Background
Colonoscopy surveillance is recommended for some patients post polypectomy. The 2002 UK surveillance guidelines classify post-polypectomy patients into low, intermediate and high risk, and recommend different strategies for each classification. Limited evidence supports these guidelines.

Objectives
To examine, for each risk group, long-term colorectal cancer incidence by baseline characteristics and the number of surveillance visits; the effects of interval length on detection rates of advanced adenomas and colorectal cancer at first surveillance; and the cost-effectiveness of surveillance compared with no surveillance.

Design
A retrospective cohort study and economic evaluation.

Setting
Seventeen NHS hospitals.

Participants
Patients with a colonoscopy and at least one adenoma at baseline.

Main outcome measures
Long-term colorectal cancer incidence after baseline and detection rates of advanced adenomas and colorectal cancer at first surveillance.

Data sources
Hospital databases, NHS Digital, the Office for National Statistics, National Services Scotland and Public Health England.

Methods
Cox regression was used to compare colorectal cancer incidence in the presence and absence of surveillance and to identify colorectal cancer risk factors. Risk factors were used to stratify risk groups into higher- and lower-risk subgroups. We examined detection rates of advanced adenomas and colorectal cancer at first surveillance by interval length. Cost-effectiveness of surveillance compared with no surveillance was evaluated in terms of incremental costs per colorectal cancer prevented and per quality-adjusted life-year gained.

Results
Our study included 28,972 patients, of whom 14,401 (50%), 11,852 (41%) and 2719 (9%) were classed as low, intermediate and high risk, respectively. The median follow-up time was 9.3 years. Colorectal cancer incidence was 140, 221 and 366 per 100,000 person-years among low-, intermediate- and high-risk patients, respectively. Attendance at one surveillance visit was associated with reduced colorectal cancer incidence among low-, intermediate- and high-risk patients [hazard ratios were 0.56 (95% confidence interval 0.39 to 0.80), 0.59 (95% confidence interval 0.43 to 0.81) and 0.49 (95% confidence interval 0.29 to 0.82), respectively]. Compared with the general population, colorectal cancer incidence without surveillance was similar among low-risk patients and higher among high-risk patients [standardised incidence ratios were 0.86 (95% confidence interval 0.73 to 1.02) and 1.91 (95% confidence interval 1.39 to 2.56), respectively]. For intermediate-risk patients, standardised incidence ratios differed for the lower- (0.70, 95% confidence interval 0.48 to 0.99) and higher-risk (1.46, 95% confidence interval 1.19 to 1.78) subgroups. In each risk group, incremental costs per colorectal cancer prevented and per quality-adjusted life-year gained with surveillance were lower for the higher-risk subgroup than for the lower-risk subgroup. Incremental costs per quality-adjusted life-year gained were lowest for the higher-risk subgroup of high-risk patients at £7821.

Limitations
The observational design means that we cannot assume that surveillance caused the reductions in cancer incidence. The fact that some cancer staging data were missing places uncertainty on our cost-effectiveness estimates.

Conclusions
Surveillance was associated with reduced colorectal cancer incidence in all risk groups. However, in low-risk patients and the lower-risk subgroup of intermediate-risk patients, colorectal cancer incidence was no higher than in the general population without surveillance, indicating that surveillance might not be necessary. Surveillance was most cost-effective for the higher-risk subgroup of high-risk patients.