Adherence to the 2015 Dutch dietary guidelines and risk of non-communicable diseases and mortality in the Rotterdam Study

This study was embedded in three sub-cohorts of the Rotterdam Study (RS), a population-based prospective cohort in Rotterdam, the Netherlands. Details of the study design and participants are described elsewhere [4]. Briefly, the first sub-cohort (RS-I) was established in 1990, and all residents aged 55 years and over living the Ommoord district of Rotterdam were invited to participate. Of the 10,215 eligible individuals, 7983 participated (78.1%). In the year 2000, the study was extended with a second sub-cohort (RS-II) of 3011 participants (out of 4472 invited, 67.3%) who had moved into the area or who had become 55 years of age since the start of the study. A third sub-cohort (RS-III) was established in 2006, in which 3932 participants aged 45 years and over were included (out of 6057 invitees, 64.9%). All participants visited the research center at baseline and subsequently every 3 to 5 years for detailed follow-up measurements.

The Rotterdam Study has been approved by the medical ethics committee according to the Wet Bevolkingsonderzoek ERGO (Population Study Act Rotterdam Study), executed by the Ministry of Health, Welfare and Sport of the Netherlands, and written informed consent was obtained from all study participants.

For the current analysis, we included all participants with reliable dietary data at baseline n = 9701; 5433 from RS-I, 1624 from RS-II, and 2644 from RS-III; Fig. 1). Per analysis, for each of the ten diseases, prevalent cases of that particular disease and participants with incomplete follow-up data on incidence on the disease were excluded, resulting in a population for analysis ranging from 6217 (for depression) to 9627 (for cancer) participants for the disease incidence analyses.

Dietary intake was assessed at baseline using validated food frequency questionnaires (FFQ). A slightly different approach was applied to the first two cohorts of the Rotterdam Study (RS-I and RS-II) than to the third cohort (RS-III). For the first two cohorts, an FFQ was applied in a two-stage approach. In the first stage, participants indicated which foods they consumed at least twice a month in the preceding year using a self-administered checklist of 170 food items. In a second stage, a trained dietician used this list to identify how often and in which amounts the foods were consumed. This FFQ was validated against fifteen 24 h food records and four 24 h urinary urea excretion samples in a subsample of the Rotterdam Study (n = 80), which demonstrated that it was able to adequately rank participants according to their intake: Pearson’s correlation for nutrient intakes with the food records ranged between 0.44 and 0.85 and Spearman’s correlation for protein intake against urinary urea was 0.67 [4]. For the third cohort, a self-administered semi quantitative FFQ was used to assess dietary intake. This FFQ was based on 389 items and was previously validated in two other Dutch populations using a 9-day dietary record [5] and a 4 week dietary history [6], which showed Pearson’s correlations for intakes of different nutrients varying from 0.40 to 0.86. For each food item, the frequency of consumption (in times per month or per week), the number of servings per day (expressed in standardized household measures) as well as the preparation methods were included. Information on portion size, type of food item, and preparation method were collected. Nutrient data were calculated from the Dutch Food Composition Table, using 1993’s version for RS-I, the 2001’s version for RS-II, and 2011’s update for RS-III to account for the changes in nutritional composition of foods. We excluded participants who had an unreliable dietary intake according to the trained nutritionist who performed the interview or because their estimated daily energy intake was implausible, for which cut-offs were set at <500 or >5000 kcal/day.

The Dutch Dietary Guidelines 2015 describe a general advice to follow a dietary pattern with more plant-based and less animal-based food and further specified recommendations for 15 food groups and for the use of dietary supplements ([1] and Supplemental Table 1). Briefly, they recommend to consume sufficient vegetables and fruits (≥200 g/day for both), whole-grain products (≥90 g/day), legumes (at least once weekly), unsalted nuts (≥15 g/day), fish (1 serving/week); and tea (3 cups/day); to replace refined cereal products by whole-grain products; to replace butter, hard margarines and cooking fats by soft margarines, liquid cooking fats and vegetable oil; and to replace unfiltered coffee by filtered coffee; and to limit the use of red meat, particularly processed meat (quantity unspecified); sugar-containing beverages; alcohol (none or ≤1 glass/day); and salt (≤6 g/day). The use of nutrient supplements is not advised except for people who belong to a group to which specific supplementation advice applies. From these items, we were unable to evaluate filtered coffee and dietary supplement use in our study population because we had no complete information on these items. For the remaining 14 items, we further quantified the guidelines, using additional information from the Netherlands Nutrition Center and Dutch food consumption surveys for the items where no specific cut-offs were provided in the original guideline (Supplemental Table 1). Subsequently, we scored every participant as adhering to this item (‘yes’ scored as 1) or not adhering to the item (‘no’ scored as 0; Table 1). As overall measure of adherence to the guidelines, we used the sum of the number of items adhered to, with a theoretical range from 0 (no adherence) to 14 (full adherence).

