Chemoprevention of colorectal cancer: systematic review and economic evaluation

Colorectal cancer

Colorectal cancer (CRC) is a malignant neoplasm arising from the lining (mucosa) of the large intestine (colon, rectum). Approximately two-thirds of CRCs occur in the colon and one-third in the rectum.

Epidemiology and incidence

Colorectal cancer is more common in developed countries than developing countries. 1 The age-standardised incidence of CRC in England in 2006 was 54 per 100,000 in males and 34 per 100,000 in females,2 and the equivalent figures for Wales in 2007 were 62 per 100,000 in males and 40 per 100,000 in females (Table 1). 3 The incidence of colon cancer is similar in males and females (39 and 36 per 100,000 respectively in England in 2006),2 whereas the incidence of rectal cancer is higher in males (22 per 100,000 in males and 13 per 100,000 in females). Colorectal cancer is the third most common cancer in the UK, with approximately 32,000 new cases annually in England and Wales in 2006–7. 2,3

Incidence of CRC increases with age. The incidence in England in 2006 was less than 6 per 100,000 in both men and women below the age of 40 years, rising to approximately 20–25 per 100,000 in people aged between 45 and 49 years. In people aged over 75 years, the annual incidence was over 300 per 100,000 for men and over 200 per 100,000 for women. 2 The median age at diagnosis is over 70 years for both colon and rectal cancer patients. 2,3

Aetiology, pathology and prognosis

Significance in terms of ill-health (burden of disease)

Colorectal cancer is a significant cause of mortality. The age-standardised mortality rate for CRC in the UK in 2006 was 18 per 100,000 population. 23 Colorectal cancer is also a significant cause of morbidity and may affect quality of life as a result of the disease itself and its treatment. Treatment of CRC has significant cost implications for the NHS. Prevention of CRC, for example through screening, also has significant associated costs.

Symptoms

Common symptoms of CRC or large polyps are abdominal pain, rectal bleeding, persisting change in bowel habit, and anaemia; more advanced tumours are likely to cause weight loss, nausea and anorexia, and abdominal pain. The early symptoms are common in the general population and can have a variety of other causes. In some patients, symptoms do not become apparent until the cancer is far advanced. 8

Symptomatic presentation

Once an individual develops symptoms associated with CRC, there are a number of potential entry routes into the bowel cancer service. The most common route is symptomatic presentation either to a general practitioner (GP) or at an Accident and Emergency (A&E) Department. Around 71% patients present via their GP, 7% through A&E and 22% elsewhere in secondary care (Department of Health Waiting Times Database 2006). 25 Of those presenting to their GP, around 13% are referred as emergency episodes (Department of Health, personal communication). 25 Upon presentation, the patient would receive a general patient consultation and some form of investigation. Initial investigations before referral are likely to depend on the nature and severity of symptoms and may include the use of simple examinations such as per rectal examination, rigid sigmoidoscopy and proctoscopy. Patients with symptoms related to CRC may already be undergoing treatment for other conditions within a secondary-care setting and may be referred for colonoscopy. Individuals who are at high-risk of developing hereditary CRC may be identified through linkage (family history) analysis or genetic testing. 25

Screening and surveillance

Cancers and large adenomas may bleed, so their presence can be detected via faecal occult blood testing (FOBT; guaiac test) which tests for blood in the stool. Meta-analysis of four randomised controlled trials (RCTs) has shown that screening by FOBT reduced the risk of death from CRC by 16% overall, and by 23% [relative risk (RR) 0.77, 95% confidence interval (95% CI) 0.57 to 0.89] in those who were actually screened. 34 Colorectal cancer screening via FOBT has been rolled out across the UK for people aged 60–69 years, and is currently being extended to people aged up to 74 years. FOBT kits are sent out by post and can be undertaken by individuals in their own homes. Individuals with a negative FOBT result will be invited for repeat screening every 2 years, while those with a positive result will be invited for endoscopic follow-up. 35,36

Individuals in whom adenomatous polyps are identified may undergo polypectomy (removal of polyps) and may undergo endoscopic surveillance, i.e. repeat examinations at regular intervals. Guidelines published in 2002 advised that patients should undergo follow-up according to their risk level as follows:

• patients with one or two adenomas of < 1 cm diameter undergo either no repeat follow-up or 5-yearly follow-up until one negative examination

• patients with three or four adenomas of < 1 cm diameter or at least one adenoma > 1 cm undergo 3-yearly follow-up until two consecutive negative examinations

• patients with five or more adenomas or three or more adenomas at least one of which is ≥ 1 cm diameter undergo an examination after 12 months before returning to 3-yearly surveillance. 19

Routine endoscopic surveillance is also recommended for other groups of patients who are considered to be at an increased risk of developing CRC, including patients with FAP, HNPCC (and their family members), ulcerative colitis or Crohn’s disease.

Diagnostic investigations

Upon referral, diagnosis of CRC may involve the use of a variety of technologies depending on the symptoms present, the fitness of the patient and the results of previous investigations. These may include haemoglobin checks, per rectal examination, proctoscopy, colonoscopy, flexible sigmoidoscopy, barium enema, computerised tomography, positron emission tomography, ultrasound, colonography, laparotomy with biopsy, X-ray and magnetic resonance imaging scans. Further tests may be undertaken to identify non-malignant colorectal disease, e.g. ulcerative colitis and Crohn’s disease. 25 Endoscopic techniques include colonoscopy, which allows visualisation of the whole colon, and flexible sigmoidoscopy, which can reach far enough to detect approximately 60% of tumours. 8 Colonoscopy is associated with some risks, including heavy bleeding (risk of approximately 1 in 150),35 bowel perforation (estimated as 1 in 1500)35 and, in very rare cases, death (estimated at 1 in 10,000). 35,37

Colorectal adenomas can be observed and removed by endoscopy before they become malignant. Approximately 90% of adenomas can be seen and removed during colonoscopy and approximately 70% during flexible sigmoidoscopy.

Treatment of colorectal cancer

In 2000, the NHS Executive published guidelines for the management of CRC in England and Wales; this guidance was updated in 2004. 8 NICE has issued guidance on various chemotherapy regimens and biological agents for the treatment of metastatic CRC,38–40 as well as guidance on chemotherapy for adjuvant treatment of stage III colon cancer,41 and on laparoscopic surgery for CRC. 42 In addition, a report on bowel cancer services was produced for the Department of Health in 2007. 25

Treatment of CRC may be curative or palliative depending on the location of the tumour and the degree to which the tumour has penetrated the bowel and spread to other organs in the body. Treatment options differ considerably for colon and rectal tumours. Curative treatment of colon cancer involves surgical resection; the majority of patients undergo open surgery, for example, left- or right-hemicolectomy. Surgery may be undertaken as an open procedure or laparoscopically. Patients may require a permanent or temporary stoma following resection. Patients with non-metastatic colon cancer may subsequently receive adjuvant chemotherapy using a variety of regimens (for example, 5-fluorouracil and folinic acid, capecitabine, oxaliplatin) either in an inpatient or outpatient setting. 25

Curative treatment of rectal cancer involves surgical resection, most commonly via anterior resection or abdominoperineal resection, depending on the location and fixity of the tumour. Surgical resection may be undertaken alongside total mesorectal excision and may be undertaken as either an open or laparoscopic procedure. Other adjuvant/neoadjuvant treatments include radiotherapy, chemotherapy and chemoradiation, each of which may be given according to a range of different regimens. Following surgical resection, patients may be followed up using a variety of diagnostic and imaging technologies including the use of ultrasound, carcinoembryonic antigen, magnetic resonance imaging, colonoscopy and computerised tomography scans. There is no currently accepted gold standard for the follow-up of patients with CRC, and schedules may differ considerably in terms of the modality and intensity according to local protocols. 25

Recurrence of CRC may be local or metastatic; however, local recurrence is less commonly reported in patients with colon cancer. Treatments of metastatic recurrence of CRC are typically palliative; however, hepatic resection and pulmonary resection may offer a chance of cure in a small proportion of patients. The mainstay of treatment for metastatic CRC involves chemotherapy; cytotoxic agents include 5-fluorouracil, capecitabine, oxaliplatin, irinotecan, tegafur with uracil, and mitomycin. Again, these may be given according to a variety of regimens across different lines of therapy. Cetuximab, a monoclonal antibody, has also been recommended for use within a specific subgroup of these patients. Supportive care may include a range of interventions for the alleviation of symptoms associated with the disease; this may include drug therapies such as antibiotics, analgesics, transfusions, corticosteroids, as well as localised radiation therapy and non-interventional therapies such as counselling and spiritual support. 25

Diagnosis and management of FAP patients

People who are FAP carriers are identified either through linkage analysis (family history) or genetic testing (direct mutation analysis) once they reach the age of around 12 years, or based on colonoscopic investigations undertaken because of symptomatic presentation. Genetic counselling may be required in families where FAP has been diagnosed. FAP patients in whom malignant bowel tumours are not found are offered ongoing annual surveillance using flexible sigmoidoscopy from the age of 13–15 years. It is recommended that at the age of about 20 years, colonoscopic surveillance should be started, alternating between flexible sigmoidoscopy and colonoscopy thereafter. Following a diagnosis of CRC, or when the patient reaches the age of 25 without a diagnosis of cancer, FAP patients are offered prophylactic surgery to remove either the colon or both the colon and rectum. One surgical option is removal of the colon via colectomy followed by connection of the small intestine (ileum) to the rectum (ileorectal anastomosis). The other option is removal of the colon and rectum via proctocolectomy plus ileoanal pouch reconstruction. The choice of surgery depends on patient preference and the number and location of polyps or cancer. Following surgery, the patient will undergo surveillance of the small intestine (duodenum) via oesophagogastroduodenoscopy (6-monthly to 3-yearly depending on severity of duodenal polyposis). If the rectum is not removed, the patient will undergo surveillance of the rectum via flexible sigmoidoscopy every 6–12 months. 11,25,43

Diagnosis and management of HNPCC patients

As with FAP, HNPCC patients are identified either through family history or through symptom-driven colonoscopy. Diagnosis of HNPCC may be based on a detailed family history (with the aid of the Amsterdam II criteria16) together with molecular screening via microsatellite instability and immunohistochemistry techniques. Genetic counselling may be required in families where HNPCC has been diagnosed. HNPCC patients undergo surveillance colonoscopy every 2 years from the age of 25 or from 5 years younger than the youngest HNPCC-affected relative (whichever is earlier). This surveillance continues until either: the patient reaches the age of 75, or until the causative mutation in that family has been excluded. Following a diagnosis of CRC, or prophylactically, patients with HNPCC are offered either surgical removal of the bowel and rectum via proctocolectomy (plus ileoanal pouch), or surgical removal of the bowel via colectomy plus ileorectal anastomosis followed by surveillance of the rectum using flexible sigmoidoscopy at 1- to 3-yearly intervals (this option is more usual than proctocolectomy). As with FAP, the choice of surgery depends on patient preferences and the location of the tumour. 11,25,43

Current cost of colorectal cancer services

A 2007 report on CRC for the Department of Health estimated the total annual treatment costs in England to be approximately £1.1 billion, the largest component of which was the cost of diagnosis (accounting for 26.4% of the total cost), followed by the cost of follow-up (24.7% of the total cost). The mean cost per patient for colon cancer treatment was estimated at £8808, and the mean cost for rectal cancer treatment was estimated at £12,037. The additional cost of screening by FOBT for those aged 60–69 every 2 years was estimated to be £112.8 million in year 1. 25

Variation in services and uncertainty about best practice

The NICE CRC guidance from 2004 states that a national survey of NHS patients carried out in 1999–2000 (just before the publication of 2-week referral guidelines for CRC) found that 34% of patients with CRC had had an appointment with a hospital doctor within 2 weeks of visiting their GP with symptoms, but 37% had waited over 3 months for their first hospital appointment and 13% waited 7 or more months. 8,44 In addition, referrals from GPs did not always adhere correctly to national guidelines in terms of whether or not to classify cases as urgent. The NHS bowel screening programme has been rolled out in England and Wales for people aged 60–69 years, and is currently being extended to people aged up to 74 years. 35

Summary of interventions

This study aims to assess various interventions for the prevention of CRC and/or adenomatous polyps in populations at differing risks for developing CRC. Decisions concerning the inclusion of interventions were taken through reference to clinical advice. The interventions included here are those which can be classed as a drug or a nutritional supplement. The assessment does not include whole-food interventions (such as fruit, vegetables, meat, fibre, garlic, green tea). Other agents such as curcumin45 and omega-3 fatty acids46 have also begun to be assessed for potential roles in chemoprevention, but were not included in this assessment because only preliminary data exist.

This study includes the following interventions: non-steroidal anti-inflammatory drugs (NSAIDs), including aspirin and cyclo-oxygenase-2 (COX-2) inhibitors; antioxidants (including vitamin A, vitamin C, vitamin E, selenium and beta-carotene); calcium, vitamin D and folic acid. Rofecoxib was explicitly excluded from the assessment because its licence has been revoked and it has been withdrawn from the market. 47

Identification of important subgroups

The use of prophylactic interventions is assessed in various populations at differing risk of CRC. One subgroup is the general population (or individuals with no increased risk for CRC); this population is described in this assessment as the ‘low-risk’ subgroup. Another subgroup consists of individuals at an increased risk for CRC because of a personal history of adenomatous polyps, personal or family history of CRC, or inflammatory bowel disease; this population is described in this assessment as the ‘intermediate-risk’ subgroup. Finally, there are individuals who are at a high risk of CRC because of one of two genetic conditions, FAP or HNPCC; this population is described in this assessment as the ‘high-risk’ subgroup.

Current usage in the NHS

Neither NICE nor the Association of Coloproctology of Great Britain and Ireland have made either a positive or a negative recommendation for chemoprevention in any subgroup of patients. 8,26 Use of NSAIDs in FAP patients has been discussed in recent reviews. 58

Anticipated costs associated with intervention

The acquisition costs of the various interventions are described in Table 3. All the agents are self-administered in tablet form and do not require administration by a health-care professional. Aspirin is available over the counter, as are many of the antioxidants (either individually or as multivitamin tablets). It should be noted that combinations of antioxidants may result in lower acquisition costs than the sum of the individual antioxidants.

Population and relevant subgroups

The use of prophylactic interventions is assessed in the following populations:

• general population or individuals with no increased risk for CRC (described in this assessment as ‘low risk’)

• individuals at increased risk of CRC (described in this assessment as ‘intermediate risk’) because of:

  1. – history of adenomatous polyps

  2. – personal history of CRC

  3. – family history of CRC

  4. – inflammatory bowel disease

• individuals at high risk of CRC (described in this assessment as ‘high risk’) because of the following genetic conditions:

  1. – FAP

  2. – HNPCC.

Interventions

The scope of this study includes interventions that can be classed as drugs or nutritional supplements used as chemopreventive agents for CRC. The study excludes whole-food interventions (such as fruit, vegetables, meat, fibre, garlic, green tea). Decisions concerning the inclusion of interventions were taken through reference to clinical advice.

The following interventions are included in this assessment:

• aspirin

• non-aspirin NSAIDs (non-steroidal anti-inflammatory drugs), including COX-2 inhibitors

• folic acid

• calcium and/or vitamin D

• antioxidants: vitamin A, vitamin C, vitamin E, selenium and beta-carotene.

Relevant comparators

Studies were included if they compared the above interventions against placebo or some other chemopreventive agent.

Outcomes

Relevant outcomes include:

• incidence/recurrence of any adenoma

• incidence of advanced adenoma

• change in polyp burden, number and size (for secondary prevention trials in FAP/HNPCC patients in whom adenomas are already present)

• incidence of CRC

• compliance and discontinuation rates

• adverse effects

• health-related quality of life (HRQoL)

• cost-effectiveness and cost–utility.

It is thought that most CRCs develop from adenomatous polyps (growths) arising from the lining of the intestine. 17,18 Therefore, adenoma incidence is used as an intermediate outcome in many trials of chemoprevention. The term ‘adenoma recurrence’ is used for studies in patients with a history of adenomas which were resected at baseline. Advanced adenomas are those which are thought to be of higher risk of developing into CRC, although there is some uncertainty about which properties are the greatest predictors of higher risk. Guidelines from 2002 advise that patients are at a higher risk if they have a higher number of adenomas or at least one adenoma with a diameter ≥ 1 cm, while there was uncertainty regarding the role of histology as a predictor of future risk. 59 In several chemoprevention trials, advanced adenomas have been defined as those which are large (≥ 1 cm) or have villous or tubulovillous features or severe dysplasia. 60–64 Detection of CRC as an end point is likely to require trials with a larger sample size and a long follow-up duration because it is thought that adenomas may be present for 10 years or more before malignancy develops, and chemopreventive agents are thought to act mainly at the point of adenoma formation. 20,21 However, it is possible that chemopreventive agents may also act at other points in the pathway, such as during the progression from adenomas to cancer.

Identification of studies

The search strategy was comprised of two key elements: (1) systematic searches to identify RCTs that evaluate the clinical effectiveness of chemopreventive agents, and (2) systematic searches to identify qualitative studies relating to individuals’ perceptions and experiences of compliance with chemopreventive agents.

Inclusion and exclusion criteria

The following inclusion and exclusion criteria were used to identify relevant studies.

Screening references and data abstraction strategy

Titles and abstracts were screened by one reviewer and 10% of citations were double-checked by a second reviewer. A kappa of 0.76 was recorded for inter-rater reliability. Full papers were examined for inclusion by two reviewers. Data were extracted by one reviewer and checked by a second reviewer. Any discrepancies were resolved by discussion.

Critical appraisal strategy

The quality of included randomised trials was assessed using criteria based on recommendations from the Centre for Reviews and Dissemination. The following criteria were used:

• What was the number (and percentage) of randomised patients excluded from the main analyses (< 5%, between 5 and 20%, or > 20%)?

• Was the study randomised and was the method described?

• Was allocation concealed and was the method described?

• Were the following blinded to treatment allocation: participants, caregiver and analyst?

