Introduction

Agriculture and food production play a critical role in mitigating climate change and meeting global climate targets [1,2,3]. Research shows that changing dietary behaviour can reduce diet-related environmental impact indicators, like greenhouse gas emissions (GHGE) [3,4,5,6,7]. Sustainable healthy diets that have lower environmental impact are generally rich in plant-based foods, such as whole grains, fruits and vegetables, legumes and nuts and seeds, with low to moderate amounts of animal-sourced foods [8]. Most food-based dietary guidelines recommend a similar dietary pattern but provide vague and generic eating advice using broad terms like ‘increase’ or ‘avoid’ [8, 9]. Other recommendations for sustainable diets with specific intake recommendations (grams per day) have been published but have been called into question by others as to whether these diets are nutritionally adequate, or culturally acceptable [3, 6, 7, 10,11,12,13,14].

To date, the evidence base for sustainable diets has been mostly theoretical and often examined using modelled dietary changes. For example, in a systematic review by Aleksandrowicz and colleagues where different dietary patterns were modelled to examine the impact of dietary change on environmental metrics [4]. Several research groups have published ‘reference diets’ which serve as a model for a sustainable diet [3, 6, 7]. These diets provide an example diet with specific intake recommendations for key food groups but have yet to be expanded to account for the variability in nutrition needs across the population. Factors such as demographics (e.g. age, sex, body size), behaviour (e.g. dietary intake, physical activity), and genetics influence an individual’s nutrition needs [15]. Using these factors to provide more tailored or targeted nutrition advice is known as personalised nutrition. Research shows that personalised nutrition advice leads to healthier eating behaviour over a longer period when compared to a traditional, ‘one-size-fits-all’ approach [16,17,18]. The multi-national Food4Me study compared multiple levels of personalised nutrition feedback to general healthy eating advice and showed that personalised nutrition advice can be delivered using a standardised decision-making framework that is scalable and reproducible [19]. The Food4Me study also showed that the messages derived from decision trees aligned with personalised advice from a dietitian and concluded the decision tree method could be effective in large populations due to its efficiency and reliability [20]. To our knowledge, personalised nutrition has yet to be tested in relation to achieving more sustainable healthy diets. Similarly, there is a research gap with respect to the effectiveness of studies providing dietary advice aimed at reducing diet-related environmental impact.

The aim of this paper is to describe the development, testing, and use of a novel personalised feedback framework designed to deliver dietary advice to reduce diet-related GHGE in a healthy and acceptable manner as part of the MyPlanetDiet RCT. This paper will provide an overview of the MyPlanetDiet dietary protocol, including the development and testing of decision trees and feedback messages used to deliver personalised dietary advice to participants.

Methods

Myplanetdiet study design

MyPlanetDiet was a 12-week single-blinded parallel RCT evaluating the capacity for a more sustainable diet to reduce diet-related GHGE in a healthy, acceptable, and safe manner. Participants were randomized into one of two groups of personalised nutrition feedback based on either a more sustainable and healthy diet (intervention) or healthy eating food-based dietary guidelines, from the Republic of Ireland and Northern Ireland (control) [21, 22]. Participants in the two groups received the same level, type, and frequency of communication with the only difference being the target diet that underpinned the advice provided to both groups.

MyPlanetDiet included two onsite visits per participant (visit 1/week 0 and visit 2/week 12). The primary outcome was change in diet-related GHGEs from baseline to endpoint. Secondary outcomes included changes in nutrient intake (energy, macronutrients, vitamins, and minerals), nutritional status (blood and urine biomarkers), health status (body weight, body mass index [BMI], waist circumference, blood pressure, metabolic function, clinical chemistry, gut microbiota composition), acceptability of the dietary changes and additional environmental indicators.

