Long-term colorectal cancer incidence after adenoma removal and the effects of surveillance on incidence: a multicentre, retrospective, cohort study

Objective Postpolypectomy colonoscopy surveillance aims to prevent colorectal cancer (CRC). The 2002 UK surveillance guidelines define low-risk, intermediate-risk and high-risk groups, recommending different strategies for each. Evidence supporting the guidelines is limited. We examined CRC incidence and effects of surveillance on incidence among each risk group. Design Retrospective study of 33 011 patients who underwent colonoscopy with adenoma removal at 17 UK hospitals, mostly (87%) from 2000 to 2010. Patients were followed up through 2016. Cox regression with time-varying covariates was used to estimate effects of surveillance on CRC incidence adjusted for patient, procedural and polyp characteristics. Standardised incidence ratios (SIRs) compared incidence with that in the general population. Results After exclusions, 28 972 patients were available for analysis; 14 401 (50%) were classed as low-risk, 11 852 (41%) as intermediate-risk and 2719 (9%) as high-risk. Median follow-up was 9.3 years. In the low-risk, intermediate-risk and high-risk groups, CRC incidence per 100 000 person-years was 140 (95% CI 122 to 162), 221 (195 to 251) and 366 (295 to 453), respectively. CRC incidence was 40%–50% lower with a single surveillance visit than with none: hazard ratios (HRs) were 0.56 (95% CI 0.39 to 0.80), 0.59 (0.43 to 0.81) and 0.49 (0.29 to 0.82) in the low-risk, intermediate-risk and high-risk groups, respectively. Compared with the general population, CRC incidence without surveillance was similar among low-risk (SIR 0.86, 95% CI 0.73 to 1.02) and intermediate-risk (1.16, 0.97 to 1.37) patients, but higher among high-risk patients (1.91, 1.39 to 2.56). Conclusion Postpolypectomy surveillance reduces CRC risk. However, even without surveillance, CRC risk in some low-risk and intermediate-risk patients is no higher than in the general population. These patients could be managed by screening rather than surveillance.


InTrODuCTIOn
Colorectal cancer (CRC) causes considerable morbidity and mortality. 1 It can be prevented by removing adenomas, known precursors. 2 Patients at increased risk of CRC following adenoma removal are recommended surveillance colonoscopy. The 2002 UK surveillance guidelines stratify patients with adenomas into three risk groups, 3 as do the European Union (EU) and US guidelines. 4 5 Low-risk patients (with 1-2 adenomas <10 mm) are recommended no surveillance or surveillance at 5-10 years; while intermediate-risk/higher-risk patients (with 3-4 adenomas <10 mm or 1-2 adenomas with at least 1≥10 mm (UK/EU), or 3-10 adenomas or at least 1≥10 mm, with villous histology, or highgrade dysplasia (US)) are recommended 3-yearly surveillance. High-risk patients (with 5 or more adenomas <10 mm, or 3 or more adenomas with at least 1≥10 mm (UK), or more than 10 adenomas (US)) are recommended colonoscopy at 1 year or within 3 years before 3-yearly surveillance.
The 2002 UK guidelines were largely based on studies using detection rates of advanced adenomas (AAs) at follow-up as a proxy for CRC, 3 6-9 which overestimates risk due to higher rates of AAs than CRC. 9 10 Moreover, as the guidelines were developed before substantial improvements in colonoscopy quality, 11 such intensive surveillance may no longer be necessary.
In 2004, there was a call for proposals to reassess surveillance requirements among intermediate-risk patients, who account for most surveillance colonoscopies. 12 There was concern that the introduction of the Bowel Cancer Screening Programme (BCSP) in 2006 would increase demand for surveillance and overwhelm endoscopy services. We developed a study that examined CRC incidence among intermediate-risk patients over a median of 7.9 years, identifying a higher-risk subgroup who benefited from surveillance and a lower-risk subgroup who could potentially forego surveillance. 13 These findings were timely as adenoma surveillance accounts for 20% of colonoscopies performed in the UK and USA, placing great pressure on endoscopy services. 14 15 Revision of the guidelines is required to minimise unnecessary colonoscopies while ensuring that patients at increased CRC risk receive surveillance. The present study examined CRC incidence among all three risk groups over a median of 9.3 years and assessed effects of surveillance on CRC incidence. We aimed to identify patient subgroups who could safely forego surveillance or receive less than currently recommended.

