Disease-free survival as a surrogate endpoint for overall survival in an adjuvant trial of curatively resected stomach cancer using individual patient data meta-analysis

Abstract

Background: In investigations of the effectiveness of oncology drugs, overall survival (OS) is considered as the gold standard end point. However, the disadvantage of OS is that it requires an extended follow-up. A reasonable candidate for a surrogate for OS in the adjuvant setting is disease free survival (DFS). DFS is defined as recurrence of stomach cancer, all second cancers, or death from any cause. Recently, some evidence has been offered for the use of DFS as a surrogate for OS in, e.g., colorectal cancer. We evaluated 3 year DFS as a surrogate for 5 year OS in adjuvant trials of stomach cancer by using individual-patient data (IPD) meta- analysis validation criteria. Methods: The GASTRIC group initiated an IPD meta-analysis of all randomized clinical trials comparing chemotherapy versus surgery alone for patients with curatively resected stomach cancer. The validity of DFS at 3 years as a surrogate for OS at 5 years was investigated by using multiple analysis techniques, which included measures of correlation between the endpoints and between the treatment effects, and the percentage of agreement in the log-rank tests for the two endpoints at the trial level. Results: Among the 31 eligible trials, at most 25 collected the date of recurrence (5,014 pts). As of December 2008, IPD were available from 13 randomized controlled trials (3,161 patients). Seventy-seven percent of patients (1,280/1,652 pts) who recurred during the follow-up period, recurred within the first 3 years after study enrollment. At the trial level, 12 out of the 13 trials yielded the same conclusions for both endpoints at the 0.05 significance level. The R2 value between both endpoints was equal to 0.97 and Spearman's rank correlation was equal to 0.92. The R2 value between hazard ratios for OS and DFS was equal to 0.85 and Spearman's rank correlation was equal to 0.87. Conclusions: Our preliminary results, which are based on 63% of the targeted data, show that the 5-year OS and the 3-year DFS are highly correlated. Additionally, the hazard ratios for DFS and OS are also strongly associated. This suggests that 3-year DFS may be a good surrogate for a 5-year OS. The results based on the most recent update of the IPD will be presented during the conference.