Calcium intake and colorectal cancer risk: Dose-response meta-analysis of prospective observational studies

International Journal of CancerVolume 135, Issue 8 p. 1940-1948 EpidemiologyFree Access Calcium intake and colorectal cancer risk: Dose–response meta-analysis of prospective observational studies NaNa Keum, Corresponding Author NaNa Keum Department of Nutrition, Harvard School of Public Health, Boston, MA Department of Epidemiology, Harvard School of Public Health, Boston, MACorrespondence to: NaNa Keum, Departments of Nutrition and Epidemiology, Harvard School of Public Health, Building 2, 3rd Floor, 655 Huntington Avenue, Boston, MA 02115, USA, Tel.: 617-432-4648, Fax: 617-432-2435, E-mail: nak212@mail.harvard.eduSearch for more papers by this authorDagfinn Aune, Dagfinn Aune Department of Public Health and General Practice, Faculty of Medicine, Norwegian University of Science and Technology, Trondheim, Norway Department of Epidemiology and Biostatistics, Imperial College London, London, United KingdomSearch for more papers by this authorDarren C. Greenwood, Darren C. Greenwood Division of Biostatistics, University of Leeds, Leeds, United KingdomSearch for more papers by this authorWoong Ju, Woong Ju Department of Obstetrics and Gynecology, Ewha Womans University, Seoul, Republic of KoreaSearch for more papers by this authorEdward L. Giovannucci, Edward L. Giovannucci Department of Nutrition, Harvard School of Public Health, Boston, MA Department of Epidemiology, Harvard School of Public Health, Boston, MA Channing Division of Network Medicine, Brigham and Women's Hospital, Boston, MASearch for more papers by this author NaNa Keum, Corresponding Author NaNa Keum Department of Nutrition, Harvard School of Public Health, Boston, MA Department of Epidemiology, Harvard School of Public Health, Boston, MACorrespondence to: NaNa Keum, Departments of Nutrition and Epidemiology, Harvard School of Public Health, Building 2, 3rd Floor, 655 Huntington Avenue, Boston, MA 02115, USA, Tel.: 617-432-4648, Fax: 617-432-2435, E-mail: nak212@mail.harvard.eduSearch for more papers by this authorDagfinn Aune, Dagfinn Aune Department of Public Health and General Practice, Faculty of Medicine, Norwegian University of Science and Technology, Trondheim, Norway Department of Epidemiology and Biostatistics, Imperial College London, London, United KingdomSearch for more papers by this authorDarren C. Greenwood, Darren C. Greenwood Division of Biostatistics, University of Leeds, Leeds, United KingdomSearch for more papers by this authorWoong Ju, Woong Ju Department of Obstetrics and Gynecology, Ewha Womans University, Seoul, Republic of KoreaSearch for more papers by this authorEdward L. Giovannucci, Edward L. Giovannucci Department of Nutrition, Harvard School of Public Health, Boston, MA Department of Epidemiology, Harvard School of Public Health, Boston, MA Channing Division of Network Medicine, Brigham and Women's Hospital, Boston, MASearch for more papers by this author First published: 13 March 2014 https://doi.org/10.1002/ijc.28840Citations: 88AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinked InRedditWechat Abstract Mechanistic and epidemiologic studies provide considerable evidence for a protective association between calcium intake and incident colorectal cancer (CRC). While the relationship has not been substantiated by short-duration randomized controlled trials (RCTs) of CRC, trials do show a benefit on adenomas, a precursor to CRC. To address some of this inconsistency, we conducted dose–response meta-analyses by sources of calcium intake, based on prospective observational studies published up to December 2013 identified from PubMed, Embase, and BIOSIS. Summary relative risks (RRs) and 95% confidence intervals (CIs) were calculated using a random-effects model. For total calcium intake, each 300 mg/day increase was associated with an approximately 8% reduced risk of CRC (summary RR = 0.92, 95% CI = 0.89–0.95, I2 = 47%, 15 studies with 12,305 cases, intake = 250–1,900 mg/day, follow-up = 3.3–16 years). While the risk decreased less steeply in higher range of total calcium intake (Pnon-linearity = 0.04), the degree of curvature was mild and statistical significance of non-linearity was sensitive to one study. For supplementary calcium, each 300 mg/day increase was associated with an approximately 9% reduced risk of CRC (summary RR = 0.91, 95% CI = 0.86–0.98, I2 = 67%, six studies with 8,839 cases, intake = 0–1,150 mg/day, follow-up = 5–10 years). The test for non-linearity was not statistically significant (Pnon-linearity = 0.11). In conclusion, both dietary and supplementary calcium intake may continue to decrease CRC risk beyond 1,000 mg/day. Calcium supplements and non-dairy products fortified with calcium may serve as additional targets in the prevention of CRC. RCTs of calcium supplements with at least 10 years of