Post-cancer diagnosis dietary inflammatory potential is associated with survival among women diagnosed with colorectal cancer in the Women’s Health Initiative

Abstract

Purpose:

Dietary factors may influence colorectal cancer (CRC) survival through effects on inflammation. We examined the association between post-CRC diagnosis inflammatory potential of diet and all-cause and cancer-specific mortality in the Women’s Health Initiative.

Methods:

The study included 463 postmenopausal women who developed CRC during follow-up and completed a food frequency questionnaire (FFQ), on average 1.7 years after diagnosis. Women were followed from CRC diagnosis until death, censoring, or the end of follow up in October 2014. Energy-adjusted dietary inflammatory index (E-DII)® scores were calculated from the FFQ and dietary supplement inventory. Cox proportional hazards models were fitted to estimate multivariable-adjusted HRs and 95% confidence intervals (CI) for all-cause, total cancer, and CRC-specific mortality with the most pro-inflammatory E-DII scores (tertile 3) as referent.

Results:

After a median 11.6 years of follow-up, 162 deaths occurred, including 77 from CRC. Lowest tertile (i.e., most anti-inflammatory) E-DII scores from diet plus supplements were associated with significantly lower all-cause mortality (HR_T1vsT3=0.49; 95% CI=0.31–0.79) compared to the most pro-inflammatory E-DII tertile. Modest associations with total cancer mortality or CRC-specific mortality were observed, though 95%CIs included 1.

Conclusions:

Consuming a dietary pattern and supplements with more anti-inflammatory potential after CRC diagnosis may improve overall survival among postmenopausal women.

Introduction

Among females in the US, colorectal cancer is the third most common cancer, and cause of cancer death, following breast and lung cancer [1]. Currently, the 5- and 10-year relative survival rates are 65% and 58%, respectively, leading to a growing number of long-term survivors who will constitute the third largest cancer survivor group in females by 2026 [2]. As colorectal cancer survivors remain at an increased risk of cancer recurrence, secondary cancer, or comorbidities which adversely impact their overall survival, investigating whether post-diagnosis modifiable lifestyle factors are associated with survival is an important strategy for improving colorectal cancer prognosis [3–5].

Although individual dietary risk factors for colorectal cancer incidence have been investigated extensively [6], there is a paucity of data on how post-diagnostic diet and supplement use affects survival among colorectal cancer patients [4,7]. As different dietary components may interact with each other to exert an effect, dietary pattern research has advantages over single-nutrient or food studies when investigating associations with disease outcomes [8]. To date, only a few studies have investigated the association between post-cancer diagnosis dietary patterns and colorectal cancer survival, producing inconsistent findings [5,9,10].

Chronic low-grade inflammation is an important mediator through which diet and supplements may influence colorectal cancer prognosis [11,12]. We previously reported that the dietary inflammatory index (DII®), a tool developed to assess the inflammatory potential of overall diet, is associated with risk of colorectal cancer in multiple studies, including the Women’s Health Initiative (WHI) [13–19]. To date, only one published study has reported no statistically significant association between post-diagnosis DII and all-cause mortality in long‐ term colorectal cancer survivors among whom dietary data were collected at a median of 6 years after cancer diagnosis [20]. However, healthier post-diagnosis diet as determined by higher scores on other a priori diet quality indices, such as the Dietary Approaches to Stop Hypertension and the American Cancer Society nutrition guidelines scores, was associated with lower risk of all-cause and colorectal cancer-specific mortality in recent analyses of the Cancer Prevention Study II Nutrition Cohort [21]. In this study, we utilized data from the WHI to investigate the association between post-diagnostic inflammatory potential of diet and supplements and all-cause mortality, total cancer mortality, and colorectal cancer-specific mortality among postmenopausal women with colorectal cancer. As the DII is primarily nutrient-based, we calculated the DII from diet plus supplements, as well as from diet only.

Materials and Methods

Study population

The WHI is a large clinical study to investigate some of the most common causes of morbidity and mortality among postmenopausal women. Details of the design of the WHI have been

described in depth [22–24]. Briefly, between 1993 and 1998, 161,808 postmenopausal women aged 50 to 79 years were enrolled from 40 sites across the US into either one or more of three randomized controlled Clinical Trials (CT) (n=68,132) or the Observational Study (OS) (n=93,676). Women were not eligible for either the CT or the OS if they had predicted survival of less than three years, conditions that adversely affect adherence or retention such as dementia and mental illness, or active participation in other randomized intervention trials [23]. The CT had additional eligibility criteria; e.g., women were excluded from participation in the Dietary Modification (DM) arm of the trial if their diet had less than 32% energy from fat or they were on a diabetic or low-salt diet [23]. The primary follow-up of CT and OS ended in 2005 and follow up continued among consenting participants in the WHI Extension Studies I (2005–2010) and II (2010–2015). Only WHI-DM and WHI-OS participants had repeated FFQs during follow-up, which permitted the assessment of post-cancer diagnosis diet and supplement use. Therefore, for the present analysis, we focused only on women from the WHI-DM and WHI-OS who were free of cancer at or before baseline except non-melanoma skin cancer, were diagnosed with primary invasive colorectal cancers during follow-up, and completed an FFQ after diagnosis (n=486; WHI-OS=192 and WHI-DM=294). Of these, we excluded 22 women (WHI-OS=11; WHI-DM=11) who reported unreasonable daily energy intake (outside the range of 600–5000 kcals/day), and one woman who did not contribute follow-up time in the cohort. Our final sample included 463 colorectal cancer survivors who were diagnosed with colon (75.6%), rectum (15.7%) or rectosigmoid cancer (8.7%). The WHI protocol was approved by the Institutional Review Boards at the Clinical Coordinating Center (CCC) at the Fred Hutchinson Cancer Research Center (Seattle, WA) and at each of the participating Clinical Centers. All participants provided written informed consent in accordance with the US Common Rule.

