Calcium intake and colorectal cancer risk: Results from the Nurses’ Health Study and Health Professionals Follow-up Study

Abstract

The relationship between calcium intake and colorectal cancer (CRC) risk remains inconclusive. We conducted this study to evaluate whether the association between calcium intake and CRC risk differs by anatomic subsite and determine the dose-response relationship for this association, as well as assess when in carcinogenesis calcium may play a role. We assessed calcium intake every 4 years and followed 88,509 women (1980–2012) in the Nurses’ Health Study and 47,740 men (1986–2012) in the Health Professionals Follow-Up Study. We documented 3,078 incident CRC cases. Total calcium intake (≥1400 vs. <600 mg/d) was associated with a statistically significant lower risk of colon cancer (multivariable relative risk: 0.78, 95%CI: 0.65–0.95). Similar results were observed by different sources of calcium (from all foods or dairy products only). The inverse association was linear and suggestively stronger for distal colon cancer (0.65, 0.43–0.99) than for proximal colon cancer (0.94, 0.72–1.22, P-_{common effects}=0.14). Additionally, when comparing different latencies, the overall pattern suggested that the inverse association appeared to be stronger with increasing latency and was strongest for intakes 12–16 years before diagnosis. Comparing total calcium intakes of ≥1400 vs. <600 mg/d for intake 12–16 y before diagnosis, the pooled RR (95% CIs) of CRC was 0.76 (0.64–0.91). Higher calcium intake was associated with a lower risk of developing colon cancer, especially for distal colon cancer. Overall inverse association was linear and did not differ by intake source. Additionally, calcium intake approximately 10 years before diagnosis appeared to be associated with a lower risk of CRC.

Introduction

Colorectal cancer (CRC) is the third most common cancer in both US women and men.¹ The World Cancer Research Foundation and American Institute for Cancer Research concluded that diets high in calcium probably decrease CRC risk.^{2, 3} Calcium may reduce CRC risk via stimulating differentiation, reducing proliferation, and inducing apoptosis.^4–6 Furthermore, a few large randomized controlled trials (RCTs),^7–9 though not all,¹⁰ have reported that calcium supplementation reduced recurrence of colorectal adenomas, precursors of most sporadic CRC. In contrast, the largest RCT to date observed no apparent benefit of 1000 mg/d of elemental calcium on CRC risk during approximately 7 years of follow-up.¹¹ Possibly because of these inconsistencies, the Institute of Medicine stated that there was insufficient evidence to suggest a benefit of calcium for CRC prevention and called for more targeted research.¹²

At least three major questions remain to be answered. First, it is unclear whether the calcium and CRC association differs according to anatomic subsite of the tumor. Second, the dose-response relationship for this association is not well-understood. While a pooled analysis of 10 cohort studies suggested no further reduction in risk for intake exceeding 1,000 mg/d,¹³ a recent meta-analysis reported linear decline in risk of CRC associated with increasing intake with the range of 250–1,900 mg/d of total calcium.¹⁴ Of note, each of analyses was based upon a single baseline assessment of calcium, and the dose-response relation may be clarified with multiple assessments of diet. Third, considering that CRC likely develops over many years, it remains unknown when exposure to calcium may play the most significant role. If calcium acts primarily in reducing risk of adenomas, as suggested by observational data and at least some RCTs,^7–9 it may require approximately a decade to observe a significant reduction incidence of CRC.

To address these questions, we examined calcium intake in relation to CRC risk in two large, well-characterized cohorts of women (the Nurses’ Health Study, NHS ^{15, 16}) and men (the Health Professionals Follow-up Study, HPFS¹⁷), with repeated measures of calcium intake over long-term follow-up. The current study extends our earlier report on calcium intake and CRC risk,¹⁸ with an additional 16 years of follow-up, and over 2,000 additional cases of incident CRC. Overall, the repeated assessment of calcium intake over 3 decades and large number of cases allow us to identify anatomic sub-site differences, better characterize the dose-response relationship, and evaluate the timing of calcium intake and CRC risk.

Methods

Study population

The NHS^{15, 16} was established in 1976, when 121,700 married female registered nurses aged 30 to 55 years residing in 11 States in the U.S. completed and returned a self-administered questionnaire. The HPFS ¹⁷ is a cohort study of 51,529 U.S. male professionals who were aged 40 to 75 years at baseline in 1986. Questionnaires have been mailed to participants in both cohorts every 2 years since baseline to collect updated information on demographics, lifestyle factors, medical history, and disease outcomes. The follow-up rate has been greater than 90% in both studies. The institutional review board at the Brigham and Women’s Hospital and Harvard T. H. Chan School of Public Health approved the studies. As approved by the committee, return of the questionnaires was considered to imply informed consent. We used 1980 in NHS and 1986 in HPFS as baseline for this study because calcium intake was first assessed in those years. We excluded participants with a history of cancer at baseline (except non-melanoma skin cancer, n = 5,592 for NHS, n =1,779 for HPFS), with ulcerative colitis or Crohn’s disease (n = 321 for NHS, n = 492 for HPFS), and individuals who did not answer the baseline dietary questionnaires (n = 27,279 for NHS, n = 492 for HPFS). The final analytic cohort included 88,509 women and 47,740 men consisting of 3,622,835 person-years through 2012.

