Evaluating the basis for use of advanced stage as a surrogate endpoint for cancer mortality in screening trials: a simulation study of meta-correlation and an alternative framework.

Background: Surrogate endpoints have been proposed to accelerate the evaluation of cancer screening but are controversial. Recent systematic reviews of advanced stage and cancer mortality in randomised trials of cancer screening have reached different conclusions regarding the suitability of advanced stage as a surrogate endpoint. We aimed to consider the suitability of trial-level meta-analysis to evaluate surrogacy and propose an alternative practical framework to evaluate potential surrogate endpoints for cancer screening trials. Methods: In this simulation study, we investigated whether a potential surrogate endpoint (advanced stage) could be used in place of cancer-specific mortality in a randomised controlled trial of cancer screening versus no screening. We focused on types of cancer screening that lead to the early detection of invasive cancer. We used in silico experiments to study meta-analysis correlation when the data are generated using a valid surrogate endpoint (one that mediates the effect of screening on mortality) or an invalid surrogate endpoint (one affected by screening without screening altering mortality). Simulations of the three surrogate scenarios considered stage-specific incidence and prognosis based on cancers of the breast, bowel, liver, lung, ovary, and prostate. Parameters in the simulations were based on stage-specific incidence, prognosis, and trial size were based on the trials included in the systematic review of Feng et al. (2024, 39 trials; source data https://github.com/hilaryrobbins/cancer_screening_endpoints); except for breast cancer. Population inclusion criteria and most parameter values in the simulations were taken from a systematic review of cancer screening trials. Alternative tests of necessary conditions for late-stage cancer incidence to be an accurate surrogate endpoint were developed. Findings: Trial-level correlation between the stage and mortality endpoints may be weak even when stage fully mediates the effect of screening on mortality. In the simulations for all cancers combined, if advanced staged mediated 100% of the screening effect and screening was effective then correlation was 0.61. Further, an invalid surrogate can also have a substantial correlation. In the simulations, if advanced stage mediated none of the screening effect the correlation across all cancers was 0.41. A series of ancillary studies to establish the utility of a surrogate endpoint will provide stronger and more timely evidence to evaluate surrogate endpoints than trial-level correlation; supported by further studies to establish a mortality benefit. Interpretation: Meta-correlation is not a reliable metric for assessing the adequacy of a surrogate endpoint for cancer screening interventions. A surrogate should be backed by causal reasoning rather than meta-analysis. To accelerate evidence generation, trials using a good candidate surrogate endpoint should be conducted in parallel with other studies to test the suitability of the endpoint. Future work is needed to determine acceptable criteria to validate cancer screening surrogate endpoints using evidence from such studies. Funding: Cancer Research UK.