Adherence to dietary recommendations and colorectal cancer risk: results from two prospective cohort studies

Abstract

Background

Understanding the association between diet and colorectal cancer (CRC) risk is essential to curbing the epidemic of this cancer. This study prospectively evaluated adherence to the Chinese Food Pagoda (CHFP), and two American Dietary Guidelines: the Alternative Healthy Eating Index (AHEI-2010) and the Dietary Approaches to Stop Hypertension (DASH) in association with CRC risk among Chinese adults living in urban Shanghai, China.

Methods

Participants included 60 161 men and 72 445 women aged 40–74, from two ongoing population-based prospective cohort studies. Associations between dietary guideline compliance scores and CRC risk were evaluated by Cox proportional hazard regression analyses, with age as time metric, and potential confounders were adjusted.

Results

We identified 1670 CRC incidence cases (691 males and 979 females) during an average 8.1 years of follow-up for men and 13.4 years for women. CHFP score was inversely associated with risk of CRC, with hazard ratios (HRs) (95% confidence intervals) of 0.88 (0.77, 1.00), 0.86 (0.75, 0.98) and 0.84 (0.73, 0.96) for the 2nd, 3rd and 4th quartiles versus 1st quartile, respectively (P_trend= 0.01). The inverse association appeared stronger for rectal cancer, individuals at younger age (< 50 years), with a lower BMI (<25 kg/m²) or without any metabolic conditions at baseline, although no multiplicative interactions were noted. No consistent association pattern was observed for the modified DASH score and the modified AHEI-2010.

Conclusions

Compliance with the Dietary Guidelines for Chinese was associated with reduced risk of CRC among Chinese adults. To maximize health impacts, dietary recommendations need to be tailored for specific populations.

Key Messages

• Compliance with the Dietary Guidelines for Chinese was associated with reduced risk of CRC, particularly rectal cancer risk among Chinese adults in urban Shanghai.

• The modified AHEI-2010 and the modified DASH compliance scores were not associated with CRC risk in the study population.

• To maximize health impacts, the study recommends that dietary recommendations need to be tailored for specific populations.

Introduction

Globally, colorectal cancer (CRC) is the third most common cancer in men and the second in women, with an estimated 1.8 million cases diagnosed in 2018.¹ The CRC incidence rate varies markedly around the world, and it has been changing over the past two decades in many countries. In the USA, CRC incidence rates have progressively declined over time, mainly attributable to an increasing use of screening to detect and remove adenomas before they become cancerous.² In contrast, CRC incidence rates have been increasing in most countries in Asia and Eastern Europe, where populations are growing and ageing and where people have adopted an increasingly Westernized lifestyle.³

Over the past few decades, many people in China have shifted their diet away from coarser grain and nutrient-rich legumes toward greater consumption of refined grains, animal-source foods (e.g. pork, processed meats, eggs, poultry, and dairy products), edible oils, fried foods, sugar and sugar-sweetened beverages and snack foods.⁴ They have also reduced physical movement with respect to occupational/domestic activities and transportation, as well as an increase in sitting time and TV viewing time.^5,6 The shift in dietary and physical activity patterns in China has been paralleled by an increased prevalence of overweight and obesity,⁷ which is likely to have had an impact on CRC incidence.⁸ For instance, in Shanghai from 1973 to 2010, CRC incidence rates (age-adjusted with the world standard) more than doubled, from 13.6 to 28.4 per 100 000 in men, and from 11.9 to 22.3 per 100 000 in women.⁹ Similarly in Hong Kong, CRC incidence more than doubled in men between 1983 and 2006, from 29.5 to 68.2 per 100 000, and rose by more than 50% in women, from 29.8 to 47.1 per 100 000.¹⁰ In 2011, an estimated 310 244 new CRC cases and 149 722 CRC deaths were diagnosed in China,¹¹ making it the fourth most common cancer and the fifth leading cause of cancer-related death.¹²

With an aim of improving cancer and chronic disease prevention in parallel, many dietary recommendations have been issued in the past two decades to guide Americans in making better choices regarding their food consumption so as to promote health and to reduce the risk of cancer and other chronic diseases.^13,14 Several diet scoring systems such as the Healthy Eating Index,¹⁵ the Alternative Healthy Eating Index-2010 (AHEI-2010)¹⁶ and the Dietary Approaches to Stop Hypertension (DASH) diet score¹⁷ have been developed. Their associations with mortality and risk of CRC have been evaluated by several cohort studies in which it was found that higher compliance levels with the AHEI-2010 and DASH diets were associated with a lower risk of developing CRC.^18–21

The Chinese Nutrition Society and Ministry of Health published the Dietary Guidelines for Chinese, namely the Chinese Food Pagoda (CHFP), in 2007. In addition to the recommendation to consume plenty of whole grains, vegetables, fruits, beans and bean products, the CHFP also recommends the consumption of dairy products, appropriate amounts of fish, poultry, eggs and lean meats, and limitation of fats and salt.^22,23 We have previously reported that greater compliance with CHFP is associated with lower mortality from all causes, cardiovascular disease and cancer in Chinese adults.²⁴ A higher healthy lifestyle index, constructed of five lifestyle factors based on smoking, alcohol use, diet (the CHFP), waist-hip ratio and exercise participation, was related to lower risk of CRC among Chinese men.²⁵ The extent to which compliance with the CHFP and two American dietary guidelines, the AHEI-2010 and DASH, is associated with CRC incidence remains unknown. Such information will provide insight into CRC prevention in China, where CRC is one of the leading causes of cancer death.¹²

Using data from the Shanghai Women’s Health Study (SWHS) and the Shanghai Men’s Health Study (SMHS), we comprehensively evaluated the associations between adherence to the CHFP, the modified AHEI-2010 and the modified DASH, and incidence of CRC.

