Total calcium, dairy foods and risk of colorectal cancer: a prospective cohort study of younger US women

Abstract

Background

Although colorectal cancer (CRC) incidence is declining among adults aged ≥65 years, CRC incidence in younger adults has been rising. The protective role of calcium in colorectal carcinogenesis has been well established, but evidence is lacking on whether the association varies by age at diagnosis. We investigated the association between total calcium intake and risk of overall CRC and CRC before age 55 years.

Methods

In the Nurses’ Health Study II (1991–2015), 94 205 women aged 25–42 years at baseline were included in the analysis. Diet was assessed every 4 years through validated food frequency questionnaires. Multivariable-adjusted hazard ratios (HRs) and 95% CIs for CRC were estimated using the Cox proportional hazards model.

Results

We documented 349 incident CRC cases during 2 202 604 person-years of follow-up. Higher total calcium intake was associated with a reduced risk of CRC. Compared with those with <750 mg/day of total calcium intake, the HR of CRC was 0.61 (95% CI, 0.38–0.97) for those who consumed ≥1500 mg/day (P for trend = 0.01). The HR per 300 mg/day increase was 0.85 (95% CI, 0.76–0.95). There was a suggestive inverse association between total calcium intake and CRC before age 55 years (HR per 300 mg/day increase, 0.87; 95% CI, 0.75–1.00), suggesting the importance of calcium intake in the younger population.

Conclusions

In a cohort of younger women, which reflects the birth cohorts, time periods and age ranges paralleling the recent rise in CRC, higher calcium intake was associated with a decreased risk of CRC.

Introduction

Colorectal cancer (CRC) is the third most common cancer among men and second among women worldwide.¹ In the USA, it is the third most common cancer and third leading cause of cancer deaths among both men and women.² In the recent decades, there was a decline in CRC incidence among US individuals aged ≥65 years by 3% annually.³ Although the overall CRC incidence in many countries including the USA is declining, partly due to increased screening, CRC incidence in younger adults has been rising; there has been a rise in CRC incidence in US individuals aged <50 years since the mid-1990s, and in those aged 50–64 years since 2011, consistently with a birth cohort effect beginning in the 1950s.^4,5 Patterns reflect an elevated CRC risk in populations born after 1950 to 1960.^5,6 It is projected that by 2030, almost 11% of colon cancers and 23% of rectal cancers will occur among US adults aged <50 years.⁷

The World Cancer Research Fund/American Institute for Cancer Research concluded that the evidence for an inverse causal association between calcium and CRC risk is probable (i.e. considered potentially actionable).⁸ A potential protective effect of calcium on CRC has been proposed by animal studies^9,10 and in vitro studies in human epithelial cells.¹¹ Almost all evidence forming the basis of recent conclusions was based on birth cohorts before 1950, when CRC rates had been dropping.^8,12 No study has evaluated the association between calcium intake and CRC diagnosed in more recent birth cohorts, over which CRC has been increasing. The causes for this increase remain largely unclear.

To better understand the reasons and contributors for the recent rise in CRC incidence among younger adults, we not only investigated whether calcium intake was associated with overall CRC risk, but also CRC risk before age 55 years, leveraging the Nurses’ Health Study (NHS) II. The NHS II cohort, for which women were born from 1947 to 1964, reflects the birth cohorts, time periods and age ranges over which the recent rise in CRC has been occurring.

Methods

Study population

The NHS II is an ongoing prospective cohort study of 116 429 female nurses aged 25–42 years that began in 1989. Participants are followed every 2 years by self-administered questionnaires on demographics, lifestyle factors and medical and other health-related information, which are complemented by assessments of dietary intake using validated semi-quantitative food frequency questionnaires (FFQs) every 4 years. This study was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required.

For the current analysis, we excluded women with inflammatory bowel disease, prior diagnosis of CRC, implausible data for baseline energy intake (<600 or >3500 kcal/d) or missing data for baseline calcium intake.