Study design and participants
We conducted a retrospective study using data from 17 UK hospitals on patients who had adenomas removed at baseline colonoscopy from 1984 to 2010 (mostly (87%) from 2000 to 2010). We used this cohort for our previous study of intermediate-risk patients. 13 16 For the present study, we obtained updated information on the cohort (eg, on surveillance examinations, cancers and deaths). This provided longerterm follow-up data for the intermediate-risk group. We additionally examined the low-risk and high-risk groups not previously analysed.
Participating hospitals were required to have lower gastrointestinal endoscopy and pathology reports recorded electronically for at least 6 years prior to study start (2006). We searched endoscopy databases for patients who had undergone colonic examination before 31 December 2010, and searched pathology databases for reports of colorectal lesions. Endoscopy and pathology reports were pseudonymised and entered into a database (Oracle Corporation, Redwood City, California, USA). Summary values for size, histology and location were assigned to lesions seen at multiple examinations. 16 After identifying patients with colonic examinations before 31 December 2010, we looked back in these patients' records to identify the first occurrence of an adenoma, defining this as baseline. Multiple examinations were sometimes required at baseline to fully examine the colon and remove detected lesions, which we grouped and defined as the baseline visit. Baseline visits sometimes spanned days or months. Subsequent colonic examinations were grouped into surveillance visits, using rules described elsewhere. 16 We excluded patients without a colonoscopy or adenoma at baseline. We also excluded patients with CRC; a prior bowel resection; inflammatory bowel disease; polyposis, juvenile polyps, or hamartomatous polyps; Lynch syndrome or family history of familial adenomatous polyposis; colorectal carcinoma in situ reported more than 3 years before baseline; missing examination dates; or missing information needed for risk categorisation.
We obtained data on cancers and deaths from National Health Service (NHS) Digital, NHS Central Register, and National Services Scotland through 2016 and entered these into the study database. We compared the cancer data with the hospital data and resolved data duplication and inconsistency issues.
The primary outcome was incident adenocarcinoma of the colorectum, including cancers with unspecified morphology but assumed to be adenocarcinomas (those located between the rectum and caecum). In situ cancers and cancers with unspecified morphology but assumed to be squamous cell carcinomas (those located around the anus) were not included as CRCs.
In line with previous methodology, 13 16 we excluded CRCs that we assumed had arisen from incompletely resected baseline lesions because we thought their inclusion could lead to biased estimates of risk and inappropriate surveillance recommendations. Namely, we excluded CRCs found in the same/adjacent colonic segment to a baseline adenoma ≥15 mm that was seen at least twice within 5 years preceding cancer diagnosis. In sensitivity analyses, we additionally excluded CRCs that satisfied only Colon some of these criteria, but that we deemed likely to have arisen from incompletely resected lesions.

