Efficacy of Emerging Technologies to Manage Childhood Obesity

Introduction

Childhood obesity is a major concern for healthcare systems and remains a challenge for medical innovation.¹ Cases of overweight or obese children (infant and young) have significantly increased worldwide to 41 million and are continuously increasing.² The huge socioeconomic transformations have significantly affected the nutrition and development of children and adolescents, particularly in developing countries. Increasing rates of urbanization has had a global effect on lifestyle, reducing the frequency of physical activity and increasing the calorie intake of nontraditional foods and “fast foods” rather than fruits and vegetables.^3,4 In addition to children showing a greater frequency of being overweight or obese, these food choices lead to nutritional deficits in growing children, which may lead to future health issues.⁵

Without intervention, overweight or obese children are more prone to carrying their condition into adulthood, possibly increasing their chances of developing diseases such as diabetes, cardiovascular diseases, and cancer.^6–8 Approximately 33% of preschool-aged children who are obese or overweight become obese adults, and 50% of obese school-aged children and 80% of adolescents carry obesity into adulthood.⁹ Therefore, it is crucial to develop strategies for the early diagnosis, prevention, and management of childhood obesity.

Currently available conventional interventions to manage weight are not always feasible as they are often inaccessible because of high cost, commute inconveniences, and lack of facilities.¹⁰ Extreme cases may require bariatric surgery, which is both expensive and risky.¹¹ Approximately 80% of children aged 12–17 years are hesitant to engage in any kind of coordinated activities, such as sports; therefore, a need for creative intervention is yet required.¹²

Over the past decade, advances in portable digital technologies that provide personalized access to relevant information have facilitated the evolution of preventive strategies and interactive aids and have become readily available. Smart digital health interventions are designed to cater to a wide number of issues because they help track and monitor health as well as collect information for analysis by a specialist to determine the required treatment strategy. These interventions include weight management programs based on websites¹³ or social media.¹⁴ They are wearable¹⁵ and are easily accessible via websites and mobile applications. These interventions are easily adaptable to different environments, affordable and time saving. Most importantly, strategies based on these technological tools keep children motivated and engaged while enjoying the comfort of their homes. Technology-based interventions for childhood obesity can be mainly classified into the following categories: (1) active video games or exergaming; (2) website, social media-based activities¹⁶ or mobile application-based activities;¹⁷ (3) machine learning (ML)-based strategies.¹⁸ A combination of those categories does also exists.

Mobile phones are the most readily and commonly available gadgets. They can be conveniently connected to wearables and sensors, providing the user with real-time data and monitoring physiological changes with internet cloud connectivity. Combining sensors and connectivity enables behavior (eg, physical activity and diet) to be monitored and influenced, and health care providers or experimenters can be alerted to changes (eg, weight and glucose). This technology offers a great opportunity to tap into new media channels that are integral to the youth culture because mobile phones are already widely used by children.¹⁹

The use of exergames, which combine exercise and video games, is another concept that is gaining popularity. These games are designed to encourage children to perform physical exercise or execute certain movements to clear levels and earn rewards in the game. They keep children motivated and engaged over time, making it easier for them to consistently follow a regular fitness program. Many available applications use this concept in combination with a wearable device. Exergames are suitable for every age group. For example, Let us move (To move!) is a program designed in the form of a video game to prevent and treat obesity in children.²⁰ Zamzee is another application for adolescents²¹ that records physical activity via a device. The number of points earned is calculated by the computer and can be exchanged for goods from partnering companies.

ML is an emerging strategy^22,23 implemented by many computerized systems used in a clinical context.^24,25 An ML-based approach provides beneficial insight into data to extract relevant information and has been actively used for the prognosis and diagnosis of several chronic conditions, particularly in patients with diabetes and sepsis.^26,27

Despite being potentially very efficacious in obesity, there are surprisingly few reviews in the literature regarding the application of ML for childhood obesity management and treatment. Available reviews mainly focus on mobile devices, wearables, and interactive electronic media or predictive models based on statistics. One review²⁸ addressed the potential use of computer decision systems (CDS) showing the main features and results of studies that use ML techniques and CDS.

