Pauline Traynard (1), Jonathan Chauvin (1), Geraldine Ayral (1)
(1) Lixoft, Antony, France
Objectives: Time-course modelling of tumour growth (TG) and tumour growth inhibition (TGI) using ordinary differential equations (ODEs) has gained much traction in recent years. It not only allows researchers to gain a better understanding of the tumour life cycle but also allows them to make educated predictions on the efficacy of new treatments and treatment regimens.
A wide range of models are available in the literature and correspond to different hypotheses on the tumor or treatment dynamics. In the absence of detailed biological knowledge, selecting the most appropriate model is a challenge. Here, we present a modular TG/TGI model library that combines sets of basic models and possible additional features and facilitates the use of frequently used tumor growth inhibition models. It is organized into a clear model selection workflow to aid modelers when selecting the most appropriate model for their needs.
This library permits to easily test and combine different hypotheses for the tumor growth kinetics and effect of a treatment, allowing to fit a large diversity of tumor size data.
The use of the library is illustrated with the modeling of a published experimental dataset comparing different schedulings of bevacizumab and pemetrexed/cisplatin in non-small lung cancer xenografts in mice [1]. The dataset is composed of five treatment groups: control, chemotherapy (pemetrexed/cisplatin), chemotherapy with bevacizumab at the same time, 3 days before or 8 days before.
Methods: The developed library allows to select:
- a basic tumor growth model within a list, among which linear, exponential, exponential-linear, Simeoni, Koch, logistic, Gompertz, Simeoni logistic, Von Bertalanffy,
- the use of the basal tumor size as a parameter to estimate or a regressor to read from the data set,
- possible additional features to combine to the basic tumor growth, such as angiogenesis or immune dynamics,
- in case a treatment should be considered, the type of drug exposure to be considered for the treatment effect : exposure characterized with a user-selected standard PK model, read as a regressor, or a constant treatment effect starting at a given time,
- the killing hypothesis assumed for the treatment: log-kill or Norton-Simon,
- the basic dynamics of treatment effects: first-order, non-linear with an Emax model and possible sigmoidicity, exponential kill,
- possible treatment effects on previously chosen additional features.
The flexible selection workflow allows to combine all the choices above, leading to a wide variety of models. In addition, the library includes a list of fully pre-defined TGI models corresponding to common models used in the field. It is documented with figures and guidelines to help the selection of an appropriate model based on the characteristics of the data.
The library has been used to easily test different hypotheses and find the appropriate model for tumor growth and treatment effect for the non-small lung cancer cancer dataset with a stepwise approach.
The step-by-step modelling workflow set up with the MonolixSuite includes visualizing the data set to characterize the tumor size dynamics, setting up and estimating several models in Monolix, assessing the uncertainty of the parameter estimates, comparing the different runs in Sycomore to select the best option, and simulating the final model with Simulx to investigate the impact of different treatments.
Results: First an appropriate growth model was selected based on the control group. Different models were compared and the Simeoni growth model was found to give the best fit. The effect of the chemotherapy was then considered. The models from the library could easily be modified to account for the modelling of pemetrexed and cisplatin’s PK with parameters from the literature. The best model combined a log-kill linear effect with a delay induced by signal transduction. Further modification of the model allowed to include the increase in the drug’s delivery brought by bevacizumab and implemented in [1].
The final model shows a good predictive power for each treatment group. Simulations of the model allow to test different schedules of bevacizumab in combination with chemotherapy and identify the most efficient.
Conclusions: The MonolixSuite and the new TG/TGI library allow an efficient modeling and diagnosis of tumor size data, as shown with an example of non-small lung cancer data.
References:
[1] Schneider, B. K., Boyer, A., Ciccolini, J., Barlesi, F., Wang, K., Benzekry, S., & Mochel, J. P. (2019). Optimal Scheduling of Bevacizumab and Pemetrexed/Cisplatin Dosing in Non-Small Cell Lung Cancer. CPT: Pharmacometrics and Systems Pharmacology, 8(8), 577–586.
Reference: PAGE 29 (2021) Abstr 9614 [www.page-meeting.org/?abstract=9614]
Poster: Drug/Disease Modelling - Oncology