2024 - Rome - Italy

PAGE 2024: Drug/Disease Modelling - Oncology
Han Liu

Comparison between multistate and parametric time-to-event models for analysis of progression-free-survival (PFS) and overall survival (OS) in oncology trials

Han Liu (1), Tamara Ray (1), Lena E. Friberg (1)

(1) Department of Pharmacy, Uppsala University, Sweden

Objectives: In oncology settings, survival analysis is often conducted to assess the relationship between predictors and efficacy endpoints like progression-free survival (PFS) and overall survival (OS). The typical analysis methods adopted are the Cox proportional hazard model and parametric time-to-event (TTE) models. However, these approaches have several limitations. Firstly, they typically ignore the interval-censored and composite nature of PFS data. Secondly, these methods do not account for the fact that intermediate events of response, progression, or second-line treatment can substantially change the risk of death, and thus bias the estimation of covariate effects. Multistate models have the potential to mitigate these issues and have been suggested as a platform model [1, 2] and have been applied for the analysis of PFS and OS in oncology trials [3, 4, 5]. The objective of this analysis was to compare the estimation of covariate effects on PFS and OS in the parametric TTE versus multistate model.

Methods: The example dataset was from patients with metastatic renal cell carcinoma receiving sunitinib enrolled in a phase II study (n=271) and a phase III study (n=369) [6,7]. PFS events occurred at the time of death or progression per RECIST 1.1 criteria and were censored at the last available date of tumor assessment if treatment was discontinued without progression. OS events occurred at death and were censored at the last known alive date due to lost-for-follow-up or termination of the study. Time 0 was the date of treatment initiation.

Parametric TTE models were developed to independently describe the occurrence of PFS events, death, censored for PFS, and censored for OS. The developed multistate model consisted of 7 states informed by tumor assessment data per RECIST 1.1 (S1-S3), active dosing records (S4-5), and survival data (S6-7): S1: Stable (SD: stable disease), S2: Response (PR: partial response or CR complete response), S3: Progression (PD: progressive disease), S4: Discontinuation due to adverse events (+28 days after the last dose), S5: Discontinuation due to other reasons (+28 days after the last dose), S6: Death, and S7: Censored for death. It was assumed that all patients started in a stable state (S1) on the date of treatment initiation. A similar model has been previously described in detail [5].

Hazards were estimated for the transitions between different states in the multistate model as well as the occurrence of different events in the parametric TTE model. Covariates tested as predictors for hazards included time since treatment initiation, age, sex, body weight, Eastern Cooperative Oncology Group (ECOG) score, and Memorial Sloan-Kettering Cancer Center (MSKCC) score. 

Results: In the multistate model, patients with worse prognostic scores based on MSKCC (high vs. intermediate vs. low-risk group) were shown to have:

  • A higher hazard of progression (0.043 vs. 0.024 vs. 0.012)
  • A higher hazard of death upon progression is described by a Weibull function with scale parameters of (0.061 vs. 0.022 vs. 0.010) and a shared shape parameter of 0.63
  • A higher probability of death upon discontinuation (0.102 vs. 0.012 vs. 0.006).

Females were found to have a 3-fold higher hazard of transiting from stable/response to discontinuation due to AEs. The transition hazard from a stable state or response to discontinuation due to AEs increased by 1.50-fold and 1.68-fold, respectively, for every 10 years increase in age.

In the parametric TTE analysis, PFS events were described by a Weibull model with a scale parameter of 0.038 vs. 0.017 vs. 0.011, and a shape of 1.00 (fixed) vs.1.00 (fixed) vs. 1.26, for patients with high, intermediate to low risk based on MSKCC. OS was described with a scale parameter of 0.019 vs. 0.007 vs. 0.005  and a shape of 1.00 (fixed) vs. 1.22 vs. 1.64. For every 10-year increase in age, the hazard of censoring for PFS was increased by 1.18-fold. Females vs. males were not identified as a significant covariate of PFS censoring.

Conclusions:  The multistate model works better in identifying covariates that are relevant to specific transitions, e.g. females vs. males on transition to discontinuation due to AEs. The estimated magnitudes of the covariate effects were different between the developed parametric TTE vs. the multistate model, and in the next step, the models' predictivity will be compared based on the calculation of the Brier score [8].



References:
[1] Beyer, U., Dejardin, D., Meller, M., Rufibach, K. & Burger, H. U. A multistate model for early decision-making in oncology. Biom. J. 62, 550–567 (2020).
[2] Lin, C.-W., Nagase, M., Doshi, S. & Dutta, S. A multistate platform model for time-to-event endpoints in oncology clinical trials. CPT Pharmacometrics Syst. Pharmacol. (2023) doi:10.1002/psp4.13069.
[3] Krishnan, S. M. et al. Multistate model for pharmacometric analyses of overall survival in HER2-negative breast cancer patients treated with docetaxel. CPT Pharmacometrics Syst Pharmacol (2021) doi:10.1002/psp4.12693.
[4] Krishnan, S. M. et al. Multistate pharmacometric model to define the impact of second-line immuno-therapies on the survival outcome of IMpower131 study. Clin. Pharmacol. Ther. (2023) doi:10.1002/cpt.2838.
[5] Liu, H. et al. A multistate modeling and simulation framework to learn dose-response of oncology drugs: Application to bintrafusp alfa in non-small cell lung cancer. CPT Pharmacometrics Syst Pharmacol (2023) doi:10.1002/psp4.12976.
[6] Motzer, R. J. et al. Sunitinib versus interferon alfa in metastatic renal-cell carcinoma. N. Engl. J. Med. 356, 115–124 (2007).
[7] Motzer, R. J. et al. Randomized phase II trial of sunitinib on an intermittent versus continuous dosing schedule as first-line therapy for advanced renal cell carcinoma. J. Clin. Oncol. 30, 1371–1377 (2012).
[8] Gerds, T. A. & Schumacher, M. Consistent estimation of the expected Brier score in general survival models with right-censored event times. Biom. J. 48, 1029–1040 (2006).


Reference: PAGE 32 (2024) Abstr 10980 [www.page-meeting.org/?abstract=10980]
Poster: Drug/Disease Modelling - Oncology
Top