follow-up are warranted to confirm a benefit of calcium supplements on CRC risk. Abstract What's New? Evidence suggests that milk intake may decrease the risk of colorectal cancer (CRC). In this meta-analysis, the authors found that calcium is responsible for this protective effect, and that the dose-response curve is virtually linear within the observed range. They also compared dietary with supplemental calcium intake, and found that both sources provide similar benefits. These results suggest that even people who are lactose intolerant or consume low levels of dairy products may still reduce their risk of CRC by taking calcium in the form of supplements and non-dairy foods. Considerable evidence suggests that milk intake may decrease the risk of CRC,1, 2 which is the third most commonly diagnosed cancer and the fourth leading cause of cancer death worldwide.3 Possible protective nutrients in milk include calcium,4-7 vitamin D if milk is fortified,8, 9 and some fat components such as conjugated linoleic acid10 and butyric acid.11 Due to the high calcium content in milk, calcium is thought to be the major nutrient that mediates the beneficial effect of milk on CRC. The involvement of calcium in the etiology of CRC is supported by several biological mechanisms. Garland and Garland et al. first hypothesized that intracelluar calcium in the colonic epithelial cells may reduce the cancer-promoting inflammatory response to bacterial flora and other agents in the colonic lumen.12 Experimental studies in animals and humans suggest that calcium may bind secondary bile acids or ionized fatty acids in the colorectal lumen, diminishing their carcinogenic effects on the colorectal mucosa.4, 5 Alternatively, evidence from in vivo and in vitro human colonic epithelial cells suggests that calcium may reduce cell proliferation and promote cell differentiation by modulating cell signaling.6, 7 In a RCT, calcium supplementation of 2,000 mg/day induced favorable changes on gene expression in the APC/β-catenin pathway in the normal mucosa of colorectal adenoma patients.13 Perturbations of this pathway is a common early event in colorectal carcinogenesis. Despite such biological plausibility, epidemiologic studies and RCTs have found inconsistent results. While a meta-analysis14 and a pooled analysis2 of observational studies found a statistically significant protective association, a recent meta-analysis of eight RCTs showed that assignment to calcium supplements without co-administered vitamin D did not statistically significantly alter the CRC risk.15 Due to the lack of strong evidence from calcium supplement trials, in The Colorectal Cancer 2011 Report from the World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR), calcium was classified as a probable protective factor of CRC rather than a convincing one.16 Several explanations could account for the apparent discrepancy. Possibly, calcium is not the causal factor, and thus it is important to consider if another nutrient in milk, particularly vitamin D, may account for the association. Also important to determine from the observational studies is whether calcium from supplements is similarly associated with lower risk of CRC as calcium from foods, particularly given that all the trials tested only supplementary calcium, and dietary and supplementary calcium might differ in bioavailability.17 Alternatively, the lack of a statistically significant finding in the meta-analysis of trials might have resulted from several methodological limitations of trials such as short follow-up period and inadequately addressed dose–response relationship. To address some of these issues, we conducted a dose–response meta-analysis of prospective observational studies that aims to characterize the shape of the relationship between calcium intake and CRC risk, examined the association by dietary and supplementary sources of calcium, and carefully considered the potential role for confounding, particularly by dietary vitamin D. Material and Methods The Meta-analysis Of Observational Studies in Epidemiology (MOOSE) checklist was followed for the design, analysis, and reporting of this meta-analysis.18 Two authors (N.K. and W.J.) participated in literature search, study selection, and data extraction independently. Inconsistency between researchers was resolved through discussion. Literature search A literature search was performed based on PubMed, Embase, and BIOSIS databases for studies published up to December 2013. Detailed search terms are provided in the online Supporting Information method. The language was limited to English and no other restrictions were imposed. Abstracts and unpublished results were not included. The reference lists of selected systematic reviews and meta-analyses, and all the articles included in our analysis were also reviewed for additional articles. Study selection To be included, studies