Methods

Study population

Details of designs and methods for the SWHS and SMHS have been described previously.^26,27 Briefly, the SWHS enrolled 74 941 women aged 40–70 years from December 1996 to May 2000, and the SMHS recruited 61 480 men aged 40–74 years from January 2002 to September 2006, with respective response rates of 92.7% and 74.1%. The SWHS was conducted in seven urban communities, whereas the SMHS was conducted in eight typical neighbourhood communities in the Changning District of Shanghai.^26,27 Approvals for using human subjects for research were obtained from the institutional review boards of the Shanghai Cancer Institute and Vanderbilt University. All participants provided written informed consent.

For the current analysis, 1604 women in the SWHS who reported a previous cancer diagnosis at their baseline interviews were excluded. We also excluded 355 participants with unreasonably high caloric intakes (>3500 kcal/day for women and >4200 kcal/day for men) or low intakes (<500 kcal/day for women and <800 kcal/day for men).²⁸ We further excluded from the study cases diagnosed within the first 2 years after study enrollment and those who had less than two years follow-up, as well as the first two years of cohort observations, to minimize the influence of preclinical cancer-related dietary pattern changes (1844 participants). Participants who had incomplete food frequency questionnaires (FFQs) at baseline were also excluded (12 participants). The final sample size for analysis was 132 606 participants (72 445 women of the SWHS, and 60 161 men of the SMHS).

Dietary assessment

The SWHS and SMHS use similar, validated, semi-quantitative FFQs that were designed to obtain information on usual food consumption habits. The SWHS FFQ consists of 77 food items, whereas the SMHS includes 81 food items commonly consumed in urban Shanghai. The frequency (five levels: daily, weekly, monthly, yearly or never) and quantity [in liangs (1 liang = ∼50 g)] for each food item/group consumed during the previous 12 months was obtained through face-to-face interviews conducted at the baseline survey. A complete FFQ was obtained from 99.9% of participants.^26,27 Both the SWHS and SMHS FFQs have fairly high validity and reproducibility as compared with multiple 24-h dietary recalls.^29,30 The correlation coefficient for major food groups were 0.41–0.66 in the SWHS³⁰ and 0.35–0.72 in the SMHS.²⁹ Daily energy and nutrient intake were derived based on the 2002 Chinese Food Composition Table.³¹

Dietary recommendation adherence score

The CHFP score was created based on 10 components, including: (i) grains; (ii) vegetables; (iii) fruits; (iv) dairy products; (v) beans; (vi) meat and poultry; (vii) fish and shrimp; (viii) eggs; (ix) fats and oils; and (x) salt, and adopted the scoring method used in creating the US Healthy Eating Index 2005.^24,32 A total CHFP score ranged from 0 (the lowest adherence) to 45 (the highest adherence). The majority of grains consumed in our study population were white rice and few refined wheat products. Information on whole grains was not collected in the component grains because of very low consumptions in the SWHS and SMHS populations. The modified AHEI-2010 score was calculated based on eight of 11 components contained in the AHEI-2010: (i) vegetables; (ii) fruits; (iii) nuts and legumes; (iv) red and processed meat; (v) eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA); (vi) polyunsaturated fatty acids (PUFA); (vii) sodium; and (viii) alcohol. A total modified AHEI-2010 score ranged from 0 to 80. The modified DASH score was calculated based on seven of the eight following components contained in the DASH eating plan: (i) vegetables; (ii) fruits; (iii) dairy; (iv) meat, poultry, fish and eggs; (v) nuts, seeds and legumes; (vi) fats and oils; and (vii) sodium. A total modified DASH score ranged from 0 to 70. Details of recommended serving size and the scoring method were described in our previous publication.²⁴ Because of the lack of relevant data in our FFQs or in the Chinese Food Composition Table, as well as very low consumption levels, whole grains, sugar-sweetened beverages and trans-fat were not included in deriving the compliance scores for the modified AHEI-2010 and the modified DASH (Supplementary Table 1, available as Supplementary data at IJE online).

