Pre- and post-diagnostic dairy intake in relation to recurrence and all-cause mortality in people with stage I-III colorectal cancer

For these analyses we used data from adults with newly diagnosed CRC who participated in the ‘COlorectal cancer: Longitudinal, Observational study on Nutritional and lifestyle factors that influence colorectal tumour cancer recurrence, survival and quality of life’ (COLON) study (NCT03191110, ClinicalTrials.gov). Participants were recruited from 11 hospitals in the Netherlands between August 2010 and February 2020, and had no hereditary CRC syndrome, a history of CRC or inflammatory bowel disease [11]. Details of this study have been described previously [11]. For the current analyses, we excluded some participants (Fig. 1), among which people diagnosed with stage IV disease (n = 150), as we expect the potential role of lifestyle to be very limited in the poor prognosis of these persons. The final analytical sample sizes were as follows: 1812 (all-cause mortality) and 1341 (recurrence) for pre-diagnostic models, and 1672 (all-cause mortality) and 1236 (recurrence) for post-diagnostic models (Fig. 1). Three participants were lost to follow-up due to moving abroad. The COLON study was approved by a medical ethics committee (region Arnhem-Nijmegen, 2009–349), and all study participants provided written informed consent.

Upon diagnosis, participants filled out a semi-quantitative food frequency questionnaire (FFQ) of 204 items [12, 13], reflecting dietary intake in the month prior to diagnosis. The exact date of diagnosis was not uniformly available, but in practice participants were included in the study by the health care practitioner only a few days after the diagnosis. The median time between this date of inclusion and completion of the first FFQ was 5 days (IQR: 3–10). At six months after diagnosis participants also completed the FFQ, reflecting dietary intake in the month prior to filling out the FFQ. Dairy product intakes and energy intakes per day were calculated, based on frequency of intake and portion sizes and using the online Dutch Food Composition Table [14], and categorized into total, low-fat and high-fat dairy. Total dairy comprised of all dairy products included in low- and high-fat dairy. Low-fat dairy included low-fat or skimmed versions of milk, yoghurt, custard, and soft curd cheese, and buttermilk. High-fat dairy included whole and condensed milk, full-fat versions of yoghurt, custard and soft curd cheese, all other cheeses, ice cream, whipped cream, and butter. Total milk included all plain and sweetened milk, regardless of fat content. Ready-made breakfast drinks were not included, because not all ready-made breakfast drinks on the Dutch market contain dairy. Total yoghurt included all plain and sweetened yoghurt, yoghurt drinks and soft curd cheese, regardless of fat content. Total cheese included all hard and soft cheese, except for soft curd cheese. Intake of dairy types and products was adjusted for total energy intake using the energy residual method [15]. Further assessment of individual food items, such as skimmed milk, low-fat milk and high-fat milk, was not possible due to low median intakes and a low variation in intake. To improve interpretability, the predicted dairy intake at the median total energy intake was added to individual residuals.

Recurrence was defined as a local or regional cancer recurrence or a distant metastasis diagnosed after initial tumour resection. Recurrence data were requested from the Netherlands Cancer Registry via the Netherlands Comprehensive Cancer Organisation (IKNL), most recently updated in February 2018. All-cause mortality data were obtained from the municipal registrations (BRP), most recently updated in September 2021.

For recurrence, follow-up time was calculated starting from date of surgery until date of recurrence, until date of death, until the date recurrence status was updated, or until lost to follow-up, whichever came first. Follow-up time for mortality was defined as starting from date of surgery until date of death, until the last date mortality status was updated, or until lost to follow-up, whichever came first. If date of surgery was unavailable (n = 71), date of filling out the questionnaires was used. For post-diagnostic analyses, follow-up time was calculated starting from date of filling out the post-diagnostic questionnaires.

