and challenging in TB-high-burden settings. Akin to Xpert MTB/RIF, a single repeat Ultra test was insufficient to confirm or rule out RR, particularly in paucibacillary samples. This study found that a repeat Ultra-confirmed RR was reliable, while an RS result and absence of a RIF result required further testing. Relying on a single repeat Ultra result would have misdiagnosed at least 15% (7/47) of true-RR cases, mostly paucibacillary, increasing the likehood of “MTB not detected” or “Trace” results. In contrast to the WHO guidance on pretest probability, the likelihood of identifying false RR did not differ between new and previously treated TB patients.

The technical reason for false-RR remains unclear. A prior case report described false-RR detection by Ultra in a lymph node aspirate with very low bacterial load [22]. Ultra's manufacturer attributed this to an unusual matrix effect, causing a bubble in the reaction tube that distorted derivative melt curves. In our study, fresh sputum, the standard matrix for TB diagnostics, was used. Mixed or evolving organism populations within patients have been suggested as an explanation for discordances in genotypic susceptibility testing; however, this is unlikely to account for patterns observed across the 54 false-RR cases. Notably, we found no evidence of heteroresistance in our data, and it seems unlikely that mixed populations would consistently be associated to very low bacillary load samples. Although end-users can consult Tm peak data and identify specific RRDR mutations, Tm values are not automatically captured by non-proprietary connectivity platforms, and melt curves are solely accessible to the manufacturer. In our study, manual Tm data analysis for a subset of samples showed expected Tm values for all probes in seven investigated true-RR cases, although Tm values of MUT probes did not match any predicted profile in 17 false-RR cases [23, 24]. Prolonged storage in Xpert's sample reagent (SR) buffer is known to affect the specificity of RR detection in paucibacillary samples, attributed to C-to-T or G-to-A substitutions from extended exposure to NaOH, a component of the SR [25]. Although our retrospective analysis could not assess individual SR incubation times, this factor alone is unlikely to explain such high rates of false-RR across sites. Post-study, we surveyed 53 health facilities, representing 64% of Ultra sites in Rwanda, revealing that 96% of studied facilities load specimens into cartridges within 8 hours after adding the SR, and none exceed the 12-hour limit recommended by Cepheid. Since 2018, all Xpert sites have participated in annual proficiency panel testing from Centers for Disease Control and Prevention (CDC) Atlanta, consistently achieving excellent results, with no major deviations indicating systemic issues such as cross-contamination, and none of the sites detected more than 1 RR-TB case on the same day.

Additionally, our findings demonstrated a high PPV for INH-susceptibility in identifying potential false RR among patients not confirmed by repeat Ultra, aligning with unpublished data from the Rwanda NTP showing relatively low INH- (3%) and RR- (2%) mono-resistance. To rapidly confirm INH resistance, robust molecular assays such as line probe assays or Xpert MTB/XDR can be directly applied to TB-confirmed specimens, with Xpert MTB/XDR being currently routinely used in Rwanda [13].

Based on these conclusions, an adjusted diagnostic flowchart has been implemented in Rwanda since May 2024, to investigate potential false RR in patients with very low bacillary loads by Ultra. Regardless of treatment history, patients undergo repeat Ultra, Xpert MTB/XDR to rule out INH resistance, and culture followed by pDST. While awaiting results, these patients receive first-line treatment with additional Ultra testing during treatment if clinical response is poor. Patients diagnosed with RR-TB by Ultra at low, medium, or high bacillary loads are referred for RR-TB treatment initiation, with a confirmatory Ultra repeat and Xpert MTB/XDR for INH and fluoroquinolone resistance.

Our study had several strengths. First, we utilized a large, nationwide population-based sample to investigate the frequency of and factors associated with false RR results by Ultra, making findings reflective of the Rwandan population and likely transferrable to other countries with similar characteristics: high TB notification rates, early TB diagnosis with universal Ultra testing, and low RR prevalence. Second, we employed robust reference standards, including Sanger sequencing, WGS, and pDST, enhancing the reliability of our findings.

Limitations in recovering sufficient MTB DNA for ropB sequencing might have resulted in underestimating the proportion of false RR among cases with very low bacillary load. Moreover, 18% of the 54 RS reference results were solely based on pDST, which may miss borderline resistance-conferring mutations or heteroresistance due to primary culture bias [26]. Our conservative composite reference standard may have led to a slight overcalling of resistance. Incomplete Xpert connectivity prevented full access to RS-TB Ultra data, hindering calculation and comparison of false-RR rates by the bacillary load as previously suggested [17]. Last, small sample size and zero count in some variables required category merging, potentially introducing sparse-data bias and masking finer distinctions in the data.

In conclusion, the low PPV of Ultra to detect RR in low prevalence settings is primarily driven by paucibacillary samples. We urge the manufacturer of the Ultra assay to revise the assay's algorithm to incorporate bacillary loads for RR determination. Given the global shift toward lower (RR-)TB prevalence settings or screening contexts in efforts to eradicate RR-TB, understanding and mitigating the factors contributing to low PPV will be important for optimizing diagnostic accuracy and improving TB patient management worldwide.

Supplementary Data

Supplementary materials are available at Clinical 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.

Notes

Author Contributions. I. C. M., B. C, d. J., L. R., and J. C. S, N., designed the study. I. C, M., F. H., D. R., H. N., J. K., W. B. d. R., W. M., Y. M. H., P. M., and J. C, S. N. participated in patients’ enrolment and data collection. I. C. M. and J. C. S. N. analyzed data and wrote the first draft of the manuscript. A. A., T. D., E. M. H. M., C. M. M., B. C. d. J., and L. R. critically revised the report. All authors approved the final version.

Acknowledgments. The authors thank the mycobacteriology section at the National Reference Laboratory and the Tuberculosis and Other Respiratory Diseases Divisions; and patients and staff in the multidrug-resistant tuberculosis clinics at Kabutare hospital for their contribution, and the Ministry of Health and Rwanda Biomedical Centre management for facilitation.

Financial support. This work was supported by the ITM's SOFI program supported by the Flemish Government - Department of Economy, Science & Innovation (DIR/av/2020/119), Belgisch Ontwikkelingsagentschap (DGD) under FA5-Rwanda Project (BE-BCE_KBO-0410057701-prg2022-13-RW), the Fonds Wetenschappelijk Onderzoek (FWO) (1SE7622N to I. C. M.), and Belgisch Ontwikkelingsagentschap (postdoctoral fellowship to J. C. S. N.). The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. I. C. M., L. R., and J. C. S. N. had full access to all data in the study and had final responsibility for the decision to submit for publication.

Author notes

L. R. and J. C. S. contributed equally to this work and share last authorship.

Potential conflicts of interest. The authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.