Dietary assessment

Women were asked to report dietary intake in the previous three months using a self-administered FFQ [25]. Nutrient intake from the FFQ was derived by linking to the University of Minnesota Nutrition Coordinating Center food and nutrient database [26]. The WHI FFQ produced comparable estimates to mean dietary intake from four 24-hour dietary recalls and a 4-day food record, with the energy-adjusted correlation coefficients ranging from 0.20 (vitamin B12) to 0.68 (magnesium) with a mean of 0.49; including supplement intake generally increased correlation coefficients [25].

In the WHI-OS, participants completed the FFQ at baseline and year 3 of follow-up. In the WHI-DM, participants completed the FFQ at baseline and year 1 of follow-up, and thereafter a random 33% subset of DM participants completed the FFQ each year on a rotating basis from year 2 to year 9 [27]. In the primary analysis, we chose the first FFQ occurring after diagnosis of colorectal cancer for each subject, which occurred on average 1.7 years after diagnosis for WHI-DM (range from 12 days to 7.2 years) and 1.4 years for OS participants (range from 7 days to 3.4 years). Dietary supplement use was assessed at baseline and annual visit for WHI-CT and at year 3 follow-up visit for OS when participants brought in their dietary supplements in their original pill bottles. Similar to the post-cancer diagnosis FFQ, supplement use was identified at the time closest to, but after, participants’ diagnoses of colorectal cancer, which occurred on average 1.4 years after diagnosis for WHI-DM (range from 2 days to 5.2 years) and 1.5 years for OS participants (range from 7 days to 3.8 years).

Description of energy-adjusted DII score

The energy-adjusted DII® (E-DII™) score was calculated to represent the inflammatory potential of an individual’s overall diet, using data from FFQ responses only (diet only) and from diet plus supplement intake combined [28]. Higher E-DII scores represent more pro-inflammatory diets while lower (i.e., more negative) E-DII scores represent more anti-inflammatory diets. Details of the development and construct validation of DII have been published previously [28–31]. Briefly, an extensive literature search was performed to identify peer-reviewed primary research articles published through 2010 that reported the association between dietary factors and six inflammatory markers (interleukin (IL)-1β, IL-4, IL-6, IL-10, tumor necrosis factor-a (TNFα), and C-reactive protein (CRP)). A total of 1,943 qualifying articles were reviewed, indexed, and scored to derive the component-specific inflammatory effect score for 45 dietary factors (i.e., components of DII), comprising macronutrients and micronutrients as well as some bioactive components [28].

Because 13 DII components, including ginger, turmeric, garlic, oregano, pepper, rosemary, eugenol, saffron, flavan-3-ol, flavones, flavonols, flavonones, and anthocyanidins were not available from the WHI FFQ, data on 32 DII components that are available in the FFQ were used to calculate the E-DII score. Because we adjusted DII components for energy using the energy density approach, energy was not included separately as one of the components of the DII. Thus, the remaining 31 components included alcohol, beta-carotene, caffeine, carbohydrates, cholesterol, total fat, saturated fat, fiber, folic acid, iron, magnesium, niacin, riboflavin, thiamin, zinc, monounsaturated fats, polyunsaturated fats, omega 3 fats, omega 6 fats, trans fat, protein, selenium, vitamins B12, B6, A, C, D, E, onion, green/black tea, and isoflavones. These WHI FFQ-derived food and nutrient intake data were first adjusted for total energy (i.e., per 1000 kilocalories) and standardized by creating a z-score for each component using mean and standard deviation data from a global energy-adjusted dietary database comprised of dietary intake from 11 populations living in different regions of the world. Then, the z-scores for each DII component were converted to percentiles and centered by doubling each percentile score and subtracting 1. From this, the energy-adjusted standardized dietary intake (expressed as centered proportions) was then multiplied by the corresponding literature-derived inflammatory effect score for each DII component. Individual scores from each DII component were then summed to determine the overall E-DII score for each individual. The DII has been evaluated for construct validity using data from the WHI, and found to be significantly associated with circulating concentrations of IL-6 and TNFα receptor 2 levels [30].

