Global obesity prevalence has reached epidemic proportions, with 37 % of men and 38 % of women now either overweight or obese( Reference Ng, Fleming and Robinson 1 ). Poor dietary choices and inadequate physical activity are the primary causes of obesity( Reference Hill, Wyatt and Peters 2 ). Current strategies for improving diet and other lifestyle behaviours, such as consuming five portions of fruit and vegetables per day( 3 ), are based on ‘one size fits all’ generalised dietary guidelines. Given that the burden of obesity is increasing( Reference Ng, Fleming and Robinson 1 ), alternative strategies for improving dietary behaviours are being developed, including predictive, personalised, preventive and participatory interventions( Reference Hood and Friend 4 ). Recent evidence suggests that genetic-based personalised nutrition (PN) improves dietary intakes more than non-personalised advice( Reference Nielsen and El-Sohemy 5 ). However, since dietary intakes tend to cluster( Reference Berrigan, Dodd and Troiano 6 , Reference Leech, McNaughton and Timperio 7 ), it may be possible to enhance the efficacy of interventions by further characterisation of participants according to their dietary and lifestyle behaviours and, subsequently, use this information to strengthen the basis for personalisation of the intervention. For example, lower intakes of fruit, vegetables and whole grains are often associated with higher intakes of red or processed meat( Reference Newby, Muller and Hallfrisch 8 ). In addition, less healthy dietary clusters are associated with increased disease risk( Reference Moore, Singer and Bradlee 9 ) and unhealthy dietary and lifestyle behaviours are associated with higher levels of sedentary behaviour( Reference Leech, McNaughton and Timperio 7 ) and mortality( Reference Khaw, Wareham and Bingham 10 , Reference van Dam, Li and Spiegelman 11 ). Clustering individuals based on whether they meet dietary recommendations may be a useful predictive tool for estimating response to an intervention( Reference Kesse-Guyot, Castetbon and Estaquio 12 – Reference Zamora, Gordon-Larsen and He 14 ) and may help to stratify or personalise interventions.
The Food4Me Proof-of-Principle study was the first Internet-based study to demonstrate that PN advice was more effective in improving dietary intakes, including lowering intakes of red meat, compared with conventional ‘one size fits all’ population-based advice( Reference Livingstone, Celis-Morales and Navas-Carretero 15 , 16 ). However, the characteristics of individuals clustered on the basis of adherence to current recommended dietary intakes of fruit and vegetables, whole grains, oily fish, dairy products and red and processed meat are unknown. Thus, the aims of the current analysis were to: (i) characterise European adults participating in the Food4Me study( Reference Celis-Morales, Livingstone and Marsaux 17 ) according to clustering based on European recommendations for healthy eating; and (ii) determine whether cluster membership predicted dietary changes following a PN intervention.
Study design and population
The Food4Me study was a 6-month, four-arm, Internet-based, randomised controlled trial (RCT) in 1607 individuals conducted across seven European countries( Reference Livingstone, Celis-Morales and Navas-Carretero 17 , Reference Livingstone, Celis-Morales and Navas-Carretero 18 ). Participants were recruited via the Food4Me website( 16 ) to emulate a web-based PN service. This was aided by local and national advertising of the study via the Internet, radio, newspapers, posters, e-flyers, social media and word of mouth. Recruitment took place between August 2012 and August 2013 in the following sites: University College Dublin (Ireland), Maastricht University (The Netherlands), University of Navarra (Spain), Harokopio University (Greece), University of Reading (UK), National Food and Nutrition Institute (Poland) and Technical University of Munich (Germany). The Research Ethics Committees at each University or Research Centre delivering the intervention granted ethical approval for the study. The Food4Me trial was registered as an RCT (NCT01530139) at Clinicaltrials.gov. All participants expressing an interest in the study were asked to sign online consent forms at two stages in the screening process.
