Evaluation of six months sputum culture conversion as a surrogate endpoint in a multidrug resistant-tuberculosis trial

Abstract

The emergence of multidrug resistant-tuberculosis (MDR-TB), defined as Mycobacterium tuberculosis strains with in vitro resistance to at least isoniazid and rifampicin, has necessitated evaluation and validation of appropriate surrogate endpoints for treatment response in drug trials for MDR-TB. The trial that has demonstrated efficacy of bedaquiline, a diarylquinoline that inhibits mycobacterial ATP synthase, possesses the requisite features to conduct this evaluation. Approval of bedaquiline for use in MDR-TB was based primarily on the results of the controlled C208 Stage II study (ClinicalTrials.gov number, NCT00449644) including 160 patients randomized 1:1 to receive bedaquiline or placebo for 24 weeks when added to an 18-24-month preferred five-drug background regimen. Since randomization in C208 Stage II was preserved until study end, the trial results allow for the investigation of the complex relationship between sustained durable outcome with either Week 8 or Week 24 culture conversion as putative surrogate endpoints. The relationship between Week 120 outcome with Week 8 or Week 24 culture conversion was investigated using a descriptive analysis and with a recently developed statistical methodology for surrogate endpoint evaluation using methods of causal inference. The results demonstrate that sputum culture conversion at 24 weeks is more reliable than sputum culture conversion at 8 weeks when assessing the outcome of adding one new drug to a MDR-TB regimen.

Conflict of interest statement

We have the following interests: Janssen funded the C208 Stage II trial. Koen Andries, Nacer Lounis, An Vandebosch and Wim Van der Elst are full-time employees of Janssen Pharmaceutica. Chrispin Kambili is employed full-time by Johnson and Johnson. Paul Meyvisch, Myriam Theeuwes and Brian Dannemann were previously employed by Janssen; all are potential stockholders of Johnson and Johnson. Paul Meyvisch is a voluntary researcher at the I-BioStat at KU Leuven and Universiteit Hasselt and is currently employed by Galapagos NV, Mechelen (Belgium). Geert Molenberghs is a full-time employee of IBioStat at UHasselt and KU Leuven, Belgium. Ariel Alonso is a full-time employee of IBioStat at KU Leuven, Belgium. Koen Andries is co-inventor and has three patents on the use of quinoline derivatives for the treatment of mycobacterial diseases (rights of which have been transferred to Johnson & Johnson): 1. WO 2004/011436 – Quinoline Derivatives and Their Use as Mycobacterial Inhibitors; 2. WO 2005/117875 – Use of Substituted Quinoline Derivatives for the Treatment of Drug Resistant Mycobaterial Diseases; 3. WO 2006/067048 – Quinoline Derivatives for the Treatment of Latent Tuberculosis. There are no further patents, products in development or marketed products to declare. This does not alter our adherence to all the PLOS ONE policies on sharing data and materials.

Figures

Fig 1. CONSORT flow diagram for the C208 Stage II trial [19].

*The modified intent-to-treat population was a subset of the intent-to-treat population that excluded nine patients (6 BDQ and 3 placebo) with Mycobacteria Growth Indicator Tube results that did not allow for primary efficacy evaluation (no evidence of culture positivity prior to first intake of blinded study drug or no results during the first 8 weeks after first intake), seven patients (3 BDQ and 4 placebo) infected with extensively drug-resistant tuberculosis, eight (4 BDQ and 4 placebo) with drug-sensitive tuberculosis, and four patients (0 BDQ and 4 placebo) for whom the multidrug-resistant tuberculosis status could not be confirmed.

Fig 2. Response rate over time in the missing = failure and multiple imputation analyses.

Fig 3. Densities and distribution of the individual causal association between ΔT and ΔS for S8, S8xS24, S24 based on MGIT, and S24 based on AFB.

Fig 4. ICA evaluated using the missing = failure and multiple imputation methods at A) Week 8 and B) Week 24.

Fig 5

Surrogate predictive value for A) S8and B) S24. The conditional probabilities of (A) ΔT given ΔS8 or (B) ΔT given ΔS24 are shown. For any individual patient, ΔT and ΔS will be -1 (Harm), 0 (Equal), and 1 (Benefit).