Statistical analysis
Sample size calculations were based on obtaining estimates of CRC incidence with a coefficient of variation of ~30%. Assuming an incidence rate of two CRCs per 1000 person-years, [17][18][19] nine CRCs and 4500 person-years in any risk subgroup would give a coefficient of variation of 33%. Thus, assuming the smallest subgroup would be 15% the size of the whole risk group, 60 CRCs were required in each risk group.
We compared baseline characteristics among patients with and without surveillance using χ² tests, including sex, age, adenoma number, size, histology, and dysplasia, presence of proximal polyps, colonoscopy completeness, bowel preparation quality, year of baseline visit, length of baseline visit (in days or months), family history of cancer/CRC, number of hyperplastic polyps and presence of hyperplastic polyps ≥10 mm. Colonoscopy completeness and bowel preparation quality were defined by the most complete colonoscopy and best preparation during baseline.
We estimated CRC incidence after baseline in each risk group. Time-at-risk started from the last examination at baseline. Timeto-event data were censored at first CRC diagnosis, death, emigration or date of complete ascertainment of cases in cancer registries.
We examined effects of surveillance and baseline characteristics on CRC incidence. Exposure to successive surveillance visits started at the last examination in each visit. When CRC was diagnosed at a surveillance visit, we did not include the visit as surveillance as it offered no protection against CRC. We used univariable Cox proportional-hazards models to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). Multivariable Cox regression was used to identify independent CRC risk factors, using backward stepwise selection based on likelihood ratio tests to retain variables with p values <0.05. Number of surveillance visits was included as a time-varying covariate. Interactions between number of surveillance visits and age or sex were assessed by including interaction parameters.
We performed Kaplan-Meier analyses to show time to cancer diagnosis and estimate cumulative CRC incidence with 95% CIs at 3 years, 5 years and 10 years. Cumulative incidence curves were compared using the log-rank test. We calculated standardised incidence ratios (SIRs) as the ratio of observed to expected CRC cases, with exact Poisson 95% CIs. Expected cases were calculated by multiplying sex-specific and 5-year agegroup-specific person-years by the corresponding incidence in the general population of England in 2007. 20 We divided each patient's follow-up time into distinct periods; in the absence of surveillance, censoring at first surveillance; after first surveillance, censoring at second surveillance; and after second surveillance to final censoring.
Using baseline risk factors, we stratified each risk group into lower-risk and higher-risk subgroups. Age was not included in the stratification criteria because older age is associated with worse colonoscopy quality and higher risks of complications; 21 nor was year or length of baseline visit which do not help define clinically relevant subgroups.
In our previous study of intermediate-risk patients, incomplete colonoscopies, colonoscopies of unknown completeness, poor bowel preparation, adenomas ≥20 mm, adenomas with highgrade dysplasia and proximal polyps were CRC risk factors. 13 16 In the present study, we used these factors to define higher-risk in a sensitivity analysis of the risk stratification criteria for intermediate-risk patients. Further sensitivity analyses excluded patients without a complete baseline colonoscopy.

Patient and public involvement
Our patient and public representatives reviewed the study proposal and results and have helped to develop plans for wider dissemination of the results.

reSulTS
There were 33 011 eligible patients in the updated cohort. Of these, we excluded 2859 with no baseline colonoscopy; 125 with CRC at baseline or a condition associated with increased CRC risk; 15 whose baseline occurred after 2010; 12 with colorectal carcinoma in situ more than 3 years before baseline; 2 with missing examination dates; 2 without adenomas; 980 whose risk could not be classified; and 44 who were lost to follow-up. Of the remaining 28 972, 14 401 (50%) were classed as low-risk, 11 852 (41%) as intermediate-risk and 2719 (9%) as high-risk (figure 1).
Patients attending surveillance were younger than nonattenders and generally more likely to have had more adenomas, an adenoma with tubulovillous histology or high-grade dysplasia, hyperplastic polyps or missing data at baseline. A greater proportion of attenders than non-attenders had a baseline visit before 2005, a baseline visit spanning more than 1 day and a family history of cancer/CRC. Non-attenders were more likely to have had an incomplete colonoscopy or poor bowel preparation. Among intermediate-risk patients, attenders were more likely to be male and have had an adenoma ≥20 mm or hyperplastic polyp ≥10 mm (online supplementary table 1).
The median age of low-risk patients was 64 years (IQR 55 to 72), 44% were women and 50% attended surveillance (table 1). The median time to first surveillance was 3.2 years (IQR 2.2 to 5.0). During a median follow-up of 9.6 years (IQR 7.2 to 12.4), 195 CRCs were diagnosed, giving an incidence rate of 140 per 100 000 person-years (table 1). Number of surveillance visits, age, adenoma histology, proximal polyps and colonoscopy completeness were independently associated with CRC incidence. Adjusting for these factors, a single surveillance visit was associated with a 44% reduction in CRC incidence compared with no surveillance. Incidence was even lower with two surveillance visits (table 1).
The median age of intermediate-risk patients was 66 years (IQR 58 to 74), 44% were women and 60% attended surveillance (table 2). The median time to first surveillance was 3.0 years (IQR 1.4 to 3.5). During a median follow-up of 9.1 years (IQR 6.6 to 12.4), 246 CRCs were diagnosed, giving an incidence rate of 221 per 100 000 person-years (table 2). Number of surveillance visits, age, adenoma dysplasia, proximal polyps, colonoscopy completeness, and year and length of baseline visit were independently associated with CRC incidence. Adenoma histology was not included in the final multivariable model because it was only associated with incidence when the unknown category was included. Adjusting for the other factors, a single surveillance visit was associated with a 41% reduction in CRC incidence compared with no surveillance. A similar reduction in incidence was seen with two surveillance visits (table 2). Figure 1 Study profile. *Not mutually exclusive. †Reasons for lost to follow-up: 19 patients had all examinations after emigrating; 22 patients were untraceable through national data sources and had no surveillance; and 3 patients had an unknown date of birth.

