We identified studies by a systematic literature search of MEDLINE electronic database (1966 through July 2013). We ran two queries, one aimed at RCTs and one aimed at observational studies (case-control or cohort). Query I was: [“Randomized Controlled Trial” (ptyp)] AND (“HMG-CoA reductase inhibitor” OR “HMG-CoA reductase inhibitors” OR “HMG-CoA reductase inhibitor” OR “HMG-CoA reductase inhibitors” OR statin OR statins OR atorvastatin OR cerivastatin OR fluvastatin OR lovastatin OR mevastatin OR pravastatin OR rivastatin OR rosuvastatin OR simvastatin). Query II was: (“HMG-CoA reductase inhibitor” OR “HMG-CoA reductase inhibitors” OR statin OR statins OR atorvastatin OR cerivastatin OR fluvastatin OR lovastatin OR mevastatin OR pravastatin OR rivastatin OR rosuvastatin OR simvastatin OR pitavastatin) AND (cancer OR cancers OR neoplasm OR neoplasms OR malignancy OR malignancies).

In addition, we browsed the reference lists of relevant narrative and systematic reviews, and asked a knowledgeable expert to identify any additional studies. We browsed the title and abstract of all identified studies to exclude any that were clearly irrelevant. The full text of the remaining articles was read to determine whether it contained information on the topic of interest.

Eligible studies for this meta-analysis were either RCTs or observational studies (case-control or cohort) evaluating any exposure to statins and the risk of colorectal cancer. In order to be included in the meta-analysis, studies had to report an estimated measure of effect size (risk ratio, rate ratio, hazard ratio or odds ratio) and its associated CI, or had to provide enough data to calculate such an effect measure and CI. In cases of multiple publications from the same population, only data from the most recent report were included.

In particular, RCTs were considered eligible if they (1) evaluated a statin therapy compared with placebo or no treatment; (2) had no other intervention difference between the experimental and the control group; (3) enrolled at least 2000 participants; (4) had a minimum duration of 2 years; and (5) reported the incidence of colorectal cancer in both arms during the trial period. The fourth criterion was used because the effects of the intervention may require long-term exposure. In addition, as colorectal cancer is a rare disease, RCTs of short duration are unlikely to register any significant number of cases.

We did not assess the methodological quality of the primary studies, since quality scoring in meta-analyses of observational studies is controversial, as it is for RCTs[28,29]. Instead, we performed subgroup and sensitivity analyses, as is recommended.

Two reviewers (Lytras T, Bonovas S) abstracted the data independently. The following information was collected from each study: (1) citation data, first author’s last name, year of publication, and country of the population studied; (2) study design; (3) number of subjects; (4) relative risk (RR) and 95%CI; (5) for RCTs, the number of events (colorectal cancer cases) in the statin and control groups; (6) definition of statin exposure; and (7) control for confounding factors by matching or adjustments, if applicable.

Risk ratios and 95%CI were calculated for each RCT by reconstructing contingency tables based on the number of subjects randomly assigned and the number of subjects with incident colorectal cancer (analysis in accordance with the intention-to-treat principle). In observational studies, we extracted the RR estimates that reflected the greatest degree of control for potential confounders. Differences in data extraction were resolved by consensus, referring back to the original article.

We included in this meta-analysis studies reporting different measures of relative risk: RCTs (risk ratio), cohort studies (rate ratio, hazard ratio), and case-control studies (odds ratio). In practice, these measures of effect yield very similar RR estimates, since the absolute risk of colorectal cancer is very low[30].

Studies were grouped on the basis of study design, and three separate meta-analyses were conducted: one each for RCTs, cohort studies and case-control studies, using both fixed-effects and random-effects models. This was done to examine consistency of results across varying study designs with different potential biases. We also compared the summary RR estimates derived from the three separate meta-analyses with a test of interaction[31].

Each meta-analysis was performed twice, assuming either a fixed-effects or a random-effects model. In the absence of heterogeneity, the fixed-effects and the random-effects models provide similar results. When heterogeneity is found, the random-effects model is considered to be more appropriate, though both models may be biased[32].

For all statistical analyses we used the R software environment[33], version 3.0.1, and the “meta” package for R[34], version 2.3-0. For RCTs we used the function “metabin” to perform meta-analysis of binary outcome data, using as input the number of colorectal cases per group in each study; the Mantel-Haenszel method[35] was used to calculate pooled estimates, and the DerSimonian-Laird method[36] to estimate between-study variance in the random-effects model. For cohort and case-control studies we used the function “metagen”, inputting the log RR and its standard error (SE) for each study; the inverse variance weighting method was used to calculate pooled estimates, and the DerSimonian-Laird method[36] to estimate between-studies variance in the random-effects model. The standard error of the log RR was calculated from the upper (U) and lower (L) limits of the 95%CI published in each study, using the formula SE = (logU-logL)/(2 × 1.96).