A majority of colorectal cancer survivors (67%) in the WHI took supplements after their cancer diagnoses, and diet and supplements together capture the most biologically robust exposure estimate. Therefore, we report E-DII score from diet plus supplements as the primary exposure of interest to characterize the overall inflammatory effect from diet and supplements on mortality outcomes, and calculated the E-DII from diet only separately.

Other covariate assessments

Age, race/ethnicity, educational attainment, and family income level were assessed at baseline using self-administered questionnaires. Recreational physical activity in metabolic equivalent tasks (MET)-hours/week and smoking status were self-reported at baseline for all WHI-OS and WHI-DM participants and updated for the WHI-DM only at years 1, 3, 6, and 9. We included baseline recreational physical activity and smoking status in our analysis to ensure consistent timing of assessments for the entire study population and to overcome the large missing proportion issue of using post-diagnosis measurements. In the WHI, participants reported various types of recreational physical activities including mild, moderate and strenuous physical activity and walking, and reported the duration of each activity (hours/week) via self-report questionnaire. The recreational physical activity was then calculated as the energy expenditure from all activities in MET-hours/week and was categorized to four levels (0 MET-hours/week; 0.1–3 MET-hours/week; 3.1–8.9 MET-hours/week; 9 or more MET-hours/week) [32,33]. Weight and height were measured using standard methods. WHI-DM participants’ weights and heights were assessed at baseline and subsequently annually from years 1 to 9, and weights and heights were measured for WHI-OS participants at baseline and year 3. To minimize missing data and be consistent with the timing of measurement of other covariates, we used baseline weight and height to calculate body mass index (BMI) as weight (kg)/height(m)² and categorized women as underweight, normal weight, overweight, and obese using the World Health Organization criteria [34].

Incident colorectal cancer cases were identified primarily through annual self-reported medical updates [35]. Reported colorectal cancer cases were locally adjudicated and assigned a diagnosis, first based on medical record review and then sent to the Clinical Coordinating Center for centralized review. Detailed characteristics of the cancer, such as stage, anatomic subsite, diagnosis date, and tumor morphology (behavior, differentiation grade, histology) were recorded according to Surveillance, Epidemiology, and End Results (SEER) coding guidelines [35,36].

Outcome ascertainment

Three mortality outcomes were examined: death from any cause, death from any cancer, and death from colorectal cancer. Vital status of participants was updated by contacts during annual clinic visits for CT and through mailings for the OS [22]. Two central adjudicators ascertained cause of death through review of recent hospitalization and autopsy records; when these records were unavailable, death certificates, medical records or other records were used as needed [35]. Regular data linkage to the National Death Index (NDI) was conducted to ensure complete mortality ascertainment [35,37].

Statistical analysis

E-DII scores were categorized into tertiles. We described baseline characteristics and tumor characteristics by means and standard errors (SE) of continuous variables, and by number and frequencies of categorical variables across E-DII tertiles. Differences in continuous and categorical variables across tertiles of the E-DII scores were tested by performing an ANOVA test and a Chi-square test, respectively.

Each participant accumulated person-years of follow-up time from diagnosis of primary colorectal cancer until death, loss to follow-up, the last NDI search date for the participant, or the end of follow up (October 2014), whichever occurred first. We used Cox proportional hazards models to examine the association between post-diagnostic E-DII tertiles and each mortality outcome, with women in the highest tertile (most pro-inflammatory scores) as the referent. E-DII score from diet plus supplements and from diet only were both examined. Age- and energy-adjusted and multivariable-adjusted HRs and 95% confidence intervals (CI) were reported for each model. Given our study design, an immortal time period may appear when no subjects were at risk of death from the date of colorectal cancer diagnosis to the date of FFQ completion. To account for this immortal time bias, we added a time-dependent covariate in the multivariable-adjusted Cox model to stratify participants’ status before and after the post-diagnosis FFQ [38]. The proportional hazard (PH) assumption was examined in all models using the Schoenfeld residual test [39]. There was no evidence that the E-DII tertiles violated the PH assumption. In analyses where covariates violated the PH assumption (i.e., WHI components and income levels), we fitted an extended Cox proportional hazards model by either stratifying on the categorical covariate or adding a time-dependent covariate as the product of follow-up time and continuous covariate [40,41]. The linear relationship between continuous E-DII score and each mortality outcome was assessed using the restricted cubic spline function within the Cox model. [42] We reported P values for non-linearity for all the analyzed associations. If linearity held (P-non linearity>0.05), we also reported the P value for trend using the E-DII score as a continuous variable in the model. Multivariable analyses were adjusted for covariates that may influence colorectal cancer survival or be related to dietary intake after cancer diagnosis [9,10,32]. We adjusted for WHI study component (WHI-OS, WHI-DM-intervention, WHI-DM-control), family income levels, age group at cancer diagnosis, race/ethnicity, education attainment, years from cancer diagnosis to FFQ completion, baseline physical activity level, smoking status at baseline, baseline BMI categories, total energy intake per day, cancer stage, and cancer grade. Cancer stage and grade were used as proxy for the unavailable cancer treatment data. In the models utilizing the E-DII from diet only, we additionally adjusted for dietary supplement use.

