Assessment of the Dutch Healthy Diet index 2015 in the Lifelines cohort study at baseline

Large epidemiological studies enable investigation of diet–disease associations, but dietary complexity necessitates pattern-based approaches. The Lifelines cohort (baseline n=167,729) uses the Flower-FFQ to collect detailed dietary intake data. The Dutch Healthy Diet index 2015 (DHD2015-index) assesses adherence to the 2015 Dutch dietary guidelines and serves as a measure of diet quality. The aims were: (1) to assess the DHD2015-index in the Lifelines cohort to facilitate diet–disease research; and (2) to compare DHD2015-index scores derived from the detailed total Flower-FFQ versus the basic heart-FFQ only, given that only about half of participants completed all petal-FFQs. This work replaces a previously retracted 2022 article.

The paper situates the work within nutritional epidemiology emphasizing dietary pattern analysis over single-nutrient approaches due to food synergies and interactions. Dietary indices are established tools to capture adherence to guidelines. The DHD2015-index, based on the Dutch 2015 food-based dietary guidelines, has been previously developed and evaluated. Prior studies report generally higher diet quality among women than men and associations of better diet quality with higher age, socioeconomic status, and physical activity, and lower quality with smoking. The Flower-FFQ was developed and validated as a modular FFQ suited to large cohorts to reduce burden while enabling detailed nutrient/food group estimates.

Study population: Lifelines is a multi-generational cohort in three northern provinces of the Netherlands, initiated 2006, with long-term follow-up. Exclusions included severe psychiatric/physical illness, life expectancy <5 years, and insufficient Dutch language proficiency. Ethical approval was granted by UMCG; informed consent obtained. Dietary assessment: The Flower-FFQ comprises a main heart-FFQ (110 items) focused on major food groups/energy/macronutrients and three petal-FFQs (covering detailed food types and supplements), totaling 212 items. Heart-FFQ was administered at baseline (2007–2013); petal-FFQs at three subsequent assessments (2011–2014, 2012–2015, 2014–2017) in randomized order; assessments collectively considered baseline dietary intake. Frequency of consumption over the previous month was collected and converted to daily energy/nutrient intake using NEVO 2011. Plausibility of energy intake was screened using Willett’s cutoffs: men <800 or >4200 kcal/day; women <500 or >3500 kcal/day. Participation: 144,093 completed heart-FFQ; 129,030 (90%) plausible. Of these, 68,698 completed the total Flower-FFQ; 59,982 (87%) plausible. For 59,881 participants, plausibility was met on both instruments. DHD2015-index assessment: The index contains 16 components (vegetables, fruits, wholegrain products [adequacy and ratio], legumes, nuts, dairy [optimum], fish, tea, fats and oils [ratio], coffee [qualitative], red meat, processed meat, sweetened beverages and fruit juices, alcohol, salt, unhealthy choices). Each component is scored 0–10 (total 0–160). Food items were mapped to components; where necessary, assumptions (e.g., proportion wholegrain vs refined) were made based on the Dutch National Food Consumption Survey; items without a basis for assumptions were excluded. Due to unavailable data on coffee filtering and salt intake in the Flower-FFQ, these components were not included. For the heart-FFQ, only the adequacy component for wholegrain products (max 5) could be assessed; the ratio subcomponent was not available. Therefore, total score ranges were 0–140 for Flower-FFQ-based DHD2015 and 0–135 for heart-FFQ-based DHD2015. Other characteristics: Sex, age, socioeconomic status (education-based categories: low, moderate, high), smoking (current, former, never), and physical activity (short questionnaire to assess health-enhancing physical activity; MVPA calculated using MET thresholds ≥4.0 to <6.5 for moderate and ≥6.5 for vigorous) were collected via questionnaires. Anthropometrics (height, weight) were measured; BMI calculated. Statistics: Normality checked via Kolmogorov–Smirnov and Q–Q plots. Continuous variables (skewed) reported as medians with IQR; DHD2015 scores treated accordingly. Mann–Whitney U tested sex differences. Trends across DHD2015-index quartiles evaluated using Jonckheere–Terpstra. Agreement between Flower-FFQ and heart-FFQ DHD2015 indices assessed by Kendall’s tau-b correlations (with Fisher’s z CI) and Bland–Altman plots. Cross-classification into quartiles evaluated agreement; >50% in same quartile considered good. Significance at p<0.05; SPSS v25 used.

