Study 38/CRUSH-TB is an international, multicenter, randomized, controlled, open-label phase 2C adaptive trial among adolescents and adults with smear- and culture-positive, drug-susceptible pulmonary tuberculosis. The primary objective is to compare the efficacy of each experimental regimen to that of the standard six-month treatment, using the intermediate endpoint of time to sputum culture negativity in liquid media. In the initial “Wave 1,” Study 38/CRUSH-TB has three arms. Due to its adaptive design, in a subsequent study, “Waves,” more arms can be added, with the comparator arm continuing enrollment.

The Centers for Disease Control and Prevention (CDC) funds TBTC to conduct Study 38/CRUSH-TB. Recruitment sites are in six countries (Canada, Haiti, South Africa, Uganda, the United States, and Vietnam) across three continents. The sociodemographic characteristics of study participants were expected to represent those populations most affected by TB worldwide.

Inclusion and exclusion criteria are listed in Table 1. Participants must have at screening an expectorated sputum specimen that is either at least 1+ positive for acid-fast bacilli on smear microscopy or positive for M. tuberculosis by nucleic acid amplification testing using Xpert^® MTB/RIF or Xpert^® MTB/RIF Ultra (“Xpert”, Cepheid Inc., Sunnyvale, CA), with semiquantitative result of ‘medium’ or ‘high’ at the sites’ laboratory of record for the trial. This approach maximizes the likelihood that a given participant has cultivable M. tuberculosis in their sputum, aligning this study with evidence from other clinical trials regarding severity of pulmonary tuberculosis based on M. tuberculosis burden detected in sputum(3, 13–15). Participants must be aged ≥12 years; children and adolescents were included to balance the importance of including children in tuberculosis trials with the specifics of this clinical trial, including flat dosing, which necessitate a population with similar to adults’ drug pharmacokinetics and disease manifestations (e.g., sputum culture positivity). Persons living with HIV with CD4 T-cell counts ≥100 cells/mm³ were eligible for enrollment.

The schedule of study procedures and assessments are provided in Table 2.

Randomization and treatment arm assignments were computer-generated. Eligible patients who satisfied all inclusion and exclusion criteria were randomly assigned in a 1:1:1 ratio to the study arms (Figure 1). Eligible patients were randomized with equal probability to each arm, using Chen’s procedure to minimize imbalance (16). Random assignment sequences were generated to limit imbalance between arms within strata while ensuring that the sequence was not predictable based on previous assignments. Randomization was stratified by site, geographical region (Africa versus not Africa), and by the presence of cavitation on chest radiograph at screening.

Doses of study drugs are shown in Table 3.

Bedaquiline is administered at a dose of 200 mg once daily for 56 days, followed by 100 mg daily for the remainder of the treatment. This dosing schedule was used in the ZeNIX TB and SimpliciTB studies, and BDQ at a 200 mg daily dose was used for eight weeks in the NC-005 study (17). Daily dosing throughout TB treatment has significant logistical advantages over the standard licensed regimen, which requires treatment to be administered three times per week after the initial 14 days. This schedule is challenging to coordinate with the other medications in this regimen, which are given daily. Further, pharmacokinetic modeling data support this dosing schedule because it produces similar bedaquiline exposures to the standard licensed dosing regimen (18) and has similar efficacy and safety (18, 19).

Delamanid is used at a dose of 300 mg once daily. This dosing provides overall daily exposures similar to the standard 100-mg dose twice daily for adults, with the convenience and programmatic feasibility advantage of once-daily dosing (20). Additionally, early bactericidal activity (EBA) data demonstrated that of the various daily doses tested (100, 200, 300, and 400 mg), maximal bactericidal activity was seen with 300 mg daily (21).

Rifabutin is used at the standard dose of 300 mg once daily. The rifamycins are considered to have the strongest sterilizing activity among established TB drug classes, but both rifampin and rifapentine reduce bedaquiline plasma concentrations by 70–80% in humans (22); however, rifabutin has a smaller impact (approximately 10% reduction in plasma concentrations) on bedaquiline pharmacokinetics (23).

