Healthy and Sustainable Diet Index: Development, Application and Evaluation Using Image-Based Food Records

There are limited methods to assess how dietary patterns adhere to a healthy and sustainable diet. The aim of this study was to develop a theoretically derived Healthy and Sustainable Diet Index (HSDI). The HSDI uses 12 components within five categories related to environmental sustainability: animal-based foods, seasonal fruits and vegetables, ultra-processed energy-dense nutrient-poor foods, packaged foods and food waste. A maximum of 90 points indicates the highest adherence. The HSDI was applied to 4-day mobile food records (mFRTM) from 247 adults (18–30 years). The mean HSDI score was 42.7 (SD 9.3). Participants who ate meat were less likely to eat vegetables (p < 0.001) and those who ate non-animal protein foods were more likely to eat more fruit (p < 0.001), vegetables (p < 0.05), and milk, yoghurt and cheese (p < 0.05). After adjusting for age, sex and body mass index, multivariable regression found the strongest predictor of the likelihood of being in the lowest total HSDI score tertile were people who only took a bit of notice [OR (95%CI) 5.276 (1.775, 15.681) p < 0.005] or did not pay much/any attention to the health aspects of their diet [OR (95%CI) 8.308 (2.572, 26.836) p < 0.0001]. HSDI provides a new reference standard to assess adherence to a healthy and sustainable diet.


Background
As evidence linking dietary behaviours to climate change is advancing, national dietary guidelines are incorporating specific advice to inform the public and guide policy [1]. People can lower the environmental impact of their dietary behaviours without sacrificing nutrient intake by adopting a more sustainable diet, and in many scenarios, can improve health and reduce risk of premature mortality [2]. By adopting more environmentally sustainable dietary behaviours, such as shifting to more plant-based proteins, people can support sustainable food systems and reduce the impact of dietary patterns on the climate [3]. The EAT-Lancet Commission on the environmental sustainability of food and the impacts of food consumption on health recommended that to achieve sustainable food systems, a shift to healthy and sustainable dietary patterns is needed [3]. Despite increasing recognition of the need for sustainable, resilient food systems for healthy diets and the The population-based sample consisted of 247 adults aged 18 to 30 years who were recruited via the Federal Electoral Roll during the Connecting Health and Technology (CHAT) study [22]. Data were collected from the same participants on two separate occasions, six months apart. Further details about the methods have been previously outlined by the authors [22]. The CHAT study was registered on the Australian and New Zealand Clinical Trials Registry (ACTRN12612000250831) and approved by the Curtin University Human Resources Ethics Committee (HR181/2011) and the Western Australian Department of Human Research Ethics Committee (#2011/90).