• Were the study groups comparable at baseline?

• Were there any co-interventions or other factors that may influence outcome?

Evidence synthesis methods

For discrete and numerical outcomes, relative risk (risk ratio) and risk difference are reported with 95% confidence intervals. For continuous outcomes, weighted mean differences were calculated using the inverse variance and reported with 95% confidence intervals. The studies were appraised in terms of clinical validity and methodological heterogeneity to determine whether statistical pooling of trial data within a meta-analysis was appropriate. Where studies were meta-analysed, the more conservative random effects model was used to account for clinical and methodological variations between trials. Statistical heterogeneity was described using the I-squared statistic, and potential reasons for any heterogeneity were discussed. Forest plots are presented for all analyses where there was more than one relevant study and sufficient data to undertake a meta-analysis. Results for all analyses, including those of single studies, are presented in summary tables.

For studies that include more than one intervention, two types of comparison are analysed and presented:

• the intervention of interest alone versus placebo alone

• the intervention of interest (plus or minus any other intervention) versus placebo (plus or minus any other intervention).

Where trials assess a combination of more than one relevant intervention versus placebo (e.g. all the participants in one trial arm received vitamin C plus vitamin E and all the participants in the other arm received placebo), outcomes data were analysed as a combined intervention. The data contributed to the meta-analysis of (for example) all antioxidants versus placebo, and also to the meta-analysis of vitamin C plus vitamin E versus placebo, but did not contribute to the meta-analysis of vitamin C alone versus placebo.

Factorial trials assessing more than one intervention of the same ‘intervention group’ (as listed, see Inclusion and exclusion criteria, e.g. ‘antioxidants’ or ‘NSAIDs’) were analysed as follows. For example, a factorial trial may include the following arms: 25% of participants receive vitamin C, 25% receive vitamin E, 25% receive both vitamins, and 25% receive placebo alone. If the trial report presented separate data on all trial arms including the placebo-alone arm, then the data contributed to the meta-analysis of all antioxidants versus placebo, and also to the meta-analyses of vitamin C versus placebo, vitamin E versus placebo, and combined vitamins C and E versus placebo. However, if the trial report only presented data on, for example, vitamin C (plus or minus other interventions) versus no vitamin C (plus or minus other interventions), and the ‘other interventions’ were of the same group (i.e. other antioxidants), then these data could not be used in the meta-analysis of all antioxidants versus placebo. These data were used in the meta-analysis of vitamin C versus placebo, as long as the study reported relative risks which were adjusted for the effects of the other antioxidants in the study; in this case, the data were meta-analysed using these adjusted relative risks directly (via the inverse variance method) rather than using event data.

Participant treatment and follow-up duration are stated to the nearest year within this report, for clarity. The way in which treatment and follow-up durations were measured was sometimes unclear in the study reports, but they generally referred to either the median value, or that which applied to the majority of patients.

With respect to the interpretation of clinical effectiveness evidence to inform the development of the health economic model, it was necessary to know the relative risks of both advanced adenomas and ‘low-risk’ adenomas (i.e. those that were not defined as advanced). ‘Low-risk’ adenoma rates were calculated by subtracting the rates of advanced adenomas from the rates of any adenomas, for studies which presented both outcomes (see Chapter 5).

Aspirin: high-risk population (FAP or HNPCC)

Aspirin: intermediate-risk population (history of adenomas or colorectal cancer

Outcome data

Effectiveness

Effectiveness results for aspirin in the intermediate-risk population (history of adenomas or CRC) are shown in Figure 3 and Table 8.

FIGURE 3.

Aspirin: meta-analyses (history of adenomas or colorectal cancer). (a) Recurrence of any adenoma: (i) aspirin ± folic acid versus placebo ± folic acid, (ii) aspirin alone versus placebo alone, (iii) aspirin + folic acid versus placebo alone; (b) Incidence of advanced adenoma: (i) aspirin ± folic acid versus placebo ± folic acid, (ii) aspirin alone versus placebo alone, (iii) aspirin + folic acid versus placebo alone; (c) Incidence of colorectal cancer: aspirin ± folic acid versus placebo ± folic acid.

TABLE 8 - Aspirin: summary of results (history of adenomas or colorectal cancer)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2; (%)	RD	LCI	UCI	I2; (%)
Recurrence of any adenoma
Baron,61,73,74 Logan,60 Benamouzig,62 Sandler79	Any aspirin (81–325 mg/day) (± folic acid 0.5–1 mg/day)	No aspirin (± folic acid)	Adenoma	4	1,3,3,3	480	1540	408	1152	0.79	0.68	0.92	34	–0.07	–0.11	–0.04	0
Baron,61,73,74 Logan,60 Benamouzig,62 Sandler79	Aspirin alone (81–325 mg/day)	Placebo alone	Adenoma	4	1,3,3,3	266	926	242	736	0.80	0.65	0.98	47	–0.07	–0.11	–0.03	4
Baron,61,73,74 Logan60	Aspirin (81–325 mg/day) + folic acid (0.5–1 mg/day)	Placebo alone	Adenoma	2	3,3	184	550	126	366	0.90	0.75	1.08	0	–0.04	–0.10	0.02	0
Incidence of advanced adenoma
Baron,61,73,74 Logan,60 Benamouzig62	Any aspirin (81–325 mg/day) (± folic acid 0.5–1 mg/day)	No aspirin (± folic acid)	Advanced adenoma	3	1,3,3	115	1281	123	894	0.66	0.51	0.84	0	–0.05	–0.07	–0.02	0
Baron,61,73,74 Logan,60 Benamouzig62	Aspirin alone (81–325 mg/day)	Placebo alone	Advanced adenoma	3	1,3,3	58	667	57	478	0.75	0.52	1.07	0	–0.03	–0.06	0.01	0
Baron,61,73,74 Logan60	Aspirin (81–325 mg/day) + folic acid (0.5–1 mg/day)	Placebo alone	Advanced adenoma	2	3,3	49	550	44	366	0.77	0.45	1.34	45	–0.03	–0.09	0.04	58
Incidence of colorectal cancer
Baron,61,73,74 Logan60	Any aspirin (81–325 mg/day) (± folic acid 0.5–1 mg/day)	No aspirin (± folic acid)	CRC	2	3,3	8	1155	8	782	0.84	0.15	4.74	49	0.00	–0.02	0.01	72

Recurrence of any adenoma

In the intermediate-risk population (history of adenomas or CRC), all four studies (2692 analysed participants) compared aspirin to no aspirin (50% of participants in the aspirin and placebo arms also received folic acid in the Logan and Cole studies); see Figure 3ai. 60,61 A meta-analysis demonstrated a statistically significant 21% reduction in the relative risk of recurrence of an adenoma of any type (RR 0.79, 95% CI 0.68 to 0.92, p = 0.002), with a moderate level of statistical heterogeneity (I² = 34%). There was a 7% reduction in the absolute risk [meta-analysed risk difference (RD) –0.07, 95% CI –0.11 to –0.04, p < 0.0001], with no statistical heterogeneity (I² = 0%).

These four studies also provided data comparing aspirin to placebo alone (n = 1662, excluding those participants who also received folic acid; Figure 3aii). The relative risk for adenoma recurrence remained similar and statistically significant at the 5% level (meta-analysed RR 0.80, 95% CI 0.65 to 0.98, p = 0.03) with a moderate level of statistical heterogeneity (I² = 47%), and the absolute risk difference remained similar.

In addition, two of the studies (Logan and Cole; 916 participants in analysis) compared aspirin plus folic acid with placebo (Figure 3aiii). 60,61 The differences in the relative and absolute risks of developing an adenoma of any type were not statistically significant at the 5% level (meta-analysed RR 0.90, 95% CI 0.75 to 1.08, p = 0.27; RD –0.04, 95% CI –0.10 to 0.02, p = 0.23), and there was no statistical heterogeneity between the studies (I² = 0%).

Incidence of advanced adenoma

Three of the studies (Logan, Cole and Benamouzig)60–62 reported incidence rates for advanced adenomas (generally defined as ≥ 1 cm in diameter, villous or tubulovillous, having severe dysplasia, or CRC; see Figure 3bi). When analysing aspirin versus no aspirin (n = 2175, including some participants who also received folic acid), a meta-analysis demonstrated a statistically significant 34% reduction in the relative risk of advanced adenoma incidence (RR 0.66, 95% CI 0.51 to 0.84, p = 0.0008) and a 5% reduction in the absolute risk (RD –0.05, 95% CI –0.07 to –0.02, p = 0.0008), with no statistical heterogeneity between the studies (I² = 0% for both analyses).

When comparing aspirin to placebo alone (three studies; n = 1662, excluding participants who also received folic acid, Figure 3bii), the risk of advanced adenoma incidence remained lower in the aspirin group but was no longer significant at the 5% level (meta-analysed RR 0.75, 95% CI 0.52 to 1.07, p = 0.12; RD –0.03, 95% CI –0.06 to 0.01, p = 0.14) with no heterogeneity (I² = 0% for both analyses).

In addition, two of the studies (Logan and Cole; 916 participants in analysis) compared aspirin plus folic acid versus placebo (Figure 3biii). 60,61 The differences in the relative risk and absolute risk of developing an advanced adenoma were not statistically significant at the 5% level (meta-analysed RR 0.77, 95% CI 0.45 to 1.34, p = 0.36; RD –0.03, 95% CI –0.09 to 0.04, p = 0.41). The meta-analysis indicated moderate (I² = 45%) and high (I² = 58%) levels of statistical heterogeneity for the relative risk and absolute risk analyses, respectively.

Incidence of colorectal cancer

Two of the studies comparing aspirin to no aspirin also reported incidence of CRC (Logan and Cole; n = 1937, 50% of participants also received folic acid; Figure 3c). 60,61 The differences in the relative risk and the absolute risk of CRC were not significant at the 5% level (meta-analysed RR 0.84, 95% CI 0.15 to 4.74, p = 0.84; RD 0.00, 95% CI –0.02 to 0.01, p = 0.82) with moderate-to-high heterogeneity (I² = 49% and 72%, respectively). As these studies were primarily designed to assess adenoma recurrence, analysis of cancer incidence is limited by the relatively short follow-up time (3 years) and the low number of CRC cases (n = 16 cases across both studies) and these results should therefore be treated with caution.

Compliance

Three of the four studies (Cole, Benamouzig and Sandler)61,62,79 selected the most compliant participants during an aspirin run-in phase before randomisation, which may increase the observed compliance (and therefore effectiveness) relative to non-trial populations. Measures of compliance differed between studies, but in all three of the above studies, 87–90% of participants reported taking the majority of their study medications at the latest time point mentioned. In the Logan study (which did not report selecting for compliant participants),60 69–77% of participants in all study arms continued to take study medication (i.e. 23–31% discontinued medication), and during the time spent on medication, 76% reported taking the majority of their medications. Benamouzig reported no differences in compliance according to gender or to increased risk of CRC (e.g. because of family history, previous adenomas, adenoma number or size). 62

Adverse effects

In terms of the studies of aspirin for adenoma prevention, event numbers were not sufficiently large to allow for a meaningful analysis of adverse event rates. 60,61,79 Dyspepsia or gastric upset rates were similar between the groups in the Logan study. 60 Included studies in the intermediate-risk population also reported data on bleeding events, gastrointestinal bleeding, peptic ulcers and cardiovascular adverse events, but event numbers were small. 60,61,79

Larger studies and reviews of the adverse effects of aspirin are discussed in the section that relates to aspirin use in the low-risk population.

Ongoing studies

An ongoing multicentre study is assessing aspirin, calcium and vitamin D (as a combination treatment) versus placebo for prevention of adenomas in participants with a history of adenomas, and aims to enrol 1000 participants, with a treatment duration of 3 years and follow-up at 3 and 5 years. The dose of aspirin is lower than that in many chemoprevention studies, at 75 mg/day (www.clinicaltrials.gov, NCT00486512).

Excluded studies

No additional studies of aspirin in the intermediate-risk population were excluded on final examination of the full text.

Aspirin: low-risk population (general population or no increased risk of colorectal cancer)

Outcome data

Effectiveness results for aspirin in the general population are shown in Figure 4 and Table 10.

FIGURE 4.

Aspirin: meta-analyses (general population or low-risk population). Incidence of colorectal cancer: aspirin alone (or ±antioxidants) versus placebo.

TABLE 10 - Aspirin: Summary of results (general population or low-risk population)

C, control; I, intervention; FU, follow-up; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2; (%)	RD	LCI	UCI	I2 (%)
Incidence of colorectal cancer
Cook,68–70 Gann,75 Farrell,21,77 Peto21,80	Any aspirin (100 mg every other day to 1500 mg/day) (± vitamin E ± beta-carotene)	No aspirin (± vitamin E ± beta-carotene)	CRC, ≤ 10 years FU	4	5,10,10,12	242	36,032	216	33,503	1.01	0.84	1.21	0	0.00	0.00	0.00	0
Farrell,21,77 Peto21,80	Aspirin alone (300–1500 mg/day)	Placebo alone	CRC, 10 years FU	2	10,10	46	5061	25	2527	0.92	0.56	1.49	0	0.00	–0.01	0.00	0
Farrell,21,77 Peto21,80	Aspirin alone (300–1500 mg/day)	Placebo alone	CRC, > 20 years FU	2	23,23	129	5061	87	2527	0.74	0.57	0.97	0	–0.01	–0.02	0.00	0
Farrell,21,77 Peto21,80	Aspirin alone (300–1500 mg/day)	Placebo alone	CRC, 10–19 years FU	2	Years 10–19 only	65	5061	53	2527	0.61	0.43	0.88	0	0.00	0.00	0.00	0

Effectiveness

Incidence of any adenoma

Of all studies assessing the use of aspirin within the general population (no increased risk for CRC), only one study (Gann; n = 22,071) reported adenoma incidence. 75 The reduction in relative risk for adenoma incidence was not significant at the 5% level (RR 0.86, 95% CI 0.68 to 1.10, p = 0.24). No trend over time for adenoma incidence was noted over the 5 years. This outcome was measured as self-reported adenomas or cancer in situ, which contrasts with the studies in the intermediate-risk population in which adenomas were assessed via colonoscopy. The results should therefore be interpreted with caution.

Incidence of advanced adenoma

No studies of aspirin in the general population assessed advanced adenoma incidence.

Incidence of colorectal cancer

Available data on the incidence of CRC is described here according to duration of follow-up. The two larger studies of lower-dose aspirin (Cook and Gann) reported CRC incidence after 10 and 5 years of follow-up respectively. 68,75 The two smaller studies of higher-dose aspirin (Farrell and Peto)77,80 were analysed in the report by Flossmann and Rothwell21 for CRC incidence over the first 10-year period since start of treatment, the second 10-year period (i.e. years 10–19), and the entire 23-year follow-up period.

The meta-analysis of data from all four studies comparing aspirin to no aspirin in the general population (n = 69,535) indicated no effect of aspirin on CRC incidence over a follow-up period of 10 years or less (RR 1.01, 95% CI 0.84 to 1.21, p = 0.92; RD 0.00, 95% CI 0.00 to 0.00, p = 0.82). The aspirin dose in these studies ranged from 100 mg every other day to 1200 mg/day. Data from the two smaller studies with a higher aspirin dose (Farrell and Peto; n = 7588; 300–1200 mg/day aspirin)77,80 was used to assess the effect of aspirin alone versus placebo, as the two larger studies also prescribed antioxidants to 50% of participants. The effect on CRC incidence over the first 10 years remained non-significant (RR 0.92, 95% CI 0.56 to 1.49, p = 0.72; RD 0.00, 95% CI –0.01 to 0.00, p = 0.72). There was no statistical heterogeneity in any of the above analyses (I² = 0%).

However, when data from the two smaller studies (Farrell and Peto; n = 7588)77,80 with a higher aspirin dose (300–1200 mg/day) were analysed over a follow-up period of up to 23 years, there was a statistically significant 26% reduction in the relative risk of CRC incidence with aspirin versus placebo (RR 0.74, 95% CI 0.57 to 0.97, p = 0.03), and a 1% reduction in absolute risk (RD –0.01, 95% CI –0.02 to 0.00, p = 0.04). Interestingly, when this analysis was restricted to the second 10-year follow-up period only (i.e. years 10–19), there was an even greater 39% reduction in relative risk (RR = 0.61, 95% CI 0.43 to 0.88, p = 0.008). There was no statistical heterogeneity in these analyses (I² = 0%).

Compliance

Two of the four studies (Cook and Gann) selected the most compliant participants during an aspirin or placebo run-in phase before randomisation,68,75 which may increase the observed compliance (and therefore effectiveness) relative to non-trial populations. Measures of compliance differed between studies, but in all four studies 70–88% of participants reported taking the majority of their study medications at the latest time point mentioned. In terms of changes in compliance over time, Cook reported that 76% of participants were compliant after 5 years, decreasing to 67% after 10 years. 68 The non-blinded study by Peto reported that 81% were compliant after 1 year but that a further 5% discontinued study aspirin during each of the next 5 years, mainly as a result of gastrointestinal symptoms. 80

Adverse effects

Adverse effect data from RCTs included in this review

Dyspepsia or gastric upset was similar between the groups in the two larger studies (Cook and Gann). 68,75 The incidence of peptic ulcers was reported to be statistically significantly higher in the aspirin group than the control group in the studies by Cook (2.7% vs 2.1%; p < 0.001)68 and Peto (p < 0.05),80 and non-significantly higher in the study by Gann (1.5% vs 1.3%; p = 0.08). 75

Gastrointestinal bleeding was statistically significantly higher in the aspirin group in one large study (Cook; 4.6% vs 3.8%; p < 0.001),68 and slightly higher in the aspirin group in the two smaller studies (one reported as not significant),77,80 whereas severe gastrointestinal bleeding was also reported to be statistically significantly higher in the aspirin group than (p = 0.02) within the study reported by Cook. 68 Gann reported bleeding events overall to be statistically significantly higher in the aspirin group (p < 0.0001). 75

No studies reported a significant between-group difference in overall incidence of stroke (Gann and Farrell)75,77 or in incidence of haemorrhagic stroke (Cook and Gann). 68,75 However, fatal strokes 77 and major/fatal haemorrhagic strokes75,77 were statistically significantly higher in the aspirin group in the studies by Farrell and Gann, although event numbers were small.