Assessment of calcium intake and other dietary factors

We used validated food frequency questionnaires (FFQs) with nearly 130 food items to obtain information on usual dietary intake over the past year in both cohorts. The FFQs were first administered in 1980 in NHS and in 1986 in HPFS, and repeated at least every 4 years thereafter in each cohort. We calculated nutrient intake by multiplying the frequency of each food consumed by the nutrient content of specified portion sizes, using composition data from the US Department of Agriculture and supplemented with other data sources. On each biennial questionnaire (except for the 1980 NHS questionnaire), we acquired the information on current use and dosage of calcium supplements and multivitamin use.¹⁸ Total calcium intake was calculated by summing calcium from dietary sources including fortified foods and supplements including multivitamins.¹⁸ We calculated the calcium from dairy sources alone by summing the contributions of all dairy products and food items containing dairy products. As reported previously,¹⁸ these food items included in the calculation of calcium from dairy sources were skim milk, 1%–2% milk, whole milk, yogurt (flavored and plain yogurt), ice cream, frozen yogurt, cottage cheese, cream cheese, other cheese, butter, clam chowder, mashed potatoes, pie (homemade), cake (homemade), cake (ready-made), donut, sweet roll (homemade and ready-made), muffin, pancake, pizza, chocolate, candy with chocolate, chocolate chip cookie (homemade and ready-made), white bread, and dark bread. Calcium from non-dairy sources was calculated by subtracting calculated dairy calcium intake from overall dietary calcium intake. All nutrients were adjusted for total energy intake using the residual method.¹⁹ The validity of the FFQs has been evaluated in 173 women from the NHS²⁰ and in 127 men from the HPFS.²¹ The energy-adjusted correlation coefficients of total calcium intake comparing the FFQ and the average of multiple 1-week diet records (four for women and two for men) were 0.63 for women^{18, 20} and 0.61 for men.²¹ The correlation coefficients for dietary calcium intake were 0.70 for women^{18, 20} and 0.60 for men.²¹

Assessment of other covariates

The baseline and biennial follow-up questionnaires included questions about CRC risk factors such as height, adult body weight, physical activity (METs-hrs/wk), cigarette smoking, sigmoidoscopy/colonoscopy screening, family history of CRC, aspirin use, and menopausal status and use of menopausal hormones.

Identification of CRC cases

Cancer and other disease outcomes have been reported by the participants in each cohort on the biennial questionnaires. Researchers received permission from study participants to obtain medical records and pathological reports on CRC, and, while blinded to exposure information, abstracted the information on anatomic location, stage, and histology. CRC as defined according to the International Classification of Diseases, Ninth Revision.²² Colon cancer (ICD-9 codes 153.0–153.4, 153.6–153.9) was further classified into proximal colon cancer (neoplasms from the cecum to the splenic flexure; ICD-9 codes 153.0, 153.1, 153.4, 153.6, 153.7) and distal colon cancers (neoplasms in the descending [153.2] and sigmoid [153.3] colon). Rectal cancer (ICD-9 codes 154.0 or 154.1) was defined as that occurring in the rectosigmoid or rectum.²² Deaths (>98%) were identified from state vital statistics records, the National Death Index,²³ reported by the families, and the postal system. Cause of death was identified from death certificates or review of medical records.

Statistical analyses

We calculated person-time for each participant from the date of baseline questionnaire return to the date of death, CRC diagnosis, or the end of follow-up (May 31, 2012 for the NHS; January 31, 2012 for the HPFS), whichever occurred first. To estimate the relative risks (RRs) and their 95% confidence intervals (95%CIs), we used a Cox proportional hazards regression model²⁴ based on a counting process data structure²⁵ for handling the left truncation and time-varying covariates (such as the cumulative updated nutrients); this Cox model was stratified simultaneously by age (in months) and year of questionnaire return (the every two year since baseline questionnaire), allowing for the finest possible control of confounding for age and secular trends. Additionally, we adjusted for established non-dietary and dietary risk factors in the multivariate models (see Table 2 for these factors and their categorizations). We further conducted sensitivity analyses in which we adjusted for intakes of fiber, phosphorous, and total fat. Additional adjustment of these factors did not materially alter the results and, thus, not included in the final model. Given that dairy foods including milk and yogurt are major food sources for calcium,²⁶ we conducted analyses controlling for them. We observed no violation of the proportional hazard assumption based on the likelihood ratio test.

Table 2.

Multivariable^* relative risk of colorectal cancer according to total calcium intake in the Nurses’ Health Study (1980–2012) and Health Professionals Follow-up Study (1986–2012)

