Sergey Gavrilov (1), Alexandra Smirnova (1), Alexander Dorodnov (1), Boris Shulgin (1,2), Kirill Peskov (1, 2)
(1) M&S Decisions LLC, Moscow, Russia (2) Computational Oncology Group, I.M. Sechenov First Moscow State Medical University, Moscow, Russia
Introduction: Overall Survival (OS) is a gold standard clinical endpoint to assess efficacy of anticancer therapies. However, direct comparison of OS data between the novel therapies is often compromised due to the lack of head-to-head cohorts in pivotal clinical trials and difference in patient’s characteristics. Development of the modeling approaches able to perform indirect efficacy comparison of oncology clinical trials with precise quantification of patients’ population characteristics could enhance the applicability of survival model-based meta-analysis (MBMA) in drug development decisions making.
Objectives: The key aim of the presented work is a development of a modeling methodology able to perform model-based meta-analysis using the individual survival data digitized from the published sources. In order to test the applicability of the proposed methodology we have applied it to quantitatively compare OS outcomes of PD-1 and PD-L1 immunotherapies in non-small cell lung cancer (NSCLC).
Methods: The individual survival data were digitized using modified approach proposed by Guyot P. et al [1, 2]. Additionally, subgroup analysis Kaplan-Meier KM curves presented within the same clinical trial were used to extract specific covariate values for the tumor histology, therapy line and PD-L1 expression level. Other covariates, such as mean cohort age, gender ratio and ECOG status, were extracted and added to the dataset on aggregate cohort value. Possible duplicates of patient entry were analyzed and not included into the final modeling dataset. At the next stage hierarchical parametrized survival models were developed and qualified using Stan statistical engine [3]. Additionally, number of the graphical model diagnostics were made and proposed to assess the quality of posterior prediction. Indirect comparison for the PD-1, PD-L1 and Chemotherapy OS outcome was performed in order to adjust the population model prediction on the effects of significant covariates and assess level of uncertainty of the estimated parameters.
Results: We reconstructed individual time-to-event data from 146 published OS Kaplan-Meier (KM) curves from 33 NSCLC clinical trials with PD-1 inhibitors (Nivolumab, Pembrolizumab), PD-L1 inhibitors (Durvalumab, Avelumab, Atezolizumab) and chemotherapy control arms. The proposed parametric survival models were able to adequately describe the existing OS outcome data on the population and individual-study level. As results of the comprehensive covariate modeling the individual covariates describing cancer histology, therapy line, PD-L1 status positivity and assay type were identified as statistically significant and incorporated into the final model. Model-based indirect comparison showed clinically non-inferior differences between PD-1 and PD-L1 inhibitor treatments after adjustment on the identified covariate effects.
Conclusions: The developed survival MBMA approach could be generalized and applied to any type of time-to-event outcome and disease. The reduction to individual patient level dataset allows to use complex parametrized baseline hazard models for survival description and to avoid cohort size correction, which can provide a model-based tool for the more precise indirect comparison of oncology clinical trials.
References:
[1] Guyot P, Ades AE, Ouwens MJ, Welton NJ. Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. BMC Med Res Methodol. 2012 Feb 1;12:9. doi: 10.1186/1471-2288-12-9.
[2] Wei Y, Royston P. Reconstructing time-to-event data from published Kaplan-Meier curves. Stata J. 2017;17(4):786-802.
[3] Carpenter B, Gelman A, Hoffman M. et al. Stan: A Probabilistic Programming Language. Journal of Statistical Software. 76 (1): 1–32. doi:10.18637/jss.v076.i01
Reference: PAGE 29 (2021) Abstr 9697 [www.page-meeting.org/?abstract=9697]
Poster: Methodology - New Modelling Approaches