Peng Man 1, Han Liu 1, Lena Friberg 1
1 Dept of Pharmacy, Uppsala University (Uppsala, Sweden)
Objectives: Immune-mediated adverse events (IMAE) have been suggested as prognostic factors for progression-free survival (PFS) and overall survival (OS) in patients receiving immune checkpoint inhibitors. However, as patients must survive long enough to experience an IMAE, such associations may be impacted by immortal time bias (ITB). Conventional methods, such as landmark analysis or Cox models with time-varying covariates, address ITB structurally but may still yield biased estimates when common causes of IMAE occurrence and survival (e.g. disease severity and immune capacity) are inadequately adjusted for. Multistate models offer an advantage by jointly capturing intermediate disease states and transitions, thereby reflecting the composite and time-dependent nature of oncology outcomes and enabling more robust estimation of covariate effects. This study applied a multistate model to address ITB when evaluating IMAE effects on survival.
The aim of this work was to evaluate whether IMAEs are independent prognostic factors for PFS and OS in atezolizumab-treated patients when addressing immortal time bias, using a multistate modeling framework that jointly characterizes disease-state transitions and adjusts for potential confounders.
Methods: Data from two atezolizumab trials (POPLAR, OAK) were analysed using a 5-state NONMEM model describing transitions among stable disease, response, progression, as defined by RECIST v1.1, death (absorbing state), and study discontinuation due to reasons other than death (absorbing state). Transition hazards were evaluated under exponential, Weibull, and Gompertz distributions. IMAE occurrence was evaluated as a time-dependent covariate. Eleven baseline covariates, including age, gender, weight, tumor cell PD-L1 expression, immune cell PD-L1 expression, number of metastatic sites, tumor size, albumin, hemoglobin, CD4+ T cell counts, CD8+ T cell counts, were investigated. Time to progression was evaluated as a predictor of progression-to-death.
Results: A total of 751 atezolizumab-treated patients were included. The stable-to-progression transition was best described by a Weibull distribution, and the progression-to-death transition by a Weibull distribution dependent on time to progression. High baseline age, high tumor PD-L1 level (TC0 or TC1/2/3), and low metastatic site number were associated with the reduced risk of stable-to-progression transition, while high CD4 count and low tumor size were associated with reduced risk of progression-to-death transition. Without covariate adjustment, IMAE reduced the risk of death (HR=0.71, 95% CI: 0.51-0.98) but not PFS (HR=0.73, 95% CI: 0.31-1.73). After adjusting for tumor burden and immune potential, IMAE was significant for progression-to-death transition (HR=0.78, 95% CI: 0.67-0.91) but insignificant for stable-to-progression transition (HR=1.45, 95% CI: 0.78-2.70).
Conclusions: Multistate models can account for ITB by jointly modeling state transitions and adjusting for confounders. After accounting for baseline covariates reflecting tumor burden, the IMAE-OS association was marginally attenuated but persisted, whereas no significant IMAE-PFS association was detected. Our analysis suggests that IMAEs may partly be surrogate predictors of survival, even after tumor burden and immune potential have been considered, emphasizing the need for cautious interpretation of IMAE-survival associations in immunotherapy.
This presentation is based on research using data from data contributor Roche that has been made available through Vivli, Inc. Vivli has not contributed to or approved, and is not in any way responsible for, the contents of this publication.
Reference: PAGE 34 (2026) Abstr 11958 [www.page-meeting.org/?abstract=11958]
Poster: Drug/Disease Modelling - Oncology