Setting and study design
This trial is a two-arm parallel, randomized, controlled study in real-world situations. This smartphone applet with built-in AI-based algorithm only takes effect where dishes prepared by recipes with quantity (in China, same dish prepared at different home are different in cooking but dishes in central kitchen or food factory are prepared strictly following standard cooking procedure [12]). Participants are enrolled sequentially until the sample size reaches requirement. This study will be carried out in a pilot company employing 3000 people and food is prepared by the central kitchen at staff canteen. All participants will choose and consume meals on their own wills during the study. Weight and blood pressure measuring tools placed at the canteen can be available to participants; therefore, they can conveniently measure and record anthropometric indicators through the applet.
Phase I. The employees will be recruited to dine in the canteen for 3 months, assigned to the intervention or control group randomly. Both groups will be asked to use the applet to record their lunch each weekday during the study. After 1 week of run-in period, the intervention group will be able to access the information of dish nutrition evaluation and PN evaluation after meal consumption, while the control group will not. Both groups will be followed up by researchers on the same time schedules for the outcome measurements.
Phase II. Full-functioned applet will be available for use by all the diners (about 800) for another 1 year. Who use the applet at least 2 days per week will be regarded as the intervention group while the others will be the control group. During this phase, metabolic indicators from the annual physical examination will be provided by the company.
Table 1
Visits and data collection schedule
|
Phase I (3 months)
|
Phase II (1 year)
|
|
Screening
|
Run-in
|
Follow-up
|
Screening
|
Run-in
|
Follow-up
|
Informed consent
|
√
|
|
|
√
|
|
|
Questionnaire
|
|
√
|
√
|
|
√
|
√
|
Dietary record
|
|
√
|
√
|
|
√
|
√
|
Anthropometric indicators
|
|
√
|
√
|
|
√
|
√
|
Metabolic indicators
|
|
|
|
|
√
|
√
|
Sample size
In Phase I, we conservatively assumed that the percentage of energy intake from fat as a key indicator will reduce by 6% after intervention. According to the findings from China National Nutrition Survey (CNNS), the SD of that indicator was 11% [13]. Thus, 85 participants in each group are expected to be enrolled with a 5% significance level, a 90% statistical power and a 15% dropout rate. In Phase II, assuming that the prevalence of metabolic syndrome after intervention is approximately 20% based on the available research evidence [14], we would need 200 participants in each group to have sufficient power to detect the given effect size.
Procedure
Inclusion and exclusion criteria
Men and women will be eligible to participate if they (1) are more than 18 years old, (2) are healthy in appearance, (3) promise to have lunch at the staff canteen during the study period, (4) agree to record food consumption of each meal on the applet. Exclusion criteria include: (1) planning to change physical activity habits during the course of study; (2) unable to follow a regular diet (e.g. on diet).
Recruitment and follow-up
Pilot. To establish the feasibility and acceptability of current protocol, we recruited 94 employees to perform a 3-week, single-arm preliminary pilot trial and users tested this applet (data collection from February 2022 to March 2022).
Phase I. Recruitment and physical examination are planned to conduct at regular intervals during 3 months. Researchers ask employees about participation in the study, collect written informed consent from participants, instruct them to use the applet as well as measure their anthropometric indicators in a standardized way. Thereafter, participants complete demographic questions (birthdate, sex and occupation) through the applet, as well as answer questions regarding their family history of NCDs, physical fitness, drug use, physical activity level (PAL), sleep, smoking status, alcohol consumption, dietary habits and nutrition service needs. The follow-up questionnaire will be repeated at the end of the phase I and questions about user experience of this applet will be added to it.
Phase II. The intervention will last for another 1 year. All the diners will access the full-functioned applet. Metabolic data will be collected from the annual physical examination by the company. Participants will complete the same electronic questionnaire at pre- and post-intervention.
Randomization and blinding
Following baseline measurements, the newly-enrolled participants will be randomized at a ratio of 1:1 to an intervention or control group. Randomization is stratified by sex and age. The trial is conducted as a double-blind study. The researchers are not blinded to allocation due to the nature of the intervention strategy, however the field investigators and participants will be blinded throughout the study, ensuring allocation concealment.
The intervention
The intervention is to provide AI-based dish nutrition evaluation and PN evaluation after meal consumption at the study canteen through the applet (Fig. 2). The applet based on privacy-preserving computing platform has two interfaces: a phone-based client and a web-based data management system. For users, this applet is a WeChat mini-program which has ease of use, high acceptance and little memory [15]. Similar to Facebook, WeChat is a very popular social software in China. The mini-program relying on WeChat can achieve health functions and provide native app-like experiences without leaving WeChat interface. For researchers, the web-based system is used to facilitate central management including user registration, recipe preparation, menu administration and information storage. The built-in AI algorithm is applied in data operation and processing to support iterative calculation.
