Search

Background: We evaluated the validity of surrogacy of progression‐free survival (PFS) or time‐to‐progression (TTP) and overall response rate (ORR) in phase II trials of pancreatic ductal adenocarcinoma (PDAC). In addition, we explored the impact of predictive variables on overall survival (OS) and developed an optimal OS model. Methods: We analyzed 1867 clinical endpoint from 619 phase II PDAC trials with a systematic search from PubMed. Endpoint correlations were determined by Spearman's rank correlation. The assessed predictive factors included PFS/TTP, treatment size, therapy type, stage, and previous treatment. The relationship between predictors and OS was explored by a gamma generalized linear model (GLM) with a log‐link function and compared with linear models. Results: The Spearman rank correlation coefficient between PFS/TTP and OS was 0.88 (95% confidence interval [CI] 0.85–0.89; p < 0.0001; n = 610) and between ORR and OS was 0.58 (0.52–0.64; p < 0.0001; n = 514). Model comparison favored the GLM model over the linear model, offering more accurate predictions for higher OS values. Consequently, PFS/TTP was the strongest predictor (pseudo‐R2 = 0.75), with 1 added median PFS/TTP month associated with 13% (95% CI 13%–14%) increase in median OS. Subgroup analysis revealed that chemotherapy conferred significantly longer OS compared to targeted therapy in 1‐Agent and 2‐Agent trials, exhibiting a "very large" and "medium" effect size, respectively (rank biserial, rrb = 0.40 [95% CI 0.22–0.56] and rrb = 0.29 [0.16–0.41], both p < 0.0001), although inconsistent efficacy in 3‐Agent trials (rrb = 0.12 [−0.07–0.30], p = 0.21). Conclusions: PFS/TTP is a more reliable surrogate than ORR and a strong predictor of OS in phase II trials of pancreatic cancer. Moreover, gamma GLM (log‐link function) is a robust tool for modeling positively skewed survival data with non‐constant variance, thus can be applied to other cancers' OS data of such nature. [ABSTRACT FROM AUTHOR]