Design: randomised noninferiority trial across eight European countries, interim analysis at 5.5 years (EPoS II)
Population: 10,799 patients after removal of high-risk adenomas
Primary outcome: cumulative incidence of colorectal cancer (10 years at final analysis; 5 years reported here)
Effect: 0.77% with first surveillance at 5 years vs 0.82% at 3 years (difference -0.05 percentage points); upper boundary of the 99.12% CI 0.68 against a 0.7-point margin

EPoS II randomised 10,799 patients across eight European countries who had had high-risk adenomas removed — an adenoma of 10 mm or more, with high-grade dysplasia or villous growth, or three to ten adenomas of any kind — to a first surveillance colonoscopy at five years or at the guideline-recommended three. The three-year group also had a colonoscopy at five years.

At an interim analysis after 5.5 years, cumulative colorectal cancer incidence at five years was 0.77% with the longer interval and 0.82% with the shorter, meeting the prespecified noninferiority criterion. Cancer stage at diagnosis did not appear to differ substantially. Five patients in total died of colorectal cancer.

The argument this settles is one endoscopy services have been having on limited evidence. Post-polypectomy surveillance is a large and growing share of colonoscopy lists, and the three-year interval has rested more on convention than on randomised data. Extending it by two years in a well-defined group is the single largest capacity release available to most units — and in India, where a symptomatic patient may wait months for a diagnostic list in the public sector, that capacity has an obvious destination.

Two constraints. This is an interim report of a trial whose primary endpoint is cancer incidence at ten years, and the comparison is five-year-only surveillance against surveillance at both three and five years.

Apply the high-risk adenoma definition precisely before extending an interval
Exclude patients with more than ten adenomas, a hereditary syndrome, or an incomplete or poorly prepared index colonoscopy
Record the index colonoscopy quality indicators; an extended interval assumes the baseline examination was complete
Where local guidelines still specify three years, document the discussion rather than changing practice silently
Direct released capacity to symptomatic diagnostic lists, which is where the benefit of this result actually lands

Don't overread it

This is an interim analysis of cancer incidence at five years; the trial's primary endpoint is incidence at ten years and has not been reached.

The statistics, in plain English

The noninferiority margin was 0.7 percentage points on the upper boundary of the confidence interval for the between-group difference; the observed boundary was 0.68, so the criterion was met with very little margin to spare. That interval is a one-sided 99.12% interval, deliberately conservative because this is an interim look, with the final ten-year analysis planned at 95.33% to hold the overall type I error at 5%. Inverse probability weighting was used for patients who did not attend their five-year colonoscopy — a statistical correction for missing data rather than observed data.