Yogurt consumption and colorectal cancer incidence and mortality in the Nurses’ Health Study and the Health Professionals Follow-Up Study

ABSTRACT

Background

Yogurt is a commonly consumed fermented food. Regular yogurt consumption may contribute to a favorable gut microbiome and gut health, but few epidemiologic studies have considered the relation between regular yogurt consumption and the incidence of and mortality from colorectal cancer.

Objectives

We used data from 2 large, prospective cohort studies, the Nurses’ Health Study and the Health Professionals Follow-Up Study, to examine the role of yogurt consumption on colorectal cancer incidence and mortality.

Methods

During 32 years of follow-up in 83,054 women (mean age at baseline, 45.7 years) and 26 years of follow-up in 43,269 men (mean age at baseline, 52.3 years), we documented a total of 2666 newly diagnosed cases of colorectal cancer in these cohorts. We modeled yogurt consumption at baseline and cumulatively updated it throughout follow-up.

Results: Baseline yogurt consumption was associated with a reduced risk of colon cancer in age-adjusted analyses (P for trend < 0.001). Associations remained statistically significant after adjusting for potential confounders, including calcium and fiber intake (P for trend = 0.03), and were restricted to proximal colon cancer. The consumption of 1 + servings per week of yogurt at baseline, compared to no yogurt consumption, was associated with a multivariable HR of 0.84 (95% CI, 0.70–0.99; P trend = 0.04) for the proximal colon cancer incidence. Latency analyses suggested that the most important window of opportunity for regular yogurt consumption to prevent colorectal cancer was 16–20 years in the past. When yogurt consumption was cumulatively updated, associations attenuated and were no longer significant. No statistically significant inverse trend was observed between yogurt consumption and the colorectal cancer mortality.

Conclusions

In these large cohorts, the frequency of yogurt consumption was associated with a reduced risk of proximal colon cancer with a long latency period. No significant inverse trend was observed for colorectal cancer mortality.

Introduction

Colorectal cancer (CRC) is the third most common cancer among women and men worldwide (1). Several important risk factors for CRC have been identified, including genetic predisposition and epigenetic factors, cigarette smoking, obesity, and low physical activity (2–8). Moreover, this cancer of the digestive tract is affected by a number of dietary factors, including regular alcohol intake and red and processed meat consumption (9, 10). In this context, the microbiome is likely an important mediator, with the intestinal microflora harboring a wealth of antagonist organisms; their equilibrium is affected by diet and creates an environment that may prevent or foster tumorigenesis of the intestinal system (11). The human microbiota is a composite of trillions of microbial cells and viruses that affect many aspects of human health and physiology (12). Fermented foods contain a large number of live microbes and thus act as probiotics, enriching the gut with beneficial bacteria that help the body absorb nutrients and enhance immune functions by stimulating phagocytosis and preventing inflammation (13). As these resident microbes nurture on nondigestible fibers (prebiotics), they produce immune-modulating metabolites, such as short-chain fatty acids (14). The consumption of fermented foods increases beneficial gut microbes by up to 10,000-fold, since a large fraction of bacteria produced during the fermentation process survive passage through the digestive system (15). Lactic acid bacteria compete for receptors or adhesion to endothelial cells, therefore preventing the access of pathogens to the intestinal epithelium, generating antimicrobial compounds, and producing proteolytic enzymes (16, 17). Moreover, lactic acid bacteria can inactivate carcinogenic substances, such as N-nitrosamines and heterocyclic aromatic amines, due to the peptidoglycans and polysaccharides in their cell walls that bind these mutagens (18–20). Lactic acid bacteria also reduce the fecal enzyme activity of b-glucoronidase, nitroreductase, and azoreductase, which convert procarcinogens to carcinogens in the colon (21).

Yogurt is among the most commonly consumed fermented foods (22). An early mention of fermented milk dates back to 76 BC when the Roman historian Plinius recommended its use to treat gastrointestinal infections (23). In the early 1900s, Metchnikoff (24) attributed the longevity of Bulgarians to their frequent consumption of yogurt containing Lactobacillus; he suggested that enrichment with these beneficial bacteria would counterbalance the toxin-producing bacteria. Few studies have considered the effect of frequent yogurt consumption on the CRC incidence (25–29). Proximal and distal colon cancer have been associated with different origins and risk factors; Fusobacterium nucleatum is more abundant in proximal than distal colon cancer (30). Hence, it is important to study these subtypes of colon cancer separately. Since gut bacteria ferment soluble fiber, the interaction between probiotics and prebiotics has recently garnered interest (31, 32).

We used data from 2 large, prospective US cohorts, the Nurses’ Health Study (NHS) and the Health Professionals Follow-Up Study (HPFS), to assess the association of yogurt consumption and the incidence of cancer of the colon and rectum. Because some of the proposed mechanisms involve factors such as immunity that could influence cancer progression and metastases independent of the cancer incidence and since other dietary and lifestyle factors have been associated with both the colorectal incidence and mortality (33), we also examined the latter.