Colorectal cancer ascertainment

Participants in the SWHS and SMHS have been followed by annual record linkage to the Shanghai Cancer Registry (SCR) and Shanghai Vital Statistics Registry (SVSR), as well as by home visits taking place every 2 or 3 years. The SCR is a population-based cancer registry which has provided cancer incidence data to the WHO/IARC publication series ‘Cancer Incidence in Five Continents’ since 1972. Cancer registration is legally mandated in Shanghai, China. All hospitals are required to notify the SCR of all newly diagnosed cancer cases.²⁷ Case ascertainment by the SCR through this case reporting system is estimated to be at least 85% complete; the remaining cancer cases are identified by the SCR by obtaining cancer death notices from the SVSR. SCR data files are linked to the SWHS and the SMHS roster annually to identify newly diagnosed cancer cases. All possible matches are checked manually and are verified by home visits. Medical charts from the diagnostic hospitals are reviewed to verify the diagnosis and collect information on cancer characteristics and treatment.^26,27 The primary outcome of the current study was incident CRC, as defined by the 9th version of the International Classification of Diseases, codes 153 to 154. Follow-up information through 31 December 2014 was used in the current study.

Statistical analysis

Baseline characteristics were described as frequencies for categorical variables and median and interquartile range (IQR) for continuous variables. The differences across subgroups were compared using nonparametric analysis (including Mann--Whitney and Kruskal-Wallis tests) for continuous variables and chi square tests for categorical variables. The association between dietary recommendation adherence scores in quartiles and CRC incidence was analysed using Cox proportional hazard regression, with age as the time metric. The follow-up time was calculated beginning 2 years after the date of study enrolment (i.e. excluding the first 2 years of cohort observation) and ending at the date of death, loss to follow-up or 31 December 2014, whichever came first. Quartile distributions of the three dietary recommendation adherence scores were derived for each gender and used in our analysis. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated and adjusted for potential confounders, using the lowest quartile as the reference group.

Based on previous studies, the following covariates were considered as potential confounders: educational attainment levels (four levels: elementary or less, middle school, high school and professional/college or higher), income levels (four levels: low, lower middle, upper middle and high), lifetime cigarette smoking (three categories: never smoking, pack-years <20 and pack-years ≥20), alcohol consumption (drinks/day continuous; one drink equals 14.18 g of alcohol), regular use of multivitamin supplement (yes/no), family history of CRC (yes/no), body mass index (BMI) (kg/m² continuous), the presence of metabolic conditions [having one or more of the following four conditions: history of hypertension, heart diseases, diabetes mellitus and obesity (BMI of 30 or higher)], physical activity (metabolic equivalent task score per hour per week, quartiles) and total energy intake (kcal/day continuous). Models were stratified by gender and birth year in 5-year intervals. The proportional hazard assumption was evaluated using scaled Schoenfeld residuals plots and calculating the correlation between Schoenfeld residuals for each covariate and the ranking of time. No evidence of violation of assumption was found. Tests for trend were conducted by treating the median value of each quartile as a continuous variable.

We also performed analysis stratified by age at 50 years old, educational attainment level (low/high), BMI at 25 kg/m² and the presence of a metabolic condition (yes/no). Age 50 was chosen for stratification because this age is commonly used for CRC screening recommendations, and incidence of CRC under age 50 has been rising in many countries.^33,34 We chose BMI 25 kg/m² for stratification because Asians are more susceptible to insulin resistance at lower BMI levels.³⁵ The log-likelihood ratio test was used to assess multiplicative interaction between quartiles of CHFP score with age, BMI, educational attainment level and metabolic conditions by comparing the models with and without the cross-product term of these variables. Analyses were also carried out to evaluate the associations between CRC incidence and the three dietary recommendation adherence scores and each food component score, treating the latter as a continuous variable. The analyses for each food component score were further adjusted for a modified total compliance score that excluded the corresponding component. Cox proportional hazard restricted cubic spline models were used to explore possible deviations from non-linear associations, with five knots specified at the median of each quintile of dietary guideline compliance scores. Additional analyses, including all cohort observations, were performed to examine the three dietary recommendation adherence scores and CRC association. SAS software (version 9.4; SAS Institute, Cary, NC, USA) was used for all analyses.

Results

We identified 1670 incident CRC cases (691 male and 979 female cases) among 60 161 SMHS participants and 72 445 SWHS participants during average follow-up of 8.1 and 13.4 years, respectively. In both sexes, compared with participants who did not develop CRC, those who did develop CRC were approximately 9 years older and had lower educational attainment, but were similar regarding income level. Notably, those who developed CRC were more likely to be obese and to have a family history of CRC. However, in only men they were more likely to smoke cigarettes, drink alcohol and have one or more metabolic conditions (Table 1).

Table 1.