Pyrazinamide is administered in two weight bands: participants weighing ≤75 kg receive a 1,500-mg dose once daily, and those weighing >75 kg receive a 2,000-mg dose once daily. These doses are selected to produce equivalent therapeutic exposures across patients of different weights, without systematically underdosing persons who weigh ≤75 kg (24), while maintaining safety (25, 26).

Regimen 1 (investigational regimen) is bedaquiline, moxifloxacin, pyrazinamide, and rifabutin (BMZRb) for eight weeks, followed by bedaquiline, moxifloxacin, and rifabutin for nine weeks (isoniazid (H)-sparing). Regimen 2 (investigational) is bedaquiline, moxifloxacin, pyrazinamide, and delamanid (BMZD) for eight weeks, followed by bedaquiline, moxifloxacin, and delamanid for 9 weeks (isoniazid-and rifamycin-sparing). Regimen 3 (control) is isoniazid, rifampin, pyrazinamide, and ethambutol (HRZE) for eight weeks, then isoniazid and rifampin (HR) for 18 weeks.

In this open-label trial, participants and clinical site staff are not blinded to the treatment assignments of individual participants for two primary reasons. Firstly, implementing blinding with placebos would have significantly increased the already high investigational groups pill burden to about 20 pills daily. We were concerned that this increase could lead to issues with adherence, negatively affect tolerability, and ultimately compromise the validity of the efficacy study results. From a physiological standpoint, the dissolution and absorption of medications in the gastrointestinal tract could be hindered by such a high pill burden, potentially reducing treatment effectiveness and raising the risk of developing drug resistance. Secondly, there is different effect of food on the absorption and bioavailability of rifampin, bedaquiline, and delamanid necessitating different instructions on food versus no food with treatment administration. Therefore, it would not have been practical to implement blinding regarding food intake during treatment administration.

Three key aspects of the study aimed to minimize ascertainment bias given the open-label design. First, the use of frequent, scheduled sputum collections for mycobacterial culture was emphasized, with the point that smear or Xpert results alone should not be used for clinical decision-making. Second, the uniform application of study visits and procedures across arms, including a pre-specified set of triggers and processes for evaluating participants who might not respond well to treatment, or, conversely, do well clinically but experience positive smear or culture at or after 17 weeks post-randomization (“possible poor treatment response” or PPTR procedures) (27). Third, keeping CDC staff members blinded to participant treatment assignment when providing study implementation guidance to the site staff members.

Participants were followed until 78 weeks post-randomization, or until the last enrolled participant completed 52 weeks post-randomization, whichever came first. The primary efficacy endpoint was time to sputum culture negative in liquid media. Secondary endpoints included long-term lack of sustained cure, safety, and additional mycobacteriology and pharmacokinetic/pharmacodynamic outcomes.

Sputum specimens were collected at each study visit, as outlined in the Schedule of Procedures/Evaluations (Table 2), and sent to the designated site laboratory for smear microscopy and mycobacteriology culture on both liquid and solid media. Phenotypic drug susceptibility testing (pDST) for at least isoniazid, rifampin, ethambutol, and a fluoroquinolone is performed on the first study isolate of M. tuberculosis and the first of any M. tuberculosis isolates obtained at or after week 17 (time of completion of the four-month regimens). Mycobacteriology laboratory procedures were harmonized across study site laboratories for 23 key elements, and test results were collected on a study-specific case report form. For each participant, the first study M. tuberculosis isolate, as well as the first of any M. tuberculosis isolates from sputum obtained at or after week 17 post-randomization, were shipped to CDC’s Tuberculosis Laboratory, where additional pDST for bedaquiline and delamanid was performed, and pyrazinamide susceptibility was assessed with pncA sequencing. Whole-genome sequencing was performed for paired isolates from the first study and at or after week 17 isolates to determine whether recurrent tuberculosis was due to relapse of the same strain or reinfection with a new strain.