Methods

Study populations

The NHS began in 1976 when 121,700 US female registered nurses aged 30–55 years provided information on their medical history and lifestyle (34). The HFPS began in 1986 when 51,529 US male dentists, pharmacists, optometrists, osteopaths, podiatrists, and veterinarians aged 40–75 years provided information on their medical history and lifestyle (35). Every 2 years, follow-up questionnaires have been sent to update both cohorts’ information. In 1980, a validated 61-item FFQ was included to assess the intake of specific foods in the NHS (36). Expanded FFQs updated dietary intakes in 1984 and every 4 years between 1986 and 2010. In 1986, a similar validated 131-item FFQ was used in the HPFS, and this was administered every 4 years between 1990 and 2010 (37). We restricted our study population to participants who returned the baseline FFQ (92,468 women in 1980 and 49,934 men in 1986). We excluded participants who left excessive blank items on the baseline FFQ (≥10 of the 61 FFQ items in 1980 for women or ≥70 on the 131-item FFQ for men; n = 373), implausibly low or high energy intake (<600 or >3500 kcal/day for women and <800 or >4200 kcal/day for men; n = 419), participants with missing information on yogurt consumption (n = 8283), and those with previously diagnosed cancer (except nonmelanoma skin cancer; n = 5565) or ulcerative colitis (n = 1305) (38). For the CRC mortality analysis, we further excluded participants who had diabetes (n = 3105) and/or cardiovascular disease (n = 3784) at baseline, since their dietary habits are likely influenced by these conditions. The final baseline populations (in 1980 for the NHS and 1986 for the HPFS) consisted of 83,054 women and 43,269 men in the CRC study and 79,725 women and 40,327 men in the CRC mortality study.

The study protocol was approved by the Committee on the use of Human Subjects in Research at Brigham and Women's Hospital and the Institutional Review Board of the Harvard T. H. Chan School of Public Health, both in Boston, Massachusetts, and by participating registries as required. The return of a questionnaire was considered to imply consent.

Ascertainment of diet

For each food on the FFQ, a commonly used unit or portion size was specified and participants were asked how often on average during the previous year the food was consumed, choosing from response categories of never or less than once per month, 1–3 times per month, 1 time per week, 5–6 times per week, 1 time per day, 2–3 times per day, 4–5 times per day, or 6+ times per day. The nutrient intake was calculated by multiplying the frequency by the nutrient composition in a standard portion size of that food and then summing the nutrient intake from all food items. The food composition database was created primarily from USDA sources (39).

The standard serving size for yogurt was one 8 oz cup (237 mL). Since 1994, information on yogurt consumption was obtained as “plain” and “flavored”; since 2010, questions on yogurt consumption have been separated into “plain yogurt,” “flavored/sweetened yogurt,” and “artificially sweetened yogurt.” The follow-up time since separating the yogurt question into subtypes is too short to provide sufficient statistical power to analyze the categories separately. For the present analysis, yogurt consumption included the sum of all yogurt types and was categorized into 3 categories: no yogurt consumption or less than 1 serving per month, 1–3 servings/month, and 1+ servings/week.

The reproducibility and validity of the FFQs in measuring food intake have been previously described (36, 37, 40–42). Self-reported yogurt consumption on the FFQ showed high validity when compared with multiple diet records, with a correlation coefficient of 0.97 (40).

Ascertainment of cancer

For both cohorts, on each mailed questionnaire we asked whether a participant was diagnosed with CRC or other diseases within the previous 2 years. When a participant reported a diagnosis of CRC, we obtained permission to review medical records and pathology reports to determine histological type, anatomic location, and cancer stage. Our study physicians reviewed each self-reported case of CRC based on the medical records and pathology reports, while blinded to any exposure data. We only included cases of CRC that were confirmed based on this record review in our analysis.

Ascertainment of death

Deaths were identified from state statistics records, the National Death Index, next of kin, and the postal system. Both cohorts ascertained deaths with more than 98% completeness (43). Causes of death, specifically from colon or rectal cancer, were identified from death certificates or reviews of medical records by study physicians.

