Failure or relapse predictors for the STREAM Stage 1 short regimen for RR-TB STREAM Sub-study Collaborators

B A C K G R O U N D : STREAM (Standardised Treatment Regimens of Anti-tuberculosis drugs for Multidrug-Resistant Tuberculosis) Stage 1 demonstrated non-inferi-or efficacy of a short regimen for rifampicin-resistant TB (RR-TB) compared to a long regimen as recommended by the WHO. The present paper analyses factors associated with a definite or probable failure or relapse (FoR) event in participants receiving the Short regimen. METHODS: This analysis is restricted to 253 participants allocated to the Short regimen and is based on the protocol-defined modified intention to treat (mITT) population. Multivariable Cox regression models were built using backwards elimination with an exit probability of P ¼ 0.157, equivalent to the Akaike Information Criterion, to identify factors independently

Patients with multidrug-resistant TB (MDR-TB) or rifampicin-resistant TB (RR-TB) have had poor outcomes for many years, as evidenced by the annual reports published by the WHO. Treatment success for MDR/RR-TB improved from 50% for the cohort starting treatment in 2012 to just 57% for the cohort starting treatment in 2017. 1 In view of the poor outcomes in patients receiving the long regimen of at least 20 months' duration recommended by the WHO, 2 Van Deun and colleagues conducted a series of observational cohort studies in Bangladesh to evaluate the possibility of developing shorter, less toxic and more effective regimens for the treatment of MDR-TB. Using a gatifloxacin (GFX) containing regimen of 9-11 months' duration, a relapse-free cure was achieved in 87.9% (95% confidence interval [CI] 82.7-91.6) of the patients enrolled. 3 Subsequently, a study in nine countries in West and Central Africa demonstrated that the 9-11 month regimen was effective in a population that included HIV coinfected patients. 4 In 2016, the WHO issued conditional recommendations, based on observational data, that patients with MDR/RR-TB who had not been previously treated with second-line drugs and in whom resistance to fluoroquinolones (FQs) and second-line injectable agents has been excluded, or is considered highly unlikely, could be treated with a MDR-TB regimen of 9-12 months instead of the conventional long regimen; the shortened regimen proposed was similar to that being tested in the STREAM (Standardised Treatment Regimens of Anti-tuberculosis drugs for Multidrug-Resistant Tuberculosis) Stage 1 study. 5 In 2019, the results of STREAM Stage 1, a multicountry non-inferiority randomised trial assessing the Short regimen in comparison with the Long WHO regimen were published. 6 The Short 9 month regimen was shown to be non-inferior to the Long (approximately 20 month) regimen using a composite endpoint of an unfavourable outcome, which included changes of treatment for adverse events, as well as death from any cause, treatment failure or relapse (FoR); this finding was subsequently confirmed using a solely TB-related outcome. 7 In view of these findings, it is important to identify which factors are associated with failure or relapse on the Short regimen.

METHODS
Data from the STREAM Stage 1 trial were used, as previously described. 6 The Union's Ethics Advisory Group and all national and local ethics committees approved the trial. All participants provided written informed consent; they were randomised in a ratio of 2:1 in favour of the Short regimen, which was identical to that used by Van Deun and colleagues in Bangladesh, 3 except that moxifloxacin was substituted for GFX because quality-assured GFX was not available. Randomisation was stratified by centre and HIV status. All participants were followed up to 132 weeks post-randomisation to determine the primary outcome. The pre-specified primary and secondary outcomes are published in the main report. 6 Variables included in the analysis The following characteristics were collected from participants at baseline: age, sex, weight, body mass index, presence of diabetes, HIV infection status, smoking status (current smoker, ex-smoker, nonsmoker) and haemoglobin level. Baseline sputum samples were collected to determine smear and culture quantified positivity as described in Table 1. Mycobacterium tuberculosis complex isolated in the local laboratories were subjected to further tests at the reference laboratory, the Institute of Tropical Medicine (ITM) in Antwerp, Belgium. These tests included determination of phenotypic resistance to isoniazid and resistance to pyrazinamide using genotyping. 6 A chest radiograph (CXR) obtained at baseline was read independently by two pulmonologists using methodology previously described. 8 CXRs with discordant readings between the two pulmonologists were assessed by a third, followed by discussion among the three readers aiming to achieve consensus. Assessments included extent of opacities, presence of cavitation and presence of costophrenic obliteration. To explore the predictive value of microbiological response to treatment in the intensive phase of the regimen the result of the Week 8 culture (positive or negative) and Week 16 smear result were also included in separate models.

