Reduction of diagnostic and treatment delays reduces rifampicin-resistant tuberculosis mortality in Rwanda

SETTING: In 2005, in response to the increasing prevalence of rifampicin-resistant tuberculosis (RR-TB) and poor treatment outcomes, Rwanda initiated the programmatic management of RR-TB, including expanded access to systematic rifampicin drug susceptibility testing (DST) and standardised treatment.OBJECTIVE: To describe trends in diagnostic and treatment delays and estimate their effect on RR-TB mortality.DESIGN: Retrospective analysis of individual-level data including 748 (85.4%) of 876 patients diagnosed with RR-TB notified to the World Health Organization between 1 July 2005 and 31 December 2016 in Rwanda. Logistic regression was used to estimate the effect of diagnostic and therapeutic delays on RR-TB mortality.RESULTS: Between 2006 and 2016, the median diagnostic delay significantly decreased from 88 days to 1 day, and the therapeutic delay from 76 days to 3 days. Simultaneously, RR-TB mortality significantly decreased from 30.8% in 2006 to 6.9% in 2016. Total delay in starting multidrug-resistant TB (MDR-TB) treatment of more than 100 days was associated with more than two-fold higher odds for dying. When delays were long, empirical RR-TB treatment initiation was associated with a lower mortality.CONCLUSION: The reduction of diagnostic and treatment delays reduced RR-TB mortality. We anticipate that universal testing for RR-TB, short diagnostic and therapeutic delays and effective standardised MDR-TB treatment will further decrease RR-TB mortality in Rwanda.

nostic delay significantly decreased from 88 days to 1 day, and the therapeutic delay from 76 days to 3 days. Simultaneously, RR-TB mortality significantly decreased from 30.8% in 2006 to 6.9% in 2016. Total delay in starting multidrug-resistant TB (MDR-TB) treatment of more than 100 days was associated with more than two-fold higher odds for dying. When delays were long, empirical RR-TB treatment initiation was associated with a lower mortality. C O N C L U S I O N : The reduction of diagnostic and treatment delays reduced RR-TB mortality. We anticipate that universal testing for RR-TB, short diagnostic and therapeutic delays and effective standardised MDR-TB treatment will further decrease RR-TB mortality in Rwanda. K E Y W O R D S : TB; Rwanda; MDR-TB programmatic management; MDR-TB diagnosis; MDR-TB treatment RIFAMPICIN (RMP) is the most powerful antituberculosis drug, and resistance to RMP is a reliable surrogate marker for multidrug-resistant tuberculosis (MDR-TB), a major challenge for TB management and control. 1 Rapid diagnostic tools, such as Xpert w MTB/RIF (Cepheid, Sunnyvale, CA, USA), have greatly increased access to RMP drug susceptibility testing (DST) and reduced diagnostic delay. 2 Globally, the vast majority (75%) of new RMPresistant TB (RR-TB) patients remained undiagnosed in 2017, 3 fuelling RR-TB transmission. 4 Logistical challenges related to sample transport and result reporting can cause diagnostic delays. 5 Other challenges, such as the lack of decentralised RR-TB treatment or access to RR-TB drugs, can cause therapeutic delays in those already diagnosed. In 2017, the global RR-TB treatment success rate was a mere 55%, 3 with diagnostic and/or therapeutic delays associated with poorer treatment outcome and pre-treatment loss to follow-up (LTFU). 2,[6][7][8][9][10][11][12][13] However, evidence on the effect of shortened delays on RR-TB mortality is lacking. 14 In Rwanda, the first patients with RR-TB were documented in 1989 at the Kigali University Hospital, Kigali, Rwanda. No standardised treatment regimen was available and little could be offered to these patients. 15 In 2005, the programmatic management of RR-TB (PMDT) was launched as a core component of the National TB Control Programme (NTP). 15 Countrywide access to RR-TB testing and MDR-TB treatment increased (Table 1). However, the effect of increased access and shortened delays on RR-TB mortality has never been investigated in Rwanda. We therefore studied data from all RR-TB patients since the start of PMDT in Rwanda-a study period of more than 10 years in order 1) to describe trends in RR-TB diagnosis and enrolment into MDR-TB treatment, 2) to compare diagnostic and therapeutic delays between different RMP DST methods and 3) to estimate the association between shortened delays and RR-TB mortality. To the best of our knowledge, this is the first such nationwide population-based study.

Design and study population
In this longitudinal retrospective analysis, we included consecutive patients diagnosed with pulmonary RR-TB who were registered between July 2005, when the PMDT started, and December 2016.