Statistical analysis

Each participant contributed person-time of follow-up from the date of return of the baseline questionnaire (1980 for NHS and 1986 for HPFS) to the date of CRC diagnosis, death, loss to follow-up, or end of analysis follow-up (June 1, 2012 for NHS and January 31, 2012 for HPFS), whichever came first. We used Cox proportional hazards regression models to calculate the HRs and 95% CIs of the association between yogurt consumption and the risk of CRC. Our primary outcome was CRC incidence and our secondary outcome was CRC mortality. The proportional hazards assumption was tested by adding interaction terms between follow-up time and yogurt consumption. To control as finely as possible for confounding by age, calendar time, and possible interactions between these 2 time scales, we stratified models jointly by age (in months) and 2-year questionnaire cycle. In multivariable models, we additionally adjusted for potential dietary and nondietary confounders, including family history of CRC, history of sigmoidoscopy or colonoscopy, BMI (in kg/m²; continuous), height in quintiles, physical activity (in metabolic-equivalent hours/week; <3, 3–9, 9–18, 18–27, or ≥27), pack-years of smoking before age 30 (continuous), current multivitamin use, regular aspirin or nonsteroidal anti-inflammatory drug (NSAID) use (≥2 tablets/week versus none), menopausal status and age at menopause (premenopausal, unknown menopause, or postmenopausal and age at menopause: <47, 47–49, 50, 51, 52–53, or 54–60 years), menopausal status and hormone use in women (premenopausal and never, past, and current users of postmenopausal hormones), total caloric intake (quintiles), alcohol consumption (in g/d; <5, 5–10, 10–15, 15–30, or ≥30), and energy-adjusted intake of folate (quintiles), vitamin D (quintiles), total fiber (quintiles), unprocessed red meat (quintiles), and processed meat (quintiles). In separate models we added total (dietary and supplemental) calcium intake (quintiles) as a covariate to examine whether calcium, an important nutrient in yogurt, may mediate any effect yogurt may have on CRC, and therefore confound any association between the microbial component of the yogurt and CRC. In additional analyses, we explored potential confounding by other dairy products by adjusting for non-yogurt dairy. In our longitudinal analyses, we updated covariate information, when available, with every questionnaire cycle. Tests for linear trend for the exposure were performed by assigning the median value of each yogurt category to all participants in that group.

We modeled the association between yogurt consumption and the CRC incidence using 2 separate approaches: 1) a baseline model using information on yogurt consumption from the baseline questionnaire only, in which yogurt intake values were derived directly from the 1980 (NHS) and 1986 (HPFS) questionnaires; or 2) a cumulative average intake model, to minimize the impact of random error in reporting of dietary intake and to best reflect long-term dietary habits (44), in which yogurt represents the cumulative average intake since 1980 (NHS) or 1986 (HPFS) from all available FFQs until the beginning of each follow-up interval. In addition, we performed latency analyses to identify the most important vulnerable period for yogurt consumption to impact CRC risk by including time lags of 4–8, 8–12, 12–16, 16–20, or 20–24 years. We also assessed the association between yogurt consumption at baseline and the cumulative average intake of yogurt and CRC mortality in the initially cancer-free cohort.

Interactions between yogurt consumption and fiber intake were explored by creating a cross-product term between yogurt and fiber and applying a likelihood ratio test.

Duplication-method time-varying Cox proportional hazards regression models were used to examine whether the associations between yogurt consumption and colon cancer differed by subtypes (proximal and distal) (45). This method allows for the estimation of separate associations of yogurt consumption in proximal and distal tumors, and tests whether yogurt consumption has statistically different regression coefficients for different colon cancer subtypes.

Heterogeneity in associations across the 2 cohorts was assessed using a test for heterogeneity (likelihood ratio test). In the absence of heterogeneity, pooled analyses for the 2 cohorts were conducted using a stratified Cox proportional hazards model, maintaining a cohort indicator in the model. All tests were 2-sided and statistical significance was assessed at the 5% level.

All data analyses were performed using SAS V9.4.

Results

During 3,393,373 person-years of follow up between 1980 and 2012, a total of 2666 incident cases of CRC (1965 colon and 579 rectal cancers) were observed in the 2 cohorts. Characteristics of the NHS and HPFS study participants included in these analyses, according to their frequency of yogurt consumption, are provided in Tables 1 and 2. On average, NHS participants consumed 0.85 servings of yogurt per week, and HPFS participants consumed 0.66 servings of yogurt per week. Participants with more frequent yogurt consumption were somewhat older, more likely to be postmenopausal, more physically active, more likely to have had a sigmoidoscopy or colonoscopy, more likely to use multivitamins, and generally had a diet found to be associated with a lower risk of CRC (lower alcohol intake, less frequent red and processed meat consumption, and more frequent calcium, vitamin D, fiber, and folate intakes). Individuals with the most frequent yogurt consumption also had slightly higher non-yogurt dairy consumption.

TABLE 1.