Characteristics of study participants by subsequent colorectal cancer diagnosis (n = 132 606)

Characteristic	Women (SWHS) number, median (IQR) or %			Men (SMHS) number, median (IQR) or %			Overall (SWHS & SMHS) number, median (IQR) or %
	Non-CRC	CRC	P-valuea	Non-CRC	CRC	P-valuea	Non-CRC	CRC	P-valuea
No. of participants	71 466	979		59 470	691		130 936	1670
CHFP score	33.9 (5.7)	33.6 (6.0)		34.1 (6.3)	33.9 (6.7)		34.0 (6.0)	33.7 (6.2)
Age (years)	50.1 (16.3)	59.2 (15.2)		52.9 (15.6)	52.8 (15.8)		51.5 (16.0)	60.5 (15.3)
Educational levels
Elementary or less	21.0	33.7	0.71	6.4	12.2	0.67	14.3	24.8	<0.01
Middle school	37.4	31.1		33.3	31.7		35.5	31.3
High school	28.0	24.5		36.2	28.4		31.8	26.5
Professional/college or higher	13.6	10.7		24.1	26.7		18.4	17.4
Income levelsb
Low	16.0	19.9	0.1	15.6	9.4	0.55	14.4	15.6	0.09
Lower middle	38.2	39.7		42.4	45.0		40.1	41.9
Upper middle	28.2	24.7		35.2	37.1		31.4	29.8
High	17.6	15.6		9.8	8.5		14.1	12.7
Lifetime smoking (pack-years)
Never smoked	97.2	97.7	0.03	30.4	34.6	0.03	66.9	71.6	<0.01
Pack-years <20	2.2	1.5		30.8	25.2		15.2	11.3
Pack-years ≥20	0.6	0.8		38.8	40.2		17.9	17.1
Alcohol consumption (drinks/day)c
Lifetime abstained	97.7	98.2	0.22	66.4	64.5	0.03	83.5	84.2	0.14
<1 drink/day	1.8	1.7		7.1	7.4		4.2	4.1
≥1 drink/day	0.4	0.1		26.5	28.1		12.3	11.7
Use of multivitamins	19.4	19.5	0.33	15.2	18.0	0.49	17.5	18.9	0.44
Body mass index (kg/m2)	23.7 (4.4)	24.4 (4.5)	<0.01	23.7 (4.0)	24.2 (4.3)	<0.01	23.7 (4.3)	24.3 (4.4)	<0.01
<22.9 kg/m2	41.1	33.9	0.03	40.8	34.4	<0.01	41.0	34.1	<0.01
23.0–24.9 kg/m2	23.8	21.5		26.1	25.1		24.9	22.9
25.0–27.4 kg/m2	20.5	24.7		22.6	23.4		21.4	24.2
27.5–29.9 kg/m2	9.5	13.3		7.9	13.3		8.8	13.3
≥30.0 kg/m2	5.1	6.6		2.6	3.8		3.9	5.5
Metabolic conditionc	30.9	41.4	0.45	35.2	49.6	0.02	32.9	44.8	0.03
Family history of CRC	2.2	3.3	0.02	2.1	3.5	0.01	2.2	3.4	<0.01
Physical activity (MET-h/week)	100.4 (57.1)	102.2 (52.1)	0.30	53.9 (44.1)	60.6 (48.7)	<0.01	78.9 (60.6)	86.5 (58.8)	<0.01
Total of energy (kcal/day)	1635 (494)	1629 (513)	0.66	1862 (619)	1862 (615)	0.36	1728 (568)	1727 (583)	0.29

Median (IQR) for continuous variables and % for categorical variables.

MET, metabolic equivalent.

^aThe differences across subgroups were compared using Mann-Whitney tests for continuous variables and χ² for categorical variables.

^bDefined as low: <¥10 000 per family per year for women and < ¥500 person per month for men. Lower middle: ¥10 000–19 999 per family per year for women and ¥500–999 per person per month for men. Upper middle: ¥20 000–29 999 per family per year for women and ¥1000–1999 per person per month for men. High: ≥¥30 000 per family per year for women and ≥¥2000 per person per month for men.

^cHaving one or more of the following four conditions: history of hypertension, heart disease, diabetes mellitus, or BMI of 30 or higher.

Participants with higher CHFP compliance scores tended to be better educated and have higher incomes than those with lower compliance scores. Conversely, participants with lower CHFP compliance scores were likely to smoke more cigarettes and to drink more alcohol, but less likely to take multivitamins than those with higher levels of CHFP compliance. Participants with a higher CHFP compliance score had a higher prevalence of metabolic conditions and family history of CRC. They also had a lower percentage of obesity and higher amount of regular exercise and physical activity (almost P <0.05) (Table 2).

Table 2.

Characteristics of study participants by quartiles of the CHFP score (n = 132 606)

