Moderate Consumption of Healthy Nordic Foods is Associated with Reduced Mortality in the Norwegian Women and Cancer Study: a Prospective Cohort Study

Background High adherence to healthy Nordic diets may enhance longevity. However, optimal intake levels of healthy Nordic foods are not known. Hence, in a large prospective cohort of women in Norway we examined all-cause mortality in relation to intake of ve food groups that are part of a healthy Nordic diet: Nordic fruits and vegetables, fatty sh, lean sh, wholegrain products, and low-fat dairy products. Methods A total of 87 899 women who completed a food frequency questionnaire between 1996 and 2004 were followed for mortality until the end of 2018. Cox proportional hazards regression models were used to examine the associations between consumption of the Nordic food groups and all-cause mortality. The food groups were examined as categorical exposures, and all but wholegrain products also as continuous exposures in restricted cubic spline models. Results A total of 9 168 women died during the 20-year follow-up. Nordic fruits and vegetables, fatty sh and low-fat dairy products were not linearly associated with mortality (p < 0.05). The optimal intake levels and hazard ratios (HR) and 95% condence intervals (CI) associated with these intakes were approximately 200 grams/day of Nordic fruits and vegetables (HR 0.84 (95% CI: 0.77–0.90)), 10–20 grams/day of fatty sh (HR 0.98 (95% CI: 0.92–1.03)) and 200 grams/day of low-fat dairy products (HR 0.94 (95% CI: 0.89–0.99)) compared to no consumption. High consumption of fatty sh ( ≥ 70 grams/day) was associated with increased mortality. Intake of wholegrain products of >120 grams/day was associated with lower mortality (HR 0.92 (95% CI: 0.85–0.99)) compared to < 60 grams per day. Lean sh consumption was not associated with mortality. After stratication by smoking status, the observed association for Nordic fruits and vegetables was only signicant in ever smokers with the optimal intake level at 250 grams/day (HR 0.78 (95% CI: 0.71-0.86)). rank women for fairly good for macronutrients (19). Another validation study that compared the relation between sh consumption registered by the FFQ and fatty acids composition in serum phospholipids concluded of in high-consuming populations was in serum phospholipids (20). In a study of the reproducibility of the FFQ, there were some indications of seasonal reporting bias, but the overall results in line with what has been found in studies on similar self-administered FFQs developed to assess habitual diet were tested with a Schoenfeld residuals observed protective effect of wholegrain products on mortality in the present analysis by meta-analyses of prospective cohort studies from the and The present results showed no further benets of 180 grams of wholegrain products per day. In the meta-analysis Aune et al. reductions in risk for whole observed up to an intake of 225 grams per day, but they found a non-linear association with all-cause mortality and a steeper reduction in risk at lower intake levels to our results, a study on wholegrain eaters by et al. in the meta-analyses found an inverse association between a calculated wholegrain score and statistical for meaningful discussions about study design and statistical analysis.

Information on vital status and cancer incidence was obtained by linkage to the National Population Registry and the Cancer Registry of Norway, using the unique 11-digit identity number assigned to all Norwegian citizens.
Participants completed a mailed self-administered questionnaire including questions about anthropometric, sociodemographic, reproductive and lifestyle factors. Most of the questionnaires included four pages of food frequency questions. The questionnaire used has previously been published elsewhere (17).
Follow-up questionnaires were mailed approximately every sixth year after recruitment.
A study on external validity of the NOWAC found no major source of selection bias (18).

Study participants
The baseline for this paper is partly the rst NOWAC mailing from 1996 to 1997 and 2003 to 2004 (response rate of 57% and 48%, respectively), and partly the second mailing (follow-up questionnaire) from 1998 to 1999 to those enrolled in 1991 to 1992 who had not been given the food frequency questions at enrolment (response rate of 81%). In total 101 316 women aged 41-76 at baseline were considered eligible for inclusion. Women who had emigrated (n = 3) and women with no follow-up (n = 13) were excluded. We further excluded women with implausible daily energy intake (< 2 500 kJ (n = 1 033) or > 15 000 kJ (n = 141)), and women with missing information on the following variables: Body mass index (BMI) (n = 2 272), physical activity (n = 8 548) and smoking habits (n = 1 407), leaving a total number of 87 899 women for the present analysis.

