Dairy Consumption and Risks of Colorectal Cancer Incidence and Mortality: A Meta-analysis of Prospective Cohort Studies

Colorectal cancer is the third most common cancer in men, ranking second in women worldwide, and more than 1.8 million new cases of colorectal cancer occurred in 2018 (1). Several studies have reported some evidence that lifestyle factors are important determinants for colorectal cancer risk (2). The relationship between dairy consumption and colorectal cancer risk has been extensively investigated, and it has been hypothesized that dairy products including calcium, vitamin D, milk lipids, and probiotics can prevent the occurrence of colorectal cancer (3–6). The World Cancer Research Fund International/American Institute for Cancer Research (WCRF) reported that consumption of dairy products probably protects against colorectal cancer (7).

Several epidemiologic studies and a few meta-analyses on dairy products and colorectal cancer incidence have been conducted (7–39), but we needed to conduct an updated meta-analysis of prospective cohort studies with the most recent data, rigorously stratified analyses by type of dairy products, and dose–response analyses. In addition, there are some prospective studies on dairy products and colorectal cancer mortality (40–46), but no meta-analysis to elucidate the long-term effects of dairy consumption on mortality from colorectal cancer has been conducted to date.

Thus, we systematically reviewed and performed a comprehensive meta-analysis of all prospective cohort studies to quantitatively assess the association between consumption of total dairy products, different types of dairy products, and colorectal cancer incidence, along with the investigation of colorectal cancer mortality according to dairy consumption.

We searched for eligible studies published up to July 2019 in PubMed and ISI Web of Science databases. The following search terms were used: “(milk, custard, yogurt, yoghurt, cheese, or dairy) combined with (colon, colorectal, rectal, rectum, neoplasm, cancer, carcinoma, or tumor).” We included prospective cohort studies to report the association between consumption of dairy products and colorectal cancer incidence or mortality, and included articles published in full-length and in English. For colorectal cancer mortality, we included studies that were performed in people who were cancer-free individuals. To be included in the analysis, the study should have had RR and 95% confidence intervals (CI) or sufficient data to calculate them. If there were several publications from the same cohort, we selected the publication with the largest number of cases.

Two investigators (S. Jin and Y. Kim) extracted data independently, according to the meta-analysis of observational studies in epidemiology guidelines (47), and any discrepancies were resolved by further discussion and referencing the original articles. The following data were extracted from each study: first author's last name, year of publication, geographic region, follow-up period or study period, baseline age, sample size, number of cases, type of dairy products, adjustment factors, and RRs and 95% CIs for association between dairy consumption and colorectal cancer incidence or mortality. If the study provided more than one RR from age-adjusted models to different multivariate models, we used the RR of the multivariate model to make the most complete adjustment for confounders.

The evaluation of the quality of the included original cohort studies was performed using the Newcastle-Ottawa Scale (NOS; ref. 48) for the following criteria: the selection of study population, comparability, and outcome assessment. We considered studies with a score of 9 or higher (of 13) to represent high quality.

The pooled RRs and 95% CIs of the highest versus lowest dairy intakes were calculated from the original studies. The natural logarithm of the RRs from each study was combined using the DerSimonian and Laird random-effects models, which incorporate both within-study and between-study variations (49). If the original study did not use the lowest category as a reference, we recalculated the RR and its 95% CI (35). When a study reported results of proximal and distal colon cancer, we first combined the two results using a fixed-effect model to obtain an overall estimate of colon cancer before combining with other studies (22). The summary estimates were presented as forest plots. We used Q statistic to assess statistical heterogeneity in the study (50) and inconsistency was quantified by the I² statistic (51). We conducted subgroup analyses stratified by sex, geographic region (Europe/United States/Asia), cancer site (colon/rectal), and duration of follow-up (<the median or ≥the median). As a way of assessing the quality of the studies, we examined whether the studies had adjusted for important covariates such as age, body mass index (BMI), smoking status, physical activity, socioeconomic status, and total energy intake, and conducted a stratified analysis by the adjustment factors.

