Effectiveness of interactive technology-assisted interventions on promoting healthy food choices: a scoping review and meta-analysis

Abstract Making healthy food choices is crucial for health promotion and disease prevention. While there are an increasing number of technology-assisted interventions to promote healthy food choices, the underlying mechanism by which consumption behaviours and weight status change remains unclear. Our scoping review and meta-analysis of seventeen studies represents 3988 individuals with mean ages ranging from 19·2 to 54·2 years and mean BMI ranging from 24·5 kg/m2 to 35·6 kg/m2. Six main outcomes were identified namely weight, total calories, vegetables, fruits, healthy food, and fats and other food groups including sugar-sweetened beverages, saturated fats, snacks, wholegrains, Na, proteins, fibre, cholesterol, dairy products, carbohydrates, and takeout meals. Technology-assisted interventions were effective for weight loss (g = –0·29; 95 % CI –0·54, −0·04; I2 = 65·7 %, t = –2·83, P = 0·03) but not for promoting healthy food choices. This highlights the complexity in creating effective interactive technology-assisted interventions and understanding its mechanisms of influence and change. We also identified that there needs to be greater application of theory to inform the development of technology-assisted interventions in this area as new and improved interventions are being developed.

By 2030, more than 38·5 % of the global adult population will be living with overweight or obesity (1) , increasing one's risk of chronic diseases including cardiometabolic diseases (2) , certain cancer (3) , musculoskeletal disorders (4) , cognitive impairment (5) , and depression (6) . Local and international health organisations have implemented public campaigns, programmes and initiatives to improve population diets but remains insufficient. For example, one study found only a 16 % more people who were exposed to a public health advertisement focusing on healthy food choices and physical activity searched up on more information on weight loss as compared with the control group (7) . The authors further reported that an advertisement targeted at lifestyle preferences and sociodemographic profiles explained 49 % of the variance in responses, highlighting the intricate interactions between individual, interpersonal and environmental (micro and macro) factors (8,9) . Individual factors include biological (e.g. appetite and hunger), psychological (e.g. emotion-trigger eating) and cognitive (e.g. preference) factors and interpersonal factors include family, cultural and peer influence (10) . Micro-environmental factors includes schools, workplace, residential neighbourhood and community health care facilities. Macro-environmental factors include the built environment (e.g. transport and infrastructure) and food environment (e.g. food availability, accessibility and advertising) (11) . In this 21st century, technology has been integrated into our everyday lives and must be added to the obesogenic system of factors. For example, technology has been used as an obesogenic vector marketing practices leverages the power of artificial intelligence to influence consumer dietary preferences towards unhealthy food choices (12) . Food can also be conveniently, cheaply and readily obtained via smartphone food delivery apps, further promoting the consumerism culture that encourages easy consumption, overconsumption and food wastage (13) . On the other hand, technology has been used to improve eating habits through smartphone apps as an interactive interventions (requiring a two-way engagement between the user and technology system (14) , and hereinafter stated as technology-assisted interventions) to enhance health promotion efforts by prolonging engagement and hence behaviour change activation (15,16) . Such apps commonly include functions of food logging, goal setting and to deliver health messages, which has been shown in various systematic reviews to result in successful weight loss (17)(18)(19)(20) . However, the underlying mechanism by which technology-assisted interventions influence weight loss, perhaps through adopting a healthy diet, remains unclear (21) .
A healthy diet generally constitutes the consumption of a balanced diet that is rich in fruits, vegetables, wholegrains, fibre, lowfat dairy products, fish, legumes, nuts, PUFA, low saturated and trans-fats, sugar, refined carbohydrates and sodium (22,23) . However, research studies on the effectiveness of technologyassisted interventions seldom evaluate all the food groups that contribute to a healthy diet. To our best knowledge, there is no review on the food choices that are commonly examined as outcomes of technology-assisted interventions. Knowing the effects of such interventions on various food choices would inform the underlying mechanism of weight loss arising from such intervention and inform future health promotion interventions (24,25) . Therefore, we aimed to scope the food choice-related outcomes assessed in studies to conduct a post hoc evaluation on the effects of technology-assisted interventions on each of the outcomes using meta-analysis, whenever statistically possible.