Age-standardized characteristics of the participants in the Nurses’ Health Study included in this analysis, averaged throughout the study period between 1980 and 2012

	Frequency of Yogurt Consumption
Characteristic	Never or <1 serving/month	1–3 servings/month	1+ servings/week
Person-years	833,869	630,288	961,735
Age, not age-standardized	57.2 (11.2)	60.3 (11.3)	62.5 (10.8)
Height, inches	64.4 (2.4)	64.5 (2.4)	64.6 (2.4)
BMI, kg/m2	25.2 (4.7)	25.5 (4.7)	25.3 (4.5)
Physical activity, METs hours/week	13.8 (16.7)	15.4 (17.5)	18.5 (18.8)
Pack-years of smoking before age 30	7.1 (5.1)	6.9 (5.3)	6.9 (5.5)
Family history of colorectal cancer, %	16.7	16.5	16.1
History of sigmoidoscopy/colonoscopy, %	13.9	18.3	20.9
Multivitamin use, %	41.7	50.0	57.3
Aspirin use, %	47.6	49.1	50.8
Nullipararity, %	6.0	5.3	5.5
Number of children among parous women	2.9 (1.0)	2.9 (1.0)	2.8 (0.9)
Age at first birth	25.2 (3.4)	25.1 (3.3)	25.0 (3.2)
Postmenopausal, %	69.4	73.3	75.1
Current postmenopausal hormone use, %	46.0	48.5	48.1
Total caloric intake, calories/day	1618 (452)	1643 (423)	1752 (428)
Total dairy consumption, servings/day	2.2 (1.4)	2.2 (1.2)	2.6 (1.3)
Total non-yogurt dairy consumption, servings/day	2.2 (1.4)	2.2 (1.2)	2.4 (1.2)
Red meat consumption, servings/week	4.2 (2. 9)	3.7 (2.7)	3.4 (2.5)
Processed meat consumption, servings/week	2.1 (2.3)	1.7 (2.0)	1.4 (1.8)
Alcohol intake, g/d	6.7 (10.8)	5.9 (9.1)	5.6 (8.0)
Folate intake, μg/d	354 (208)	387 (200)	425 (197)
Calcium intake, mg/d	801 (329)	905 (333)	1047 (334)
Vitamin D intake, IU/d	314 (224)	350 (220)	390 (215)
Fiber intake, g/d	14.9 (4.9)	16.4 (4.6)	17.5 (4.7)

Values are means (SD) or percentages and are standardized to the age distribution of the study population. Abbreviation: METs, metabolic equivalents.

TABLE 2.

Age-standardized characteristics of the participants in the Health Professionals Follow-Up Study included in this analysis, averaged throughout the study period between 1986 and 2012

	Frequency of Yogurt Consumption
Characteristic	Never or <1 serving/ month	1–3 servings/month	1+ servings/week
Person-years	441,948	253,817	271,717
Age, not age-standardized	62.8(11.2)	63.3(11.1)	64.3(11.0)
Height, inches	70.1 (2.8)	70.2 (2.8)	70.2 (2.8)
BMI, kg/m2	25.6 (3.5)	25.7(3.5)	25.5 (3.5)
Physical activity, METs hours/week	27.2 (27.4)	31.1(28.3)	35.8 (30.8)
Pack-years of smoking before age 30	11.3 (6.5)	10.9(6.7)	10.4 (6.5)
Family history of colorectal cancer, %	13.7	13.3	13.3
History of sigmoidoscopy/colonoscopy, %	21.1	25.6	27.6
Multivitamin use, %	40.4	48.3	55.5
Aspirin use, %	45.0	48.8	50.3
Total caloric intake	1934 (560)	1942(535)	2056 (541)
Total dairy consumption, servings/day	2.0 (1.5)	2.0 (1.3)	2.5 (1.4)
Total non-yogurt dairy consumption, servings/day	2.0 (1.5)	2.0 (1.3)	2.2 (1.3)
Red meat consumption, servings/week	4.3 (3.2)	3.7 (2.9)	3.4 (2.8)
Processed meat consumption, servings/week	2.5 (3.0)	2.0 (2.5)	1.7 (2.3)
Alcohol use, g/d	12.1 (15.4)	10.4 (13.0)	9.7 (11.6)
Folate intake, μg/d	496 (247)	554 (248)	608 (254)
Calcium intake, mg/d	861 (367)	932 (359)	1052 (364)
Vitamin D intake, IU/d	398 (267)	437 (260)	479 (259)
Fiber intake, g/d	20.6 (6.5)	22.6 (6.3)	23.9 (6.4)

Values are means (SD) or percentages and are standardized to the age distribution of the study population. Abbreviation: METs, metabolic equivalents.

Frequent yogurt consumption at baseline (1980 for NHS and 1986 for HPFS) was inversely related to CRC incidences in age-adjusted analyses among both women and men (P for trend <0.01; Supplemental Tables 1 and 2). The inverse relation remained but was no longer statistically significant after adjustment for potential confounding variables, including a family history of CRC, history of sigmoidoscopy/colonoscopy, BMI, height, physical activity, smoking before age 30, current multivitamin use, regular aspirin or NSAID use, menopausal status and age at menopause, menopausal status and hormone use in women, total caloric intake, alcohol consumption, and energy-adjusted intakes of folate, calcium, vitamin D, total fiber, unprocessed red meat, and processed meat for colon cancer. The association appeared specifically for proximal cancer, and only among women (P for trend = 0.06 for proximal cancer), but not among men (Supplemental Tables 1 and 2). The 2 cohorts were combined in the absence of significant heterogeneity, and both age-adjusted as well as multivariable estimates suggested significant associations with CRC (P for trend = 0.05; Table 3). The consumption of 1 or more servings per week of yogurt at baseline was associated with an HR of 0.89 (95% CI, 0.80–1.00) for CRC incidence. When subtypes of CRC were considered separately, the protective association with regular yogurt consumption was restricted to proximal colon cancer (HR = 0.84; 95% CI, 0.70–0.99). No important multivariable association was observed between yogurt consumption at baseline and distal colon or rectal cancer. However, duplication-method Cox proportional hazards analyses did not detect a statistically significant difference between the subtypes of CRC in their relation to yogurt consumption.