We planned a priori to test if there was heterogeneity of the post-diagnostic E-DII score and all-cause mortality relationship by physical activity, BMI, smoking status, cancer site, cancer stage and tumor grade.. In sensitivity analysis, we excluded participants from the WHI-DM-intervention arm because dietary intervention could change an individual’s dietary habits in the long-term. Dietary intake may change during cancer treatment due to side effects, therefore we performed another sensitivity analysis by excluding individuals whose FFQs were collected within 6 months of diagnosis. Finally, we adjusted for baseline E-DII score in the respective multivariable-adjusted models of E-DII from diet plus supplement and E-DII from diet only.

The data that support the findings of this study are available by request from WHI; all requests must conform to the Publications and Presentations Policy of the WHI. All statistical analyses were conducted using SAS® version 9.4 (Cary, NC). All tests were two-sided with p values <0.05 considered statistically significant if not otherwise noted.

Results

After a median 11.6 years of follow-up, 162 deaths occurred, including 77 from colorectal cancer. As shown in Table 1, compared to women with the most pro-inflammatory E-DII scores (tertile 3), those consuming more anti-inflammatory diets and supplements had lower energy intake after cancer diagnosis, lower BMI at baseline, but higher education level, and were more likely to be enrolled in the WHI-OS or DM-intervention arm and more likely to be diagnosed with an earlier stage of colorectal cancer.

Table 1.

Demographic, lifestyle, and clinical characteristics of 463 postmenopausal women diagnosed with colorectal cancer in the WHI-DM and OS by tertiles of E-DII score from diet plus supplements, Women’s Health Initiative, 1993–2015.

	Most anti-inflammatory diet E-DII tertile 1 (−6.80, −3.91) Median: −4.65	E-DII tertile 2 (−3.90, −2.09) Median: −3.16	Most pro-inflammatory diet E-DII tertile 3 (−2.08, 3.25) Median: −0.37	P-valuea
N	155	154	154
	Mean (SE)	Mean (SE)	Mean (SE)
Age at colorectal cancer diagnosis (years)	67.8 (0.5)	67.5 (0.6)	68.0 (0.5)	0.84
Years from colorectal cancer diagnosis to FFQ	1.6 (0.09)	1.5 (0.09)	1.6 (0.1)	0.61
Years from colorectal cancer diagnosis to death from any cause	7.4 (0.9)	6.9 (0.5)	6.1 (0.5)	0.31
Total energy intake after cancer diagnosis (kcal/day)	1331 (35)	1549 (50)	1620 (57)	<.001
BMI at baseline (kg/m2)	27.4 (0.4)	28.9 (0.4)	30.0 (0.5)	0.001
Physical activity at baseline (Met-hours/week)b	12.7 (0.9)	11.2 (1.0)	9.3 (1.1)	0.06
	N(%)c	N (%)c	N (%)c
WHI components				0.001
WHI OS	75 (48.4)	58 (37.7)	47 (30.5)
WHI DM-intervention	42 (27.1)	37 (24.0)	31 (20.1)
WHI DM-control	38 (24.5)	59 (38.3)	76 (49.4)
Race/ Ethnicity				0.11
White non-Hispanic	135 (87.1)	126 (81.8)	122 (79.2)
Hispanic/Latino	2 (1.3)	3 (2.0)	6 (3.9)
Black/African-American	9 (5.8)	21 (13.6)	20 (13.0)
Other	9 (5.8)	4 (2.6)	6 (3.9)
Income level				0.41
<20,000	26 (16.8)	25 (16.2)	34 (22.1)
20,000–49,999	70 (45.2)	77 (50.0)	81 (45.51)
>=50,000	50 (32.3)	45 (29.2)	44 (24.72)
missing	9 (5.8)	7 (4.6)	16 (8.99)
Education level				0.01
High school or below	36 (23.2)	49 (31.8)	65 (42.2)
Some college	57 (36.8)	36 (23.4)	45 (29.2)
College	19 (12.3)	20 (13.0)	15 (9.7)
Post graduate	42 (27.1)	49 (31.8)	28 (18.2)
Missing	1 (0.6)	0 (0)	1 (0.7)
Cancer stage				0.001
Localized	96 (61.9)	69 (44.8)	77 (50.0)
Regional	54 (34.8)	79 (51.3)	61 (39.6)
Distant	3 (1.9)	4 (2.6)	15 (9.7)
Unknown	2 (1.3)	2 (1.3)	1 (0.7)
Smoking status at baseline				0.17
Never smoked	77 (49.7)	79 (51.3)	65 (42.2)
Past smoker	62 (40.0)	69 (44.8)	72 (46.8)
Current smoker	13 (8.4)	6 (3.9)	15 (9.7)
Missing	3 (1.9)	0 (0)	2 (1.3)
Cancer site				0.22
Colon cancer	127 (81.9)	115 (74.7)	105 (68.2)
Rectum cancer	18 (11.6)	23 (14.9)	31 (20.1)
Rectosigmoid cancer	9 (5.8)	15 (9.7)	16 (10.4)
At two of the above three locations simultaneously	1 (0.7)	1 (0.7)	2 (1.3)
Grade of cancer differentiation				0.38
Well differentiated	16 (10.3)	9 (5.8)	16 (10.4)
Moderately differentiated	106 (68.4)	97 (63.0)	100 (64.9)
Poorly differentiated	22 (14.2)	33 (21.4)	26 (16.9)
Unknown/not done	10 (6.5)	15 (9.7)	10 (6.5)
Anaplastic	1 (0.7)	0 (0)	2 (1.3)

