Mathilde Marchand 1, Tong Lu 2, Rucha Sane 2, Sravanthi Cheeti 2, Qi Liu 2, Pascal Chanu 3
1 Certara Drug Development Solutions (Paris, France), 2 Clinical Pharmacology, Genentech (South San Francisco, USA), 3 Clinical Pharmacology, Genentech (Lyon, France)
Background: In advanced breast cancer, progression-free survival (PFS) is a primary registrational endpoint and a critical metric for early development decisions, especially when overall survival (OS) data remains immature. While Tumor Growth Inhibition (TGI) models effectively characterize longitudinal tumor dynamics to predict OS, reconstructing PFS solely from target-lesion dynamics (per RECIST) is often incomplete. This approach fails to account for progression driven by non-target lesions, new lesions, or clinical deterioration. We sought to develop a modeling framework that reconstructs PFS by integrating RECIST-driven dynamics, TGI-linked OS predictions, and an additional component representing mechanisms beyond target-lesion growth.
Methods: The framework was developed using data from INAVO120, a Phase III study evaluating inavolisib versus placebo in combination with palbociclib and fulvestrant in PIK3CA-mutated, HR+/HER2− metastatic breast cancer [1]. It consisted of three integrated components:
(i) a TGI model describing longitudinal target-lesion dynamics,
(ii) a TGI–OS model linking tumor kinetics to mortality risk, and
(iii) a discrete-time hazard model capturing additional progression mechanisms not reflected in target-lesion growth.
Target-lesion trajectories were described using a TGI model estimated on the longitudinal sum of the longest diameter of target lesions (SLD) data. Virtual tumor profiles were simulated under treatment-specific parameter distributions, and RECIST-like progression times were reconstructed from simulated dynamics. Overall survival times were generated through an established TGI–OS linkage, connecting individual tumor dynamics to mortality risk using a parametric multivariate survival model. To account for progression events not driven by measurable target-lesion growth, an additional visit-driven discrete hazard model with event probabilities evaluated between scheduled assessments and parameterized as a function of tumor growth rate of the target lesion TGI model, was implemented to capture non-target, new-lesion, or clinical progression. Indeed, it was assumed that the treatment effect derived from SLDs would apply equally to other sources of disease progression. For each virtual patient, progression-free survival (PFS) was defined as the minimum of (i) simulated target-lesion progression, (ii) simulated additional progression events, or (iii) death, reflecting the composite nature of the PFS endpoint.
Results: In the clinical study, centrally assessed PFS demonstrated a significant treatment benefit (stratified hazard ratio (HR) for progression or death, 0.50; 95% CI, 0.36–0.68; P<0.001), providing the benchmark for the evaluation of the model-based predictions [1]. The TGI model provides individual estimates of tumor dynamics parameters, including KG, and enables the generation of full simulated tumor profiles. The associated multivariate TGI–OS model links tumor growth dynamics (logKG) with baseline prognostic factors, such as alkaline phosphatase and number of metastatic sites, to predict mortality risk. Model performance was validated by comparing simulated replicates against observed Kaplan–Meier PFS profiles and also the HR. Reconstructing PFS from target-lesion dynamics and OS resulted in a systematic underprediction of the event fraction. However, the proposed multi-component approach introducing a discrete-time hazard model for non-target lesions successfully reproduced the observed Kaplan–Meier curves, accurately capturing treatment arm separation and event timing consistent with the INAVO120 study, resulting in a simulated HR distribution across replicates of 0.56 [0.41 – 0.75]
Conclusions: We developed an integrated PFS modeling framework that addresses the inherent limitations of approaches focusing only on target lesions. By combining target-lesion dynamics with OS predictions and a stochastic component for non-target progression, this method improves the fidelity of PFS reconstruction while maintaining a mechanistic TGI-OS basis. This framework provides a robust tool for oncology drug development, enabling more accurate assessments of Phase III probability of success based on early-phase (Phase Ib/II) tumor dynamics.
References:
[1] Turner NC, et al. Inavolisib-Based Therapy in PIK3CA-Mutated Advanced Breast Cancer. N Engl J Med. 2024 Oct 31;391(17):1584-1596. doi: 10.1056/NEJMoa2404625. PMID: 39476340.
[2] Stein, W. D. et al. 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 17, 907–917 (2011).
Reference: PAGE 34 (2026) Abstr 12188 [www.page-meeting.org/?abstract=12188]
Poster: Drug/Disease Modelling - Oncology