Methods
We conducted a scoping review according to the Arksey and O'Malley framework (26) and reported our findings according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist (Appendix 1) (27) .

Search strategy
We searched through seven electronic databases (i.e. Embase, CINAHL, PubMed, PsycINFO, The Cochrane Library, Scopus and Web of Science) for articles published from inception through 22 March 2022. An initial search of PubMed was first conducted using keywords and medical subject headings (MeSH) terms derived from the concepts of 'food choice' and 'technology' to identify more keywords and index terms. The search was then refined according to each database using Boolean operators AND and OR and keywords shown in Appendix 2.

Study selection
Titles and abstracts screening were first performed by HSJC according to a prespecified set of eligibility criteria defined using the population, intervention, comparison, outcome and study design (PICOS) framework.
Population: Studies on adults aged above 18 years, normal or overweight were included. Studies on subjects with existing health conditions and pregnancy were excluded as the dietary requirements may be different.
Intervention: Studies on technology-assisted interventions were included. Studies that focused on food labelling, message reminders or other non-interactive provision of nutritional information were excluded. Studies that used a virtual supermarket as a test setting or an online home grocery delivery service with no other technology-assisted components were excluded.
Outcomes: Studies that measured food choice in terms of purchase or consumption were included. Studies that measured alcohol consumption only were excluded as alcohol is not a common part of one's daily diet.
Study design: Randomised controlled trials Studies that were not in the English language or was without version in the English language were excluded. Once the duplicated articles were removed, full texts of the articles were screened independently by HSJC and SC to further shortlist the articles to be included in this review. Interrater agreement was calculated for methodological quality assessment using the Cohen's Kappa statistic.
Of a total of 1324 articles retrieved, 526 duplicated articles were removed, resulting in 798 articles screened for eligibility using titles and abstracts. After removing 749 articles and adding three articles found through reference hand searching, forty-nine full-text articles were retrieved and screened for eligibility. Thirty-two articles were removed with reasons as shown in Fig. 1, resulting in a final seventeen articles included in this review. Sixteen articles reporting thirty-five unique outcome results were included in the meta-analysis. Interrater agreement for the risk of bias (RoB) was k = 0·822, P =< 0·001, indicating a strong level of interrater agreement.

Data extraction
A form was created using an excel spreadsheet to extract information according to the following headings: authors, year of publication, outcomes measured, measurement unit of each outcomes measured, country, sample size, sample characteristics, programme name, intervention type, intervention components, duration, intervention group condition, control group condition, delivery mode (i.e. individual or group), mean age, percentage of male subjects, socio-economic status, educational level, baseline weight, weight measurement instrument, baseline BMI, follow-up time point(s), attrition rate by the time of analysis, presence of comparison between participants retained and lost to follow-ups, method of missing data management (e.g. intention-to-treat (ITT)/per-protocol (PP) analysis), presence of protocol registration, and presence of funding. Data extraction was first piloted on three articles, and additional headings were added. Measures of central tendency (mean or mean difference) and variance (standard deviation or standard error) on each outcome were extracted in its raw form.

Methodological quality and certainty of evidence assessment
The methodological quality of the included articles was assessed using the Cochrane's RoB tool. Articles were rated as low, unclear or high RoB according to six domains namely random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, outcome data completeness, and selective reporting (28) .

Data analysis
Effect size estimates were converted to standard mean differences expressed as Hedges' g and pooled using random effects models. Hedges' g was used to correct for the small number of studies included in the meta-analysis (four to seven studies) where a magnitude of 0·2 = small, 0·5 = moderate, 0·8 = large and 1·2 = very large (29) . The Hartung-Knapp-Sidik-Jonkman (HKSJ) was used for to adjust the random effects models instead of the more widely used DerSimonian-Laird (DL) method as it has been shown to result in less false-positive estimates, especially in small samples and high heterogeneity (30) . Between-study heterogeneity was assessed using Cochrane's Q statistics and quantified by I 2 statistics where a statistic of 50 % indicates heterogeneity (31) .
All analyses were performed using R version 4.1.3.

