BACKGROUND: The duration and cost of cancer clinical trials could be reduced if a surrogate endpoint were used in place of survival. We did a meta-analysis to assess the extent to which two surrogates, tumour response and time to progression, are predictive of mortality in metastatic colorectal cancer and non-small-cell lung cancer. METHODS: Summary data (median time to progression, proportion of patients responding to treatment, and median overall survival) from randomised trials of first-line treatment in colorectal cancer (146 trials) and lung cancer (191 trials) were identified. Data were extracted and analysed by linear regression. We used prediction bands for trials with 250, 500, and 750 patients to identify the surrogate threshold effect that would predict a significant difference in survival. FINDINGS: Response to treatment and time to progression correlated with improvement in survival for both lung cancer (p<0.0001 and p=0.0003, respectively) and colorectal cancer (p<0.0001 for both). To predict a significant survival gain in colorectal cancer trials, an improvement of 20% in the number of patients responding to treatment was required in trials with 750 patients, increasing to 26% in trials with 500 patients and 38% in trials with 250 patients. In lung cancer trials, the same prediction required differences in response of 18% for 750 patients, 21% for 500 patients, and 30% for 250 patients. For time to progression for both cancer types, the incremental gain needed to predict a survival improvement was a median of 1.8 months for trials with 750 patients, 2.2 months for 500 patients, and 3.3 months for 250 patients. INTERPRETATION: Irrespective of trial size, large differences in tumour response rate are needed to predict a significant survival benefit. If surrogates are chosen as the primary endpoint in a clinical trial, time to progression is the preferred measure because more modest and achievable differences are needed for a significant survival benefit. Trials in metastatic lung cancer and colorectal cancer should measure survival as their primary outcome unless the surrogate outcome difference is anticipated to exceed the threshold effect size. [ABSTRACT FROM AUTHOR]