	Total calcium intake (mg/d)						P-trend
	<600	600–<800	800–<1000	1000–<1200	1200–<1400	≥1400
Colorectal cancer
NHS
No.cases (n=1,828)	313	459	397	306	187	166
Age-adjusted	1 (ref)	0.96 (0.83, 1.11)	0.81 (0.69, 0.94)	0.75 (0.63, 0.88)	0.67 (0.56, 0.81)	0.57 (0.47, 0.69)	<0.001
MVRR	1 (ref)	1.00 (0.86, 1.16)	0.87 (0.74, 1.02)	0.85 (0.71, 1.01)	0.80 (0.65, 0.97)	0.72 (0.58, 0.90)	0.01
HPFS
No.cases (n=1,250)	201	358	270	187	112	122
Age-adjusted	1 (ref)	0.91 (0.76, 1.09)	0.78 (0.65, 0.94)	0.83 (0.67, 1.01)	0.79 (0.63, 1.00)	0.74 (0.59, 0.92)	0.01
MVRR	1 (ref)	1.00 (0.84, 1.20)	0.93 (0.76, 1.13)	1.04 (0.84, 1.28)	1.01 (0.79, 1.29)	0.98 (0.76, 1.25)	0.89
Pooled
No.cases (n=3,078)	514	817	667	493	299	288
Age-adjusted	1 (ref)	0.94 (0.84, 1.05)	0.80 (0.71, 0.89)	0.78 (0.68, 0.88)	0.72 (0.61, 0.84)	0.64 (0.50, 0.83)	0.01
MVRR	1 (ref)	1.00 (0.89, 1.12)	0.89 (0.79, 1.01)	0.93 (0.76, 1.13)	0.89 (0.70, 1.12)	0.83 (0.62, 1.12) **	0.36
Colon cancer
NHS
No.cases (n=1,427)	236	361	308	243	144	135
MVRR	1 (ref)	1.02 (0.86, 1.21)	0.86 (0.71, 1.04)	0.85 (0.70, 1.05)	0.79 (0.62, 1.00)	0.75 (0.59, 0.97)	0.02
HPFS
No.cases (n=996)	162	292	212	152	94	84
MVRR	1 (ref)	1.00 (0.82, 1.22)	0.89 (0.71, 1.10)	1.02 (0.81, 1.30)	1.04 (0.79, 1.36)	0.82 (0.62, 1.09)	0.43
Pooled
No.cases(n=2,423)	398	653	530	395	238	219
MVRR	1 (ref)	1.01 (0.89, 1.15)	0.87 (0.76, 1.00)	0.92 (0.78, 1.10)	0.90 (0.68, 1.17)	0.78 (0.65, 0.95) **	0.05
Proximal colon cancer
NHS
No.cases (n=856)	114	201	201	155	92	93
MVRR	1 (ref)	1.13 (0.89, 1.44)	1.12 (0.87, 1.44)	1.08 (0.82, 1.42)	0.97 (0.71, 1.33)	1.00 (0.72, 1.38)	0.89
HPFS
No.cases (n=426)	59	129	99	62	43	34
MVRR	1 (ref)	1.14 (0.83, 1.57)	1.05 (0.74, 1.48)	1.00 (0.68, 1.47)	1.16 (0.76, 1.77)	0.83 (0.53, 1.31)	0.47
Pooled
No.cases(n=1,282)	173	330	300	217	135	127
MVRR	1 (ref)	1.14 (0.94, 1.37)	1.09 (0.89, 1.34)	1.05 (0.84, 1.32)	1.03 (0.80, 1.33)	0.94 (0.72, 1.22)	0.57
Distal colon cancer
NHS
No.cases (n=528)	112	147	103	80	48	38
MVRR	1 (ref)	0.92 (0.71, 1.19)	0.65 (0.48, 0.87)	0.65 (0.47, 0.91)	0.65 (0.44, 0.95)	0.53 (0.35, 0.82)	<0.001
HPFS
No.cases (n=364)	70	107	67	54	33	33
MVRR	1 (ref)	0.91 (0.66, 1.24)	0.70 (0.49, 1.01)	1.00 (0.68, 1.47)	0.97 (0.63, 1.51)	0.81 (0.52, 1.28)	0.79
Pooled
No.cases(n=892)	182	254	170	134	81	71
MVRR	1 (ref)	0.91 (0.75, 1.11)	0.67 (0.53, 0.84)	0.80 (0.52, 1.22)	0.78 (0.52, 1.17)	0.65 (0.43, 0.99) **	0.25
Rectal cancer
NHS
No.cases (n=401)	77	98	89	63	43	31
MVRR	1 (ref)	0.90 (0.66, 1.23)	0.90 (0.64, 1.25)	0.84 (0.58, 1.24)	0.94 (0.61, 1.45)	0.75 (0.46, 1.23)	0.31
HPFS
No.cases (n=254)	39	66	58	35	18	38
MVRR	1 (ref)	1.02 (0.67, 1.53)	1.12 (0.73, 1.72)	1.10 (0.67, 1.79)	0.90 (0.50, 1.63)	1.68 (1.02, 2.78)	0.06
Pooled
No.cases(n=655)	116	164	147	98	61	69
MVRR	1 (ref)	0.94 (0.74, 1.21)	0.97 (0.75, 1.27)	0.93 (0.69, 1.26)	0.93 (0.65, 1.31)	1.12 (0.51, 2.48) **	0.79

^*Multivariable relative risks were adjusted for age (in month), race (Caucasian vs. non-Caucasian), adult BMI (< 25, 25 –< 27.5, 27.5 –< 30, or ≥ 30 kg/m²), smoking (0, 1–10, or > 10 pack-years), history of colorectal cancer in a parent or sibling (yes or no), history of sigmoidoscopy/colonoscopy (yes or no), physical activity (< 3, 3–< 27, ≥ 27 MET-hrs/wk), regular aspirin use (yes, no), alcohol consumption (0 –< 5, 5 –< 15, or ≥ 15 g/d), 25-hydroxyvitamin D (25(OH)D) scores (tertiles), current energy-adjusted total intake of, folate, red meat, processed meat (all in tertiles), and postmenopausal hormone use (premenopausal, never, past, or current user; women only).

^**P-values for heterogeneity by sex were 0.08 for colorectal cancer, 0.67 for colon cancer, 0.64 for proximal colon cancer, 0.19 for distal colon cancer, and 0.03 for rectal cancer.

We used energy-adjusted total calcium intake as the main exposure variable and conducted sub-analyses according to source of calcium (i.e., from all foods, dairy products only, or calcium supplements). To better represent long-term dietary intake and minimize random measurement errors in dietary assessments, we calculated the cumulative average intake of total calcium. Taking the HPFS as an example, the incidence of CRC from 1986 through 1990 was related to the total calcium intake from the 1986 questionnaire, and the incidence of CRC from 1990 through 1994 was related to the average intake of total calcium from the 1980, 1986, and 1990 questionnaires, and so forth. When appropriate, we used the cumulative average for those variables, including alcohol consumption, energy-adjusted total intake of, folate, red meat, processed meat. We evaluated the association between total calcium intake (<600, 600–<800, 800–<1000, 1000–<1200, 1200–<1400, ≥1400 mg/d) and risk of overall CRC and by anatomic subsite.

To examine the dose-response relationship, we used non-parametric restricted cubic splines,^{27, 28} stratifying by age, year of questionnaire return, and adjusted for the aforementioned CRC risk factors. To reduce the influence of outliers, we excluded participants with extremely high values (i.e., top 1% total calcium intake, the resulted maximum intake was 1,942 mg/d in women and 2,197 mg/d in men). To formally test the non-linear associations, we used a likelihood ratio test to compare the model including the linear and cubic spline terms selected by a stepwise regression procedure with the model including only the linear term for calcium.