Characteristic	Quartiles of CHFP score number, median (IQR) or %
	Q1	Q2	Q3	Q4	P-valuea
No. of participants	33 153	33 150	33 152	33 151
CHFP score	28.1 (3.6)	32.5 (1.6)	35.3 (1.3)	38.2 (2.0)
Age (years)	51.1 (16.9)	51.6 (16.6)	51.5 (15.7)	52.0 (15.4)
Educational levels
Elementary or less	19.9	15.9	12.5	9.6	<0.01
Middle school	39.1	36.6	34.9	31.3
High school	28.7	31.1	32.6	34.3
Professional/college or higher	12.3	16.4	20.0	24.7
Income levelsb
Low	20.3	15.4	12.1	10.0	<0.01
Lower middle	42.7	41.3	39.9	36.7
Upper middle	26.4	30.2	33.3	35.5
High	10.6	13.1	14.7	17.8
Lifetime cigarette smoking (pack-years)
Never smoked	61.0	65.9	68.6	72.2	<0.01
Pack-years <20	14.5	15.2	15.5	15.4
Pack-years ≥20	24.5	18.9	15.9	12.4
Alcohol consumption (drinks/day)
Lifetime abstained	78.0	83.2	85.6	87.2	<0.01
<1 drink/day	4.1	4.2	4.2	4.5
≥1 drink/day	17.9	12.6	10.2	8.3
Use of multivitamins	13.0	16.1	18.2	22.5	<0.01
Body mass index (kg/m2)	23.7 (4.4)	23.7 (4.3)	23.7 (4.2)	23.7 (4.1)	0.11
<22.9 kg/m2	41.3	40.8	41.3	40.1	0.25
23.0–24.9 kg/m2	24.1	24.7	24.7	25.8
25.0–27.4 kg/m2	21.0	21.3	21.7	21.9
27.5–29.9 kg/m2	9.1	9.1	8.7	8.6
≥30.0 kg/m2	4.5	4.1	3.6	3.6
Metabolic conditionc	31.8	32.9	32.6	34.7	<0.01
Family history of CRC	1.9	2.1	2.3	2.4	<0.01
Physical activity (MET-h/week)	77.3 (61.7)	78.7 (61.4)	79.2 (60.5)	80.2 (60.2)	<0.01
Total of energy (kcal/day)	1769 (670)	1731 (587)	1719 (546)	1703 (488)	<0.01

Median (IQR) for continuous variables and % for categorical variables.

^aThe differences across subgroups were compared using the Kruskal-Wallis tests for continuous variables and χ² for categorical variables.

^bDefined as low: <¥10 000 per family per year for women and <¥500 person per month for men. Lower middle: ¥10 000–19 999 per family per year for women and ¥500–999 per person per month for men. Upper middle: ¥20 000–29 999 per family per year for women and ¥1000–1999 per person per month for men. High: ≥¥30 000 per family per year for women and ≥¥2000 per person per month for men.

^cHaving one or more of the following four conditions: history of hypertension, heart disease, diabetes mellitus, or BMI of 30 or higher.

The CHFP compliance score was inversely associated with the risk of CRC in the age-, gender- and energy-adjusted model, as well as in the multivariate adjusted model. Approximately 15% reduced risk of CRC was found when comparing the 2nd, 3rd and 4th (highest) quartiles (HR_{Q2 vs Q1}: 0.88; 95% CI: 0.77, 1.00, HR_{Q3 vs Q1}: 0.86; 95% CI: 0.75, 0.98 and HR_{Q4 vs Q1}: 0.84; 95% CI: 0.73, 0.96) with the 1st (lowest) quartile of CHFP compliance scores, with P_trend = 0.01 in the multivariate adjusted model. Per standard deviation (SD), increase of the CHFP compliance score was associated with a 6% reduced risk of CRC (HR: 0.94; 95% CI: 0.90, 0.99; P = 0.02). However, neither the modified AHEI-2010 nor the modified DASH score was associated with the risk of CRC, with one exception. In the model with adjustment for age, gender and total energy, the 4th quartile of DASH score was associated with a reduced risk of CRC (HR_{Q4 vs Q1}: 0.86; 95% CI: 0.75, 0.98; P_trend = 0.05) when compared with the lowest quartile. Non-linear association was not found for any of the three adherence scores (Table 3).

Table 3.

HRs (95% CIs) for colorectal cancer by quartiles of dietary recommendation adherence scores

	Quartiles of dietary recommendation adherence scores				Ptrend	HR (95% CI) per SD increase	P	P for curvea
	Q1	Q2	Q3	Q4
CHFP score
No. of cases/no. of participants	469/33 153	415/33 150	396/33 152	390/33 151
Age-, gender-, and energy- adjusted modelb	1.00	0.86 (0.76, 0.98)	0.84 (0.73, 0.96)	0.81 (0.71, 0.93)	<0.01	0.93 (0.88, 0.97)	<0.01	0.76
Multivariate Model 1c	1.00	0.88 (0.77, 1.00)	0.86 (0.75, 0.98)	0.84 (0.73, 0.97)	0.01	0.94 (0.90, 0.99)	0.02	0.73
Multivariate Model 2d	1.00	0.88 (0.77, 1.00)	0.86 (0.75, 0.98)	0.84 (0.73, 0.96)	0.01	0.94 (0.90, 0.99)	0.02	0.74
Modified AHEI-2010
No. of cases/no. of participants	471/33 152	411/33 151	412/33 152	376/33 151
Age-, gender-, and energy- adjusted modelb	1.00	0.91 (0.80, 1.04)	0.93 (0.81, 1.06)	0.89 (0.78, 1.02)	0.12	0.96 (0.91, 1.00)	0.06	0.86
Multivariate Model 1c	1.00	0.92 (0.80, 1.05)	0.94 (0.83, 1.08)	0.91 (0.79, 1.05)	0.27	0.97 (0.92, 1.02)	0.17	0.88
Multivariate Model 2d	1.00	0.92 (0.80, 1.05)	0.94 (0.83, 1.08)	0.91 (0.79, 1.05)	0.27	0.96 (0.92, 1.01)	0.13	0.89
Modified DASH score
No. of cases/no. of participants	428/33 153	405/33 150	441/33 150	396/33 153
Age-, gender-, and energy- adjusted modelb	1.00	0.90 (0.79, 1.03)	0.96 (0.84, 1.10)	0.86 (0.75, 0.98)	0.05	0.95 (0.90, 0.99)	0.03	0.90
Multivariate Model 1c	1.00	0.92 (0.80, 1.05)	0.98 (0.86, 1.12)	0.90 (0.78, 1.03)	0.23	0.97 (0.92, 1.02)	0.18	0.90
Multivariate Model 2d	1.00	0.92 (0.80, 1.05)	0.98 (0.86, 1.12)	0.90 (0.78, 1.03)	0.23	0.96 (0.91, 1.01)	0.15	0.90