The primary safety endpoint was the proportion of participants with a Grade 3 or higher adverse event during 26 weeks from randomization. Adverse events are graded by the site investigator, according to the Common Terminology Criteria for Adverse Events (version 5.0) (28). All adverse events submitted by sites to CDC are reported and reviewed in real-time by pharmacovigilance staff members.

Based on Phillips et al. (29) 90 assessable participants per arm were adequate to measure the long-term clinical endpoint with sufficient accuracy to make reliable predictions for future phase 3 trials. This sample size is also adequate to demonstrate a statistically significant reduction in time to culture negativity.

The hazard ratio for time to culture conversion for the moxifloxacin substitution, as compared to HRZE control, was shown to be 1.73 in the OFLOTUB phase 2 trial (30) with 50 participants per arm, but was 1.17 in REMoxTB with 600 participants per arm (13). The hazard ratio for the 35mg/kg rifampin arm in the PanACEA MAMS-TB trial was 1.78, 1.68 for the PaMZ arm in the NC-002 phase 2 trial, and 1.7 and 2.0 for BPaZ in the NC-005 phase 2 trial when bedaquiline was given at the licensed loading dose (400 mg daily for 2 weeks followed by 200 mg three times per week for 24 weeks (18) or daily 200mg dosing, respectively (31). The hazard ratio for the noninferior HZPM arm in Study 31/ACTG 5349 compared to the control arm was 1.38 (3). Based on these results, a hazard ratio of 1.7 could be considered an absolute minimum target for a phase 2 study.

In the REMoxTB trial, approximately 65% of 600 participants in the control arm had achieved culture conversion on MGIT by 12 weeks (13). Similar results were seen in the much smaller PanACEA MAMS-TB trial (32) although this finding was higher (86%) in the larger Study 31/ACTG 5349, with rigorous laboratory testing and reporting standards (3).

Assuming 65% of participants in the control arm achieve culture conversion by 12 weeks, 100 participants per arm would give 81% power to detect a hazard ratio of 1.6 and 90 participants per arm would give 86% power to detect a hazard ratio of 1.7. Therefore 90 participants per arm would be sufficient to give at least 85% power to detect a hazard ratio of 1.7, and greater power for higher hazard ratios, and greater power if more participants achieve culture conversion (as was seen in Study 31/ ACTG 5349).

The only participants excluded from the primary efficacy analysis were those who were late exclusions (Table 1). We expect 6% of these (6% of randomized participants were late exclusions in S31/A5349), and therefore, we will enroll 96 participants on each arm (a total of 288) to account for these late exclusions.

Intention-to-treat (ITT) analysis group included all enrolled participants who received a treatment assignment. The modified intention-to-treat analysis group (mITT) is the primary analysis population of interest for efficacy analyses, comprising all participants in the ITT population, except for those who were late exclusions. The safety analysis group was the primary analysis population of interest for safety analyses, which included all participants in the ITT population who received at least one dose of study medication.

The primary analysis of all efficacy endpoints is conducted in the mITT group. The primary analysis of the primary efficacy estimand (Table 4), time to culture negative in liquid media was performed using Cox proportional hazards models stratified by randomization stratification factors (geographical region and presence of cavitation). If there was a larger than expected contaminated or missing culture results data, we conducted multiple imputation using Fully Conditional Specification.

The secondary efficacy outcome of lack of sustained cure was analyzed using a time-to-event analysis, with the difference in proportion of participants meeting the designated endpoint at 52 weeks post-randomization estimated using Kaplan-Meier estimates. These comparisons between arms were a non-inferiority comparison using a 12% non-inferiority margin (NIM). Justification for the 12% NIM is based on a cure rate of untreated TB of 30% (33), and a 95% control arm cure rate, which gives a standard treatment effect of 65% for M1 (34). A NIM of 12% (M2) means that more than 80% of the treatment effect is preserved, exceeding the general recommendation in the Food and Drug Administration guidance of preserving at least 50% of the treatment effect. This secondary analysis was also repeated at 78 weeks post-randomization.

