An analysis of a prospective cohort study

The European Potential Investigation into Most cancers and Diet (EPIC) research (http://epic.iarc.fr/) is an ongoing multicenter, potential cohort research investigating metabolic, dietary, life-style, and environmental components in relation to most cancers and different continual illnesses. Between 1992 and 2000, greater than 500,000 volunteers (25 to 70 years) had been recruited from 10 European nations (23 administrative facilities): Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, and the UK. Many of the members had been recruited from the overall inhabitants residing in a given geographic space, city, or province. Exceptions had been the cohorts of France (feminine members of a medical insurance scheme for college workers), Utrecht (breast most cancers screening attendees), Ragusa (blood donors and their spouses), and Oxford (primarily vegetarian and wholesome eaters). All members gave written knowledgeable consent and accomplished questionnaires on their weight loss plan, life-style, and medical historical past. The research was authorised by the native ethics committees and by the Inside Assessment Board of the Worldwide Company for Analysis on Most cancers. Particulars of the research design, recruitment, and knowledge assortment have been beforehand printed [32–34].

Of the 521,324 members enrolled, 451,390 had been included within the analyses, with 46,627 recorded deaths between 1992 and 2014. We excluded members with lacking life-style or dietary data, these with an excessive ratio of power consumption to power requirement (prime and backside 1%, as these values had been thought of physiologically implausible [34]), volunteers with null follow-up, these with prevalent illness at baseline (historical past of most cancers, cardiovascular illnesses [CVDs], and diabetes), and all members from the EPIC-Greece cohort, as a result of administrative constraints (S1 Fig).

An intensive and standardized phenotypic characterization was carried out for every participant upon enrollment. Questionnaires had been used to gather sociodemographic data, instructional degree (standardized for the entire cohort), private and familial historical past of illnesses, life-style (e.g., smoking, alcohol use, and bodily exercise), and menstrual and reproductive historical past of ladies. Anthropometric measurements (e.g., top, weight, waist, and hip circumferences) had been carried out in all facilities (besides France, Oxford, and Norway: self-reported knowledge) [35]. Top and weight had been complemented with out there self-reported values or imputed with center-, age-, and gender-specific common values when lacking. Physique mass index (BMI) was calculated as weight divided by top squared (kg/m²).

Common dietary consumption was assessed for every particular person at recruitment utilizing country- or center-specific validated dietary questionnaires (DQs) developed to seize the geographical specificity of a person’s weight loss plan over the previous 12 months. The kind of DQ used differed in accordance with research facilities and included self- or interviewer-administered semiquantitative meals frequency questionnaires (FFQs) with an estimation of particular person common parts or with the identical normal portion assigned to all members or weight loss plan historical past questionnaires combining an FFQ and 7-day dietary data. In most facilities, DQs had been self-administered, apart from Ragusa, Naples, and Spain, the place face-to-face interviews had been carried out. In depth quantitative DQs had been utilized in northern Italy, the Netherlands, and Germany, which had been structured by meals in Spain, France, and Ragusa. Semiquantitative FFQs had been utilized in Denmark, Norway, Naples, Umeå, and the UK, whereas an FFQ was mixed with a 7-day report on scorching meals in Malmö [33]. Put up-harmonization of DQ knowledge was carried out, following standardized procedures (e.g., decomposing recipes into substances), to acquire a standardized meals checklist for which the extent of element is comparable between nations. The EPIC meals composition database contains greater than 11,000 meals and beverage gadgets reflecting the specificities of every nation [36].

(Bio)range might be partitioned into 3 parts: richness, evenness, and disparity (Fig 1) [37]. However, our research focuses solely on DSR, beforehand really helpful as essentially the most acceptable measure of meals biodiversity for dietary consumption research, as we intention to tell food-based interventions and coverage based mostly on a easy, crosscutting indicator of human and planetary well being [26]. We argue that species evenness, which is outlined as a wonderfully equal distribution of meals and drinks within the weight loss plan, is neither desired from a dietary [38] nor environmental safety perspective [39]. Therefore, dietary evenness requires an arbitrary a priori collection of a relative abundance unit (e.g., power, vitamins, weight, quantity, and frequency) and “healthfulness” weighting components [40,41]. Furthermore, in contrast to for species richness, there may be presently no consensus on the measurement of species evenness (e.g., Shannon entropy, Berry–Simpson, and Pielou’s index) [37]. Dietary disparity is outlined for our analysis functions because the consumption of meals with distinct human well being [42] and ecosystem attributes [43], relatively than the extra slender, however well-established measures of dietary meals group range [44]. Equally to evenness, ecological metrics of species dissimilarity are based mostly on an inconsistent choice (and quantity) of phylogenetic, useful, and/or morphological traits (e.g., Rao’s quadratic range and Jaccard index) [37].

