Surrogacy between pathological complete response and overall survival: An individual patient data analysis of eight RCTs on neoadjuvant treatment plus surgery for esophageal cancer.

Abstract

498 Background: Overall survival (OS) is the gold standard endpoint of treatment efficacy but requires an extended follow-up period. This study aimed to determine the validity of pathological complete response (pCR) as a surrogate endpoint for OS. Methods: The individual-level surrogacy between categorical outcome and OS was assessed using Kendall correlation coefficient, τ. Because no method has been reported to estimate τ between categorical variables and OS, we proposed the new method using an inverse probability of censoring weighting (IPCW) estimator adjusted for tied data that occur when applied to categorical outcomes. To evaluate the validity of this new method, we conducted simulations to compare its estimation accuracy with existing methods. A systematic review for all phase III RCTs comparing therapies in perioperative settings for resectable advanced esophageal and gastroesophageal junction cancer was performed and individual patient data (IPD) were requested from all included trials. Finally, surrogacy between pCR and OS was assessed using our proposed method. Results: Across all scenarios with varying τ values, sample sizes, and censoring proportions, the proposed method showed the lowest bias compared to the existing statistical methods. In total, 23 eligible trials were identified and IPD were available from eight RCTs (JCOG1109, JCOG9907, JCOG9204, FFCD9901, FFCD9102, SAKK75/08, CROSS, and KOK), including 1688 patients who received neoadjuvant therapy. For trial-level analysis, pCR showed a strong correlation with a determination of correlation of 0.65 when limited to neoadjuvant chemotherapy (NAC) trials. For individual-level analysis, in patients who received NAC, the hazard ratio (HR) for OS of pCR was 0.25 (95% CI: 0.13 - 0.47), and τ between OS and pCR was 0.225. While in patients who received neoadjuvant chemoradiotherapy (NACRT), the HR was 0.54 (95% CI: 0.45 - 0.66), and τ was 0.189. Additionally, τ between OS and pathological grade (pGrade), OS and tumor regression grade (TRG) was 0.196 and 0.143, respectively. As a reference, hypothetical data showed that in order to achieve a τ of 0.8 between pCR and OS, it is necessary to be able to clearly identify the prognosis to the extent that an HR of 0.09 (95% CI: 0.08 - 0.11) can be obtained. Conclusions: While pCR is sensitive enough to stratify prognosis and exhibit strong trial-level surrogacy, no individual-level surrogacy was demonstrated, making them insufficient to replace OS. It should be noted that there are substantial differences in prognostic stratification and potential use of pCR as primary endpoints in clinical trials.