We used the Bayesian methodology proposed in Phillips et al. (29) to estimate the predictive probability of regimens not exceeding certain targets of the proportion of TB-related unsuccessful outcomes (lack of sustained cure) in potential future phase 3 trials. We will conduct a simulation study to estimate the probability of non-inferiority of regimens in potential future phase 3 trials under various scenarios. The calculation of the posterior predictive distribution in Phillips et al. (29) used a fixed value for the estimate of the control arm for the future phase 3 study; an obvious extension is to use the observed data on the control arm from this study, in addition to prior information on control arm performance from other recent trials.

We developed a decision framework in advance of trial opening to rank intervention regimens in terms of safety and efficacy, using results from the analysis of the primary efficacy endpoint (time to culture negativity in liquid media) as well as secondary efficacy endpoints (in particular the long-term efficacy endpoints).

Sampling for pharmacokinetic (PK) analysis of study tuberculosis drugs is performed for all study participants as a component of the main Study 38/CRUSH-TB protocol (Table 2). Plasma PK samples are shipped to a designated quality-assured laboratory, where concentrations of tuberculosis drugs and their metabolites are measured.

We tested a novel molecular biomarker (the RS ratio^®) that quantifies ongoing M. tuberculosis rRNA synthesis (35). This marker represents a first-in-class, new category of pharmacodynamic marker that monitors the impact of a drug on a bacterial cellular process, rather than bacterial burden. This marker provides a fundamentally new type of information that is distinct from and complementary to conventional culture-based measures of bacterial burden.

The RS ratio is anticipated to provide value early in treatment and at treatment completion (collection schedule in Table 3). Regimens are compared based on the proportion of participants who continue to have a detectable RS ratio at the end of treatment and in the post-treatment period. RS Ratio analyses are exploratory to supplement the primary analyses and therefore no correction for multiple testing will be applied.

The adaptive design of this trial allows for possible introduction of novel regimens. Whether or not new drugs in the development pipeline (36) display synergistic activity with components of the BMZ backbone is under evaluation in preclinical animal models. Introduction of additional arms to this trial via adaptive design can be considered in the future, as preclinical and clinical results emerge.

If new arms are added to the trial, the randomization allocation ratio will continue from that time forward, with equal numbers randomized to each arm (i.e., a 1:1:1:1 allocation ratio, etc.). Recruitment to an intervention arm will cease when the required sample size for that arm is reached (and the primary analysis for that arm will then proceed). Recruitment will continue in the control arm and any remaining arms until the required per-arm sample size is reached for all intervention arms. This approach will ensure a 1:1 ratio between each intervention arm and its contemporaneously randomized controls. The primary analyses of existing and any new arms compared to control will always only include control arm participants contemporaneously randomized to the arm in question.

To evaluate the success of the experimental arms, interim efficacy and safety analyses were conducted to have an early assessment of which experimental arm will likely demonstrate an adequate efficacy and safety profile to continue to a phase III trial. Bayesian methodology was used to assess the predictive probability of the experimental regimens obtaining predefined efficacy and safety thresholds. For efficacy, the successful experimental arm should have at least an 80% probability that the hazard ratio of the time to culture conversion in liquid media is ≥ 1.3 compared to the control arm. For safety, the successful experimental arm should have at least an 80% probability of not having a higher proportion of grade 3 or higher adverse events during 26 weeks from randomization than the control arm. Both efficacy and safety threshold should be met for the successful arm. A Bayesian approach to the Cox proportional hazards regression model was used for the efficacy threshold and a Bayesian analysis with Beta conjugate priors was used for the safety threshold. In both cases, non-informative priors were used.