Age at baseline was calculated as the number of years between the date of birth and date of baseline visit to the research center. Information on smoking status and duration was collected through self-report and categorized as never, past, or current smoker. Information on paid employment at baseline (yes/no), and educational level was self-reported at baseline. Educational level was categorized into: primary education with or without a partially completed higher education (primary); lower vocational or lower secondary education (lower); intermediate vocational education and or general secondary (intermediate); or higher vocational or university education (higher). Women’s menopausal status at baseline (post-menopausal, yes/no) was self-reported.

Information on daily energy intake was obtained from the previously described FFQ. Two different questionnaires were used to assess physical activity: For RS-I and RS-II, physical activity was assessed using a validated adapted version of the Zutphen Physical Activity Questionnaire [15]; and for RS-III, physical activity was assessed using the LASA Physical Activity Questionnaire (LAPAQ) [9]. Both questionnaires included questions on walking, cycling, gardening, sports, and housekeeping. Data were recalculated into metabolic equivalent of task (MET)-hours per week for each participant [16]. As different methods were used to estimate physical activity in the different cohorts, MET-hours of the participants were divided into cohort-specific quintiles. At baseline, weight (kg) and height (cm) were measured at the research center, and BMI (kg/m²) was calculated.

To assess the association of adherence to the dietary guidelines with mortality risk and incidence of diseases, we used multivariable Cox proportional hazard models. We tested for non-linearity by including a quadratic term of the dietary guideline score in the models. Hazard ratios (HRs) for mortality or disease incidence were calculated per item increase in adherence to the guidelines and for quintiles of adherence to the dietary guidelines with the lowest quintile as reference.

The basic model (model 1) was adjusted for cohort, age at dietary assessment, and sex. The confounder model (model 2) was additionally adjusted for educational level, employment status, smoking status, physical activity, and energy intake. Because we considered BMI to be a potential intermediate in the association of diet quality with disease, we additionally included BMI in a separate model (model 3). We evaluated effect modification by age, sex, total energy intake, and BMI by including the interaction term of the dietary guideline score with the covariable in models 2 and 3. In addition, an interaction term between smoking and the dietary guideline score was evaluated for COPD and lung cancer; and interaction of menopausal status and the diet score was evaluated for breast cancer.

To examine the robustness of our findings, we performed several sensitivity analyses. First, we repeated all analyses in each of the three cohorts separately. Second, we repeated analyses after excluding participants who died (for the mortality analysis) or developed the disease of interest (for disease incidence) within the first two years after follow-up. Third, among women, we additionally adjusted our findings for menopausal status. Fourth, we additionally adjusted for duration of smoking for analyses with COPD and lung cancer as outcomes. Fifth, to check if associations were not driven by one specific guideline for a specific food group, we repeated our main analyses by excluding one of the 14 individual guidelines from the total dietary guideline score one at a time.

To assess the proportion of disease attributable to poor adherence to the 2015 Dutch Dietary Guidelines in the Netherlands, we computed Population Attributable Risks (PAR) for diseases that were significantly associated with adherence to dietary guidelines using the following equation: PAR% = 100 × P_e(RR − 1)/(P_e(RR − 1) + 1), where P_e is the prevalence of the exposure (i.e., adherence to dietary guidelines). We calculated PAR for less than 25%, less than 50%, or less than 75% adherence to the guidelines.

To reduce potential bias associated with missing data, missing values of covariables were multiple imputed (n = 10 imputations) [17], according to the Fully Conditional Specification method (predictive mean matching), assuming no monotone missing pattern. As effect estimates were similar before and after imputation, we only report pooled effect estimates after the multiple imputation procedure. Analyses were performed using SPSS version 21.0 (IBM Corp., Armonk, NY, USA).

