Optimizing Sunitinib Dosing in mRCC: Addressing Confounding Bias and Immortal Time Bias in Exposure- and Toxicity-survival Analyses using a Multistate Survival Modeling Framework
Han Liu, Tamara Ray, Lena Friberg
Uppsala University
Objectives:
Sunitinib was approved by the FDA in 2006 and remains the most cost-effective first-line treatment for metastatic renal cell carcinoma (mRCC)[1]. The labeled 50 mg daily 4/2 (four weeks on, two weeks off) oral regimen was established as the maximum tolerated dose (MTD)[2]. Since only 50 mg was tested in a large population under the 4/2 regimen, determining causal PKPD relationships is challenging[3].
Administering sunitinib at the MTD often leads to dose reductions (~32%) and treatment discontinuation (~8%) due to adverse events (AEs) in patients with mRCC[4]. This dosing strategy aligns with the historical “more is better” paradigm[5], assuming that maximizing dose enhances efficacy despite the increased risk of AEs[6,7]. Better outcomes have been observed in patients with above-median exposure[6], driving therapeutic drug monitoring (TDM) efforts to escalate doses in tolerable patients with below-median exposure[8,9]. Additionally, common AEs, such as hypertension, have been linked to prolonged progression-free survival (PFS) and overall survival (OS)[10,11], supporting toxicity-guided dosing[12].
However, the observed correlations between toxicity and survival, as well as between exposure and survival, do not confirm causality:
The hypertension-OS link may be subject to immortal time bias, as patients must survive and receive treatment long enough to develop toxicities.
The exposure-survival analysis relies on Kaplan-Meier curves stratified by median exposure over the treatment duration without adjusting for confounders that could simultaneously influence PK and survival.
This study aimed to address these limitations in analyses to improve sunitinib dosing by:
Mitigating immortal time bias in assessing blood pressure and survival associations.
Correcting for confounders in the analysis of exposure-PFS and OS relationships.
Methods:
Data were pooled from three phase II and one phase III trials [4,13–15] and analyzed in NONMEM 7.5 [16]:
- Population PK model: Patients with available PK data from the 50 mg 4/2 (n=293) or 37.5 mg continuous daily (n=96) regimens were included. Previously developed PK models for sunitinib and active metabolite SU12262[17] were refined. Weight-based allometric scaling was applied, and covariates (baseline demographics, lab values, and disease factors) were examined for inter-individual PK variability.
- Multistate survival model: All patients with survival data (n=806) were included. PFS and OS were described with six states: stable (S1), response (S2), progression (S3), discontinuation due to AEs (S4) or other reasons (S5), and death (S6)[18]. Transition hazards (?ij from state i to j) were estimated, testing predictors including all covariates examined in the PK analysis, study, and time.
- Toxicity-survival analysis: Using the developed multistate survival model, time-varying systolic blood pressure (SBP) metrics, including the occurrence of hypertension (SBP >140 mmHg, 0 = no prior occurrence, 1 = has occurred), absolute SBP, and relative change in SBP from baseline, were tested on transition hazards from stable to response (?12) and progression (?13). Hypertension occurrence before progression (0 = no, 1 = yes) was tested on ?36 (progression to death).
- Exposure-survival analysis: the multistate model was re-estimated in patients on the 50 mg 4/2 regimen with available PK data (n=293). Combined exposure (sunitinib + SU12662) was tested on all transitions. To mitigate immortal time bias from high dose reduction rates — where later-progressing patients have more possibilities for dose reduction — AUCss at cycle 1 was used, as no reductions occurred during this cycle.
Results:
Population PK model:
Sunitinib and SU12662 were described by two two-compartment models linked via the pre-systemic and hepatic formation of SU12662.
Females had 34.3% higher FM (fraction of sunitinib metabolized to SU12662), 6.3% lower sunitinib CL, and 12.2% lower SU12662 CL, leading to a 22.7% increase in active drug AUCss,c1 compared to males.
In older patients, sunitinib and SU12662 CL decreased, resulting in a 20.7% higher AUCss,c1 in 70-year-old than in 50-year-old male patients.
Higher albumin (42 ± 5.65 g/L) and hemoglobin (130 ± 19.1 g/L) increased bioavailability, raising AUCss,c1 by 13.7% and 2.4% per 10 g/L, respectively.
Multistate Survival Model:
?12 (stable to response) decreased over time since the first dose (Weibull).
?14 (stable to discontinuation due to AEs) increased 2.85x (2.85-fold) in females than males, and 1.31x per 10 g/L hemoglobin decrease.
?13 (stable to progression) increased with decreases in hemoglobin (1.14x/10 g/L), albumin (1.45x/10 g/L), and age (1.43x/10 years).
?36 (progression to death) decreased with a longer time to progression (log-logistic). ?36 increased with decreases in hemoglobin (1.07x/10 g/L) and albumin (1.44x/10 g/L), and increase in baseline tumor size (1.33x/100 mm).
Toxicity-Survival Analysis:
Predictors of ?13 (baseline hemoglobin, albumin, and age) correlated with baseline SBP, confounding the SBP-?13 relationship. No SBP metrics predicted ?13 after adjusting for confounders.
Hypertension (SBP>140 mmHg) before progression was linked to a lower post-progression death hazard (?36, HR=0.65, ?OFV= -18.9), but this effect became non-significant after adjusting for time to progression and baseline covariates (hemoglobin, albumin, and tumor size). The adjustment accounted for:
— Immortal time bias: slower progression led to longer treatment, higher hypertension risk, and lower death risk, creating an apparent hypertension-survival link.
— Confounding bias: hemoglobin and albumin confounded the SBP-?36 association.
Exposure-Survival Analysis:
The hazard of stable to discontinuation due to AEs (?14) was 1.95-fold higher per 1 µg·h/mL increase in AUCss,c1 (sunitinib + SU12662, median: 2.27, interquartile range: 1.71–2.90 µg·h/mL).
Age, albumin, and hemoglobin confounded the exposure-survival relationship. Patients with older age or higher albumin and hemoglobin had higher AUCss,c1, and lower progression hazard (?13). After adjusting for confounders, AUCss,c1 was no longer a significant predictor of ?13.
Despite a 1.23-fold higher median AUCss,c1 in females, sex was only associated with ?14 (removed after adding AUCss,c1), confirming the lack of exposure-PFS or OS relationship.
Conclusions:
After accounting for immortal time and confounding bias:
— Hypertension was not significantly associated with survival benefits;
— High exposure was related to a higher risk of AE-related treatment discontinuations and did not improve survival benefits.
MTD-based dosing may cause excessive toxicity without enhancing survival, especially in females and elderly patients. A lower starting dose may reduce toxicity and improve the quality of life.
TDM may help identify patients with high exposure, enabling early dose reductions to minimize the risk of treatment discontinuation due to AEs.
Reference: PAGE 33 (2025) Abstr 11789 [www.page-meeting.org/?abstract=11789]
Oral: Drug/Disease Modelling - Oncology