Data collection
Patients were assigned a unique ID on treatment initiation, or laboratory ID for those who did not start treatment. Patient files were retrieved from their respective health facilities. The National MDR-TB register and the National Reference Laboratory (NRL) registers were reviewed to extract relevant data.
A standardised data collection tool (see Supplementary Data) was used to capture demographics, baseline clinical characteristics, type of RMP DST used for diagnosis, dates of sample collection for RMP DST, dates that RMP DST results were available at the RR-TB testing laboratory, date of MDR-TB treatment initiation, the programmatic Table 1 Timeline showing changes to the programmatic management of MDR-TB in Rwanda   Period/year  Intervention  Implementation context   Before 2005  Passive surveillance of drug-resistant TB  Selected patients' samples (mostly those failing and relapsing from  the WHO Category 2 regimen) were shipped to collaborating  laboratory for DST, but no standard treatment for MDR/RR-TB  2005 Programmatic management of RR-TB and systematic surveillance of drug-resistant TB among retreated patients The programmatic management of RR-TB (PMDT) was launched as a core component of the NTP. 15  Expending access to RR-TB rapid diagnostic Xpert testing was decentralised to the district hospital level and became easily accessible to all 515 peripheral health facilities as first-line diagnostics for all those aged 55 years and presumptive TB patients with HIV coinfection, as well as all smear-positive TB patients. 15 The shorter MDR-TB treatment regimen was introduced. 36 All patients diagnosed with RR-TB were considered eligible for the shorter MDR-TB regimen (previously only those patients diagnosed with rifampicin and isoniazid resistance, i.e., MDR-TB, were eligible for the long regimen). Effective coordination between diagnostic centres and the Rwandan health facilities where TB care is provided was essential outcome (death before treatment initiation, death during treatment, treatment success-cure or treatment completion, treatment failure, treatment discontinuation and LTFU) and the outcome date. Two trained MDR-TB experienced nurses collected the data. They were supervised by the investigators. Data were double-entered in a dedicated EpiData database (EpiData Association, Odense, Denmark) by two encoders. Discordances were resolved by referring to the source.
Definition of variables RR-TB diagnostic delay was defined as the number of days between the date of collection of the first sputum sample that led to RR-TB diagnosis and the date that RMP DST results became available at the laboratory. RR-TB therapeutic delay was defined as the number of days between the date that RMP DST results became available at the laboratory and the date that MDR-TB treatment was started. Total RR-TB delay was defined as the sum of RR-TB diagnostic delay and RR-TB therapeutic delay.
A binary outcome ''mortality'' was constructed. Patients were categorised as ''dead'' if they died before or during MDR-TB treatment or as ''survived throughout treatment'', after excluding those with treatment failure or LTFU. In a sensitivity analysis, patients with treatment failure or LTFU were considered as dead (worst case scenario). It is plausible that the majority of these patients with MDR-/RR-TB and without (sufficient) treatment died.
Data analysis RR-TB incidence was calculated as an average of cases registered in three recent years (2014, 2015 and 2016) divided by the estimated population size (2012 Demographic and Health Survey), and was expressed per 100 000 population. The equality-of-medians test was used to compare the median delays within groups. RR-TB diagnostic delay and RR-TB therapeutic delay were categorised to facilitate logistic regression. Univariate analysis was conducted for each independent variable with mortality as the outcome variable. To assess the effect of both types of delay on mortality, adjusted for potential confounders, variables with P , 0.2 on univariate analysis were included, together with both delay variables in the multivariable logistic regression analysis. Statistical significance was set at 0.05. Kaplan-Meier techniques were used to estimate survival. Follow-up time started on the date of the first sputum sample collection that led to RR-TB diagnosis, the outcome event was death. Patients who survived throughout treatment were censored on the date of treatment outcome. STATA v14.2 (Stata Corp, College Station, TX, USA) was used for data analysis.

Ethics
The study protocol was approved by the Rwanda National Ethical committee ( Figure 1), 730 (92.7%) of whom were included in the primary RR-TB mortality analysis, while 39 (5.0%) were excluded from any analysis and 18 were only included in the sensitivity analysis.
Of the 730 eligible RR-TB patients, more were male (57.5%) and the median age was 34 years (interquartile range [IQR] 27-43). The human immunodeficiency virus (HIV) co-infection status was documented for 698 (95.6%), 291 (39.9%) of whom were HIV co-infected (Table 2); this remained stable over the study period (P ¼ 0.28, data not shown). The majority of the patients (n ¼ 390, 53.7%) came from Kigali City, with an estimated RR-TB incidence of 2.38 per 100 000 population per year, with the lowest incidence in Northern Province (0.21/ 100 000).