TABLE 3.

Associations between yogurt consumption at baseline and incidences of colorectal cancer and its subtypes

Yogurt consumption	Number of cases	Person-years	Age-adjusted HR (95% CI)	Multivariable HR (95% CI)	Multivariable HR without calcium (95% CI)
Colorectal cancer
Never or <1 serving/month	1755	2,063,751	1	1	1
1–3 servings/month	484	669,760	0.92 (0.83–1.02)	0.97 (0.87–1.07)	0.96 (0.87–1.07)
1+ servings/week	427	659,862	0.80 (0.72–0.89)	0.89 (0.80–1.00)	0.88 (0.79–0.99)
P for trend			<0.0001	0.05	0.03
Colon cancer
Never or <1 serving/month	1297	2,064,185	1	1	1
1–3 servings/month	359	669,877	0.92 (0.82–1.04)	0.97 (0.86–1.09)	0.96 (0.85–1.09)
1+ servings/week	309	659,970	0.78 (0.69–0.89)	0.87 (0.76–0.99)	0.86 (0.75–0.97)
P for trend			0.0001	0.03	0.02
Proximal colon cancer
Never or <1 serving/month	745	2,064,696	1	1	1
1–3 servings/month	201	670,025	0.89 (0.76–1.04)	0.92 (0.79–1.08)	0.92 (0.79–1.08)
1+ servings/week	179	660,091	0.78 (0.66–0.92)	0.84 (0.70–0.99)	0.84 (0.71–1.00)
P for trend			0.003	0.04	0.05
Distal colon cancer
Never or <1 serving/month	523	2,064,868	1	1	1
1–3 servings/month	151	670,059	0.97 (0.81–1.16)	1.04 (0.86–1.25)	1.02 (0.85–1.22)
1+ servings/week	122	660,133	0.78 (0.64–0.95)	0.91 (0.74–1.12)	0.87 (0.71–1.07)
P for trend			0.01	0.36	0.18
Rectal cancer
Never or <1 serving/month	378	2,065,022	1	1	1
1–3 servings/month	103	670,115	0.89 (0.71–1.11)	0.93 (0.75–1.17)	0.93 (0.75–1.17)
1+ servings/week	98	660,162	0.85 (0.68–1.06)	0.95 (0.76–1.21)	0.95 (0.76–1.20)
P for trend			0.16	0.72	0.72

Baseline was 1980 in NHS and 1986 in HPFS. Data are from 83,054 participants in the NHS and 43,269 participants in the HPFS (2666 events) between 1980/1986 and 2012. HRs and 95% CIs were generated by Cox proportional hazards analyses adjusted for age, 2-year follow-up cycle, family history of colorectal cancer, history of lower gastrointestinal endoscopy, BMI, height, physical activity, pack-years of smoking before age 30, current multivitamin use, regular aspirin or nonsteroidal anti-inflammatory drug use, parity in women and age at first birth in women, menopausal status and age at menopause, menopausal status and hormone use in women, total caloric intake, alcohol consumption, and energy-adjusted intakes of folate, calcium, vitamin D, total fiber, unprocessed red meat, and processed meat. Abbreviations: HPFS, Health Professionals Follow-Up Study; NHS, Nurses’ Health Study.

Since the calcium content of the yogurt (and other dairy products) is a possible confounder for potential microbial protection, we also ran multivariate models without adjusting for calcium intake. The results for proximal colon cancer remained unchanged, whereas the association with distal colon cancer became somewhat stronger, suggesting partial confounding. Adjustment for non-yogurt dairy did not appreciably change the results for any of the CRC subtypes (data not shown). We further explored the interaction between yogurt consumption and fiber intake; we observed an interaction of borderline statistical significance (P = 0.05 for model with calcium; P = 0.049 for model without calcium). Interestingly, the CRC risk was lowest among participants with high yogurt and low fiber intakes (Supplemental Table 3). Individuals who consumed 1 or more servings of yogurt per week and had a mean fiber intake of about 15 grams per day had a 21% (95% CI, 7–32%) lower incidence of CRC than those who consumed the same amount of fiber but less than 1 serving of yogurt per week. The CRC incidence did not differ in participants with the highest fiber intake (mean, 23 grams per day), compared to individuals with low yogurt and low fiber consumption.

