Predictive Modeling to Study the Treatment-Shortening Potential of Novel Tuberculosis Drug Regimens, Toward Bundling of Preclinical Data

Abstract

Background Given the persistently high global burden of tuberculosis (TB), effective and shorter treatment options are needed. Here, we explore the relationship between relapse and treatment length as well as inter-regimen differences for two novel anti-TB drug regimens using a mouse model of TB infection and mathematical modeling. Methods Mycobacterium tuberculosis-infected mice were treated for up to 13 weeks with bedaquiline and pretomanid combined with moxifloxacin and pyrazinamide (BPaMZ) or linezolid (BPaL). Cure rates were evaluated 12 weeks after treatment completion. The standard regimen of isoniazid, rifampicin, pyrazinamide, and ethambutol (HRZE) was evaluated as a comparator. Results Six weeks of BPaMZ was sufficient to cure all mice. In contrast, 13 weeks of BPaL and 24 weeks of HRZE did not achieve 100% cure rates. Based on mathematical model predictions, 95% probability of cure was predicted for BPaMZ, BPaL and HRZE to occur at 1.6, 4.3, and 7.9 months, respectively. Conclusion This study provides additional evidence for the treatment-shortening capacity of BPaMZ over BPaL and HRZE. To optimally utilize preclinical data for predicting clinical outcomes, and to overcome the limitations that hamper such extrapolation, we advocate bundling of available published preclinical data into mathematical models.

Model-Based Meta-Analysis of Relapsing Mouse Model Studies from the Critical Path to Tuberculosis Drug Regimens Initiative Database

“…We accomplished this in the present study through a relatively simple logistic regression approach, which utilized observed binary relapse data for individual mice to estimate the INT and SLP parameters that describe the logit-linear relationship between relapse probability and treatment duration for each regimen. It is noted that this is analogous to that applied previously by Mourik, et al (29), as well as more recently by Mudde et al (30), which utilized a sigmoid maximum effect (Emax) model. Although based on slightly different assumptions, both mathematical models enable the calculation of model-based parameters of (which was not certified by peer review) is the author/funder.…”

“…This is significant in that it allows for robust “apples-to-apples” comparisons of all regimens in the dataset, despite many not being evaluated together in the same experiment. That is important in light of the T 50 estimates presented elsewhere for certain regimens (29, 30), as the impact of the significant covariates identified herein as well as other influential covariates that may not be known at this time, must be considered for cross-study comparisons. In this regard it is noted that the ability to make cross-study comparisons within this meta-analysis were bolstered by the presence of reference regimens (e.g., HRZ/HR), which helped to partition inter-study variability from treatment effects, especially for those regimens for which data were available from only a single study or where the number of treatment durations was limited.…”

“…Aside from the similarity in the underlying mathematical models, the present analysis extends beyond that of models based on single relapse studies (29, 30). Given the significantly larger dataset of more than 1500 mice from 28 studies, the present model-based meta-analysis methodology was able to account for inter-study covariate and random effects on the INT and SLP parameters.…”

Model-Based Meta-Analysis of Relapsing Mouse Model Studies from the Critical Path to Tuberculosis Drug Regimens Initiative Database

“…We accomplished this in the present study through a relatively simple logistic regression approach, which utilized observed binary relapse data for individual mice to estimate the INT and SLP parameters that describe the logit-linear relationship between relapse probability and treatment duration for each regimen. It is noted that this is analogous to that applied previously by Mourik et al ( 15 ) and, more recently, by Mudde et al ( 16 ), which utilized a sigmoid maximum effect (E max ) model. Although based on slightly different assumptions, both mathematical models enabled the calculation of model-based parameters of interest and the derivation of continuous relapse probability versus treatment duration profiles, as seen in Fig.…”

Design principles to assemble drug combinations for effective tuberculosis therapy using interpretable pairwise drug response measurements

“…We, therefore, sought to prioritize the candidates further by asking whether they also outperform bedaquiline, pretomanid, and linezolid (BPaL). This combination is better than the SOC in the RMM (Xu et al, 2019, Mudde et al, 2021, Berg et al, 2021) and is a highly effective combination in the clinic, where it has been used to dramatically shorten the treatment time of multidrug-resistant TB (MDR TB) (Conradie et al, 2020).…”