^a P for non-linear test using restricted cubic spline regression.

^bCox proportional hazards model was stratified by gender and birth year in 5-year intervals and adjusted for total of energy (continuous kcal/day).

^cMultivariate Model 1: Cox proportional hazards model was stratified by gender and birth year in 5-year intervals and adjusted for education levels, income levels, lifetime smoking (three categories: never smoked, pack-years <20, pack-years ≥20), alcohol consumption (continuous drinks/day only for CHFP and DASH), use of multivitamins, family history of CRC, physical activity (continuous MET-h/week), total of energy (continuous kcal/day).

^dMultivariate Model 2: Multivariate Model 1 with additional adjustment for BMI (continuous kg/m²) and metabolic conditions.

Table 4 shows multivariate-adjusted HRs (95% CI) for CRC associated with the CHFP compliance score by anatomical site of cancer. The CHFP compliance score was inversely associated with the risk of rectal cancer. A 23–24% reduced risk of rectal cancer was found when comparing the two highest quartiles (HR_{Q3 vs Q1}: 0.77; 95% CI: 0.62, 0.97 and HR_{Q4 vs Q1}: 0.77; 95% CI: 0.62, 0.97) with the lowest quartile of the CHFP compliance score, with P_trend = 0.01. Per SD, increase in the CHFP compliance score was associated with a 10% reduced risk of rectal cancer (HR: 0.90; 95% CI: 0.83, 0.98; P = 0.01). No association between the CHFP compliance score and colon cancer risk was observed. However, the association direction for colon and rectal cancer had the same the confidence intervals for risk estimates overlapped. Non-linear association was not found for the CHFP compliance score in any of the anatomical specific analyses (Table 4).

Table 4.

HRs (95% CIs) for colorectal cancer by quartiles of Chinese Food Pagoda score by anatomical site of cancer

	Quartiles of CHFP score				Ptrend	HR (95% CI) per SD increase)	P	P for curve a
	Q1	Q2	Q3	Q4
Colon cancer
No. of cases/no. of participants	285/33 153	246/33 150	258/33 152	253/33 151
Age-, gender-, and energy- adjusted modelb	1.00	0.84 (0.71, 1.00)	0.90 (0.76, 1.07)	0.87 (0.73, 1.03)	0.13	0.96 (0.90, 1.02)	0.18	0.28
Multivariate Model 1c	1.00	0.85 (0.72, 1.01)	0.92 (0.77, 1.09)	0.89 (0.74, 1.06)	0.26	0.97 (0.91, 1.03)	0.32	0.27
Multivariate Model 2d	1.00	0.85 (0.72, 1.01)	0.91 (0.77, 1.08)	0.88 (0.74, 1.05)	0.23	0.96 (0.91, 1.03)	0.27	0.27
Rectal cancer
No. of cases/no. of participants	184/33 153	169/33 150	138/33 152	137/33 151
Age-, gender-, and energy- adjusted modelb	1.00	0.89 (0.72, 1.10)	0.74 (0.59, 0.92)	0.72 (0.58, 0.90)	<0.01	0.88 (0.81, 0.95)	<0.01	0.21
Multivariate Model 1c	1.00	0.92 (0.75, 1.14)	0.78 (0.62, 0.97)	0.78 (0.62, 0.98)	0.01	0.91 (0.84, 0.98)	0.02	0.19
Multivariate Model 2d	1.00	0.92 (0.74, 1.13)	0.77 (0.62, 0.97)	0.77 (0.62, 0.97)	0.01	0.90 (0.83, 0.98)	0.01	0.19

^a P for non-linear test using restricted cubic spline regression.

^bCox proportional hazards model was stratified by gender and birth year in 5-year intervals and adjusted for total of energy (continuous kcal/day).

^cMultivariate Model 1: Cox proportional hazards model was stratified by gender and birth year in 5-year intervals and adjusted for education levels, income levels, lifetime smoking (three categories: never smoked, pack-years <20, pack-years ≥20), alcohol consumption (continuous drinks/day only for CHFP and DASH), use of multivitamins, family history of CRC, physical activity (continuous MET-h/week), total of energy (continuous kcal/day).