^a.P value was calculated from ANOVA test for categorical variables and from Chi-Square test for categorical variables

^b.Total expenditure of energy from recreational physical activity (includes walking, mild, moderate and strenuous physical activity in kcal/week/kg)

^c.The sum of percentages in certain E-DII tertiles for some categorical variables may not add up to 100% because of rounding

In the multivariable adjusted Cox proportional hazards model, there was a non-linear relationship for E-DII from diet plus supplement with all-cause mortality (P-non linearity=0.001) and with colorectal cancer specific mortality (P-non linearity=0.02), with participants in the first tertile (most anti-inflammatory diet and supplement) having the lowest risk while those in second tertile having elevated mortality risk although the confidence intervals included 1, compared to the highest tertile (most pro-inflammatory group). There was 51% lower risk of all-cause mortality for women with the most anti-inflammatory E-DII scores including diet and supplements compared with the most pro-inflammatory E-DII tertile (HR_T1vsT3=0.49; 95%CI=0.31–0.79) (Table 2). We observed a non-statistically significant reduced risk for E-DII from diet and supplements with total cancer mortality (HR_T1vsT3=0.57; 95%CI=0.29–1.10) and with colorectal cancer-specific mortality (HR_T1vsT3=0.58; 95%CI=0.28–1.22) (Table 2). The association with all-cause mortality was attenuated and no longer statistically significant when E-DII scores from diet only were used as the exposure variable (HR_T1vsT3=0.72; 95%CI=0.46–1.12; Table 3).

Table 2.

Association between post-cancer diagnosis E-DII score from diet plus supplements and mortality outcomes among 463 colorectal cancer subjects in the WHI-DM and OS, Women’s Health Initiative, 1993–2015.

	Most anti-inflammatory diet E-DII tertile 1(−6.80, −3.91)	E-DII tertile 2 (−3.90, −2.09)	Most pro-inflammatory diet E-DII tertile 3 (−2.08, 3.25)	P-non linearitya	P-trendb
N	155	154	154
All-cause mortality
Number of deaths	37	60	65
Age and energy - adjusted HR (95% CI)	0.44 (0.29–0.66)	0.84 (0.59–1.20)	1.00 (reference)	<.0001	NA
Multivariabl e-adjusted HR (95% CI)c	0.49 (0.31–0.79)	1.10 (0.74–1.64)	1.00 (reference)	0.001	NA
Total cancer mortality
Number of deaths	21	32	35
Age and energy - adjusted HR (95% CI)	0.49 (0.28–0.84)	0.85 (0.52–1.37)	1.00 (reference)	0.07	0.12
Multivariabl e-adjusted HR (95% CI)d	0.57 (0.29–1.10)	1.06 (0.60–1.87)	1.00 (reference)	0.10	0.58
Colorectal cancer-specific mortality
Number of deaths	13	31	34
Age and energy - adjusted HR (95% CI)	0.33 (0.17–0.63)	0.86 (0.53–1.39)	1.00 (reference)	0.002	NA
Multivariabl e-adjusted HR (95% CI)c	0.58 (0.28–1.22)	1.19 (0.67–2.12)	1.00 (reference)	0.02	NA

^a.P-non linearity was assessed using the restricted cubic spline function within the Cox proportional hazards regression model where three knots at 5^th, 50^th and 95^th percentile of the exposure variable were added. A significant P-non linearity value indicates the relationship is not linear.

^b.P-trend was calculated by using continuous E-DII variable in the Cox proportional hazards model if the linear relationship assessed with restricted cubic spline held, otherwise, NA was reported.

^c.Model was stratified by WHI components (OS, DM-intervention, DM-control) due to PH assumption violation and was adjusted for age group at baseline (<=66, >66 years old), race/ethnicity (White non-Hispanic, Hispanic/Latino, Black/African-American, other), smoking status at baseline (never smoked, past smoker, current smoker, missing), income levels (<20,000, 20,000–49,999, >=50,000, missing), cancer stage (localized, regional, distant, unknown), education (high school or below, some college, college, postgraduate, missing), years from cancer diagnosis to FFQ (continuous), baseline physical activity in MET-h/week (0, 0–3,3–9,9+, missing), total energy intake per day (continuous), body mass index at baseline (underweight, normal, overweight, obese, missing), cancer differentiation grading (anaplastic, well differentiated, moderately differentiated, poorly differentiated, unknown), with the covariate of time-dependent status before and after post-diagnosis FFQ in the model.

^d.Model was stratified by WHI components and income levels due to PH assumption violation and adjusted for other covariates listed in c.

Table 3.