Study participants and recruitment

The Human Research Ethics Committee in University College Dublin (LS-21-51-Davies-OSullivan) (affirmed by Faculty of Medicine, Health and Life Sciences Research Ethics Committee, Queen’s University Belfast MHLS_21_109) and the Clinical Research Ethics Committee of the Cork Teaching Hospitals in University College Cork (ECM 4 (cc) 10/8/2021 & ECM 3 (f) 19/10/2021) granted ethical approval for the study. Prior to beginning the study, participants received a Participation Information Leaflet and asked to sign consent (Appendix A). The study was carried out in line with the principles set forth in the Declaration of Helsinki. Participants were able to discontinue at any time at which point no further data was collected. Participants also had the option to withdraw their consent at which point data collected from the individual was destroyed. A standard operating procedure (SOP) was developed and agreed upon by all study site investigators. The SOP (version 1.0, 29 March 2022) for MyPlanetDiet is described here using SPIRIT reporting guidelines [23]. The protocol included procedures for conducting the trial and how to manage deviations. Participants who deviated from protocol continued the trial as close to protocol as possible and deviations were noted in participant logs.

The study aimed to recruit 360 participants evenly across three universities: University College Dublin, University College Cork, and Queen’s University Belfast. Participants were recruited through advertisements via email, radio, posters, public speaking events and social media. Sample size (80% power, 5% significance) was calculated based on achieving a 20% difference in GHGE between intervention and control diets. Allowing for potential age and sex differences and a 25% potential dropout rate, a total of 360 participants was required across the three study sites.

Healthy adults aged 18–64 years who consumed a moderate-to-high GHGE diet (self-reported red meat intake of ≥ 3 portions per week) were eligible to take part. Screening questions included demographic characteristics (e.g. age, sex) and lifestyle behaviours, including questions relating to habitual intake of critical food groups such as red meat, white meat, fish, eggs, and plant protein. The following exclusion criteria were applied;

  • Pregnancy, currently breastfeeding or females planning to become pregnant.

  • Following a medically prescribed diet.

  • Diagnosis of an acute or chronic condition that may interfere with the outcomes of the study. Conditions that are excluded include (but are not limited to) diabetes mellitus, inflammatory bowel disease, recent history/ongoing cancer treatment.

  • Immunocompromised or have a suspected immunodeficiency.

  • Excessive alcohol intake (> 28 units of alcohol consumed per week).

  • Known food allergies.

  • Regular consumption of a single high-dose vitamin or mineral supplement.

  • Participation in another research study.

  • Inability to read, write or understand English.

Randomisation and blinding

Following consent, participants were stratified by sex (female or male) and age (≤ 40, > 40 years), and randomised using site-specific blocked randomisation list to the intervention or control. As each new participant was recruited, they were allocated to the next available and ID code based on respective sex and age. Participants were blinded to their study; however, researchers were not.

Dietary intake and analysis

Dietary intake was measured prior to commencing the study or attending visit 1 (referred to from this point onwards as the “baseline” dietary assessment) with follow-ups at week 6 and week 12 (Table 1). Participants completed 3-online 24-hour recalls on non-consecutive days and a food frequency questionnaire at each timepoint using Foodbook24, a validated online dietary assessment tool [24]. Dietary intake data from the three 24-hour recalls was downloaded directly from Foodbook24 and includes nutrient composition as previously described [25]. Food frequency data were not used for the present analysis. Participants enrolled in MyPlanetDiet received personalised nutrition feedback based on mean daily intakes of nutrients and key food groups (as detailed below). A database was created to standardise how foods in the Foodbook24 food list contribute to relevant food groups. Single foods were directly matched with the relevant food group. Composite dishes were disaggregated to food group level using 3–5 online recipes. The mean ingredient amount (g) per 100 g of recipe was calculated accounting for cooking factors derived by McCance and Widdowson Composition of Food Integrated Dataset [26]. Ingredients weighing less than 10 g of the total recipe weight (< 1%) were excluded. The total weight of the recipe and the total weight of each ingredient were calculated, and the mean of each ingredient of the 3–5 recipes was calculated to show the average contribution per 100 g of recipe. Each ingredient was matched with a relevant food group; for example, a cheeseburger will contribute to red meat, dairy, and vegetable food groups.