Participants were randomised to receive non-personalised, generalised dietary advice (Control) or one of three levels of PN (Level 1, Level 2 or Level 3). Briefly, non-personalised dietary advice was based on national dietary recommendations in each of the seven European countries. These ‘standardised’ recommendations included advice on energy intake and on the consumption of fruits and vegetables, whole grains, fish, dairy products, meat, type of fat and salt. Participants randomised to Level 1 received personalised dietary advice on how their intakes of these food groups at baseline compared with guideline amounts. Participants randomised to Level 2 received advice based on their dietary intake (as for Level 1) and also on their baseline phenotypic data. The phenotypic feedback was based on anthropometric measurements and nutrient- and metabolic-related biomarkers. Participants randomised to Level 3 received advice based on their dietary intake, phenotypic and genotypic data collected at baseline. The genotypic feedback was based on specific variants in five nutrient-responsive genes selected specifically for the study. Further details are provided elsewhere( Reference Celis-Morales, Livingstone and Marsaux 17 ).
Screening questionnaires and dietary intakes
Participants eligible for inclusion in the RCT completed an online questionnaire to collect detailed information on sociodemographic, health and anthropometric characteristics and dietary habits. Following completion of this questionnaire, participants were asked to complete an online FFQ to estimate usual dietary intake. This FFQ, which was developed and validated for the study( Reference Forster, Fallaize and Gallagher 19 , Reference Fallaize, Forster and Macready 20 ), included 157 food items consumed frequently in each of the seven recruitment countries. Intakes of foods and nutrients were computed in real time using a food composition database based on McCance and Widdowson’s The Composition of Foods ( 21 ). Intakes of nutrients were assessed based on standardised recommendations (see online supplementary material, Supplemental Table 1) for dietary intakes of foods and food groups( Reference Celis-Morales, Livingstone and Marsaux 17 ), which were integrated and harmonised across eight European countries (UK, Ireland, Germany, The Netherlands, Spain, Greece, Poland and Norway)( 22 – 25 ). The following four food group recommendations were used in the present analysis: (i) eat at least five portions of fruit and vegetables every day (operationalised as ≥400 g); (ii) eat at least three portions of wholegrain products daily (≥50 g); (iii) eat at least one portion of oily fish per week (≥150 g); and (iv) eat less than three portions of red or processed meat per week (≤450 g)( Reference Celis-Morales, Livingstone and Marsaux 17 ). The Healthy Eating Index 2010 (HEI) was derived based on intakes of the following components: ratio of MUFA + PUFA to SFA, protein, salt, ‘empty calories’, refined grains, seafood and plant protein, fruit, whole fruit, vegetables, greens and beans, whole grains and dairy products( Reference Guenther, Casavale and Reedy 26 ).
Personalised feedback report
Participants randomised to PN received personalised reports via email at baseline, month 3 and month 6 of the intervention based on diet, anthropometric measurements and physical activity. Using information on the individual’s intakes of nutrients, algorithms were used to rank information on need for dietary change and to provide participants with three specific dietary, food-based goals. For participants randomised to Level 2 and Level 3, the dietary advice was also based on phenotypic data (Level 2) and phenotypic plus genotypic data (Level 3). Reported intakes were compared with recommended intakes and determined to be adequate, high or low. If intakes were too high or too low, contributing foods were identified and specific messages developed to advise change in intake of those foods. Dietary intakes relative to recommendations were illustrated using a three-colour sliding scale: green representing ‘Good, no change recommended’, amber representing ‘Improvement recommended’ and red representing ‘Improvement strongly recommended’. For the genotype-based information, risk was indicated using ‘Yes’ or ‘No’ according to whether the participant did or did not carry the higher-risk variant for each of the five nutrient-related genes included in the study. Additionally, each report contained a personalised message from the dietitian/nutritionist to the participant. Further details of the protocol are provided elsewhere( Reference Celis-Morales, Livingstone and Marsaux 17 ).