had to be a prospective observational study (cohort studies analyzed with nested case-control, case-cohort, or prospective cohort approaches) investigating the relationship between calcium intake and incident cancer of the colon or rectum; providing a dose of calcium intake, RR (hazard ratio or risk ratio), 95% confidence interval (CI), category-specific or total number of cases, and category-specific or total number of either person-years or noncases for at least three categories. Retrospective studies were excluded to minimize recall and selection bias. When there were several publications from the same cohort, the publication with the largest number of cases was selected. This process of study selection and reasons for exclusion are summarized in Figure 1. Authors of three publications2, 19, 20 were contacted for additional information (e.g. category-specific cases, category-specific person-years) and they all provided the requested data. A total of 21 publications were included to extract data for 20 prospective observational studies (Supplementary table 1). Figure 1Open in figure viewerPowerPoint Flowchart of study selection. Data extraction From each study, the following information was extracted: category-specific dose of calcium intake (range, mean, or median), the most fully adjusted RRs and their 95% CIs, the first author's last name, publication year, study design, study name, country of the study cohort, exclusion criteria, sex, age at baseline, sample size, number of cases, follow-up period, number of person-years, types of calcium intake (total = dietary + supplementary, dietary, supplementary), types of CRC (CRC, colon cancer (CC), rectal cancer (RC)), dietary assessment method (type, whether it had been validated), number of dietary measurements (baseline only, updated), outcome ascertainment method, variables adjusted for, variables reported to be not adjusted for based on the statistical criteria, reverse causation (whether it was addressed, whether sensitivity analysis excluding cases during the first few years follow-up changed the results substantially). Statistical analysis For the linear dose–response analysis assuming a linear relationship between calcium intake and CRC risk, the method described by Greenland and Longnecker21 was used to calculate study-specific RRs (linear slopes) and 95% CIs from the natural logs of extracted RRs and 95% CIs across categories of calcium intake. Study-specific RRs and variance/covariance matrices were pooled using a random effects model to calculate the summary RR and 95% CI. This method for linear dose–response analysis requires that the distribution of cases, the distribution of person-years or noncases, and RRs with the variance estimates are known for at least three quantitative exposure categories. When studies did not provide distributions of cases or person-years, authors were contacted or approximations were made if possible (e.g., if studies analyzed calcium intake by quartiles, the category-specific number of person-years was estimated by dividing the total number of person-years by 4). For studies that showed results separately for colon and rectal cancer or for men and women, category-specific RRs and variances were combined using a fixed effects model with inverse variance weight to obtain combined estimates for CRC or for both sexes, before calculating study-specific RRs and CIs. In each study, the mean or median value of calcium intake in each category was assigned to the corresponding RR. When the lowest or highest categories were open-ended, the length of the open-ended interval was assumed to be the same as that of the adjacent interval. Calcium intakes reported in densities (mg/kcal) were converted into absolute intakes (mg/day) using the mean or median daily energy intake of the study population. Forest plots of the linear dose–response results are presented for RRs for a 300 mg daily increment of calcium intake (equivalent to calcium content in one serving (250 mL) of milk). To examine a potential non-linear relationship between calcium intake and CRC risk, fractional polynomial models were used.22 The best-fitting second order fractional polynomial model was determined as the one with the lowest deviance. The linear and non-linear models were compared using a likelihood test to test for non-linearity. Heterogeneity in the relationship between calcium intake and CRC across studies was assessed by Q test and quantified by I2.23 To identify sources of heterogeneity, subgroup analyses and meta-regression were conducted based on a priori selected variables related to etiologic heterogeneity and potential effect modifiers. To assess study quality, subgroup analyses and meta-regression were conducted by variables concerning methodological characteristics. Potential for small study effects, such as publication bias, was assessed using Egger's test24 