Fig 1. Partitioning meals biodiversity in 2 dietary patterns, which each consist of fifty foods and drinks gadgets.

Distinct species are indicated by their colour. Richness is absolutely the variety of species: In each dietary patterns, it is the same as 5. Evenness is the equitability of the species abundance distribution within the weight loss plan: In dietary sample A, all species are current in an equal abundance (e.g., frequency) and so it’s completely even, whereas dietary sample B may be very uneven since it’s dominated by the yellow [Zea mays (maize)] species. Disparity is the extent of similarity between species: For instance, crimson [Bos taurus (cow)] and pink [Gallus gallus (chicken)] species are extra related to one another, e.g., nutritionally and taxonomically, than the purple [Solanum melongena (aubergine)] species.

https://doi.org/10.1371/journal.pmed.1003834.g001

Due to this fact, for the 451,390 members included in our evaluation, meals biodiversity in a person’s weight loss plan was calculated based mostly on absolutely the variety of distinctive organic species in every (composite) meals, drink, and recipe, utilizing the European Meals Security Authority’s FoodEx2 meals classification and outline system [45] together with the detailed EPIC meals classification system (NCLASS). Meals gadgets consumed “by no means or lower than as soon as per thirty days” (on common) had been recalled below one class; accordingly, these species didn’t depend towards DSR. Furthermore, portions (g/day) had been disregarded for total DSR computation, since our curiosity is the sum of distinct species consumed per 12 months (i.e., DQs recalled dietary consumption over the previous 12 months). Nevertheless, relative portions had been thought of throughout sensitivity analyses, which excluded species consumed in trivial quantities (see under). Moreover, though a species might be consumed a number of instances per 12 months, doubtlessly from numerous useful meals teams (e.g., rooster meat and eggs, that are nutritionally disparate), via a “biodiversity conservation” lens, it contributes just one species to a person’s DSR in all eventualities (taxonomically equivalent: Gallus gallus). Right here, we contemplate that prime meals biodiversity is thus a mix of a number of species that act synergistically and complimentary for each human and environmental well being (e.g., ecological and internet dietary advantages from the Mesoamerican mixture of corn, beans, and squash, often known as the “three sisters”) [21,46].

Trigger-specific mortality knowledge had been coded in accordance with the tenth revision of the Worldwide Statistical Classification of Illnesses, Accidents, and Causes of Dying (ICD-10) [47]. Causes of demise assessed embody coronary heart illness [i.e., coronary heart disease (CHD) (ICD-10 codes: I20 to I25) and CVD other than CHD (I00 to I99, excluding I20 to I25)], most cancers (C00 to D48), illnesses of the respiratory system (J00 to J99), and illnesses of the digestive system (K00 to K93). Whole mortality (foremost consequence) was outlined as mortality from all causes, besides exterior causes of demise (S00 to T98 and V01 to Y98).

Statistical analyses had been preplanned and adopted the plan detailed within the venture proposal that was submitted to and authorised (March 2019) by the EPIC Steering Committee (see S1 Text). Unadjusted absolute demise charges had been calculated because the variety of instances per 10,000 person-years in Q₅ and Q₁ of DSR, respectively. Associations between meals biodiversity [DSR per year; count variable and quintiles (Qs)] and whole and cause-specific mortality had been characterised [hazard ratio (HR) and 95% confidence interval (CI)] utilizing multivariable-adjusted Cox proportional hazard regression fashions with age as the first underlying time variable. The Breslow methodology was adopted for dealing with ties. Examination of the Schoenfeld residuals, in accordance with follow-up time (years) for Qs of DSR, confirmed that the assumptions of proportionality had been glad. General survival curves by Q of DSR had been generated utilizing the Kaplan–Meier methodology (S2 Fig). Individuals contributed person-time to the mannequin till their date of demise, their date of emigration/loss to follow-up, or end-of-follow-up, whichever occurred first. P-values for linear development had been calculated with using the Wald take a look at of a pseudo-continuous rating variable, based mostly on the median variety of species consumed per 12 months for every Q of DSR. Nonlinear associations between DSR and whole mortality had been examined nonparametrically with restricted cubic splines [48]. P-value for nonlinear development was calculated with using the chance ratio take a look at, evaluating the mannequin with solely the linear time period to the mannequin with the linear and the cubic spline phrases.