Anthropometric, sociodemographic and physical activity measures
Detailed standardised online instructions were given for participants to self-measure and self-report their body weight, height and waist circumference via the Food4Me website( 16 ). BMI was estimated from body weight and height. Self-reported measurements were validated in a sub-sample of the participants (n 140) and showed a high degree of reliability( Reference Celis-Morales, Livingstone and Woolhead 27 ). Physical activity level and time spent in sedentary behaviours were estimated from triaxial accelerometers (TracmorD, Philips Consumer Lifestyle, The Netherlands). Participants self-reported smoking habits and occupation. Occupations were grouped according to the European classifications of occupations and their salaries (the European-wide average salary for each occupation was compared with the mean overall salary; if the sd of the salary was >0·5, they were placed in Group 1, between 0·5 to −0·5 were placed into Group 2, and <−0·5 were placed into Group 3) as follows: Group 1, Professional and managerial (professionals; managers); Group 2, Intermediate (armed forces occupations; technicians and associate professionals; clerical support workers); Group 3, Routine and manual (craft and related trades workers; plant and machine operators and assemblers; service and sales workers; elementary occupations; skilled agricultural, forestry and fishery workers)( 28 , 29 ). Categories for ‘Students’ and ‘Retired and unemployed’ were added.
Data were analysed using the statistical software packages Stata version 13 and IBM SPSS Statistics Version 22.0. Clusters of dietary recommendations were generated based on whether participants met the following four food group recommendations at baseline and were coded as 0 or 1 accordingly: (i) eat at least five portions of fruit and vegetables every day (operationalised as ≥400 g); (ii) eat at least three portions of wholegrain products daily (≥50 g); (iii) eat at least one portion of oily fish per week (≥150 g); and (iv) eat less than three portions of red or processed meat per week (≤450 g). Clusters were derived using the SPSS Two Step cluster analysis procedure( Reference Chiu, Fang and Chen 30 ). Small pre-clusters were generated based on the log-likelihood distance criterion (Step 1) and were merged into distinct groups using agglomerative hierarchical clustering (Step 2). Automatic selection and the Bayesian information criterion were used to determine the optimal number of clusters. Robustness and stability of the final clusters were re-evaluated by random ordering of cases (four times). This clustering methodology identified the percentage of participants within each cluster who met recommended intakes of each of the four food groups of public health importance. Logistic regression was used to test for significant differences across categorical variables and ANOVA was used for continuous variables. Tukey pairwise comparisons were used to test for significant differences between clusters. Analyses were adjusted for age, sex, country, BMI, physical activity level and smoking, except when those (or related) variables were being assessed (i.e. analyses were not adjusted for BMI when assessing BMI, body weight or waist circumference). Results were deemed significant at P<0·05. To exclude extreme intakes of the food groups used for clustering, the top and bottom 3 sd of these intakes were excluded prior to clustering.
Of the 5562 individuals who registered on the Food4Me website, 1607 were randomised into the study and a total of 1480 provided baseline data on dietary intakes( Reference Celis-Morales, Livingstone and Marsaux 17 ).
Dietary adequacies across Food4Me cohort
Recommended intakes for nutrients are summarised in the online supplementary material, Supplemental Table 1. On average, 50 % of individuals met the recommendations for total fat (Supplemental Table 2). The percentage of individuals who met the recommendations for SFA, MUFA and PUFA intake was 54, 24 and 36 %, respectively. Only 56 % of individuals met the recommendation for carbohydrate intake, whereas 91 % of individuals had adequate protein intakes. Only 7 and 46 % of individuals met the recommendations for salt and dietary fibre intakes, respectively. Meeting recommended micronutrient intakes ranged from 61 % (folate) to 99 % (vitamin B12).
As summarised in the online supplementary material, Supplemental Table 2, approximately half (52 %) of participants reported consuming at least five portions of fruit and vegetables per day and 32 % consumed at least one portion of oily fish per week. Nearly three-quarters (74 %) of participants consumed more than three servings of whole grains per day and approximately half of participants (51 %) consumed less than three servings of red meat per week. Fourteen per cent of individuals met the recommendation for dairy product intake (≥600 g/d).
