“Weighting” the Evidence: How Much Bedaquiline Is Enough?

The approval of bedaquiline in 2012 by the Food and Drug Administration was a major milestone in the treatment of tuberculosis (TB), representing the first approval of a TB medication from a new class in 40 years. Although bedaquiline has now become a cornerstone of drug-resistant TB therapy, it is worth remembering that very little was known about it at the time. Bedaquiline’s approval was based on the results of three small Phase II trials, comprising a total of 330 study participants who had actually received the drug (1–3). Drug-drug interactions had largely been unexplored, and there were appropriate concerns about QT-interval prolongation and an unexplained excess number of deaths in the treatment arm of one trial that resulted in a black-box warning (4). Nevertheless, given the high mortality associated with drug-resistant TB and the poor treatment options at the time (5, 6), the approval of bedaquiline was heralded as the beginning of a new era in the management of drug-resistant TB (7).

Now, more than 10 years later, it is fair to say that bedaquiline has indeed been a “game changer.” South Africa led the global rollout, providing critical observational data and demonstrating the drug’s safety and effectiveness in a programmatic setting (8). To date, nearly 50,000 patients have received bedaquiline in South Africa alone, and a large meta-analysis of international multidrug-resistant TB cohorts has shown that patients receiving bedaquiline had 60% lower mortality than matched controls (9). World Health Organization guidelines now recommend that all patients with rifampin-resistant TB receive a regimen containing bedaquiline (10).

With bedaquiline’s safety profile and its role in management now largely established, attention has turned to optimizing its use. Current high-priority research questions include identifying the best combination medications (11–13), defining the emergence and mechanisms of bedaquiline resistance (14), and establishing the ideal duration of therapy (15, 16). It is this last question that Trevisi and colleagues (pp. 1525–1532) have addressed in this issue of the Journal (17).

Bedaquiline is currently approved for a 6-month course (originally as part of an 18- to 24-month total duration of combination therapy for rifampin-resistant TB), as this is the only duration that was tested in the Phase II trials by the manufacturer. The limited 6-month course was chosen, however, to accelerate completion of those trials (18) and fit within traditional approaches to TB therapy, which consisted of an initial “intensive phase” followed by a “continuation phase” with fewer medications. Bedaquiline has a very long terminal half-life of more than 5 months, and there is pharmacologic reason to believe that its effects may continue well after the medication is stopped. Such an effect, however, has never been quantified, and among clinicians, the question has lingered as to whether a longer course of bedaquiline might yield superior treatment outcomes. At face value, this would seem to be a question that cannot be answered using programmatic observational data, given immortal time bias and indication bias. Trevisi and colleagues, however, bravely embrace the challenge and have attempted to tackle the question using data from the multinational endTB observational study and the technique of target trial emulation with inverse probability weighting.

To receive a longer course of bedaquiline, one must, by definition, survive longer. Examining a group of such patients retrospectively, then, would theoretically select a healthier cohort and suggest a benefit from the longer treatment course. By contrast, indication bias results when some patients are more or less likely to be prescribed a longer course of bedaquiline. One could imagine, for example, that clinicians might extend bedaquiline therapy in someone who was doing poorly, failing therapy, or receiving a background regimen containing fewer active medications. A cohort of such patients might have worse treatment outcomes; thus, one may incorrectly infer that the longer duration of bedaquiline is associated with inferior outcomes. The traditional way to avoid these biases is, of course, to randomize study participants to different durations of therapy, but absent such a trial, advanced biostatistical methods are needed.

Trevisi and colleagues have conducted an interesting analysis of a rich and important dataset. The endTB observational study prospectively enrolled patients with multidrug-resistant TB from 16 countries between 2015 and 2018 (19). All participants received bedaquiline and/or delamanid in addition to an individualized regimen, following World Health Organization recommendations at the time. Regimens included different combinations of second-line medications, including linezolid, clofazimine, and fluoroquinolones, among others, and more than a third of participants received an injectable medication. In using target trial emulation, Trevisi and colleagues imagine a hypothetical clinical trial in which participants are randomized to three different durations of bedaquiline therapy (i.e., ⩽6 months, 7–11 months, and ⩾12 mo). Using the endTB observational data (n = 1,468), they then created three clones—essentially copying the data—for each endTB participant and assigned each clone to one of the three treatment arms of the hypothetical trial. Clones who deviated from their assigned arm (i.e., who received longer or shorter durations of bedaquiline) were censored but still contributed statistical weight and survival data to those two arms up to the point of censoring. The authors then fit a logistic model to determine the probability of a given participant remaining uncensored and used those results to apply inverse probability weighting when analyzing the probability of success and comparing the three arms.

The study found no difference in treatment success associated with longer durations of bedaquiline and, when the authors ran an unadjusted analysis without inverse probability weighting, they did find an increased probability of treatment success with a ⩾12-month course of bedaquiline compared with a 6-month course. The authors highlight this as a likely spurious finding, which the inverse probability weighting served to correct.

Although the methodology of inverse probability weighting is well established (20), target trial emulation is not entirely intuitive. The nonstatistician reader may or may not be persuaded that the technique adequately mitigates underlying biases (i.e., that patients who received longer courses of bedaquiline did so for a nonrandom reason), but the authors are to be applauded for making the most of available data in the absence of an actual randomized trial. Nevertheless, there remains the possibility that potential confounders were unmeasured or inadequately assessed.

What now? The study’s findings are reassuring that a 6-month course of bedaquiline is likely sufficient for most patients, but given that treatment success rates with bedaquiline regimens are generally so high, demonstration that a longer course is superior would be hard to show. Many clinicians may wonder more about subgroups for whom they would be more inclined to prolong therapy, such as those with cavitary disease, those who remain culture positive after some specific time point, or patients with HIV and a CD4 count of less than 50 cells per cubic millimeter. Additionally, patients in the endTB cohort largely received the longer 18- to 24-month regimen recommended at the time. The applicability of the study findings to the newer, shorter regimens remains to be seen.

As data on bedaquiline continue to accrue from international drug-resistant TB programs and clinical trials, it is vital that the TB community continues to ask important research questions and refine our use of bedaquiline. Finally, ensuring access to bedaquiline (and pretomanid, linezolid, etc.) and achieving global equity in the diagnosis and treatment of drug-resistant TB are paramount. Regardless of treatment duration, failure to support patients and ensure treatment completion will lead to widespread bedaquiline resistance and the eventual squandering of this vital medication.