Systematic reviews of adverse effects (and benefits) of aspirin

There is a large existing literature regarding adverse effects (and benefits) associated with the use of aspirin. A previous systematic review of aspirin for chemoprevention of CRC prepared by Dube et al. for the US Preventive Services Task Force provides a summary of 12 good-quality systematic reviews assessing adverse effects and health benefits associated with aspirin use. 76 All-cause mortality and mortality due to cardiovascular events were not statistically significantly affected when aspirin was given as primary prevention, yet both outcomes were reported to be statistically significantly lower for aspirin use in the secondary prevention of cardiovascular disease. Myocardial infarction was significantly reduced in both settings. Stroke incidence was not statistically significantly affected in the primary prevention setting, while in the secondary prevention of cardiovascular disease, aspirin appeared to significantly reduce the risk of ischaemic stroke but to significantly increase the risk of haemorrhagic stroke. 76

In the review by Dube et al. , aspirin was consistently associated with statistically significant increases in the risk of gastrointestinal bleeding (RR 1.6–2.5 in systematic reviews of RCTs) and in the risk of adverse gastrointestinal symptoms such as nausea and dyspepsia. A dose effect has been suggested for aspirin-induced gastrointestinal toxicity. In one systematic review, the risk of gastrointestinal bleeding was 2.5% in patients taking more than 100 mg aspirin per day, and 1.1% in patients taking less than 100 mg/day. In another, the risk of gastrointestinal bleeding was higher in patients taking 1600 mg/day than in those taking 300 mg/day. 76

A recent meta-analysis of aspirin in the prevention of cardiovascular disease analysed six primary prevention trials (n = 95,000 individuals at low cardiovascular risk) and 16 secondary prevention trials (n = 17,000 individuals at high cardiovascular risk, i.e. history of myocardial infarction or stroke). 145 In the primary prevention trials, the relative risk of serious vascular events per person-year was reduced by 12% [incidence of 0.51% per year in the aspirin groups and 0.57% per year in the control groups; RR 0.88 (95% CI 0.82 to 0.94); RD 0.07% per year]. This was largely the result of a reduction of approximately one-fifth in the relative risk of non-fatal myocardial infarction. There was no significant difference in the risk of stroke or deaths from vascular causes. However, the risk of internal bleeding increased by approximately one-third in the primary prevention trials. In the secondary prevention trials, the relative risk of serious vascular events per person-year was reduced by 19% (incidence of 6.7% vs 8.2% per year; RR 0.81, 95% CI 0.75 to 0.87, RD 1.49% per year). There were reductions of approximately one-fifth in the relative risk of coronary events and total stroke, but a non-significant increase in risk of haemorrhagic stroke. Therefore, it was concluded that the benefits of aspirin outweighed the risk of harm in secondary prevention of cardiovascular disease, but not necessarily in primary prevention.

Ongoing studies

No ongoing studies of aspirin in the low-risk population were identified.

Excluded studies

No additional studies of aspirin in the low-risk population were excluded on final examination of the full text.

Summary of results for aspirin

Four studies of aspirin (81–325 mg/day) in individuals with a history of adenomas (three studies) or history of CRC (one study) demonstrated a statistically significant 21% reduction in the relative risk of adenoma recurrence. Incidence of advanced adenomas also showed a significant 34% reduction when comparing aspirin versus no aspirin (this was no longer significant when comparing aspirin alone with placebo alone). Aspirin combined with folic acid produced a non-statistically-significant reduction in adenomas and advanced adenomas. Of the four studies of aspirin in the general population, two large studies giving a relatively low dose of aspirin (100–325 mg every other day) showed no effect on CRC over a follow-up period of 5–12 years. Two smaller studies in which a higher dose of aspirin was administered (300–1500 mg/day) showed no effect on CRC over the first 10 years of follow-up, but demonstrated a significant 26% reduction in CRC incidence over the full 23-year follow-up period, and an even greater reduction when analysing years 10–19 only. In addition, aspirin (600 mg/day) was assessed in FAP patients in one study and in HNPCC patients in another study; both studies reported no statistically significant reduction in adenoma incidence. Aspirin is associated with an increased risk of upper gastrointestinal toxicity including nausea and dyspepsia, peptic ulcers and gastrointestinal bleeding, as demonstrated in the larger studies included here and in a review which collated systematic reviews of adverse effects of aspirin,76 with higher aspirin doses being associated with greater risk of toxicity. Previous studies, particularly in individuals at higher risk of cardiovascular disease, have indicated that aspirin may reduce the risk of myocardial infarction and ischaemic stroke but increase the risk of haemorrhagic stroke.

Non-aspirin NSAIDs: high-risk population (FAP or HNPCC)

Non-aspirin NSAIDs: intermediate-risk population (history of adenomas)

Outcome data

Effectiveness results for non-aspirin NSAIDs in the intermediate-risk population (history of adenomas) are shown in Figure 5 and Table 14. The Ladenheim study90 did not report the recurrence of adenoma or CRC (only adenoma regression) and was therefore excluded from meta-analyses.

FIGURE 5.

Non-aspirin NSAIDs: meta-analyses (history of adenomas). (a) Recurrence of any adenoma: celecoxib versus placebo; (b) Incidence of advanced adenoma: celecoxib versus placebo.

TABLE 14 - Non-aspirin NSAIDs: summary of results (history of adenomas)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Recurrence of any adenoma
Arber,63,84 Bertagnolli64,84,86,87	Celecoxib alone (400 mg/day)	Placebo alone	Adenoma	2	3	N/A	1453	N/A	1165	0.66	0.60	0.72	0	N/C	N/C	N/C	N/C
Incidence of advanced adenoma
Arber,63,84 Bertagnolli64,84,86,87	Celecoxib alone (400 mg/day)	Placebo alone	Advanced adenoma	2	3	N/A	1453	N/A	1165	0.45	0.35	0.58	0	N/C	N/C	N/C	N/C
Incidence of colorectal cancer
Arber63,84	Celecoxib alone (400 mg/day)	Placebo alone	CRC	1	3	6	933	1	628	4.03	0.48	33.4	N/C	0.00	0.00	0.01	N/C

Effectiveness

Recurrence of any adenoma

In the intermediate-risk population (history of adenomas or CRC), two studies (Arber and Bertagnolli; n = 2618 participants in analysis) compared celecoxib with placebo (Figure 5a). 63,64 A meta-analysis demonstrated a statistically significant 34% reduction in the relative risk of recurrence of an adenoma of any type (RR 0.66, 95% CI 0.60 to 0.72, p < 0.00001), with no statistical heterogeneity (I² = 0%). This was based on a meta-analysis of reported relative risks rather than on raw event data, because the study authors calculated the relative risk using data from both the 1-year and 3-year time points. Therefore, it was not possible to calculate an absolute risk difference.

Ladenheim90 reported the effect of sulindac compared with placebo on the regression of colorectal polyps ≤ 1 cm in diameter which were left in situ over the 4-month study duration. The number of participants whose polyps disappeared or regressed significantly was five of 22 (23%) in the sulindac group and three of 22 (14%) in the placebo group (risk ratio 1.67, 95% CI 0.45 to 6.14).

Incidence of advanced adenoma

Two studies (Arber and Bertagnolli; n = 2618 participants) also reported incidence rates for advanced adenomas (generally defined as adenoma ≥  1 cm in diameter, villous or tubulovillous adenoma, adenoma with severe dysplasia, or CRC; Figure 5b). 63,64 A meta-analysis demonstrated a statistically significant 55% reduction in the relative risk of advanced adenoma incidence (RR 0.45, 95% CI 0.35 to 0.58, p < 0.00001), with no statistical heterogeneity (I² = 0%).

Incidence of colorectal cancer

One study (Arber)63 also reported incidence of CRC, with no significant difference in relative risk at the 5% level (RR 4.03, 95% CI 0.48 to 33.4). These data should be treated with caution because of the relatively short follow-up time in terms of detecting cancer (3 years) and the low number of CRC cases (six in the celecoxib arm and one in the placebo arm).

Compliance

Both studies selected the most compliant participants during a placebo run-in phase before randomisation, which may increase the compliance estimates relative to non-trial populations. In both studies, 68–79% of participants reported taking the majority of their study medications, with similar compliance between arms in each study.

Adverse effects

Adverse effect data from RCTs included in this review

In terms of the celecoxib studies included in this review, serious cardiovascular events were more frequent in the celecoxib groups than the placebo groups; this was significant in the Bertagnolli study64 whereas there was a non-significant increase in the Arber study. 63 Both trials were stopped early because of this increased cardiovascular risk. Rates of gastrointestinal ulceration or haemorrhage were slightly, but not significantly, higher in the celecoxib groups in the Arber study,63 but Bertagnolli reported no difference between the groups. 64 Renal or hypertensive disorders were significantly higher in the celecoxib group in the Arber study63 and in one of the two celecoxib groups in the Bertagnolli study. 64 In the Ladenheim study of sulindac, four of 22 patients (18%) discontinued sulindac because of adverse events, three of which were thought to be related to sulindac (one anaemia, two moderate-to-severe heartburn) and one was thought to be unrelated (urosepsis); none of the 22 (0%) patients discontinued placebo. 90

Systematic reviews of adverse effects of NSAIDs

There is a large existing literature regarding adverse effects of NSAIDs. A previous systematic review of non-aspirin NSAIDs for chemoprevention of CRC prepared by Rostom et al. on behalf of the US Preventive Services Task Force provides a summary of systematic reviews assessing adverse effects associated with NSAID use. 88 These data are summarised below.

All-cause and cardiovascular-specific mortality were not shown to be statistically significantly affected in the majority of reviews. Three reviews demonstrated an increased risk of serious cardiovascular events associated with COX-2 inhibitors; this risk was greatest in patients at higher cardiovascular risk. Six reviews demonstrated statistically significant increases in risk of acute myocardial infarction with COX-2 inhibitors. Five reviews demonstrated no statistically significant increase in risk of stroke with COX-2 inhibitors. Two COX-2 inhibitors, rofecoxib and valdecoxib, were recently withdrawn from use because of concerns about their cardiovascular toxicity. The risks of hypertension and renal toxicity may also be increased with COX-2 inhibitors. 88

In terms of NSAIDs as a whole, the systematic reviews summarised by Rostom et al. 88 reported increased risk of peptic ulceration and gastrointestinal haemorrhage with non-aspirin NSAID use. The increase in absolute risk of complicated peptic ulcers (perforation, obstruction or bleeding) was 0.48% across RCTs and 0.22% across cohort studies, with a higher risk in individuals who were older or had a history of peptic ulcers or cardiovascular disease. The risk for upper gastrointestinal toxicity as a result of NSAID use can be reduced through use of a gastroprotective agent such as misoprostol. Twelve systematic reviews assessed gastrointestinal adverse effects of COX-2 inhibitors specifically (as opposed to all NSAIDs). Several of these reviews demonstrated no statistically significant difference in gastrointestinal bleeding or ulceration with COX-2 inhibitors compared with placebo. One review demonstrated a statistically significant increase in risk of ulcers in patients receiving celecoxib at a dose of 400 mg/day, but not at a dose of 200 mg/day. A trial of rofecoxib reported an increase in risk of peptic ulceration. 88

Ongoing studies

An ongoing phase 3 RCT is assessing celecoxib versus placebo for prevention of adenoma incidence in participants with a history of resected stage 1 colon cancer (adenocarcinoma), and aims to enrol 1200 participants, with a treatment duration of 3 years; follow-up duration not specified (www.cancer.gov/clinicaltrials/NSABP-P-3; www.clinicaltrials.gov NCT00087256).

Excluded studies

A study by Meyskens et al. assessed sulindac plus DFMO (or eflornithine) versus placebo in participants with a history of adenomas. 154 DFMO inhibits synthesis of polyamines which are involved in cell proliferation, inhibits colon carcinogenesis in animal models, and may act additively with sulindac. This study was excluded because the availability of oral DFMO is restricted (DFMO is available as a cream for hair removal).

A study by Baron et al. of rofecoxib versus placebo in participants with a history of adenomas153 was excluded because its licence has been revoked and it has been withdrawn from the market. 47

Non-aspirin NSAIDs: low-risk population (general population or no increased risk of colorectal cancer)

No relevant published or ongoing studies were identified investigating NSAID use in populations without increased risk of CRC.

Summary of results for non-aspirin NSAIDs

In terms of non-aspirin NSAIDs, two studies of celecoxib (400 mg/day) in individuals with a history of adenomas demonstrated a statistically significant 34% reduction in the relative risk of adenoma recurrence and a statistically significant 55% reduction in the relative risk of advanced adenoma incidence. There were insufficient data to analyse the effect on CRC incidence. A small study of sulindac in patients with the FAP genotype showed a non-statistically significant reduction in adenoma incidence, whereas five studies of NSAIDs (sulindac, celecoxib or tiracoxib) in FAP patients with existing adenomas demonstrated reductions in polyp number and size, some statistically significant. The two celecoxib trials in individuals with a history of adenomas were stopped early because of an increased risk of serious cardiovascular events, which was statistically significant in one of the studies. A review that collated systematic reviews of adverse effects88 also demonstrated increased risk of serious cardiovascular events with COX-2 inhibitors, the risk being greatest in patients with pre-existing cardiovascular risk factors. Two COX-2 inhibitors, rofecoxib and valdecoxib, were recently withdrawn from use because of concerns about their cardiovascular toxicity; a study of rofecoxib was therefore excluded from this review. COX-2 inhibitors may also increase the risks of hypertension and renal toxicity. NSAIDs can also cause upper gastrointestinal toxicity, although the risk is lower for COX-2 inhibitors than for some other types of NSAID.

Folic acid: high-risk population (FAP or HNPCC)

No studies were identified that investigated the use of folic acid as a chemopreventive agent within high-risk populations for CRC (i.e. FAP and HNPCC).

Folic acid: intermediate-risk population (history of adenomas)

Outcome data

Effectiveness

Effectiveness results for folic acid in the intermediate risk population (history of adenomas) are shown in Figure 6 and Table 16.

FIGURE 6.

Folic acid: meta-analyses (history of adenomas). (a) Recurrence of any adenoma: (i) folic acid ± aspirin versus placebo ± aspirin, (ii) folic acid alone versus placebo alone, (iii) folic acid + aspirin versus placebo alone; (b) Incidence of advanced adenoma: (i) folic acid ± aspirin versus placebo ± aspirin, (ii) folic acid alone versus placebo alone, (iii) folic acid + aspirin versus placebo alone; (c) Incidence of colorectal cancer: folic acid ± aspirin versus placebo ± aspirin.

TABLE 16 - Folic acid: summary of results (history of adenomas)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Recurrence of any adenoma
Cole,61,73,74 Logan60	Any folic acid (0.5–1 mg/day) (± aspirin 81–325 mg/day)	No folic acid (± aspirin)	Adenoma (two high-quality studies)	2	3	336	933	311	907	1.05	0.93	1.18	0	0.02	–0.03	0.06	0
Cole,61,73,74 Logan60	Folic acid alone (0.5–1 mg/day)	Placebo alone	Adenoma (two high-quality studies)	2	3	152	383	126	366	1.16	0.97	1.39	0	0.05	–0.02	0.12	0
Cole,61,73,74 Logan60	Folic acid (0.5–1 mg/day) + aspirin (81–325 mg/day)	Placebo alone	Adenoma	2	3	184	550	126	366	0.90	0.75	1.08	0	–0.04	–0.10	0.02	0
Incidence of advanced adenoma
Cole,61,73,74 Logan60	Any folic acid (0.5–1 mg/day) (± aspirin 81–325 mg/day)	No folic acid (± aspirin)	Advanced adenoma	2	3	109	933	94	907	1.13	0.84	1.51	21	0.01	–0.02	0.04	8
Cole,61,73,74 Logan60	Folic acid alone (0.5–1 mg/day)	Placebo alone	Advanced adenoma	2	3	60	383	44	366	1.34	0.77	2.36	55	0.04	–0.03	0.11	46
Cole,61,73,74 Logan60	Folic acid (0.5–1 mg/day) + aspirin (81–325 mg/day)	Placebo alone	Advanced adenoma	2	3	49	550	44	366	0.77	0.45	1.34	45	–0.03	–0.09	0.04	58
Incidence of colorectal cancer
Cole,61,73,74 Logan60	Any folic acid (0.5–1 mg/day) (± aspirin 81–325 mg/day)	No folic acid (± aspirin)	CRC	2	3	8	948	9	926	0.87	0.34	2.25	0	0.00	–0.01	0.01	0

The Jaszewski study did not report event data (only mean numbers of adenomas) and was therefore excluded from meta-analysis. 108 The Cole study reported follow-up data for two intervals (years 1–3 and years 4–8). Only the results from the first interval (3 years follow-up) are presented here, because only 607 of 1021 randomised patients (59%) agreed to be followed up beyond 3 years, and only 501 patients (49%) agreed to continue taking study medications beyond 3 years.

Recurrence of any adenoma

Two studies (Logan and Cole; 1840 participants in analysis) compared folic acid (with or without aspirin) versus no folic acid (Figure 6ai). 60,61 The relative risk and absolute risk difference for developing an adenoma of any type were not statistically significant at the 5% level (meta-analysed RR 1.05, 95% CI 0.93 to 1.18, p = 0.45; RD 0.02, 95% CI –0.03 to 0.06, p = 0.43), and there was no statistical heterogeneity (I² = 0%).

Two studies (Logan and Cole; 749 participants in analysis) compared folic acid alone versus placebo (Figure 6aii). 60,61 The relative risk and absolute risk difference for developing an adenoma of any type were not statistically significant at the 5% level, as demonstrated in a meta-analysis (RR 1.16, 95% CI 0.97 to 1.39, p = 0.11; RD 0.05, 95% CI –0.02 to 0.12, p = 0.14), and there was no statistical heterogeneity (I² = 0%).