Current literature generally suggests that calcium intake was associated with a lower risk of CRC.^{2, 3} Importantly, to ensure statistical power we used CRC for the time lagged analyses because the number of cases is reduced with increasing time-lag. To evaluate the timing of total calcium intake and CRC risk, we performed time-lagged analyses²⁹ to compare the difference in risk estimates for different latency periods. For example, in the NHS, for latency of 0–4 years, we used calcium intake in 1980 for CRC cases diagnosed from 1980 to 1984, calcium intake in 1984 for cases diagnosed from 1984 to 1986, and so forth. For the 4–8 year latency, follow-up started in 1984 and calcium intake in 1980 was used for cases diagnosed from 1984 to 1988. We then pooled results and tested for heterogeneity by sex using the Q statistic.^{30, 31}

Because calcium absorption is influenced by vitamin D, we examined the associations with total calcium intake (each 300 mg/day increase, equivalent to the calcium content in 250 mL of milk) by categories of total vitamin D intake and predicted 25-hydroxyvitamin D (25(OH)D) scores³² to evaluate the potential interactions between these two nutrients. Additionally, given that calcium inhibited the promotion of colon carcinogenesis in rats by cured meat,³³ we evaluated the interaction between calcium intake and processed meat, a risk factor for distal colon cancer in this population.³⁴ Lastly, as exploratory analyses, we examined whether the calcium-CRC association modified by factors including body mass index (BMI), family history of CRC, regular aspirin use. For these analyses, we used a Wald test to examine whether the beta coefficients of the cross-product term between calcium intake and these factors were statistically significant. We conducted all analyses using the SAS software (SAS Institute, Inc., Version 9.2, Cary, NC). All statistical analyses were two-sided with a P-value less than 0.05 indicating statistical significance.

Patient involvement

No patients were involved in setting the research question or the outcome measures, nor were they involved in the design and implementation of the study. There are no plans to involve patients in dissemination.

Results

We documented 3,078 incident CRC cases (1,828 in women and 1,250 in men) during up to 32 years of follow-up. Total calcium intake generally increased over time in this study (Supplementary Table 1). In addition, we calculated the Spearman correlation between total calcium intakes at different follow-up times. For NHS, the correlations were 0.37 for intake between 1986 and 1998, 0.23 between 1986 and 2010, and 0.37 between 1998 and 2010. Correlations for HPFS for the same comparisons were 0.43, 0.29, 0.39, respectively. With regard to major sources of total calcium intake, in women, dairy products contributed to 30–35% (relatively stable over follow-up), calcium supplement use contributed to about 25% (in early 1990s) to 35% (in 2000s), and multivitamin use contributed to about 3% (in early 1990s) to 8% (in 2000s). Similar to women, in men dairy products contributed to 34–45%, and multivitamin use contributed to about 3% (in early 1990s) to 8% (in 2000s). However, calcium supplement use was less prevalent but more recent, and contributed to about 3% (in early 1990s) to 8% (in 2000s) of total calcium intake. Individuals with higher total calcium intake tended to smoke less, have higher intake of total folate, total vitamin D, total phosphorous, and dietary fiber, and used multivitamins more frequently, but consumed less alcohol, red meat, processed meat, or total fat. In men, individuals with higher total calcium intake tended to use more aspirin and were more likely to have had a sigmoidoscopy/colonoscopy (Table 1).

Table 1.

Baseline characteristics of participants by frequency of total calcium intake in the Nurses’ Health Study (1980) and Health Professionals Follow-up Study (1986)