Association between post-cancer diagnosis E-DII score from diet only and mortality outcomes among 463 colorectal cancer subjects in the WHI-DM and OS

	Most anti-inflammatory diet E-DII tertile 1 (−5.96, −2.25)	E-DII tertile 2 (−2.24, −0.19)	Most pro-inflammatory diet E-DII tertile 3 (−0.18, 3.82)	P-non linearitya	P-trendb
N	154	155	154
All-cause mortality
Number of deaths	41	62	59
Age and energy - adjusted HR (95% CI)	0.59 (0.39–0.88)	0.97 (0.68–1.40)	1.00 (reference)	0.01	NA
Multivariable-adjusted HR (95% CI)c	0.72 (0.46–1.12)	1.01 (0.68–1.50)	1.00 (reference)	0.08	0.40
Total cancer mortality
Number of deaths	20	39	29
Age and energy - adjusted HR (95% CI)	0.59 (0.33–1.06)	1.28 (0.79–2.09)	1.00 (reference)	0.09	0.16
Multivariable-adjusted HR (95% CI)c	0.86 (0.45–1.66)	1.36 (0.80–2.34)	1.00 (reference)	0.48	0.76
Colorectal cancer specific mortality
Number of deaths	14	36	28
Age and energy - adjusted HR (95% CI)	0.45 (0.24–0.87)	1.23 (0.74–2.03)	1.00 (reference)	0.01	NA
Multivariable-adjusted HR (95% CI)c	0.75 (0.36–1.57)	1.43 (0.81–2.52)	1.00 (reference)	0.20	0.97

^a.P-non linearity was assessed using the restricted cubic spline function within the Cox proportional hazards regression model where three knots at 5^th, 50^th and 95^th percentile of the exposure variable were added. A significant P-non linearity value indicates the relationship is not linear.

^b.P-trend was calculated by using continuous E-DII variable in the Cox proportional hazards model if the linear relationship assessed with restricted cubic spline held, otherwise, NA was reported.

^c.Model was stratified by WHI components (OS, DM-intervention, DM-control) due to PH assumption violation and was adjusted for age group at baseline (<=66, >66 years old), race/ethnicity (White non-Hispanic, Hispanic/Latino, Black/African-American, other), smoking status at baseline (never smoked, past smoker, current smoker, missing), income levels (<20,000, 20,000–49,999, >=50,000, missing), cancer stage (localized, regional, distant, unknown), education (high school or below, some college, college, postgraduate, missing), years from cancer diagnosis to FFQ (continuous), baseline physical activity in MET-h/week (0, 0–3,3–9,9+, missing), total energy intake per day (continuous), body mass index at baseline (underweight, normal, overweight, obese, missing), cancer differentiation grading (anaplastic, well differentiated, moderately differentiated, poorly differentiated, unknown), supplement intake (yes/no), with the covariate of time-dependent status before and after post-diagnosis FFQ in the model.

After excluding women from the WHI-DM-intervention (n=353), associations were attenuated and not statistically significant for E-DII from diet and supplement. The association for E-DII from diet plus supplements with all-cause mortality was HR_T1vsT3=0.74 (95%CI=0.43–1.28), and for E-DII from diet only, the HR_T1vsT3=0.90 (95%CI=0.54–1.49) (Table 4). In the sensitivity analysis including subjects who completed FFQs after 6 months of cancer diagnosis, we found a slightly stronger effect for E-DII from diet plus supplements on all-cause mortality (HR_T1vsT3=0.40, 95%CI=0.24–0.67), and for E-DII from diet only (HR_T1vsT3=0.66, 95%CI=0.40–1.07) (Table 4). Adjusting for the baseline E-DII score slightly strengthened the association between post-diagnosis DII and colorectal cancer specific mortality compared to the main analyses (HR_T1vsT3=0.40; 95%CI=0.18–0.88; Supplemental Table 1, Online Resource 1), but associations of E-DII from diet only did not change materially (Supplemental Table 2, Online Resource 2).

Table 4.

Sensitivity analyses of multivariable-adjusted HRs and 95% CIs of all-cause mortality in relation to post-cancer diagnosis E-DII scores among colorectal cancer subjects in the WHI-DM and OS, Women’s Health Initiative, 1993–2015.

	Most anti-inflammatory diet E-DII tertile 1 (−6.80, −3.91)	E-DII tertile 2 (−3.90, −2.09)	Most pro-inflammatory diet E-DII tertile 3 (−2.08, 3.25)	P-non linearitya	P-trendb
Sensitivity analysis of excluding DM-intervention arm
E-DII from diet plus supplements
Number of deaths/total	31/113	47/117	46/123
Multivariable-adjusted HR (95% CI)c	0.74 (0.43–1.28)	1.35 (0.84–2.18)	1.00 (reference)	0.01	NA
E-DII from diet only
Number of deaths/total	35/115	44/110	45/128
Multivariable-adjusted HR (95% CI)d	0.90 (0.54–1.49)	1.05 (0.66–1.68)	1.00 (reference)	0.35	0.96
Sensitivity analysis of excluding participants whose FFQs completed within 6 month of cancer diagnosis
E-DII from diet plus supplements
Number of deaths/total	31/129	44/122	51/123
Multivariable-adjusted HR (95% CI)c	0.40 (0.24–0.67)	0.76 (0.47–1.24)	1.00 (reference)	0.002	NA
E-DII from diet only
Number of deaths/total	32/128	42/120	48/126
Multivariable-adjusted HR (95% CI)d	0.66 (0.40–1.07)	0.89 (0.55–1.42)	1.00 (reference)	0.14	0.19