Assessment of total calcium intake

FFQs were mailed to the NHS II participants, starting in 1991. The items for dairy products included skimmed/low-fat milk, whole milk, ice cream, cream, sherbet, yogurt, cottage/ricotta cheese, cream cheese and other cheese. Butter intake was not included in the analysis because it is nutritionally distinct from other dairy foods.¹³ Total calcium intake was calculated by adding dietary calcium intake including calcium from fortified foods and supplemental calcium intake from multivitamins and calcium supplements. We further classified total dairy food intake into high-fat and low-fat dairy food intakes. High-fat dairy food intake included intakes of whole milk, ice cream, cream, cream cheese and other cheese. Low-fat dairy food intake included intakes of skimmed/low-fat milk, sherbet, yogurt and cottage/ricotta cheese.

The validity and reproducibility of the FFQ have been described in detail elsewhere.^14–17 The correlation coefficients between four 1-week dietary records and the 1980 FFQ from the NHS I, a similar prospective cohort, were reasonably high at 0.56 for energy-adjusted total calcium and 0.51 for dietary calcium.¹⁸ For dairy foods specifically, the correlation coefficients comparing intakes based on four 1-week dietary records collected over a 1-year period to a 61-item FFQ were 0.57 for hard cheese and 0.94 for yogurt.¹⁶

Assessment of covariates

We calculated body mass index (BMI) based on weight and height reported at baseline and weight updated every 2 years. We assessed total energy, red and processed meat, dietary fibre, vitamin D, folate and alcohol intakes from the FFQ. We also assessed overall diet quality using the Alternative Healthy Eating Index 2010 (without alcohol intake), where a higher score has been associated with reduced risks of cardiovascular disease, diabetes mellitus^19,20 and CRC.²¹ Smoking habits were updated biennially for calculation of pack-years smoked. Physical activity reported on the baseline and biennial follow-up questionnaires was validated previously.²² A metabolic equivalent of task score (METS) based on energy expenditure was assigned to each type of physical activity and the total physical activity was calculated by multiplying the METS by the mean time spent in each activity and then summing across all activities. The NHS II participants also regularly provided updated information on regular use of aspirin and other non-steroidal anti-inflammatory drugs, use of post-menopausal hormone therapy and family history of CRC among first-degree relatives.

Ascertainment of CRC

Our primary end point was incident CRC. We requested written permission to collect medical records and pathology reports from participants who reported having CRC on biennial questionnaires to confirm diagnosis. We also identified unreported, lethal CRC cases through family members, the postal system and the National Death Index. For deaths caused by previously unreported CRC, we requested consent for medical records acquisition from the next of kin. Physicians who were masked to exposure status reviewed medical records to verify CRC diagnosis and recorded histopathologic findings, anatomical location (proximal colon, distal colon or rectum) and other clinically relevant information.

Statistical analysis

We prospectively evaluated the association between total calcium intake and risk of CRC. We analysed calcium intake by creating categories of intakes based on 250-mg/day increments with collapsed categories if limited by case numbers. Person-time was measured from the return of the baseline dietary questionnaire in 1991 until the date of CRC diagnosis, death from any cause or the administrative end of follow-up (June 2015), whichever came first. We used energy-adjusted calcium intake, which was calculated from the regression-residual method to minimize variation due to energy intake and related measurement error.²³ To better capture long-term intake and reduce within-person variations, we used cumulative averages for dietary factors, including total calcium intake. Cox proportional hazard models were used to estimate age- and multivariable-adjusted hazard ratios (HRs) and 95% CIs. The Cox model was stratified by age and year of questionnaire return. Covariates were chosen a priori based on established and suspected risk factors for CRC.²⁴ The base model was adjusted for age. The first multivariable model was further adjusted for non-dietary risk factors for CRC that included White ethnicity, height, BMI, smoking pack-years, physical activity, alcohol intake, regular use of aspirin, non-steroidal anti-inflammatory drug use, menopausal status and post-menopausal hormone use, family history of CRC and history of lower endoscopy within the past 10 years. In the second multivariable model, we additionally adjusted for dietary factors including total energy intake, red/processed meat intake, dietary fibre intake, total folate intake and Alternative Healthy Eating Index 2010. In the third multivariable model, we additionally adjusted for total vitamin D intake, which may correlate with calcium intake. The counting process data structure was used to deal with time-varying covariates and left truncation. Test for trend was performed using the median of total calcium intake category as a continuous variable. We assessed potential nonlinear associations using restricted cubic splines.^25,26 We presented continuous calcium intake results with a 300-mg/day increment of calcium intake, which is approximately equivalent to the calcium content for one serving of milk (250 mL).¹²