Dish nutrition evaluation
Dish nutrition evaluation is used to determine whether the content of fat, sodium and sugar in dish is benefit for health.
Two independent datasets are used for our study. First is the Chinese food composition database [16], which is publicly available. This database includes nutritional values for over 1110 food items and corresponding food groups. Second, we create a recipe dataset of all the dishes supplied in this canteen. The recipe includes the raw weights of materials (including ingredients and condiments) and their edible proportion, as well as gross and single-portion cooked weights of each dish. Weight measurements are conducted by field investigators with expertise in nutrition under the same standard criteria.
We construct a full dish/nutrition dataset by connecting the recipe dataset to Chinese food composition database. Thus, the food groups, energy and nutrients for each dish are automatically calculated using Algorithm 1. Food groups are classified as cereals & tubers (grains, potatoes and tubers), vegetables (excluding legumes), fruits (including citrus), livestock and poultry meat, eggs and products, seafood, dairy, nuts, soybeans and products, cooking oil, salt and sugar. Plant food stands for cereals & tubers, vegetables, fruits, soybeans and products. Animal food stands for livestock and poultry meat, seafood, eggs and products. Nutrients include protein, fat, carbohydrate, cholesterol, sodium, calcium, iron, zinc and vitamin C.
Algorithm 1
Calculation of food groups, energy and nutrients for each dish
\(dish=\left(food item\left[1\right], food item\left[2\right], \dots , food item\left[m\right]\right)\) 1
\({Weight}_{food group,dish}={\sum }_{i\in dish}Raw Weight\left[i\right]\times Edible Proportion\left[i\right]\times \frac{{Cooked Weight}_{single-portion dish}}{{Cooked Weight}_{gross dish}}\) 2
\({Energy}_{dish}={\sum }_{i\in dish}Raw Weight\left[i\right]\times Edible Proportion\left[i\right]\times \frac{Energy\left[i\right]}{100}\times \frac{{Cooked Weight}_{single-portion dish}}{{Cooked Weight}_{gross dish}}\) 3
\({Nutrient}_{dish}={\sum }_{i\in dish}Raw Weight\left[i\right]\times Edible Proportion\left[i\right]\times \frac{Nutrient\left[i\right]}{100}\times \frac{{Cooked Weight}_{single-portion dish}}{{Cooked Weight}_{gross dish}}\) 4
1 By matching two datasets, food item represents raw material (ingredient or condiment) in dish.
2 Similar food items are merged into predetermined food groups.
3 According to Chinese food composition database, energy[j] refers to content of energy in 100g edible portion of food item[j].
4 According to Chinese food composition database, nutrient[j] refers to content of nutrient in 100g edible portion of food item[j].
The intervenors can browse and choose dishes on the ordering interface of the applet. Colored dots are displayed next to the dishes’ names by a “traffic light” approach to indicate whether the dishes are benefit for health. The judgement of three colors is based on the contents of fat, sodium and sugar (green = reaching the dietary recommendations, yellow = between the recommendations and average intakes among Chinese population [13], red = above the upper limit of intakes). The dietary recommendations in this study are: no more than 8g fat, 500mg sodium and 4.5g sugar in 100g dish (raw weight except for condiments). The cutoffs of nutrient contents according to the definition of “traffic lights” are listed in Table 2.
Table 2
The cutoffs of nutrient contents for dish nutrition evaluation1
Nutrient content
|
I
|
II
|
III
|
Fat (g/100g2)
|
< 8
|
8–20
|
> 20
|
Sodium (mg/100g2)
|
< 500
|
500–1000
|
> 1000
|
Sugar (g/100g2)
|
< 4.5
|
4.5-9
|
> 9
|
1 Green light for the dish represents all three indices within the range in the I column, red indicates at least 1 index within the range in the III column and yellow includes all the others.
2 100g refers to 100g edible portion of dish.
PN evaluation after meal consumption
PN evaluation after meal consumption is used to illustrate whether food intakes are inadequate, adequate or excessive.
A meal may consist of different dishes in varying portions. To obtain the actual intake of each chosen dish, nutritional values are multiplied by portions and non-discarded proportion that are input by participants on the ordering interface. Subsequently, to calculate the meal consumption, the consumptions of corresponding food groups, energy and nutrients from different dishes are summed up (Algorithm 2). The whole-day consumption can be further calculated according to self-reported contribution of three meals to total daily food intake.
