Causal Inference Approaches to Estimating the Effect of Ibrutinib Dose Modifications on Progression‐Free Survival in Chronic Lymphocytic Leukemia

Dose modifications (DMs) are common in hematology and oncology, yet their causal effect on clinical outcomes remains uncertain. Conventional survival analyses may yield biased results since DM is a time‐varying exposure and is often confounded by patient frailty and disease severity. To address these challenges, we applied a causal inference framework to real‐world data from 130 patients with chronic lymphocytic leukemia treated with ibrutinib to estimate the effect of DM on clinical outcomes. A Bayesian prognostic risk (PR) model was developed to characterize patients' baseline risk, incorporating patient‐level clinical trial data from RESONATE‐2 as prior information. Baseline (b‐sIPW) and time‐updated stabilized inverse probability weights (tu‐sIPW) were developed to improve covariate balance between patients with and without DM and to account for timing‐varying confounding. Baseline Cox models were compared with longitudinal Cox models in which DM was specified as a time‐varying exposure. In baseline Cox analyses, DM appeared to be associated with shorter progression‐free survival (PFS). However, in longitudinal Cox models accounting for time‐varying exposure and confounding using stabilized inverse probability weighting, the estimated DM effect was attenuated and no longer statistically significant. These findings suggest that the apparent detrimental effect of DM on PFS is potentially explained by baseline and time‐varying confounding, and improper handling of time‐varying exposure, highlighting the importance of causal inference methods when evaluating DM causal effects in oncology therapies.