Patient treatment outcome and factors associated with death
Of the 730 patients included in the overall mortality analysis, 49 (6.7%) died before starting MDR-TB treatment, while 63 (8.6%) died during treatment (  (Figure 4). In a multivariable analysis, total delay of at least 100 days was independently associated with mortality (adjusted odds ratio [aOR] 2.45, 95%CI 1.35-4.46) ( Table 5). After including patients who failed treatment and those lost to follow-up as dead, and those with unknown status of HIV under the HIV-coinfected category, and excluding all RR-TB diagnosed with very low bacillary load (Mycobacterium tuberculosis, n ¼ 31) using Xpert in sensitivity analyses, a total delay of at least 100 days was still significantly associated with mortality (respectively aOR 2.82, 95%CI 1.60-4.97), aOR 3.71 (95%CI 2.09-6.58) and aOR 2.38 (95%CI 1.31-4.34) ( Table 6).
HIV co-infection (aOR 2.27, 95%CI 1.35-3.82) and age (.54 years; aOR 4.83, 95%CI 2.10-11.10) remained associated with increased mortality (Table  5). Other variables such as sex, MDR-TB treatment clinic and MDR-TB treatment regimen were not significantly associated with mortality (Table 5). Figure 5 shows that patients with a shorter diagnostic delay and those who were HIV-negative were more likely to survive (P , 0.001 and P¼ 0.01, respectively).

DISCUSSION
To the best of our knowledge, this is the first nationwide RR-TB population-based study to show that a shortened diagnostic delay is associated with a decline in mortality. In Rwanda, RR-TB mortality dropped from 30.8% in 2006 to 6.9% in 2016. Mortality was more than two-fold higher in patients with a total delay of 100 days than in those with a delay of ,35 days. Delays in diagnosing and treating RR-TB have been reduced through the nationwide scale-up of rapid molecular RMP DST since 2014 (Table 1), when 100% of RR-TB patients were diagnosed using molecular diagnostics, either Xpert (81.3%) or LPA (18.7%).
Our study showed that MDR/RR-TB mortality decreased as delays were reduced, mainly due to the implementation of Xpert. This findings contrasts with some previous studies which did not show an effect of the use of Xpert on TB mortality. 18,19 On the other hand, our findings complement those from a Peruvian study conducted in patients already started  on MDR-TB treatment, which showed a decreased odds (0.5) of death and increased odds (1.4) of treatment success associated with the implementation of rapid phenotypic (microscopic-observation drug susceptibility) and genotypic (LPA) DST. 13 Consistent with findings from other studies, 13 HIV coinfection and older age were associated with mortality, although HIV-coinfected and older patients were diagnosed more rapidly than HIVnegative and younger patients.  Although the clinical decision to start MDR-TB treatment before laboratory confirmation is challenging, it resulted in a lower mortality during the period that rapid molecular DST was not available and when diagnostic delays were very long. 7 This should not be surprising. In patients with multiple risk factors, a susceptible RMP DST result is unlikely to lower the post-test probability below the treatment threshold, the minimal level of probability of having RR-TB that is required to start MDR-TB treatment. 20 Hence, empirical RR-TB treatment seems justified in such situations, especially when an effective treatment is available and when delay is associated with mortality. 21 In our study, the treatment success rate was relatively high (84.7%). 13 The high rate of treatment success is in line with low rates of resistance to second-line drugs such as fluoroquinolones 22,23 and second-line injectables. None of the over 400 RR-TB  patients routinely tested since 2010 displayed resistance to these drugs (NRL Rwanda, unpublished findings), confirming that the standardised MDR-TB treatment used was effective in curing RR-TB and halting its spread. In fact, before the implementation of PMDT in Rwanda, the estimated prevalence of RR-TB among new TB patients rose from 1.3% (95%CI 0.7-2.1) in 1993 to 3.9% (95%CI 2.5-5.7) in 2005, 24,25 likely fuelled by ongoing transmission. 26 Before RR-TB testing became easily accessible, the majority of the RR-TB patients diagnosed before 2013 had a primary RR-TB strain, but received multiple rounds of ineffective RMP-based treatment, while spreading RR-TB in their communities. This probably explains the increase of RR-TB prevalence among new TB patients, as shown in 2005. 25 Since the implementation of PMDT, drug resistance surveys revealed a statistically significant decline in RR-TB prevalence among new TB patients, from 3.9% (95%CI 2.5-5.7) in 2005 to 1.4% (95%CI 0.7-2.1) in 2015. 25,27 Thus, the combination of early RR-TB diagnosis with effective MDR-TB treatment likely reduced transmission and probably explains the decrease in RR-TB prevalence, as observed in the 2015 survey. 27 Nonetheless, the programme should be cautious of sole dependence on rapid molecular testing because 1) a shortage of reagents (e.g., cartridges) could completely paralyse the diagnostic system, 28,29 2) the increased sensitivity of the newly developed Xpert w MTB/RIF Ultra (Cepheid) may complicate its use in detecting true treatment failures or relapses, as persistent DNA may not reflect active disease, 30 3) commercial rapid molecular diagnostics miss RMP resistance-conferring mutations outside the 81 base pair RMP resistance-determining region (RRDR), such as Val170Phe and Ile491Phe, 31,32 and 4) falsepositive RR-TB results on Xpert may be associated with low bacillary load in the sample. 33 However, a sensitivity analysis which excluded all RR-TB identified using Xpert with very low bacillary load did not affect the interpretation of the final model.
Our study had several important strengths. First, 85.4% of all patients diagnosed with RR-TB in Rwanda over a period of more than 10 years were included in the study. Our findings thus represent the current reality of the MDR/RR-TB programme in Rwanda. Second, in contrast with most studies on RR-TB outcomes, data from patients who died before starting treatment were included. Survival bias was therefore limited. Third, we validated data by comparing data collected from different sources (patient files, the national MDR-TB register and NRL registers). When discrepancies were identified, sources were consulted a second time. Finally, a sensitivity analysis including those who failed treatment or were lost to follow-up, was used to confirm the findings of the primary analysis.
Our study also had some important limitations. We did not collect data on the delay between the day a patient became at risk for MDR-/RR-TB and the date RMP DST was requested. In addition, we did not collect data on the severity of TB disease, such as extensiveness of TB on X-ray, the bacterial load in diagnostics or patient body mass index. Moreover, second-line drug susceptibility was not tested systematically and could not be taken into account. However, it should be noted that resistance to second-line drugs is extremely rare in Rwanda. Another important limitation was the lack of systematic data on timing of antiretroviral therapy initiation. The strong association between HIVcoinfection and RR-TB mortality could thus not be explored further. Finally, as this was a retrospective study, we could only adjust for those variables that were routinely collected.
In conclusion, diagnostic and treatment delays were strongly associated with RR-TB mortality in Rwanda. As the NTP were able to control delays, mortality declined from 30.8% in 2006 to 6.9% in 2016. Other factors that likely contributed to improved MDR-/RR-TB outcomes were universal RMP resistance surveillance enhanced by performance-based financing and the implementation of effective standardised MDR-TB treatment. We anticipate that universal testing of all TB patients for RR-TB, short diagnostic and therapeutic delays, and effective MDR-TB treatment will further reduce RR-TB prevalence in Rwanda.