Diet-related environmental impact data

All foods in the Foodbook24 database were assigned GHGE and water footprint values per 100 g of food using life cycle assessment data (LCA) published by Colombo and colleagues [27] in the UK, taking account of the proportion of local production and imported foods which is similar in Ireland [27, 28]. The environmental data published by Colombo and colleagues included LCA data from previous studies which encompass over 50 LCAs [27]. Colombo and colleagues further refined the environmental database to include 266 foods or food groups [27]. A stepwise procedure mapped foods from the Foodbook24 database to environmental data, starting with foods that mapped directly. The next step mapped composite dishes to their relevant food groups using the recipe database described in the previous section. The final step took foods that did not have respective environmental values and allocated them to a similar food group. The data was quality controlled, and a syntax was created to connect the food consumption database with the environmental impact database.

Sample and data collection

Participant data was collected at five timepoints (Table 1). Following screening and consent, participants completed a health and lifestyle questionnaire and the first dietary assessment (3-online 24-hour recalls and 1 food frequency questionnaire) at baseline prior to attending visit 1. The health and lifestyle questionnaire included socio-demographics, health behaviours, and self-reported anthropometrics which was used to estimate energy (kcal) requirements for the individual [29]. Participants were invited to attend two onsite visits at a study centre at the start of week 0 (visit 1) and at the end of week 12 (visit 2). At each visit, fasting anthropometry (height, body composition, hip circumference, waist circumference), clinical measurements (blood pressure) and biological samples (blood and urine samples and optional faecal samples) were collected. Biological samples will be used to measure changes in metabolic health, nutrient status, and gut microbiome composition. More detail on biological sample collect and data analysis can be found in Appendix B. All data was collected using standard operating procedures. Participants completed questionnaires onsite during visits including food waste, stage of change and diet-change tolerability using Qualtrics (Qualtrics XM, Seattle, USA), an online survey tool. The 36-item food waste questionnaire was developed and validated by Stancu and colleagues and included self-reported food waste quantities, behaviours associated with food waste, and attitudes towards food waste [30]. Participants completed a stage of change questionnaire at visit 1 which used an algorithm previously used in dietary change studies to assess readiness for dietary change [31]. Participants completed a tolerability questionnaire (adapted from a previous dietary intervention) at visit 2 to test the acceptability of the dietary changes, general well-being of participants, and ease of dietary changes [32].

Table 1 MyPlanetDiet data and sample collection
Full size table

Participant communication

All study resources, participant communication guides and nutritionist training were designed to ensure consistent participant interaction and engagement across groups and study sites. MyPlanetDiet participants were blinded, therefore no feedback message included language related to sustainability. A trained study nutritionist discussed the personalised nutrition feedback with each participant in-person using a feedback report (described later) at onsite visit 1 (week 0) and followed up with participants at weeks 1, 3, 6 and 9 to check-in on progress and reiterate dietary targets and improve adherence to study protocols.

Data management and analysis

An independent central data monitor was appointed from the project consortium. A data management plan was developed and agreed by all partners which included plans for data monitoring, sharing and dissemination. Upon signing informed consent, participants were allocated a study code which was used to store participant data during the trial. A file linking a participant’s information to their study code was stored during the trial in site-specific password protected files that only named researchers had access to. Data was deidentified upon study completion. Only project researchers will have access to datasets until grant completion. Dietary assessment and questionnaire data were exported from Foodbook24 and Qualtrics. Anthropometry data was input by researchers from case report forms. Data dictionaries were created and standardised across sites to code data. Deidentified data was merged across sites and quality controlled by researchers. Upon conclusion of the study, identifiable data was destroyed in accordance with relevant data protection acts. Biological samples will be destroyed after ten years to comply with the study’s ethical approval. Primary and secondary outcomes such as diet-related GHGEs, nutrient intake, nutrition status, and health status will be compared between intervention and control groups using general linear model (GLM) two-way repeated measures analysis of covariance (ANCOVA), controlling for covariates such as sex, age, BMI, and energy intake. Pearson’s correlation will be used to determine possible covariates. Results will be disseminated through academic journals, conference presentations and public speaking engagements.