Algorithm 2
Calculation of food groups, energy and nutrients for meal
$$meal=\left(dish\left[1\right], dish\left[2\right], \dots , dish\left[n\right]\right)$$
$${Weight}_{food group,meal}={\sum }_{j\in meal}{Weight}_{food group, dish\left[j\right]}\times Portions\left[j\right]\times \left(1-discarded proportion\left[j\right]\right)$$
$${Energy}_{meal}={\sum }_{j\in meal}{Energy}_{dish\left[j\right]}\times Portions\left[j\right]\times \left(1-discarded proportion\left[j\right]\right)$$
$${Nutrient}_{meal}={\sum }_{j\in meal}{Nutrient}_{dish\left[j\right]}\times Portions\left[j\right]\times \left(1-discarded proportion\left[j\right]\right)$$
According to individual’s biological profile and total energy expenditure (TEE), the specific recommended intake of the individual will be determined. TEE can be calculated using resting energy expenditure (REE) multiplied by PAL [17]. REE is estimated by Schofield equation regarding age, sex and body weight [18]. PAL is categorized into 1.5 for light, 1.75 for moderate, and 2.0 for vigorous physical activity [19]. Based on Chinese Food Recommendation [20], TEE ranging from 1000 kcal/day to 3000 kcal/day can be divided to 11 ranks and nutrition needs for a balanced diet pattern vary with different TEE ranks. Taking self-reported contribution of three meals to total daily food intake into consideration, recommendations of food groups and nutrients per meal can be assessed according to Chinese Dietary Guidelines [21] and Chinese Dietary Reference Intakes (DRIs) [19], respectively.
By comparison with recommended intake, the PN evaluation regarding actual intake will be provided for participants. The key evaluation indices for meal consumption are described in Table 3, including fat, sodium, meat and vegetables. For each index, we assign a health score with a corresponding mark to illustrate the comparisons between the current consumption and the national dietary recommendations. For example, a mark of three plus (+++) for fat indicates that the percentage of energy intake from fat is far from recommendation. Evaluation of fat and sodium is given greater weight than others, with a score of 0–3, because these two dietary factors are more related to burden of NCDs in China [22]. Finally, an aggregate score of all the indices will be obtained, which represents the healthiness of whole meal consumption, 0 being the unhealthiest, and 10 being very healthy.
Table 3
The key evaluation indices for meal consumption
Evaluation index
|
Range
|
Health score
|
Mark
|
Fat: percentage of energy intake from fat, %
|
< 20
|
2
|
-
|
|
20-<30
|
3
|
Good
|
|
30-<40
|
2
|
+
|
|
40-<50
|
1
|
++
|
|
50-<60
|
0.5
|
+++
|
|
> 60
|
0
|
++++
|
Sodium: actual intake, mg
|
< 1000
|
3
|
Good
|
|
1000-<1400
|
2
|
+
|
|
1400-<1800
|
1
|
++
|
|
1800-<2200
|
0.5
|
+++
|
|
> 2200
|
0
|
++++
|
Meat: actual/recommended intake ratio
|
< 0.8
|
1
|
-
|
|
0.8-<1.2
|
2
|
Good
|
|
1.2-<1.4
|
1
|
+
|
|
> 1.4
|
0
|
++
|
Vegetables: actual/recommended intake ratio
|
< 0.5
|
0
|
--
|
|
0.5-<0.7
|
1
|
-
|
|
> 0.7
|
2
|
Good
|
Outcomes
In Phase I, dietary pattern, body weight or blood pressure optimizing is expected after the intervention. The primary outcome for the intervention effectiveness is dietary intakes. Anthropometric indicators including weight, body mass index (BMI), body composition and blood pressure are the secondary outcome. In Phase II, body metabolism normalization is expected after this period.
Quality control
The quality control team was established prior to this study. All the researchers involved in this study must attend the complete operation training, including study protocol and standard operating procedures of participants’ data collection. In addition, both on-site and on-line review for data verification will be implemented, to ensure the quality and consistency of organizational operations, and stabilize food supply at the canteen.
Statistical analysis
The Mann-Kendall test will be implemented to examine the temporal trends of lunchtime food supply based on average “traffic light” score and animal/plant food ratio. For dietary record, we hypothesized that at least 5000 person-meal observations would be collected and analyzed over the follow-up period. Generalized linear mixed models with intervention, time and 2-way interaction as fixed factors, will be used to examine the intervention effects on each outcome. A two-sided p value < 0.05 will be considered to indicate statistical significance.
Trial status
The recruitment for the trial was initiated in September 2022, but the field work of Phase I suspended in December 2022 due to the COVID-19 epidemic. Subsequent work is under arrangement at the time of submission. This protocol was completed before the research team had received any data.