Median age of the participants at baseline was 64.1 years (95%-range 49.0–82.8) (Table 1 and Supplemental Table 2). Participants had a median energy intake of 2089 kcal/d (95% range 1155–34,891) and had a median dietary guideline adherence score of 7 (95% range 3–10). None of the participants fully adhered to the guidelines. For the individual dietary guideline items, adherence to the recommendation to limit sugar-containing beverages was high (82.9% of the population), whereas adherence to the recommended legume and nut intake was low (14.3 and 17.0% of the study population, respectively). Characteristics of study participants enrolled in the cohort but without information on dietary intake (n = 5225) are presented in Supplemental Table 3. This group was on average slightly older and more often had a lower educational level as compared to the participants with dietary data (n = 9701).

Median follow-up time for mortality was 13.5 years (range 0–27.0), during which 4592 out of 9701 participants died (Table 2). After adjustment for confounders (model 2), adherence to the dietary guidelines was associated with a lower mortality risk (HR 0.97 per item adherence; 95% CI 0.95, 0.98). This effect estimate did not change after additional adjustment for BMI (model 3, Table 2). Participants in the highest quintile of adherence to the guidelines had on average a 14% lower risk of dying than participants in the lowest quintile (HR 0.86; 95% CI 0.78, 0.95; p-trend over the quintiles <0.001; Table 2).

Median follow-up time for incidence of non-communicable diseases ranged from 7.3 years for T2D to 11.8 years for stroke. Further details on follow-up time and number of cases per disease provided in Table 3. After adjustment for confounders (model 2), adherence to the dietary guidelines was significantly associated with a lower risk of stroke (HR 0.95; 95% CI 0.92, 0.99) and COPD (HR 0.94; 95% CI 0.91, 0.97), but not significantly with risk of HF, T2DM, or dementia (Table 3). Furthermore, higher adherence to the guidelines was also associated with a lower risk of developing colorectal cancer (HR 0.90; 95% CI 0.84, 0.96), but not with breast cancer. Inverse associations of dietary guideline adherence with lung cancer and CHD in basic models (model 1) were driven by confounders since associations were no longer significant in model 2. Finally, adherence to the guidelines was associated with a borderline lower risk of depression (HR 0.97; 95% CI 0.95, 0.999). Additional adjustment for BMI (model 3) did not change the effect estimates for any of the diseases as compared to model 2 (Table 3). Quadratic terms of the dietary guideline score were not significant for any of the outcomes. In line with this, analyses with the dietary guidelines adherence score in quintiles supported findings for the continuous score and associations with stroke, colorectal cancer, COPD, and depression appeared to be approximately linear (Supplemental Table 4).

We examined effect modification by age, sex, total energy intake, and BMI for all outcomes. There was a significant interaction of the dietary guideline sore with sex on stroke (p = 0.008), but not with any of the other diseases (p > 0.20). Stratification by sex showed that the association of adhering to the dietary guidelines with stroke was only present among men (HR 0.90; 95% CI 0.85, 0.95) and not among women (HR 0.99; 95% CI 0.95, 1.04). Further stratification of the women by menopausal status at baseline showed no clear differences in associations of diet quality with stroke between post-menopausal women versus peri-menopausal or pre-menopausal women. We also observed a significant positive interaction between age and adhering to the guidelines for all-cause mortality risk (p = 0.03), suggesting a less strong association with older age; and a negative interaction for depression (p = 0.02), suggesting stronger associations with older age. Stratification by age groups showed that, indeed, associations of adhering to the guidelines with mortality risk were stronger at younger ages [e.g. HR 0.95 (0.92; 0.99) for <65 years and HR 0.98 (0.96, 0.999) for ≥65 years]; whereas associations with depression were only present at older ages [e.g. HR 1.00 (0.96; 1.05) for <65 years and HR 0.96 (0.92; 1.00) for ≥65 years]. Furthermore, there was a positive interaction of energy intake with the guideline score for depression (p = 0.02); and a negative interaction of BMI with the dietary guideline score for breast cancer (p = 0.02), but not with any of the other outcomes. We observed no interaction of the dietary guideline score with smoking status for COPD (p = 0.64) or for lung cancer (p = 0.70); or with menopausal status for breast cancer (p = 0.56).