Selective outcome reporting or publication bias was assessed using the Begg and Mazumdar adjusted rank correlation test[37] and the Egger regression asymmetry test[38]. To evaluate whether the results of the studies were homogeneous, we used the Cochran’s Q test with a 0.10 level of significance. We also calculated the I² statistic[39] that describes the percentage variation across studies that is due to heterogeneity rather than chance. Negative values of I² were put equal to zero, so that I² lies between 0% (i.e., no observed heterogeneity) and 100%. We regarded an I² value less than 40% as indicative of “not important heterogeneity” and a value higher than 75% as indicative of “considerable heterogeneity”[40].

All P-values are two-tailed. For all tests (except for heterogeneity), a probability level less than 0.05 was considered statistically significant. This work was performed according to the guidelines proposed by the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group[41], and the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) statement[42].

The results of our search process are presented on Figure 1. We identified and analyzed 40 independent studies that met the predefined inclusion criteria[4,5,24,25,39-74]. Eight out of 40 were RCTs of statins for cardiovascular outcomes[4,5,43-48], 13 were cohort studies[49-61], and 19 were case-control studies[25,26,62-78]. Sixteen studies had been included in the previous meta-analysis[4,5,25,45-48,60,61,72-78] while 24 were newly published[26,43,44,49-59,62-71], of which 2 RCTs[43,44], 11 cohort studies[49-59] and 11 case-control studies[26,62-71].

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Summary of evidence search and selection.

In total, the 40 studies involved more than 8.2 million subjects. The number of colorectal cancer cases ranged from 32 to 245 in the RCTs, from 76 to 6637 in the cohort studies, and from 56 to 36736 in the case-control studies. Three out of the 40 studies had been published solely in abstract form[73-75], and one was an academic dissertation[62]. One study from Japan[54] was a pooled analysis of two randomized trials and one cohort study, providing a single summary RR for statin use and colorectal cancer; we treated this as a single cohort study.

Seven out of eight RCTs were placebo-controlled[4,5,43-45,47,48], while one RCT was a non-blinded trial comparing statin treatment with a usual care control group[46]. All RCTs reported site-specific cancer outcomes, including colorectal cancer, as secondary endpoints. We were thus able to conduct a post hoc analysis of these trials and calculate risk ratios for colorectal cancer in an intention-to-treat analysis. All observational studies[24,25,47-76] evaluated exposure to statins and risk of colorectal cancer, and were controlled for various potential confounding factors by matching or adjustments. The publication dates of the studies included in the meta-analysis ranged between 1996 and 2013. Study designs, along with the RR estimates and 95%CI are shown in Table 1 for the RCTs, in Table 2 for the cohort studies, and in Table 3 for the case-control studies.

Eight large RCTs contributed to the analysis[4,5,43-48]. A total of 77994 individuals participated in these trials; 39002 in treatment groups and 38992 in control groups. The participants had a mean follow-up of approximately 5.0 years and a total experience of 390000 person-years. Five trials[4,43-45,47] reported a lower risk of colorectal cancer in the treatment group, while the other three trials[5,46,48] reported a higher risk (Table 1). None was statistically significant. The overall rate of colorectal cancer on all 8 RCTs was 0.83% in the statin group (325 incident cases) and 0.93% in the control group (361 incident cases). We found a modest but not statistically significant protective association ( about 10% risk reduction) of statin use against colorectal cancer, both under the assumption of a fixed-effects model (RR = 0.90, 95%CI: 0.78-1.04), or a random-effects model (RR = 0.89, 95%CI: 0.74-1.07) (Table 4). Figure 2 shows the forest plot of the RR estimates and 95%CI from the individual trials and the pooled results. The Cochran’s Q test had a P-value of 0.23 and the corresponding I² statistic was 25%, both indicating little variability between studies that cannot be explained by chance. The P-values for the Begg’s and the Egger’s tests were P = 0.22 and P = 0.31, respectively, both suggesting a low probability of selective outcome reporting or publication bias. Figure 3 shows the contribution of the randomized studies (represented by the blue points) in the overall funnel plot, while Figure 4 shows the incidence of colorectal cancer in the statin group against the incidence in the control group, across the 8 RCTs (L’Abbé plot)[79]. This plot demonstrates no substantial heterogeneity between studies, and no relation of statin effect to baseline risk of colorectal cancer.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Forest plot: results from individual studies and meta-analyses. The RR and 95%CI for each study are displayed on a logarithmic scale. Pooled estimates are from a random-effects model.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Funnel plot of observed relative risk against standard error (as a surrogate of study size) for all studies analyzed. RCT: Randomized controlled trials.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 L’Abbé plot of the incidence of colorectal cancer in the experimental (statin) group, against the incidence in the control group, across the analyzed randomized controlled trials (n = 8).