To examine dose–response relationships across various categories of dairy intake, we used a generalized least-squares trend (GLST) estimation analysis based on the method developed by Greenland and Longnecker (52–54). We first estimated study-specific slopes from the correlated natural logarithm of RRs across intake categories in each study (9, 15–17, 19–22, 24, 26, 28, 30, 32, 34, 35) and then combined the GLST-estimated study-specific slopes with studies that reported slope estimates (23, 27) to derive an overall average slope. If the study reported the intake as serving per day or week, we converted the intake to gram of intake per day using standard units of 244 g for milk, 43 g for cheese, 244 g for fermented milk, and 177 g for dairy products based on serving sizes provided in the ‘U. S. Department of Agriculture Food and Nutrient Database for Dietary Studies’ (55). A total of 10, 13, eight, and seven studies were included in the dose–response analysis for total dairy (15–17, 19, 21, 23, 24, 27, 30, 34), milk (16, 19–24, 26, 28, 30, 32, 34, 35), cheese (9, 16, 20–23, 30, 35), and fermented milk consumption (9, 15, 16, 21, 22, 30, 35) and colorectal cancer incidence, respectively. Finally, we assessed publication bias through Egger test (56). A two-sided P < 0.05 was considered to be statistically significant. All statistical analyses were performed using Stata/SE version 14.0 Software (Stata Corporation).

We identified a total of 31 prospective studies (35 publications) including 24 studies (28 publications; Supplementary Fig. 1S) with 2,314,942 participants and 24,964 cases on association between dairy consumption and colorectal cancer incidence (8–35) and seven studies with 466,053 participants and 2,302 deaths on association between dairy consumption and colorectal cancer mortality (40–46). We have summarized the characteristics of the studies included in the meta-analysis in Tables 1 and 2. Regarding dairy consumption and colorectal cancer incidence, 14 studies provided RRs of total dairy (13, 15–17, 19, 21, 23–25, 27, 30, 31, 33, 34), 20 studies provided RRs of milk (8, 10, 12, 14, 16, 18–24, 26, 28–32, 34, 35), 11 studies provided RRs of cheese (9, 11, 14, 16, 20–23, 30, 31, 35), and 11 studies provided RRs of fermented milk (9, 11, 15, 16, 18, 21–23, 30, 31, 35). By geographic region, 12 studies were conducted in Europe (8–10, 15–17, 20, 22, 23, 28, 30, 32, 33, 35), 10 studies in the United States (11–14, 18, 19, 21, 24, 27, 29, 31, 34), and two studies in Asia (25, 26). The follow-up periods of 24 studies ranged from 5 to 31.1 years (mean, 12.8 years). The subjects were all older than 15 years at baseline. All of the studies adjusted for age (8–35), and most of the studies adjusted for BMI (kg/m²; refs. 9, 11, 12, 14–19, 21–25, 27–35), smoking (8, 11, 12, 14–16, 18–21, 23–25, 27–35), physical activity (11, 12, 14, 15, 18, 19, 21, 23–25, 27–31, 33–35), socioeconomic status (13, 15–17, 19, 22, 25–28, 30, 31, 33, 35), and total energy intake (9, 11, 13, 16, 17, 19, 21–27, 29, 30, 32–35). In terms of quality assessment, the studies included in the meta-analysis had a mean score of 10.7 out of a possible 13, and all of the studies had a score of more than 9, indicating high quality.

For colorectal cancer mortality, two studies provided RRs of total dairy products (42, 46), six studies provided RRs of milk (40, 41, 43–46), three studies provided RRs of cheese (40, 43, 44), and three studies provided RRs of fermented milk (43–45). By geographic region, three studies were conducted in the United States (40, 42, 46) and four studies in Japan (41, 43–45). The follow-up periods of seven studies ranged from 9.15 to 27 years (mean, 17.4 years). The subjects were all older than 18 years at baseline. All of the studies adjusted for age (40–46), and some studies adjusted for smoking (42–44, 46), BMI (44, 46), physical activity (44, 46), and total energy intake (42, 46). Regarding quality assessment, the studies had a mean score of 10.4 out of a possible 13. Six studies had a score more than 9, indicating high quality (40, 42, 46), and one study had a score of 8, indicating good quality (41).

