The association of social and food preparation location context with the quality of meals and snacks consumed by young adults: findings from the MYMeals wearable camera study

A sub-sample of young Australian adults were recruited from the larger MYMeals cross-sectional study [16]. Recruitment methods included electronic newsletters, noticeboards, social media, fundraising events (Relay For Life), letters of invitation using names provided by the Australian Electoral Commission and word of mouth by participants. Individuals were eligible for the study if they were: (i) aged 18–30 years; (ii) resided in NSW, Australia; (iii) owned or had access to a smartphone; (iv) could read and write English and (v) consumed at least one meal, snack or beverage purchased outside the home per week. Participants were excluded if they: (i) did not meet the criteria; (ii) were pregnant; (iii) were lactating; or (iv) had an eating disorder. Quotas for geographic location (metropolitan versus non-metropolitan, defined by the Accessibility/Remoteness Index of Australia [17]) and socio-economic status (SES) (higher versus lower using the Socio-Economic Indexes for Areas (SEIFA) [18]) were assigned using participants residential postcode to ensure the sample was from diverse demographic backgrounds.

Participants completed the online initial screening questions, provided informed consent, and completed the basic demographic questionnaire. Demographic questions included: gender (male, female or prefer not to say); age (18–24 or 25–30 years); residential postcode and if they would like to participate in the camera sub-study. Those that expressed interest in this sub-study were randomly selected to participate until a 20% sub-sample of the larger study was recruited. Participants were briefed about the requirements of the camera sub-study by the research team via telephone before receiving the camera.

A total of 216 participants were recruited into the validation sub-study but only 133 participants (62%) were included in this analysis. Participants were excluded from analysis for the following reasons: (i) had camera data of less than 8 h per day across the 3 days (n = 48); (ii) did not complete all 3 days of data collection (n = 21); (iii) withdrew from the study for personal or employment reasons (n = 5); (iv) did not have camera data (n = 4); (v) had incorrect camera settings (n = 3) or (vi) failed the selection criteria (n = 2).

A wearable camera, Autographer, along with a portable charger, connecting cables, information card, instruction manual and postage paid reply bag was provided to participants via post mail. The five-megapixel camera measured 90 mm × 37.5 mm, weighed 58 g and captured images with a 136° wide angle lens that was saved on to the 8 GB internal memory. Researchers instructed individuals to wear the camera on a lanyard around the neck positioned on the sternum. The device was programmed to capture point-of-view images in 30 s intervals. Occasionally, the device captured images more frequently if it sensed changes in movement, light, or temperature. Participants were asked to wear the camera for 3 consecutive days such that across the population the study start days were distributed to capture both weekdays and weekends. Cameras were to be worn for all waking hours, and participants were instructed to undertake their normal activities and to charge the camera overnight. To be included in the dataset, participants were required to capture at least 8 h of images per day. Participants were able to remove the camera or to close the lens cover to temporarily cease image recording when privacy was required (i.e., in bathrooms), on government premises or if they (and/or others) felt uncomfortable having images taken in certain settings. Participants were provided with the opportunity to review and remove any images for privacy reasons prior to returning the cameras to researchers via postage paid mail upon completion of the study. All captured images were transferred from the internal camera storage onto a secure university research data storage facility. Data access was restricted to approved members of the research team who were briefed on the strict privacy protocol. Anonymity of participants and third parties in image data was protected by obscuring faces and any other identifying information. Participants received AU $100 voucher on receipt of the camera.

Self-reported weight (kg) and height (cm) were collected in the questionnaire administered at the end of the 3-day recording [16]. Weight and height were used to calculate Body Mass Index (BMI = weight kg/height m²). This self-reported weight and height has been validated in a previous paper [19].

Participants’ residential postcode was used to assign the participant as being from a higher SES area (top five SEIFA deciles) or lower SES area (bottom five SEIFA deciles) [18]. In the rare event the participant’s residential postcode did not have an associated SEIFA value (n = 2), SES was assigned based on the SEIFA decile of the adjacent postcode area.