Pooled cohort fashions had been stratified by intercourse, age at recruitment (1-year intervals), and research heart (“strata” choice in proc phreg, SAS) and multivariable-adjusted for confounding components utilizing a 5% change-in-estimate criterion for β-coefficients (utilized to all variables reported in Table 1, as a result of restricted information on components regarding DSR): smoking standing (present, 1 to fifteen cigarettes/day; present, 16 to 25 cigarettes/day; present, 26+ cigarettes/day; present, pipe/cigar/occasional; present/former, lacking; former, give up 11 to twenty years; former, give up 20+ years; former, give up ≤10 years; by no means; unknown), instructional degree, as a proxy variable for socioeconomic standing [longer education (including university degree, technical, or professional school); secondary school; primary school completed; not specified], marital standing (single, divorced, separated, or widowed; married or residing collectively; unknown), bodily exercise (Cambridge index: energetic; reasonably energetic; reasonably inactive; inactive; lacking), alcohol consumption at recruitment (g/day), whole power consumption (kcal/day), the 18-point relative Mediterranean weight loss plan rating, as an indicator for an total nutritious diet [49], the consumption of crimson and processed meat (g/day) [50], and fiber consumption (g/day; i.e., to mirror carbohydrate high quality [51], similar to complete grains). Attainable multicollinearity of (dietary) variables included in our fashions had been assessed by “collin” and “vif” choices in proc phreg, SAS (all situation indices <30 and variance inflation components <3, respectively). When knowledge on categorical covariates had been lacking, a “lacking class” was launched to the mannequin.

We carried out a number of sensitivity analyses to check the robustness of our all-cause mortality outcomes. First, we stratified our multivariable-adjusted fashions to estimate associations by intercourse and nation (owing to the various element of foods and drinks gadgets captured by DQs) individually. Moreover, we eliminated Mediterranean weight loss plan rating, crimson and processed meat, fiber, and whole power consumption from the fashions to look at their function as mediators, relatively than confounders. To evaluate the potential for residual confounding, we carried out subgroup analyses in accordance with main potential categorical impact modifiers: instructional degree, smoking standing, marital standing, and bodily exercise. Moreover, we used a “full instances” strategy, excluding members with lacking/unknown knowledge on covariates. Not all foods and drinks gadgets acquired a particular FoodEx2 species code, however relatively stored a generic NCLASS classification (e.g., “different root greens,” which counted as one species towards a person’s total DSR). Due to this fact, we reran our fashions dropping these generic foods and drinks gadgets. Analyses had been additionally carried out together with DSR with out the bottom 5% and 10% of species intakes for every EPIC meals group (see methodological reasoning above). As well as, we repeated our potential analyses for species richness inside every foremost EPIC meals group adjusted for total DSR (minus itself) to research whether or not a number of meals teams had been chargeable for the noticed associations [52]. Furthermore, from the 46,627 deadly occasions, we excluded deaths inside the first 3 (n = 2,969) and 6 years (n = 7,928) of follow-up to permit adequate delay between baseline dietary evaluation and mortality, thereby limiting reverse causality of subclinical illness. Findings from sensitivity analyses, which aren’t totally different (i.e., secure route, energy, and development of affiliation) from these utilizing all the EPIC cohort, aren’t proven. To evaluate potential residual confounding from unmeasured or uncontrolled confounders, E-values had been used [53,54].

Preliminary traits from 451,390 eligible members (71% feminine; median age 51 years) in accordance with Qs of DSR are proven in Table 1. After a median follow-up of 17 years (7,506,482 person-years), 44,892 deaths from nonexternal causes occurred, amongst which 19,284 from most cancers, 11,353 from illnesses of the circulatory system, 2,479 from illnesses of the respiratory system, and 1,386 from illnesses of the digestive system. From the 11,858 gadgets included within the EPIC meals checklist, 80% had been assigned FoodEx2 species codes (248 distinctive values; 78% of whole kcal/day; see S1 Table), whereas 16% acquired a generic NCLASS code (100 distinctive values, e.g., “different citrus fruits”), and 4% had been labeled as “not relevant” (e.g., added salt and water). In the entire cohort, participant’s [median (P₁₀–P₉₀)] DSR was 68 (40 to 83) species per 12 months. Bos taurus (cow), Triticum aestivum (frequent wheat grain), Sus scrofa (home pig), and Solanum tuberosum (potato) contributed most to self-reported whole dietary power consumption (i.e., roughly 45%) with [mean % (SD)] 19% (8), 16% (8), 4% (3), and 4% (3) kcal/day, respectively. When evaluating the fifth Q of DSR (highest; largely represented by the predominately vegetarian, “health-conscious” EPIC-Oxford (30%) and “omnivorous” German (35%) cohorts) towards the primary (lowest) Q, our findings point out massive variations in median dietary vegetable richness (22 versus 10 species), fruit, nuts, and seed richness (11 versus 5 species) and condiment richness (7 versus 2 species). In France, elevated DSR throughout Qs was noticed as a result of a major optimistic gradient in greens richness solely (7 versus 24 species).