Colon
The median age of high-risk patients was 67 years (IQR 61 to 74), 29% were women and 66% attended surveillance (table 3). The median time to first surveillance was 1.5 years (IQR 1.0 to 3.0). During a median follow-up of 8.4 years (IQR 5.7 to 11.2), 84 CRCs were diagnosed, giving an incidence rate of 366 per 100 000 person-years (table 3). Number of surveillance visits, adenoma dysplasia and colonoscopy completeness were independently associated with CRC incidence. Adjusting for these factors, a single surveillance visit was associated with a halving of CRC incidence compared with no surveillance. Attendance at subsequent visits was associated with further incidence reductions (table 3).
There were no significant interactions between number of surveillance visits and age or sex (all p values ≥0.05). Each risk group was then divided into lower-risk and higher-risk subgroups using the identified baseline risk factors.

low-risk group
The higher-risk subgroup of low-risk patients comprised those with incomplete colonoscopies, colonoscopies of unknown completeness, tubulovillous or villous adenomas, or proximal polyps at baseline (n=9166, 64%); lower-risk patients had none of these (n=5235, 36%) (table 4). Higher-risk patients were older, more likely to have had a baseline visit before 2005, and had more surveillance than lower-risk patients (online supplementary table  2). Surveillance was associated with lower CRC incidence in the higher-risk but not the lower-risk subgroup; however, estimates in the lower-risk subgroup were imprecise owing to few CRCs (table 4).

Intermediate-risk group
The higher-risk subgroup of intermediate-risk patients comprised those with incomplete colonoscopies, colonoscopies of unknown completeness, adenomas with high-grade dysplasia or proximal polyps at baseline (n=7114, 60%); lower-risk patients had none of these (n=4738, 40%) (table 4). Higher-risk patients were older, more likely to have had a baseline visit before 2005, and had more surveillance than lower-risk patients (online supplementary table 2). Surveillance was associated with reduced CRC incidence in the higher-risk but not the lower-risk subgroup, although estimates in the lower-risk subgroup were imprecise (table 4).
Without surveillance, cumulative CRC incidence at 10 years was 2.6% (95% CI 2.1 to 3.3) in the whole intermediate-risk group, differing significantly between the lower-risk (1.3%, 95% CI 0.8 to 2.1) and higher-risk (3.7%, 95% CI 2.9 to 4.7) subgroups ( After first surveillance, cumulative CRC incidence still differed between the risk subgroups (table 5; figure 3), although incidence in the higher-risk subgroup was now similar to that in the general population (SIR 1.00, 95% CI 0.73 to 1.33) and was lower in the lower-risk subgroup (SIR 0.59, 95% CI 0.34 to 0.96) (table 5).
When we additionally included poor bowel preparation and adenomas ≥20 mm in the classification of higher risk, the proportion of patients classed as higher risk increased to 74%. Incidence rates and effects of surveillance on incidence remained similar (data not shown).