Assessment of Healthy and Sustainable Dietary Behaviours
The protocol outlining the justification and the methods used to assess individual H&S dietary behaviours have previously been published by the authors [8]. Briefly, 4-day image-based mFR TM were used, in which participants captured images of all the foods and beverages they consumed over four consecutive days. The H&S dietary behaviours were assessed by a trained analyst using images captured with the mFR TM , including (1) intake of animal-based foods, including ruminant meat, pigs, poultry, fish, seafood, dairy foods and eggs; (2) intake and seasonality of fruits and vegetables (including legumes and beans) and other plant-based foods high in protein (including nuts, seeds and tofu); (3) intake of ultra-processed energy-dense nutrient-poor (EDNP) foods and beverages, as defined by Monteiro and colleagues [23], including sugar-sweetened beverages (SSB) and alcohol; (4) use of individually packaged foods (see Figure 1); and (5) food (plate) waste behaviours. The theoretically driven HSDI contains twelve individual items related to H&S dietary behaviours. The items chosen to be included in any diet quality index are a compromise between what information is available and what information is practical to include, which is often driven by the dietary assessment method used to collect the data. Therefore, deciding on the dietary components included and excluded involves an element of researcher subjectivity. Details explaining greater detail of the components included in the HSDI have been described by the authors, previously [8]. The individual items and their respective weightings can be seen in Table 1. The influence of dietary behaviours on human health were given the highest weighting, followed by evidenced impact on the environment. This was due to stronger evidence in this area of dietary intake and health outcomes. A maximum score of either five or ten points was allocated to each component of the HSDI. To determine the weighting, each component in the HSDI was categorised into The theoretically driven HSDI contains twelve individual items related to H&S dietary behaviours. The items chosen to be included in any diet quality index are a compromise between what information is available and what information is practical to include, which is often driven by the dietary assessment method used to collect the data. Therefore, deciding on the dietary components included and excluded involves an element of researcher subjectivity. Details explaining greater detail of the components included in the HSDI have been described by the authors, previously [8]. The individual items and their respective weightings can be seen in Table 1. The influence of dietary behaviours on human health were given the highest weighting, followed by evidenced impact on the environment. This was due to stronger evidence in this area of dietary intake and health outcomes. A maximum score of either five or ten points was allocated to each component of the HSDI. To determine the weighting, each component in the HSDI was categorised into one or more of the following elements: impact on human health and/or impact on the environment. For example, ultra-processed EDNP foods and beverages affect health (contributing excess kilojoules (kJ) and contributing to chronic disease risk) and the environment through the use of land, water, electricity, transport, packaging, storage and disposal [24]. Therefore, these foods and beverages were given a maximum weighting of ten points each. Another example is food waste, which has a direct negative impact on the environment (resources used to dispose of waste and landfill) and a potential influence on health, as fresh fruit and vegetables are perishable and often thrown away, creating a barrier for purchase and consumption. However, due to limited evidence for the latter, food waste was assigned a maximum of five points (for an average of ≤10% of edible plate waste over the 4-day mFR). Ten points were allocated to behaviours that positively or negatively align with both a healthy and sustainable diet (such as vegetables), and a maximum of five points were allocated to other dietary behaviours, such as food packaging. The HSDI used a continuous weighting system based on increments of food group serves, or proportions of total intake for behaviours without set recommendations, such as seasonality of fruits and vegetables and plate waste. In a review of diet quality indices, the method of using continuous scales (opposed to simple cut offs or binary scales) has been identified as a superior method because the intake of many foods has a "U-shaped" effect [25], and continuous scales allow for more variability and provide more sensitivity within the index. For the present study, determining the categories for the continuous scale involved developing a maximum score for each component (as described above), then dividing it into five categories, each with different scores assigned. In circumstances where these dietary behaviours were reflected in the ADGs (such as fruit and vegetables, and milk, yoghurt and cheese) the maximum weighting assigned to the item was determined by the ADGs recommended daily number of serves [6]. In circumstances where the behaviour was not reflected in the ADGs, such as seasonality, food waste or individual food packaging, categories were created from the scientific literature.
Some food groups in the ADGs were separated into several items in the HSDI due to notable differences in the environmental impact of these foods. For example, the "lean meat, poultry, fish and alternatives" food group was separated into "ruminant meat and pig", "poultry and fish" and "non-animal protein alternatives". To minimise foods being picked up in two or more components of the index, for example, legumes and beans appear in both the "vegetable" and "lean meat and alternatives" food group, were assigned to the "non-animal protein foods" group. Some foods, such as EDNP foods may have been counted in both the categories of "individually packaged foods" and "ultra-processed EDNP foods". However, to minimise repetition, foods such as ice cream were only counted as ultra-processed EDNP foods, and not "milk, yoghurt and cheese".

Statistical Analyses
A purpose-built Microsoft Access Database was used to record the components of an H&S diet. This study involved a secondary analysis of 4-day mFR TM collected from a population-based sample at baseline (n = 247) and six months (n = 220). Once the secondary analysis of all mFR TM (approximately 12,000 food images) was complete, the dietary data was exported to SPSS Version 22 and merged with the participants' anthropometric and demographic characteristics. SPSS Version 22 was used for all data analyses and p-values of less than 0.05 were considered statistically significant.
Five stages of analysis were conducted to assess the following: 1.
Descriptive statistics about the sample, including demographic, anthropometric, and dietary variables.

