A patient stratification tool for selecting tuberculosis treatment regimen composition and duration
Erwin Dreesen (1,2), Michael Morimoto (1), Vincent Chang (1,3), Gustavo E. Velásquez (3,4), Pieter Van Brantegem (1), Marjorie Z. Imperial (1,3), Radojka M. Savic (1,3)
(1) Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, San Francisco, California, United States of America; (2) Department of Pharmaceutical and Pharmacological Sciences, KU Leuven, Leuven, Belgium; (3) UCSF Center for Tuberculosis, University of California, San Francisco, San Francisco, California, United States of America; (4) Division of HIV, Infectious Diseases, and Global Medicine, University of California, San Francisco, San Francisco, California, United States of America.
Objectives: Tuberculosis (TB) treatment regimens endorsed by guidelines lack the flexibility needed to provide optimally safe and effective regimens for all individuals. Developing shorter treatment regimens for TB is a global priority and stratified medicine is a priority research area. TB regimens comprising various drug combinations have been tested, yet the treatment-shortening potential of individual drugs has not been quantified.
Our objective was to develop a model-based stratification algorithm that informs optimal treatment composition and duration for participants with drug-susceptible TB (DS-TB) in future clinical trials.
Methods: We performed an individual patient-level data meta-analysis using data from five international randomized controlled Phase 3 trials (DMID 01-009, RIFAQUIN, REMoxTB, OFLOTUB, and Study 31/A5349) [1–5]. The pooled dataset included data from 6,346 participants with DS-TB allocated to 13 study arms and treated with nine four- and six-month regimens including isoniazid (H), rifampin (R), pyrazinamide (Z), ethambutol (E), moxifloxacin (M), and/or rifapentine (P).
We developed a parametric time-to-event (TTE) model to describe time to TB-related unfavorable outcomes for a maximum of 18 months after start of treatment (NONMEM v7.5). We evaluated regimen composition, number of treatment days, and baseline phenotype as predictors of unfavorable outcomes. Model-building was guided by OFV comparisons, Kaplan-Meier VPC plots, and receiver operating characteristic (ROC) analyses using the model development dataset and an independent validation dataset (RIFASHORT) [6].
We compiled model parameter estimates into a stratification algorithm, classifying participants as easier-to-treat, moderately harder-to-treat, and harder-to-treat with each of the regimen compositions. We used the model to assess in silico trial designs in which each participant’s regimen composition and treatment duration were varied, aiming at a 95% cure rate 18 months after start of treatment. We developed an interactive treatment stratification tool based on the final model using the Shiny package (v1.7.4.1) in R.
Results: Of the 6,346 participants, 551 (9%) had an unfavorable outcome. The median number of treatment days was 119 [interquartile range, 118–151]. The hazard for unfavorable outcomes, described using a surge function, was higher among those living with HIV, with higher smear grade, when assigned male sex at birth, with cavitary disease, and with lower BMI. The hazard for unfavorable outcomes decreased with the number of treatment days (8% [RSE 5%] decrease per seven treatment days), E-to-Mhigh-dose substitution (26% [informed by Study 31 database; unpublished] hazard reduction), and R-to-P substitution (41% [RSE 16%]). The hazard for unfavorable outcomes increased with H-to-Mhigh-dose substitution (22% [RSE 68%] increase). The model had an area under the ROC curve (AUROC) of 0.70 (95% CI, 0.67–0.73).
We categorized participants as easier-to-treat (<16 weeks; 574/6,346; 9%), moderately harder-to-treat (16 to <26 weeks; 4,971/6,346; 78%), and harder-to-treat (≥26 weeks; 801/6,346; 13%) considering their phenotype at baseline and regimen composition. Isoniazid, rifapentine, and high-dose moxifloxacin compensated for phenotypic vulnerability.
The model had similar discrimination and calibration with the validation dataset as compared with the model development dataset (AUROC, 0.73; 95% CI, 0.62–0.83), confirming that easier-to-treat participants can be treated with a four-month HRZE regimen, while moderately harder-to-treat disease needs at least a six-month HRZE regimen to achieve a 95% cure rate.
Conclusions: We developed and externally validated a TTE model describing TB-related unfavorable outcomes. This model was the basis for a flexible patient stratification algorithm for tailoring the choice of regimen (5 multidrug combinations) and treatment duration (2–9 months) to the right phenotype. Our results provide confidence that a stratified medicine approach allows the treatment flexibility needed to enhance equity and inclusiveness in TB care. Our evidence-based interactive stratification tool can inform the selection of TB regimen compositions and durations for each patient phenotype, thereby optimizing cure rates for all patients and informing the design of innovative Phase 3/4 enrichment or treatment stratification trials.
References:
[1] Johnson JL et al. Am J Respir Crit Care Med (2009) 180:558–563.
[2] Gillespie SH et al. N Engl J Med (2014) 371:1577–1587.
[3] Merle CS et al. N Engl J Med (2014) 371:1588–1598.
[4] Jindani A et al. N Engl J Med (2014) 371:1599–1608.
[5] Dorman SE et al. N Engl J Med (2021) 384:1705–1718.
[6] Jindani A et al. NEJM Evid (2023) 2:2:–.