Giovanni Di Veroli1
1Certara
Introduction Modelling of longitudinal tumor size metrics has gained traction in Oncology to complement traditional endpoints. Several models have been developed to model Tumor Growth Inhibition (TGI) but none appear to be tailored to Immuno-oncology (IO) trials in which a few patients (typically 5-10%) pseudo-progress, a phenomenon where tumour size initially increases prior to decreasing. Well capturing TGI in these patients is essential (often a substantial fraction of responders), particularly if attempting to predict overall survival (OS) [1]. Objectives We aimed to develop and test a new TGI model that can be used to analyse clinical trials where pseudo-progression is present. Methods A new model has been derived by introducing a pseudo-progression term in the standard model developed by Stein et al [2]. The regular Stein model incorporates two components, a growing term (exp(kg t)) and a shrinking term (exp(-kk t)). These combine into an analytical solution that can capture either tumor growth only, or shrinkage and growth, or only shrinkage (TS = TS0 (exp(kg t) + exp(-kk t) – 1)). The model was adapted by introducing an initial phase with a linear growth (G·t) into the analytical solution. This initial phase is activated for a duration t such that 0=t=dps. Profiles are automatically classed into pseudo-progression vs. non-pseudo progression by inference of dps and a population dps cut-off. The model was initially explored in parameters space. It was then used to model a TGI dataset for an undisclosed IO drug. The dataset was composed of TGI data for 578 patients. In this dataset, 9% of patients displayed pseudo-progression or steep changes in tumor size. Both the Stein and the new models were used to model the data via the Stochastic Approximation Expectation Minimization (SAEM) algorithm by Monolix 2024R1 (https://monolixsuite.slp-software.com/). Observations were censored to account for lower limit of quantification and early drop-out. An exponential random-effect model was used for each parameter, assuming a normal distribution with mean 0 and variance ?2. A correlation was also modelled between kg and kk while residual errors were modelled as a combination of proportional and additive errors. Results Exploring the model in parameters space confirmed its ability to capture pseudo-progression. When the regular Stein model and our new model were applied to the IO TGI dataset, the new model was able to capture all atypical or pseudo-progression cases which could not be captured by the regular Stein model. In all other cases, the same profile was obtained with the two approaches. The fixed effects for kg and kk obtained with the two methods were as follows: Modelskg [/week] (%R.S.E.)kk [/week] (%R.S.E.) Stein3.8 10-3 (15.2%)1.5 10-2 (11.8%) Di Veroli4.8 10-3 (11.2%)3.4 10-2 (9.2%) The standard deviation obtained with the two methods were as follows: Models?kg (%R.S.E.)?kk (%R.S.E.) Stein1.8 (6.9%)1.4 (6.1%) Di Veroli1.4 (6.8%)1.2 (6.8%) The parameters kg and kk were found to be similar for both models for patients that did not pseudo-progress. For patients that displayed pseudo-progression, the new model provided parameters that better reflected the post-pseudo-progression phase of tumor longitudinal changes. Visual predictive checks (VPC) were similar for the two models while goodness-of-fit and individual weighted residuals (IWRES) over time plots, as well as Bayesian information criterion (BIC; 7438.4 vs. 8346.6 for Di Veroli and Stein respectively) all favoured the new model. Conclusions We have developed a new mathematical model to support TGI modelling in Immuno-Oncology trials where pseudo-progression is observed. The model can be applied to all patients independently of their response type, and its parameters can be easily interpreted in terms of the various phases of tumor progression: pseudo-progression (if present), shrinkage and (re)growth.
[1] Bruno R, Bottino D, De Alwis DP, Fojo AT, Guedj J, Liu C, et al. Progress and opportunities to advance clinical cancer therapeutics using tumor dynamic models. Clin Cancer Res 2020;26:1787–95. [2] Stein WD, Gulley JL, Fojo T, Schlom J, Madan RA, Dahut W, Figg WD, Ning Y-M, Arlen PM, Price D, Bates SE. Tumor Regression and Growth Rates Determined in Five Intramural NCI Prostate Cancer Trials: The Growth Rate Constant as an Indicator of Therapeutic Efficacy. Clin Cancer Res (2011) 17 (4): 907–917
Reference: PAGE 33 (2025) Abstr 11406 [www.page-meeting.org/?abstract=11406]
Poster: Drug/Disease Modelling - Oncology