Dietary assessment
Diet was assessed using a semi-quantitative food frequency questionnaire (FFQ). The FFQ was designed to measure the typical diet during the past year with special emphasis on sh consumption. The response options were given with four to seven frequency categories ranging from never/seldom to six or more per week. Questions about portion size were included for some food items as natural units, such as number of carrots, or household units, such as tablespoons.
The FFQ used in NOWAC has been validated in several studies. Hjartåker et al. reported that the FFQ's ability, when compared to information from repeated 24hour dietary recalls, was good to rank women for foods eaten frequently and fairly good for macronutrients (19). Another validation study that compared the relation between sh consumption registered by the FFQ and fatty acids composition in serum phospholipids concluded that habitual intake of sh in highconsuming populations was re ected in serum phospholipids (20). In a study of the reproducibility of the FFQ, there were some indications of seasonal reporting bias, but the overall results were in line with what has been found in studies on similar self-administered FFQs developed to assess habitual diet (21).
The Norwegian Weight and Measurement Table with standardised portion sizes and weights was used to convert the consumption of food items to grams (22), and information about the nutrient content in foods was obtained from the Norwegian Food Composition Database (23). The calculations of daily intake of food items, energy and nutrients were made using a statistical program for SAS (SAS Institute Inc., Cary, NC, USA) developed at the Department of Community Medicine, University of Tromsø, for the NOWAC cohort. Missing frequency values were treated as no consumption, and missing portion sizes were set to the smallest portion size asked for.

Exposures
We included ve Nordic food groups that can be extrapolated to our dietary guidelines in the analysis. The questions in the FFQ that formed the basis for the construction of the food groups has been described previously (24). Based on the criteria set by Olsen et al., we have included fruits and vegetables produced in the Nordic climate without the use of external energy and that were available from the FFQ: broccoli/cauli ower, cabbage, carrots, swede, mixed vegetables (commonly a frozen mix of carrots, broccoli and cauli ower) and apples/pears (7). Consumption of Nordic fruits and vegetables was divided into four categories (grams/day): < 100, 100-199, 200-299, ≥ 300.
We analysed lean and fatty sh separately because they are speci ed in our dietary guidelines, and are sources of speci c essential nutrients such as vitamin D and omega-3 fatty acids from fatty sh, and iodine from lean sh (14). Fatty sh was classi ed as sh with ≥ 4% fat in the meat (salmon, trout, herring, mackerel), and was categorised in four categories (grams/day): < 5, 5-14, 15-29, ≥ 30. Lean sh was classi ed as sh containing < 4% fat in the meat (cod, haddock, plaice) but excluding products like sh cakes, sh balls, sh spread and stew, and was categorised in four categories (grams/day): < 15, 15-29, 30-44, ≥ 45.
Low-fat dairy products, comprised of semi-skimmed milk (≤ 1.5%), skimmed milk (0.1% fat) and yoghurt (≤ 3.4 % fat). It was chosen to include this food group as it is part of the Baltic Sea Diet Score, and because it is the main source of iodine in the Norwegian diet. Consumption was categorised into four categories: non-consumers, ≤ 200, 201-400, > 400 grams/day. Wholegrain products included wholegrain bread and breakfast cereals and was categorised into four categories (grams/day): < 60, 60-119, 120-179, ≥ 180.

Confounders
Covariates included in the analysis were chosen based on literature and selected with the use of Directed Acyclic Graphs (DAGs). DAGs are a tool that can help in selecting confounding factors to include in the statistical analysis when the purpose is to study causal relationships (25). The selection of confounding factors through DAGs are based on the assumption of causation between included variables in the DAG, hence wrong assumptions could lead to misspeci ed models. The strength is however that possible colliders are also identi ed in a DAG, reducing the risk of introducing bias in the statistical models. The following confounders were included in the analysis: physical activity, body mass index (BMI), smoking status and intake of energy, alcohol and processed meat.