Fourteen studies, including 1,686,419 participants, examined the association between total dairy consumption and colorectal cancer incidence. Comparing the highest versus lowest consumption categories, the pooled RR was 0.79 (95% CI, 0.74–0.85; Fig. 1A). The inverse association was similar for men (RR, 0.76; 95% CI, 0.65–0.88) and women (RR, 0.77; 95% CI, 0.71–0.83; Table 3). By cancer site, the pooled RRs for colon and rectal cancer were 0.77 (95% CI, 0.69–0.86) and 0.81 (95% CI, 0.67–0.98). By geographic region, the inverse association tended to be slightly stronger in Europe (RR, 0.75; 95% CI, 0.63–0.88) than in the United States (RR, 0.80; 95% CI, 0.75–0.86), while only one study from Asia showed no significant association (RR, 0.98; 95% CI, 0.82–1.17).

For milk consumption, 20 studies with 2,070,491 participants were included in the analysis, showing a pooled RR of 0.81 (95% CI, 0.76–0.86; Fig. 1B). By sex, the inverse association tended to be stronger in men (RR, 0.77; 95% CI, 0.70–0.85) than in women (RR, 0.88; 95% CI, 0.80–0.97; P_difference = 0.08). The inverse association was very similar for colon and rectal cancer, and did not vary by geographic region.

For cheese consumption, 11 studies including 1,025,510 participants, showed no significant association overall (RR, 0.95; 95% CI, 0.83–1.08; Fig. 1C). Stratifying by geographic region, the pooled RR from seven studies in Europe was 0.87 (95% CI, 0.78–0.97), while a nonsignificant positive association was shown in four studies from the United States (RR, 1.16; 95% CI, 0.93–1.46; P_difference = 0.04).

Similarly, fermented milk consumption, with 11 studies including 1,097,570 participants, showed no significant association overall (RR, 0.96; 95% CI, 0.87–1.05; Fig. 1D). By sex, the inverse association tended to be stronger for women (RR, 0.86; 95% CI, 0.78–0.95) than for men (RR, 0.97; 95% CI, 0.87–1.08). By geographic region, there was a significant inverse association in Europe (RR, 0.91; 95% CI, 0.85–0.98), while there was a nonsignificant positive association in the United States (RR, 1.11; 95% CI, 0.88–1.40).

In the stratified analysis by duration of follow-up, there was no significant difference for total dairy, milk, and cheese consumption. For fermented milk consumption, however, there was a significant difference in RRs (P_difference = 0.03), showing an inverse association in studies with long follow-up times (≥10 years; RR, 0.89; 95% CI, 0.82–0.96), but not in those with short follow-up times. For the adjustment factors, studies with strong adjustment tended to show slightly stronger inverse associations overall, but no significant difference was found.

In addition to the analysis of highest versus lowest dairy consumption, we also conducted a dose–response meta-analysis with available data from 13 cohort studies. The pooled RR for a 400 g/day increment of total dairy from 10 studies was 0.86 (95% CI, 0.81–0.90; Table 4). For a 200 g/day increment of milk consumption, the pooled RR from 13 studies was 0.94 (95% CI, 0.90–0.97). For cheese and fermented milk consumption, inverse associations were found in the analysis. For a 50 g/day increment of cheese consumption, the pooled RR from eight studies was 0.93 (95% CI, 0.89–0.97). For a 200 g/day increment of fermented milk consumption, the pooled RR from seven studies was 0.93 (95% CI, 0.88–0.99).