At diagnosis and six months after diagnosis, participants filled out a general questionnaire on demographics, anthropometrics, and lifestyle habits, including questions about age (continuous), sex (male/female), education (low, medium, high), body weight (kg), height (cm), smoking status (current, former, never), and calcium and vitamin D supplement use (including multivitamins, yes/no). Physical activity was assessed using the Short QUestionnaire to ASsess Health-enhancing physical activity (SQUASH) [16]. Moderate-to-vigorous physical activity (hours/week) included all activities with a metabolic equivalent value ≥ 3 [17]. Clinical data, such as disease stage (I-III), tumour location (proximal colon (caecum to the transverse colon), distal colon (splenic flexure to the sigmoid colon), rectum (rectosigmoid junction and rectum)), American Society of Anesthesiologists (ASA) classification (I-IV) and type of treatment (only surgery, surgery and chemotherapy, surgery and radiotherapy, surgery and chemotherapy and radiotherapy, no or unknown surgery, unknown treatment), were collected via the Dutch ColoRectal Audit [18].

To assess possible confounding by other dietary factors previously associated with CRC risk [2], we also calculated total dietary intake of wholegrains, red meat, processed meat, dietary fiber, and alcohol, as defined in [2].

Population characteristics are presented as medians [interquartile range (IQR)] or numbers (percentage). Sex-specific quartiles of pre- and post-diagnostic dairy intake were calculated.

Cox proportional hazards regression analyses were used to calculate Hazard Ratios (HR) and 95% Confidence Intervals (95%CI) for the associations between pre- and post-diagnostic dairy intake, and recurrence and all-cause mortality. No violation of the proportionality assumption for the Cox models was observed when visually inspecting log–log curves. For categorical analyses, the lowest quartile was used as the reference category. P_trend values were computed over quartiles of dairy intake using the medians of the corresponding quartiles. For continuous analyses increments were 100 g/day for total dairy, low-fat dairy, high-fat dairy, milk, and yoghurt, and 30 g/day for cheese. We first created a crude model with the exposure and outcome variables, that was adjusted for age and sex. Then, potential confounders were added to the model one by one, and were included when they changed the HR with > 10%. Based on literature, the following covariates were evaluated: disease stage, primary tumour location, BMI, education level, smoking status, moderate-to-vigorous physical activity, use of calcium supplements, use of vitamin D supplements, and total dietary intake of wholegrains, dietary fiber, red meat, processed meat and alcohol. Additionally, for the post-diagnostic dairy intake analyses, having a stoma, type of treatment, and BMI, smoking status, supplement use and dietary intake at 6 months post-diagnosis were evaluated for confounding. Finally, the adjusted model included age, sex, stage of disease and total energy intake (as part of the energy adjustment method [15]).

We also evaluated restricted cubic splines (RCS) to study linearity of the associations between dairy exposures and CRC prognosis, using the adjusted model. Knots were placed at the 5th, 50th, and 95th percentiles, and the graphs were truncated at the 1st and 99th percentile. The median intake of the first quartile of each exposure was used as the reference.

For linear associations, HRs are displayed continuously as well as in quartiles with p for trend, while for non-linear associations, only quartiles of dairy intake are displayed. All associations are visualized in plots.

Statistical analyses were performed using R Statistical Software (version 4.0.5). P-values below 0.05 and 95% confidence intervals not containing 1 were considered statistically significant.

Participants were on average 66 years (IQR: 61–72) at CRC diagnosis, and 37% was female (Table 1). Median BMI was 26.2 kg/m² (IQR: 24.0–28.7) and 10% of participants were current smokers. Furthermore, 66% had a tumour located in the colon, and stage III (43%) was more common than stage I (26%) or stage II (27%) (Table 1).