Physical activity
Physical activity level was included based on self-report on a ten-point scale estimating physical activity at home, at work, exercising and walking. A validation study showed that self-reporting was able to rank women according to their level of physical activity (26). Physical activity was categorised as low (1)(2)(3)(4) points), medium (5-6 points) or high (7-10 points). BMI BMI was calculated based on self-reported height and weight (kg/m 2 ), and was categorised in four categories: < 20, 20-24.9, 25-29.9, ≥ 30 kg/m 2 . Selfreported weight and height has been found to provide valid ranking of BMI in NOWAC (27).

Smoking status
The smoking variable was computed by combining information on smoking status (never, former and current), with age at smoking initiation for those who have ever smoked. For current smokers who started smoking before the age of 20 we also included information about pack-years (number of cigarettes smoked per day, divided by 20, multiplied by number of years smoked). Twenty or more pack-years was de ned as heavy smoking, and 0-19 pack-years was de ned as moderate. Further adjustments for pack-years did not change the confounding effect of smoking. Smoking exposure was then divided into six categories: never smoker, current heavy smoker early starter (age at start smoking < 20), current moderate smoker early starter, current smoker late starter (age at start smoking ≥ 20), former smoker early starter, former smoker late starter.

Intake of energy, alcohol, and processed meat
The calculations of daily intake of nutrients, food items and energy has been described in the dietary assessment segment above. More speci cally, the calculated total energy intake was based on approximately 85 food frequency questions that cover the habitual diet of the women. Energy intake was computed as a continuous variable (kJ per day) excluding energy from alcohol.
Intake of alcohol was calculated based on three questions about intake of alcoholic beverages and was computed as a categorical variable to get a group of non-consumers, and categories representing lower and higher intake (grams/day): non-consumers, 0-5, > 5.

Outcome
The women were followed from return of the FFQ and until death or censoring, which was the date of emigration or end of follow-up on 31 December 2018.

Statistical methods
Population characteristics and dietary factors by healthy Nordic food group categories were analysed using χ² tests for categorical covariates and Kruskal-Wallis tests for continuous covariates. Distribution of covariates is presented across consumption categories of the Nordic food groups as mean (and standard deviation) for age, as median intake (and 10th-90th percentile) for energy, and percentages (%) for the covariates expressed categorically.
Spearman's rank-order correlation was used to test the association between the Nordic food groups and is presented as the correlation coe cient (r s ).
Cox proportional hazards regression models with age as the underlying time scale were used to examine the associations between consumption of the ve Nordic food groups and all-cause mortality. Estimates from the Cox regression models are presented as age-adjusted and multivariable-adjusted estimates. The Nordic food groups were mutually adjusted for in the multivariable-adjusted model, and the results were also adjusted for: physical activity, BMI group, smoking status, alcohol intake, estimated intake of energy and processed meat. Both models examined the Nordic food groups expressed as categorical exposures, and four of the Nordic food groups were further examined in the multivariable-adjusted model as continuous exposures in restricted cubic splines. The wholegrain products variable, which is only based on two FFQ frequency questions, was not examined in restricted cubic splines, as the distribution of values could not be approximated to a continuous variable.
Number of knots in the restricted cubic splines was determined by testing and comparing models with three, four and ve knots by the Akaike and Bayesian information criteria. This test was chosen because unlike the likelihood-ratio test and Wald testing procedures, the models do not have to be nested to compare how well the different models t the data. Models with the smallest AIC value were judged to t the data better, resulting in three knots at xed percentiles (10, 50, 90) of the distribution (29). The p-value for non-linearity in the restricted cubic spline analysis was calculated by performing Wald testing, which tests the null hypothesis that the coe cient of the second spline is equal to zero. Proportional-hazards assumptions were tested with a Schoenfeld residuals test.
All models were strati ed by subcohorts (n = 5), which were constructed by grouping together the FFQs that are most similar regarding the food frequency questions included, and which were completed closest together in time, as the data were collected over a period of almost ten years. We explored potential interactions between the Nordic food groups and smoking habits by adding a product term in the mutually adjusted models. If a statistically signi cant interaction effect was observed, we performed analyses strati ed by never and ever smokers.