and Begg's test.25 Diverse sensitivity analyses including the influence analysis were performed to explore robustness of the results. For statistical significance, two-sided α was set at p = 0.05. All statistical analyses were conducted using STATA 12 (StataCorp, College Station, TX). Results Overall calcium intake: Primary analysis While total calcium intake is the exposure measure from which the best dose–response relationship between calcium intake and CRC risk could be inferred, not all studies have information on supplementary calcium intake, limiting their investigation only to the effect of dietary calcium intake. Since total calcium intake is reasonably approximated by dietary calcium intake if the prevalence of calcium supplement use is low (<5%) in the study cohort, our primary analyses included studies that reported total calcium intake and studies that only examined dietary calcium intake but their cohorts have a low prevalence of calcium supplement use. Fifteen cohort studies19, 26-39 were included in the primary dose–response meta-analysis of calcium intake (10 total, 5 dietary) and CRC risk (12 CRC, 3 CC), providing a total of 12,305 cases among 1,415,597 participants with mean total calcium intake ranging approximately between 250 and 1,900 mg/day. From the linear dose–response analysis, the summary RR for a 300 mg/day increase in total calcium intake was 0.92 (95% CI = 0.89–0.95), with moderate heterogeneity (I2 = 47%, Pheterogeneity = 0.02) (Fig. 2a). Small study effects, such as publication bias, were not evident (PEgger = 0.34, PBegg = 0.73). In sensitivity analyses omitting one study at a time, the results were not influenced greatly by any of the studies. In other sensitivity analyses such as excluding studies19, 28 that used CC as the outcome and replacing three cohort studies19, 26, 39 included in the pooling project2 with the pooling project itself, the results did not change materially (data not shown). Figure 2Open in figure viewerPowerPoint Linear dose–response meta analyses on CRC risk associated with a 300 mg/day increase in (a) total calcium intake, (b) total calcium intake in populations with an appreciable use of calcium supplements, (c) total calcium intake in populations with a low (<5%) use of calcium supplements, (d) supplementary calcium intake. RR = Relative Risk for a 300 mg/day increase in calcium intake of each type; CI = confidence interval. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] Slight non-linearity was apparent, with CRC risk decreasing more steeply in lower range of total calcium intake than in higher range (Pnon-linearity = 0.04) (Fig. 3). Albeit statistically significant, the degree of curvature was mild. Compared to 250 mg/day of total calcium intake, the summary RR was 0.82 (95% CI = 0.71–0.95) at 1000 mg/day and further reduced to 0.74 (95% CI = 0.65–0.85) at 1,750 mg/day. However, the scatter plot suggested that the decreasing rate of reduction in CRC risk with increasing total calcium intake may be strongly driven by the study by Jarvinen et al.,35 which reported a statistically non-significantly increased risk of CRC at the highest total calcium intake observed. When excluding this study, the dose–response curve showed no evidence of non-linearity (Pnon-linearity = 0.08). Figure 3Open in figure viewerPowerPoint Non-linear dose–response meta-analysis on total calcium intake and CRC risk (Pnon-linearity = 0.04; reference: 250 mg/day) RR = relative risk compared to 250 mg/day of total calcium intake. As a sensitivity analysis aiming to derive the calcium and CRC relationship from the most inclusive data, we repeated the same linear and non-linear dose–response meta-analyses in 20 studies19, 20, 26-43 that reported total (n = 10) or dietary (n = 10) calcium intake. Results did not change materially (data not shown). Overall calcium intake: Secondary analysis In order to further investigate if the dose–response relationship between total calcium intake and CRC risk differs by supplementary calcium use, the primary dataset including 15 studies (10 total, 5 dietary)19, 26-39 was divided into the following two subsets. Subset representing populations with an appreciable use of calcium supplements Ten cohort studies19, 26-33, 39 that reported total calcium intake including supplementary calcium were included (10,601 cases, 1,186,722 participants, range = 250–1,700 mg/day). In the linear dose–response analysis, the summary RR for a 300 mg/day increase was 0.93 (95% CI = 0.89–0.96) with moderate heterogeneity (I2 = 53%, Pheterogeneity = 0.02) (Fig. 2b). Small study effects such as publication bias were not evident (PEgger = 0.44, PBegg = 0.92). In sensitivity analyses such as omitting one study at a time, excluding studies19, 28 that used CC as the outcome, and replacing three cohort studies19, 26, 39 included in the pooling project2 with the pooling project itself, the