For total dairy consumption, the pooled RR of the highest versus lowest categories from two studies was 0.71 (95% CI, 0.54–0.93; Table 5). However, each type of dairy consumption did not show a significant association with colorectal cancer mortality. For milk consumption, the pooled RR from six studies was 0.98 (95% CI, 0.86–1.12). When stratified by geographic region, the pooled RR in studies from the United States was 0.73 (95% CI, 0.57–0.93), while Japanese studies had no association with milk consumption (pooled RR, 1.06; 95% CI, 0.95–1.18; P_difference = 0.02). For cheese consumption, the pooled RR from three studies was 1.18 (95% CI, 0.95–1.47). For fermented milk consumption, the pooled RR from three studies was 0.84 (95% CI, 0.63–1.12).

We found no evidence of publication bias for the meta-analysis of association between dairy products and colorectal cancer incidence (Egger P > 0.1 in all analyses) and colorectal cancer mortality (Egger P > 0.2 in all analyses).

This meta-analysis of 31 prospective cohort studies evaluated the potential associations between dairy consumption and colorectal cancer incidence or mortality. On the basis of the results of colorectal cancer incidence, we found significant inverse associations for the highest versus lowest categories of total dairy and milk consumption. For cheese and fermented milk consumption, there were weak inverse associations overall, while the associations tended to be stronger in studies conducted in Europe and those with strong adjustment. The results of dose–response meta-analysis support that dairy consumption is inversely associated with colorectal cancer incidence. For colorectal cancer mortality, we found lower death from colorectal cancer in people with high total dairy consumption, compared with those with low consumption, but the data were limited. Overall, in the meta-analyses, there was no evidence of heterogeneity among the studies.

A few meta-analyses have been conducted on association of colorectal cancer risk for the highest versus lowest intake of dairy products (36–38). A previous meta-analysis that showed a significant inverse association of total dairy intake indicated no association by cancer site (37), while our study showed significant associations by cancer site. In this meta-analysis, two more recent prospective studies were included (31, 33), and the two previously included studies were replaced with recent studies including the largest number of cases (30, 34), compared with the study by Aune and colleagues (37). In addition, our meta-analysis was stratified by sex and geographic region. Unlike the previous meta-analysis of milk consumption that showed a significant inverse association only with colon cancer (36), in our analysis, high milk consumption was inversely associated with both colon and rectal cancer risks. Another meta-analysis reported that high milk consumption was inversely associated with reduced risk of colorectal cancer only in men (38). However, in our meta-analysis, the inverse associations were found both in men and women. Compared with the study by Ralston and colleagues (38), our study included additional six prospective studies (28, 29, 31, 32, 34, 35), and the three previously included studies were replaced with recent studies including the largest number of cases (18, 22, 30). For cheese consumption, no significant associations were found in the previous meta-analyses, but our study showed a significant inverse association of cheese consumption with colorectal cancer risk in Europe. In this meta-analysis, we included additional four prospective studies (9, 22, 31, 35), and one study was replaced with the largest study (30). For fermented milk consumption, the previous meta-analysis had no significant association with colorectal cancer risk and had no association when stratified by sex (38), while our study showed that fermented milk consumption was associated with colorectal cancer in women, but not in men. In a similar fashion to the finding of cheese consumption, we found a significant inverse association for fermented milk consumption in Europe, but not in the United States. The results of our dose–response meta-analysis of total dairy and milk consumption were consistent with those from the WCRF Continuous Update Project (7), showing inverse associations of colorectal cancer risk per 400 g/day of total dairy consumption and 200 g/day of milk consumption. For cheese consumption, there was no significant association in the project, while we found 7% lower risk of colorectal cancer with 50 g/day of cheese consumption in the dose–response analysis. For fermented milk consumption, there was no dose–response meta-analysis conducted previously, and in our analysis, we found a significant inverse association of colorectal cancer risk with 200 g/day of fermented milk consumption.