Results
During a median of 20 (range 0.2-23) years of follow-up, 9 168 women died, mainly from cancer (n = 4 719) and cardiovascular diseases (n = 1 668). Table 1 shows the total population distribution and number of deaths by consumption categories of the Nordic food groups, and the median intake within the categories. The correlation matrix (Table 2) shows that lean and fatty sh were most strongly correlated of the ve food groups, but the correlation was still quite low (r s = 0.21).  Table 3 gives the distribution of included covariates across consumption categories of the Nordic food groups. The oldest women were in the highconsumption group of lean sh, and the greatest age span across intake categories was within this food group (ranging from 51.0 years old in low consumers to 53.5 in high consumers). Similar tendencies were seen within the food group fatty sh, where mean age ranged from 51.0 in low consumers to 53.2 years in high consumers. Within the other Nordic food groups the age differences across intake categories were minimal. We see a general tendency of women in the high-consuming categories within the Nordic food groups being more physically active, and more likely to be never smokers except among high consumers of lean and fatty sh. Across all food groups, energy intake was higher in the higher-consumption categories. The proportions of women reporting overweight (BMI 25.0-29.9 kg/m 2 ) and obesity (BMI ≥ 30 kg/m 2 ) were higher among high consumers of Nordic fruits and vegetables and lean sh, whereas the opposite was observed within the wholegrain products group.
Due to the high number of participants, even marginal differences in the distribution of covariates across consumption categories of the healthy Nordic food groups were statistically signi cant (p < 0.05). The restricted cubic spline regression showed a signi cant J-shaped association for the food groups Nordic fruits and vegetables (Fig. 1A), low-fat dairy products (Fig. 1B) and fatty sh (Fig. 1C).
For Nordic fruits and vegetables, the nadir (the intake level associated with lowest mortality) was observed at 200 grams/day (HR 0.84 (95% CI: 0.77-0.90) compared to no consumption (Fig. 1A). We observed a signi cant interaction between smoking status and Nordic fruits and vegetables, and thus strati ed analyses are also presented.
After strati cation by never/ever smokers, the observed association was only signi cant in ever smokers with the nadir at 250 grams/day (HR 0.78 (95% CI:  (Fig. 2). Furthermore, consumption of Nordic fruits and vegetables > 500 grams/day increased mortality among never smokers, but there were only 33 deaths registered at this consumption level.
For low-fat dairy products the nadir was observed at 200 grams/day (HR 0.94 (95% CI: 0.89-0.99) compared to no consumption, and high consumption (> 800 grams/day) increased mortality (Fig. 1B). For fatty sh the nadir was observed at an intake level of 10-20 grams per day (20 grams/day: HR 0.98 (95% CI: 0.92-1.03)), but this was not signi cantly better than not consuming fatty sh at all (Fig. 1C). Excessive consumption on the other hand was associated with increased mortality from 70 grams/day (HR 1.09 (95% CI: 1.01-1.17)). Consumption of lean sh was neutral in relation to mortality (Fig. 1D). Table 4 gives the results when the intake is categorised into consumption groups. These con rmed the ndings presented in Fig. 1. In addition, we note that the intake of wholegrain products, both of 120-179 grams/day and of ≥ 180 grams/day compared to < 60 grams per day, was associated with lower mortality (HR 0.92 (95% CI: 0.85-0.99)) ( Table 4). In the strati ed analysis, the median consumption of Nordic fruits and vegetables was 173 grams/day in never smokers and 159 grams/day in ever smokers (Table 5). An intake between 100-199 grams/day compared to < 100 grams/day was associated with reduced mortality among never smokers in similar strength as in the unstrati ed analysis (HR 0.90 (95% CI 0.81-0.99). However, for ever smokers intake above 100 grams/day was bene cial (Table 6). To minimise the chance of reverse causation we performed sensitivity analysis, starting follow-up two years after enrolment. Results did not change (Supplementary Figure 1), except that the association with low-fat dairy products and reduced mortality was slightly attenuated in the restricted cubic spline regression model due to a wider con dence interval believed to be caused by loss of cases (Supplementary Figure 1B). As ndings for Nordic fruits and vegetables in part could re ect the in uence of the consumption of other fruits and vegetables (24), we made further adjustments including other fruits and vegetables in the multivariable-adjusted model, but this did not in uence the results (Supplementary Figure 2).