Endpoint
To examine the factors associated with a TB-related outcome, the STREAM primary endpoint of unfav- ourable outcomes was re-classified on an individual basis according to the likelihood that they represented a FoR event on a 5-point Likert scale: definite, probable, possible, unlikely and highly unlikely. Further details on the methodology are available elsewhere. 7 In this analysis, the outcome of interest is whether or not the participant experienced a definite or probable FoR event in a time to event analysis. Participants not classified as experiencing a definite or probable FoR event were censored at the point they met the primary outcome criteria for unfavourable or unassessable, or at the end of follow-up (132 weeks post-randomisation) if they had a favourable outcome in the primary analysis. 6

Statistical analysis
All analyses were based on the protocol-defined modified intention to treat (mITT) population. This was defined as all randomised participants with a positive culture at baseline, except for those who 1) were randomised in error, or 2) had isolates taken before randomisation or up to Week 4 subsequently found to be susceptible to rifampicin or resistant to both aminoglycosides and FQs on a phenotypic DST at the central reference laboratory. Participants who were classified as unassessable in the STREAM primary analysis are included in these analyses. Thus, those who were lost to follow-up after 76 weeks but had culture converted when last seen and those whose treatment changed following proven reinfection with an exogenous strain of M. tuberculosis were included and were classified as unlikely to have had a FoR event. These participants were censored at the time last seen and at the time of change of regimen, respectively.
A multivariable Cox regression model was built using backwards elimination with an exit probability of P ¼ 0.157 equivalent to the Akaike Information Criterion. 9 HIV infection status, age and sex were considered a priori to be important confounding factors and were included in all models. All models were stratified by country and were restricted to individuals with complete data. The most prevalent category was set as the reference in all models. Landmark analyses, with time at risk starting at Week 8 and Week 16, were used to assess the association between Week 8 culture and Week 16 smear results and FoR events, respectively. A Kaplan-Meier graph was structured using the KMunicate project format. 10 Analyses were performed using Stata v16.1 (Stata-Corp, College Station, TX, USA).

RESULTS
A total of 253 participants were randomised to the Short regimen (Table 1). One third (85 participants) were co-infected with HIV, of whom 60 (70.6%) were on antiretroviral therapy (ART) at randomisation.
On the Short regimen, 25/253 (9.9%) participants had an FoR event, a Kaplan-Meier probability of an FoR event by Week 132 of 0.11 (95% CI 0.07-0.15; Figure). In the multivariable analysis of factors known at baseline, the factors remaining independently associated with an FoR event were male sex, a heavily positive baseline smear grade, HIV coinfection and the presence of costophrenic obliteration (Table 2). When the Week 8 culture result was included in the multivariable analysis, this was found to be associated with an FoR event, in addition to the four factors described above.
The third model included the smear result at Week 16 (the end of the intensive phase) and, in addition, the presence of diabetes at start of treatment or ever having smoked were significant but not the Week 8 culture.