^dMultivariate Model 2: Multivariate Model 1 with additional adjustment for BMI (continuous kg/m²) and metabolic conditions.

We did not find that the CHFP compliance score and CRC risk association was modified by gender, age, obesity status or metabolic conditions or educational attainment, in the stratified analysis. However, the inverse association between CHFP compliance score and CRC appeared to be slightly stronger among participants who were younger than 50 years, had a lower BMI (<25 kg/m²) or who were without any metabolic conditions at baseline (Table 5). Additional analyses, including all cohort observations, showed similar associations between the CHFP compliance score and risk of CRC (Supplementary Table 2, available as Supplementary data at IJE online).

Table 5.

HRs (95% CIs) for colorectal cancer by quartiles of Chinese Food Pagoda score in stratified analyses^a

		Quartiles of CHFP score
	CRC/participants	Q1	Q2	Q3	Q4	P trend	P for interaction factor
Women	979/72 445	1.00	0.90 (0.75, 1.06)	0.89 (0.75, 1.06)	0.83 (0.69, 0.99)	0.06	0.76
Men	684/60 161	1.00	0.84 (0.68, 1.04)	0.80 (0.65, 0.99)	0.84 (0.68, 1.05)	0.08
Age <50 years	346/58 410	1.00	0.87 (0.65, 1.15)	0.78 (0.58, 1.04)	0.63 (0.45, 0.86)	<0.01	0.28
Age ≥50 years	1324/74 196	1.00	0.88 (0.76, 1.02)	0.88 (0.76, 1.03)	0.90 (0.77, 1.05)	0.16
BMI <25 (kg/m2)	953/87 124	1.00	0.92 (0.77, 1.09)	0.81 (0.67, 0.97)	0.81 (0.68, 0.98)	0.01	0.36
BMI ≥25 (kg/m2)	717/45 482	1.00	0.83 (0.67, 1.02)	0.93 (0.76, 1.15)	0.87 (0.70, 1.07)	0.30
Non-metabolic conditions	922/88 832	1.00	0.87 (0.73, 1.03)	0.79 (0.66, 0.95)	0.76 (0.63, 0.91)	<0.01	0.32
Metabolic conditions	748/43 774	1.00	0.89 (0.72, 1.09)	0.95 (0.78, 1.17)	0.95 (0.77, 1.17)	0.72
Low education	934/66 044	1.00	0.87 (0.73, 1.03)	0.96 (0.80, 1.15)	0.84 (0.70, 1.02)	0.16	0.20
High education	736/66 562	1.00	0.88 (0.71, 1.09)	0.75 (0.60, 0.93)	0.82 (0.67, 1.01)	0.03

^aCox proportional hazards model was stratified by gender and birth year in 5-year intervals and adjusted for education levels, income levels, lifetime cigarette smoking (three categories: never smoked, pack-years <20, pack-years ≥20), alcohol consumption (continuous drinks/day), use of multivitamins, family history of CRC, BMI (continuous kg/m²), metabolic conditions, physical activity (continuous MET-h/week), total of energy (continuous kcal/day) and except for corresponding variable used for stratification.

Discussion

In this prospective population-based cohort study, we found that higher CHFP compliance scores were associated with ∼15% lower risk of CRC and 23% lower risk of rectal cancer when comparing the highest compliance score (the 3rd and 4th quartiles) with the lowest score (the 1st quartile) after adjustment for multiple potential confounders (P_trend <0.05). The association appeared to be stronger among individuals who were younger than age 50 at baseline, who had no metabolic condition and whose BMI was less than 25 kg/m², although no multiplicative interaction was noted. The CHFP compliance and CRC association was attenuated in the multivariable adjusted model with BMI, metabolic conditions, family history of CRC, lifestyle factors and demographic covariates. In general, the modified AHEI-2010 and the modified DASH compliance scores were not associated with CRC risk in our study population.

The Chinese Dietary Guidelines and the two well-known US dietary guidelines were developed with the objective to achieve a balanced diet with adequate and diverse nutrients and food intakes which could prevent chronic diseases, including cancer. Several cohort studies have previously shown that a higher AHEI or DASH score was associated with lower risk of CRC and rectal cancer in the USA. In the National Institutes of Health-AARP Diet and Health Study, Reedy et al. (2008) reported that a higher AHEI was associated with 17–29% lower risk of CRC.¹⁸ Another study found that a higher AHEI also predicted a 46% lower risk of rectal cancer and 25% lower risk of distal colon cancer among males.¹⁸ A higher DASH score was found to be associated with a reduced risk of CRC of 19–25% for men and 16–21% for women.²⁰ In the Women’s Health Initiative Observational Study, a higher DASH score, but not AHEI score, was associated with reduced CRC risk.¹⁹ In our study, compliance with the modified DASH and the modified AHEI-2010 was not related to the CRC risk among our study population. This observation may be explained by the fact that the modified DASH and modified AHEI-2010 were originally designed for prevention of hypertension and cardiovascular disease^16,17 among North Americans. In the Nurses Health Study and Health Professionals Follow-Up Study, the highest quintile of the AHEI-2010 was associated with only 6% reduced risk of cancer, whereas it was associated with 31%, 24% and 20% decreased risk of coronary heart disease, cardiovascular disease and stroke, respectively.³⁶ Moreover, the modified DASH combined fish, meat, poultry and eggs into one component and recommended a low consumption of these foods. These foods may have different influences on CRC risk.³⁷ In addition, information on whole grain, sugar-sweetened beverages and trans-fat intakes was not included when we derived the compliance scores for the modified AHEI-2010 and the modified-DASH, due to the lack of relevant data in our FFQs and in the Chinese Food Composition table.