2.
The specific dietary differences between participants with the lowest, middle, and highest total HSDI scores. This was conducted by separating the participants' total HSDI scores into tertiles using the SPSS rank function. One-way ANOVA was used for continuous variables (age and body mass index (BMI)) and the Chi-Squared test for all remaining categorical variables.

3.
The relationship between the components of the index to assess if H&S dietary behaviours are related. This was conducted using the non-parametric test, Spearman's correlation coefficient.

4.
Regression analyses were conducted to assess which variables help determine the characteristics of those who are in the lowest tertile for total HSDI score and whether any individual variables were predictors of overall HSDI score. Univariate regression analyses were conducted to identify which individual variables predict those most at risk of being in the lowest tertile of HSDI score (20-38 out of 90). Univariate regression analyses were then conducted after adjusting for age, sex and BMI. Multivariate regression analyses were conducted to see which variables continued to determine those most at risk of being in the lowest tertile when including all variables in the model.

5.
The test-re-test reliability of the index was assessed by comparing individual components and the overall HSDI score of participants who completed the CHAT study (mFR TM collected at baseline and at the six-month visit (n = 220)).

Results
Descriptive statistics from the 247 participants are outlined in Table 2. One participant was excluded from the analysis due to an incomplete food record. 77.2% of the participants were White, 58.5% had a BMI in the healthy weight range (18.5-24.9 kg/m 2 ), 32.1% were classified as either overweight or obese, and 37.4% reported taking vitamin supplements. The mean intake of fruit was 0.9 (±0.7) serves per day and the mean intake of vegetables was 1.8 (±1.0) serves per day. More than half (51.4%) of the fruits and vegetables consumed were in season in Western Australia at the time of consumption, and 20% of edible food prepared and served was assessed as edible plate waste. The intake of individual components of the HSDI are shown in Table 2. The mean intake of ultra-processed EDNP foods and sugar sweetened beverages (one serve is equivalent to about 600 kJ) over 4 days was 2.7 (±1.4) and 1.0 (±1.0) serves per day, respectively, and participants consumed a mean of 1.9 (±1.4) individually packaged EDNP items (such as a chocolate bar or can of SSB) and 1.6 (±1.2) individually packaged healthy items (such as a bottle of water or small tub of yoghurt) per day.
People who reported taking vitamin supplements in the self-reported written questionnaire were significantly more likely to have a higher HSDI score than those who did not (p < 0.005). Those who reported paying a lot of attention to the health aspects of their diet were more likely to have a higher total HSDI score than those who reported not thinking much or at all about the health aspects of the food they eat (p < 0.0005). There were statistically significantly differences in participants in each tertile and scores for individual components of the index (Table 3), which is expected, as the tertiles were ranked on total HSDI score, taking into account all components. However, the intake of seasonal fruits and vegetables; ruminant animal meat and pigs; and milk, yoghurt and cheese were exceptions to this. There were no significant differences detected between the tertiles. See Table 3 for the differences between the HSDI scores for each tertile.  Spearman's correlation test results indicate participants who ate ruminant meat and pigs were significantly less likely to eat vegetables (p < 0.001) ( Table 4). Those who consumed milk, yoghurt and cheese were significantly more likely to eat vegetables (p < 0.05). In addition, those who ate non-animal protein foods, such as legumes, tofu, nuts and seeds were significantly more likely to eat more fruit (p < 0.001), vegetables (p < 0.05) and dairy foods (p < 0.05). The strongest association found was between the intake of individually packaged EDNP foods and ultra-processed EDNP foods (p < 0.001) and EDNP beverages (p < 0.001) ( Table 4).  The univariate analyses showed that those not taking vitamin supplements were more likely to have an HSDI score in the lowest tertile (OR = 1.855, 95%CI [1.059, 3.250], p < 0.05) ( Table 5). This relationship was still significant after adjusting for age, sex and BMI, however, it was ruled out as a predictor for being in the lowest tertile of HSDI scores once all variables were taken into account in the multivariate regression. Participants who reported currently smoking were significantly more likely to be in the lowest tertile (OR = 3.407, 95%CI [1.065, 10.904], p < 0.05). However, after adjusting for age, sex and BMI, no significant association was observed. The strongest predictor of the likelihood of being in the lowest tertile for total HSDI score was dietary health consciousness. After adjusting for all other variables in the multivariate regression model, those who reported only taking a bit of notice (OR = 5.276, 95%CI [1.765, 15.619], p < 0.005) or not thinking much or at all about the health aspects of their diet (OR = 8.308, 95%CI [2.572, 26.836], p < 0.0001) were more likely to be in the lowest tertile of HSDI scores. Table 5 shows results from the univariate and multivariate regression analyses. The test-re-test reliability of the HSDI was assessed using data collected from the same sample on two different occasions, six months apart (n = 220) ( Table 6). The results indicated significant differences between the baseline and six-month visit for all components of the index, with the exception of non-animal protein foods and poultry, fish and eggs. The difference between the total HSDI scores for participants from baseline to the six-month visit was 4.1 points (p < 0.0005), with the six-month visit having improved HSDI scores (closer adherence to a healthy and sustainable diet). Table 6. Paired-sample t-test to assess the test-re-test reliability of the HSDI between data collected baseline and six months. Presented as HSDI scores * on participants who completed the study (n = 220).