We performed various stratified and sensitivity analyses. Because recent data suggest a rise in CRC patients younger than age 55 years,⁵ we examined calcium in relation to CRC cases under age 55 years in an additional analysis. We also performed analyses based on anatomical sites (proximal colon, distal colon or rectum)²⁷ and BMI. Test for interaction was performed by including a cross product of continuous calcium intake and potential effect modifier into the multivariable models and utilizing a Wald test to assess statistical significance of the interaction. Different sources of calcium intake (dietary calcium or supplemental calcium) were also evaluated. We assessed associations with high-fat dairy and low-fat dairy separately. We also investigated milk²⁸ (skimmed milk plus whole milk) and yogurt²⁹ separately as they are rich in calcium and previously studied in relation to reduced risk of colorectal neoplasia. In addition, to minimize the possibility that undiagnosed CRC may have contributed to recent changes in total calcium intake, we employed lag analysis in which we excluded the first 8 years of follow-up and added an 8-year lag period between the calcium assessment and follow-up period to address the possibility of changes in behaviours during the pre-clinical phase and evaluate the etiologically relevant period. For example, calcium intake in 1991 was used to estimate the CRC risk in 1999 and calcium intake in 1995 was used to estimate the CRC risk in 2003. We chose 8 years because our diet questionnaires are updated every 4 years and calcium intake ∼10 years before diagnosis was associated with a lower CRC risk.³⁰

All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC).

Results

Among the 94 205 women studied after baseline exclusions, we documented 349 incident CRC cases (median age at diagnosis: 53 years) during 2 202 604 person-years of follow-up from 1991 to 2015. The median calcium intake was 1059 mg/day. At baseline, those with higher calcium intake tended to be older, be White and have a lower BMI; were more likely to have a family history of CRC; were less likely to be pre-menopausal, have history of diabetes, smoke cigarettes and eat red and processed meat; consumed more dietary fibre, total folate and total vitamin D; used more aspirin and multivitamins; and were more likely to be physically active and have a healthy dietary pattern (Table 1).

In the age- and multivariable-adjusted analyses, higher total calcium intake was associated with a reduced CRC risk (Table 2). In the fully adjusted model, compared with women who consumed <750 mg/day of total calcium intake, the HR for those who consumed ≥1500 mg/day of total calcium intake was 0.61 (95% CI, 0.38–0.97) (P for trend = 0.01). There was no evidence of a nonlinear association (P for nonlinearity = 0.44). The HR per 300-mg/day increase was 0.85 (95% CI, 0.76–0.95). There was a moderate correlation between total calcium intake and total vitamin D intake (r = 0.62). HRs for other confounders in the fully adjusted model were reported. Most variables were based on a continuous measure and whereas they were generally in the expected direction, many did not reach statistical significance (Supplementary Table S1, available as Supplementary data at IJE online). These are established risk factors, but our sample size was modest and some factors might have weaker association for younger women.

During the follow-up, 216 women were diagnosed with CRC before age 55 years. We observed a suggestive inverse association between total calcium intake and CRC diagnosed at age <55 years (HR per 300-mg/day increase, 0.87; 95% CI, 0.75–1.00) (Table 3). Although we had limited number of cases (n = 111) of CRC before age 50 years, results were compatible with the overall results (HR per 300-mg/day increase, 0.86; 95% CI, 0.72–1.03, for the fully adjusted model without vitamin D and HR per 300-mg/day increase, 0.93; 95% CI, 0.77–1.13, for the fully adjusted model with vitamin D).