Two studies (Logan and Cole; 916 participants in analysis) compared folic acid plus aspirin versus placebo (Figure 6aiii). 60,61 The relative risk and absolute risk difference for developing an adenoma of any type were not statistically significant at the 5% level, as demonstrated in a meta-analysis (RR 0.90, 95% CI 0.75 to 1.08, p = 0.27; RD –0.04, 95% CI –0.10 to 0.02, p = 0.23), and there was no statistical heterogeneity (I² = 0%).

A third, lower-quality, smaller study (Jaszewski; 94 participants in analysis) reported that the mean number of recurrent adenomas was significantly higher in the placebo group (OR 2.77, 95% CI 0.06 to 0.84, p = 0.025). 108

Incidence of advanced adenoma

Two studies (Logan and Cole; 1840 participants in analysis) compared folic acid (with or without aspirin) versus no folic acid (Figure 6bi). 60,61 The relative and absolute risk difference between groups for developing an advanced adenoma were not statistically significant at the 5% level (meta-analysed RR 1.13, 95% CI 0.84 to 1.51, p = 0.43; RD 0.01, 95% CI –0.02 to 0.04, p = 0.034), and these suggested moderate and low levels of statistical heterogeneity (I² = 21% and 8%) respectively.

Two studies (Logan and Cole; 749 participants in analysis) compared folic acid alone versus placebo (Figure 6bii). 60,61 The relative and absolute risk differences between groups for developing an advanced adenoma were not statistically significant at the 5% level (meta-analysed RR 1.34, 95% CI 0.77 to 2.36, p = 0.30; RD 0.04, 95% CI –0.03 to 0.11, p = 0.24). These were subject to high (I² = 55%) and moderate (I² = 46%) levels of statistical heterogeneity respectively.

Two studies (Logan and Cole; 916 participants included in the analysis) compared folic acid plus aspirin versus placebo (Figure 6biii). 60,61 The relative and absolute risk differences between groups for developing an advanced adenoma were not statistically significant at the 5% level (meta-analysed RR 0.77, 95% CI 0.45 to 1.34, p = 0.36; RD –0.03, 95% CI –0.09 to 0.04, p = 0.41). These analyses suggested moderate (I² = 45%) and high (I² = 58%) levels of statistical heterogeneity respectively.

Incidence of colorectal cancer

Two studies (Logan and Cole; 1874 participants included in the analysis) compared folic acid with or without aspirin versus controls without folic acid (Figure 6c). 60,61 The differences in the relative and absolute risks of developing CRC were not statistically significant at the 5% level (meta-analysed RR 0.87, 95% CI 0.34 to 2.25, p = 0.77; RD 0.00, 95% CI –0.01 to 0.01, p = 0.72), and there was no statistical heterogeneity (I² = 0%). These results should be interpreted with caution because of the relatively short follow-up duration and the low event rate (total of eight cases in the folic acid arms and nine cases in the control arms).

Compliance

Two studies (Logan and Cole) reported a compliance rate of between 87 and 90%. 60,61 The Jaszewski study required at least 90% of pills to be taken, but results were not reported for the trial period. 108 Compliance was measured by periodic self-report in two studies60,61 and pill counting in two studies. 60,108

Adverse effects

Two studies (Logan and Cole) reported on serious adverse events. 60,61 Neither study reported any significant differences between folate and non-folate groups in terms of death, bleeding, stroke, myocardial infarction, vascular events or dyspepsia. The Logan study reported no difference between groups for non-CRC but the Cole study did report a significant difference: incidence of non-CRC was higher in the folic acid (54/516) than the placebo group (32/505), p = 0.02. However, authors reported that this was a result of the high baseline rate of prostate cancer in the folic acid group (24/516 vs 9/505, p = 0.01).

Ongoing studies

Two ongoing studies are assessing folic acid for prevention of adenomas in participants with a history of adenomas (www.clinicaltrials.gov, NCT00512850 and NCT00002650). One aims to enrol 1000 participants and has a planned follow-up duration of 4 years (NCT00512850). Another small study (80 participants planned) aims to assess the effect of high-dose folic acid (dose not specified), with treatment and follow-up durations of 1 year (NCT00002650).

Excluded studies

No additional studies of folic acid in the intermediate risk population were excluded on final examination of the full text.

Folic acid: low-risk population (general population or no increased risk of colorectal cancer)

Outcome data

Effectiveness

Effectiveness results for folic acid in the general population (or populations with no increased risk for CRC) are shown in Figure 7 and Table 18.

FIGURE 7.

Folic acid: meta-analyses (general population or low-risk population). Incidence of colorectal cancer: (a) folic acid + vitamin B12 ± vitamin B6 ± antioxidants versus placebo ± antioxidants: low risk, (b) folic acid + vitamin B12 + vitamin B6 + antioxidants versus placebo + antioxidants: low risk.

TABLE 18 - Folic acid: summary of results (general population or low-risk population)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Incidence of colorectal cancer
Lonn,106,107 Zhang,105 Zhu109,110	Any folic acid (2.5 mg/day to 20 mg/day) (+ B vitamins ± antioxidants)	No folic acid (± antioxidants)	CRC	3	5 + 6 + 7	68	5523	60	5539	1.13	0.77	1.64	7	0.00	0.00	0.01	36
Lonn,106,107 Zhang105	Folic acid (2.5 mg/day) + vitamin B6 + vitaminB12 + antioxidants	Placebo + antioxidants	CRC	2	5 + 7	68	5479	59	5485	1.15	0.82	1.63	42	0.00	0.00	0.01	63
Zhu109,110	Folic acid alone (20 mg/day for 1 year then 20 mg twice weekly for next year)	Placebo alone	CRC	1	6	0	44	1	54	0.41	0.02	9.76	N/C	–0.02	–0.07	0.03	N/C

Incidence of any adenoma

No studies of folic acid in the general population reported adenoma incidence.

Incidence of advanced adenoma

No studies of folic acid in the general population reported advanced adenoma incidence.

Incidence of colorectal cancer

Three studies (11,062 participants in analysis) compared folic acid plus B vitamins (with or without antioxidants) versus placebo (± antioxidants; Figure 7a). 105,106,109 A meta-analysis demonstrated that the relative and absolute risk differences between groups for developing CRC were not statistically significant at the 5% level (RR 1.13, 95% CI 0.77 to 1.64, p = 0.54; RD 0.00, 95% CI 0.00 to 0.01, p = 0.76). There were low (I² = 7%) and moderate (I² = 36%) levels of statistical heterogeneity, respectively.

Two studies (Zhang and Lonn; 10,964 participants in analysis) compared folic acid plus B vitamins plus antioxidants versus placebo (± antioxidants; Figure 7b). 105,106 The relative and absolute risk differences between groups for developing CRC were not statistically significant at the 5% level, as demonstrated in a meta-analysis (RR 1.15, 95% CI 0.82 to 1.63, p = 0.42; RD 0.00, 95% CI 0.00 to 0.01, p = 0.42). The analyses of relative and absolute risk suggested moderate (I² = 42%) and high (I² = 63%) levels of statistical heterogeneity, respectively.

One study (Zhu; 98 participants in analysis) compared folic acid plus B vitamins versus placebo alone. 109 There was one case of CRC among 54 participants receiving placebo, and 0 cases among 44 participants receiving folic acid.

Compliance

Two studies (Lonn and Zhu) reported that at least 90% of pills were taken as required in the intervention groups over the treatment period. 106,109 Compliance was measured by pill counting in two studies (Lonn and Zhu), by self-report in two studies (Lonn and Zhang) and serum concentration levels in one study (Zhu). In the Zhang study, compliance, defined as taking at least 66% of pills, was 83% over the study period by the intervention group. 105

Adverse effects

None of the studies reported any serious adverse events associated with the study treatments.

Ongoing studies

No ongoing studies of folic acid in the low-risk population were identified.

Excluded studies

No additional studies of folic acid in the low-risk population were excluded on final examination of the full text.

Summary of results for folic acid

Two studies of folic acid (0.5–1.0 mg/day) in individuals with a history of adenomas showed no statistically significant difference in the relative risk of adenoma recurrence or advanced adenoma incidence, with the event rates actually being slightly higher in the folic acid groups. Three studies of folic acid (2.5 mg/day in two studies and 20 mg/day in one study) in populations with no increased baseline risk of CRC demonstrated no statistically significant effect of folic acid on the relative risk of CRC, with the rates being slightly higher in individuals receiving folic acid. However, the duration of follow-up was 5–7 years, which may not be long enough to detect an effect on cancer incidence. No studies reported any difference in serious adverse event rates between the folic acid and placebo groups (except for one study reporting a higher incidence of cancers other than those of the colorectum in the folic acid group, which was thought to be the result of the higher baseline rate of prostate cancer in that group).

Calcium and/or vitamin D: high-risk population (FAP or HNPCC)

Calcium and/or vitamin D: intermediate-risk population (history of adenomas)

Outcome data

Effectiveness

Effectiveness results for calcium in the intermediate-risk population (history of adenomas) are shown in Figure 8 and Table 21.

FIGURE 8.

Calcium: meta-analyses (history of adenomas). (a) Recurrence of any adenoma: calcium alone versus placebo alone; (b) Incidence of advanced adenoma: calcium alone versus placebo alone; (c) Incidence of colorectal cancer: calcium alone versus placebo alone.

TABLE 21 - Calcium: summary of results (history of adenomas)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Recurrence of any adenoma
Baron,99,100 Bonithon-Kopp96	Calcium alone (1200–2000 mg/day)	Placebo alone	Adenoma	2	3	155	585	195	601	0.82	0.69	0.98	0	–0.06	–0.11	–0.01	0
Incidence of advanced adenoma
Baron,99,100 Bonithon-Kopp96	Calcium alone (1200–2000 mg/day)	Placebo alone	Advanced adenoma	2	3	35	585	47	601	0.77	0.50	1.17	0	–0.01	–0.05	0.03	57
Incidence of colorectal cancer
Baron,99,100 Bonithon-Kopp96	Calcium alone (1200–2000 mg/day)	Placebo alone	CRC	2	3	1	668	4	678	0.34	0.05	2.14	0	0.00	–0.01	0.00	0

Recurrence of any adenoma

Two studies (Baron and Bonithon-Kopp; 1186 analysed participants) compared calcium alone with placebo (Figure 8a). 96,99,100 A meta-analysis demonstrated an 18% relative reduction in the risk of developing an adenoma of any type for those taking calcium, and this was statistically significant at the 5% level (RR 0.82, 95% CI 0.69 to 0.98, p = 0.03). There was a 6% absolute reduction in the risk of developing an adenoma of any type (RD –0.06, 95% CI –0.11 to –0.01, p = 0.03) for those taking calcium. There was no statistical heterogeneity in either analysis (I² = 0%).

The third, smaller study (Hofstad; 93 participants in the analysis) reported that calcium plus antioxidants (vitamins C and E, beta-carotene and selenium) significantly reduced the risk of polyp recurrence compared with placebo. Event data were not reported but the odds ratio was reported as 0.31 (95% CI 0.11 to 0.84). 103

The study reported by Baron also presented follow-up data beyond the 4-year treatment period for participants with a subsequent colonoscopy; however, this was not controlled for post-trial calcium intake. 155 In the interval 0–5 years after the end of the trial (i.e. years 5–9 after the trial started), 347 patients had a further colonoscopy (37% of the 930 originally randomised) and the relative risk of adenoma was 0.63 (95%CI 0.46 to 0.87, p = 0.005). The calcium group therefore had a 37% relative reduction in the risk of an adenoma of any type, which was statistically significant at the 5% level. In the interval 6–10 years after the end of the trial (i.e. years 10–14 after the trial started), 424 patients had a further colonoscopy (46% of the 930 originally randomised) and the relative risk of adenoma was 1.09 (95% CI 0.85 to 1.39, p = 0.511), i.e. the reduction in risk was not shown to continue to this time point.

Incidence of advanced adenoma

Two studies (Baron and Bonithon-Kopp; 1186 analysed participants) compared calcium alone with placebo (Figure 8b). 96,99,100 The reductions in the relative and absolute risks of developing advanced adenoma were not statistically significant at the 5% level (meta-analysed RR 0.77, 95% CI 0.50 to 1.17, p = 0.21; RD –0.01, 95% CI –0.05 to 0.03, p = 0.61). The statistical heterogeneity was 0% for the relative risk and high (57%) for the risk difference.

The study reported by Baron also presented follow-up data beyond the 4-year treatment period for participants with a subsequent colonoscopy; however, this was not controlled for post-trial calcium intake. 155 In the interval 0–5 years after the end of the trial (i.e. years 5–9 after the trial started), 347 patients had a further colonoscopy (37% of the 930 originally randomised) and the relative risk of advanced adenoma was 0.85 (95%CI 0.43 to 1.69, p = 0.65), which was not statistically significant at the 5% level. In the interval 6–10 years after the end of the trial (i.e. years 10–14 after the trial started), 424 patients had a further colonoscopy (46% of the 930 originally randomised) and the relative risk of advanced adenoma was 1.10 (95% CI 0.65 to 1.88, p = 0.717), which was not statistically significant at the 5% level.

Incidence of colorectal cancer

Two studies (Baron and Bonithon-Kopp; 1146 participants in analysis) compared calcium alone with placebo (Figure 8c). 96,99,100 The reductions in the relative risk and absolute risk of developing CRC associated with calcium use were not statistically significant at the 5% level (meta-analysed RR 0.34, 95% CI 0.05 to 2.14, p = 0.25; RD 0.00, 95% CI –0.01 to 0.00, p = 0.22). There was no statistical heterogeneity in either analysis (I² = 0%). These results should be interpreted with caution because of the relatively short follow-up duration and the low event rate (one of 668 in the calcium arms and four of 678 in the placebo arms).

Compliance

Compliance was measured by periodic self-report in two studies96,99,100 and by pill counting or renewal in two studies. 96,103 Two studies reported that between 69 and 76% in the intervention group took 80% or more of pills. 96,103 In the Bonithon-Kopp study, this was statistically significantly lower than the rate of compliance in the control group (p = 0.04). 96 The Baron study reported that at least 90% of pills were taken by between 79 and 85% of participants. 99,100

Adverse effects

The Bonithon-Kopp study reported a statistically significant difference between the calcium and placebo groups in terms of all side effects (26/176 vs 12/178, p = 0.043), yet there was no statistically significant difference in terms of the ‘major side effects’ of severe diarrhoea or abdominal pain (six of 176 vs three of 178, p = 0.31; p-value generated by reviewers). 96 The Hofstad study also reported the number of cases of diarrhoea and abdominal pain, but found no significant differences between groups (five versus seven and four versus four for intervention and control groups respectively). 103 The Baron study reported that ‘medical symptoms and complications were not associated with treatment assignment’. 99,100

Ongoing studies

An ongoing multicentre study is assessing aspirin, calcium and/or vitamin D (as a combination treatment) versus placebo for prevention of adenomas in participants with a history of adenomas, and aims to enrol 1000 participants, with a treatment duration of 3 years and follow-up at 3 and 5 years (www.clinicaltrials.gov, NCT00486512).

Excluded studies

No additional studies of calcium and/or vitamin D in the intermediate-risk population were excluded on final examination of the full papers.

Calcium and/or vitamin D: low-risk population (general population or no increased risk of colorectal cancer)

Outcome data

Effectiveness

Effectiveness results for calcium and/or vitamin D in the general population (or in populations with no increased risk for CRC) are shown in Figure 9 and Table 23.

FIGURE 9.

Calcium and/or vitamin D: meta-analyses (general population or low-risk group). Incidence of colorectal cancer: (a) calcium ± vitamin D versus placebo alone, (b) calcium alone versus placebo alone, (c) calcium + vitamin D versus placebo alone.

TABLE 23 - Calcium: summary of results (general population or low-risk population)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Incidence of colorectal cancer
Lappe,97 Wactawski-Wende101	Any calcium (1000–1500 mg/day) (± vitamin D 400–1100 IU/day)	Placebo alone	CRC	2	4 + 7	169	19,067	156	18,394	0.62	0.11	3.40	58	0.00	–0.01	0.00	39
Lappe97	Calcium alone (1400–1500 mg/day)	Placebo alone	CRC	1	4	0	445	2	288	0.13	0.01	2.69	N/C	–0.01	–0.02	0.00	N/C
Lappe,97 Wactawski-Wende101	Calcium (1000–1500 mg/day) + vitamin D (400–1100 IU/day)	Placebo alone	CRC	2	4 + 7	169	18,622	156	18,394	1.08	0.87	1.34	0	0.00	0.00	0.00	0

Incidence of any adenoma

No studies of calcium in the general population assessed adenoma incidence.

Incidence of advanced adenoma

No studies of calcium in the general population assessed advanced adenoma incidence.

Incidence of colorectal cancer

Two studies (37,461 participants in analysis) compared calcium with or without vitamin D with placebo (Figure 9a). 97,101 There was no statistically significant difference between groups in the relative risk of developing CRC (meta-analysed RR 0.62, 95% CI 0.11 to 3.40, p = 0.58); however, this analysis suggested a high level of statistical heterogeneity (I² = 58%). There was no statistically significant difference between groups in the absolute risk of developing CRC (RD 0.00, 95% CI –0.01 to 0.00, p = 0.80), but there was a moderate level of statistical heterogeneity (I² = 39%).

One study (Lappe; 733 participants in analysis) compared calcium alone with placebo (Figure 9b). 97 The incidence of CRC was none of 445 in the calcium arm and two of 288 in the placebo arm.

Two studies (37,016 participants in analysis) compared calcium plus vitamin D with placebo (Figure 9c). 97,101 The relative and absolute risk differences between groups for developing CRC were not statistically significant at the 5% level (meta-analysed RR 1.08, 95% CI 0.87 to 1.34, p = 0.51; RD 0.00, 95% CI 0.00 to 0.00, p = 0.56) and no statistical heterogeneity was evident (I² = 0%).