Social robots are also being developed and tested in various healthcare concepts. Social robots have significant potential to assist healthcare workers and deliver customized care to children because they are more capable than other types of interactive technologies to positively influence the behavior and motivation of the human interactants. This superior potential derives from the fact that social robots, unlike other assistive technologies, are embodied agents with anthropomorphic or zoomorphic features. In some cases, robots also have a limited ability to navigate or manipulate the physical environment, enabling them to engage in embodied interaction with humans and share a concrete focus of attention (object or activity) with them. Furthermore, the ability to express emotions and interest is much more developed in person-like agents, such as robots, compared with impersonal technologies, such as mobile phone or tablet applications.

Robinson et al²⁹ described the potential of robots and demonstrated that an autonomous social robot can efficaciously motivate adults to reduce the consumption of high-calorie snacks, aiding a low-powered therapy on dietary intake with no human intervention. Building on this and analogous examples of robots acting as coaches and lifestyle advisors, a “virtuous” theoretical approach to robot design has been proposed.^30,31 This “virtuous robotics” approach can be applied to obesity management where robots can be programmed and deployed to cater to specific health and psychological requirements, motivate children to perform physical activities, and encourage healthier eating habits. Robots are suited for this motivational/assistive role because they are perceived as toys by children and can be a source of companionship, provide comfort and relief, and gain the trust of children, which is indispensable to give children explicit instructions and help in reforming their dietary habits.

Several reviews have explored obesity management using different technologies in children and adolescents and compared their efficacy in managing and preventing obesity via various channels.^32,33 No reviews were specifically dedicated to analyzing the efficacy of robots and AI-based strategies compared with other technology-based techniques for obesity management in children and adolescents that have received more attention to date. The present systematic review aims to fill the gap by analyzing and comparing research studies published between the years 2010–2021 to map how ML techniques are used and how they compare with non-AI-based technological strategies in obesity management for children.

Materials and Methods

The present review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines using the following strategic approach. Studies included in this review were identified systematically by searching databases using the following keywords: obesity, obesity management, children, technology, exergaming, robots, machine learning, and overweight. The databases consulted were PubMed, IEEE Xplore, Scopus, and Google Scholar. Moreover, a few references of significant studies^34,35 were scrutinized to extract more relevant articles suited to our review for further investigation and were shortlisted according to our criteria.

Inclusion and Exclusion Criteria

Studies fulfilled the inclusion criteria if they were (1) published in English; (2) published within the years 2010–2021; (3) based on obesity prevention or management (4) included participants aged ≤18 years; (5) were randomized controlled trials (RCTs) or interventions; (6) Used AI, robots or ML techniques.

Studies were excluded if they were (1) not in English; (2) had no reliable outcomes; (3) had participants aged >18 years; (4) did not use any types of technology in their research; and (5) were reviews, book chapters, thesis or conference material.

Quality Assessment Strategy

A modified Delphi list was used to assess the quality of each study.³⁶ The assessment quality was improved by considering the following elements: age range, sample size, and intervention duration. The latter two were considered with the understanding that a smaller sample size and a shorter intervention duration might negatively impact generalizability and reproducibility.

Additional considerations were given to a study’s purpose, whether they were preventive, screening-based, experimental or treatment-based. Also, the methods of the study were taken into consideration if they were activity/video game based, web/mobile based or ML based. The choice of elements included in the quality assessment was guided by a previously published systematic review conducted by Nouchi et al.³⁴

Results

A total of 1244 articles obtained from various databases were screened and duplicate articles were rejected. The titles and abstracts of the remaining 250 articles were screened for relevancy, and 75 articles were identified for full-text reading, 26 were finalized for review. A review of each study was performed, and information related to the layout and purpose of the study, as well as details of the participants (age, gender, size, and characteristics) were extracted.

The selection strategy for our review is illustrated (Figure 1). Table 1 shows an overview of the study characteristics, Table 2 outlines the key findings of the included literature, and Table 3 shows the assessment of the quality of studies using a delphi list. The studies were of various types, such as RCTs, experimental studies, or screening studies. The overall sample size analyzed ranged from 17 to 18,818, and the age of the participants ranged from 9 months to 18 years. Most studies were not gender specific and included both males and females, except for one study by Staiano et al that only included females.⁴²

Table 1 Characteristics of the Included Literature

Table 2 Key Findings of the Studies

Table 3 Quality Assessment of Included Studies

Figure 1 Search strategy used for selection of studies.