Personalised nutrition feedback

Development of personalised nutrition feedback

The process of delivering personalised nutrition feedback included 4 stages: assessing dietary intake, using decision trees, selecting feedback messages, and compiling a feedback report (Fig. 1). This standardised process has been previously used and described in more detail [19, 24]. An individual’s dietary intake can be assessed based on their intake of critical food groups, such as fruit and vegetables, whole grains, dairy and protein foods. Food groups included in the personalised feedback are described in more detail later in this manuscript. Recommended intakes are used in decision trees to create a stepwise process for providing personalised feedback messages. Feedback messages included a specific intake target (e.g. grams per day per food group) and tips for how to achieve their target. For example, if a participant received a message to increase their whole grain intake, they would be provided with suggestions for how to swap refined grains for whole grains in their diet. Feedback messages were compiled into a report and provided to an individual. Feedback reports included 5 sections. ‘Your diet targets’ included individuals’ current intakes (by weight or servings) of key food groups compared to their personalised targets. ‘Your personalised feedback’ had actionable feedback messages derived from decision trees. Messages included specific advice to increase, reduce or maintain intake of food groups. ‘General feedback’ included considerations that could not be specifically personalised such as limiting ‘treat food’ intake. ‘What are your food groups’ included tables to define and explain food groups to the participant. ‘Your nutrient profile’ was a visual representation of baseline intakes, compared to recommended intakes using European Food Safety Authority (EFSA) DRVs [19] for 13 key nutrients and a traffic light system (Fig. 1).

Fig. 1
figure 1

Stages of delivering personalised feedback

Full size image

Personalised dietary advice was provided to the intervention and control groups to control for mode of delivery. To develop the personalised dietary feedback for the intervention group, a review of published literature and existing food-based dietary guidelines was conducted to identify critical food groups for a more sustainable diet [33]. Reference diets were compiled and compared to mean daily intakes from the national food consumption survey in Ireland (National Adult Nutrition Survey, NANS) to assess the compatibility between the two [34]. Mean food group intakes and patterns of consumption from NANS informed the personalised feedback messaging. For example, the personalised feedback recommended fewer larger serving sizes of 140–160 g for meat compared to healthy eating guidelines to reflect current dietary patterns. An intervention diet was developed from identified reference diet recommendations, taking into consideration current dietary intakes using data from NANS [3, 6, 7]. Five tiers of dietary advice depending on estimated energy needs (≤ 2249 kcal, 2250-2449 kcal, 2500-2749 kcal, 2750-2999 kcal, ≥ 3000 kcal) were created. Estimated energy needs were calculated by estimating resting metabolic rate with self-reported height and weight using the Mifflin-St Jeor equation and using activity factors which corresponded with self-reported physical activity using the short-form International Physical Activity Questionnaire (IPAQ) [29, 35]. Prior to generating personalised feedback, participants were screened for underreporting and underreporters were given an opportunity to repeat dietary assessments before progressing on the study as was done in the Food4Me study [36]. Dietary advice for each energy group was designed to meet energy needs and maintain similar macronutrient distribution and micronutrient intake. Thirty decision trees were created, five groups of six decision trees separated based on an individual’s daily energy requirement, ranging from 2249 to 3000 + kcal in 250 kcal increments. Decision trees provided feedback on six food groups in the following order: meat, plant protein, fish, dairy, fruit and vegetables, starches. The order of feedback in the intervention was set based on the highest potential impact to reduce diet-related GHGEs.

Personalised nutrition feedback for the control was based on healthy eating guidelines as described previously [24, 37]. Unlike the intervention, the control provided only one tier of recommendations for all energy needs. Decision trees provided feedback on five food groups in the following order: fruit and vegetables, wholegrain, dairy, fish, and red meat. The priority order of the feedback was decided based on the order of advice provided in healthy eating guidelines [21].

The median intake recommendations (grams per day) for intervention and control groups are presented in Table 2. The intervention provided specific recommendations for starchy vegetables, dark green vegetables, red/orange vegetables, and plant proteins.