R É S U M É
C O N T E X T E : En 2005, en réponseà une prévalence croissante de la tuberculose résistanteà la rifampicine (RR-TB) et aux résultats médiocres du traitement, le Rwanda a initié la gestion programmatique de la RR-TB, notamment l'expansion de l'accès systématique au test de pharmacosensibilité de la rifampicine et au traitement standardisé. Parallèlement, la mortalité liéeà la RR-TB a significativement diminué de 30,8% en 2006à 6,9% en 2016. Un retard de traitement de la TB multirésistante (MDR-TB) de plus de 100 jours aété associéà un risque plus de deux fois supérieur de décès. Quand les délais ontété longs, la mise en oeuvre d'un traitement empirique de la RR-TB aété associéeà une diminution de la mortalité. C O N C L U S I O N : La diminution du retard au diagnostic et au traitement a réduit la mortalité de la RR-TB. Nous nous attendonsà ce que le test systématique de RR-TB, un délai court de diagnostic et de traitement et un traitement standardisé efficace de la MDR-TB diminue encore la mortalité de la RR-TB.

R E S U M E N M A R C O D E R E F E R E N C I A:
En el 2005, en respuesta a un aumento en la prevalencia de la tuberculosis resistente a rifampicina (RR-TB) y los desenlaces terapéuticos desfavorables, se inició en Rwanda el manejo programá tico de la RR-TB, que incluía la expansión del acceso a las pruebas sistemá ticas de sensibilidad a rifampicina y al tratamiento normalizado. días a un día y la mediana del retraso terapéutico pasó de 76 a 3 días. De manera simultá nea, se observó una importante disminución de la mortalidad por RR-TB, de 30,8% en el 2006 a 6,9% en el 2016. Un retraso total del comienzo del tratamiento de la tuberculosis multirresistente (MDR-TB) superior a 100 días se asoció con una posibilidad de mortalidad superior al doble. Donde los retrasos eran prolongados, el comienzo empírico del tratamiento contra la RR-TB se asoció con una menor mortalidad. C O N C L U S I Ó N: La disminución de los retrasos en el diagnóstico y el tratamiento disminuyó la mortalidad por RR-TB. Se propone que la realización de la prueba de resistencia a rifampicina a todos los pacientes, un corto lapso hasta el diagnóstico y el comienzo del tratamiento y un tratamiento eficaz normalizado de la MDR-TB disminuirán aún má s la mortalidad por RR-TB en Rwanda.