Clustering of individuals according to whether they met the intake recommendations for dairy products, fruits and vegetables, oily fish, red meat and whole grains at baseline did not create clear clustering due to the low percentage of individuals who met the recommendation for dairy products (two clusters). Exclusion of dairy products as a clustering variable provided improved clustering, as estimated by the silhouette measure of cohesion and separation (average silhouette: 0·3 v. 0·5; four clusters; see online supplementary material, Supplemental Table 3). Cluster 1 (C1) was the largest (n 475) and was particularly characterised by individuals meeting the recommended intake for oily fish (100 % of individuals); 74 and 69 % of C1 members met the recommendations for whole grains and fruit and vegetables, respectively, whereas only 46 % met the recommendation for red meat. Cluster 2 (C2; n 398) was the second largest and was particularly characterised by all members meeting recommendations for whole grains (100 %) and red meat (100 %), only 50 % meeting the recommendation for fruit and vegetables and no one meeting the recommendation for oily fish. All individuals in cluster 3 (C3; n 348) met the recommendation for whole grains, but no one met the recommendation for oily fish or red meat, whereas only 48 % met the recommended intake for fruit and vegetables. None of the participants in cluster 4 (C4; n 259) met the recommended intake for either oily fish or whole grains; only 50 and 29 % of C4 members achieved the recommended intakes for red meat and fruit and vegetables, respectively (Supplemental Table 3).
Dietary intakes by cluster
Intakes of oily fish and fruit and vegetables were higher in C1 than in C2, C3 and C4 (P<0·05), and wholegrain intakes were higher in C1, C2 and C3 than in C4 (Table 1; P<0·05). Red meat intake was lower in C1, C2 and C3 than in C4 (P<0·05). Intakes of fruit juice, eggs, chicken, white fish, fish products, beans and lentils and low-fat dairy products were higher in C1 than C4, whereas intakes of non-wholegrain products were lower (P<0·05). Participants in C2 had lower intakes of chips and pizza and fried foods than C3 and C4 (P<0·05; Table 1). Total energy intake and ratio of energy intake to BMR were higher in C1 than in C2 and C4 and higher in C3 than in C2 (P<0·05; Table 1). Individuals in C1 derived higher percentages of energy intake from PUFA and protein than those in C2 and C4 (P<0·05) and individuals in C2 derived a higher percentage energy from carbohydrates than participants in C3 and C4 (P<0·05). In contrast, individuals in C1 had lower percentage energy intakes from total fat and SFA than those in C4 (P<0·05) and higher percentage energy intake from MUFA than participants in C2 and C3 (P<0·05). Individuals in C1 had lower percentage energy intake from sugar than those in C2 (P<0·05). Participants in C1 consumed more dietary fibre and salt than those in C2 and C4 (P<0·05).
EI:BMR, ratio of energy intake to BMR.
* ANOVA was used to test for significant differences across clusters. Models were adjusted for age, sex, BMI, physical activity level, smoking habits and country; significant P values are highlighted in bold. Post hoc Tukey tests were performed to test for significant differences between clusters; superscript numbers denote where the differences lie across the clusters. For example, 2 indicates that the mean value is significantly different from that in cluster 2 (P<0·05).
† P values are also adjusted for total energy intake.
More individuals in C1 met the recommendations for total fat intake (51 %), SFA (62 %), PUFA (42 %) and dietary fibre (56 %) than C4 cluster members (see online supplementary material, Supplemental Table 4). Fewer individuals in C1 met the recommendation for protein intake (86 %) than those in C2 (97 %) and C3 (93 %). Furthermore, fewer individuals in C1 met the recommendation for salt intake (5 %) than C2 (11 %) and C4 (17 %; Supplemental Table 4).
