A joint model of tumor size dynamics and survival with new lesions in EGFR mutation-positive non-small cell lung cancer patients with treatment of gefitinib or carboplatin and paclitaxel.

Abstract

e20697 Background: Tumor dynamics have been serving as significant predictors in diagnosis, staging, prognosis and treatment of patients with non-small cell lung cancer (NSCLC). The purpose of this study is to propose a joint model that links the longitudinal tumor burden to progression free survival (PFS) with the appearance of new lesion in a population of NSCLC patients with gefitinib treatment or Carboplatin/Paclitaxel. Methods: The model was extended to the estimation of new lesion based on a previously developed tumor size and survival joint model. The derivative of the tumor loads and the probability of new lesions served as biomarkers in the survival submodel. Parameters were estimated from the posterior distribution in a Bayesian framework and numerical study was realized with R and STAN. A total of 434 NSCLC patients with EGFR mutation positive treated with gefitinib or chemotherapy (carboplatin+paclitaxel) from IPASS (‘NCT00322452’) were used to construct the model. Predictions were performed on the IFUM study (‘NCT01203917’) with 102 EGFR mutation positive NSCLC patients. Results: The model performed well in PFS prediction in both within-sample and out-of-sample estimations. Further improvement of model specifications is necessary since the tumor load developing rate and appearance of new-lesion negatively impacted survival predictions. About 90% accuracy was realized by the joint model when recapitulating the outcomes from the response evaluation criteria in solid tumors (RECIST). The appearance of new lesion contributed less than tumor size in accommodating drug effect when comparing progression-specific hazards. Conclusions: This Bayesian joint model well recapitulated the outcomes from the RECIST with sequentially updated tumor size that linked to survival predictions. New insights of relative predictive values were provided by the joint model regarding the components of RESICT.