^a.P-non linearity was assessed using the restricted cubic spline function within the Cox proportional hazards regression model where three knots at 5^th, 50^th and 95^th percentile of the exposure variable were added. A significant P-non linearity value indicates the relationship is not linear.

^b.P-trend was calculated by using continuous E-DII variable in the Cox proportional hazards model if the linear relationship assessed with restricted cubic spline held, otherwise, NA was reported.

^c.Model was adjusted for WHI components, age group at baseline, race/ethnicity, smoking status at baseline, income levels, cancer stage, education, years from cancer diagnosis to FFQ, baseline physical activity in MET-h/week, total energy intake per day, body mass index at baseline, cancer differentiation grading, with the covariate of time-dependent status before and after post-diagnosis FFQ in the model.

^d.Model was adjusted for all variables in c and also dietary supplement intake

As shown in the Supplemental Table 3 (Online Resource 3), in the stratified analyses, the protective effect of the most anti-inflammatory diets and supplements on all-cause mortality was present among cancer survivors with BMI>=25kg/m², diagnosed with colon cancers (as opposed to rectal cancers), or with regional/distant cancer stage or well/moderately differentiated cancer grade, but not among their counterparts However, given the small sample size and case number, tests for interaction were under-powered.

Discussion

In this prospective study of postmenopausal women who were diagnosed with colorectal cancer, consumption of more anti-inflammatory dietary patterns and supplements after cancer diagnosis was associated with a lower risk of all-cause mortality. Associations between E-DII scores and total cancer mortality or colorectal cancer-specific mortality were weaker and confidence intervals included1. An anti-inflammatory diet alone, with adjustment for supplement use, also was associated with lower risk of all-cause mortality while comparing the most anti-inflammatory to pro-inflammatory tertile, though the effect size was smaller and not statistically significant. Given the small study size, findings need confirmation in future, larger prospective studies.

Although many prior studies have examined pre-diagnostic diet and colorectal cancer incidence [6] and mortality [43,44], only a few studies focused on the effect of post-diagnostic dietary patterns on colorectal cancer survival [20, 21, 45]. We previously reported that anti-inflammatory DII scores were associated with better diet quality scores on other dietary indices including the Alternate Healthy Eating Index (AHEI), Dietary Approaches to Stop Hypertension (DASH) and HEI-2010 [46]. Thus, our results showing that an anti-inflammatory diet after colorectal cancer diagnosis was associated with lower all-cause mortality is consistent with some previous studies reporting associations between diet quality after colorectal cancer diagnosis and mortality, including the aforementioned Cancer Prevention Studies II Nutrition Cohort [21]. In one study of 1,009 stage III colon cancer patients enrolled in a randomized adjuvant therapy trial (median follow-up of 5.3 years with 251 deaths), a higher Western dietary pattern after diagnosis was associated with a significantly worse overall survival (HR_Q5VSQ1=2.32; 95%CI=1.36–3.96, P-trend<0.001), though a “prudent” dietary pattern was not associated with all-cause mortality [9]. In the Nurses’ Health Study (NHS), 1,201 women diagnosed with stage I–III colorectal cancer were followed for a median of 11.2 years and 435 deaths were documented [10]. In this study, the AHEI-2010, alternate Mediterranean Diet score (aMED). DASH score and two a posteriori dietary patterns, Western and prudent, were computed based on FFQs completed at least 6 months after diagnosis. Among these five dietary patterns, only higher AHEI-2010 score was significantly associated with lower overall mortality (HR_Q5vsQ1=0.71, 95%CI=0.52–0.98, P-trend=0.01), and none of the dietary patterns were associated with colorectal cancer-specific mortality [10]. In a German prospective cohort study, in which diet was assessed at a median of 6 years after cancer diagnosis in 1,404 colorectal cancer survivors, better adherence to two a priori-defined post-diagnostic dietary patterns [Modified Mediterranean Diet Score (MMDS) and healthy Nordic Food Index (HNFI)] were associated with 52% and 37%, respectively, lower all-cause mortality [5]. In addition, two other studies reported significant associations between post-diagnostic glycemic load and dietary insulin load in relation to overall mortality among colorectal cancer survivors [47,48]. The combined analyses of the NHS and Health Professionals Follow-up Study (HPFS) found that higher post-diagnostic insulin load assessed within 4 years of diagnosis among 2,006 colorectal cancer patients, was also associated with higher colorectal cancer specific mortality (HR_Q5vsQ1=1.82, 95%CI=1.20–2.75, P-trend=0.006) [48].