Analysis by tumour anatomical sites showed stronger inverse associations between total calcium intake and risks of rectal and distal colon cancers compared with the proximal colon cancer although the P for heterogeneity between colon vs rectal cancer was 0.29 (Supplementary Table S2, available as Supplementary data at IJE online). Stratified analysis by BMI suggested that the potential beneficial role of calcium is stronger for those with BMI <25 kg/m² compared with those with BMI ≥25 kg/m², but the P for interaction was 0.73 (Supplementary Table S3, available as Supplementary data at IJE online). We also evaluated differences in sources of calcium intake. A strong association was observed for dietary calcium, especially dairy calcium (HR comparing highest vs lowest tertile, 0.64, 95% CI, 0.47–0.88), but not for supplemental calcium (HR comparing highest vs lowest tertile, 0.91, 95% CI, 0.66–1.26) (Supplementary Table S4, available as Supplementary data at IJE online). Compared with high-fat dairy food intake (HR comparing highest vs lowest tertile, 0.94, 95% CI, 0.71–1.26), low-fat dairy food intake (HR comparing highest vs lowest tertile, 0.82, 95% CI, 0.60–1.12) seemed to be more strongly associated with a reduced CRC risk. Among specific dairy foods, milk and yogurt had suggestive inverse associations with CRC (Supplementary Table S4, available as Supplementary data at IJE online). The association between total calcium intake and colorectal cancer became slightly stronger when we introduced an 8-year lag period (Supplementary Table S5, available as Supplementary data at IJE online).

Discussion

In this cohort of younger women, higher total calcium intake was associated with a reduced risk of CRC. In addition, higher total calcium intake was suggestively associated with a reduced risk of CRC among participants aged <55 years. The NHS II cohort reflects the birth cohorts, time periods and age ranges, paralleling the recent rise in CRC.

There is evidence supporting an association between calcium intake and risk and mortality of CRC. A meta-analysis of observational studies on calcium intake and CRC risk reported that each 300-mg/day increase in total calcium intake was associated with an 8% reduced risk of CRC (summary RR, 0.92; 95% CI, 0.89–0.95) and each 300-mg/day increase in supplemental calcium was associated with a 9% reduced risk of CRC (summary RR, 0.91; 95% CI, 0.86–0.98).¹² In addition, calcium has been suggested to lower the risk of death among patients with CRC.³¹ A causal association is also supported by randomized–controlled trials of calcium supplementation and colorectal adenomas.³²

Although the evidence for an inverse association of calcium intake and CRC is strong, it is still unknown whether calcium intake is associated with CRC that occurs at younger ages, especially in more recent birth cohorts, which have experienced a rise in CRC incidence. Accumulating evidence suggests that calcium insufficiency or deficiency is not only encountered in the elderly but also in the younger population in Europe³³ and the USA.³⁴ In the National Health and Nutrition Examination Survey data, <10% of women aged >50 years achieved adequate dietary calcium intakes and only a minority of men and women met recommended total calcium intakes.³⁵ The recent increase in CRC incidence among young adults in the USA may partly be explained by the recent increase in calcium deficiency and insufficiency. Even if other factors are primarily causing the recent rise in CRC, inadequate calcium intake may be a contributing factor that accentuates the rise. In our study, total calcium intake was inversely associated with risk of CRC in younger women, with median age at CRC diagnosis of 53 years in contrast to 66 years in the general US population.³⁶ In addition, higher total calcium intake was marginally associated with a decreased CRC risk before age 55 years, underlining the potential role of calcium in CRC prevention among young adults. Vitamin D intake, which is highly correlated with calcium intake, was shown to be associated with early-onset CRC among individuals aged <50 years.³⁷ Our results showed that calcium was associated with CRC even after adjusting for vitamin D intake. However, more studies are warranted to explore the interactive effects of calcium and vitamin D in CRC among old and/or young adults.

Findings from epidemiologic studies of calcium intake and colorectal subsites have been inconclusive. In the NHS I and Health Professionals Follow-up Study, an inverse association was observed between calcium intake and distal colon cancer, but not proximal colon cancer.³⁸ In our study, we observed slightly strong inverse associations for calcium intake and risks of rectal and distal colon cancers compared with the proximal cancer. Previous research suggests that rectal cancer incidence in younger adults is rising faster than colon cancer,³⁹ which could have affected our results among young women aged 25–42 years at baseline. In addition, compared with other subtypes (e.g. ‘serrated pathway’), calcium intake has been inversely associated with subtype 1 (‘conventional pathway’), which is concentrated in the distal colon and rectum.⁴⁰ A recent study demonstrated that the efficacy of calcium against colorectal adenomas could differ by BMI, with less effect in those with a higher BMI.⁴¹ Our results suggested that those with BMI <25 kg/m² were likely to benefit slightly more from calcium (HR per 300-mg/day increase, 0.81; 95% CI, 0.69–0.94) than those with BMI ≥25 kg/m² (HR per 300-mg/day increase, 0.89; 95% CI, 0.76–1.04), but the P for interaction was 0.73.

