Exploring the use of constant and time-varying observation-based versus model-based tumour size metrics for early predictions of drug efficacy on overall survival in malignant pleural mesothelioma
Aurélie Lombard (1), Hitesh Mistry (2,3), Sonya C. Chapman (5), Ivelina Gueorguieva (5), Leon Aarons (3,4), Kayode Ogungbenro (3,4)
(1) Eli Lilly and Company, Neuilly, France (2) Division of Cancer Sciences, (3) Division of Pharmacy and Optometry, (4) Centre for Applied Pharmacokinetic Research, School of Health Sciences, Faculty of Biology, Medicine and Health, Manchester Academic Health Science Centre, University of Manchester, Manchester, United Kingdom (5) Eli Lilly and Company, Bracknell, United Kingdom
Objectives: Being able to characterise drug efficacy at an early stage during oncology clinical trials is of primary necessity for patients and drug developers. PK/PD frameworks have been developed to (1) characterise patients’ tumour size (TS) profile, (2) establish a link between TS metrics and overall survival, (3) identify optimal metrics which will best predict drug effect on overall survival and allow the assessment of drug efficacy at an early stage after treatment initiation, (4) help to design future clinical trials and facilitate cancer drug development [1,2]. Here, we sought to (1) explore the relationship between constant and time-varying observation-based versus model-based TS metrics and overall survival and (2) identify an early marker of drug efficacy for patients with malignant pleural mesothelioma (MPM).
Methods: Data was obtained from a randomized phase III clinical trial showing a significant survival benefit of the cisplatin-pemetrexed combination compared to cisplatin alone in patients with MPM. Constant and time-varying relative change in TS from baseline (RCFB) metrics were derived using imaging and model-predictions from a tumour growth inhibition model [3]. We explored the relationship between pre-treatment patient characteristics, treatment-related metrics and overall survival. We compared two models, one using observed RCFB and one using model-predicted RCFB to describe drug effect on overall survival. The validation of those potential biomarkers as a surrogate for overall survival was explored using the following Prentice criteria [4]: (1) the treatment has an effect on the true endpoint, (2) the treatment has an effect on the biomarker, (3) the biomarker correlates to the true endpoint, (4) the biomarker fully captures the treatment effect on the true endpoint.
Results: Pre-treatment ECOG performance status, TS and albumin levels were identified as baseline risk factors. The observed RCFB at the first measurement during treatment and the model-predicted RCFB at week 8 after treatment initiation were strongly associated with overall survival and both fully captured drug effect on overall survival.
The two RCFB variables were observed to have a nonlinear relationship with log(Hazard Ratio (HR)). When using a typical linear relationship between the model-predicted RCFB and the log(HR), week 2 after treatment initiation was identified as the best time point to assess drug efficacy in MPM patients rather than week 8 when taking into account the nonlinear relationship. Therefore, the linearity of the relationship between log(HR) and continuous potential covariates should always be checked to prevent any erroneous conclusion.
The relationship between the model-predicted time-varying RCFB and log(HR) was observed to evolve over time. For example, a 20% reduction from tumour size at baseline was associated with a HR of 0.66 and 0.84 at week 8 and 14 after treatment initiation, respectively. This phenomenon might be linked to the increase of the range of RCFB values over time due to deeper response-to-treatment and the emergence of resistance. Therefore, using a constant coefficient induced an instability in the model when trying to establish the time-varying RCFB-log(HR) relationship, with final parameter estimates varying depending on the initial estimates. Modelling the change in the RCFB coefficient over time would provide a full description of this relationship. However, a simpler approach was adopted to provide easier implementation during drug development and modelling the RCFB as a longitudinal variable was not retained. The observation-based and the prediction-based TS metrics that best correlated with overall survival were identified at a single time point.
Conclusions: This work suggests that the observed RCFB at the first TS measurement during treatment, when performed around the end of the 2nd cycle, and the model-predicted RCFB at week 8 would be two early biomarkers that could fully capture drug effect on overall survival. The model-based approach was not superior to observations-based approach at predicting drug effect on overall survival, suggesting that the simpler and less time-consuming observation-based approach could adequately differentiate the clinical benefit between two chemotherapy treatments in MPM patients. These results would need to be confirmed across different cancer types and treatment line before making any generalisation.
References:
[1] Claret L, Girard P, Hoff PM, Van Cutsem E, Zuideveld KP, Jorga K et al. (2009) Journal of Clinical Oncology 27 (25):4103-4108.
[2] Tate SC, Andre V, Enas N, Ribba B, Gueorguieva I (2016) European Journal of Cancer 66 (suppl C):95-103.
[3] Lombard A, Mistry H, Chapman SC, Gueorguieva I, Aarons L, Ogungbenro K (2021) European Journal of Pharmaceutical Sciences 1;161:105781.
[4] Prentice RL (1989) Statistics in Medicine 8 (4):431-440.