A meta-analysis of six studies found that every 90 g/day increase in whole grains reduced the risk of CRC by 17%,³⁸ but the association between sugar-sweetened beverages and trans-fat intake and the risk of CRC still lacks conclusive evidence.^39,40 However, consumption of sugar-sweetened beverages and trans-fats are positively associated with weight gain, obesity, diabetes mellitus, heart disease and cardiovascular disease,^40,41 which are potential risk factors of CRC.^42–44 This might be the reason why we found no meaningful impact on the hazard ratios of the modified AHEI and the modified DASH score. However, we did find that compliance with the CHFP recommendation was associated with reduced CRC risk, particularly rectal cancer risk. These results suggest that dietary recommendations may need to be tailored to specific populations in order to maximize health impact.

The main components of the CHFP recommend higher intakes of grains, fruits, vegetables, beans and bean products, dairy products, appropriate amounts of fish and shrimp, lower intakes of meat and poultry, eggs, and limitation of fat and salt. Evidence from epidemiological studies have linked high intakes of fibre,⁴⁵ fruits and vegetables,^46,47 fish⁴⁸ and dairy products⁴⁹ to lower risk of CRC. Studies have also shown that high consumption of red or processed meat,^50,51 as well as moderate or high alcohol consumption,⁵² were associated with increased risk of CRC. When individual food components of the CHFP were analysed, we found that only high consumption of fruits and dairy products were associated with decreased CRC and rectal cancer risk (Supplementary Table 3, available as Supplementary data at IJE online). The observed inverse association between fruits and the development of CRC might be largely attributed to their high level of nutrients and numerous bioactive compounds, including fibre, vitamins C and E, folic acid, flavonoids, polyphenols and limonene.¹⁴ These nutrients and phytochemicals might act as potential anti-tumorigenic agents through increasing anti-oxidative activities, which could inhibit cellular damage and exposure to reactive oxygen species.^53,54 In addition, dairy products are a rich source of calcium, which may protect against CRC development via its ability to bind unconjugated bile acids and free fatty acids in the colorectum.⁵⁵ Calcium may also reduce cell proliferation, promote cell differentiation, prevent colonic K-Ras mutations and inhibit haeme-induced promotion of colon carcinogenesis.^56–58 Consumption of dairy products provides lactic acid-producing bacteria and many bioactive compounds such as lactoferrin, vitamin D and short-chain fatty acid butyrate, which may have protective effects against CRC.⁵⁹

The most notable strength of our study is that it was based on two large prospective population-based cohort studies that had high initial enrolment rates and high follow-up rates. The SWHS/SMHS interviews used validated FFQs that covered around 90% of commonly consumed foods in urban Shanghai²⁶ and showed fairly high validity and reproducibility as compared with multiple 24-h dietary recalls.^29,30 Extensive information on covariates allowed a full adjustment for potential confounding efforts.

Several limitations in our study should be considered when interpreting our findings. Dietary intake in free-living individuals is difficult to capture, and measurement errors are inevitable. Such misclassification is likely to be non-differential, and thus lead to an underestimate of the true association. However, possible bias due to the influence of preclinical cancer-related dietary pattern changes is likely, as evidenced by the analysis that included all study observations, which resulted in attenuated risk estimates (Supplementary Table 2, available as Supplementary data at IJE online). Furthermore, dietary guideline compliance scores are often associated with higher education and income and healthy lifestyle factors such as not smoking, drinking less alcohol, taking regular multivitamins and being physically active. Although we carefully adjusted for these confounders, residual confounding from other unknown or unmeasured confounders cannot be excluded. The aforementioned lack of data in the Chinese Food Composition table for some dietary components that are included in the modified AHEI score and the modified DASH score is also a drawback. In addition, the component grains in the CHFP score were mostly white rice with few refined wheat products, with exclusion of whole grains because of very low consumption in the study population. Finally, since our participants were recruited from urban communities in Shanghai, the findings of this study may not necessarily be generalizable to the rural population.

In conclusion, we found that adherence to Chinese dietary recommendations was inversely associated with CRC risk, particularly rectal cancer risk among urban Chinese. Efforts to promote compliance with the CHFP should be considered as a means to slow down the trend of increasing CRC incidence in China.

Funding

The Shanghai Women’s Health Study and Shanghai Men’s Health Study were supported by the US National Institutes of Health (R37 CA070867 and UM1 CA182910 to W.Z., UM1 CA173640 and R01 HL079123 to X.O.S.).