NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Rostom A, Dube C, Lewin G. Use of Aspirin and NSAIDs to Prevent Colorectal Cancer [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2007 Mar. (Evidence Syntheses, No. 45.)

Cover of Use of Aspirin and NSAIDs to Prevent Colorectal Cancer

Use of Aspirin and NSAIDs to Prevent Colorectal Cancer [Internet].

Show details

Appendix 4. Detailed Report Supporting Sections

Study Identification

Literature search

A highly sensitive electronic search strategy was developed in Medline (Appendix 1) and modified for the other databases. For questions 1 and 2, three main concepts were included: NSAIDs or chemoprevention; CRC or intestinal polyps; and relevant study designs. Study designs sought were randomized controlled trials (RCTs; using the Cochrane highly sensitive search strategy plus “versus”)1 and case-control and cohort studies (using a filter for observational study designs based on the SIGN filter [available at, visited December 1, 2004] and terms suggested in McKibbon).2 A comprehensive retrieval strategy for NSAIDs was derived from the indexing in both Medline and Embase, reviewer nominated terms, and previous reviews.37 Terms were derived from the National Cancer Institute (NCI) Cancer topic searches (available at ) for “Colorectal Cancer.” Adding the term “Adenomatous polyps,” increased the sensitivity by removing major emphasis and by retrieving text words in all fields rather than just title. The search was limited to English language reports, and non-human studies were excluded. The search was validated by testing to see if it retrieved included studies from two recent major reviews.79

Databases searched were Medline 1966-November week 3, 2004, Embase 1980-week 47 2004 (publication years 2003-2005 only) and CENTRAL, The Cochrane Library Issue 4, 2004. Pubmed Cancer subset was searched for non-Medline material. That search used the NCI search strategy for CRC with free text terms for NSAIDs and COX-2 inhibitors (COX-2 inhs) and was also limited to English. The search was conducted on December 1, 2004 and yielded seven items. The strategy is reproduced in Appendix 1.

A supplemental search was made of non-Medline material from the PubMed Cancer subset. This search used the NCI search strategy for CRC with free text terms for NSAIDs and COX-2 inhs and was limited to English. The search was conducted on December 1, 2004 and yielded seven items. The strategy is reproduced in Appendix 1.

Additional material potentially relevant to the economic analysis question (question 4 of the task order) was sought in Medline (1966 to November Week 3 2004), HealthStar (1987 to November 2004), Embase (1980 to 2004 Week 50), Cochrane Library 4th Quarter 2004 NHS EED and HTA. TRIP ( database was searched December 14, 2004.

A search strategy to detect recent systematic reviews of NSAID that appeared to address harm was developed and run in Medline (2003 to November Week 3 2004). Cochrane Database of Systematic Reviews (CDSR) and DARE (Cochrane Library, 3rd Quarter 2004) were searched for all systematic reviews related to NSAIDS, without date restrictions.

A weekly monitoring strategy was implemented to detect emerging information on CV harms associated with COX-2 inhs. We ran the harms search against new material added to Medline or PreMedline with the exception that a concept restricting the results to CV disease was added and the search was not limited to systematic reviews but included all new reports of NSAIDS and harms. In addition, any records that related to CV disease matching our search strategy for NSAIDS (Appendix 1) were obtained — these included case reports, comments, consensus development, conference proceedings, editorials or letters, news or newspaper articles, published erratums, as well as retracted publications. We also monitored the FDA news digest and Health Canada's Health Product Information mailing list for announcements related to COX- 2 inhs and CV harms (monitoring dates Jan 14, 2005–May 26, 2005).

Literature Synthesis and Preparation of Systematic Evidence Review

Study selection methods

1) Effect of NSAIDs on the risk of CRC and/or CRA. A calibration exercise was conducted prior to the initiation of study selection. Screening of articles for inclusion was conducted for each screening level by two members of the review team (see Appendix 2 for screening forms). An initial screening level to identify potentially relevant articles was followed by a relevance assessment to identify articles meeting inclusion criteria. Conflicts were resolved by consensus.