Table 2 Median and range of personalised nutrition feedback recommendations (grams per day) by group
Full size table

Decision tree testing

The theoretical effectiveness of the intervention feedback process was assessed by modelling the expected effects on individuals’ diets. A convenience sample (n = 20) from the MyPlanetDiet RCT was selected and used to model theoretical outcomes and variation from the control and intervention groups (control and intervention) of the study’s personalised nutrition feedback system. The first five participants who joined the study in each stratification group (females aged 18–40 years, females aged 41–64 years, males aged 18–40 years, males aged 41–64 years) were included in the analysis.

Each participant’s baseline dietary assessments (3x online 24-hour recalls) were entered into Nutritics software (Research Edition, v5.85, Dublin, Ireland). Each participant’s food intake was considered separately and inputted by day (Day 1, Day 2, and Day 3) and sorted by ascending food code. A two-step process was undertaken. Firstly, an individual’s dietary intake was considered within the MyPlanetDiet feedback framework for both intervention and control, resulting in recommended changes to the diet for both groups. In step 2, conditional flow charts were used (Table 3) to adjust baseline diets in accordance with the appropriate feedback messages derived in step 1. Table 3 presents an example of how diets were modelled for fruit and vegetable intake in modelled control diets. Changes were made in order of ascending food code to reduce researcher bias in diet changes. Adjustments were made until the participant’s food intake aligned with feedback messages. Taking the vegetable example presented in Table 3; if a participant consumed 160 g of vegetables, then 80 g of unsalted, boiled mixed vegetables was added to their daily intake to meet the 240 g target. Step 2 was completed for all food groups and repeated for both the intervention and control dietary feedback. Detailed description of dietary modelling protocol is included in Online Resource 1. Following adjustments according to the theoretical changes, the modelled diets were reanalysed using Nutritics. As such, on completion of this exercise, 3 versions of each participant’s diet (Baseline, modelled control, and modelled intervention) were calculated and exported for analysis.

Table 3 Example of modelled changes to diets, shown for fruit and vegetables in control group
Full size table

Statistical analysis

Statistical analysis was performed using IBM SPSS Statistics version 27 (IBM Corp., Armonk, NY, USA). Data are presented as n (%) or means ± SE where appropriate. Variables were checked for normality using Shapiro-Wilk tests and histograms. Mean daily nutrient intakes and environmental indicators were calculated for baseline, modelled control and modelled intervention diets and analysed using GLM repeated measure one-way ANOVA where macro- and micro-nutrient intakes and absolute environmental values were controlled for energy intake. Mean energy (kcal) from baseline, modelled control and modelled intervention were compared to estimated energy needs using paired samples t-tests. Individuals included in the present analysis were assigned EFSA DRVs for critical nutrients depending on sex and age. Each individual’s mean daily micronutrient intakes were compared to their corresponding EFSA DRVs (PRIs or AIs).

Results

Environmental indicators

The presented analysis was based on 20 baseline assessments from MyPlanetDiet, evenly represented by sex and age group. The participants had a mean age of 40 years and were representative of the total MyPlanetDiet RCT (Supplemental Table 3 Online Resource 1). Diet-related GHGE were highest among baseline diets (5.5 ± 0.4 kg CO2 equivalents per day) (p = 0.006) (Table 4). Modelled control diets had a mean diet-related GHGE of 5.4 ± 0.3 kg CO2 equivalents/day, 3% lower than baseline diets. GHGE associated with modelled intervention diets was 15% lower relative to baseline with mean daily GHGE of 4.7 ± 0.3 kg CO2 equivalents. There were significant differences in energy intake across baseline, modelled control, and modelled intervention diets (Table 5) with modelled intervention diets having the highest energy due to the energy-tiered feedback structure (p < 0.001). When GHGE were adjusted for energy intake, there were larger decreases in GHGE in both modelled control (-7%) and modelled intervention (-34%) diets relative baseline (p < 0.001). Baseline diets remained the highest for diet-related GHGE when adjusted for energy (2500 kcal) emitting 7.1 ± 0.5 kg CO2 equivalents on average per day. Modelled intervention diets’ mean GHGE/2500 kcal was 29% lower than the control diets. Baseline diets were adjusted based an individual’s respective energy needs resulting in mean GHGE of 7.6 ± 0.7 kg CO2 equivalents, 38% higher than unadjusted baseline intakes (Online Resource 1). Total water footprint (litres per day) was higher in modelled control diets relative to baseline (+ 5%) but when adjusted for energy (2500 kcal) the control diets had 4% lower water footprint (p < 0.001). The modelled intervention diets had the lowest mean water footprint in both total water footprint (838.7 ± 99.2 L of water/day) and water footprint per 2500 kcal (873.2 ± 118.7). The modelled intervention diets had 23% lower mean water footprint per 2500 kcal relative to baseline and 20% lower relative to the modelled control (p < 0.001). In baseline diets adjusted for energy needs, the mean water footprint was 31% higher than baseline diets as reported by participants (Online Resource 1).