Data on CRC incidence in the studies of calcium should be interpreted with caution because of the relatively short follow-up durations in terms of detecting cancer outcomes (4 and 7 years in the two studies).

Compliance

The two studies reported a compliance rate (> 80% of pills taken) in the intervention groups of ≥ 60% (Wactawski-Wende)101 or 74–86% (Lappe). 97 Compliance was measured by periodic weighing of pill bottles in both studies.

Adverse effects

Neither study reported any serious adverse events. There were no statistically significant differences between groups in terms of major symptoms or any major disease outcomes, including cardiovascular disease, any cancer or death.

Ongoing studies

No ongoing studies of calcium and/or vitamin D in the low-risk population were identified.

Excluded studies

No additional studies of calcium and/or vitamin D in the low-risk population were excluded on final examination of the full papers.

Summary of results for calcium and/or vitamin D

Two studies of calcium (1200–2000 mg/day) in individuals with a history of adenomas demonstrated a statistically significant 18% reduction in the relative risk of adenoma recurrence. Two studies of calcium (1000–1500 mg/day) plus vitamin D (400–1100 IU/day) in populations with no increased baseline risk of CRC demonstrated no statistically significant effect on the relative risk of CRC. However, the duration of follow-up was 4 to 7 years, which may not be long enough to detect an effect on cancer incidence. One study assessed calcium in patients with adenomas due to FAP, and reported no significant reduction in polyp number or progression. No study reported any serious adverse events associated with calcium and/or vitamin D.

Antioxidants: high-risk population (FAP or HNPCC)

No studies of antioxidant chemoprevention use within populations at high-risk for CRC (i.e. FAP and HNPCC) were identified.

Antioxidants: intermediate-risk population (history of adenomas or colorectal cancer)

Outcome data

Effectiveness

Effectiveness results for antioxidants in the intermediate-risk population (history of adenomas) are presented in Figure 10 and Table 25.

FIGURE 10.

Antioxidants: meta-analyses (history of adenomas). Recurrence of any adenoma: (a) any antioxidants versus placebo alone; (b) any antioxidants (±low-fat diet ±bran) versus no antioxidants; (c) vitamins A, C, E versus placebo alone; (d) vitamins C and E versus placebo alone.

TABLE 25 - Antioxidants: summary of results (history of adenomas)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Recurrence of any adenoma
Bonelli,115 Greenberg,111,112 MacLennan,126 McKeown-Eyssen,129 Ponz de Leon,121 Roncucci121,124	Any antioxidant (± low-fat diet ± bran)	No antioxidant (± low-fat diet ± bran ± lactulose)	Adenoma	6	1.5, < 2, 2, 4, 4, 5	316	1013	199	632	0.78	0.54	1.14	74	N/C	N/C	N/C	N/C
Bonelli,115 Greenberg,111,112 McKeown-Eyssen,129 Ponz de Leon,121 Roncucci121,124	Any antioxidant	Placebo alone	Adenoma	5	1.5, < 2, 2, 4, 5	262	857	158	482	0.67	0.42	1.07	75	–0.10	–0.22	0.02	79
Roncucci,121,124 Ponz de Leon121	Vitamins A,C,E	Placebo alone	Adenoma	2	< 2	12	106	38	111	0.35	0.07	1.68	83	–0.21	–0.42	0.01	70
Greenberg,111,112 McKeown-Eyssen	Vitamins C,E	Placebo alone	Adenoma	2	2, 4	98	275	89	254	1.02	0.81	1.28	0	0.00	–0.08	0.08	0
Greenberg111,112	Beta-carotene	Placebo alone	Adenoma	1	4	66	184	68	187	0.99	0.75	1.29	N/C	0.00	–0.10	0.09	N/C
Greenberg111,112	Vitamins C,E + beta-carotene	Placebo alone	Adenoma	1	4	66	175	68	187	1.04	0.79	1.36	N/C	0.01	–0.09	0.11	N/C
Bonelli115	Vitamins A,C,E + selenium + zinc	Placebo alone	Adenoma	1	5	20	117	31	116	0.64	0.39	1.05	N/C	–0.10	–0.20	0.01	N/C

Recurrence of any adenoma

Five studies (1339 participants in analysis) compared any antioxidant (alone or in any combination) with placebo (Figure 10a). 112,115,121,124,129 A meta-analysis demonstrated that the differences in relative and absolute risks of developing at least one adenoma were not significant at the 5% level (RR 0.67, 95% CI 0.42 to 1.07, p = 0.09; RD –0.10, 95% CI –0.22 to 0.02, p = 0.10). There was a high level of statistical heterogeneity observed in both estimates (I² = 75% and I² = 79%, respectively).

Six studies (1706 participants in analysis) compared any antioxidant (alone or in any combination) against no antioxidant (Figure 10b). This analysis included participants from MacLennan who had received 25 g/day wheat bran and/or a low-fat diet, in addition to placebo. 126 There was no statistically significant difference between groups in the relative risk of developing at least one adenoma (meta-analysed RR 0.78, 95% CI 0.54 to 1.14, p = 0.20). This analysis suggested a high level of statistical heterogeneity (I² = 74%).

A number of discrete analyses were undertaken for different combinations of antioxidants (see Table 25 and Figure 10c–d). No statistically significant results were found for any of the five analyses; the interventions were neither effective nor harmful.

A seventh study (Hofstad; 93 participants in the analysis) reported that calcium plus antioxidants (vitamins C and E, beta-carotene and selenium) significantly reduced the risk of adenoma recurrence compared with placebo. Event data were not reported but the odds ratio was reported as 0.31 (95% CI 0.11–0.84). 103

Incidence of advanced adenoma

No studies of antioxidants in the intermediate-risk population reported incidence of advanced adenoma, as defined in studies of other interventions (i.e. adenoma ≥ 1 cm in diameter, villous or tubulovillous adenoma, adenoma with severe dysplasia or CRC). Three studies reported separate data on large adenomas or adenomas with other abnormalities (Roncucci, MacLennan, McKeown-Eyssen);124,126,129 however, these data are not presented here because of the lack of consistency in definitions of advanced adenoma.

Incidence of colorectal cancer

No studies of antioxidants in the intermediate-risk population reported CRC incidence.

Compliance

Three studies reported a compliance rate in the intervention group. McKeown-Eyssen gave a compliance range over the study period which was 89–100%, measured by urine concentrations of vitamin C. 129 Roncucci measured the compliance rate by periodic self-report as 85%. 124 Bonelli did not indicate how compliance was measured, but reported the overall value to be 63.7%. 115 However, the compliance from Bonelli varied greatly between the two study centres from 40.5% to 96.1%. 115 Greenberg did not provide compliance rates but indicated that five participants had stopped taking tablets because of their presumed toxicity. 112

Adverse effects

Two studies reported minor adverse events. Roncucci reported that 1.4% of participants in the antioxidants intervention group experienced pruritus with no skin alteration and withdrew from the trial. 124 In the same trial, some participants were receiving lactulose and 4.9% had lactulose-induced diarrhoea. Bonelli reported that 12.5% of participants in the intervention group and 2.9% of participants in the placebo group experienced minor adverse events but did not state what these were. 115

Ongoing studies

An ongoing phase 3 RCT is assessing selenium versus placebo for prevention of adenoma recurrence in participants with a history of colorectal adenomatous polyps, and is aiming to enrol 2050 participants, with treatment and follow-up durations of 5 years, or up to 7.5 years in some participants (www.cancer.gov/clinicaltrials/UARIZ-00–0430–01).

Excluded studies

Hofstad103 did not report event data (only mean numbers of adenomas) and this study was therefore excluded from meta-analysis. No further studies of antioxidants in the intermediate-risk population were excluded on final examination of the full papers.

Antioxidants: low-risk population (general population or no increased risk of colorectal cancer)

Outcome data

Effectiveness

Effectiveness results for antioxidants in the general population (or populations with no increased risk for CRC) are shown in Figure 11 and Table 27.

FIGURE 11.

Antioxidants: meta-analyses (general population or low-risk population). Incidence of colorectal cancer: (a) any antioxidants (± aspirin, simvastatin, ramipril) versus no antioxidants; (b) vitamin C versus no vitamin C (adjusted for other antioxidants); (c) vitamin E alone versus placebo alone; (d) vitamin E versus no vitamin E (adjusted for other antioxidants); (e) selenium alone versus placebo alone; (f) beta-carotene (± aspirin) versus no beta-carotene: studies with event data; (g) beta-carotene (± aspirin and adjusted for other antioxidants) versus no beta-carotene: studies with or without event data.

TABLE 27 - Antioxidants: summary of results (general population or low-risk population)

C, control; I, intervention; I², measure of heterogeneity; LCI, lower 95% confidence limit; N/C, not calculable; RD, absolute risk difference; RR, relative risk; UCI, upper 95% confidence limit.

Studies	Intervention (I)	Control (C)	Outcome	No. of studies	Follow-up (years)	I: n	I: N	C: n	C: N	RR	LCI	UCI	I2 (%)	RD	LCI	UCI	I2 (%)
Incidence of any adenoma
Malila (Virtamo)133–136	Any antioxidant (vitamin E and/or beta-carotene)	Placebo alone	Adenoma	1	6	119	11,651	27	3887	1.47	0.97	2.23	N/C	0.00	0.00	0.01	N/C
Malila (Virtamo)133–136	Vitamin E alone	Placebo alone	Adenoma	1	6	47	3890	27	3887	1.74	1.09	2.79	N/C	0.01	0.00	0.01	N/C
Malila (Virtamo)133–136	Beta-carotene alone	Placebo alone	Adenoma	1	6	28	3883	27	3887	1.04	0.61	1.76	N/C	0.00	0.00	0.00	N/C
Malila (Virtamo)133–136	Vitamin E + beta-carotene alone	Placebo alone	Adenoma	1	6	44	3878	27	3887	1.63	1.01	2.63	N/C	0.00	0.00	0.01	N/C
Incidence of colorectal cancer
Zhu,109,110 Duffield-Lillico,137–139 HPS,114,140,141 Hennekens,142 SU.VI.MAX,114,130–132 CARET,114,127,128 Lippman,116 Virtamo,133–136 HOPE-TOO107,125	Any antioxidant (± aspirin, simvastatin, ramipril)	No antioxidant (± aspirin, simvastatin, ramipril)	CRC	9	5, 5, 5.5, 6, 7, 7–8, 10, 12, 12	981	90,744	673	58,178	1.00	0.88	1.13	25	0.00	0.00	0.00	24
Gaziano,113 Lin118–120	Vitamin C (adjusted for other antioxidants; studies with no event data)	No vitamin C (adjusted for other antioxidants)	CRC	2	8, 9	NR for some studies	NR for some studies	NR for some studies	NR for some studies	0.84	0.64	1.10	0	N/C	N/C	N/C	N/C
Virtamo,133–136 Lippman116	Vitamin E alone (studies with event data)	Placebo alone	CRC	2	5.5, 12	142	16,023	135	15,983	1.05	0.83	1.33	0	0.00	0.00	0.00	N/C
Gaziano,113 HOPE-TOO,107,125 Lee,69,70,122,123 Lin,118–120 Lippman,116 Virtamo133–136	Vitamin E (adjusted for other antioxidants; studies with or without event data)	No vitamin E (adjusted for other antioxidants)	CRC	6	5.5, 7, 8, 9, 10, 12	NR for some studies	NR for some studies	NR for some studies	NR for some studies	0.99	0.86	1.14	0	N/C	N/C	N/C	N/C
Duffield-Lillico,137–139 Lippman116	Selenium alone	Placebo alone	CRC	2	5.5, 7–8	72	9373	79	9325	0.77	0.37	1.62	68	–0.01	–0.02	0.01	75
Hennekens,142 Virtamo,133–136 Zhu109,110	Beta-carotene (± aspirin; studies with event data)	No beta-carotene (± aspirin)	CRC	3	6, 12, 12	266	18,436	250	18,376	1.09	0.78	1.51	54	0.00	0.00	0.00	41
Hennekens,142 Virtamo,133–136 Zhu,109,110 Lin118–120	Beta-carotene (± aspirin, and adjusted for other antioxidants; studies with or without event data)	No beta-carotene (± aspirin, and adjusted for other antioxidants)	CRC	4	6, 9, 12, 12	NR for some studies	NR for some studies	NR for some studies	NR for some studies	1.11	0.84	1.47	26	N/C	N/C	N/C	N/C
Virtamo133–136	Vitamin E + beta-carotene alone	Placebo alone	CRC	1	12	90	7278	75	7287	1.20	0.89	1.63	N/C	0.00	0.00	0.01	N/C
CARET114,127,128	Vitamin A + beta-carotene alone	Placebo alone	CRC	1	10	127	9420	123	8894	0.97	0.76	1.25	N/C	0.00	0.00	0.00	N/C
Lippman116	Vitamin E + selenium alone	Placebo alone	CRC	1	5.5	77	8703	60	8696	1.28	0.92	1.79	N/C	0.00	0.00	0.00	N/C
HPS114,140,141	Vitamins C, E + beta-carotene (± simvastatin)	Placebo (± simvastatin)	CRC	1	5	117	10,269	140	10,267	0.84	0.65	1.07	N/C	0.00	–0.01	0.00	N/C
SU.VI.MAX114,130–132	Vit CE + beta-carotene + selenium + zinc	Placebo alone	CRC	1	7.5	21	6481	24	6536	0.88	0.49	1.58	N/C	0.00	0.00	0.00	N/C

Incidence of any adenoma

Malila et al. (15,538 participants in analysis) compared any antioxidant (alone or in any combination) with placebo. 134 There was no significant difference between groups in the relative risk of developing at least one adenoma (RR 1.47, 95% CI 0.97 to 2.23, p = 0.07). The absolute risk difference was 0.00 (95% CI: 0.00 to 0.01 p = 0.04).

Incidence of advanced adenoma

No studies of antioxidants in the low-risk population reported incidence of advanced adenoma.

Incidence of colorectal cancer

Nine studies (148,922 participants in analysis) compared any antioxidant (alone or in any combination) against no antioxidant (Figure 11a). 109,116,122,127,130,133,134,137,140,142 A meta-analysis demonstrated no significant difference between groups in the relative or absolute risk of developing CRC (RR 1.00, 95% CI 0.88 to 1.13, p = 0.94; RD 0.00, 95% CI 0.00 to 0.00, p = 0.94). There was a low level of statistical heterogeneity (I² = 25% and I² = 24%, respectively).

A number of discrete analyses were undertaken for different combinations of antioxidants (see Table 27 and Figure 11b–g). Of the 14 analyses undertaken, only two provided statistically significant results. In the ATBC study, Malila found a 74% relative increase in the risk of developing at least one adenoma in participants receiving vitamin E compared with placebo alone (RR 1.74, 95% CI 1.09 to 1.79, p = 0.02). 134 Malila found that there was a 63% relative increase in the risk of developing at least one adenoma in participants receiving vitamin E and beta-carotene compared with placebo alone (RR 1.63, 95% CI 1.01 to 2.63, p = 0.04). 134 However, given the small numbers of both participants and events, as well as the large number of separate analyses (uncorrected for repeated significance testing), these results should be treated with caution. The remaining 12 analyses showed no statistically significant difference between the intervention and placebo groups; the intervention was neither effective nor harmful (Table 27).

Compliance

All 12 studies reported a compliance rate for the intervention group. Seven studies provided average compliance rates over the whole trial period that ranged between 73 and 93%. 109,118,122,127,133,134,137,140 Lin, Lee and Duffield-Lillico determined compliance by patient self-report alone. 118,122,137 Goodman determined compliance by pill counting for 85% of participants and self-report for 15% of participants. 127 Zhu and Malila 1999/Virtamo 2003 used pill counting and random blood sampling109,133,134 whereas the HPS employed pill counting alone. 140

Hercberg and Hennekens provided end of trial compliance rates that were 74% and 80% respectively. 130,142 Five studies provided compliance rates at the start or during the trial and at the end of the trial. 113,116,118,122,125 Initial compliance rates in these studies ranged between 76 and 94.2%. Final compliance rates ranged between 65 and 89.2%.

Adverse effects

Seven studies did not report adverse events. Gaziano reported a 74% increased risk in haemorrhagic strokes for the vitamin E intervention group when compared with the placebo group [hazard ratio (HR) 1.74, 95% CI 1.04 to 2.91]. 113 Lippmann reported a statistically significant increase in alopecia and grade 1–2 dermatitis in the selenium group. 116 Lee reported a 6% increase in the risk of epistaxis in the vitamin E group compared with the placebo group (RR 1.06, 95% CI 1.01 to 1.11; p = 0.02). 122 Zhu reported two false jaundices in the beta-carotene group of one study as a result of hypercarotenemia. 109 Hennekens reported a statistically significant increase in the minor side effects of yellowing of the skin and minor gastrointestinal problems such as belching, in the beta-carotene group. 142

Ongoing studies

No ongoing studies evaluating the use of antioxidants as chemoprevention in the low-risk population were identified.

Excluded studies

Reid et al. 138 presented data on the number of participants developing at least one adenoma after receiving 200 µg selenium or a placebo. This study was excluded because the participants were not certified as polyp-free at the beginning of the trial. The CRC incidence in the selenium and placebo groups in this study is included in our analyses using data from Duffield-Lillico. 137

Summary of results for antioxidants

There were six studies of antioxidants (including vitamins A, C and E, beta-carotene or selenium) in individuals with a history of adenomas. Doses and combinations varied between studies. Statistically significant differences in relative risk of adenoma recurrence were not demonstrated, either when all antioxidants were analysed together or when the separate combinations were assessed separately. There were 12 studies of antioxidants in populations with no increased risk of CRC, with follow-up durations between 5 and 12 years. Across the nine studies comparing antioxidants to no antioxidants, there was no statistically significant effect on incidence of CRC. In addition, one study assessing the effect of antioxidants on adenoma incidence in the low-risk population did not demonstrate a statistically significant effect. Of 14 discrete analyses for different combinations of antioxidants in the low-risk population, one study reported a statistically significant increase in relative risk of adenoma incidence in participants receiving vitamin E or vitamin E plus beta-carotene; however, this should be interpreted with caution because of the number of analyses undertaken. Reported side effects of antioxidants included pruritus (vitamins A, C, E), epistaxis (vitamin E), statistically significant increase in risk of haemorrhagic stroke (vitamin E), alopecia and dermatitis (selenium), yellowing of the skin and belching (beta-carotene). Other reviews have shown that antioxidants did not reduce gastrointestinal cancer incidence (beta-carotene and vitamin A possibly increasing the risk),114 and that vitamin A, vitamin E and beta-carotene may increase overall mortality. 156

Inclusion and exclusion criteria

Population/setting:

• adults only

• UK only (first search and screen); non-UK (second search and screen).