Sociodemographic, anthropometric and health characteristics by cluster
Individuals in C1 were on average 4·5 years older than those in C4 (P<0·05; Table 2). Body weight was significantly lower in C1 than in C3, and in C2 compared with C3 and C4 (P<0·05). Individuals in C1 had 1·4 kg/m2 lower BMI and 5·0 cm lower waist circumference than participants in C3 (P<0·05) and physical activity level was higher in C1 than C2 and C4 (P<0·05). Eleven per cent more individuals in C4 wanted to lose weight than in C1 (P<0·05; Table 2) and C4 was characterised by more current smokers than C1 (P<0·05). Twelve per cent more individuals in C1 than in C4 had a professional or managerial occupation, and similarly 7 % more individuals had a manual occupation in C4 compared with C1 (P<0·05; Table 2). No other significant differences were observed (Table 2).
Values are presented means and standard deviations or percentages.
* ANOVA and logistic regression were used to test for significant differences across clusters in continuous and categorical variables, respectively. Analyses were adjusted for age, sex, BMI, physical activity level, smoking habits and country; significant P values are highlighted in bold. Post hoc Tukey tests (continuous data) and logistic regression (categorical) were used to test for significant differences between clusters; superscript numbers denote where the differences lie across the clusters. For example, 2 indicates that the mean value or percentage is significantly different from that in cluster 2 (P<0·05).
Changes in Healthy Eating Index by cluster after 6-month intervention
Baseline and follow-up HEI scores and their components are presented in Table 3. There were no significant differences in changes in HEI between clusters for those randomised to non-personalised dietary advice. In contrast, for individuals who received PN advice (based on information of current diet alone or combined with information on phenotype and genotype), changes in HEI differed between clusters (P<0·001). There were bigger improvements in HEI for participants in C4 compared with C1 and C2 (P<0·05) and in C2 compared with C4 (P<0·05; Fig. 1). There were no significant differences in changes in HEI between clusters when PN was stratified by Level 1, Level 2 or Level 3 (data not shown).
* Fatty acid ratio is the ratio of unsaturated fatty acids (mono- and polyunsaturated fatty acids) to saturated fatty acids.
*ANOVA was used to test for significant differences across clusters. Models were adjusted for age, sex, BMI, physical activity level, smoking habits and country; significant P values are highlighted in bold. Post hoc Tukey tests were used to test for significant differences between clusters; superscript numbers denote where the differences lie across the clusters. For example, 2 indicates that the mean value is significantly different from that in cluster 2 (P<0·05).
Exclusion of participants with reported intakes more than 3 sd above or below the mean dietary intakes of whole grains, oily fish, red meat and fruit and vegetables revealed similar clusters (see online supplementary material, Supplemental Table 5). The pattern of the main results remained the same, with individuals in C3 and C4 making greater changes in HEI at month 6 than those in C1, as did participants in C4 compared with those in C2 (P<0·05).
Based on our secondary analysis in the Food4Me Proof-of-Principle study, we identified four distinct clusters of individuals according to their adherence to current European dietary recommendations. Individuals in C1 and C2 met more dietary recommendations than those in C3 and C4. Moreover, on average, individuals in C1 and C2 had a healthier diet, lower BMI and waist circumference, and smoked less, compared with those in C3 and C4. When randomised to a 6-month PN intervention, participants in C4 made the greatest improvements in their diets (as estimated by HEI), compared with participants receiving non-personalised ‘one size fits all’ generalised advice. The present study is the first to investigate clusters of adherence to European dietary recommendations and to determine the responsiveness of cluster members to PN advice.
Comparison with other studies
Previous studies have used cluster analysis to categorise individuals( Reference Ocké 31 ). We used cluster analysis to categorise individuals based on their adherence to current European food-based dietary guidelines at baseline for participants in the Food4Me intervention study. This approach identified groups of individuals who differed in the number, and groupings, of dietary recommendations they met. Clusters where more individuals met the recommendations were characterised by being slightly older and in more highly educated occupations, which is a well-established characteristic of healthy dietary clusters( Reference Kant 32 ).