Discussion
We conducted a scoping review and meta-analysis to provide an overview of the effectiveness of interactive technology-assisted interventions on commonly targeted food choice outcomes and consequently weight loss. The common food choice outcomes reported in the included articles were total calorie consumption and consumption of vegetables, fruits, healthy food and fats. Although more than 50 % of the included studies reported significant interventional effects on their respective outcomes, our meta-analysis only found significant interventional effects on weight loss. Given that weight loss results from a caloric deficit either from a decrease in caloric intake or an increase in caloric expenditure, the non-significant interventional effect on total caloric consumption remains unclear. Though four studies reported findings on both weight loss and total calorie intake, only one study reported consistent findings for the effectiveness of an interactive technology-assisted intervention resulting in a decrease in total calorie intake and significant weight loss (40) . One study (39) reported a significant weight loss but non-significant change in total calorie intake, while two studies reported the opposite (34,37) . One reason for this inconsistency could be due to the small number of studies included in the       meta-analysis, and the fact that there was a wide range of sample sizes thus the statistical weight could not be proportionally distributed by sample size. Another reason could be due to the proportionate increase in healthy food consumption (especially energy-dense food groups like protein and wholegrain instead of calorie-light food groups like vegetables) and decrease in unhealthy food consumption, leading to no change in calorie intake when aggregated (49) . Lastly, this could indicate the complexity in weight loss such that it should not be understood merely as an equation of calories intake and output, but also as an outcome of food quality. Further studies are necessary to ascertain the optimal dietary composition for weight loss, considering important biopsychosocial factors such as demographics (50) , environments, lifestyles (i.e. sleep, meal frequency and physical activity) (51) , resources, genetics (52) and gut health (53) which may not be as effectively influenced, if possible at all to do so, by interactive technology-assisted interventions.
Given the general prevalence of technology in our daily lives, it was surprising to have identified only a small number of studies that have developed, piloted and evaluated the use of technology-assisted interventions to influence food choices and consequently weight loss outcomes. Nevertheless, from the studies identified, 53 % were published in the last 4 years (from 2018), indicating a clear trend of more technology-assisted interventions being explored. In particular, the use of smartphone app-based intervention is the dominant choice, moving away from website-based and mobile text message-based interventions. It was not possible, in this scoping review, to analyse the effectiveness of the different types of technology-assisted interventions because of the heterogeneity of the interventions, but as a critical mass of similar interventions are tested and published this analysis should be conducted in future reviews to evaluate the effectiveness of these interventions.
From our reviewed studies, we also identified that the development of the technology-assisted interventions and the studies   Interactive technology on food choices 1257 evaluating them should be more theoretically informed. Majority of the studies (58·8 %) did not use or specify an underpinning theory or framework. Greater and effective use of theory going forward would be important in advancing the research and development of interventions in this area. The technology employed in the interventions are a means of delivering interventions, but these interventions should be theoretically informed to target specific levers of informed behaviour change. Introducing a technology without clearly understanding how it might lead to behaviour change should not be an intervention. This scoping review is not without limitations. Firstly, it might have been possible that some studies on the effects of interactive technology-assisted interventions on the consumption of various foods may have been excluded due to lack of mention about food choice, leading us to preclude these relevant studies. However, when identifying studies in this review, we searched using a list of commonly targeted food groups to ensure that we were able to identify the relevant studies. Secondly, with the small number of studies reviewed, an even smaller number of studies was included in the meta-analysis, and thus this could have introduced biased estimates. We tackled this problem by adjusting the random effects models with the HKSJ method, which is a well-established method for such situations (54) . Thirdly, due to the heterogeneity of the technology-assisted interventions identified, we were not able to conduct further needed analysis to compare between the types of interventions. Lastly, the studies reviewed here spanned a wide age range. Given that there might be differences in the level of affinity with technology and across age groups, this could have been an influential in some studies included in this review. Future intervention studies might consider exploring potential age differences as part of their evaluation process. This, together with the heterogeneity of interventions identified will be the remit of a similar review conducted in the future as the body of knowledge expands.
The above notwithstanding, in this scoping review, we have provided an overview of the available evidence on the use of technology-assisted interventions to improve food choices and its effectiveness on weight-related outcomes. Our meta-analysis found that technology-assisted interventions were effective for weight loss outcomes but not for improving food choices. This could be due to the heterogeneity within the small number of interventions identified in this review as this field is still in its nascency. We identified that there needs to be greater application of theory to inform the development of technology-assisted interventions in this area as new and improved interventions are being developed.