High-risk group
The higher-risk subgroup of high-risk patients included those with incomplete colonoscopies, colonoscopies of unknown completeness or adenomas with high-grade dysplasia at baseline (n=902, 33%); lower-risk patients had none of these (n=1817, 67%) (table 4). The subgroups were similar regarding sex, age, year of baseline visit and number of surveillance visits (online supplementary table 2). Surveillance was associated with reduced CRC incidence in the higher-risk but not the lower-risk   †The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, a tubulovillous or villous adenoma, or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors. ‡The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, an adenoma with high-grade dysplasia or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors. §The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness or an adenoma with high-grade dysplasia at baseline; the lower-risk subgroup included patients with none of these factors. Table 5 Cumulative colorectal cancer (CRC) incidence and age-standardised and sex-standardised incidence ratios  P values calculated with the log-rank test to compare incidence in the lower-risk and higher-risk subgroups of each risk group. *One minus the Kaplan-Meier estimator of the survival function was used to estimate the cumulative incidence of colorectal cancer. †Expected numbers of colorectal cancers were calculated by multiplying the sex and 5-year age-group-specific observed person-years by the corresponding incidence rates in the general population of England in 2007. ‡The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, a tubulovillous or villous adenoma, or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors. §The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, an adenoma with high-grade dysplasia or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors. ¶The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness or an adenoma with high-grade dysplasia at baseline; the lower-risk subgroup included patients with none of these factors. SIR, standardised incidence ratio. Table 5 Continued subgroup, although estimates in the lower-risk subgroup were imprecise (table 4). Without surveillance, cumulative CRC incidence at 10 years was 5.7% (95% CI 4.0 to 8.3) in the whole high-risk group, differing significantly between the lower-risk (3.8%, 95% CI 2.1 to 6.8) and higher-risk subgroups (9.9%, 95% CI 6.2 to 15.7) (table 5; figure 4). Compared with the general population, CRC incidence was higher in the whole high-risk group (SIR 1.91, 95% CI 1.39 to 2.56) and higher-risk subgroup (SIR 3.55, 95% CI 2.34 to 5.17), but not significantly different in the lower-risk subgroup (SIR 1.10, 95% CI 0.64 to 1.76) (table 5).

Colon
After first surveillance, cumulative CRC incidence at 10 years was 5.6% (95% CI 3.1 to 9.8) in the whole high-risk group, 4.4% (95% CI 1.8 to 10.6) in the lower-risk subgroup and 7.8% (95% CI 3.8 to 15.4) in the higher-risk subgroup (table 5; figure 4). Compared with the general population, CRC incidence was not significantly different in the whole high-risk group (SIR 1.34, 95% CI 0.86 to 1.99) or lower-risk subgroup (SIR 1.01, 95% CI 0.52 to 1.76), but remained higher in the higher-risk subgroup (SIR 1.97, 95% CI 1.02 to 3.44). After a second surveillance visit, CRC incidence was no longer higher in the higher-risk subgroup than in the general population (table 5).
In the main analysis, we excluded CRCs assumed to have arisen from incompletely resected baseline lesions; those found in the same/adjacent colonic segment to a baseline adenoma ≥15 mm that was seen at least twice within 5 years preceding cancer diagnosis (intermediate-risk group, n=38; high-risk group, n=12). In sensitivity analyses, we additionally excluded CRCs that satisfied only some of these criteria, but that we deemed likely to have arisen from incompletely resected lesions (low-risk group, n=6; intermediate-risk group, n=29; high-risk group, n=7). This negligibly affected the results (data not shown). Excluding patients without a complete baseline colonoscopy (low-risk group, n=2682; intermediate-risk group, n=2885; high-risk group, n=365) had little impact (online supplementary tables 3-7), although high-grade dysplasia was no longer significant in intermediate-risk patients (online supplementary table 4).