Demographic and anthropometric information were de-identified and stored on REDCap electronic capture tools hosted at The University of Sydney [20, 21]. Ethics approval was obtained by the Institution’s Human Research Ethics Committee (2016/546) on the 15th of July 2016.

An image coding schedule and coding manual were developed and refined using an iterative process. A sample coding sheet and a summarised coding manual was provided (supplementary file 1). Prior to image analysis, training sessions regarding annotation rules and protocols were held to ensure coding was accurate and reproducible. A 90% agreement threshold was considered acceptable inter-rater agreement. Within the test dataset of 3557 images from a randomly selected trial participant, answers were generated by one researcher (V.C). Inter-rater reliability was tested with one other coder (A.D.) for eating episode label (100%), food types (100%), screen type (100%), social interaction (100%), preparation location (93%), consumption location (100%), overall rating (100%) and meal length (100%). Not all variables coded in the test dataset were required in analysis for this study.

Coding commenced in March 2019 and concluded in September 2020. The unique numerical identifiers of the captured images for each participant were tabulated in Microsoft Excel (Microsoft Corporation, Redmond, WA, USA) for annotation. Images were reviewed for image clarity (codable or not codable) and for the consumption of food (yes or no) by an Accredited Practising Dietitian (V.C). If food was consumed and the image was classified as codable, four coding domains were assessed, as illustrated in Fig. 1. As food was consumed over a period of time, all images captured (from the first to the last image when the food was observed to be consumed) were considered part of the same eating episode.

The first coding domain classified eating episodes as: breakfast (consumed between waking time to 11 am); lunch (12 pm to 3 pm); dinner (6 pm to 9 pm) or snack (consumed before or after any of the main meals). The second coding domain was the classification of the meals and snacks as consisting of food items that were predominately from the recommended FFG or discretionary (i.e., foods that should be restricted) by visual inspection of the images. These food classifications were defined above and are the current system used in the official Australian Guide to Healthy Eating [10]. FFG foods are defined in the AGHE as nutritious foods and include: (a) grain (cereal) foods mostly wholegrain and/or high cereal fibre varieties; (b) fruit; (c) vegetables and legumes/beans; (d) lean meat and poultry, fish, eggs, tofu, nuts and seeds and legumes/beans; and (e) milk, yoghurt, cheese and/or alternatives, mostly reduced fat [5, 10]. Discretionary foods are defined in the AGHE as foods that are high in added saturated fat and/or added sugars, and/or salt, and/or low in fibre [10] and are sometimes referred to as energy dense and nutrient-poor items in other countries.

In the instance that meal or snack consisted of multiple food items that were a mixture of FFG and discretionary foods, the overall rating of the meal or snack was assigned according to the predominant food items by observed volume. For example, if a participant was observed to consume a grilled steak (FFG item) with steamed vegetables (FFG item) and a small side of fried potato chips (discretionary item), the meal was rated as predominately FFG as most of the meal consisted of food from the FFG.

If a meal or snack was observed to be consumed but its components could not be clearly identified, the meal was rated as unclear to reduce the possibility of misclassifying ambiguous image data (see Fig. 2e). Images that could not be coded for any reason (e.g. poor lighting conditions, blurry due to rapid movement or blocked by an object) were classified as not codable (see Fig. 2f).

The third coding domain classified where the food item was prepared: (i) within the home; (ii) FOH (including both purchased and non-purchased meals and snacks); (iii) both inside and outside the home; or (iv) unclear. Non-purchased meals and snacks FOH could include free food samples at food outlets. As each component of meals and snacks were individually reviewed, items could be prepared from both inside and outside the home. For example, a participant may have consumed a sandwich purchased from a café (FOH) and a fruit salad prepared within home in the same episode. If a meal or snack was observed to be consumed but its preparation location could not be coded for any reason (e.g. components could not be clearly identified), it was annotated as unclear. The location of consumption was coded but not included in analysis as this study aimed to understand the association of where the food was prepared rather than consumption location. This may be useful to inform future health policies and food labelling regulations.