Interventions:

• anti-inflammatory agents (NSAIDs, aspirin), micronutrients (vitamins, minerals, folate, selenium, calcium) and dietary supplements generally.

Outcomes:

• attitudes, perceptions and beliefs surrounding the taking of such agents and supplements.

Study designs:

• qualitative: interviews, focus groups, open-ended questionnaires

• quantitative (satisfaction surveys).

Other:

• English language only

• dates: 2003 onwards.

Exclusion criteria:

• studies excluded if they do not fulfil the above criteria.

Identification and screening of studies

Two searches were performed by an information specialist (D Papaioannou) after the development and piloting of an appropriate search strategy. The first strategy sought studies that corresponded to the target population for the quantitative review, i.e. attitudes towards the stated agents reported in UK studies from the last 5 years. Consequently, the search used terms describing the agents of interest: NSAIDs, aspirin, vitamins, minerals, folate, selenium, calcium and dietary supplements generally, combined with a published, validated filter for identifying qualitative studies, together with the Medical Subject Headings (MeSH) term ‘qualitative research’,166 and a pragmatically developed filter to identify UK studies (see Appendix 1 for the MEDLINE search strategy). The following databases were searched for published and unpublished material: MEDLINE, PREMEDLINE, CINAHL, EMBASE, AMED, ASSIA, IBSS, PsycINFO, Science Citation Index, and Social Science Citation Index. Searches for recently completed and unpublished research or grey literature were conducted using the Health Management Information Consortium (HMIC) and the King’s Fund database. Searches were performed for citations from 2003 onwards, to increase relevance to current views and attitudes.

Three reviewers (C Carroll, A Booth, K Cooper) each screened a third of the citations for relevance (based on the inclusion criteria) and any references identified for potential inclusion were discussed within the review team. Disagreements were either resolved by discussion or the full paper was retrieved to make a definitive judgment. Full papers of all relevant and potentially relevant citations were then screened using the same process.

This search (screen 1) produced only five papers that satisfied the initial UK-only inclusion criteria;167–171 four of the studies focused on aspirin/NSAIDs168–171 and one focused on folic acid;167 none considered other agents or supplements, and no population used the agents for chemoprevention.

Given the apparent limitations of the sample, a decision was made to broaden the search beyond the original proposed scope of the review, to identify non-UK papers exploring people’s views of taking any of these agents for the prevention or chemoprevention of CRC. This second search combined free text terms and related MeSH keywords for prevention and chemoprevention, with a filter for CRC, and the published, validated filter for identifying qualitative studies, with the addition of the MeSH term ‘qualitative research’, cited above (see Appendix 1 for an example search strategy). The following databases were searched for published and unpublished material from 2003 onwards: MEDLINE, PREMEDLINE, CINAHL, EMBASE, AMED, ASSIA, IBSS, PsycINFO, Science Citation Index, and Social Science Citation Index. Searches for recently completed and unpublished research or grey literature were conducted on HMIC and the King’s Fund database.

The same process of study screening described above was followed for the results of this second search except for removal of the limitation to studies performed in the UK (screen 2). No non-UK studies were found to meet the revised inclusion criteria. The team therefore decided to revisit the original set of references from the first search (screen 3), but this time removed the UK-only limitation that had formerly been employed in screen 1. This resulted in the identification of 10 new non-UK studies that did satisfy the revised criteria (screen 3). The reference lists of all included studies were checked for additional, relevant studies, and some supplementary, unstructured, non-systematic searching testing various, different combinations of terms was performed by two members of the project team (A Booth, K Cooper). These supplementary searches were not intended to be sensitive or comprehensive, but were undertaken as a means of identifying additional studies that may be relevant.

The overall aim of this iterative, pragmatic approach to searching was to identify a set of studies providing relevant information on views and attitudes towards the taking of chemopreventive agents.

Data extraction strategy

The resulting included studies were extracted using a form developed specifically for this review (see Appendix 3), and piloted on one included paper.

Critical appraisal strategy

The quality of included studies was assessed using a number of criteria derived from relevant critical appraisal checklists, principally for qualitative studies and surveys, the most dominant study designs. 172–174 Using a small number of key quality assessment criteria, so-called prompts, that apply universally to qualitative studies, rather than appraising studies using large numbers of potentially irrelevant checklist questions, has been recommended as an approach for critical appraisal of qualitative studies. 175 It may not always be appropriate to exclude qualitative research studies simply on the basis of quality assessment of the study design175,176 so the aim of this assessment was to comment generally on the relative quality of the included studies, based on the following criteria common to the cited checklists for both types of studies described above: the reporting of their sampling strategies, and the reporting and conduct of methods of data collection and analysis. Satisfactory, good or very good studies provided details on all or most of these criteria, and the least good studies might only report on perhaps as few as one. The aim of this assessment was to explore quality as an explanation for any differences in the results of otherwise similar studies, and to consider its impact on the internal validity of the review. This is an accepted approach for quality assessment in the systematic review of qualitative data. 177

Evidence synthesis methods

Data for analysis consisted of either verbatim quotations from study participants or findings reported by authors that were clearly supported by study data. The latter included qualitative data from surveys, such as ‘75% of respondents said that they were concerned about side effects of NSAIDs’ or ‘27% of respondents reported that the views of family and friends affected their decision-making’.

Following discussions within the team regarding the competing merits of a grounded theory-type approach (starting with a completely blank sheet) and an augmentative approach (building on an existing model or framework) the decision was to go with the latter, a combination of framework analysis178 and framework synthesis. 179–181 It was felt that a flexible framework-based approach does not militate against the identification or discovery of additional concepts. Indeed the existence of the best fit model could arguably be seen as privileging study-specific insights over the generic observations made within the pre-existing model. Nevertheless, the review team recognises that this approach represents a methodologically innovative middle ground on the continuum between grounded theory-type and framework-based syntheses and acknowledges the need for further evaluation.

Framework analysis involves the identification of a priori themes against which to map data from the studies identified for the review. However, in this case these predetermined themes were generated from a relevant pre-existing model in the literature, as is the case with framework synthesis. This model was identified in the following way. Key concepts (e.g. influence of family, health practitioners, media, considerations of cost etc.) were identified from the preliminary scoping literature searches and from an initial examination of articles for potential inclusion, before a definitive judgement. These concepts in combination were used to inform supplementary searches to try to identify existing models or frameworks within the general topic area of attitudes to vitamins and supplements. However, these supplementary searching processes failed to find a model or framework that was completely prespecified to the team’s satisfaction. Nevertheless they did identify a framework that represented a ‘best fit approach’ and included the majority of the key concepts. This ‘contingent’ framework formed the categories under which the data extracted from the qualitative studies were summarised. This model was a conceptual model (not evidence-based, i.e. based on the author’s views, not research) of young women’s views and attitudes to the taking of micronutrients (including vitamins, minerals). 182 This model provided an initial framework and pre-existing themes against which to map and code the data from studies identified for this review. A list of themes was derived from the conceptual model and, if relevant, extracted data were coded against these pre-existing themes. New themes were created where data did not translate into any of the pre-existing themes. In this way, a relevant existing model acted as the basis for the synthesis and could be built-on, reduced or added to by these new data. Dixon-Woods et al. 183 suggest that such methods are most appropriate where concepts are relatively ‘secure’ and ‘well specified’. Framework synthesis also usually excludes studies of lower quality, but this was not done in this case, and so differs from the published method. 181

This method of synthesis therefore used current methodological developments within qualitative data synthesis for systematic review and the production of accompanying conceptual models and frameworks. It offered a means not only to reinforce, critique and develop an existing published model, conceived for a specific population, but also, by starting from the principle of a priori determined themes rather than the generation of theory grounded in data, produced a process of synthesis that was relatively rapid when compared with certain other more interpretative forms of synthesis (e.g. thematic analysis). 184

Two reviewers (C Carroll, K Cooper) each extracted data from half of the included studies using a form developed for this review. The reviewers assigned data to preset themes and, where relevant, created new themes that better captured or described the data. Each reviewer then checked and examined critically the extraction and assignment of data to themes performed by the other. A third reviewer (A Booth) independently examined the extracted data and, using a framework analysis approach, generated a new list of themes, independent from the published conceptual model. The team of reviewers then discussed the data and resulting list of themes, including themes recognised from the pre-existing model and any new themes generated by the study data. A consensus was reached on which preset themes were supported by the data; whether any of the new themes identified by all three reviewers mapped into any of the preset themes or into one another; and on the finalised list of new themes. These themes or concepts were then used to develop a new, revised conceptual model, drawing on the earlier model, to describe and explain people’s views around the taking of aspirin, NSAIDs and micronutrients, such as vitamins, minerals, folic acid and other dietary supplements.

Synthesis

The following list of themes was derived from the conceptual model of factors determining use of micronutrients among women of reproductive age,182 and constituted the a priori themes against which data were extracted from included studies (Box 1).

The synthesis generated the accompanying model presented in Figure 13. This model describes the processes involved in an individual’s decision about whether or not to take NSAIDs, aspirin, folic acid, vitamins, minerals, calcium or other dietary supplements. External factors, such as the influence of health professionals and family members, and internal factors, such as a person’s own experience or health, all have an impact on an individual’s perceived need for an agent or supplement, and their subsequent decision about whether or not to take it. These processes, described in the model Figure 13, correlate with versions of the Perceived need and Decision elements of the Huffman182 model of micronutrient use among women, and the Contemplation and Determination elements of Prochaska and Velicer’s 1997 Stages of Change model regarding the development of health behaviours. 202

FIGURE 13.

Conceptual model to describe views and experiences of adults concerning the taking of micronutrient supplements, aspirins and NSAIDs.. Adapted from Huffman 2002182 and Prochaska and Velicer 1997.202

A third, vital element in the decision-making process is the influence of the perceived risks or benefits associated with an agent or supplement, which appear as Benefits and Negative factors affecting use in the Huffman model. 182 However, the model resulting from findings of this review highlights the even more complex nature of the influence of risk and benefit on decision-making and use. The perceived risks or benefits of an agent or supplement can either directly influence an individual’s decision to use it, or may inform a personal assessment of the trade-off between these two elements, which in turn may also affect the decision-making process.

This process then results in the use or non-use of the agents in question, the so-called Action Stage of the Prochaska and Velicer’s 1997 model. 202 Maintenance or continuity of use may be affected by three elements: the risks (side effects) and/or benefits experienced by taking the agents or supplements; the physical properties of the agents, which may facilitate or act as a barrier to continued use; and an individual’s age or gender, which may predict the likelihood of their maintenance of a course of therapy.

All of the factors and processes described in this evidence-based model, determining the decision to take, and to continue to take, NSAIDs, aspirin, folic acid, vitamins, minerals, calcium or other dietary supplements, have emerged from qualitative literature published in the previous 5 or 6 years. The decision process runs from the first stages of perceived need, on the left, through decision-making itself, to final non-use or use, and maintenance of use on the right. The arrows representing Prochaska and Velicer’s 1997 Stages of Change model, below our model, flow in the same direction, capturing the similar stages and processes of contemplation, determination and action. The details of this evidence are provided as follows.

External factors affecting decision-making

Internal factors affecting decision-making

The balance between benefit and risk in decision-making

Use and the maintenance of use

Methods

A systematic review of economic evaluations of chemoprevention for CRC was carried out as part of a separate project, which aimed to identify all economic evaluations of health-care interventions within the CRC health-care service. 33 The inclusion criterion for the economic evaluations of chemoprevention for CRC was broadly defined as economic evaluations describing any population given any form of chemoprevention. The search criteria are presented in Appendix 1. This review presents the current evidence base around economic evaluations of chemopreventive agents for CRC. The results of the review determine whether it is necessary to develop a health economic model for this evaluation.

Results

The electronic literature searches identified 4311 potentially relevant citations to economic analyses of bowel cancer technologies. Of these, only 11 citations appeared to relate to economic evaluation of agents used in the chemoprevention of CRC. A total of eight full papers were obtained for detailed inspection, of which six studies met the inclusion criteria.

Justification of excluded studies

Emmons et al. 214 was excluded from the review following full paper screening because the intervention considered within the evaluation was not specifically chemoprevention. This paper assessed the CRC risk reduction factor of interventions such as red meat consumption, fruit and vegetable intake, multivitamin intake, alcohol, smoking and physical inactivity; the majority of which did not meet our definition of chemoprevention. Ladabaum215 was excluded because this was a secondary reporting of the work described by Ladabaum in two of the included studies. 216,217 Figure 14 shows the summary of the study selection and exclusion.

FIGURE 14.

Summary of economic evaluation selection and exclusion.

Ladabaum et al.: Aspirin as an adjunct to screening for prevention of sporadic colorectal cancer

Ladabaum et al. 216 present the methods and results of a cost-effectiveness analysis of aspirin chemoprevention (325 mg/day) versus CRC screening and versus no prevention within the general US population. The analysis also compares a combination of screening and aspirin chemoprevention against screening alone and against no prevention. Screening using both colonoscopy every 10 years and flexible sigmoidoscopy every 5 years alongside annual FOBT is considered within the model. A state transition model was used to simulate a cohort of people aged 50 to 80 years using annual cycles. The choice of age group for the model population was not justified by the study authors. The analysis was undertaken from the perspective of a third-party payer (Medicare) and the outcome of the model is the cost per life-year saved (LYS). Health-related quality of life was not considered within the model.

The health states used within the model include normal (no polyps or cancer), polyp, cancer (localised, regional, distant) and death. Age-specific polyp incidence rates were taken from autopsy studies. CRC incidence was derived to match age-specific Surveillance, Epidemiology and End Results (SEER) data,222 assuming that 90% of CRCs arise from pre-existing polyps (hence 10% of cancers are assumed to arise de novo) and that cancer progresses from localised to regional (2 years in each state) to disseminated cancer. Persons within the model may experience screening or aspirin-related complications, both of which are associated with a probability of death. During any model cycle, the authors report that individuals may die from CRC and CRC treatment or other causes. It is, however, unclear what is meant by CRC treatment in this context. The risk of death due to CRC was derived from the SEER Cancer Statistics Review 222 while other-cause mortality rates were derived from US life tables. The model incorporates the sensitivity and specificity of FOBT, flexible sigmoidoscopy and colonoscopy. Screening is assumed to be associated with a participation rate of 25%,223 whereas aspirin is assumed to be associated with perfect compliance. The reduction of CRC due to aspirin was assumed to be 30% (range 5–55%) through equal reductions in adenoma incidence and cancer progression rates, based on an analysis of 12 studies which assessed the impact of aspirin use on CRC incidence. It is unclear how the authors have used these 12 studies to derive the risk reduction parameter; the use of a meta-analysis is not mentioned within the paper.

Cost components include CRC care by stage (localised, regional, distant),224 aspirin and surveillance and treatment of complications. 225 All costs are presented in 1998 US dollars. Both costs and life-years were discounted at a rate of 3%. One-way sensitivity analyses and a partial probabilistic sensitivity analyses (PSA) were undertaken. The authors did not consider uncertainty within all of the disease natural history parameters; however, the incidence of adenomas and the probability of CRC following adenoma development were varied within the PSA.

The results of the model suggest that aspirin chemoprevention dominates no aspirin or screening in 38% of the PSA iterations; however, if colonoscopy screening was available, the addition of chemoprevention would result in a cost per LYS of around $150,000 in 1998 US dollars (approximately £120,000 in 2008 UK pounds). The model predicts that if flexible sigmoidoscopy and FOBT were currently used in practice, the addition of aspirin chemoprevention would be dominated. Conversely, if screening using flexible sigmoidoscopy/FOBT or colonoscopy was adopted in addition to chemoprevention, the cost per LYS was estimated to be $26,000 or $31,000, respectively in 1998 prices (approximately £21,000 and £25,000 in 2008 UK pounds). Similar results were presented for both the deterministic outcomes and the probabilistic outcomes; however, all of the probabilistic results were not presented within the paper. The one-way sensitivity analysis suggested that the cost-effectiveness results are highly dependent on the effectiveness of aspirin, the rate of screening participation and the rate of aspirin-related complications. Results were improved if it was assumed that aspirin decreases cardiovascular death as well as CRC incidence.

The model presented by Ladabaum et al. has several limitations:

• Uniform distributions were used to describe the uncertainty in all model parameters. Ranges used to characterise uncertainty surrounding model parameters do not appear to be evidence-based.

• The model uses constant transition probabilities for simulating the disease natural history component rather than adjusting for age. With increasing age, polyp incidence will increase and more aspirin-related complications are likely to occur.

• Compliance associated with aspirin chemoprevention is assumed to be 100%, which is highly unlikely in practice.

• The analysis is presented from a Medicaid perspective; however, it is unlikely that Medicaid would fund chemoprevention within the general population.

• The characterisation of uncertainty surrounding the disease natural history model component is limited.

• No model validation was reported.