All studies focused on technology-based interventions. Most studies^{13,37–39,41–45,49–54} involved overweight or obese participants, a few^{13,35,36,46–48,52,53} included healthy participants, and others did not specify the characteristics of the participants.^40,55–59 The technology-based interventions were classified into three categories: (1) active video games or exergaming; (2) web, internet-based activities or mobile application-based activities; (3) ML-based strategies;

The duration of the intervention ranged from 1 week³⁵ to 3 years,⁴¹ and the studies that used ML-based algorithms did not always specify the time period for data collection. Intervention intensity was reported in most studies, but not all,^40,54–59 and was highly heterogeneous. Ten studies^{13,35,41,42,46,47,49,51–53} were RCTs that compared technology-driven intervention with a control group and used a non-technology-based intervention⁵⁶ in which the control group played video games that were sedentary compared with the intervention group. The outcome of the studies varied according to the aim, but all comprised a body mass outcome, such as body mass index (BMI), BMI z-score, or BMI percentile. Other outcomes commonly observed included other anthropometrical, diet-based, and physical activity measurements.

Active Video Game Interventions

Twelve studies^35,37–47 investigated the efficacy of active video games or exergaming. Lindberg et al³⁵ incorporated the gaming program Running Othello into the South Korean school physical education (PE) curriculum for the intervention group, while the control group used printed leaflets and the regular curriculum. Positive results were observed in the intervention group as children were more engaged and showed increased heart rates compared with the control group. In another study, Argarini et al³⁷ investigated the efficacy of moderate intensity exergaming on children with a BMI in the 85th percentile. Exergames were played using Xbox 360, and dance and sports-based games were played for four weeks. The results showed reduced BMI and improved fundamental movement skill, with no significant discrepancy between male and female participants. Adamo et al³⁸ used the interactive video game, “GameBike” and included stationary cycling for two groups: one group performed stationary cycling with music and the other group used GameBike while performing stationary cycling. Regularity of exercise, energy consumption, aerobic fitness, body composition, and the risk of heart disorder in overweight and obese children were monitored. The music group performed considerably better and used more strength than the GameBike group. There were no statistically significant differences between or within groups with respect to BMI.

Staiano et al³⁹ performed exergaming with Kinect and Xbox 360 gaming consoles using four simultaneous exergames, whereas the control group was required to maintain their regular level of physical activity. Statistically significant differences were observed in the intervention group, which showed improved scores, increased physical activity, and improved cardiometabolic health. Ruggiero et al⁴⁰ developed the exergame “MyPlatePick” that combines nutrition education and physical activity to promote movement and encourage healthy behavioral changes. Positive changes in eating behavior and increased physical activity were noticed.

Trost et al⁴¹ evaluated the effect of exergaming on children participating in a group-based program for managing weight. Improvements in physical activity were observed in the gaming group compared with the control group, which only took part in the program and not in the exergaming. The active video game group showed reduced weight and BMI. Another study by Staiano et al⁴² examined the effects of dance-based exergaming on females matched to a control group who followed their normal level of physical activity. No statistically significant differences were observed between the groups at follow-up.

Rio et al⁵⁰ tested a platform in which weekly group sessions with training were conducted on healthy-eating habits and active video games, and the participants developed their own vocal projects based on healthy activities. This study documented a positive effect on the intervention group in terms of nutritional knowledge and adherence to Mediterranean diet. Wagner et al⁴⁵ studied the efficacy of dance-based exergaming on obese adolescents’ observed capability to exercise, self-regulation, and BMI compared with a wait-list control group. A statistically significant increase in the perceived ability to exercise was found based on self-reporting by the participants in the intervention group compared with the control group. However, no major variations were found for BMI z-score.

Duman et al⁴⁴ used aerobics along with music and video games in their eight-week study and found a positive outcome in the final measurements with the exercise program using video games. Maddison et al⁴³ investigated the effect of video games on physical activity and fitness compared with standard video games. They found a decrease in BMI and BMI z-score of active group participants compared with the children in the control group.