Discussion
Moderate consumption of Nordic fruits and vegetables and low-fat dairy products were associated with reduced all-cause mortality, while excessive intake of low-fat dairy products was associated with increased morality during follow-up. Intake of wholegrain products estimated to be approximately in line with current recommendations for wholegrains of 70-90 grams/day was associated with reduced mortality, as was higher consumption. Consumption of both lean and fatty sh in line with dietary guidelines was within a non-signi cant bene cial range, but excessive consumption of fatty sh was associated with increased mortality during follow-up. In contrast, lean sh consumption level had no impact on total mortality.
Thus, there was a J-shaped trend with Nordic fruits and vegetables, fatty sh and low-fat dairy products and mortality, implicating that risk changes might not be linear with increasing intake of some healthy Nordic food groups.
The maximum bene t of consuming Nordic fruits and vegetables was achieved at around 200 grams/day, which is below the recommended intake of all fruits and vegetables of ve servings per day (30)(31)(32)(33). Non-linear inverse associations of fruit and vegetable intake with total mortality have recently been shown in two meta-analyses (34,35). While the maximum bene t was observed at higher consumption levels in both studies ( Miller and colleagues also found in the PURE study that optimal health bene ts of fruit and vegetable consumption could be achieved at a more modest intake level than currently recommended (around three to four servings per day) (36).
Potentially, subgroups of fruit and vegetable consumption such as the selected Nordic varieties have distinct health effects due to variations between different fruits and vegetables in nutritional properties (33), but other underlying dietary factors could also play a role in variations between dose-response relationships across populations.
The inverse association between Nordic fruits and vegetable consumption and mortality seemed stronger in former and current smokers than in never smokers. Also, the optimal consumption level was estimated to be higher in ever smokers than never smokers. Similar tendencies were reported in the European Prospective Investigation into Cancer and Nutrition which also included a subsample of women from NOWAC (37). In addition, a meta-analysis of prospective cohort studies on the association between consumption of fruits and vegetable and risk of lung cancer found stronger associations with lung cancer among smokers. Potentially antioxidant properties of fruits and vegetables are protective against increased oxidative stress caused by smoking (38).
The observed protective effect of wholegrain products on mortality in the present analysis is supported by meta-analyses of prospective cohort studies including populations from the US, Europe and Asia (39,40). The present results showed no further bene ts of consuming > 180 grams of wholegrain products per day. In the meta-analysis by Aune et al. reductions in risk for whole grains were observed up to an intake of 225 grams per day, but they found a non-linear association with all-cause mortality and a steeper reduction in risk at lower intake levels (40). Compared to our results, a study on Norwegian wholegrain eaters by Jacobs et al. included in the meta-analyses found an inverse association between a calculated wholegrain consumption score and mortality, with the highest score being most bene cial. This score was calculated based on slices of bread multiplied by percentages of wholegrain and was thus based on more detailed information on wholegrain consumption than we had access to (41).
The impact of dairy intake on mortality has been extensively studied, with contradicting results (42,43). The divergence between studies could be due to variation between different types of dairy products being investigated (i.e., total dairy, speci c categories of dairy such as milk, yoghurt, cheese, low-fat/highfat dairy), but also the quality of the underlying diet in different populations could affect the association between dairy consumption and mortality. For example, in a population with little access to complete proteins and essential nutrients from other animal sources than dairy, consumption could be differently associated to all-cause mortality compared to in a population with access to such nutrients from multiple food sources. Still, even when comparing results on low-fat milk consumption as a speci c dairy category and mortality in Nordic populations, one study nds an increased mortality (44) while another nds no association (45). It is noted that the fat content in yoghurt, which was part of the low-fat dairy products in present study could be up to 3.4 %, and therefore not considered low-fat within the yoghurt subcategory of dairy products. Hence, our results are not directly comparable with these studies. Our analysis showed a non-linear association with low-fat dairy and mortality, much in line with what Ding et al. found for total dairy consumption in three prospective cohort studies in women and men (46).
As in the present analysis, several large cohort studies have not been able to show any reduced mortality linked to frequent sh consumption (47,48). In line with our results, Engeset et al. found a non-linear trend with fatty sh consumption and mortality in the European Prospective investigation into Cancer and Nutrition cohort, which included a part of our sample (48). Also, a study on sh consumption and mortality in a cohort of Swedish men and women found a Ushaped association between consumption of sh and all-cause mortality, which was more pronounced in women (49). Further, when they considered lean and fatty sh separately, they found no associations between consumption of lean sh and mortality, but a markedly more pronounced association between fatty sh consumption and mortality. We observed that consumption up to the recommended 200 grams of fatty sh/week (29 grams/day) was within a nonsigni cant bene cial range, but when intake reached 70 grams/day there was a signi cantly increased mortality. In the cohort on Swedish men and women they reported higher mortality amongst women who consumed 80 grams sh per day compared to the median intake level (49).
Even though sh is a good source of essential nutrients, it is also a source of environmental contaminants such as dioxins, which are classi ed as carcinogens, and accumulates in the adipose tissue (14,50,51). While lean sh store fat in the liver, fatty sh store it in the llet itself, and such contains more of these substances compared to lean sh. One can speculate if this is related to the observed increased mortality with high consumption of fatty sh but not with lean sh. Nevertheless, our ndings do not support the part of the dietary guideline underlining that at least 200 grams a week should be fatty sh, as this conveys the impression that consuming more than this is better (14).
The search for optimal intake levels of foods and an ideal composition of the diet should be emphasised in studies on sustainable healthy regional diets, both for health and to reduce the burden of food production on the environment. However, establishing optimal intake levels of foods for health is not straightforward, given the limitations inherent in FFQs to give precise estimates of actual food intake and that the health effect is dependent on the underlying dietary pattern. For example, even though we found that approximately 200 grams/day of Nordic fruits and vegetables and 120-179 grams/day of wholegrain products (in models mutually adjusted for healthy Nordic food groups and energy intake), is optimal for longevity in this study, substituting processed meat with increased intake of these foods and for example lean sh (which was neutrally associated with mortality) could be bene cial for both health and the environment. Nevertheless, public health messages advocating an "increased" intake of certain foods without pointing to speci c intake levels gives the impression that the more we can eat of these foods the better health will be, and this might not be the case. To identify optimal food composition of a healthy Nordic diet in a public health perspective, substitutional analyses are highly relevant for further research.