Arguedas et al.: Surveillance colonoscopy or chemoprevention with COX-2 inhibitors in average-risk postpolypectomy patients: a decision analysis

Arguedas et al. 218 present a cost-effectiveness analysis of celecoxib (200 mg twice daily) against no prevention, surveillance colonoscopy every 3 years and surveillance colonoscopy every 5 years in ‘average’-risk patients who have undergone prior adenoma resection (complete colonoscopy and polypectomy). The authors adopted a Markov framework, with the model following a cohort of people aged 50 for 10 years from the point of resection using an annual cycle length. No justification is provided concerning the choice of population age or the time horizon used within the analysis. The authors report the cost per LYS and a cost per ‘high-grade’ adenoma prevented from the perspective of a third-party payer (Medicare).

The health states included within the model were normal (no polyps or cancer), low-grade adenoma, high-grade adenoma, CRC and death. Constant probabilities rather than age-related probabilities were used to represent the development of polyps and CRC. The adenoma formation rate (incidence) and the incidence of CRC were estimated from the National Polyp Study (NPS). 226 The model assumes that the incidence of polyps after the first colonoscopy is half the incidence after the index colonoscopy, as experts suggest that approximately 50% of the polyps found in the NPS represented polyps missed during the index colonoscopy. 158 The rates of progression from low-grade to high-grade polyps and CRC were estimated from Stryker et al. 20 The model optimistically assumes 100% compliance with both colonoscopy and celecoxib and that the sensitivity of colonoscopy is 100%; more conservative assumptions were tested within the sensitivity analysis. The authors assumed that celecoxib would reduce the incidence of adenomas by 50%. This value was varied from 0 to 100% within one-way and two-way sensitivity analyses. The annual incidence of ulcers due to aspirin was based on the CLASS study (Celecoxib Long-term Arthritis Safety Study)227 and the model assumes that there would be a discontinuation rate from aspirin of 1% following 6 months of chemoprevention. Health-related quality of life outcomes were not incorporated within the model and no information is reported in the paper around mortality rates due to any cause.

Endoscopy costs within the model were based on Medicare reimbursement rates and the cost of celecoxib was based on the Drug Topics Red Book. 228 The overall cost of treating CRC was taken from a study by Brown et al. ,229 which used SEER data to calculate average costs of treating CRC. All costs were valued at 1999 prices. All costs and health outcomes were discounted at a rate of 3%. One-way sensitivity analysis was carried out around the effectiveness of celecoxib. Two-way sensitivity analysis was carried out around the yearly cost of celecoxib and the effectiveness of the drug. Probabilistic sensitivity analysis was not presented within the paper.

The incremental cost-effectiveness ratio (ICER) for surveillance versus no prevention was estimated to be $27,970 per LYS in 1999 US dollars (approximately £22,000 in 2008 UK pounds). The ICER for celecoxib chemoprevention versus surveillance was estimated to be $1,715,199 per LYS (approximately £1,319,000 in 2008 UK pounds). The cost per LYS remains greater than £1 million in 2008 UK pounds within the one-way sensitivity analysis around the effectiveness of celecoxib. The two-way sensitivity analysis suggests that if the annual drug cost was below $150 (£115 in 2008 UK pounds) and with a risk reduction in the incidence of adenomas of greater than 50%, chemoprevention may become a reasonable option in comparison with colonoscopy surveillance.

The model presented by Arguedas et al. is subject to several limitations:

• The effectiveness of celecoxib is based on an assumption by the authors rather than clinical trial evidence (50% assumed risk reduction in adenoma incidence).

• The probability of adenoma formation (incidence) and CRC incidence is assumed to be constant and unrelated to age.

• The model assumes 100% compliance with both colonoscopy and celecoxib and that the sensitivity of colonoscopy is 100%.

• There is a limited analysis of uncertainty.

• There is no consideration of chemoprevention alongside endoscopic surveillance.

• HRQoL is not considered within the model.

• No model validation was presented.

Suleiman et al.: Chemoprevention of colorectal cancer by aspirin: a cost-effectiveness analysis

Suleiman et al. 219 assess the cost-effectiveness of aspirin chemoprevention (325 mg daily) and colonoscopy surveillance (once every 10 years; every 3 years in patients with polyps) versus no prevention, both individually and in combination. This model is based upon a previous model of CRC screening developed by Sonnenberg et al. 230 A Markov process model was used to simulate the experience of a cohort of 100,000 people aged 50 years over a lifetime horizon using an annual cycle length. No justification is provided concerning this starting age. The model assesses the cost per LYS as a result of prevention within the general US population from a third-party payer perspective (Medicare). The states within the model are:

1. a state after a negative colonoscopy without polyps/disease-free on aspirin prophylaxis

2. a state after colonoscopy plus polypectomy

3. a state after developing CRC

4. death.

The natural history of CRC is not explicitly modelled.

The model used a 1% annual polyp incidence rate to calculate the number of polypectomies and repeat colonoscopies after polypectomy. This rate was not related to age or whether the person had had polyps in the past. The annual age-specific incidence of CRC was estimated using SEER data. 222 Death from other causes was also included within the model based on standard US life tables. Colonoscopy and aspirin chemoprevention were assumed to reduce the incidence of CRC but have no effect on reducing the incidence of polyps. This is contrary to the clinical evidence identified within the systematic review described within Chapter 3 (see Results: aspirin), which suggests that aspirin does have an impact upon polyp incidence. Importantly, the model does not include a disease natural history component. No relationship between adenomas and cancer is specified, for example, it is unclear which cancer incidence rates are applied to persons with a negative screen result, a group that includes both true-negative and false-negative test results. Overall, the methodology adopted within this model which ‘tried to reduce the complex natural history of CRC to few essential states and avoid transition assumptions for which little or no published data existed’,230 resulted in a model that appears to fail to capture the events relevant to an assessment of either CRC screening or chemoprevention. The relative risk of CRC incidence was assumed to be 75% using colonoscopy surveillance and 50% using aspirin chemoprevention. These estimates were based on a range of sources including the NPS226 and the assumed values were tested within a one-way sensitivity analysis. The source of the estimate of the effectiveness of the combination of aspirin and colonoscopy in reducing CRC incidence is unclear; this was assumed to be 87.5%, a much higher relative effectiveness than either method of prevention alone. The model includes a probability of perforation and bleeding due to colonoscopy and polypectomy; however, there the model does not include the possibility of death due to prevention.

The costs of aspirin, colonoscopy, polypectomy and associated adverse events were taken from Medicare Payments in 2000. 230 The total cost of CRC was based on the uplifted costs from a study by Lee et al. 231 All costs were valued in 2000 US dollars. All costs and life-years were discounted at a rate of 3%. One-way sensitivity analyses were undertaken to explore the impact of the cost of chemoprevention and the preventive effectiveness of both colonoscopy and daily aspirin on resulting cost-effectiveness estimates. No further analysis of uncertainty was undertaken.

The results of the model suggest that the cost per LYS associated with chemoprevention versus no prevention is $47,249 in 2000 US dollars (approximately £35,000 in 2008 UK pounds). The use of aspirin chemoprevention and colonoscopy surveillance compared with colonoscopy surveillance alone results in a much higher estimated cost per LYS of $227,607 (approximately £167,000 in 2008 UK pounds). The authors suggest that based upon the one-way sensitivity analysis the costs of chemoprevention would need to fall below $70 per person per year (£51 in 2008 UK pounds) to become more cost-effective than colonoscopy.

The model presented by Suleiman et al. has several limitations:

• The disease natural history is not modelled.

• The model does not appear to account for costs and outcomes associated with cancer treatment if the cancer is curable.

• The model does not seem to account for mortality due to perforation.

• The model uses constant transition probabilities. With increasing age, polyp incidence will increase and more aspirin-related complications are likely to occur (although there is also some evidence to suggest that the effectiveness of aspirin may increase over time).

• It is not clear how compliance is dealt with in the model.

• No HRQoL has been accounted for within the model.

• The analysis is presented from a Medicaid perspective; however, it is unlikely that Medicaid would fund chemoprevention within the general population.

• There is a limited analysis of uncertainty.

• No model validation has been carried out.

Ladabaum et al.: Potential effect of cyclo-oxygenase-2-specific inhibitors on the prevention of colorectal cancer: a cost-effectiveness analysis217

Ladabaum et al. 217 assess the potential cost-effectiveness of COX-2 inhibitors (400 mg twice daily) versus no prevention. This is assessed with and without screening surveillance consisting of either colonoscopy or flexible sigmoidoscopy alongside FOBT. The same model previously developed by Ladabaum and described previously was used for this analysis. 216 The population considered within this analysis is people who:

• are at average risk of CRC

• have one first-degree relative who have previously been diagnosed with CRC

• have two first-degree relatives with/who have previously been diagnosed with CRC.

This model includes the effectiveness, costs and complications associated with COX-2 inhibitors rather than aspirin; however, all other model parameters and assumptions are as described within the 2001 paper. 216 The reduction in incidence of CRC as a result of COX-2 inhibitors is estimated to be 30% (range 0–100%) based on two studies assessing the use of COX-2 inhibitors on the reduction of CRC; one of which was in the FAP population. 89,232

The results of the model suggest that for a person at average risk of CRC, the cost per LYS will be $233,300 in 1998 US dollars (approximately £187,000 in 2008 UK pounds). For people with one and two first-degree relatives with CRC the cost per LYS decreases to $80,300 (£64,000 in 2008 UK pounds) and $56,700 (£45,000 in 2008 UK pounds), respectively. If CRC screening is current standard practice the cost per LYS of chemoprevention in addition to CRC screening is estimated to be greater than $195,000 (£156,000 in 2008 UK pounds) for all patient groups assessed. The one-way and two-way sensitivity analyses suggest that the key drivers of cost-effectiveness are the acquisition costs of COX-2 inhibitors and the effectiveness of COX-2 inhibitors; however, varying these parameters within plausible ranges is unlikely to affect the conclusions of the analysis.

Limitations of the model are as described in the 2001 paper described above. 216

Hur et al.: The cost-effectiveness of aspirin versus cyclo-oxygenase-2-selective inhibitors for colorectal carcinoma chemoprevention in healthy individuals

Hur et al. 221 estimate the costs and QALYs associated with celecoxib (400 mg twice daily) versus enteric-coated aspirin (325 mg/day) in healthy men aged 50 years. Aspirin or COX-2 inhibitors are not compared against ‘no prevention’ within the analysis. The analysis was undertaken from a US societal perspective over a 10-year time horizon. The authors also report other outcomes in terms of mortality and complication rates. The differences in costs and QALYs are not combined to produce a cost-effectiveness ratio.

A Markov model of the complications associated with aspirin and celecoxib was produced alongside the cardioprotective benefit of aspirin using a monthly cycle length. The natural history of CRC was not incorporated into the model; instead the authors assumed that the potential cancer benefits would be equally effective for either therapy. There is, however, no evidence to suggest that this is the case. Health-related quality of life utility values were taken from a study by Fryback et al. 233 Utility estimates were adjusted by age and gender using the ‘Beaver Dam Health Outcomes Study’,233 and adjusted to reflect recuperation from complications. The probability that an individual will experience complications (including ulcer, ulcer perforation, bleed, myocardial ischaemia) was taken from several literature sources including Sanmuganathan et al. ,234 the PHS235 and Spiegel et al. 236 The model also includes a probability of death due to these complications. Importantly, death due to other causes was not included within the model.

The costs of aspirin and celecoxib were based on the Drug Topics Red Book228 and the costs of treating most of the complications associated with aspirin or celecoxib were based on a study by Spiegel et al. 236 Costs were valued in 2000 US dollars. All costs and outcomes were discounted at a rate of 3%. One-way sensitivity analyses were undertaken to examine the impact of starting age, length of follow-up, the cost of the drug and the discount rate on cost-effectiveness outcomes. Threshold analyses were also reported around several additional parameters including the relative excess myocardial infarction rate for aspirin compared with COX-2 inhibitors, the relative ulcer rate for aspirin compared with COX-2 inhibitors, the cost of celecoxib and the relative bleeding rate of aspirin compared with COX-2 inhibitors.

The results of the model suggest that aspirin is associated with an increase in QALYs (7.60 vs 7.57) and lower costs ($181 vs $23,403) than celecoxib, i.e. aspirin dominates celecoxib. The results were not sensitive to the start age, the length of follow up or the discount rate. The cost of celecoxib affected the model results substantially; however, this is unlikely to alter the conclusions.

The model reported by Hur et al. is subject to several limitations:

• The report presents a cost-effectiveness analysis comparing aspirin and celecoxib; however, it is not known whether either chemopreventive agent would be cost-effective compared with a ‘do nothing’ alternative.

• The model assumes that the efficacies of aspirin and celecoxib are equivalent (no prevention of cancer is incorporated).

• The analysis is presented from a Medicaid perspective; however, it is unlikely that Medicaid would fund chemoprevention within the general population.

• There is a limited analysis of uncertainty.

Dupont et al.: Aspirin chemoprevention in patients with increased risk for colorectal cancer: a cost-effectiveness analysis220

Dupont et al. 220 present a cost-effectiveness analysis of aspirin chemoprevention (325 mg daily) and colonoscopy surveillance (every 3 or 5 years depending on polyp risk) versus no prevention, individually and in combination. A cohort of individuals with prior adenoma resection (complete colonoscopy and polypectomy) without FAP or HNPCC are simulated from age 50 over a lifetime within a Markov model using annual cycles. No justification was provided within the paper concerning the starting age of the model cohort. Outcomes are presented in terms of the cost per LYS and the cost per case of CRC prevented. The analysis was carried out from the perspective of a third-party payer (Medicare).

The natural history of the disease was modelled using the health states normal (no polyps or cancer), low-grade adenoma, high-grade adenoma, carcinoma in situ, advanced CRC and death. The annual probability of developing polyps was obtained from the placebo group in the study by Baron et al. 74 and was not age-related. The relative risk of polyp incidence was assumed to be 0.85 based on three studies of aspirin use within this population. 62,74,79 The initial colonoscopy was assumed to be 100% successful in removing all polyps, but a miss rate of 8% for subsequent colonoscopies was assumed. 237 All colonoscopies were assumed to be complete and compliance with aspirin and colonoscopy was assumed to be 100%; these assumptions are both highly optimistic. The probabilities of developing complications associated with aspirin or colonoscopy were drawn from a number of journal sources including the paper by Ladabaum et al. reviewed here. 216 The model assumes that death may occur as the result of a colectomy, aspirin or colonoscopy complication, CRC or other causes.

The cost of aspirin was based on the costs used in Suleiman et al. 219 and Ladabaum et al. ,216 both described above. The cost of colonoscopy and polypectomy, complications of colonoscopy and aspirin, initial side effects of aspirin and CRC resection were based on Medicare reimbursement rates. The cost of treating incurable CRC was based upon several sources including Arguedas et al. 218 and Ladabaum et al. 216 described above. A formal price year for model results was not reported; for the purposes of uplifting the costs, the authors assume that it is 2006 given that it was published in 2007. Costs and life-years were discounted at a rate of 3%.

Probabilistic sensitivity analysis and a one-way sensitivity analysis were undertaken. A scenario analysis where the length of time between colonoscopy surveillance was increased was also reported. No uncertainty was considered around the reduction in polyp incidence associated with aspirin. Results were presented incrementally. Aspirin is predicted to cost $87,609 per LYS compared with no prevention (approximately £50,000 in 2008 UK pounds). The use of aspirin chemoprevention in addition to colonoscopy compared with colonoscopy alone is estimated to cost $60,942 per LYS (approximately £35,000 in 2008 UK pounds). These results presented by Dupont et al. are less favourable than other studies in terms of the use of aspirin as a chemopreventive agent compared with no prevention strategy; however, they are more favourable than those from other studies when considering the use of aspirin in addition to colonoscopy surveillance.

The model presented by Dupont et al. is subject to several limitations:

• The model assumes that the initial colonoscopy is 100% successful in removing all polyps within the base case.

• The model assumes that the compliance rate is 100% for both surveillance and aspirin use.

• The model uses constant transition probabilities. With increasing age, polyp incidence will increase and more aspirin-related complications are likely to occur (although there is also some evidence to suggest that the effectiveness of aspirin may increase over time).

• HRQoL is not considered within the model.

• The analysis is presented from a Medicaid perspective; however, it is unlikely that Medicaid would fund chemoprevention within the general population.

• The model does not consider any uncertainty around the reduction in polyp incidence as a result of aspirin use.

• No model validation has been carried out.

Current chemoprevention model limitations

This review identifies a number of key limitations employed within existing economic evaluations of CRC chemoprevention, outlined in Box 2 below.

This review highlights the limited number of economic assessments that have been undertaken around chemopreventive agents. The only chemopreventive agents that have been modelled are aspirin and COX-2 inhibitors, and there are numerous limitations associated with these economic evaluations as described above. None of the economic evaluations undertaken to date are appropriate for use within this analysis; hence a novel health economic model has been developed.

Methods

Model structure

In 2003, Tappenden et al. 238 developed a state transition model to assess the cost-effectiveness of alternative options for CRC screening. This model has been further developed to incorporate the expected costs and outcomes associated with chemoprevention. This model is comprised of three inter-related model components: a model of the natural history of CRC, a model of screening and surveillance interventions, and a mortality model.

Natural history submodel

The natural history of CRC was modelled as a series of transitions between the following mutually exclusive health states: normal epithelium, low-risk adenomas, high-risk adenomas, Dukes’ A, Dukes’ B, Dukes’ C and Dukes’ D as shown in Figure 15.

FIGURE 15.

Model schematic of disease natural history.