Pooled multivariable evaluation indicated that common DSR consumption was inversely related to whole mortality (P_Wald < 0.001 for development), in that members with low DSR (Q₁; <48 species per 12 months) had notably larger mortality charges than people with average (Q₃; 64 to 72 species per 12 months) or excessive DSR (Q₅; ≥81 species per 12 months) (see S2 Table). The corresponding pooled HRs (95% CIs) had been 0·80 (0.76 to 0.83) for average DSR and 0.63 (0.59 to 0.66) for top DSR as compared with low DSR (Fig 2). Absolute mortality charges amongst members within the highest and lowest fifth of DSR had been 65.4 and 69.3 instances/10,000 person-years, respectively (see S2 Table).

Fig 2. Inverse affiliation between larger DSR per 12 months and whole mortality price within the EPIC cohort, 1992 to 2014.

Multiadjusted fashions had been stratified for heart, age at recruitment (1-year intervals, timescale), and intercourse and adjusted for baseline alcohol consumption (g/day), bodily exercise (Cambridge index: energetic; reasonably energetic; reasonably inactive; inactive; lacking), marital standing (single, divorced, separated, or widowed; married or residing collectively; unknown), smoking standing and depth of smoking (present, 1 to fifteen cigarettes/day; present, 16 to 25 cigarettes/day; present, 26+ cigarettes/day; present, pipe/cigar/occasional; present/former, lacking; former, give up 11 to twenty years; former, give up 20+ years; former, give up ≤10 years; by no means; unknown), instructional degree [longer education (including university degree, technical or professional school); secondary school; primary school completed; not specified], baseline power consumption (kcal/day), baseline fiber consumption (g/day), baseline crimson and processed meat consumption (g/day), and an 18-point Mediterranean weight loss plan rating [49]. P-values remained statistically vital after adjustment for a number of testing utilizing the Benjamini–Hochberg methodology. CI, confidence interval; DSR, dietary species richness; EPIC, European Potential Investigation into Most cancers and Diet; HR, hazard ratio; Q, quintile.

https://doi.org/10.1371/journal.pmed.1003834.g002

Our findings point out barely stronger relationships amongst males (see S3 Table). As an example, a 10-species increment in DSR was related (95% CI) with a 14% to 17% and 6% to eight% discount in all-cause mortality charges amongst men and women throughout roughly 20 years of follow-up, respectively. General, outcomes for mortality charges from nonexternal causes had been constant throughout 8 nations (P-value and P_Wald each <0·05; see S4 Table). Within the UK, we noticed total larger DSR and a subsequent smaller distinction between members with decrease and better scores (Q₁; <71, Q₅; ≥82 species per 12 months). The related protecting impact of DSR was not considerably modified when eradicating potential dietary mediators, amongst main subgroups (though the bottom HRs had been reported amongst present people who smoke and members with secondary training) or full instances, when dropping generic foods and drinks codes, or exclusion of the bottom 5% (Q₁; <47, Q₅; ≥80 species per 12 months) and 10% species consumption (Q₁; <46, Q₅; ≥78 species per 12 months) from every EPIC meals (sub)group. Moreover, our noticed associations weren’t defined by species richness inside one single meals group, suggesting a optimistic cumulative impact of total DSR (see S5 Table). Restricted graphical, however statistically vital (P_LR<0.001), proof of nonlinearity was noticed for all-cause mortality (S3 Fig).

In multivariable analyses, a 10-species increment in DSR was inversely related to the speed of demise [HR (95% CI)] as a result of digestive illness [0.80 (0.76 to 0.86)], respiratory illness [0.84 (0.80 to 0.88)], coronary heart illness [0.88 (0.86 to 0.90)], and most cancers [0.93 (0.92 to 0.95); all P_Wald < 0.001 for trend; Table 2].

The massive E-values for whole and cause-specific mortality recommend that residual confounding is prone to be low, conditional on the measured covariates in our fashions (see S6 Table).