The fourth coding domain assessed the social context during meal and snack consumption. This domain consisted of the two sub-domains: (A) social interactions and (B) screen use.

Sub-domain A of the fourth coding domain assessed if the participant engaged in any social interactions during the eating episode. Social interactions were recorded if the participant was seen to engage with another person (or people) as indicated by conversation, smiling or body language. The people that the participant interacted with were classified into one of five categories: (i) family and/or partner, (ii) friend, (iii) colleague and peer, (iv) other or (v) mixed. Categories were assigned according to the observed interaction location and age of the person. For example, if the person only interacted with the participant at the workplace they would be coded as colleague and peer. However, if the young adult interacted with an older adult within a home setting, they would be assumed to a parent and coded as family and/or partner. If participants interacted with people from different categories during an eating episode, for example, interacting with their family members and friends at a group gathering, it was classified as a mixed interaction.

Sub-domain B of the fourth coding domain identified if any electronic screens were used during the eating episode. This was defined as a participant engaging with a screen (such as touching a smart phone). Screen types included: (i) mobile phone, (ii) laptop or computer, (iii) television, (iv) other screen type and (v) mixed type (i.e. more than one screen type used). For example, the mixed screen type was used if the participant was on their mobile phone whilst watching television.

Figure 2 provides an example of coding.

Mixed binary logistic regression was used to understand the association between preparation location, social interaction, screen use, gender and SES on the overall classification of meals and snacks as predominately from the FFG or discretionary. In the mixed model, a random effect for participants was included to control for correlation among repeated measures from the same participant. Due to small numbers, social interaction and screen use categories were reduced to two categories: (a) none or (b) present. For social interactions, “present” covered: (i) family and/or partner, (ii) friend, (iii) colleague and peer, (iv) other and (v) mixed. For screen use, “present” included: (i) mobile, (ii) laptop or computer, (iii) television, (iv) other screen type and (v) mixed type. Meals and snacks prepared as a combination inside and outside the home were grouped with outside the home category as this only occurred for 23 episodes. Meals and snacks rated as unclear were not included in the analysis as they could not be categorised as predominately from the FFG or discretionary. Data were analysed using SPSS version 25 (IBM Corporation).

The demographics of participants and camera wear times are presented in Table 1. The sample comprised of 55% female participants and 55% of the sample were aged in the 18–24 years bracket. The average BMI was 25.1 kg/m² (SD: 5.2) and the majority (60%) were within the healthy weight range. Regarding SES, 65% were from a higher SES area and 35% from a lower SES area. Two participants reported postcodes that had no SEIFA assigned [18], one reported a postcode of 2005 that was changed to 2007 (SEIFA decile 8, higher SES area) and the other was 2308 that was changed to 2307 (SEIFA decile 2, lower SES area).

A total of 1840 eating episodes (main meal or snack) were identified, 7% (35,422 out of the 487,912) images reviewed images captured food consumption. Eating episodes were excluded from analysis if the preparation location could not be determined (n = 8) and if they were rated as unclear (3% of all eating episodes; breakfast n = 8; lunch n = 16; dinner n = 19 and snacks n = 14). A total of 1775 eating episodes were included in analysis of which 289 were identified as breakfast (16%), 353 as lunch (20%), 322 as dinner (18%) and 811 as snacks (46%). Of the main meals 79% of breakfast, 69% of lunch and 72% of dinner were rated as consisting of items mostly from the FFG. However, a higher proportion of snacks was identified as consisting of predominately discretionary items (60%).

Main meal (breakfast, lunch, or dinner) and snack consumption characteristics are shown in Figs. 3, 4and5.