Table 4 Diet-related greenhouse gas emissions and water footprint of baseline, modelled control, and modelled intervention diets
Full size table

Nutrient intakes

Nutrient intakes were significantly different between baseline, modelled control, and modelled intervention diets (p < 0.05) (Table 5). Energy intakes were highest in the modelled intervention diets (2562.5 ± 112.9 kcal/day) compared to baseline (1975.4 ± 100.1 kcal/day) and control (2088.3 ± 341.5 kcal/day). Mean energy from the modelled control diets and baseline diets were significantly different to mean estimated energy needs (p < 0.001) (Online Resource 1). The modelled intervention diets had highest percentage of energy from carbohydrates and the lowest percentage of energy from fat and saturated fats. Percent energy from protein was highest in the modelled control diets, while protein intake relative to body weight was the same in modelled control and modelled intervention diets and higher than baseline. Intakes of calcium, iodine, vitamin C and vitamin B12 were highest in modelled control diets. Fibre, iron, zinc, folate, vitamin A and sodium were highest in the modelled intervention diets. Intakes of critical micronutrients increased in both modelled diets relative to participants’ baseline diets.(Fig. 2)

Table 5 Mean baseline nutrient intakes compared to mean nutrient intakes of modelled control and intervention diets
Full size table

Nutrient intakes relative to EFSA recommendations

A greater proportion of modelled control diets and modelled intervention diets fell within dietary reference value (DRV) relative to baseline diets [38]. Modelled control diets had the highest proportion of individuals above the DRV for vitamin A, vitamin B12, vitamin C, calcium, and iodine [38]. Modelled intervention diets had the highest proportion above the DRV for vitamin B6, iron and zinc [38]. No baseline or modelled diet met the EFSA adequate intake (AI) for vitamin D (15 µg/day) [38]. Baseline diets and modelled control diets were below the respective DRV for zinc. The present DRV used for zinc is based on the highest level of phytate intake, 12.7 mg/day for females and 16.3 mg/day for males [38]. In the modelled intervention diets, 35% of diets were above the DRV for zinc [38].

Fig. 2
figure 2

Percent of individual diets meeting EFSA DRV (PRI or AI); EFSA, European Food Safety Authority; DRV, dietary reference value; PRI, population reference intake; Vitamin D, Vitamin B12 and Iodine; EFSA adequate intake (no PRI available); Zinc requirements use highest phytate recommendations for both males and females

Full size image

Discussion

This paper describes the protocol of a dietary intervention study and the development and testing of personalised nutrition feedback for a more sustainable and healthy diet. The personalised feedback was designed to be nutritionally adequate and acceptable while decreasing diet-related GHGE. If demonstrated to be effective, use of the presented personalised nutrition feedback can be easily scalable due to the reproducibility of the method [20]. Dietary advice was tailored to an individual based on their baseline dietary intake and their nutrition needs. Each individual received actionable feedback messages for critical foods groups with specific intake recommendations on a daily (e.g. one serving per day of dark green vegetables) or weekly (e.g. have oily fish once per week) basis. To our knowledge, the work presented here is the first to describe the development of a standardised approach for providing personalised healthy and sustainable dietary advice to individuals. Several other research groups have published reference diets for sustainable healthy eating, but have used one energy tier for recommendations, such as 2500 kcal or 10 MJ per day [3, 6, 7]. Such reference diets along with consideration of national dietary intakes were used as a basis for the present work, which was then expanded to consider how personalised factors influence nutrition needs and necessary dietary changes.