Web-Based/Mobile-Based Studies

Eight studies^{13,36,48–53} used interventions based on internet or smart phone applications. Yang et al³⁶ developed a mobile-based HAPPYME platform that asks students to complete quests involving physical activity and healthy eating habits. The platform sends feedback to their parents who can then monitor the performance. Curiel et al⁴⁸ developed a video game-based mobile application that enabled participants to complete quests based on physical activity and healthy eating. Another study by Nawi et al⁴⁹ used the internet-based intervention ObeseGo! that provides information about healthy lifestyle and diet nutrition via the internet. This intervention showed no significant effect on BMI scores. Chen et al⁵⁶ developed a web-based program that comprised self-assessment activities to determine problem-solving skills and an interactive software program customized to Chinese food that allowed the user to check nutritional information about the food they ate. Fiechtner et al⁵⁴ developed a computerized clinician decision support system that monitored family behavior and found that living within one mile of a supermarket encouraged healthy eating habits.

ML-Based Studies

Five studies^55–59 used ML-based strategies for their interventions. Lingren et al⁵⁵ created both rule-based and ML-based algorithms using structured and unstructured data from two large electronic health records to detect severe early childhood obesity and children at a long-term risk of developing complications. The rule-based algorithms performed better overall. Rios-Julian et al⁵⁶ used ML to estimate the feasibility of developing an automated screening tool for diagnosing obesity using anthropometric variables. Fergus et al⁵⁷ used a wearable accelerometer to collect data that was used to evaluate physical tasks, and a multilayer perceptron neural network was used to classify physical activities according to type of activity. Dugan et al⁵⁸ used ML on the collected data to develop an algorithm that could predict childhood obesity in children aged >2 years using data collected before their second birthday. Six different ML techniques, namely, RandomTree, RandomForest, J48, ID3, Naïve Bayes, and Bayes, trained using the Child Health Improvement through Computer Automation system data showed that a precise, sensitive model could be formed. Balbir et al⁵⁹ developed ML algorithms to predict the risk of young people becoming obese in the future with a 90% accuracy.

Discussion

The objective of this systematic review was to assess the use of technology-based interventions in the management and prevention of childhood obesity and evaluate the use of ML and other technological strategies. Various technologies have been used, including video game-based interventions, mobile-based interventions, web-based interventions, phone communication-based interventions and machine learning-based interventions. It is important to note that along with these, multicomponent mediations, including both dietary intake and physical activity, are required to effectively overcome obesity.

Various parameters, such as BMI, body fat percentage, weight, heart rate, and BMI z-scores, have been used in these interventions to measure the progress and success of the interventions. Among the 26 studies under review, 21 studies showed positive significant outcomes on their respective outcome measures.

The present review classified the selected literature broadly into three categories: active video games or exergaming; internet or mobile-based interventions; and ML-based interventions. The two categories of web- or internet-based and mobile-based strategies were combined as one. The search did not yield any articles based on robot-based techniques to manage obesity in children; therefore, no such studies were included for that category, but one study⁶⁰ was found to be using a social robot to manage and spread awareness about obesity in children of Saudi Arabia; however, since it was not an RCT or an intervention with outcomes, it was not included in the review literature.

Because of the complex etiology of childhood and adolescence obesity and the diversity of its possible influencing factors (eg, biological, psychological, dietary, behavioral, and cultural), conventional statistical methods, relying on linear models, have limitations as the analyses they offer typically have a limited predictive power. Most cases use the same prediction values transformed slightly, and it is difficult to find models capable of handling multidimensional data. In contrast, ML-based strategies are valuable tools for multidimensional datasets that involve complex relationships among multidomain variables. ML techniques can search for complicated nonlinear relations between prediction variables and response variables in an automated manner and do not require manual predefinition.⁶¹ Thus, the predictivity, ease of use, and robustness of ML-based models are superior to those of conventional models.³⁶ Five studies incorporated ML strategies in their interventions. Most of these studies scanned existing electronic health records to prevent or detect obesity. Lingren et al⁵⁵ detected severe cases of childhood obesity, and Rios-Julian et al⁵⁶ determined the possibility of developing automated screening tools to detect obesity based on anthropometric variables using ML. Dugan et al⁵⁸ and Singh et al⁵⁹ developed algorithms to predict obesity in children and youngsters.