As shown in Fig. 3, most main meals were rated as containing items predominately from the FFG if they were prepared within the home (69% of all breakfasts, 48% of all lunches and 57% of all dinners). Snacks were more likely to be rated as comprising of mostly discretionary items rather than from the FFG regardless of if they were prepared at home (39% of all snacks) or FOH (21% of all snacks).

As seen in Fig. 4A, most meals and snacks were consumed alone rather than with social interactions regardless of whether they were rated as comprising of mostly items that belonged to the FFG (58% of all breakfasts, 42% of all lunches, 33% of all dinners and 28% of all snacks) or discretionary (16% of all breakfasts, 16% of all lunches and 36% of all snacks). Dinner was the most social meal with 59% of meals consumed with social interactions. The most common type of social interaction during dinner meals was with family and/or partner regardless of if the dinners were identified to be mostly composed of items from the FFG (24%) or discretionary (11%).

As shown in Fig. 4B, most meals and snacks were consumed with at least one screen type in use regardless of if they were rated as consisting of mostly foods from the FFG or discretionary items (69% of all breakfasts, 66% of all lunches, 59% of all dinners and 58% of all snacks consumed). Mobile phones were the most common screen type used during main meals (used at 35% of breakfast, 32% of lunch and 19% of dinner episodes). Mobile phones (17%) and laptops or computers (17%) were the most used devices during snacks. Out of all meals and snacks, television was most watched during dinner (13%).

As indicated in Fig. 5A, females contributed n = 1018 eating episodes and males contributed n = 757 eating episodes. Females were more likely to consume main meals that were rated as consisting of foods mostly from the FFG (46% of all breakfasts, 41% of all lunches and 41% of all dinners) than discretionary (10% of all breakfasts, 14% of all lunches and 14% of all dinners). Similarly, males were also more likely to consume main meals where most of the items were from the FFG (33% of all breakfasts, 28% of all lunches and 31% of all dinners) than discretionary (11% of all breakfasts, 16% of all lunches and 14% of all dinners). Both genders were more likely to consume snacks that were made up of discretionary items (37% consumed by females and 23% by males) than from the FFG (22% consumed by females and 18% consumed by males).

As shown in Fig. 5B, participants classified as residing in a higher SES area contributed n = 1162 eating episodes and those residing in a lower SES area contributed n = 613 eating episodes. Those that resided in higher SES areas consumed more main meals that contained items mostly from the FFG (56% of all breakfasts, 49% of all lunches and 49% of all dinners) than discretionary (breakfast 11%, lunch 18% and dinner 18%). Similarly, those classified as residing in a lower SES area also consumed more main meals that contained items that belong within the FFG (22% of all breakfasts, 20% of all lunches and 22% of all dinners). Both SES groups were more likely to consume snacks that had a higher proportion of discretionary items (high SES 37% and low SES 23%) than components that belonged to the FFG items (high SES 27% and low SES 13%).

Lunch, dinner and snacks prepared within the home were more likely to consist of food items that belonged to the FFG than those FOH (Table 2). Snacks (OR = 3.2, 95% CI 2.2–4.8) were more than three times as likely to be predominately from the FFG when made at home. Lunch was 4.8 times (OR = 4.8, 95% CI 2.7–8.6) and dinner was 14.8 times more frequently categorised as containing items largely from the FFG (OR = 14.8, 95% CI 7.6–28.6) when prepared at home. Preparation location was not significantly associated with breakfast quality.

Breakfast consumed by those from a higher SES area (OR = 3.2, 95% CI 1.4–7.4) were more likely to consist of foods that were mostly from the FFG than those consumed by participants from a lower SES area. Additionally, lunch eaten by females (OR = 2.0, 95% CI 1.1–3.8) and higher SES participants (OR = 1.9, 95% CI 1.0–3.7) were almost twice as likely to be rated as being predominately from the FFG than those eaten by males or those residing in lower SES locations.

Social interaction and screen use were not found to be significantly associated with the meal or snack being composed of items mostly from the FFG or discretionary foods.