(M-090) A Novel Approach to Capturing the Dynamics of Nontarget and New Tumors in Mechanistic Models of Progression-Free Survival

Jamie Connarn, Ph.D. – Senior Principal Scientist, Amgen; Khamir Mehta, Ph.D. – Director, Amgen; Saurabh Modi, Ph.D. – Senior Scientist, Amgen; Vijay Upreti, Ph.D. – Director, Amgen

Objectives: Progression-free survival (PFS) is an important endpoint in oncology trials that is determined by applying the RECIST criteria to longitudinal measurements of tumor size and appearance of new tumors. While traditionally modeled as a whole, practitioners have recently begun directly modeling the underlying longitudinal components of PFS via mechanistic models (Yu, 2020) (Baaz, 2023). Specifically, they use tumor growth inhibition (TGI) models to model the size of target lesions over time, and they jointly use a hazard function to model progression of new and nontarget tumors.
This presents several advantages. First, the model is fit using the entire longitudinal history of data instead of a single progression time, providing more statistical power and allowing for earlier prediction of PFS. Second, a more detailed, mechanistic approach is more generalizable and allows for the modeling of PFS under novel conditions such as combination therapies or new complex dosing regimens.
We explore whether these existing models can be further improved by using data on nontarget and new tumors to directly model their size longitudinally, rather than treating them as a time-to-event data.

Methods: We introduce a way to directly model the size of nontarget and new tumors using existing longitudinal data. Our approach directly estimates the continuous size of nontarget and new lesions, as well as their measured size at scan times by utilizing their entire available longitudinal history. We illustrate the method and its advantages on publicly available data from a trial in colorectal cancer.

Results: Our approach provides an even further mechanistic approach closer to the true data-generating process of longitudinal nontarget and new lesion data. This has several practical advantages to the hazard approach such as the ability to account for the different types of progression and handle non-measurable patients.

Conclusions: The size of nontarget and new lesions can be modeled longitudinally in a TGI model using existing data on nontarget and new lesions. This approach presents several advantages over current time-to-event based methods.

Citations: [1] Baaz, M. T. (2023). Model‐based prediction of progression‐free survival for combination therapies in oncology. CPT: Pharmacometrics & Systems Pharmacology, 1227-1237.

[2] Yu, J. N. (2020). A new method to model and predict progression free survival based on tumor growth dynamics. CPT: Pharmacometrics & Systems Pharmacology, 177-184.