When the analysis was restricted to trials that evaluated statin therapy compared with placebo[4,5,43-45,47,48], the results did not substantially change (fixed-effects model: RR = 0.86, 95%CI: 0.74-1.01; random-effects model: RR = 0.85, 95%CI: 0.71-1.03; Cochran’s P = 0.30 and I² = 17%; Begg’s P = 0.18 and Egger’s P = 0.24). Similarly, after stratifying the data in two subgroups (lipophilic[5,45,47] vs lipophobic statins[4,43,44,46,48]), we did not find any statistically significant association between lipophilic or lipophobic statins and risk of colorectal cancer (Table 4). Last, to explore whether the results were dominated by a single study, we performed a “leave-one-out” sensitivity analysis, removing one study at a time (Figure 5A). This approach confirmed the stability of our results.

Open in New Tab   Full Size Figure   Download Figure

Figure 5 ‘‘Leave-one-out” sensitivity analysis for the three meta-analyses: pooled estimates are from random-effects models with one study omitted at a time. A: Randomized controlled trials, B: Cohort studies, C: Case-control studies.

Thirteen cohort studies[49-61] evaluating exposure to statins and colorectal cancer risk were included in the meta-analysis (Table 2, Figure 2). Approximately seven million patients participated in these studies, with the occurrence of 30019 colorectal cancer cases. Four cohort studies reported not an overall, but two or more “correlated” subgroup effect-estimates[49,53,55,58]; however, based on the available data, we were able to calculate study-specific crude RR estimates for these four studies for the purpose of our meta-analysis (Table 2).

Statin use was associated with a modest reduction in the risk of colorectal cancer, and this association reached statistical significance both under a fixed-effects model (RR = 0.96, 95%CI: 0.93-0.99) and under a random-effects model (RR = 0.91, 95%CI: 0.83-1.00) (Table 4, Figure 2). However, the Cochran’s Q test had a P-value lower than 0.001 and the corresponding I² statistic was 83%, indicating substantial heterogeneity between studies. In the sensitivity analysis (Figure 5B), omitting any single study did not lower the I² further than 78%. The P-values for the Begg’s and the Egger’s tests were P = 0.54 and P = 0.22, respectively, suggesting a low probability of publication bias.

Nineteen case-control studies[25,26,62-78] evaluated exposure to statins and colorectal cancer risk (Table 3). A total of 1.3 million patients participated in these studies, of which 100000 were colorectal cancer cases. Once again, statin use was associated with a similar modest reduction in the risk of colorectal cancer, which was statistically significant under both a fixed-effects model (RR = 0.93, 95%CI: 0.91-0.96) and a random-effects model (RR = 0.92, 95%CI: 0.87-0.98) (Table 4, Figure 2). We found substantial heterogeneity between studies; the Cochran’s Q test had a P-value lower than 0.001 and the corresponding I² statistic was 64%. In the “leave-one-out” sensitivity analysis, we identified the study by Vinogradova et al[26] as contributing most to the between-studies variability, but not to a crucial degree; excluding this study from the analysis lowered the I² to 50%. The P-values for the Begg’s and the Egger’s tests were P = 0.46 and P = 0.27, respectively, suggesting a low probability of publication bias.

When the analysis was limited to studies published in full-text, i.e., excluding those published solely in abstract form[73-75], the results did not appreciably change (Table 4). The association between statins and colorectal cancer risk remained statistically significant assuming either a fixed-effects model (RR = 0.94, 95%CI: 0.91-0.98), or a random-effects model (RR = 0.90, 95%CI: 0.83-0.99). The Cochran’s Q test had a P-value lower than 0.001, and the corresponding I² was 69%. The P-values for the Begg’s and the Egger’s tests were P = 0.47 and P = 0.19, respectively, but the funnel plot was slightly asymmetric, indicating a small likelihood of publication bias. Thus, selective publication of smaller case-control studies with statistically significant results might have occurred to some extent. It should be noted, however, that the result of our meta-analysis of case-control studies was fairly robust in the “leave-one-out” sensitivity analysis (Figure 5C); removal of any single study did not alter the statistical significance of the pooled estimate.