The dietary feedback framework was tested using a small sample of the MyPlanetDiet RCT to assess the effectiveness and impact of the personalised nutrition feedback on environmental and nutrient intake outcome measures. Diets modelled to the intervention recommendations had significantly lower GHGE relative to both baseline diets and modelled control diets. Energy intake in the modelled intervention diets were significantly higher due to the energy tiered feedback system which uses an individual’s estimated energy needs to provide dietary feedback. When each group’s diet-related GHGE were adjusted for energy intake (GHGE per 2500 kcal), larger differences in diet-related GHGE relative to baseline (-34%) were observed. Previously published research which modelled the environmental impacts of dietary changes found diet-related GHGE were significantly reduced (20–50%) with more plant-based diets [4, 27, 39, 40]. Our initial analysis aligns with previous research groups’ findings and with the primary outcome of MyPlanetDiet (20% decrease in GHGE). However, despite significant reductions in GHGE, the mean diet-related GHGE of the modelled intervention diets is still 2.3–2.7 kg CO2-eq above the recommended GHGE limit from previous reference diets [7, 14]. Due to the inclusion criteria of MyPlanetDiet, participants were screened and included if they followed key behaviours of moderate to high emitting diets. MyPlanetDiet was designed to be acceptable to those who consume such diets by considering dietary preferences (i.e. distribution of meat intake or portion sizes) from nationally representative dietary intake data. This consideration may impact the scope of reducing environmental impact.

Diet-related GHGE were marginally lower in the modelled control diets (-3% compared to baseline), but higher than previous modelled research which has estimated reductions in diet-related GHGE by 13% with adherence to food-based dietary guidelines [5]. However, while all participants in MyPlanetDiet were told to limit their intakes of discretionary foods, this was not a personalised feedback message in the study and was not modelled in the present work. Reducing discretionary foods in the diet is likely to reduce energy intakes, GHGE and water footprint and improve energy balance. Research suggests that eating within an energy balance is a core component for making a diet more sustainable and healthier [5].

The results of the modelled data suggest that adherence to both control and intervention diet recommendations would lead to improved nutrient intakes. Only the modelled intervention diets had mean values for percent total energy from carbohydrates, fat, and protein and total grams fibre within EFSA DRV macronutrient ranges [38]. The modelled intervention had the lowest percent energy saturated fat (10%) and was closest to EFSA reference value of < 10% [38]. Mean macronutrient distributions of the modelled intervention diets were comparable to other reference diets [6, 7]. Micronutrient intakes were significantly higher in both control and intervention diets relative to baseline. The intervention diets had mean micronutrient intakes above EFSA DRVs apart from iodine and vitamin D [38]. Control diets had mean micronutrient intakes above EFSA DRVs except for vitamin D and zinc [38]. However, no diet (baseline, modelled control, or modelled intervention) was nutritionally adequate for all nutrients for all individuals based on EFSA DRVs. Previous research has concluded that sustainable diets could lack sufficient intakes of vitamin D, vitamin B12 and iodine [6, 10, 13]. These findings may be indicative of the need for more animal sourced foods in the diet, increases in fortification of plant-based foods or supplementation. It is worth noting that baseline diets had fewer individual diets meeting adequate intakes for iodine and vitamin B12 relative to the modelled intervention diets, despite having more animal sourced foods. Yet contributions to critical micronutrients in the modelled intervention diets were predominantly from plant-based sources, and previous research has linked such diet patterns with increased risk of nutrient inadequacies due to reduced bioavailability [10, 41,42,43,44,45]. Future research should consider the difference in animal- and plant-based contributions to nutrient intakes and the corresponding impacts to health and nutrition status.