DISCuSSIOn
This is the largest study examining long-term CRC incidence following adenoma removal and the effects of surveillance on CRC incidence. We obtained data from 17 hospitals on 28 972 patients who underwent baseline colonoscopy and polypectomy and were followed for a median of 9.3 years. Stratifying the cohort into low-risk (50%), intermediate-risk (41%) and highrisk (9%) groups according to the 2002 UK surveillance guidelines, 3 we identified heterogeneity in CRC incidence and in the effects of surveillance on CRC incidence among each risk group.
Our analyses showed that patients in the low-risk group were indeed at low risk of CRC. Even among the two-thirds of the group at higher CRC risk than the rest owing to an incomplete colonoscopy, colonoscopy of unknown completeness, tubulovillous or villous adenoma, or proximal polyps at baseline, CRC incidence was similar to that in the general population, without any surveillance. Among the remaining one-third, CRC incidence without surveillance was lower than in the general population. In a resource-constrained setting, it is important to consider the opportunity costs of performing surveillance in a particular patient group; we think that patients remaining at increased CRC risk following a highquality baseline colonoscopy, as compared with the general population, should be prioritised. Given this, and considering the risks of colonoscopy, we think that patients classified as Figure 2 Cumulative colorectal cancer incidence after baseline in the low-risk group. Cumulative colorectal cancer incidence with no surveillance (censoring at first surveillance) for the whole low-risk group (A) and the lower-risk and higher-risk subgroups (B). Cumulative colorectal cancer incidence after a single surveillance visit (censoring at second surveillance) for the whole lowrisk group (C) and the lower-risk and higher-risk subgroups (D). 95% CIs are shown for each curve. The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, a tubulovillous or villous adenoma, or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors.
low-risk do not require surveillance and they could instead be managed by screening.
Our results corroborated our previous finding that surveillance is warranted for most but probably not all intermediate-risk patients. 13 Among intermediate-risk patients with incomplete colonoscopies, colonoscopies of unknown completeness, adenomas with high-grade dysplasia or proximal polyps at baseline (60% of the risk group), CRC incidence without surveillance was higher than in the general population and a single surveillance visit conferred substantial protection against CRC. Among patients without these characteristics, CRC incidence was lower than in the general population after baseline colonoscopy, indicating that surveillance is not necessary.
Incidence of CRC was high in the high-risk group; without surveillance, rates were double that in the general population. Cumulative incidence at 10 years was 6% both without surveillance and with one surveillance visit, falling to 3% with two visits. High-risk patients might therefore benefit from attending two surveillance visits, although studies are needed to define the optimum interval between first and second visits. When we stratified the high-risk group into subgroups, estimates were too imprecise to draw clear conclusions.
Our findings suggest that surveillance is warranted for high-risk patients (n=2719) and the higher-risk subgroup of intermediate-risk patients (n=7114) (34% of our cohort), but not for the lower-risk subgroup of intermediate-risk patients (n=4738) or low-risk patients (n=14 401) (66% of our cohort), who could instead be managed by screening. In the BCSP in England, surveillance is recommended for intermediate-risk and high-risk patients only. 23 In this setting, numbers of surveillance colonoscopies could be reduced by a third if the lower-risk subgroup of intermediate-risk patients forewent surveillance.
Patients returning to the BCSP would be screened biennially with the faecal immunochemical test (FIT), which replaced the faecal occult blood test in June 2019. 24 Although FIT was introduced with a relatively high positivity threshold of 120 µg haemoglobin per gram of faeces, the threshold may be lowered over time if endoscopy capacity increases, which would improve FIT sensitivity for adenomas and early CRCs. 25 It is important that patients returning to screening are reminded to see their general practitioner if lower gastrointestinal symptoms occur.
Several baseline characteristics were repeatedly predictive of CRC, including older age, incomplete colonoscopies, adenomas with high-grade dysplasia and proximal polyps. This aligns with our previous study of intermediate-risk patients, 13 16 and other studies describing these as risk factors for incident advanced neoplasia. 26 27 These findings reinforce the importance of a thorough baseline colonoscopy with complete resection of detected lesions. Incomplete resection might be implicated in the elevated risk among patients with high-grade dysplasia or proximal polyps, Figure 3 Cumulative colorectal cancer incidence after baseline in the intermediate-risk group. Cumulative colorectal cancer incidence with no surveillance (censoring at first surveillance) for the whole intermediate-risk group (A) and the lowerrisk and higher-risk subgroups (B). Cumulative colorectal cancer incidence after a single surveillance visit (censoring at second surveillance) for the whole intermediate-risk group (C) and the lower-risk and higher-risk subgroups (D). 95% CIs are shown for each curve. The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness, an adenoma with high-grade dysplasia or proximal polyps at baseline; the lower-risk subgroup included patients with none of these factors.