A third level of screening was included (for questions 1 and 2, efficacy) to discriminate the different levels of evidence, as follows: 1) RCT parallel design; 2) RCT crossover design; 3) RCT factorial design; 4) controlled clinical trial (non-RCT), 5) multiple prospective cohort; 6) at least one prospective cohort and one retrospective cohort; 7) case-control study; and, 8) cost-effectiveness analysis.

Data abstraction (see Appendix 2 data abstraction form) was performed by one reviewer using the SRS system and electronic forms and was checked by a second reviewer.

Inclusion/exclusion criteria(efficacy)

Design: RCTs, controlled clinical trials, and observational studies (cohorts and case-control) were considered for inclusion if they fulfilled the population, intervention, and outcome criteria detailed below.

Population: Studies were considered for inclusion if the study population included participants at ‘average’ risk for CRC (i.e., no known risk factors for CRA or CRC other than age), or a personal or family history of CRA or sporadic CRC. Studies involving participants with familial adenomatous polyposis (FAP) or hereditary non-polyposis colon cancer syndromes (Lynch I or II) were excluded since these syndromes account for a small percentage of CRC and represent distinct clinical entities with specific management strategies. Secondary prevention studies of patients with a personal history of CRC were also excluded.

Interventions: Included studies assessed the efficacy or effectiveness of ASA, or non-ASA NSAIDs including COX-2 inhs.

Outcomes: The incidence of CRA and/or CRC; reductions in CRC-related mortality or overall mortality.

2) Harms: The GI, CV, and renal harms associated with the use of ASA, non-ASA NSAIDs, and COX-2 inhs were sought through identification of systematic reviews. Authors of systematic reviews were contacted by electronic email in the case that incomplete harms data was provided in the systematic review report. The timeline for response was set a priori by the review team as 2 weeks. If no response was obtained, the review was excluded on the basis of incomplete data.

The GI harms included: incidence of endoscopic ulcers; clinically important ulcers (POB-perforation, obstruction, bleeding, and PUB- perforation, obstruction, bleeding or symptomatic ulcer); and, GI related mortality. The CV harms included: stroke; myocardial infarction (MI); congestive heart failure; hypertension; and, CV related deaths. The renal harms included: acute renal failure; chronic renal failure; need for dialysis; ans, renal related mortality.

3) Cost-effectiveness: A specific search was conducted to identify cost-effectiveness analyses that addressed the question of the cost-effectiveness of chemoprevention with ASA or non-ASA NSAIDs for the prevention of the above listed endpoints.

Quality assessment

We used predefined criteria from the USPSTF to assess the internal validity of included systematic reviews, trials and observational studies, which we rated as “good”, “fair”, or “poor” (see Appendix 3) (US Preventive Services Task Force Procedure Manual. Revised July 2004).


Figure 2 and Table 1 illustrate a hierarchical framework that identifies key characteristics that were expected to be common to all the included studies. This framework was used to facilitate study grouping and subsequent data analysis. Studies were initially grouped by the “disorder” (i.e., adenoma vs. CRC), study design, study population and exposure, as these characteristics were expected to be reliably and consistently reported. Subsequently, studies were subcategorized based on the measures of dose effect and secondary outcomes, such as comparisons of polyp size, histology, morbidity or mortality. Evidence tables based on the illustrated framework were created to aid in subgroup generation (Appendix 8). We anticipated the dose effect measure to be the most heterogeneous outcome reported between studies. Measuring the dose effect depends on the intervention dose, the frequency and duration of use, and potentially, whether the use is current and ongoing or was at some remote time in the past. In RCTs, dose effect was precisely defined, while in observational studies we anticipated variability in the definitions of “regular use” and that the dose, per se, would be rarely reported. When possible, data was reported based on a precise dose or based on defined levels of dose intensity within a study (such as low moderate or high dose). When this was not possible, we defined various levels of dose effect based on the duration of regular use of an intervention (e.g., regular use for: 1–3 years; 4–6 years; 7–9 years; and 10 or more years in the case of CRC). For CRC, studies of regular use of the intervention for <1 year or intermittent use for <5 years were not included in any meta-analytical pooling. This level of exposure was not felt to be “relevant” given the presumed mechanism of chemoprevention of CRC. Intermittent use for >5 years was considered separately, and as a meta-analytical subgroup analysis with the “lower dose” exposure group. For adenoma incidence outcomes, shorter durations of exposure were considered.