Discussion
The aim of this study was to develop a theoretically derived Healthy and Sustainable Diet Index (HSDI) to determine a density score, and then apply the index on dietary intake using images captured by the mobile food record (mFR™). The findings suggest people who reported taking vitamin supplements had an increased likelihood of having diets more aligned with an H&S diet, regardless of age, sex or BMI. Previous studies have found those who are least at risk of poor nutrient intake are more likely to use nutritional supplements [27]. However, no research has examined the relationship between sustainable dietary behaviours and supplement use. Another key finding was that dietary health consciousness was the only independent predictor of one's likelihood of being in the lowest tertile of HSDI scores, when all other variables were included in the multivariate regression model. These findings support the inclusion of a measure of dietary health consciousness (e.g., "Which of the following best describes how you feel about the health aspects of your diet?") in future sustainable diet research. This question has been associated with high level of concern regarding the impact of the environment on the food supply [28] and EDNP food, SSB intake [29] and support for government interventions to promote dietary guidelines [30].
The evaluation of the HSDI found that individual components were related to each other, with people who displayed one behaviour being significantly more or less likely to also display another behaviour. For example, the intake of non-animal protein foods was associated with fruit (p < 0.001) and vegetable intake (p < 0.001), and all were aligned with an H&S diet. Conversely, the intake of EDNP foods, beverages and individually packaged foods were associated with one another, and all were unsupportive of a sustainable diet. The ability to detect these associations shows an element of sensitivity within the index. Further research applying principle component analysis to the HSDI scores will help determine if there are independent components of the Index and if any components can be ruled out.
To date, there is no evidence available on the use of individually packaged foods by consumers in Australia. The novel findings in this study show these foods and beverages, both EDNP and healthy individually packaged items, are consumed daily (1.9 ± 1.4 items and 1.6 ± 1.2 items per day, respectively). This dietary behaviour is of concern considering a sustainable diet, as ultra-processed EDNP foods and beverages: (1) provide minimal, if any, nutritional value; (2) encourage the overconsumption of kilojoules above energy requirements; and (3) require resources (such as water and electricity) for the extensive levels of food processing, and (4) food packaging. On a positive note, these findings confirm the value of recommending a dietary pattern for individual health and sustainability. Conducting image-based dietary assessment using the mFR TM to measure the use of individually packaged foods and edible food waste in real time was a unique aspect of this study.
The low mean HSDI scores for fruits (4.6 ± 3.3 out of a possible points) and vegetables (3.9 ± 2.1 out of a possible points) were unsurprising, as national population dietary surveys have shown poor compliance with recommendations for these food groups [31]. The high intake of EDNP foods were similarly expected, and the mean scores (2.1 ± 2.4 out of a possible 10 points) reflect the estimated 36% of the total daily energy intake of Australian adults coming from EDNP foods [31].
The study sample was predominantly female in the healthy weight range, although the mean BMI for both males and females was at the upper end with 24.7 and 24.1 kg/m 2 , respectively. The strengths of this study were that a population-based sample of participants was recruited via the Federal Electoral Roll and data were collected from the same participants on two separate occasions, six months apart. This enabled the test-re-test reliability of the index to be evaluated using a paired sample t-test.