Colon
as advanced and proximal polyps have been associated with greater risks of incomplete resection. 28 Some proximal polyps in our study may have been serrated lesions which are often proximally located, flat, and difficult to see and remove. 29 Unfortunately, serrated lesions were not consistently classified in the era of our data.
Half of low-risk patients, 60% of intermediate-risk patients and 66% of high-risk patients attended surveillance. Nonattenders were older than attenders, more likely to have had an incomplete baseline colonoscopy or poor bowel preparation, and in the intermediate-risk group were more likely to be female, consistent with the literature. 13 16 30 That 30%-40% of intermediate-risk and high-risk patients had no surveillance suggests some underuse of surveillance colonoscopy. Unfortunately, we had no information on why patients did not attend surveillance, but reasons may have included patient comorbidities, objections to colonoscopy or process errors.
Among low-risk patients, first surveillance occurred after a median of 3.2 years, earlier than recommended. 3 This has been observed elsewhere. 31 32 Possible explanations include slow adoption of guidelines and concern about postcolonoscopy CRCs. There was greater adherence to recommended surveillance intervals for intermediate-risk and high-risk patients.
Besides the present study and our previous study of intermediaterisk patients, 13 16 only one other study has compared CRC risk following adenoma removal with that in the general population in the absence and presence of surveillance. 17 This study included 5779 patients who underwent baseline colonoscopy from 1990 to 1999. Among patients with an AA (adenoma ≥10 mm, with high-grade dysplasia or villous histology) at baseline, CRC risk without surveillance was four times that in the general population and surveillance substantially reduced this risk. By contrast, among patients with non-AAs, CRC risk without surveillance was similar to in the general population and surveillance did not affect CRC risk. The study was limited, however, by the small sample size and age of the data.
Strengths of the present study include the large, high-quality data set, comprising detailed data from 17 hospitals on baseline and surveillance colonoscopies. The hospitals included general and teaching hospitals located throughout the UK. Few data were missing and follow-up was complete for nearly all patients. Most baseline colonoscopies were performed after the introduction of colonoscopy quality initiatives in 2001. 11 Nevertheless, 20% of patients did not have a complete baseline colonoscopy. Exclusion of these patients had little impact, however, indicating that the findings are applicable in the modern era of high-quality colonoscopy.
Limitations include the observational design, meaning we cannot assume that surveillance caused the reductions in CRC incidence. However, we adjusted for potential confounders and still saw a large effect of surveillance on incidence. Use of routine data means that misclassification may have occurred; however, this would Colon Figure 4 Cumulative colorectal cancer incidence after baseline in the high-risk group. Cumulative colorectal cancer incidence with no surveillance (censoring at first surveillance) for the whole high-risk group (A) and the lower-risk and higher-risk subgroups (B). Cumulative colorectal cancer incidence after a single surveillance visit (censoring at second surveillance) for the whole high-risk group (C) and the lower-risk and higher-risk subgroups (D). 95% CIs are shown for each curve. The higher-risk subgroup included patients with an incomplete colonoscopy or colonoscopy of unknown completeness or an adenoma with high-grade dysplasia at baseline; the lower-risk subgroup included patients with none of these factors.
likely be non-differential, producing underestimations of effects. More patients attending surveillance were missing baseline data than non-attenders, particularly for colonoscopy completeness and bowel preparation quality, which is a potential source of bias. Some follow-up colonoscopies may have been for symptoms rather than surveillance. Additionally, as patients were stratified into risk groups by baseline adenoma size and number, we could not interpret the individual effects of these characteristics. Finally, although the follow-up period was long, the full benefit of surveillance on CRC incidence may not manifest until after 10 years.

COnCluSIOn
A large proportion of patients with adenomas do not remain at increased CRC risk following a complete baseline colonoscopy and polypectomy, compared with the general population. In our cohort, this was true for low-risk patients, and intermediaterisk patients without high-grade dysplasia or proximal polyps. Surveillance is probably not necessary for these patients and routine screening would suffice, although patients should be reminded to contact their general practitioner if lower gastrointestinal symptoms occur. Conversely, surveillance is warranted for high-risk patients, and intermediate-risk patients without a complete baseline colonoscopy or with high-grade dysplasia or proximal polyps, whose risk was higher than in the general population before surveillance. Incorporating these findings into guidelines could reduce surveillance colonoscopies by a third, while ensuring that patients at increased risk are protected. Twitter amanda J cross @DramandaJcross