results did not change materially (data not shown). The test for non-linearity was not statistically significant (Pnon-linearity = 0.11). Subset representing populations with a low use of calcium supplements The remaining five studies34-38 reporting dietary calcium intake were included. Of note, three studies19, 33 that reported total calcium intake also provided estimates for the effect of dietary calcium intake on CRC risk among non-supplement users. Thus, a total of eight studies19, 33-38 were included in this subgroup (3,770 cases, 451,568 participants, range of intake = 250–1,900 mg/day). In the linear dose–response analysis, the summary RR for a 300 mg/day increase was 0.90 (95% CI = 0.85–0.96, I2 = 27%, Pheterogeneity = 0.21) (Fig. 2c). Small study effects such as publication bias were not evident (PEgger = 0.98, PBegg = 1.00). In sensitivity analyses such as omitting one study at a time and excluding studies19 that used CC as the outcome, the results did not change materially (data not shown). The test for non-linearity was not statistically significant (Pnon-linearity = 0.07). Supplementary calcium intake: Primary analysis The presence of a relatively high number of calcium supplement users in the U.S. provided an opportunity to investigate the dose–response association of supplementary calcium intake with CRC risk. Six cohort studies26, 27, 30-33 from the U.S. were included (five CRC, one CC), providing a total of 8,839 cases among 920,837 participants with mean supplementary calcium intake ranging approximately between 0 and 1,150 mg/day. In the linear dose–response analysis, the summary RR for a 300 mg/day increase was 0.91 (95% CI = 0.86–0.98), with moderate heterogeneity (I2 = 67%, Pheterogeneity =0.01) (Fig. 2d). Small study effects such as publication bias were not evident (PEgger = 0.43, PBegg = 0.85). In sensitivity analyses omitting one study at a time, the results were not influenced greatly by any of the studies. The test for non-linearity was not statistically significant (Pnon-linearity = 0.11). Supplementary calcium intake: Secondary analysis To examine indirectly if supplementary and dietary calcium form differential relationships with CRC risk, six studies26, 27, 30-33 were identified that provided results in the entire population regarding both total and dietary calcium intake (Table 1). Adding supplementary calcium intake on top of background dietary calcium intake did not change the linear dose–response relationship, as the summary RR for a 300 mg/day increase was similar between total calcium (0.93, 95% CI = 0.89–0.97, I2 = 62%, Pheterogeneity = 0.02, range of intake = 350–1,700 mg/day) and dietary calcium (0.92, 95% CI = 0.90–0.95, I2 = 0%, Pheterogeneity = 0.85, range of intake = 300–1,200 mg/day). Both estimates were free of small study effects, such as publication bias, and robust to the influence of any single study (data not shown). Subgroup analyses While extensive subgroup analyses were done for total calcium intake, only sex-specific subgroup analyses were conducted for supplementary calcium intake due to the limited number of studies available (Supplementary table 2). In investigating potential etiologic heterogeneity between CC and RC, there was no evidence of heterogeneity between subsites, but statistical significance of an inverse association was limited to CC. In exploring heterogeneity by potential effect modifiers and by methodological characteristics, there was no evidence of between-subgroup heterogeneity. A statistically significant inverse association was observed in most subgroups. In some subgroups with a statistically marginally-significant inverse association, exclusion of the study by Jarvinen et al.35 improved statistical significance and reduced within-subgroup heterogeneity. Discussion Our linear dose–response analysis supports an approximately 8% decreased risk of CRC associated with a 300 mg/day increase in calcium intake, which holds for both dietary and supplementary calcium, for both men and women, and more consistently for CC than for RC. While there was statistically significant evidence for a non-linear relationship with CRC risk decreasing less steeply at higher total calcium intake, the degree of curvature was mild suggesting that, overall, a linear association is a reasonable summary of the dose–response relationship within the observed calcium intake of 250 to 1,900 mg/day. The statistical significance of non-linearity was further discounted because the test for non-linearity was sensitive to the presence of the study by Jarvinen et al.,35 which was atypical in several respects; it was the only study without a validated dietary questionnaire, and it had the highest calcium intakes observed, the smallest number of cases, the youngest age range because the study recruited anyone aged 15 years or older, and the longest mean follow-up. Limitations of the