Dairy consumption may reduce the risk of colorectal cancer through several potential mechanisms. The main anticancer ingredient contained in dairy products is considered to be calcium (3, 6). In the intestinal lumen, calcium can bind to intestinal fatty acids and secondary bile acids, thereby reducing their cytotoxicity as well as tumorigenic exposure to the epithelium (57), inhibiting colonic epithelial cell proliferation, and inducing terminal differentiation (58). Laboratory studies have shown that calcium induces apoptosis in colonic epithelial cells and alters K-ras gene mutations in the colon (4, 59). Epidemiologic studies have also shown that increased calcium intake can reduce the risk of colorectal cancer (60). Vitamin D is included in dairy products, and it can decrease p-P38 MAPK activation in lamina propria leukocytes and NF-κB activation in colonic epithelial cells, which prevent inflammation-related colon cancer by inhibiting inflammatory responses during early-stage carcinogenesis (61). In addition, lactoferrin and certain fatty acids, such as butyric acid, are associated with beneficial effects on colorectal cancer (3, 62). The study by Zanabria and colleagues has shown that milk fat globule film has an anti-colorectal cancer effect (63), and sphingolipids in milk can also inhibit colorectal cancer (64). Several mechanisms have proven that substances contained in dairy products can reduce the risk of colorectal cancer, but the consumption of dairy products also increased the circulating levels of insulin-like growth factor-1 (IGF-1; refs. 65, 66). The increased levels of IGF-1 could increase the risk of colorectal cancer, but the positive association was generally weak (65, 67).

There were strong inverse associations of colorectal cancer incidence with the highest versus lowest intakes of total dairy products and milk, but weak associations were found with cheese or fermented milk consumption. The most possible reason was that the result for total dairy consumption may be mainly driven by the intake of milk because milk was the largest proportion of total dairy consumption in most populations, and the highest categories of total dairy consumption (mean, 700 g/day; ∼4 servings/day) and milk consumption (mean, 433 g/day;∼2 servings/day) were higher than cheese (mean, 48 g/day; ∼1 serving/day) and fermented milk (mean, 189 g/day; ∼1 serving/day). Thus, calcium intake through total dairy products and milk was also higher than cheese and fermented milk. Lactose in milk may increase the bioavailability of calcium (68), but it is lost in cheese production, and β-galactosidase in fermented milk hydrolyzes lactose during storage and digestion (69). The previous studies showed that bacteria used for the production of yogurt can be effective in preventing the initiation of carcinogenesis (70, 71). When the data were analyzed separately by geographic region, a significant inverse association with high consumption of cheese and fermented milk was found only in studies from Europe, but not in the United States. The difference in RRs between Europe and the United States may be due to the difference in consumption, and the highest category of cheese consumption in Europe (mean, 59.82 g/day) was more than double from the United States (mean, 23.58 g/day). However, it should be noted that only four risk estimates were available to investigate this association in studies from the United States. For fermented milk consumption, our study showed a significant inverse association in women, but not in men. Although it is unknown why an association between fermented milk consumption and colorectal cancer risk exists among women but not among men, it could be related to relatively higher intake of fermented milk in women than men. The study by Murphy and colleagues, included in our meta-analysis, provided a detailed information on the amount of yogurt intake (equivalent to fermented milk in our study) according to quintiles of intakes, separately by men and women (30). Although total dairy intake was higher for all of the quintiles in men than women in the study, yoghurt consumption tended to be higher in women than men for all of the quintiles. Both cheese and fermented milk consumption showed relatively weak associations with colorectal cancer incidence in the main analyses, but when limited to studies with strong adjustment, we found significant inverse associations. In addition, fermented milk consumption showed an inverse association when limited to studies with long follow-up times.

For association between colorectal cancer mortality and dairy consumption, there was no previous study for the meta-analysis. We found 29% lower mortality from colorectal cancer in subjects with high total dairy consumption compared with those with low consumption. All of the studies included in the analysis of total dairy consumption were conducted in the United States. For milk consumption, there was no significant association for colorectal cancer mortality overall. However, when stratified by geographic region, the studies from the United States showed 27% lower mortality in subjects with high milk consumption. Japanese studies had no association for colorectal cancer mortality, showing a significant difference in RRs by geographic region. The possible reason for the different results is that the cutoffs for the highest milk intake categories varied by geographic region. The studies from Japan used “every day or daily” as a cutoff for the highest milk consumption, while those from the United States used “three or more servings per day” as a cutoff. For cheese and fermented milk consumption, we found no significant association for colorectal cancer mortality. Only a few studies were included in the analysis of cheese consumption, one from the United States and two from Japan, and in the analysis of fermented milk consumption, three from Japan.

