Marjorie Z. Imperial (1), Vincent Chang (1) Patrick P. J. Phillips (1), Payam Nahid (1), and Rada M. Savic (1)
1UCSF Center for Tuberculosis, University of California, San Francisco, USA.
Objectives:
Innovation in tuberculosis (TB) therapy is desperately needed and there is an increase in number of novel regimens undergoing evaluation in drug development programs.[1] In Phase 3 trials, the primary clinical endpoint is a composite outcome of failure at the end of treatment or relapse after stopping treatment, which can involve up to 24 months follow-up.[2] To reduce cost and save time, a biomarker that informs the effect of intervention on clinical endpoint is necessary. The current standard Phase 2 biomarker for TB therapy are sputum based: binary month 2 sputum culture conversion or time to sputum culture conversion.[3] Few tools exist that link Phase 2 and Phase 3 outcomes and those developed do not account for important factors identified to increase risk of poor outcomes leading to failed clinical trials.[2], [4] TB stakeholders are increasingly seeking to integrate innovative clinical trial approaches and tools into their decision making to facilitate early and effective deployment of the best regimens. Additionally, epidemiologic models strongly suggest that the best regimens to end the TB epidemic require high cure rates.[5] Here, we developed parametric models that link Phase 2 biomarker measures, patient phenotypes, and treatment duration to Phase 3 clinical endpoints. Then, we applied our tool to design novel innovative late stage trials aimed to reach high cure rates: i. stratified medicine trial design versus one size fits all, ii. therapeutic vaccines with post treatment effects, and iii. interventional trials that improve adherence.
Methods:
Individual level data from five Phase 3 trials were available (n=4036) that evaluated nine novel regimens with short durations. Each trial contributed data on the clinical endpoint and longitudinally measured biomarker (sputum culture) with various schedules.
A joint model that links biomarker measures and clinical endpoints was developed. Month 2 sputum culture conversion status and time to sputum culture conversion were tested as independent predictors of time to clinical endpoints. The most significant biomarker measure based on model discrimination (area under receiver operating characteristic curve, ROC) was retained in the model. Binary outcomes were modeled with logistic regression and time to event outcomes were modeled with parametric survival models with hazard risk described by the surge function. Baseline characteristics and treatment duration were also tested as predictors of outcomes using a stepwise model selection approach. Models were evaluated with visual predictive checks and model discrimination.
Clinical trial simulations were performed with the final joint model. First, simulations to predict biomarker measures for each individual were performed using relevant hazard ratios (HR) for novel treatment strategies relative to standard of care (SOC). Second, biomarker measures and significant patient phenotypes were used to predict individual level clinical endpoints. Cure rates were used to evaluate novel trial strategies.
The simulation tool was applied to inform clinical trial designs. For each application, assumptions were made. For one size fits all designs, the novel regimen was assumed to improve biomarker measures by fixed HR for all patients. For stratified medicine approaches, optimal duration was based on a published risk stratification algorithm.[2] For therapeutic vaccines, the duration effect was assumed to be extended due to post treatment effect of the vaccine (i.e. 26-week SOC plus vaccine assumed to equate to at least 28 weeks of SOC). For adherence interventions, improved adherence was also assumed to extend the duration effect. Target Phase 2 biomarker HR required for target Phase 3 cure rates were determined for each application to reach at least 0.80 probability of noninferior cure rates (margin = 6%) relative to SOC based on 500 simulations.
Sensitivity analysis were performed with 500 simulations to assess the impact of uncertainty in Phase 2 biomarker HR and in the link between the Phase 2 and Phase 3 outcomes on probability of showing noninferiority.
Parametric estimation was conducted in NONMEM 7.4 using the Lapacian method. Interactive simulation tools were developed using the shiny and mlxR (simulx) packages in R 3.4.
Results:
Model
Quantitative time to sputum culture conversion better predict TB clinical endpoints compared to binary month 2 culture conversion status (ROC=0.69 vs. 0.64). Inclusion of bacterial burden measures, HIV status, cavitation status, and sex also improved predictions (ROC=0.74). Our model suggests that low and high risk patients who culture convert within the first 2 months of treatment require additional treatment for 2 and 4 months, respectively, after conversion to reach 90% cure rates with standard regimens.
Biomarker Target HR
Clinical trial simulations showed that a potent novel 4-month one size fits all regimen requires a Phase 2 biomarker HR of 2 to show noninferior cure rates relative to SOC. To treat low risk patients with the same potent regimen, but for 2 months, a hazard ratio of 3.5 is required. For therapeutic vaccine development, a hazard ratio of 1.4 is required to increase effectiveness by 50% relative to SOC alone.
A potent novel 4-month regimen with a true Phase 2 biomarker HR of 2 has a 0.90 probability of showing noninferior cure rates compared to SOC. Uncertainty in the true HR with relative standard deviations of 20% and 50%, decreased this probability to 0.88 and 0.79, respectively. In the joint model, the relative standard error for the link between Phase 2 and Phase 3 outcomes was 15%; when accounting for this uncertainty, probability of showing noninferiority remained the same.
Trials That Improve Cure Rates
All evaluated interventions and treatment strategies showed great potential for improved cure rates leading to superiority trial designs. Efficient adherence intervention, where adherence is increased from 78% to 95% can improve cure rates to 94% for SOC compared to 90% without intervention. Furthermore, stratified medicine strategies where optimal duration is selected based on patient phenotypes can reach cure rates as high as 98%.
Conclusions:
We have linked current standard biomarker measures, patient phenotypes and treatment characteristics to TB clinical endpoints. Based on the identified relationships, we provide evidence-based biomarker targets for novel innovative clinical trial designs. Our tool has been used to design and evaluate all proposed clinical trial strategies investigated in this study, which are now in development and implementation stages. Our strategies are predicted to improve cure rates such that superiority trials are currently being designed, an unprecedented shift from noninferiority goals. As more quantitative and sensitive biomarker measures are discovered (blood based), our model can be extended to include identified relationships with clinical endpoints.
References:
[1] C. Lienhardt and P. Nahid, “Advances in clinical trial design for development of new TB treatments: A call for innovation,” PLOS Med., vol. 16, no. 3, p. e1002769, Mar. 2019.
[2] M. Z. Imperial et al., “A patient-level pooled analysis of treatment-shortening regimens for drug-susceptible pulmonary tuberculosis,” Nat. Med., vol. 24, no. 11, pp. 1708–1715, Nov. 2018.
[3] P. P. J. Phillips, K. Fielding, and A. J. Nunn, “An Evaluation of Culture Results during Treatment for Tuberculosis as Surrogate Endpoints for Treatment Failure and Relapse,” PLoS One, vol. 8, no. 5, p. e63840, May 2013.
[4] R. S. Wallis, T. Peppard, and D. Hermann, “Month 2 Culture Status and Treatment Duration as Predictors of Recurrence in Pulmonary Tuberculosis: Model Validation and Update,” PLoS One, vol. 10, no. 4, p. e0125403, Apr. 2015.
[5] E. A. Kendall et al., “Priority-Setting for Novel Drug Regimens to Treat Tuberculosis: An Epidemiologic Model,” PLOS Med., vol. 14, no. 1, p. e1002202, Jan. 2017.
Reference: PAGE () Abstr 9570 [www.page-meeting.org/?abstract=9570]
Poster: Oral: Drug/Disease Modelling