We further explored cumulatively updating yogurt consumption throughout follow-up. Frequent yogurt consumption was associated with lower colon cancer incidences in age-adjusted analyses for both proximal and distal colon cancer (Table 4), with the association with proximal colon cancer restricted to women (Supplemental Table 4) and the association with distal colon cancer restricted to men (Supplemental Table 5). Adjustment for potential confounding variables diminished the association to nonsignificance, with 1 or more servings of yogurt per week associated with an HR of 0.92 (95% CI, 0.79–1.08) for proximal colon cancer and an HR of 0.97 (95% CI, 0.80–1.17) for distal colon cancer (Table 4). Latency analyses suggested that the most important window of opportunity for regular yogurt consumption to prevent CRC was 16–20 years in the past (Supplemental Table 6).

TABLE 4.

Association between cumulatively updated yogurt consumption and incidences of colorectal cancer and its subtypes

Yogurt consumption	Number of cases	Person-years	Age-adjusted HR (95% CI)	Multivariable HR (95% CI)	Multivariable HR without calcium (95% CI)
Colorectal cancer
Never or <1 serving/month	1088	1,275,817	1	1	1
1–3 servings/month	691	884,105	0.90 (0.82–1.00)	0.97 (0.88–1.07)	0.96 (0.87–1.06)
1+ servings/week	887	1,233,452	0.84 (0.76–0.92)	0.97 (0.87–1.07)	0.95 (0.86–1.05)
P for trend			0.0004	0.60	0.37
Colon cancer
Never or <1 serving/month	786	1,276,097	1	1	1
1–3 servings/month	527	884,263	0.94 (0.84–1.06)	1.01 (0.90–1.13)	1.00 (0.89–1.12)
1+ servings/week	652	1,233,671	0.83 (0.74–0.92)	0.95 (0.84–1.07)	0.93 (0.83–1.05)
P for trend			0.001	0.32	0.18
Proximal colon cancer
Never or <1 serving/month	422	1,276,437	1	1	1
1–3 servings/month	312	884,456	0.98 (0.85–1.14)	1.02 (0.88–1.19)	1.03 (0.88–1.20)
1+ servings/week	391	1,233,920	0.85 (0.74–0.98)	0.92 (0.79–1.08)	0.93 (0.80–1.09)
P for trend			0.02	0.20	0.24
Distal colon cancer
Never or <1 serving/month	347	1,276,477	1	1	1
1–3 servings/month	209	884,554	0.92 (0.77–1.09)	1.01 (0.84–1.21)	0.99 (0.82–1.18)
1+ servings/week	240	1,234,029	0.79 (0.66–0.93)	0.97 (0.80–1.17)	0.91 (0.76–1.09)
P for trend			0.007	0.70	0.29
Rectal cancer
Never or <1 serving/month	247	1,276,586	1	1	1
1–3 servings/month	129	884,632	0.74 (0.59–0.91)	0.79 (0.63–0.99)	0.79 (0.63–0.99)
1+ servings/week	203	1,234,081	0.84 (0.69–1.02)	0.99 (0.80–1.22)	0.98 (0.80–1.21)
P for trend			0.28	0.59	0.62

Data are from 83,054 participants in the NHS and 43,269 participants in the HPFS (2666 events) between 1980/1986 and 2012. HRs and 95% CIs were generated by Cox proportional hazards analyses adjusted for age, 2-year follow-up cycle, family history of colorectal cancer, history of lower gastrointestinal endoscopy, BMI, height, physical activity, pack-years of smoking before age 30, current multivitamin use, regular aspirin or nonsteroidal anti-inflammatory drug use, parity in women and age at first birth in women, menopausal status and age at menopause, menopausal status and hormone use in women, total caloric intake, alcohol consumption, and energy-adjusted intake of folate, calcium, vitamin D, total fiber, unprocessed red meat, and processed meat. Abbreviations: HPFS, Health Professionals Follow-Up Study; NHS, Nurses’ Health Study.

During 3,313,192 person-years of follow-up, 1086 cases of fatal CRC were observed in these 2 prospective cohorts combined. Analyses of fatal CRC revealed a lower risk for yogurt eaters, especially when yogurt consumption was updated throughout follow-up. These trends were attenuated and nonsignificant in multivariable models (Table 5).

TABLE 5.

