The NHS-Galleri trial has demonstrated feasibility for multi-cancer screening trial design where all participants provide a 'sample' for screening, but only samples from the intervention arm are tested and acted upon during the trial. We assess efficiency of analysis methods when the control arm may be retrospectively tested at time of analysis. Analyses considered are: (1, traditional) by randomised allocation with all events included; (2, 'intended-effect') nested in those who tested positive in both arms and all events therein; and (3, targeted) by randomised allocation but with endpoint 'test-positive event'. They are compared using approximate statistical methods and scenario analysis. Provided the number who die from cancer after a test-positive sample is a small fraction of the total number who die from cancer, intended-effect and targeted analyses require a much smaller sample size to evaluate cancer-specific mortality than the traditional approach. Intended-effect analysis has a smaller sample size requirement than targeted analysis. This gain is only substantial when the risk of cancer death in test positives is high. Intended-effect or targeted analysis will substantially reduce the sample size needed to evaluate cancer-specific mortality in blood-based screening trials. Targeted analysis requires many fewer retrospective tests and avoids potential effects of needing to inform those whose stored samples test positive. Trialists should consider the trade-off of costs between sample size and retrospective testing requirements when choosing the analysis.
Read full abstract