study Combining many studies allowed us to robustly investigate calcium intake over a wide range, but the resulting heterogeneity was significant and subgroup analyses did not identify statically significant sources of between-subgroup heterogeneity. However, as suggested in the forest plots, heterogeneity was largely attributable to variations in the magnitude of RRs than to differences in the directionality of RRs. Furthermore, heterogeneity was lower in subgroups that adjusted for intake of red meat, dietary fiber, folate intake, screening, and NSAID use, when the analysis was restricted to CC, and when the analysis was stratified by gender. Excluding the study by Jarvinen et al.35 substantially reduced heterogeneity in several subgroups. While no evidence for small study effects such as publication bias was found in this analysis, we were not able to investigate potential etiologic heterogeneity between proximal and distal CC associated with calcium intake due to the selective reporting of results on subsite-specific CC. Measurement error in the assessment of calcium intake is of concern because it could have affected the magnitude and shape of the dose–response relationship between calcium intake and CRC risk. Meta-analysis is inevitably prone to any measurement error in the studies included. While most of the included studies used validated dietary questionnaires whose ability to assess relative calcium intake was tested to be reasonable, such questionnaires do not necessarily guarantee the validity in terms of measuring absolute calcium intake, on which the dose–response meta-analysis relied. Thus, measurement errors in assessing absolute calcium intake within each study compromise the validity of our quantitative findings. Further measurement error is introduced in the procedure of meta-analysis, as the investigation of the dose–response relationship based on results provided for a categorical calcium intake necessitates some assumptions. For instance, assigning mean or medium calcium intake of a category to the corresponding RR, assigning the length of the adjacent interval to an open-ended lowest or highest category, and converting calcium intake reported in density to absolute intake using the mean or median energy intake of the study population, are all potential sources of measurement error. While some robustness to measurement error is suggested by the finding that total, dietary, and supplementary calcium intake, despite having possibly different measurement errors, consistently showed an inverse linear association, we cannot completely ruled out differential effects of different measurement errors on different true RRs producing an appearance of robustness. Altogether, measurement errors from diverse sources are inevitable, and while their direction of bias cannot be predicted, they are generally anticipated to attenuate the true effect.44 Thus, our quantification of a linear association as an approximately 8% reduced risk of CRC with a 300 mg/day increase in calcium intake may be an underestimation of the true calcium effect. Strengths of the study First, this analysis is based on strong biological plausibility as explained in the introduction and our findings are also supported by RCTs with adenoma endpoint, a precursor to CRC. By including 15 prospective studies, we had adequate statistical power to assess the shape of dose–response relationship over a wide range and for diverse sources of calcium intake. Since we included only prospective studies, our findings are less likely to be explained by recall and selection bias. While unmeasured or residual confounding and reverse causation are of concern, our extensive subgroup analyses showed that none of the methodological aspects was a statistically significant source of heterogeneity. Furthermore, a statistically significant association persisted in most strata that adjusted for confounders and addressed reverse causation. Since most studies had a long duration of follow-up, our meta-analysis was advantageous in mitigating the influence of reverse causation and accounting for a potential long induction period relating calcium to CRC risk. Findings from our meta-analyses help address the critical confounding by dietary vitamin D. Considering that vitamin D has been suggested to protect against CRC8, 9 and fortified milk is the major common source of both dietary calcium and dietary vitamin D in many countries, confounding by dietary vitamin D complicates the distinction of an effect of calcium itself from that of dietary vitamin D in epidemiologic studies. Our subgroup analyses provide three lines of evidence against confounding by dietary vitamin D. A linear inverse association between calcium intake and CRC risk remained statistically significant, first in the subgroup of studies that controlled for dietary vitamin D intake or stated no confounding from dietary