DISCUSSION
The primary objective of STREAM Stage 1 was to assess, in people with rifampicin-resistant pulmonary TB who have no evidence of resistance to FQs or aminoglycosides on line-probe assays, whether the Short regimen developed in Bangladesh was noninferior to the Long regimen recommended by the WHO in 2011. 2 The composite primary endpoint included not only unfavourable treatment outcomes, such as failure during treatment or relapse, but also death from any cause, loss to follow-up and treatment changes for adverse events. An alternative outcome based on an assessment as to whether the event that met the primary endpoint definition was likely to be an FoR event on a five-point Likert scale, ranging from definite to highly unlikely, has been published. 7 The FoR rates reported here are similar to the failure/relapse rates reported in the West/Central Africa study report and the Bangladesh, Benin, Niger and Cameroon cohorts. 11,12 A comparison of observational studies of the effectiveness of the WHOrecommended shorter and longer regimens reported that the risk difference for relapse or failure was slightly higher with the shorter regimens. 13 The current paper has assessed the association of baseline factors and two microbiological assessments during the intensive phase with a FoR event in the Short regimen. In the multivariable analysis, there were five factors associated independently with a FoR event. Four of these were not unexpected: male sex, HIVpositive status, a heavily positive baseline smear and failure to culture convert by 8 weeks. The finding that is difficult to explain was the association of costophrenic obliteration with a poor outcome. We consider that this association is likely to be an incidental finding, although it also occurred in a similar analysis of the Long regimen (not reported here).
HIV-positive status was predictive of a poor treatment outcome; this is consistent with findings in several studies, [14][15][16] although in the West/Central Africa study all relapses occurred in HIV-negative patients and in the review of shorter regimens, the frequency of relapse or failure did not differ significantly by HIV status. 11,12 The use of ART improves treatment outcomes and reduces the risk of relapse in HIV-positive patients. [17][18][19] Moreover, advanced HIV infection and low CD4 count have been associated with poor treatment success. 18,[20][21][22] Antiretroviral therapy that is properly timed has been shown to improve treatment success and reduce mortality. 23,24 In the present study, 71% of the HIV-positive participants on the Short regimen were on ART at the time of initiation of MDR-TB treatment.
A systematic review and meta-analysis revealed treatment success was less likely among HIV-positive MDR-TB patients despite high ART uptake, which may be due to poor adherence and complications in management. 25 Initiation of an optimal ART regimen, and close monitoring for treatment adherence and ART failure can maximise the benefit of therapy. In a large retrospective South African cohort of patients with MDR-TB, opportunistic infections were reported as the main causes of death among HIV coinfected patients on ART. 26 There was, however, no evidence from the present study that those receiving ART on entry had better outcomes than HIV coinfected participants not on ART at that time. A study by Magis-Escurra et al. reported a low rate of relapse- free cure among HIV-positive patients despite culture conversion rates, which were not significantly different from those in HIV-negative patients. 27 Our findings suggest that HIV-positive patients treated with shorter regimens should be closely monitored for signs of treatment failure or relapse. Modification of the regimen may need to be considered in individual cases to reduce the risk of treatment failure or relapse.
Higher smear grade at baseline and culture positivity after 8 weeks of treatment predicted a poor treatment outcome, which is consistent with several reports. 14,28 A KwaZulu-Natal study reported culture conversion to be poorly predictive of a successful treatment outcome but an important predictor of survival among patients with resistance to rifampicin, FQs and aminoglycosides. 29 In the univariable analysis, cavitary disease and extent of opacities or advanced disease were radiologically predictive of FoR. This is consistent with several reports and may be reflective of a higher bacillary load. 15,21,30 Patients with extensive lung lesions may benefit from longer duration of treatment. A favourable radiological response may correlate with bacteriological and clinical response; 31 the role of CXR in treatment monitoring may need to be investigated further. In our study, CXRs were only taken at baseline and it was not possible to assess the radiological evolution of pulmonary lesions.
A patient-level pooled analysis of treatment-shortening regimens for drug-susceptible pulmonary TB reported patients with high smear grades and cavitation may require treatment durations of more than 6 months for a successful outcome. 32 Likewise, patients with RR-TB at risk for relapse receiving shorter regimens may benefit from tailored treatment durations. In the present study, clinicians were encouraged to consider extending the intensive phase of treatment for patients whose smears were slow to convert. This was not done consistently and consequently it was difficult to assess whether this approach was beneficial.
Smoking was associated with FoR in the third model; several studies have reported an association with poor treatment outcomes of TB and recurrent TB. 33 Unfortunately, we did not routinely collect information on alcohol use; it was therefore not possible to assess its association with FoR, although this may be associated with male sex and smoking. A meta-analysis and observational studies have reported alcohol use as a predictor of poor treatment outcomes in both drug-susceptible and MDR-TB. 14,28,34 There was some evidence in our study to suggest that participants with a low baseline body mass index (BMI) had a higher risk of FoR than those with a normal BMI, in the range 18.5-24.9 kg/m 2 ; an association between low BMI and poor outcomes has been shown elsewhere. 16,30,35 Failure to gain weight has also been shown to be predictive of an unfavourable response to treatment in underweight MDR-TB patients. 36 Further analyses of BMI evolution in STREAM patients are planned.
It is important to consider the value of adherence in preventing treatment failure and improving the relapse-free cure rate. Even minimal non-adherence and missed doses can adversely impact treatment outcomes. 32,37 Adherence support strategies, which can be achieved more readily in the context of a clinical trial, should also be implemented alongside the delivery of all TB treatment. It is likely that participants experiencing a severe adverse drug reaction may adhere less well to treatment and consequently have a less favourable outcome following treatment modification. 38 The use of bedaquiline, linezolid and later-generation FQs in regimens for MDR-and extensively drugresistant TB has been associated with improved treatment success. 19,39,40 Some studies have reported that the use of moxifloxacin as compared with GFX in shorter regimens was associated with FoR, and that GFX is more effective in overcoming low-level FQ resistance and in preventing acquired resistance during treatment. 11,41 The WHO recently recommended an all-oral shorter regimen with levofloxacin and bedaquiline as the core drugs for selected patients with MDR/RR-TB. 42 The inclusion of the most effective drugs such as bedaquiline and linezolid, has the potential to improve treatment outcomes of shorter regimens. The results of STREAM Stage 2 will further guide evidence-based recommendations on all-oral shorter regimens.
There are a number of limitations of the present analysis. These include the size of the study population and the inability to include certain factors which have been shown to be predictors in other studies, in particular alcohol consumption, adherence (for which the data from the present study were not considered to be sufficiently reliable to include) and CD4 counts in HIV-positive participants.
In conclusion, the results presented here have identified factors associated with FoR when patients with RR-TB are treated with the STREAM 9-11 month regimen. The role of the risk factors should be considered when monitoring response to treatment. It is important to note that resistance to FQs, which would be expected to have a poor outcome with this regimen, was largely excluded from this patient population. Such patients are likely to require the addition of one of the new generation of drugs in the regimen to ensure a successful outcome.
International Development (USAID) through the TREAT TB R É S U M É C O N T E X T E : L'essai clinique STREAM (Standardised Treatment Regimens of Anti-tuberculosis drugs for Multidrug-Resistant Tuberculosis) É tape 1 a démontré l'efficacité non inférieure d'un schéma court pour le traitement de la TB résistanteà la rifampicine (RR-TB) par rapport au schéma long recommandé par l'OMS. Cet article analyse les facteurs associésà une récidive où a unéchec (FoR) confirmé(e) ou probable chez les participants sous schéma court.