	Total calcium intake (mg/d)
	<600	600–<800	800–<1000	1000–<1200	1200–<1400	≥ 1400
Women – Nurses’ Health Study
No. (%)	34,138(39%)	24,290(27%)	14,733(17%)	8,326(9%)	3,998(4.5%)	3,024(3.5%)
Age, years*	46.5(7.0)	46.8(7.2)	46.8(7.3)	46.7(7.4)	47.1(7.5)	47.1(7.4)
White, %	96.4	98.0	98.3	98.4	98.5	97.7
Body mass index, kg/m2	24.0(4.2)	24.0(4.2)	24.1(4.1)	24.2(4.1)	24.5(4.4)	24.3(4.3)
Activity, METs-hrs/week	12.5(18.0)	14.2(19.7)	15.2(21.5)	15.3(21.0)	15.4(24.1)	17.3(29.5)
Family history of colorectal cancer, %	7.9	7.8	7.9	7.9	7.6	7.9
Regular aspirin use,** %	33.1	33.4	32.7	31.5	31.6	29.1
Past smoking, %	31.2	34.4	34.7	34.5	34.5	31.7
Current smoking, %	23.3	19.9	18.8	18.3	18.3	19.4
Multivitamin use, %	28.5	33.8	37.9	40.6	41.5	49.6
History of sigmoidoscopy/colonoscopy, %	9.9	10.0	10.0	10.6	10.9	9.6
Postmenopausal status, %	45.0	44.0	44.2	43.9	44.0	45.8
PMH use, %	18.3	18.5	18.8	19.3	18.6	20.1
Total energy intake, kcal/d	1573(513)	1546(484)	1565(518)	1602(481)	1575(470)	1560(530)
Dietary calcium intake, mg/d	456(97)	690(60)	882(71)	1078(87)	1264(115)	1523(324)
Supplemental calcium intake, mg/d	1(11)	3(24)	7(43)	14(67)	23(102)	142(442)
Alcohol, g/day	7.8(12.5)	6.2(9.6)	5.5(8.8)	4.8(8.1)	4.2(7.6)	3.4(6.7)
Total folate intake, ug/d	311(228)	365(236)	399(254)	417(262)	437(284)	589(686)
Total vitamin D, IU/d	238(227)	309(238)	377(252)	451(268)	516(268)	724(659)
Red meat, servings/wk	3.2(2.3)	2.4(1.8)	2.1(1.7)	1.9(1.5)	1.7(1.4)	1.3(1.3)
Processed meat, servings/wk	1.3(1.9)	1.2(1.8)	1.0(1.6)	0.9(1.6)	0.7(1.3)	0.6(1.2)
Total phosphorous, mg/d	962.2(156)	1125(155)	1262(164)	1404(167)	1541(178)	1764(256)
Total fat, g/d	73.6(14.3)	69.9(12.6)	66.9(12.8)	64.9(12.8)	63.1(13.2)	60.2(13.8)
Total fiber, g/d	15.4(5.6)	17.4(6.2)	18.1(6.9)	17.9(6.9)	17.7(7.2)	17.4(7.6)
Men in Health Professionals Follow-up Study
No. (%)	10,817(23%)	13,820(29%)	9,049(19%)	5,328(11%)	3,737(8%)	4,989(10%)
Age, years*	54.2(9.6)	54.2(9.8)	54.8(9.9)	54.8(9.9)	55.6(10.0)	56.4(9.7)
White, %	93.4	95.9	96.6	97.0	97.3	97.1
Body mass index, kg/m2	25.6(3.4)	25.6(3.3)	25.5(3.2)	25.4(3.3)	25.4(3.4)	25.4(3.3)
Activity, METs-hrs/week	18.2(27.2)	20.7(28.4)	22.1(31.6)	21.5(31.1)	22.2(31.3)	22.6(29.4)
Family history of colorectal cancer, %	8.8	8.4	8.3	8.6	8.5	8.7
Regular aspirin use,** %	27.0	28.9	30.2	31.0	31.1	31.5
Past smoking, %	43.7	43.1	41.0	40.8	40.0	39.9
Current smoking, %	12.4	9.7	8.2	8.9	8.8	7.5
Multivitamin use, %	50.8	57.4	62.1	67.3	69.8	77.4
History of sigmoidoscopy/colonoscopy, %	24.2	26.4	27.0	27.0	26.6	27.8
Total energy intake, kcal/d	1958(639)	1996(606)	1956(632)	2109(633)	2048(568)	1891(587)
Dietary calcium intake, mg/d	499(76)	683(77)	845(114)	982(189)	1110(260)	1232(466)
Supplemental calcium intake, mg/d	7(21)	21(55)	52(103)	118(180)	184(255)	603(636)
Alcohol, g/day	15.5(19.3)	11.8(15.0)	9.6(13.4)	9.9(14.2)	8.8(12.3)	7.8(12.0)
Total folate intake, ug/d	382(211)	446(227)	498.1(252)	530(290)	549(304)	661(397)
Total vitamin D, IU/d	273(242)	339(253)	408(279)	489(292)	547(301)	706(403)
Red meat, servings/wk	2.2(1.9)	1.9(1.6)	1.6(1.5)	1.7(1.5)	1.6(1.4)	1.3(1.3)
Processed meat, servings/wk	1.4(2.0)	1.3(1.8)	1.1(1.8)	1.2(1.9)	1.1(1.8)	0.9(1.6)
Total phosphorous, mg/d	1181(168)	1316(156)	1427(169)	1516(187)	1606(214)	1733(326)
Total fat, g/d	73.5(14.6)	72.2(13.3)	70.1(13.7)	70.6(13.8)	69.3(13.7)	68.0(15.0)
Total fiber, g/d	19.1(6.4)	21.2(6.5)	22.3(7.1)	21.6(7.6)	21.6(7.7)	22.0(8.1)

^*Values are means (SD) or percentages and except for the data on age and the percentage who were users, all data were standardized to the age distribution of the study population.

^**Regular aspirin use was defined as 2 or more tablets/wk.

Total calcium intake (≥ 1400 vs. < 600 mg/d) was associated with a statistically significant lower risk of colon cancer (0.78, 0.65–0.95), and the association appeared stronger for distal colon cancer (0.65, 0.43–0.99) than for proximal colon cancer (0.94, 0.72–1.22, P-_{common effects}=0.14) (Table 2). The results remained essentially the same after further adjustment for total dairy foods or major components such as milk or yogurt. For example, in women, the multivariable RR of total calcium intake changed from 0.75 (0.59–0.97) to 0.77 (0.61–0.98) after adjusting for total dairy foods. Similar results were observed by different sources of calcium (from all foods or dairy products only, Supplementary Table 2). Calcium supplement intake at a higher dose was suggestively associated with a lower risk (women only, Supplementary Table 3). Moreover, the association with CRC appeared to be linear within the intake range (P-value for curvature > 0.20), the multivariable RRs per a 300 mg/d increase were 0.92 (0.88–0.97) for CRC, 0.91 (0.87–0.97) for colon cancer, 0.97 (0.91–1.04) for proximal colon cancer, 0.83 (0.76–0.92) for distal colon cancer, and 0.94 (0.85–1.05) for rectal cancer in the NHS. The corresponding RRs were 1.00 (0.95–1.05), 0.98 (0.92–1.03), 0.97 (0.88–1.06), 0.98 (0.90–1.08), and 1.10 (1.00–1.21) in the HPFS, respectively.

We conducted latency analyses to further evaluate the timing of total calcium intake in relation to CRC risk. Overall, the inverse association for total calcium intake appeared only apparent for intakes at least 8–12 years before diagnosis, with the strongest association being observed for intakes 12–16 years before diagnosis, although inverse association was observed in women for 0–4 year lagged analysis (all P-value for heterogeneity by gender>0.05; Table 3).

Table 3.