Our result from the lagged analysis demonstrated that the potential effect of calcium became slightly stronger when taken years prior to diagnosis. A previous study found that the inverse association between calcium and CRC was apparent for intakes at least 8–12 years before diagnosis, suggesting that the latency for CRC related to calcium intake is ∼10 years.³⁰ Given that it also takes ∼10 years for a CRC precursor (e.g. adenoma) to develop into cancer, it could be hypothesized that calcium plays a role in early stages of CRC. Both observational data and randomized–controlled trials support that calcium is protective for adenoma.^32,42 A recent study on the NHS II found that higher adolescent dairy intake was associated with a lower risk of rectal and advanced adenomas later in life,⁴³ further highlighting the importance of calcium intake early in life.

A possible mechanism of calcium on colorectal carcinogenesis is that calcium binds secondary bile acids and ionized fatty acids (both of which have been hypothesized to promote epithelial colon cell proliferation⁴⁴) to form insoluble soaps in the lumen of the colon that could reduce the proliferative stimulus of these compounds on colon mucosa.⁴⁴ It is also suggested that calcium can directly decrease epithelial cell proliferation⁴⁵ and induce cell differentiation, possibly mediated through the calcium-sensing receptor.⁴⁶ In addition, in a randomized trial, low-fat dairy foods that are rich in calcium reduced proliferation and normalized differentiation of colonic epithelial cells.⁴⁷

The strengths of our study are the prospective design, the long-term follow-up of more than 20 years and repeated measurement of dietary and lifestyle factors. In addition, the younger age of our participants at enrolment (ages 25–42 years) enabled us to prospectively investigate young-onset CRC. To minimize the influence of measurement error, we used the cumulative average of the total calcium intake. We acknowledge several limitations in our study. Although our analyses controlled for established and suspected CRC risk factors, our results could be affected by residual and unmeasured confounding. In addition, our study population consisted of female nurses (mostly White), which could reduce the generalizability of the results, though for overall CRC, calcium has shown inverse associations across a spectrum of more diverse populations.⁴²

In conclusion, we found evidence that higher total calcium intake is associated with a decreased CRC risk in younger women. Our results support that avoiding low-calcium status may be beneficial for CRC prevention among young adults.

Ethics approval

This study was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required.

Data availability

Information including the procedures for obtaining and accessing data from the Nurses’ Health Study II is described here: www.nurseshealthstudy.org/researchers. Codes used for our analyses are available here: https://github.com/githubhskim/calcium_manuscript.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

The study was conceived and designed by H.K. and E.L.G. H.K. analysed and drafted the manuscript. J.H. checked the accuracy of the analysis. All authors revised the manuscript critically and approved the final version of the manuscript. E.L.G. is the guarantor for the paper. Jinhee Hur ([email protected]) and Edward Giovannucci ([email protected]) contributed equally to this study as co-corresponding authors. The corresponding authors attest that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Funding

This work was supported by the US National Institutes of Health grants (U01 CA176726, R21 CA230873 to K.W., R03 CA197879 to K.W., R21 CA222940 to K.W., R21 CA252962 to X.Z., R00 CA215314 to M.S.) and by the American Cancer Society Mentored Research Scholar Grant (RSG NEC-130476 to X.Z., MRSG-17–220-01-NEC to M.S.).

Acknowledgements

The authors would like to acknowledge the contribution to this study from central cancer registries supported through the Centers for Disease Control and Prevention’s National Program of Cancer Registries (NPCR) and/or the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) Program. Central registries may also be supported by state agencies, universities and cancer centres. Participating central cancer registries include the following: Alabama, Alaska, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, Florida, Georgia, Hawaii, Idaho, Indiana, Iowa, Kentucky, Louisiana, Massachusetts, Maine, Maryland, Michigan, Mississippi, Montana, Nebraska, Nevada, New Hampshire, New Jersey, New Mexico, New York, North Carolina, North Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Puerto Rico, Rhode Island, Seattle SEER Registry, South Carolina, Tennessee, Texas, Utah, Virginia, West Virginia, Wyoming.

Conflict of interest

None declared.

References

Author notes