The HSDI is the first diet quality index to use image-based food records to assess dietary behaviours that influence health outcomes (e.g., EDNP foods and beverages) and those that significantly burden the environment (e.g., individually packaged foods and food waste), which are often not assessed using traditional dietary assessment methods. In the absence of a gold standard for an H&S diet and the rudimentary evidence on the environmental impact of specific food groups in an Australian context, challenges arose when it came to its evaluation. Although validation of all dietary assessment methods is important to measure whether they accurately achieve their goals, first, a 'gold standard' is required [32]. Such a standard requires a strong evidence base, such as the Dietary Guidelines for good health, and does not exist regarding an H&S diet in Australia. First, a reference standard is required [32]. The present study developed a new reference standard to examine H&S diets using images, which can be used in future studies and applied to a larger population group and wider age range. The HSDI maximum score of 90 points was developed as a result of equal weighting of the elements of a healthy and/or sustainable diet that could be retrospectively assessed using the mFR TM images. There is potential to modify the weighting of individual components of the HSDI, and include additional components, as evidence on the environmental impact of foods evolves, and if the HSDI is applied to different settings.
The HSDI demonstrated its ability to assess the multidimensional nature of an H&S diet by incorporating 12 components into the index and finding significant associations between behaviours. Future research involving the application of the HSDI to a larger, more diverse sample, the collection of markers of health outcomes (such as blood lipid profiles as a risk factor for cardiovascular disease), and additional dietary behaviours, such as the use of nutritional supplements, will strengthen the evaluation of the index.
Similar to other methods, using the mFR TM to assess diet is accompanied by limitations. The primary limitation being participants forgetting to capture an image of an eating occasion. This can be minimised by the ability to set alerts on the mobile device to remind participants to capture images of all foods and beverages consumed. These alerts have previously been described by Ahmad and colleagues [33]. Another limitation is the potential for estimation error inherent in dietary assessment methods involving humans, including the use of a trained analyst in this study. The advancements in the use of mFR TM technology toward automated image analysis in the future may reduce this error. The influence of social desirability bias is a potential limitation of this study. In addition, due to the secondary analysis of existing data, the dietary behaviours of focus were limited to those collected from existing mobile food records during the CHAT study. As this study involved a secondary analysis of all mFR TM collected at baseline and at the six-month return visit of the CHAT study, the authors were limited by the data collected in this study. The future directions of this research could amend the mFR TM to collect additional information from participants, such as the assessment of nutritional supplement use through images and prompts to ask whether their food waste and packaging was put into landfill, recycling or composted.
Further research exploring consumer interest and awareness of H&S diets and modifications to the existing mFR TM app would strengthen the proposed method. For example, short survey questions could be included in the app to measure variables, such as dietary health consciousness or supplement use, to increase the level of detail collected for the dietary assessment method.

Conclusions
Dietary guidelines to encourage and promote behaviours that support good health and environmental sustainability are needed to inform and guide nutrition policy. Evidence on how current dietary patterns align with healthy and sustainable diets will highlight the need for these guidelines in Australia. This study provides a new reference standard for the Healthy and Sustainable Diet Index. Using the image-based mobile food record TM , dietary behaviours known to have a greater impact on the environment can be assessed without placing additional burden on users. The novel index using the mFR TM is a prediction model that can be applied to other population groups and datasets to further evaluate its ability to measure adherence to a healthy and sustainable diet.