We compared pairwise the pooled RR estimates derived from the three separate analyses with a test of interaction[31]. We found no statistically significant differences between estimates, either between those assuming a fixed-effects model (RCTs vs cohort studies, z = 0.81, P = 0.42; RCTs vs case-control studies, z = 0.49, P = 0.62; cohort studies vs case-control studies, z = 1.23, P = 0.22) or those assuming a random-effects model (RCTs vs cohort studies, z = 0.18, P = 0.86; RCTs vs case-control studies, z = 0.36, P = 0.72; cohort studies vs case-control studies, z = 0.30, P = 0.76).

In addition, we performed a combined analysis of observational studies, i.e., cohort and case-control studies (Table 4). Statin use was again associated with a modest reduction in the risk of colorectal cancer, which was statistically significant assuming either a fixed-effects model (RR = 0.94, 95%CI: 0.92-0.96) or a random-effects model (RR = 0.92, 95%CI: 0.87-0.96; n = 32). The Cochran’s Q test had a P-value lower than 0.001 and the corresponding I² statistic was 75%, indicating substantial between-studies variability. The P-values for the Begg’s and the Egger’s tests were P = 0.36 and P = 0.16, respectively, both suggesting a very low probability of publication bias.

Combining all 40 studies for analysis yielded very similar results (Table 4). This is expected, as this particular analysis was dominated by the observational studies (36 studies; 8.1 million participants). These studies accounted for 92.3% and the 98.5% of the weight in the fixed- and the random-effects model, respectively.

Finally, we attempted to analyze the effect of statins separately in colon and rectal cancer. Six cohort studies (out of 13) and five case-control studies (out of 19) provided results by colorectal cancer subsite. In these 11 studies, statins did not appear to have an effect on colon cancer, assuming either a fixed-effects model (RR = 0.97, 95%CI: 0.93-1.02), or a random-effects model (RR = 0.95, 95%CI: 0.87-1.04). The Cochran’s Q test had a P-value of 0.02, and the corresponding I² was 54%. The p-values for the Begg’s and the Egger’s tests were P = 0.31 and P = 0.39, respectively. As regards rectal cancer, the fixed-effects model suggested no effect of statins (RR = 0.98, 95%CI: 0.91-1.05) but the random-effects model suggested a statistically significant effect (RR = 0.78, 95%CI: 0.62-0.97). The Cochran’s Q test had a P-value lower than 0.001 and the corresponding I² was 79%, indicating substantial heterogeneity. The P-values for the Begg’s and the Egger’s tests were P = 0.59 and P = 0.06, respectively, which highlights a significant potential for selective outcome reporting bias in this analysis.

Statins are some of the most widely prescribed drugs worldwide, as a result of their proven efficacy in the primary and secondary prevention of cardiovascular morbidity and mortality. Besides their main mechanism of action through the reduction of serum cholesterol, by means of competitive inhibition of hepatic 3-hydroxy-3-methylglutaryl-coenzyme A reductase, statins exert a variety of “pleiotropic” effects. It has been suggested that statins might play a role in cancer chemoprevention, and data from in vitro and animal model studies have been encouraging.

The relation between statins and colorectal cancer has been the focus of a growing body of epidemiological research, with often inconsistent results ranging from very protective to moderately harmful.

The authors sought to update their previous systematic review and meta-analysis to reflect the current totality of evidence on statins and colorectal cancer risk.

The authors believe it is not the end of the road, as has been suggested, for statins and colorectal cancer; rather a new approach is needed. One needs to focus more on basic research and pharmacogenomics, and perform epidemiological studies and clinical trials on high risk populations that might be more likely to benefit from statins, either as primary chemoprevention or as an adjuvant to treatment. In the meantime, the use of statins should remain restricted to the approved indications.

Cancer chemoprevention is an area of research that focuses on cancer prevention through pharmacological, biological, and nutritional interventions. Meta-analysis is a systematic and quantitative integration of the results of a set of independent studies. It allows for an objective appraisal of the epidemiological evidence, which may lead to resolution of uncertainty and disagreement.

The article is interesting and welcome. It is written by a team with experience in the meta-analysis and who has published scientific papers in the field of colorectal cancer. Choice trials were correctly made and motivated. Statistical analysis is laborious and thorough. The results are analyzed and discussed with competence. The risk reduction of colon cancer is properly justified, as the need for future studies. Figures are properly made, and the bibliography is correctly written.