• Sample characteristics: Among Flower-FFQ completers (n=59,982), 40% men; median age men 47 (36–56), women 46 (38–54). Among heart-FFQ completers (n=129,030), 41% men; median age men 45 (36–54), women 44 (35–52). - DHD2015-index scores: Higher in women than men. Flower-FFQ medians: men 75 (65–85), women 81 (70–91). Heart-FFQ medians: men 68 (58–77), women 73 (63–82). - Associations: Higher DHD2015-index associated with older age, higher SES, more MVPA; inversely associated with smoking. Nutrient associations: positive with protein, dietary fiber, micronutrients; inverse with energy, carbohydrate, and fat intake (observed for both FFQs). - Agreement between Flower-FFQ and heart-FFQ indices: Median difference in total scores: men +5.8 (Flower higher), women +6.6. Excluding the grain ratio component from Flower-FFQ score reduced median differences to +3.8 (men) and +4.5 (women). Kendall’s tau-b for total scores: men 0.67 (95% CI 0.66–0.68), women 0.66 (0.66–0.67). Bland–Altman mean differences: men +5.9 (LOA −8.9 to 20.7), women +6.7 (LOA −8.3 to 21.7). Cross-classification into quartiles: same quartile 59% (men) and 60% (women); adjacent 37% (men) and 36% (women); non-adjacent 4% (both). - Component-level findings: Flower-FFQ scores higher than heart-FFQ for vegetables, wholegrain intake, fish, and fats/oils; lower for fruit and tea; differences generally small except fats/oils (median difference 4.8 points in men; 8.2 in women). Correlations ranged from poor for fats/oils (r=0.16; both sexes) to perfect for legumes, nuts, and alcohol (r=1.00); wholegrain adequacy correlations were acceptable to good (men r=0.38; women r=0.50). Highest component scores generally for red meat and alcohol; lowest for unhealthy choices.

The study achieved its aims by computing the DHD2015-index in the Lifelines cohort and comparing index calculations from detailed (Flower-FFQ) versus basic (heart-FFQ) dietary data. The strong agreement (tau-b ~0.66–0.67) and favorable cross-classification indicate that the heart-FFQ-based index can effectively rank individuals by diet quality when the more detailed Flower-FFQ data are unavailable. Nonetheless, component-level discrepancies—particularly the poor agreement for fats and oils and lower agreement for wholegrain intake—highlight the added value of the detailed petal-FFQs and the need for assumptions when using the heart-FFQ, which may reduce precision for certain components. Sex differences (higher scores in women) and associations with demographic and lifestyle factors align with prior literature, reinforcing construct validity. Given that ranking individuals is often the key requirement in diet–disease epidemiology, the heart-FFQ-based index is acceptable for large-scale analyses, with the Flower-FFQ remaining the preferred source for optimal component resolution and total score accuracy.

The DHD2015-index, reflecting adherence to the 2015 Dutch dietary guidelines, was computed for the Lifelines cohort using both Flower-FFQ and heart-FFQ data. The Flower-FFQ provides more optimal information and should be preferred for assessing the DHD2015-index. However, the heart-FFQ-based index shows good agreement in ranking participants by diet quality and is suitable when Flower-FFQ-derived indices are not available, acknowledging limited detail for some components. These indices can support investigations of diet–disease associations within Lifelines. Future work may focus on improving component-specific precision (e.g., fats/oils, grains) using targeted questions or calibration approaches.

• Dietary intake was self-reported via FFQs, which are prone to recall bias and social desirability bias. - The Flower-FFQ’s modular administration over approximately five years may not perfectly capture stable intake if dietary changes occur within that period, although dietary patterns are assumed relatively stable. - FFQs are less accurate for estimating absolute intake levels; they are better suited for ranking individuals. - Missing data on coffee preparation (filtered/unfiltered) and salt intake in the Flower-FFQ necessitated exclusion of these components from index calculations; the heart-FFQ lacked sufficient detail for the grain ratio component, potentially impacting total scores and component precision.