Clustering of dietary intakes and adequacies has been investigated in relation to several health outcomes( Reference Leech, McNaughton and Timperio 7 , Reference Newby, Muller and Hallfrisch 8 , Reference Ma, Bertone and Stanek 33 ) and can be strong predictors of these outcomes( Reference Wirt and Collins 34 ). A recent review of dietary clusters and health outcomes by the US Department of Agriculture( 35 ) concluded that the strongest evidence for an association between unhealthy dietary patterns and increased disease risk is for CVD, followed by obesity and then type 2 diabetes. The review concluded that there was a lack of studies assessing dietary intakes at follow-up and using a universal and quantitative indicator of dietary intake. Our study is in line with these recommendations as we utilised the HEI, which is a validated estimate of dietary adequacy, and we assessed dietary change using the same instrument at both baseline and follow-up. Although more limited, some prospective and RCT studies have investigated the effect of clustering on changes in health outcomes( Reference Kesse-Guyot, Castetbon and Estaquio 12 , Reference Gao, Beresford and Frank 36 , Reference Jacobs, Sluik and Rokling-Andersen 37 ) and some studies have used adherence to dietary recommendations to derive clusters( Reference Kesse-Guyot, Castetbon and Estaquio 12 – Reference Zamora, Gordon-Larsen and He 14 , Reference Cheng, Gerlach and Libuda 38 , Reference Park, Murphy and Wilkens 39 ). Dietary recommendations used in studies included in the systematic review by the US Department of Agriculture( 35 ) varied according to the study, but all included a measure of fruit and vegetable, wholegrain and meat intake.
To our knowledge, no previous research has evaluated the impact of clustering of dietary recommendations on the response to a PN intervention. We observed that individuals in the cluster where the fewest recommendations were met (C4) reported the biggest improvement in HEI following PN intervention but there were no differences between clusters in response to conventional, non-personalised dietary advice. Given that adverse lifestyle behaviours and the prevalence and risk of death from obesity-related diseases are strongly socio-economically patterned( Reference Di Cesare, Khang and Asaria 40 ), it is important that appropriate interventions are targeted to those most in need of improved lifestyle. While research on the development and implementation of PN interventions and their effects on changing diets is in its infancy( Reference Celis-Morales, Lara and Mathers 41 ), the findings from the present study provide encouragement that PN interventions can be more effective than current ‘one size fits all’ interventions and that they may be particularly effective among individuals with the poorest diets. There have been concerns that PN may be taken up only by the ‘worried well’( Reference Ferguson 42 ), who already have adequate dietary intakes. However, our findings suggest that PN is most effective in people who have the least adequate diets, and therefore the greatest need for improvement in dietary intakes, with the potential for significant reductions in disease risk.
Strengths and limitations
The present study had a number of strengths. Our findings are derived from a relatively large number of participants who were broadly representative of European adults from seven different European countries( Reference Livingstone, Celis-Morales and Navas-Carretero 18 ). The Food4Me RCT collected extensive information on anthropometrics, physical activity and sociodemographic and health-related data, which contributed to detailed characterisation of participants in the clusters. Our study design allowed us to estimate changes in dietary intakes using the same validated instrument at baseline and at month 6. Furthermore, we quantified responses using the HEI, which has been shown to be an effective indicator of overall diet quality( Reference Guenther, Casavale and Reedy 26 ) and, therefore, a better measure of overall dietary change than outcomes based on single foods or nutrients.