We found a stronger association with all-cause mortality using the E-DII from diet plus supplements compared to the association using E-DII calculated from diet only. One possible reason is that results may be confounded by socioeconomic status or other healthy lifestyle factors that tend to coincide with supplement use. We analyzed major characteristics between supplement users and non-users and found supplement users had lower BMI and were more likely to have localized cancer stage compared to non-users. We controlled for these potential confounders in the multivariable model. However, residual or unmeasured confounding may still be present. Another possible explanation is that in this population, supplement intake may have an anti-inflammatory effect or other beneficial health effects which improve survival, especially for all-cause mortality. While dietary supplements are not recommended for cancer prevention [6] and have not reduced risk of mortality in randomized clinical trials among the general population [49,50], there is some evidence suggesting dietary supplements may reduce mortality among cancer survivors [51–53]. In the overall WHI, multivitamin use was not associated with risk of colorectal cancer or total mortality [54]. However, among WHI subjects diagnosed with invasive breast cancer during follow-up (n=7,728), multivitamin use reported at baseline and close to the time of diagnosis was associated with reduced breast cancer mortality during mean follow-up of 7.1 years [52]. Similar analyses among WHI participants diagnosed with colorectal cancer have not been performed. Colorectal cancer survivors have reported increasing multivitamin and other supplement use after diagnosis [55,56], thus, examining the association between supplement use and mortality after colorectal cancer diagnosis is an important area for future research.

We did not observe a statistically significant association between E-DII and colorectal cancer-specific mortality. This may be due to limited statistical power, but anti-inflammatory diets may also be related to other causes of mortality. In a previous investigation in WHI, post-diagnostic E-DII was associated with CVD mortality among breast cancer survivors [38]. Comorbidities at diagnosis among colorectal cancer survivors are prevalent; the most common comorbidities include diabetes, chronic obstructive pulmonary disease, and congestive heart failure, which may be major contributors to death in our population [57]. In our study, while comorbidity data were available, these data were not time-specific to the time of cancer diagnosis, and given the small size of our study, we had limited ability to further investigate associations with these cause-specific mortalities. The association between post-diagnostic E-DII from diet+supplements and all-cause mortality was attenuated and became non-significant when the DM-intervention arm was excluded, which may be due to the reduced sample size or the DM-intervention arm having a stronger E-DII and mortality association compared to the other arms as a result of their overall healthier lifestyle.

There is growing evidence for the strong connection between chronic inflammation and cancer development and progression [58,59]. Inflammation plays a role in colorectal cancer maintenance, progression, and metastasis, thus worsening survival among cancer patients [59] which is mainly through NF-κB pathways [60]. Epidemiologic data also suggest that systemic inflammation in cancer patients predicts higher overall mortality and non-cancer mortality [61–63]. Proposed mechanisms include upregulation of innate immune/inflammatory responses [62], elevated systematic oxidative stress [64], cyclo-oxygenase (COX)-2 overexpression, and p53 alteration [61].

Strengths of our study include a well-characterized prospective cohort of postmenopausal women with relatively detailed information on colorectal cancer patients and potential confounders, follow-up for a median of 11.6 years, and diet assessed prospectively. The DII, which has been evaluated for construct validity in the WHI, assesses the overall dietary quality with regard to inflammatory potential [30]. Limitations of our study include possible misclassification of E-DII score because diet may change during follow-up, and diet and supplement use were assessed only at one time point after cancer diagnosis in our study. However, DII scores were found to be relatively stable in the longitudinal timeframe in the WHI-OS and DM previously [65]. Measurement error in dietary assessment is another unavoidable limitation, which could exist with any dietary assessment tool. In the WHI specifically, a previous calibration study found self-reported protein and total energy intake was underreported from the FFQ as compared to recovery biomarkers [66]. In addition, 13 dietary components of the DII were not available from the WHI FFQ. However, the range of DII scores may rely more on the amount of DII components actually consumed rather than on the number of DII components included in the scoring [67]. Any misclassification in dietary inflammatory potential discussed above is likely to be non-differential, and subsequently may have attenuated our results toward the null. Residual or unmeasured confounding may still be a possibility in our study; for example, we used cancer stage and grade as proxy for the unavailable data on colorectal cancer treatment. In addition, data on intake of anti-inflammatory drugs, presence of a stoma, or comorbidities after cancer diagnosis were not available in this study. The small sample size and number of outcomes limited statistical power to perform interaction tests between E-DII score and important covariates, and some stratified associations were underpowered. Finally, our sample included only women who were postmenopausal and the majority are White non-Hispanic, therefore, generalizability is limited.

In summary, in this prospective study of postmenopausal women who were diagnosed with colorectal cancer, higher consumption of more anti-inflammatory diets and supplements after diagnosis was associated with lower risk of all-cause mortality, and was more modestly associated with lower total cancer and colorectal cancer-specific mortality. Our findings suggest that consuming dietary patterns and supplements with more anti-inflammatory potential after cancer diagnosis may improve overall survival among women diagnosed with colorectal cancer. Future large cohort studies are needed to replicate these findings.