We searched the literature using PubMed, CINAHL and Scopus for all studies on physical activity or exercise and colon polyps through April 2010. We employed the terms exercise and physical activity in combination with colon polyps using the terms colon polyp, colon adenoma, colorectal polyp, colorectal adenoma and adenomatous polyps. We also utilised a previous review of the data (Samad et al, 2005
; Lee and Oguma, 2006
; World Cancer Research Fund/American Institute for Cancer Research, 2007a
) and manual searches of the reference lists of identified manuscripts. We included recurrent, incident and prevalent cases of colon polyps. We did not limit studies by type of physical activity or study sample demographics.
Our search yielded 89 potential articles. We excluded reviews, non-human studies, editorials/comments/letters to the editor, studies without colon polyps as an outcome, studies where physical activity was only included as a covariate (and thus no measure of association was presented), and where no metric for effect estimate precision (P-value, s.e., confidence interval (CI)) was provided. Combined with searches from the reference sections of manuscripts and previous reviews, this yielded 20 manuscripts. From each manuscript, we abstracted the sample size (including number of cases), gender, years of follow-up or type of control sample, case definition, physical activity domain, adenoma detection method, sample definition criteria and results. We also abstracted the variables that each study used in its most adjusted analysis. Data extraction was performed by a single investigator (KYW). Where studies included more than one type of physical activity without a summary measure, we included only leisure time physical activity, which is the major modifiable component of energy.
Previous meta-analyses have suggested that results for adenomatous polyps need to be presented separately from hyperplastic or malignant polyps. (Botteri et al, 2008
) Although we did not restrict our analysis to studies where data was limited to adenomatous polyps, we did consider those results separately. Specifically, we excluded results for hyperplastic polyps where feasible. We also identified studies considered to be the ‘best approach' using criteria similar to those used in a previous meta-analysis (Botteri et al, 2008
), namely, studies that met all of the following: (1) limited the outcome to only adenomatous polyps; (2) all individuals received a full colonoscopy; and (3) the study population excluded anyone with inherited colorectal cancer syndromes, inflammatory bowel disease, a history of colon polyps or cancer, or a previous colon resection.
Meta-analysis of random effects was used to allow for the heterogeneity of results across studies. (Mosteller and Colditz, 1996
) Data were processed in SAS, and the analyses were performed using R-package ‘meta' (SAS Institute Inc., Cary, NC, USA). A summary forest plot was generated in Stata (StataCorp LP, College Station, TX, USA). As most studies reported RRs or odds ratios (ORs) and their associated 95 percent CIs, we used these data as summary statistics for each study. First, we derived the s.e. of log (RR or OR) using the 95 percent CI, with the expression: (log (upper limit) – log (lower limit))/2*
1.96. These s.es were used as weights for summary effect estimates in the meta-analysis. We visually examined publication bias using Funnel plots, and employed the rank correlation method to formally test for bias. (Begg and Mazumdar, 1994
) Where studies reported results separately for men and women, we included both estimates when reporting the overall association. To evaluate the potential effects of limiting results to only adenomatous polyps, we conducted exploratory analysis in the subset of those studies. We also included results separately for large/advanced adenoma, if the data were presented as such in the original manuscript. We also conducted exploratory analyses limited to those studies defined as the ‘best approach'. To test sub-analysis differences (large vs
all adenomas; best approach vs
all studies), we partitioned ‘total heterogeneity' into between-group and within-group heterogeneity, and used the ‘between-group' heterogeneity index as the test statistic against χ2
distribution with 1 degree of freedom. (Cooper and Hedges, 1994