A limitation of the study is that our data were self-reported via the Internet, which may have introduced measurement error. However, the validity of Internet-based, self-reported anthropometric data is high( Reference Pursey, Burrows and Stanwell 43 ) and has been confirmed in the present study( Reference Celis-Morales, Livingstone and Woolhead 27 ). We were not able to include dairy products as a dietary recommendation in the present analyses due to so few individuals meeting the recommendation. However, dairy products do not have a recommended intake in the UK and so habitual diets would not necessarily be expected to comply with this recommendation, even if they were very health conscious. Dietary intakes were estimated by an FFQ, which is known to be subject to misreporting error( Reference Macdiarmid and Blundell 44 ), but this was minimised by validating our FFQ against a 4 d weighed food record( Reference Fallaize, Forster and Macready 20 ). Moreover, our estimation of dietary change was based on the HEI, which is a validated indicator of overall diet( Reference Guenther, Casavale and Reedy 26 ) and which may be less susceptible to reporting errors than approaches measuring change in specific nutrients or individual foods. Our study participants were almost solely Caucasian; thus, further research in wider ethnicity groups is required to generalise our findings to other populations. One of the primary aims of the Food4Me Proof-of-Principle study was to evaluate change in intakes of food groups across four treatment arms. Thus, although the present study is a secondary analysis of these data, clustering was based on how individuals adhered to food group recommendations and included four clusters. As a result, our analyses are likely to be powered to detect differences between clusters.
Implications of findings
Our findings suggest that the efficacy of PN in modifying dietary intakes depends on the clustering of adherence to dietary recommendations, with those with the poorest diets benefiting most from the PN intervention. As a result, the implementation of PN-based interventions in individuals with the least healthy diets may help to address health inequalities. Understanding the characteristics of individuals within coherent clusters which are linked with their responsiveness to interventions may help in the design and implementation of more effective health promotion actions. Future PN interventions may benefit from tailoring PN advice based on clustering of overall dietary behaviours rather than on single nutrients or foods.
We identified four distinct clusters of individuals based on adherence to current food-based dietary recommendations. Participants in the cluster where the fewest recommendations were met (C4) reported significantly greater improvements in their diets (as estimated by the HEI) following a 6-month trial of PN, whereas there was no difference between clusters for those randomised to the Control, non-personalised dietary intervention.
Acknowledgements: The authors would like to thank Antoneta Granic for her valuable input into the design of the cluster analysis. J.A.M. and S.N.-C. are grateful to CIBERobn Fisiopatología de la Obesidad y Nutrición (Instituto Carlos III, Madrid, Spain) for general support in research. Financial support: This work was supported by the European Commission under the Food, Agriculture, Fisheries and Biotechnology Theme of the 7th Framework Programme for Research and Technological Development (265494). The European Commission had no role in the design, analysis or writing of this article. Conflict of interest: None of the authors had a personal or financial conflict of interest. Authorship: K.M.L. and C.C.-M. are joint first authors. Author responsibilities were as follows: Y.M., I.T., C.A.D., E.R.G., L.B., J.A.L., J.A.M., W.H.M.S., H.D., M.G. and J.C.M. contributed to the research design. J.C.M. was the Food4Me Proof-of-Principle study leader. C.C.-M., C.F.M.M., H.F., C.B.O., C.W., A.L.M., R.F., S.N.-C., R.S.-C., S.K., L.T., C.P.L., A.S., M.C.W., E.R.G., L.B. and J.C.M. contributed to the developing the standardised operating procedures for the study. C.C.-M., S.N.-C., R.S.-C., C.W., C.B.O., H.F., C.F.M.M., A.L.M., R.F., S.K., L.T., C.P.L., A.S., M.C.W. and J.C.M. conducted the intervention. C.C.-M., C.F.M.M. and W.H.M.S. contributed to physical activity measurements. K.M.L. and C.C.-M. wrote the paper, performed the statistical analysis for the manuscript. All authors contributed to a critical review of the manuscript during the writing process. All authors approved the final version to be published. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects/patients were approved by the Research Ethics Committees at each University or Research Centre delivering the intervention. The Food4Me trial was registered as an RCT (NCT01530139) at Clinicaltrials.gov. All participants expressing an interest in the study were asked to sign online consent forms at two stages in the screening process. These consent forms were automatically directed to the local study investigators to be counter-signed and archived.
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1368980016001932