Effect estimates did not differ greatly between the three individual cohorts, although the variation was generally larger in the two youngest cohorts with shorter follow-up time and smaller sample sizes (Supplemental Table 5). Effect estimates of analyses from which we excluded participants who died or developed the disease of interest within the first two years of follow-up were similar to those obtained for the complete study population for all diseases, except for depression (Supplemental Table 6). For diet quality and depression, there was no longer an association after excluding incident depression cases in the first two years (n = 784) from our analyses (HR 0.99; 95% CI 0.96, 1.02; Supplemental Table 6). Additional adjustment for menopause status or for duration of smoking did not affect the results (data not shown). Finally, excluding each of the dietary components of the dietary guideline adherence score one by one resulted in similar associations with incident disease and mortality risk as observed for the total dietary guideline score (data not shown).

Based on the prevalence of adherence to the guidelines in our cohort and observed associations with mortality and disease, we estimated PAR proportions based on our study population. We estimated that when less than half of the dietary guidelines was adhered to (i.e., a score <7), this has most impact on colorectal cancer, with a PAR% of 29.8%, followed by COPD (PAR% = 19.8%) and stroke (PAR: 16.4%) (Supplemental Table 7).

We used a large population-based cohort with long term follow-up, information on several important potential confounders and a broad range of accurately measured incidence of diseases to evaluate the most recent dietary guidelines. Incident diseases were identified based on combined information from questionnaires, detailed measurements at our research center, and continuous monitoring of medical records. However, to interpret the findings some limitations need to be considered. First, we used two different FFQs, composed of different numbers of items. For example, the FFQ used in RS-I and RS-II had less detailed items on types of fish and legumes than the FFQ used in RS-III. Although this may have implications when studying specific nutrients, we expect that the FFQs are equally capable of estimating overall food-based dietary quality, since sensitivity analyses showed no major differences in dietary quality between cohorts. Unfortunately, the items in the FFQs did not distinguish between filtered versus unfiltered coffee or salted versus unsalted nuts, which are important distinctions in the new dietary guidelines. This may have led to an underestimation of the magnitude of the associations for the outcomes that have been particularly associated with coffee and salt consumption in the previous literature such as cardiovascular disease, type 2 diabetes and dementia [36‐38]. For some food groups, no quantitative cut-offs were provided in the dietary guidelines, e.g. for legumes the guideline is to eat them weekly, and for SCBs the recommendation is to minimize intake. For these components, the authors based their used cut-offs on additional information from the Netherlands Nutrition Center and Dutch food consumption surveys. Different interpretations may have resulted in slightly different cut-off values for these food groups [39].

Although a strength of our study is the use of multiple endpoints to provide a full overview of the construct validity and potential impact of the Dutch dietary guidelines, this consequently required multiple statistical tests, which may have increased the risk of chance findings. Furthermore, although we adjusted for many confounding variables and conducted several sensitivity analyses, conclusions regarding the causality of the observed associations cannot be made. Replication of these analyses in other populations and studies on the associations of diet quality with preclinical disease risk factors, such as hypertension, dyslipidemia, or brain pathologies, are required to provide stronger evidence on potential causality and to better understand underlying pathways.

The Dutch food-based dietary guidelines were developed on the basis of evidence for associations of nutrients, foods, and dietary patterns with incidence of major chronic diseases in cohorts and/or randomized controlled trials. We now also show that adherence to these combined dietary guidelines is associated with a lower risk of some, but not all, of these diseases. Implying that the 2015 Dutch dietary guidelines have moderate criterion validity for preventing incidence or major chronic diseases. However, every item higher adherence to the dietary guidelines was associated with a 3% lower all-cause mortality risk and a reduction in risk of stroke, depression, and colorectal cancer, again emphasizing the importance of a healthy diet in lowering risk of several chronic diseases. In our population, we observed that adherence to the dietary guidelines was suboptimal, with a median score of 8 out of a maximum of 14. None of the participants had the maximum score. In line with a recent analyses of dietary intake of 885 Dutch adults, we observed that dietary guideline adherence was particularly low for intake of legumes, nuts, and fish [39]. Although this may partly be explained because e.g. legume intake is a relatively new component in the guidelines, also current legume, nut and fish intake is low among the Dutch population, suggesting there is plenty of opportunity to improve dietary quality in the Netherlands [2]. Poor adherence to the dietary guidelines attributed most to colorectal cancer, COPD and stroke with PARs varying from 16 to 30% for following less than 50% of the dietary guidelines. This suggests that many cases of these diseases can be attributed to poor adherence to dietary guideline, and that policies to improve adherence to these new dietary guidelines can have vast implications for public health.