Time lagged analyses on colorectal cancer according to total calcium intake in the Nurses’ Health Study (1980–2012) and Health Professionals Follow-up Study (1986–2012)

	Total calcium intake (mg/d)						P-trend
	<600	600–<800	800–<1000	1000–<1200	1200–<1400	≥1400
0–4 yr lag
NHS
No.cases (n=1,481)	287	277	185	173	152	407
	1 (ref)	0.93 (0.78, 1.10)	0.80 (0.66, 0.96)	0.81 (0.65, 0.97)	0.84 (0.68, 1.03)	0.72 (0.61, 0.86)	0.02
HPFS
No.cases (n=999)	157	262	207	116	94	163
	1 (ref)	1.09 (0.88, 1.33)	1.00 (0.81, 1.25)	0.86 (0.67, 1.11)	1.00 (0.77, 1.32)	0.98 (0.77, 1.24)	0.42
Pooled
No.cases (n=2,480)	444	539	392	289	246	570
	1 (ref)	0.99 (0.85, 1.16)	0.89 (0.71, 1.11)	0.82 (0.70, 0.96)	0.90 (0.75, 1.07)	0.83 (0.62, 1.11)**	0.32
4–8 year lag
NHS
No.cases (n=1,418)	255	301	202	177	126	357
	1 (ref)	1.16 (0.98, 1.37)	1.06 (0.87, 1.28)	1.03 (0.84, 1.26)	0.94 (0.75, 1.18)	0.97 (0.81, 1.16)	0.16
HPFS
No.cases (n=888)	184	221	173	107	84	119
	1 (ref)	0.81 (0.66, 0.99)	0.81 (0.65, 1.01)	0.79 (0.61, 1.01)	0.91 (0.69, 1.19)	0.78 (0.61, 1.00)	0.11
Pooled
No.cases (n=2,306)	439	522	375	284	210	476
	1 (ref)	0.97 (0.68, 1.38)	0.93 (0.72, 1.21)	0.91 (0.70, 1.19)	0.93 (0.78, 1.10)	0.89 (0.72, 1.09)**	0.04
8–12 year lag
NHS
No.cases (n=1,236)	302	264	161	160	111	238
	1 (ref)	0.89 (0.75, 1.06)	0.77 (0.64, 0.94)	0.89 (0.73, 1.09)	0.84 (0.67, 1.06)	0.78 (0.65, 0.95)	0.11
HPFS
No.cases (n=700)	146	204	133	91	47	79
	1 (ref)	0.99 (0.80, 1.23)	0.88 (0.69, 1.13)	0.99 (0.75, 1.30)	0.72 (0.51, 1.02)	0.84 (0.63, 1.13)	0.09
Pooled
No.cases (n=1,936)	448	468	294	251	158	317
	1 (ref)	0.93 (0.81, 1.06)	0.81 (0.70, 0.95)	0.92 (0.78, 1.08)	0.80 (0.66, 0.97)	0.80 (0.68, 0.94)**	0.03
12–16 year lag
NHS
No.cases (n=1,046)	224	215	139	147	98	223
	1 (ref)	0.89 (0.74, 1.08)	0.79 (0.63, 0.98)	0.89 (0.72, 1.11)	0.80 (0.62, 1.02)	0.74 (0.60, 0.91)	0.05
HPFS
No.cases (n=482)	118	144	100	43	29	48
	1 (ref)	0.93 (0.72, 1.19)	0.94 (0.71, 1.24)	0.68 (0.47, 0.98)	0.66 (0.43, 1.01)	0.83 (0.58, 1.19)	0.14
Pooled
No.cases (n=1,528)	342	359	239	190	127	271
	1 (ref)	0.90 (0.75, 1.05)	0.84 (0.71, 1.00)	0.81 (0.63, 1.05)	0.76 (0.61, 0.94)	0.76 (0.64, 0.91)**	0.02

^*Multivariable relative risks were adjusted for age (in month), race (Caucasian vs. non-Caucasian), adult BMI (< 25, 25 –< 27.5, 27.5 –< 30, or ≥ 30 kg/m²), smoking (0, 1–10, or > 10 pack-years), history of colorectal cancer in a parent or sibling (yes or no), history of sigmoidoscopy/colonoscopy (yes or no), physical activity (< 3, 3–< 27, ≥ 27 MET-hrs/wk), regular aspirin use (yes, no), alcohol consumption (0 –< 5, 5 –< 15, or ≥ 15 g/d), 25-hydroxyvitamin D (25(OH)D) scores (tertiles), current energy-adjusted total intake of folate, red meat, processed meat (all in tertiles), and postmenopausal hormone use (premenopausal, never, past, or current user; women only).

^**P-values for heterogeneity by sex were 0.07 for 0–4 yr lag, 0.56 for 4–8 yr lag, 0.55 for 8–12 yr lag, and 0.67 for 12–16 yr lag.

Because of the observed strongest association with distal colon cancer, we further evaluated whether this association was modified by total vitamin D intake, predicted 25(OH)D score, processed meat intake, BMI, and other factors. Although all P-values for interactions ≥ 0.10 (Table 4), higher calcium intake (per a 300 mg/d increase) appeared to be associated with a lower risk among individuals with low total vitamin D intake (0.76, 0.64–0.92), with high processed red meat intake (0.87, 0.77–0.99), or who were obese (0.68, 0.59–0.83), smokers (0.81, 0.69–0.95) or physically inactive (0.82, 0.70–0.96). We conducted additional interaction analyses on overall CRC and observed similar patterns (data not shown).

Table 4.

Multivariable^* relative risk of distal colon cancer according to total calcium intake (every 300 mg/day increase) and other lifestyle and nutritional factors in the Nurses’ Health Study (1980–2012) and Health Professionals Follow-up Study (1986–2012)

