The study was approved by the Human Participants Ethics Committee of the University of Auckland (ref 12330). A novel DIETCOST programme [18] was developed for researchers, using Python, to model the cost of healthy and current household diets using a list of commonly consumed foods, a set of min and max quantity/serves constraints for each, and specified food group and nutrient intakes based on dietary guidelines (healthy diets) and nutrition survey data (current diets for different population groups).
The programme was applied in New Zealand as a case study for the total population and specific ethnic population groups.
The reference household comprised a 45-year old man, a 45-year old woman, a 14-year old boy and a 7-year old girl.
Inputs
The following inputs were prepared as Excel files:
The lists of common foods for the total population and the different ethnic population groups (Māori and Pacific populations) were derived from the New Zealand adult nutrition survey and the New Zealand children’s food and drink survey [19,20,21,22]. Foods within the different groups were included if consumed by more than 5% of the population. The resulting common foods list for Māori and Pacific populations was additionally checked by experts from Toi Tangata (for Māori) and Pacific Heartbeat (for Pacific) and a few items were deleted or added. Some foods were only included for children or only for adults dependent on consumption frequencies from surveys and advice from the expert panels. The final common foods list contained about 120-133 food and takeaway items dependent on the population group.
The energy requirement for healthy adult diets was calculated using the Body Weight Calculator [23] based on a weight derived from a Body Mass Index (BMI) of 23 kg/m2, a mean population height [24], and moderate physical activity. The energy requirement for healthy children’s diets was based on the recommended energy requirements per KJ/kg per day from FAO/WHO/UNU [25] for moderate physical activity. The target weight was calculated using the 50th percentile BMI from the CDC growth charts [26] using mean height [24]. The energy requirement for current adult diets was based on the current BMI and moderate physical activity as over half of New Zealand adults meet the physical activity guidelines [24]. The energy requirement for current children’s diets was based on actual weight [24] and moderate physical activity as most children meet the New Zealand physical activity guidelines [27]. The additional energy required for the actual weight was calculated using a validated equation [28] for the excess energy intake per unit excess weight in childhood.
The daily food group and nutrient targets included serves of fruit, starchy and non-starchy vegetables, dairy, protein sources and grains, percentage of energy from fats, saturated fats, carbohydrates, protein, and total sugars, amount (g/mg) of fibre, red meat and sodium, and for certain scenarios the percentage of energy from alcohol and/or discretionary foods. For the healthy diet, these targets were derived from the serve sizes recommended in the New Zealand Eating and Activity Guidelines [29, 30] and the acceptable macronutrient distribution ranges, upper limit (sodium) and suggested dietary target (fibre) from the Nutrient Reference Values for Australia and New Zealand [31]. For the current diets, these targets were derived from average intakes reported in the nutrition surveys [19,20,21,22] and for sodium using a later survey which performed 24-h urine collection [32]. For current diets, about 30% of variation was allowed around the average population intakes for each of the targets, except for energy intake, where only 1.5% of variation was allowed.
The common foods list was used to generate the list of foods in the current and healthy diet baskets. The current diet basket for the total population contained 100 foods (including takeaways, excluding alcohol), while the healthy diet basket contained 73 foods (excluding takeaways and alcohol). Takeaway items and discretionary foods were included in the standard current diets but not in the standard healthy diets. Alcohol was only included in the diets for specific scenarios.
Compared to the current diet, the healthy diet basket contained a higher variety of fruits and vegetables, healthier versions of common foods (e.g. canned tomatoes without added salt, some wholegrain or wholemeal products, low fat yoghurt) and a limit on the consumption of red meat (maximum 100 g per day). For each common food, minimum and maximum serve sizes were set based on nutrition survey data, to avoid unrealistic amounts of any one food in the generated meal plans.
Food composition data
Nutrient composition data and edible cooking factors for the common foods were used from the New Zealand Food Composition Database [33] and the New Zealand Nutritrack database of packaged food products [34].
Food prices data
Two sources of prices data were used:
-
Prices were collected in spring (November 2016) in 12 Auckland supermarkets and their nearest fruit and vegetables store, in areas with different levels of deprivation. For each common food the cheapest price was collected, and the original price was also collected if the cheapest price was discounted. Prices of fruit and vegetables were collected in supermarkets as well as fruit and vegetable stores. In addition, for 6 out of 12 supermarkets, if the cheapest food was a generic item, the cheapest branded item was also collected. Based on consultations with Toi Tangata, a few common fruits and vegetables (e.g. feijoas, kamo kamo, puha, watercress) were included with zero cost for Māori, as these would always be gifted or gathered, not purchased. Based on advice from Pacific Heartbeat, for Pacific households prices were collected in 3 different additional supermarkets in South Auckland and fruit and vegetables were only priced from fresh produce markets.
-
The Food Price Index (FPI) data for New Zealand [35] was used for the period 2007-2016 to examine trends in the cost of current New Zealand population diets over time. As some healthier options for certain food groups were not included in this dataset (e.g. low salt or low fat products, wholegrain foods, butter but no margarine), trends in cost of healthy diets were not assessed using the FPI. Items in the FPI are selected based on their expenditure in the Household Economic Survey. Prices are collected monthly from 56 supermarkets across 12 regional centres and from fresh fruit and vegetable stores, fish shops, butchers, convenience stores, restaurants and takeaway food outlets [35].
Interface
The programme user interface [18] allows the user to specify the daily targets for the food groups and nutrients for all household members for current and healthy diets separately. In addition, the interface allows the user to specify whether or not to include takeaway meals, alcohol and discretionary foods as part of the diets.
The minimum serve size difference between any two generated individual meal plans was set at half a serve for any common food in this study.
The programme algorithm uses the Mersenne Twister as a random number generator to specify the starting meal plan and the starting value in grams for each of the common foods. If a meal plan meets all targets/constraints and is not already in the list of matching meal plans, it is added to the results. If it doesn’t (i.e. it fails some constraint), the algorithm will then try to fix that constraint (by raising/lowering the amount of some item that affects that constraint randomly between the min and max amount for that food item). If the modification results in a matching meal plan the meal plan is added to the results, and the same procedure starts again until the specified number of iterations has been run (Fig. 1). If the modification does not result in a matching meal plan, the algorithm will continue to try to resolve one of the failing constraints in a subsequent iteration. All success meal plans are independent from each other.
For each individual household member, the current and healthy diet scenarios were run multiple times with 1million, 2million and 20million iterations respectively to find the right number of iterations needed to accurately estimate the average cost of healthy and current household diets.
Validation of menu plans
A random selection of meal plans (N = 8, 4 for the healthy and 4 for the current diets) for the different household members generated by the programme was validated by a research assistant who made fortnightly household meal plans with the same amount of common foods manually to make sure that the meal plans were acceptable. A few food items were linked within the programme to ensure the resulting meal plans are acceptable, e.g. the programme code specified that the total number of serves of milk needs to be higher than or equal to the total number of serves of breakfast cereals and the total number of serves of spreads equal or lower than the total number of serves of bread and crackers.
Statistical analysis
All possible combinations of two-weekly meal plans for the four individual household members were assembled into two-weekly household diets for healthy and current diets separately. The range and distribution of the cost of the fortnightly household meal plans and the contributions of each food group and discretionary foods, alcohol and takeaways to the cost of the diets was calculated.
The impact of different prices, diets and policy scenarios on the cost differential between healthy and current household diets was also calculated.