Strengths and limitations
The strengths of this study include a large sample size, a high number of deaths and the long follow-up (median 20 years), providing enough statistical power in the analysis. Linkage to registry is a strength as all deaths are con rmed. Further, the risk of sampling bias is considered low due to the selection of women through the National Registry. Another strength is that a validated questionnaire was used to assess food intake and covariates (19-21, 26, 27).
The study is, however, limited by having only one assessment of diet, as dietary habits probably have changed during follow-up. Recalling the habitual diet with the use of FFQ could lead to recall error and misclassi cation of dietary exposures, but this is expected to be non-differential. In addition, the FFQ was not designed to measure all foods that are part of a healthy Nordic diet and hence does not capture all relevant food components such as wild berries and vegetables like kale or distinguish between speci c varieties of Nordic wholegrains such as rye and barley. Furthermore, precise assessment of dietary exposure is di cult and measurement errors are inevitable in nutritional epidemiology. Also, even though we adjusted for possible confounding factors that were unevenly distributed across intake categories of the Nordic food groups, residual confounding due to imprecise assessment of these factors as well as unmeasured factors is likely.
In particular, these results must be interpreted with caution as the moderate consumers are probably more representative of what most people eat, while both low and high consumers can be different in many ways (e.g., extreme dieters, vegans, people with allergies).

Conclusion
Moderate consumption of foods that are part of a healthy Nordic diet is either signi cantly better for or does not compromise longevity, compared to low or high intake, among middle-aged and older women. Consumption of Nordic fruits and vegetables was most bene cial in women that were either current or former smokers, and the optimal intake level seemed to be higher among these women compared to never smokers. These ndings implicate that dietary interventions might be especially important for people with higher mortality due to smoking.
Moderate intake of many food groups facilitates a varied diet, which is also part of the dietary guidelines, and this can be good for both health and the environment. Our results indicate that we need to assess linear as well as non-linear associations between food intake and health outcomes.