To perform the analysis of regular NSAID use versus nonuse, we compared the risk of outcome (CRC or adenoma according to the case) in the patients exposed to the equivalent of 3 days or more per week for at least a year, with nonexposed patients. When the frequency or duration of use was not reported, we compared the risk of outcome in patients with “any NSAID use” with nonexposed patients.

Ascertainment of NSAID exposure and outcomes in the observational studies were also anticipated to be variable. The method of ascertainment of exposure of each study was abstracted and recorded in evidence tables to facilitate data analysis and interpretation. In cases where statistical pooling was possible, sensitivity analyses were conducted based on the methods of ascertainment used (i.e., questionnaire, database).

We only pooled when I2 was 50% or less, which Higgins et al.10 refer to as “moderate” heterogeneity. In the event that pooling was conducted, we used the reported relative risks (RRs) and corresponding 95% confidence intervals (CIs), rather than using cell counts from 2×2 tables. This was because analysis results from observational studies require adjustment for potential confounders, so that unadjusted cell counts are not appropriate. For generating the CIs, we computed the CI width by subtraction and then divided by 2*1.96 to obtain the standard error. Calculations were performed on a log scale because the standard errors of relative risks are most appropriately represented on this scale. In one case, in which no CIs were provided,11 we chose a CI from two different reported estimates. We then computed the standard error based on that imputed CI. The pooling was conducted using inverse-variance weighting and a random effects model.12 In all cases, we used study groupings that appeared to be clinically comparable.

Heterogeneity in efficacy data

Included studies were grouped using the methods detailed above in an effort to form logical, usable subgroups and to reduce heterogeneity. In the generated subgroups, heterogeneity was further assessed through the use of Forest plots and through the use of the I squared test for heterogeneity. Statistical pooling of results was only considered if clinically and statistically appropriate.

Analysis of harms and cost-effectiveness data

For the harms and cost-effectiveness data, the results of the included systematic reviews and cost-effectiveness studies were summarized and presented as a qualitative systematic review.


Optimizing search strategies to identify randomized controlled trials in Medline. 04 Oct 2; Ottawa, Canada: XII Cochrane Colloquium, 2004.
McKibbon A. PDQ Evidence-based principles and practice. Hamilton: B.C. Decker Inc., 1998.
Rostom, A., Dube, C., Jolicoeur, E., Boucher, M., and Joyce, J. Gastroduodenal ulcers associated with the use of non-steroidal anti-inflammatory drugs: a systematic review of preventative pharmacological interventions. Ottawa, Canada: CCOHTA, 2004. Technology Report Issue 38.
Rostom A, Dube C, Wells G et al. Prevention of NSAID-induced gastroduodenal ulcers. Cochrane Database Syst Rev 2002;(4):CD002296. [PubMed: 12519573]
Asano TK, McLeod RS. Non steroidal anti-inflammatory drugs (NSAID) and Aspirin for preventing colorectal adenomas and carcinomas. Cochrane Database Syst Rev 2004;(2):CD004079. [PubMed: 15106236]
Rostom A, Dube C, Wells G et al. Prevention of NSAID-induced gastroduodenal ulcers. Cochrane Database Syst Rev 2002;(4):CD002296. [PubMed: 12519573]
Asano TK, McLeod RS. Non steroidal anti-inflammatory drugs (NSAID) and Aspirin for preventing colorectal adenomas and carcinomas. Cochrane Database Syst Rev 2004;(2):CD004079. [PubMed: 15106236]
Giovannucci E. NSAIDs: Role in prevention of colorectal cancer. In: Rose BD, editor. MA: Wellesley, 2004.
Giovannucci E. NSAIDs: Role in prevention of colorectal cancer. In: Rose BD, editor. UpToDate. Wellesley, MA: 2004.
Higgins JP, Thompson SG, Deeks JJ. et al. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557–560. [PMC free article: PMC192859] [PubMed: 12958120]
Paganini-Hill A. Aspirin and colorectal cancer: the Leisure World cohort revisited. Prev Med. 1995;24(2):113–115. [PubMed: 7597009]
DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177–188. [PubMed: 3802833]
PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to pubmed

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...