Advertisement

P48 Joint Modelling Measurable Target Lesions, EQ5D/Utility and Overall Survival: Do We Still Need Partitioned Survival Models?

      Objectives

      The biologic effect of treatment is typically analyzed using tumor burden or PFS, the latter often leads to loss of information due to being based on the categorization of RECIST criteria. Whereas, tumor burden is complete and rational summarization of the process which captures data on the primary mechanism through which most treatments are expected to act. The objective of this study is to estimate overall survival using joint model which simultaneously models patient’s longitudinal tumor burden (TB) trajectory, EQ-5D/utility and survival by accounting for the association between the three outcomes.

      Methods

      We simulated data based on a clinical trial (n=110), repeated measure (940 observations). Tumor burden is defined as sum of the longest diameters of measurable prespecified target lesions. Joint model using current value association structure and several distribution was fitted to predict survival. Time was included as a random effect. No covariates were included in the model. We compared joint model with independently fitted models for OS and PFS.

      Results

      The median follow-up visit was 25 months (range: 22-29 months). Joint model showed a positive statistical association to TB, indicating that a higher value of TB increases the risk of death and a negative statistical association to EQ-5D/utility, indicating that proximity to death is linked to lower utility values. While the restricted mean estimates did not show differences between independently fitted models and the joint model, the overall uncertainty around long-term outcomes is dramatically reduced.

      Conclusions

      Multivariate Joint modelling is a powerful tool which can simultaneously model multiple longitudinal biomarkers and survival that are relevant in economic evaluation and leads to more efficient treatment effect. Given the limitations associated with partitioned survival model, i.e. the structural assumptions that survival functions modeled are independent, this approach provides a new way to conduct cost-effectiveness analysis for metastatic oncology treatments.