The model uses age-dependent transition matrices to estimate the number of individuals in each health state for each model cycle through a process of iterative matrix multiplication. Transitions between model health states are calculated on a cohort-level basis using an annual cycle length and a lifetime horizon. Health states were defined according to the true underlying histological state of the individual. Health states describing the presence of neoplasia (adenomas and/or cancer) were defined in terms of the ‘index’ lesion; that is the adenoma with the greatest malignant potential or the most advanced tumour. Individuals with adenomas were defined as ‘low-risk’ or ‘high-risk’ to reflect current guidelines for endoscopic surveillance following adenoma removal. 59 Cancer stage was modelled according to the Turnbull modification of the Dukes’ staging system. 239,240

Screening submodel structure

The potential impact of earlier detection and removal of screen-detected adenomatous polyps, the detection and treatment of CRC, and the ongoing surveillance of high-risk individuals in whom adenomas are detected are simulated processes within the model. Test characteristics associated with screening and colonoscopy are defined in terms of the probability of achieving a positive or negative test result given an individual’s true underlying histological state (i.e. sensitivity and specificity). The impact of the screening test, follow-up colonoscopy and treatment of detected polyps and cancers is modelled by redistributing the model cohort across the health states at the point of screening and surveillance. Individuals in whom adenomas are found are assumed to undergo polypectomy via snare diathermy and are subsequently assigned a higher risk of adenoma recurrence. 241 Individuals identified as having high-risk polyps are divided into two groups; higher-risk and intermediate-risk. This terminology is unrelated to the use of the terms high-risk (i.e. FAP/HNPCC) and intermediate-risk (i.e. history of adenomas) within the clinical review. Individuals identified as having low-risk polyps continue to be classified as low-risk within the model. Individuals who are identified as being at an intermediate risk of developing CRC following screening are assumed to be invited to attend a 3-yearly colonoscopic surveillance programme, whereas those at higher risk are assumed to enter into an annual colonoscopic surveillance programme. It is assumed that individuals with preclinical detectable CRC may present symptomatically during any model cycle; the probability of clinical presentation was estimated through model calibration; this probability is assumed to increase according to cancer stage. These parameters are shown in Table 31. Individuals in whom previously undetected CRC is identified are assumed to enter into a clinical management state. Individuals in whom neither adenomas nor cancer are detected are assumed to be reinvited to attend CRC screening during the next round in 2 years. Clinical management states are modelled according to Dukes’ stage as prognosis, treatment options, HRQoL impacts and costs differ considerably between stages.

TABLE 31 - Disease natural history parameters

Natural history model transition probability	Lower 95% CI	Upper 95% CI
Normal epithelium to low-risk polyp – age 30	0.001	0.003
Normal epithelium to low-risk polyp – age 40	0.000	0.002
Normal epithelium to low-risk polyp – age 50	0.004	0.006
Normal epithelium to low-risk polyp – age 60	0.003	0.010
Normal epithelium to low-risk polyp – age 70	0.008	0.014
Normal epithelium to low-risk polyp – age 80	0.000	0.009
Normal epithelium to low-risk polyp – age 90	0.000	0.024
Normal epithelium to low-risk polyp – age 100	0.000	0.026
Low-risk polyp to high-risk polyp	0.048	0.109
High-risk polyp to Dukes’ A	0.037	0.086
Normal epithelium to Dukes’ A	0.000	0.000
Dukes’ A to Dukes’ B	0.574	1.000
Dukes’ B to Dukes’ C	0.507	1.000
Dukes’ C to Dukes’ D	0.577	0.998
Dukes’ A to Dukes’ A clinical	0.058	0.102
Dukes’ B to Dukes’ B clinical	0.162	0.285
Dukes’ C to Dukes’ C clinical	0.404	0.550
Dukes’ D to Dukes’ D clinical	0.534	0.999

Although based on the same conceptualisation of the underlying disease process and clinical pathways, the decision node differs for the analysis of chemoprevention within the general population and the intermediate-risk groups. For the economic analysis of chemoprevention within the general population, the simulation begins with a cohort aged 30, at which point the prevalence of polyps and CRC is assumed to be zero. The model uses age-dependent transition matrices as described above to calculate the number of individuals in each health state for each annual cycle over a lifetime horizon. For the intermediate-risk population, the analysis begins immediately after the point of postpolypectomy.

Mortality submodel

The model includes three types of mortality: other-cause mortality, CRC-specific mortality and death following endoscopic perforation. The probability of dying from other causes is modelled as an age-dependent probability, which is dependent on the age of the cohort during each model cycle, based on UK life tables. 242 Dukes’ stage-specific mortality rates are represented using a mixed model, which divides the population into terminal and non-terminal cancer patients. This analysis is based upon 1-, 3- and 5-year survival data produced by the Northern and Yorkshire Cancer Registry and Information Service (see Figure 16). 243 The probability of death due to endoscopic perforation was modelled according to the experience of existing screening studies. 59,37

FIGURE 16.

Survival estimates for CRC patients.243

Model parameters

The following sections outline the methods used to derive specific types of parameter values within the model.

Methods for calibrating the natural history model

Importantly, model parameters such as transition probabilities for preclinical disease and the probabilities of symptomatic presentation are largely unobservable through empirical study. These unknown parameters were estimated through a process of model calibration. This involved fitting intermediate model outputs to available data on CRC incidence and mortality,243 cancer and polyp detection rates at screening (NHS Cancer Screening Programme, personal communication) and colorectal autopsy evidence244 using a Metropolis–Hastings algorithm (an analytic Markov Chain Monte Carlo sampling approach). This algorithm allows the calculation of a set of possible permutations of transition probabilities used to represent the disease natural history. Rather than assuming that one particular combination of transition probabilities is ‘correct’ within a base case, the base model is probabilistic and includes several sets of possible permutations of natural history transition probabilities. In describing the disease natural history, there is structural uncertainty around whether there are a proportion of cancers which arise de novo (i.e. there may be some cancers that do not arise following polyp incidence). The calibration process allows a small proportion of cancers to arise de novo. As this proportion is unknown, a prior distribution was used assuming that a mean of 5% of cancers arise de novo (95% CI 1% to 13%) within the model calibration process. This resulted in an average of around 1–2% of cancers arising de novo within the model. The ranges around the parameters used to simulate the disease natural history are shown in Table 31.

Methods for modelling the effectiveness of chemoprevention

The systematic review presented in Chapter 3 highlighted considerable evidence concerning the effectiveness of chemoprevention in the general population and intermediate-risk groups. However, the evidence base leaves a number of relevant questions unanswered. In particular, key uncertainties concern the length of time required for chemoprevention to start to become effective in reducing adenomas, whether the effectiveness is constant over time or gradual, and the length of time over which chemoprevention is likely to continue to be effective, the relationship between compliance and effectiveness, and whether additional benefits accrue when an individual has stopped taking the chemopreventive agent. Furthermore, none of the studies identified within the review attempted to establish an optimal treatment duration, and limited data are available surrounding optimal dosing regimens. Owing to these uncertainties, it is unclear how chemoprevention would be adopted in practice within England and Wales. Based on published literature and expert opinion, the model assumes the following within the base-case analysis.

• The general population will begin chemoprevention at age 40 or 50 (both ages tested within analyses) Polyp prevalence estimates suggest that polyps generally begin to appear when people are aged over 40 years. 245

• The intermediate-risk population will be offered chemoprevention following a polypectomy Within the model several ages are tested; the polypectomy is assumed to be undertaken at ages 60, 64 or 70 and chemoprevention is assumed to begin a year later at ages 61, 65 or 71.

• Chemoprevention will not be taken beyond age 80 years in either the intermediate-risk population or the general population This is based on the assumption that chemoprevention will have an effect on reducing CRC up to 15 years post-treatment21 because of the time lag between adenoma incidence and CRC development. Hence chemoprevention is unlikely to be beneficial beyond age 80 years. This is varied from age 50 to age 80 within the analyses.

Assumptions around treatment dosing are based upon the included studies describing the effectiveness of chemoprevention. The following dosing was assumed within the model:

• aspirin: 300 mg/day

• celecoxib: 400 mg/day

• calcium: 1200 mg/day.

The following additional effectiveness assumptions have been made within the base-case model:

• The effects of chemoprevention manifest at the point at which the individual begins taking the agent A study by Baron et al. reported a positive effect of rofecoxib after 1 year of treatment. 153 Given the absence of evidence to the contrary, it is assumed that all chemopreventive agents begin to take effect within the first year of treatment. It should be noted that the initial impacts of chemoprevention would be upon adenoma incidence rather than CRC incidence.

• The relative risk of developing low-risk adenomas will be constant over the period during which the individual is taking chemoprevention Only two studies identified within the review report results at more than one follow-up time (Arber and Bertagnolli)63,64. These studies suggest that the relative risk of developing adenomas when given celecoxib is likely to be similar at 1 year and at 3 years. Therefore, the model assumes that this assumption can be applied to other chemopreventive agents and over longer periods of time such that the relative impact of chemoprevention upon the number of cancers is constant while chemoprevention is being taken.

• The preventive effects for adenomas of all chemoprevention agents drop off at the point at which the patient stops taking the agent Flossmann and Rothwell21 suggest that the impact of chemoprevention upon CRC incidence decreases around 15 years after treatment with chemoprevention has stopped; this is approximately consistent with the reported average time taken to progress from a polyp to CRC suggesting that the main impact of chemoprevention is likely to be on polyp incidence.

The impact of the above assumptions is tested within the model analysis to examine potentially optimal indications for chemoprevention (see Results). It should be noted that within the model it is assumed that chemoprevention will continue to be taken for 10 or 20 years; however, no trials have assessed treatment duration greater than 5 years; hence the effectiveness of taking chemoprevention over this longer time frame is not known.

Preventive effectiveness estimates were based only upon studies that compare chemoprevention with placebo; studies in which patients received other chemoprevention agents in both arms have not been included. It should be noted that there are heterogeneities between the studies used to estimate effectiveness of the chemopreventive agents such as age and treatment time. These are discussed in more detail in Chapter 3.

The clinical trials identified within the systematic review provide evidence for a wide range of chemopreventive agents. Outcomes reported within the clinical trials varied according to the population receiving chemoprevention. Within the general population trials, the number of people developing CRC was reported but the number of people developing polyps was not, which is likely to be because of ethical concerns around undertaking colonoscopies on healthy individuals within the general population. Within the intermediate-risk population, the number of people with advanced adenomas and the number of people with any adenomas were commonly reported; however, the follow up and sample sizes were insufficient for CRC incidence to be reported. In the instances where CRC incidence was reported in the intermediate-risk group, too few events had occurred to suggest an effect. Many of the chemopreventive agents were shown to be ineffective in preventing adenomas within intermediate-risk populations, or in preventing CRC in the general population. Those chemopreventive agents that were reported to be ineffective were not included in the economic analysis because they are expected to be dominated. For the purposes of the model analysis, the chemopreventive agents were considered to be ineffective if they had a pooled relative risk of developing polyps/CRC of 1.0 or greater. In instances whereby the meta-analysis of available trial data resulted in a mean relative risk that was below 1.0, the chemopreventive agent has been included in the analysis.

The systematic review of clinical effectiveness suggests that aspirin chemoprevention is likely to be effective within the general population. Other NSAIDs have not been tested within the general population, and calcium and antioxidants currently have limited trial follow-up which may not be sufficient to demonstrate any effects of chemoprevention upon cancer incidence. The only chemopreventive agent assessed within the model for the general population is therefore aspirin. The chemopreventive agents assessed within the modelling framework for the intermediate-risk group are aspirin, celecoxib and calcium. The evidence around the use of antioxidants for chemoprevention provided variable results. The results of some of these trials are contradictory and the effectiveness of both individual antioxidants and combinations of antioxidants is unclear. Antioxidants were therefore not included within the main analysis.

Methods for modelling the effectiveness of chemoprevention

The current evidence suggests that chemoprevention, if effective, will have an effect upon the incidence of low-risk adenomas. Studies also suggest that there may be some impact upon the incidence of high-risk adenomas, but there is currently no evidence to suggest whether this is as a result of the reduction in low-risk adenomas or if there is an additional direct impact upon high-risk adenomas. There is also currently limited evidence to suggest whether there would be a direct impact upon CRC. The model assumes that chemoprevention has a direct impact upon polyp incidence and hence an indirect impact upon the incidence of CRC. The model assumes that there would be no direct impact upon CRC incidence. The model also assumes that people would discontinue receiving chemoprevention if they were diagnosed with CRC.

General population

The relative risk for CRC incidence for patients receiving aspirin within the general population was based upon a meta-analysis of two studies77,80 for which the long-term CRC incidence data have been reported in Flossmann and Rothwell. 21 The results of this meta-analysis are shown in Table 32.

As described previously, there is a probability of progressing through each stage of the disease based upon calibrated disease natural history parameters. Within the model, a relative risk can be applied to these transition probabilities based upon the estimated effect of chemoprevention upon each transition. Assuming that chemoprevention has an equivalent impact upon the incidence of low-risk polyps and the incidence of high-risk polyps, the calibrated disease natural history model was used to estimate these relative risks based on fitting the number of cancers within the model to match the number of cancers reported within the trials for both the control group and the aspirin chemoprevention group. Flossmann and Rothwell21 suggest that the chemopreventive effect upon CRC incidence is minimal for the first 10 years; however, from 10 to 19 years the relative risk for CRC incidence is estimated to be 0.61. This is consistent with the current model of disease natural history, which suggests that on average it takes more than 10 years to progress from normal epithelium to adenoma incidence to cancer incidence. The model therefore uses the relative risk from years 10–19 from Flossmann and Rothwell21 and matches the number of cancers for both the control group and the aspirin chemoprevention group at this time point. This means that the relative risk of CRC incidence will gradually decrease from 1 at initiation of chemoprevention (when there will be an effect upon adenoma incidence only) to 0.61 at 10–19 years. This process was undertaken using the solver add-on in excel using a model where screening is not currently rolled out so as to correspond to the trials.

These figures were derived within a framework that does not include screening so it is not possible to calculate the relative risks of the transitions from postpolypectomy to low-risk or high-risk polyps. Within the intermediate-risk population model, the relative risk associated with the transition from postpolypectomy to low-risk polyps is assumed to be the same as the relative risk associated with the transition from low-risk polyps to high-risk polyps, whereas the relative risk associated with the transition from postpolypectomy to high-risk polyps is allowed to be different (see Intermediate-risk population). Therefore, within the general population model, it is assumed that the relative risk associated with the transition from postpolypectomy to low-risk polyps is equivalent to that derived for the relative risks associated with the transitions from normal epithelium to low-risk polyps and from low-risk polyps to high-risk polyps. Using this method, because the relative risk of the incidence of CRC was so low (0.61), the relative risk associated with these transitions was low (0.0419). Within the intermediate-risk model, the relative risk associated with the transition from postpolypectomy to high-risk polyps was consistently lower than the relative risk associated with the transition from postpolypectomy to low-risk polyps. Within the general population model, the relative risk associated with the transition from postpolypectomy to high-risk polyps was therefore assumed to be zero. As such a small proportion of patients follow this pathway within the general population model, this assumption would not have a large impact upon the model results. These assumptions are shown in Figure 17.

FIGURE 17.

Relative risk (RR) assumptions for chemoprevention. TP, transition probability.

For the probabilistic sensitivity analysis, a distribution was fitted to the relative risk shown in Table 32 and a value was sampled from this distribution. This sampled value was then used within solver to estimate the relative risks for the transition probabilities outlined in Figure 17. This process was repeated 1000 times to generate a distribution for these relative risks.

TABLE 32 - Relative risk of incidence of CRC in general population

Chemopreventive agent	RR for incidence of CRC at 10–19 years	Source
Aspirin	0.61 (95% CI 0.43 to 0.88)	Two trials: Farrell,77 Peto80 Long-term follow-up data presented in: Flossmann21

Intermediate-risk population

Within the clinical trials, the number of people with low-risk adenomas was not provided; however, the number of people with any adenoma and the number of people with high-risk adenomas was provided within the intermediate-risk group. The number of people with low-risk adenomas was therefore estimated as the difference between these two values. The following calculation was undertaken to calculate event rates of low-risk adenomas in each arm of the study:

No. of low-risk adenomas = [ ( Cumulative rate of all​ adenomas − Cumulative rate of high-risk adenomas ) / 100 ] × Total number of patients in study arm

The result of this calculation was then used within the meta-analysis (see Table 33).

TABLE 33 - Relative risk of incidence of low-risk and high-risk polyps in the intermediate-risk population

95% CI in parentheses.

Chemopreventive agent	RR for incidence of low risk polyps (95% CI)	RR for incidence of high risk polyps (95% CI)	Source
Aspirin	0.93 (0.76 to 1.14)	0.75 (0.52 to 1.07)	Logan,60 Cole,61 Benamouzig62
Celecoxib	0.77 (0.69 to 0.86)	0.45 (0.35 to 0.58)	Arber,63 Bertagnolli64
Calcium	0.84 (0.68 to 1.03)	0.77 (0.50 to 1.17)	Bonithon-Kopp,96 Baron99

The relative risk of developing low-risk and high-risk adenomas for patients receiving aspirin was based upon a meta-analysis of three studies. 60–62 Within both celecoxib studies identified by the review,63,64 the event numbers at the 3-year follow-up were biased by the intervention following the 1-year follow-up. The cumulative proportion of events therefore provided a more reasonable approximation for calculating the relative risks. Two calcium studies were identified within the review for the intermediate-risk population. 96,99 Within Bonithon-Kopp,96 high-risk adenomas were described as either > 1 cm or villous; the former of which was used to calculate incidence of low-risk adenomas. Within Baron,99 the relative risk of developing high-risk adenomas was not reported; however, the authors were contacted and were able to provide these data for use within the analysis. Both of these calcium studies also provided CRC incidence; however, event numbers were close to zero because of the short follow-up of the trials. Therefore these studies could not be used for the purposes of validation. The relative risks of the incidence of low-risk and high-risk polyps in the intermediate-risk population are shown in Table 33.

For the intermediate-risk population, it was assumed that an equivalent relative risk can be applied to the transition probability from the postpolypectomy state to low-risk polyps and from low-risk polyps to high-risk polyps. A second relative risk was assumed to apply to the probability of transiting from normal epithelium following a polypectomy to high-risk polyps (Figure 17). The calibrated disease natural history model was used to estimate these relative risks based on fitting the number of low-risk polyps and high-risk polyps within the model to match the number of low-risk polyps and high-risk polyps reported within the trials for both the control group and the chemoprevention group. This process was undertaken using the solver add-on in excel as in the general population model. The samples for the PSA values were generated in the same way as those for the general population model.