Participants in the highest quartile of total pre-diagnostic dairy intake were older, less often current smokers, and more often had a lower level of education, compared to participants in the lowest quartile of total dairy products. Also, they consumed less processed meat and alcohol, consumed more dietary calcium, and less often used calcium supplements. Participants in the highest quartile of low-fat dairy intake were less often current smokers, consumed more dietary calcium and less alcohol, more often used vitamin D supplements, and less often had an ASA score of I, compared to participants in the lowest quartile. Participants in the highest quartile of high-fat dairy products were older and more often had a lower level of education, consumed more dietary calcium, and less processed meat and alcohol and had a lower total energy intake than participants in the lowest quartile (Table 1). Median pre-diagnostic energy-adjusted intakes of total, low-fat and high-fat dairy were 270 (IQR: 177–383), 165 (IQR: 70–281), and 77 (IQR: 46–126) g/day, respectively (Table 1). Median post-diagnostic energy-adjusted intakes of total, low-fat and high-fat dairy were 282 (IQR: 184–392), 161 (IQR: 70–286), and 81 (IQR: 48–136) g/day, respectively (data not shown). Pearson correlation coefficients between pre- and post-diagnostic intakes ranged between 0.41 and 0.61 (data not shown).

For pre-diagnostic analyses, we observed 176 recurrences during a median follow-up time of 3.0 years (IQR: 2.0–4.2), and 301 participants died during a median follow-up of 5.9 years (IQR: 3.9–7.2). For post-diagnostic analyses, we observed 153 recurrences that occurred during a median follow-up time of 2.6 years (IQR: 1.7–3.8), and 245 participants died during a median follow-up of 5.5 years (IQR: 3.8–6.8).

As shown in Table 2 and Supplementary Fig. 1, when comparing the highest to the lowest quartile, a higher pre-diagnostic intake of total dairy was associated with a lower risk of recurrence (HR: 0.51, 95% CI 0.33–0.80; P_RCS: 0.06), and tended to be associated with all-cause mortality (HR_{Q4 vs Q1}: 0.72, 95% CI 0.52–1.01; P_RCS: 0.18). A higher intake of low-fat dairy before diagnosis was associated with a lower risk of recurrence (HR: 0.42, 95%CI 0.26–0.67, P_RCS: 0.008; Table 2 and Fig. 2A) and all-cause mortality (HR_{Q4 vs Q1}: 0.58, 95% CI 0.41–0.81; P_RCS < 0.001). RCS revealed a significant non-linear association between pre-diagnostic low-fat dairy intake and all-cause mortality, with intakes higher than the median of the first quartile (i.e. 21 g/day) associated with a reduced risk of mortality (Fig. 2A). Contrastingly, a higher intake of high-fat dairy before diagnosis tended to be associated with an increased risk of all-cause mortality (HR_{Q4 vs Q1}: 1.41, 95% CI 0.98–2.01; P_RCS: 0.030). Pre-diagnostic high-fat dairy intake was not associated with recurrence. A higher yoghurt intake before diagnosis seemed associated with a lower risk of all-cause mortality (HR_{Q4 vs Q1}: 0.70, 95% CI 0.50–0.99; P_RCS: 0.09), but not with recurrence. No associations were observed between total milk or cheese intake before diagnosis and recurrence or all-cause mortality.

As shown in Table 3 and Supplementary Fig. 1, post-diagnostic total dairy intake was not associated with recurrence or all-cause mortality. A higher intake of low-fat dairy after diagnosis tended to be associated with a lower risk of all-cause mortality (HR_{Q4 vs Q1}: 0.76, 95% CI 0.53–1.08; P_RCS: 0.011; Table 3 and Fig. 2B). RCS revealed a significant non-linear association between post-diagnostic low-fat dairy intake and all-cause mortality, with intakes higher than the median of the first quartile (i.e. 20 g/day) associated with a reduced risk of mortality (Fig. 2B). Post-diagnostic low- and high-fat dairy intakes were not associated with recurrence. A higher intake of high-fat dairy after diagnosis was associated with an increased risk of all-cause mortality (HR_{Q4 vs Q1}: 1.60, 95% CI 1.10–2.33; P_RCS: 0.006). Post-diagnostic milk, yoghurt, and cheese intake were not associated with recurrence or all-cause mortality.

Except for pre- and post-diagnostic low-fat dairy in relation to all-cause mortality, we observed no evidence of non-linearity in our RCS analyses, and findings of these analyses were in line with the Cox proportional hazard regression analyses (Fig. 2 and Supplementary Figure 1).