	NHS		HPFS		Pooled results
Factors	No. cases	RR	No. cases	RR	No. cases	RR	P-interaction
BMI (kg/m2)							0.15
<25	264	0.80 (0.70, 0.92)	117	1.01 (0.86, 1.20)	381	0.90 (0.71, 1.13)
25–<30	166	0.98 (0.83, 1.15)	201	1.00 (0.89, 1.14)	367	1.00 (0.90, 1.10)
30+	98	0.66 (0.52, 0.85)	46	0.72 (0.49, 1.05)	144	0.68 (0.55, 0.83)
Family history of colorectal cancer							0.54
No	433	0.81 (0.73, 0.90)	306	0.98 (0.89, 1.09)	739	0.89 (0.74, 1.08)
Yes	95	0.91 (0.72, 1.14)	58	0.98 (0.76, 1.27)	153	0.94 (0.79, 1.12)
Aspirin use
No	339	0.86 (0.76, 0.97)	228	1.00 (0.89, 1.12)	567	0.93 (0.80, 1.08)
Yes	189	0.77 (0.66, 0.90)	136	0.94 (0.79, 1.11)	325	0.85 (0.70, 1.03)
Smoking							0.92
Never	243	0.89 (0.78, 1.02)	157	1.03 (0.90, 1.17)	400	0.96 (0.84, 1.10)
>–<10 pack year	73	0.85 (0.65, 1.10)	28	1.16 (0.85, 1.58)	101	0.98 (0.72, 1.32)
10+ pack year	212	0.75 (0.64, 0.87)	179	0.88 (0.76, 1.02)	391	0.81 (0.69, 0.95)
Physical activity (METs/wk)							0.49
<3	183	0.81 (0.68, 0.96)	30	0.93 (0.62, 1.38)	213	0.82 (0.70, 0.96)
3–17	289	0.83 (0.73, 0.94)	208	0.95 (0.84, 1.08)	497	0.89 (0.78, 1.02)
27+	56	0.94 (0.70, 1.26)	126	1.07 (0.92, 1.25)	182	1.04 (0.91, 1.19)
Alcohol intake (g/day)							0.94
0–<5	348	0.83 (0.74, 0.93)	148	1.04 (0.91, 1.19)	496	0.93 (0.74, 1.16)
5–<15	109	0.64 (0.51, 0.80)	94	1.05 (0.87, 1.27)	203	0.82 (0.50, 1.34)
15+	71	1.04 (0.79, 1.35)	122	0.77 (0.62, 0.96)	193	0.89 (0.67, 1.18)
Total fat intake							0.64
Tertile 1	154	0.87 (0.75, 1.02)	133	0.82 (0.68, 0.99)	287	0.85 (0.76, 0.96)
Tertile 2	191	0.72 (0.60, 0.85)	108	1.08 (0.91, 1.29)	299	0.88 (0.59, 1.32)
Tertile 3	183	0.84 (0.71, 1.01)	123	1.07 (0.93, 1.23)	306	0.95 (0.76, 1.20)
Total vitamin D intake							0.76
Tertile 1	187	0.74 (0.59, 0.93)	126	0.81 (0.60, 1.09)	313	0.76 (0.64, 0.92)
Tertile 2	184	0.82 (0.68, 0.98)	127	0.96 (0.79, 1.17)	311	0.88 (0.75, 1.03)
Tertile 3	157	0.88 (0.77, 1.02)	111	1.04 (0.91, 1.18)	268	0.96 (0.82, 1.12)
Vitamin D predicted score							0.57
Tertile 1	247	0.82 (0.71, 0.95)	151	0.92 (0.78, 1.09)	398	0.86 (0.77, 0.96)
Tertile 2	156	0.83 (0.69, 0.99)	121	0.90 (0.76, 1.07)	277	0.87 (0.77, 0.98)
Tertile 3	125	0.78 (0.64, 0.94)	92	1.08 (0.93, 1.26)	217	0.92 (0.67, 1.27)
Total phosphorous intake							0.96
Tertile 1	197	0.78 (0.63, 0.97)	109	0.99 (0.85, 1.15)	306	0.89 (0.71, 1.12)
Tertile 2	169	0.80 (0.65, 1.00)	121	0.96 (0.80, 1.16)	290	0.89 (0.74, 1.06)
Tertile 3	165	0.80 (0.67, 0.95)	134	0.98 (0.83, 1.16)	299	0.89 (0.72, 1.09)
Total processed meat intake							0.16
Tertile 1	153	0.86 (0.74, 1.01)	104	1.07 (0.93, 1.23)	257	0.96 (0.78, 1.19)
Tertile 2	177	0.78 (0.66, 0.93)	116	0.94 (0.78, 1.13)	293	0.86 (0.71, 1.02)
Tertile 3	198	0.84 (0.71, 1.00)	144	0.91 (0.76, 1.09)	342	0.87 (0.77, 0.99)
Total fiber intake							0.61
Tertile 1	158	0.88 (0.74, 1.05)	144	0.87 (0.69, 1.10)	302	0.88 (0.76, 1.01)
Tertile 2	190	0.71 (0.59, 0.85)	116	1.19 (0.99, 1.43)	306	0.92 (0.55, 1.52)
Tertile 3	180	0.87 (0.74, 1.02)	104	0.94 (0.80, 1.10)	284	0.90 (0.81, 1.01)

^*Multivariable relative risks were adjusted for age (in month), race (Caucasian vs. non-Caucasian), adult BMI (< 25, 25 –< 27.5, 27.5 –< 30, or ≥ 30 kg/m²), smoking (0, 1–10, or > 10 pack-years), history of colorectal cancer in a parent or sibling (yes or no), history of sigmoidoscopy/colonoscopy (yes or no), physical activity (< 3, 3–< 27, ≥ 27 MET-hrs/wk), regular aspirin use (yes, no), alcohol consumption (0 –< 5, 5 –<15, or ≥ 15 g/d), 25-hydroxyvitamin D (25(OH)D) scores (tertiles), current energy-adjusted total intake of folate, red meat, processed meat (all in tertiles), and postmenopausal hormone use (premenopausal, never, past, or current user; women only). Of note, variables examined in this table were not adjusted for.

In the NHS, the median values were 51.9 for tertile 1, 60.8 for tertile 2, and 71.0 for tertile 3 for total fat intake (g/d), the corresponding values were 962.0, 1134.7, and 1345.0 for total phosphorous intake (mg/d), 158.9, 306.3, and 543.7 for total vitamin D (IU/d), 0, 0.1, and 0.2 for processed meat (servings/d), and 9, 13.6, and 17.4 for total fiber intake (g/d).