This meta-analysis has several strengths. This is the comprehensive meta-analysis of prospective cohort studies to assess the association of dairy consumption with colorectal cancer incidence as well as mortality, and no evidence of publication bias was found. This study included the most recent prospective data and the largest number of participants and colorectal cancer cases. We conducted rigorously stratified analyses by type of dairy products, sex, cancer site, geographic region, and follow-up times. In addition, we investigated the linear association between dairy consumption and colorectal cancer incidence through a dose–response meta-analysis, although the number of studies was limited. For dairy consumption and colorectal cancer mortality, it is, to the best of our knowledge, the first meta-analysis to investigate the long-term effects of dairy consumption on mortality from colorectal cancer.

Despite these strengths, some limitations of this meta-analysis should be acknowledged. This meta-analysis was based on observational studies. Although we used multivariate RRs that reflected the greatest degree of adjustment for potential confounders, we were unable to completely solve the problem of residual or unmeasured confounding inherent in the observational studies. Moreover, the individual studies adjusted for different covariates. Especially, it is very important to control for total calorie intake, at the very least, as well as obesity to investigate the associations between dietary intake and cancer (72), but some of the studies did not adjust for the factors. We carried out the stratified analysis by adjustment for important confounders including BMI and total energy intake, which showed similar results or even stronger inverse associations when limited to studies with strong adjustment. For the measurement of dairy intake, many of the studies included in the meta-analysis assessed it at baseline only. The possible misclassification of dairy consumption categories would have probably led to the underestimation, rather than overestimation, of results, and thus the risk of colorectal cancer may be even stronger. We found a slightly stronger inverse association in studies with repeated measurements, but the data were limited. Another limitation is that the cutoffs for highest versus lowest consumption categories varied among the studies. As a way to compensate for this limitation, we performed a dose–response meta-analysis. We could not perform a meta-analysis by fat content in dairy products because of limited data. The two previous studies examined the association of colorectal cancer risk according to whole-fat, low-fat, and nonfat milk consumption, but the fat content of milk in the studies had little effect on the risk of colorectal cancer (30, 34). For association between dairy consumption and colorectal cancer incidence, most of the studies were conducted in Europe or in the United States. Only one study for milk consumption and one study for total dairy consumption were conducted in Asia. Because of the limited data, it was hard to investigate the association of dairy consumption with colorectal cancer incidence in Asian population. In addition, the consumption of dairy products in Asia is lower than that in Europe and the United States. The average consumption of dairy products is more than two servings per day in North America, Europe, and Australia, but less than one serving per day in Asia and Africa (73), and dietary patterns in the countries are also different (7). Therefore, our findings may not be generalizable to Asian population. Although we conducted the first meta-analysis of association between dairy consumption and colorectal cancer mortality, only a few studies were available for the analysis. In addition, there was no study conducted in Europe, while three studies were conducted in the United States and four studies in Japan.

In conclusion, the results of this comprehensive meta-analysis provide quantitative evidence that high dairy consumption may lower the risk of colorectal cancer incidence and mortality. The evidence from dairy consumption and colorectal cancer incidence is scarce in Asian population, and the available data from dairy consumption and colorectal cancer mortality are relatively limited. Further large prospective studies which examine the association between each type of dairy products and colorectal cancer incidence or mortality in different populations are warranted.

No potential conflicts of interest were disclosed.

S. Jin: Data curation, formal analysis, writing–original draft, writing–review and editing. Y. Kim: Formal analysis, methodology, writing–review and editing. Y. Je: Conceptualization, supervision, funding acquisition, methodology, project administration, writing–review and editing.

The National Research Foundation of Korea, funded by Ministry of Science, ICT, and Future Planning, supported this project through grant no. NRF-2018R1D1A1B07045353 (to Y. Je).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.