Longitudinal Model-Based Biomarker Analysis of Exposure-Response Relationships in Adults with Pulmonary Tuberculosis

Abstract

The identification of sensitive, specific, and reliable biomarkers that can be quantified in the early phases of tuberculosis treatment and predictive of long-term outcome is key for the development of an effective short-course treatment regimen. Time to positivity (TTP), a biomarker of treatment outcome against Mycobacterium tuberculosis, measures longitudinal bacterial growth in mycobacterial growth indicator tube broth culture and may be predictive of standard time to stable culture conversion (TSCC). In two randomized phase 2b trials investigating dose-ranging rifapentine (Studies 29 and 29X), 662 participants had sputum collected over 6 months where TTP, TSCC, and time to culture conversion were quantified. The goals of this post hoc study were to characterize longitudinal TTP profiles and to identify individual patient characteristics associated with delayed time to culture conversion. In order to do so, a nonlinear mixed-effects model describing longitudinal TTP was built. Independent variables associated with increased bacterial clearance (increased TTP), assessed by subject-specific and population-level trajectories, were higher rifapentine exposure, lower baseline grade of sputum acid-fast bacillus smear, absence of productive cough, and lower extent of lung infiltrates on radiographs. Importantly, sensitivity analysis revealed that major learning milestones in phase 2b trials, such as significant exposure-response and covariate relationships, could be detected using truncated TTP data as early as 6 weeks from start of treatment, suggesting alternative phase 2b study designs. The TTP model built depicts a novel phase 2b surrogate endpoint that can inform early assessment of experimental treatment efficacy and treatment failure or relapse in patients treated with shorter and novel TB treatment regimens, improving efficiency of phase 2 clinical trials. (The studies discussed in this paper have been registered at ClinicalTrials.gov under identifiers NCT00694629 and NCT01043575.).

Keywords: Mycobacterium; biomarker; mycobacterium; rifampin; time to culture conversion; treatment.

Figures

FIG 1

Individual trajectories of the relation between assay time (time to positivity) and time on treatment for rifampin and rifapentine. A heat map indicating the proportion of observations available at each (nominal) study time (weeks) that lies at the censoring limit (44 days) is included above each plot. (A) Rifampin raw data, grouped by dose (mg/kg). Colored lines represent median trajectories over each arm. (B) Rifapentine raw data, grouped by dose (mg/kg). Colored lines represent median trajectories over each study and arm combination. (C) Rifapentine raw data, grouped by flat dose. Colored lines represent median trajectories over each dose, as shown in the key. (D) Rifapentine raw data, grouped by quintiles of AUC0–24. Colored lines represent median trajectories over each steady-state AUC range, as shown in the key. Abbreviations: AUC, area under the concentration-time curve from 0 to 24 h.

FIG 2

Visual predictive checks for probability of culture conversion. Base and final models of rifampin (top) and rifapentine (bottom). Observed proportion of culture-negative patients (black line), simulated proportion of culture-negative patients (dashed line), and simulated 95% confidence intervals around the simulated proportion (blue band).

FIG 3

Exposure-response plots for truncated rifapentine base models. Plots show model-predicted time-to-culture conversion for logistic models fitted to the full data time course (red) and data up to 16 weeks (tan), 12 weeks (green), 8 weeks (blue), and 6 weeks (purple). Note that enough data need to be present when truncation occurs at 6, 8, or 12 weeks.

FIG 4

Selected model-predicted trajectories for covariate combinations for rifapentine. (Left) Covariate combinations with high exposure (AUC); (right) the same covariate combinations with low exposure (AUC). Red lines represent median simulated trajectories, and blue lines represent 90% percentiles of trajectories calculated using distribution of interindividual-variance shift in parameter only. Solid portions of trajectories indicate model-predicted trajectories corresponding to observed data (up to black horizontal dashed line at the ULOQ of 44), whereas dashed portions of trajectories represent model-predicted trajectories above the ULOQ.