The approach used to model diets was consistent for each individual. For example, carrots were added into the diets of each individual who were not yet meeting their recommendation for red/orange vegetables in the intervention diet. Carrots are rich in beta-carotene and contribute to retinol equivalents (vitamin A), while other red vegetables such as red bell peppers would be higher in vitamin C, a micronutrient which only 60% of modelled intervention diets were above the EFSA PRI as opposed to 95% for vitamin A. Similarly, changes were made to daily diets based on ascending food code. This helped to eliminate researcher bias in the diet modelling, but also may have decreased diet variability across the sample. For example, many individuals had eggs, poultry, and pork in the same day, but eggs have a lower food code in Foodbook24. Eggs therefore remained in the diet over poultry or pork in these cases affecting the nutrient intake results. However, it is important to note that the findings of the decision tree testing are based on systematic decisions designed to replicate adherence to the personalised nutrition feedback. Individuals will respond to dietary advice differently in real life, which will be examined in future analysis from MyPlanetDiet. The variability in dietary intakes will be assessed using the study’s dietary assessments, and participants’ perceptions of the dietary advice will be captured in the diet-change tolerability questionnaire. The analysis of dietary changes over time and the acceptability of such changes are two outcomes that will be reported in future work.

Strengths and limitations

The MyPlanetDiet RCT is the first of its kind with the primary outcome of reducing environmental impact through personalised dietary advice at an individual level. To our knowledge, the study will be the first to test adherence to a more sustainable diet and to assess the safety and nutrition adequacy of such a diet in a human intervention study. Biological samples collected during the study will be used to assess markers of metabolic health, micronutrient status, amino acid status and gut microbiota composition. Participants on the study also completed tolerability questionnaires to determine the acceptability of the dietary change recommendations.

The nutrition feedback provided in the study has been refined, analysed, and tested to minimise risk of nutrient inadequacies. The decision trees have been tested using a modelled, personalised approach with a robust method designed to reduce bias. In this modelled approach, foods were exclusively modelled based on feedback participants received, then based on daily intake by ascending food code. While this kept the diet modelling consistent for each individual, it may skew nutrition data due to reduced diversity within critical food groups which impacts micronutrient intakes as described above in relation to red/orange vegetables. Similarly, a limitation of the dietary assessments is that the data is based on what individuals chose to report at one point in time. Further context to dietary assessment such as broad history of energy intakes, micronutrient intakes, or weight history could not be considered in the present analysis. Lastly, the only modelled changes that were made to individual diets were the changes specifically advised to each individual in their personalised nutrition feedback. Discretionary foods and other foods outside the feedback’s target food groups were unchanged. While we can anticipate individuals will make other dietary changes to compensate for the recommended dietary changes, these are unable to be captured or estimated in this type of analysis. The results of the MyPlanetDiet study will show the reality of dietary changes individuals made as part of the RCT’s dietary assessments.

Conclusions and future perspectives

The methods and study described will fill a critical gap in understanding the effectiveness, nutritional adequacy, safety, and acceptability of more sustainable dietary advice. Based on modelled data presented here, it is anticipated that the recommended dietary changes of both the intervention and control groups will decrease environmental impact and improve nutrient intakes. However, as we transition to a more sustainable healthy diet it is important to consider the potential impact that dietary change will have on the food system. For example, compared to current intakes, sustainable diets may increase fish consumption, so it is important to develop alternative practices to deliver a resilient fish supply without overfishing open waters. Future MyPlanetDiet results will help to better understand how individuals change their diets in response to sustainable dietary advice. This perspective should then be considered as part of a food systems approach alongside other system-level changes such as ethical and sustainable food production and processing as we move to protect our environment. Similarly, future results will report diet-related GHGE and water footprint of diets pre- and post-intervention, but modelled data shows that the diet recommendations are effective in reducing GHGE in line with the study’s primary outcome and power calculation. While previous research and the current analysis shows that a more sustainable diet would not be nutritionally adequate, modelled data shows macro- and micronutrient intakes will improve. Nutrition status will be tested with biological data collected in the MyPlanetDiet intervention. It is unclear how participants will alter their diets to compensate for the recommended dietary changes, but these results will be reported as part of the MyPlanetDiet study.