Association between yogurt consumption and colorectal cancer mortality

Yogurt consumption	Number of cases	Person-years	Age-adjusted HR (95% CI)	Multivariable HR (95% CI)	Multivariable HR without calcium (95% CI)
Baseline
Never or <1 serving/month	740	2,013,468	1	1	1
1–3 servings/month	174	654,889	0.79 (0.67–0.93)	0.86 (0.73–1.02)	0.85 (0.72–1.01)
1+ servings/week	172	644,835	0.76 (0.64–0.90)	0.90 (0.75–1.07)	0.88 (0.74–1.04)
P for trend			0.002	0.26	0.18
Cumulatively updated
Never or <1 serving/month	458	1,238,440	1	1	1
1–3 servings/month	258	863,422	0.74 (0.63–0.86)	0.83 (0.71–0.97)	0.82 (0.70–0.96)
1+ servings/week	370	1,211,329	0.72 (0.62–0.83)	0.91 (0.78–1.07)	0.88 (0.76–1.03)
P for trend			0.0003	0.61	0.38

Data are from 1086 events in 79,725 participants in the NHS and 40,327 participants in the HPFS between 1980/1986 and 2012. HRs and 95% CIs were generated by Cox proportional hazards analyses adjusted for age, 2-year follow-up cycle, family history of colorectal cancer, history of lower gastrointestinal endoscopy, BMI, height, physical activity, pack-years of smoking before age 30, current multivitamin use, regular aspirin or nonsteroidal anti-inflammatory drug use, parity in women and age at first birth in women, menopausal status and age at menopause, menopausal status and hormone use in women, total caloric intake, alcohol consumption, and energy-adjusted intake of folate, calcium, vitamin D, total fiber, unprocessed red meat, and processed meat. Abbreviations: HPFS, Health Professionals Follow-Up Study; NHS, Nurses’ Health Study.

Discussion

In these 2 large, prospective cohort studies, frequent yogurt consumption was associated with a reduced incidence of colon cancer. Yogurt consumption in the past, possibly decades earlier, was inversely associated with proximal colon cancer, whereas long-term regular consumption was slightly more inversely related to distal colon cancer; however, the latter association vanished after adjustment for potential confounding variables, particularly calcium intake. While the association between frequent yogurt consumption and proximal colon cancer characterized by long latency was independent of the calcium content of the yogurt, calcium (but not dairy consumption other than yogurt) explained some of the inverse association between more recent yogurt consumption and distal cancer, confirming results from the previously conducted European Prospective Investigation into Cancer and Nutrition (EPIC). Similar results were found in a previous analysis (46) and in the pooling project of prospective cohorts (47).

Fermented foods enrich the microbiota with favorable bacteria, which support immune function and reduce inflammation. Considerable interest has been paid to the role of the microbiome in colorectal carcinogenesis, but few studies have examined the contribution of specific probiotic foods. In rodent models, yogurt ingestion but not nonfermented milk supplementation reduced the number and size of chemically induced colorectal tumors (48). In other mouse models, yogurt intake lowered chemically induced colorectal carcinoma by enhancing cellular apoptosis and modulating the inflammatory immune response (49–51).

Population-based studies have yielded mixed results. In a case-control study in Los Angeles county, an inverse association was observed for regular yogurt consumption and colon cancer, with an odds ratio of 0.83 (95% CI, 0.70–0.98) for 10 servings of yogurt per month compared to no yogurt consumption (25). A similar inverse association with fermented milk was found in the Netherlands Cohort Study (26). In EPIC, yogurt consumption was also inversely related to colon cancer (HR = 0.88; 95% CI, 0.77–1.00; comparing the highest quartile to the lowest quartile of yogurt consumption) among the 477,122 participants. After adjustment for dietary calcium, the inverse association was no longer significant (27). While in the Italian EPIC population of 45,241 women and men the inverse association between yogurt consumption and CRC was more pronounced even when controlling for calcium and other nutrients (HR = 0.65; 95% CI, 0.48–0.89; comparing the top with the bottom tertile of yogurt intake) (28), no relation was evident in the Spanish Prevenciόn con Dieta Mediterránea study comprising 7216 participants (29). Among these prior studies, proximal and distal colon cancer associations were considered only in the EPIC cohort; the inverse association between yogurt consumption (highest quartile of intake) and CRC was more pronounced for distal cancer, possibly reflecting the calcium content (27). Yogurt consumption was assessed at baseline with a mean of 11 years of follow-up, and calcium intake was not adjusted for in the analyses.

Possible explanations for the diverse findings include differences in the study populations’ genetic backgrounds, their different lifestyle and dietary habits, and different study designs and dietary assessment methods. The distal colon cancer finding might have resulted at least in part from confounding by calcium intake. Additionally, the consumption of fermented foods, and specifically yogurt, is less frequent in the United States than in Northern, middle, and Eastern Europe, and plain yogurt without additives like sugar, fruit, or flavorings is considerably more common in Europe (52–56). Among the EPIC participants, the top quintile consumed about 1 serving of yogurt per day (27), whereas in the Harvard cohorts the top quintile consumed closer to 1 serving per week. In the United States and in Europe, yogurt must be fermented with at least Lactobacillus bulgaricus and Streptococcus thermophilus; other bacteria such as Lactobacillus acidophilus and other strains of Lactobacillus and Bifidobacterium are often added (57, 58).