In the HPFS, the median values were 57.7 for tertile 1, 69.8 for tertile 2, and 81.4 for tertile 3 for total fat intake (g/d), the corresponding values were 1212.5, 1406.0, and 1636.0 for total phosphorous intake (mg/d), 194.9, 371.3, and 661.6 for total vitamin D (IU/d), 0, 0.1, and 0.3 for processed meat (servings/d), and 16.0, 21.2, and 27.6 for total fiber intake (g/d).

Discussion

Our study found that calcium intake was linearly associated with a significant lower risk of colon cancer, primarily distal colon cancer, and mainly apparent for intakes about 10 years before diagnosis. The inverse association between higher calcium intake and CRC risk is consistent with the majority of the previous studies^{2, 3}. Consistent with our previous report¹⁸ and some, but not all, previous studies,^{2, 3} we found that higher calcium intake was significantly associated with risk of distal colon cancer but not proximal colon cancer or rectal cancer. Although none of the associations were significantly different by sex (P-value for heterogeneity by sex = 0.19), the significant inverse association for distal colon cancer appeared to be primarily driven by results in women. In addition, the only statistically significant association observed in men was the positive association seen for rectal cancer but based on small number of cases (n = 38). This observation of stronger inverse association with distal colon cancer might be due to chance. Alternatively, it is biologically plausible. First, calcium has been hypothesized for decades to bind to free ionized fatty acids and bile acids in the lumen of the bowel to neutralize their toxicity effects.³⁵ The high pH observed in the distal colon, but not in the proximal colon increases the ionization of fatty acids and free un-conjugated bile acids by about fivefold.³⁶ Thus, higher calcium intake could have a stronger effect in preventing distal colon cancer because of the higher concentrations of fatty acids and bile acids.³⁶ Second, although the inverse association was not statistically modified by other factors, we observed some hints that inverse association appeared stronger among individuals with high processed meat intake, a risk factor for distal (but not for proximal) colon cancer in our cohorts.³⁴ Nonetheless, more studies with larger sample sizes are warranted to clarify any difference in calcium intake by tumor location.

The linear (inverse) association between total calcium intake and risk of CRC within the studied intake range of total calcium (99^th percentile: 1,942 mg/d in women, 2,197 mg/d in men) is consistent with a recent meta-analysis of 15 studies that reported that the risk of CRC continued to decrease within the observed range of 250–1,900 mg/d of total calcium intake.¹⁴ Additionally, we found that the inverse association was generally observed for total calcium, dietary calcium, dairy calcium and calcium supplements (in women), which was consistent with majority of previous studies.^{14, 37} In contrast to women’s stable and long-term calcium supplement use, their use in men tended to be much less prevalent and more recent, which might partly explain the weaker association for total calcium in men. According to National Health and Nutrition Examination Survey,²⁶ at least 35% of US adults use dietary supplements containing calcium but median total calcium intake was 650 mg/d for non-calcium supplement users and 1,000 mg/d for calcium-supplement users for adults over 50 years old. Given that calcium is a simple, modifiable, inexpensive agent and the benefit of calcium intake on CRC is expected to continue beyond 1,000 mg/d, both calcium supplement users and non-users may further reduce their CRC risk through additional calcium intake. This is particularly true for populations with relatively low calcium intake.

Although there was a suggestive trend for total calcium intakes 4–8 years before CRC diagnosis, the overall pattern of the entire time lagged analyses did suggest that the overall inverse association was apparent for calcium intakes at least 8–12 years before diagnosis with the strongest association being observed for intakes 12–16 years before diagnosis. This finding suggested that the latency for CRC associated with calcium intake is about 10 years, which is similar to the progression time of adenomas that develop into cancers. Because most sporadic CRC develops from the adenoma-carcinoma sequence, our findings not only add another line of evidence to support the notion that calcium is a promising chemopreventive agent for CRC but also suggest that higher calcium intake may reduce early development of carcinogenesis. Laboratory studies have shown that extracellular calcium may reduce proliferation of human colonic cells and promote cell differentiation by activating various signaling pathways.⁴ Malfunction of one of the pathways, APC/WNT/CTNNB1(β-catenin), is an early and common event in CRC pathogenesis.^38–42 Studies have reported that exposure of human colon tumor cells⁴³ and normal colon cells⁴⁴ to extracellular calcium suppresses CTNNB1 transcriptional activation, suggesting a beneficial role of calcium in the early development of CRC.

The strengths of our study include its prospective design and high follow-up rate, a large number of cases, and the inclusion of 2 independent cohorts of women and men. The detailed and repeated measurement of calcium intake beginning in the 1980s allowed us to study different latency for up to 3 decades. The detailed assessments of CRC risk factors enabled the evaluation of the independent association with calcium intake. Our study has some limitations. First, because of the observational nature of this study, we cannot rule out the possibility of residual confounding. However, the results were robust after adjustment for known CRC risk factors, total dairy consumption, or intake of commonly consumed dairy foods (e.g., milk, yogurt). Second, we had limited power for subgroup analyses. Lastly, relatively low calcium intake from non-dairy sources (i.e., <15%) in this population limited our ability to test its potential effect on CRC risk.

In conclusion, we found that the higher calcium intake was associated with a lower risk of developing colon cancer. The inverse association was linear and more pronounced for distal colon cancer. Additionally, overall inverse association for total calcium intake appeared apparent for intakes approximate 10 years before diagnosis. Considering the potential concern that diets high in calcium may have adverse consequences, such as an increased risk of prostate cancer,⁴⁵ future study is warranted to identify population subgroups that may most likely to benefit from higher calcium intake in the context of CRC primary prevention.

Impact and Novelty.

We found that higher calcium intake was significantly associated with a lower risk of developing colon cancer, especially for distal colon cancer. The overall inverse association was linear and did not differ by source of intake (from all foods or dairy products). Additionally, we found that calcium intake approximately 10 years before diagnosis appeared to be associated with a lower risk of CRC. Results from this study may contribute to future dietary guidelines of calcium intake for CRC primary prevention.