The stronger association of proximal colon cancer with yogurt consumption at baseline observed in our study may reflect the longer latency period, relevant for the cancer of this site in its multistage process. Indeed, our latency analyses point towards a 16–20-year latency for the association between yogurt consumption and overall CRC incidence, which is consistent with the stronger associations found with the baseline models. This points towards a role of yogurt consumption in the earlier phases of CRC initiation, and thus emphasizes its potential importance in primary prevention. Consistent with our previous findings, we found a stronger association between calcium intake with distal colon cancer, but no relation with proximal colon cancer, suggesting that confounding by calcium intake does not explain the association between yogurt intake and proximal colon cancer (46). Calcium intake confounds the association with distal colon cancer.

Recently, the infection hypothesis for CRC has been rekindled. Findings from our 2 cohorts have suggested a role for the Fusobacterium nucleatum in colon carcinogenesis, which was restricted to proximal colon cancer (30). The inverse association between yogurt consumption and proximal colon cancer in our cohorts is of interest in this context, as lactic acid bacteria in the intestine can suppress the growth of pathogenic microbiota, thereby countering an infection with pathogens (59, 60). This may also explain the long latency period for yogurt consumption to affect proximal colon cancer, since an imbalance in gut microbiota is likely an early step in cancer initiation.

Yogurt production employs a broad spectrum of bacteria. While Lactobacillus bulgaricus and Streptococcus thermophilus are essential cultures in yogurt, different brands of yogurt use different additional bacteria species for milk fermentation (57, 58, 61). In most yogurt production, live cultures of Lactobacillus bulgaricus, Streptococcus thermophiles, and other probiotic bacteria are added to pasteurized milk to convert it to yogurt during fermentation (61, 62). The number of bacteria diminishes over time and with higher temperature storage conditions. Of note, some commercially available yogurt is pasteurized after fermentation, which inactivates nearly all live bacteria, including the probiotic ones. However, the beneficial metabolites produced by the bacteria, such as short-chain fatty acids, remain largely intact and may still render some of the favorable effects associated with consuming fermented foods (63). Among fermented foods, yogurt generally harbors fewer bacteria than kefir, kimchi, or sauerkraut, which are likely to have more profound probiotic effects (64–67). However, none of these other foods are consumed with sufficient frequency on a population level to allow exploration in an observational study; hence, evaluating their preventive potential for CRC is challenging.

The observation of a potential interaction between yogurt consumption and fiber intake in our cohorts was unexpected but may lend support to the particular benefit of yogurt consumption if fiber intake is low. Prebiotic fiber might enhance the effect of a probiotic food like yogurt; however, the majority of fiber intake in this population is insoluble, and hence is not fermented by gut bacteria. We are not aware of any previous studies exploring this possible interaction. Future studies need to confirm this interaction, which was of borderline statistical significance in our data.

Limitations of our study include the relatively modest amount of yogurt consumption in our study population (and in the United States in general) and the possibility of errors in self reports, although our validation studies have demonstrated high validity of self-reported yogurt consumption by the FFQ. While we were able to control for numerous potential confounding variables, residual or unmeasured confounding may have affected our results. Observational studies on yogurt consumption are generally challenged by the diversity of bacterial cultures used in milk fermentation, which prohibits inference about specific bacteria. Strengths include the inclusion of data from 2 large, well-maintained, longitudinal cohorts with repeated dietary assessments (reducing nondifferential misclassification), extensive follow-ups, continuously updated assessments of numerous potential confounders, and verified disease endpoints.

In summary, we observed a significant trend in proximal colon cancer risk reduction with more frequent consumption of yogurt decades earlier. Whether diet in adolescence and early adulthood may induce long-lasting changes in the microbiota remains to be elucidated. For the distal colon cancer incidence, long-term yogurt consumption and the beneficial effect of its calcium content may be more relevant. Residual confounding, however, cannot be excluded as an explanation for our findings, especially since yogurt consumption is associated with many healthy behaviors, not all of which may have been assessed or were even quantifiable. Future studies need to confirm these findings, in particular with respect to heterogeneity in the anatomic subsites.

Notes

The Nurses’ Health Study is supported by federal research grants from the National Cancer Institute, National Institutes of Health (P01CA87969 and UM1CA186107). The Health Professionals Follow-Up Study is supported by federal research grants from the National Cancer Institute, National Institutes of Health (P01CA55075 and U01CA167552). The funders had no role in the design, implementation, analysis or interpretation of the data.

Supplemental Tables 1–6 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Data described in the manuscript, code book, and analytic code will be made available pending application to the Nurses’ Health Study and Health Professionals Follow-Up Study investigators with coverage of costs.

Abbreviations used: CRC, colorectal cancer; EPIC, European Prospective Investigation into Cancer and Nutrition; HPFS, Health Professionals Follow-Up Study; NHS, Nurses’ Health Study.