If a comparison is to be made between ML-based techniques and the other technological methods, it is difficult at this point in the research as ML studies are more focused on screening the databases to identify patterns, which can help in prediction and other strategies involving a different approach by various methods of training, games and information communication. It is not possible to compare two entirely different methods with different objectives, even though the main objective stands in managing and prevention of obesity, the method used is vastly different for them. Since the search did not reveal any AI and robot-based interventions, it can be concluded that this area has a lot of potential for future and needs to be actively researched.

On observing closely most of the studies, it can be broadly recognized as preventive, educational, activity-based, data screening, and treatment-based studies. It was perceived that the preventive and educational interventions were focused more on making the participants aware of the different kinds of food and their impacts on the body. They used various gaming methods to educate children, or internet/telephone-based methods to make children aware of the evil effects of unhealthy foods, high sugar diets, etc. They also highlighted the importance of physical activities and their rewards.

Michie⁶² outlines effective techniques in healthy eating and physical activity interventions, stating what makes an intervention effective in successfully changing the lifestyle or behavior of the participants for the better. These were the basic techniques according to S. Michie to bring about change in behavior: formation of an intention, setting up a specific goal, providing performance feedback, behavior self-monitoring and behavior goal reviews. Most of our studies did follow this pattern to some extent, where they had a targeted goal, set up intervention, tracked performance and feedback was taken even after post-intervention to assess self-monitoring efficacy.

Activity-based intervention were focused on exergaming, which kept children engaged and motivated to continue the program. The games came with levels and rewards, which kept the child interested and as a result, continued to play games. These activity-based methods were used for both managing obesity and in prevention strategies.

Even though varied in nature, the studies included mostly had very similar primary outcome measures like BMI score, body fat, BMI-z score, physical activity duration, weight measures, etc. Most studies focused on both the genders and did not do any gender-specific study except one study, which was a female-based study.⁴²

The studies included in this review were of average quality, as only 5 of the 26 studies used ML or advanced techniques to detect or prevent obesity and did not specify the intervention duration. Furthermore, five studies did not define the characteristics of the participants. None of the studies used simple robots (without AI) or AI-based robots for the management of childhood obesity.

Overall, it can be seen that these different approaches to cater well to the childhood obesity issue have been effectively using technologies to keep children engaged and motivated when carrying out the required activities. Overall, it can be seen that these different approaches to cater well to the childhood obesity issue have been effective using creative methods to keep children engaged and motivated to carry out the required activities, using standard measures as primary outcomes, it gives a standardized way of evaluating the efficacy of each of these studies. Most studies showed positive results in both preventive and management-based approaches.

However, children require a different type of care than elderly people, and robots may be a good platform to fulfill these unique needs.^63,64 Diseases, such as cancer and diabetes, can disrupt a child’s normal life, affect their social needs, and pose various challenges during the treatment of illness and change in lifestyle, which in turn also affect the mental health of the child.^65,66 Social robots may prove to be helpful in the management of chronic illnesses because they can provide effective support through encouragement, education to adhere to healthy behaviors, and entertainment useful to mitigate stress thereby providing a sense of comfort and companionship.⁶⁷ As such, robots have been shown to be helpful for treating various conditions,⁶⁸ and it is reasonable to expect that they can have similar effects on the treatment of obesity, helping children emotionally and motivating them and subsequently leading them to a healthier lifestyle. While ML helps screening datasets to identify children at risk and make predictions about their condition, AI and robotics could be used to develop user-centric strategies based on the real-time customization of the treatment program.

Conclusion

Although almost all the studies analyzed were statistically sound and showed positive outcomes, our review shows that this area still lacks the use of robots and AI-based technologies.

A more efficacious intervention should be developed in the future by combining a more diverse set of techniques. Studies should focus on the positive aspects of each category and develop an integrated system where there is a physical entity, such as a robot with AI and ML capabilities, that can conduct a physical activity program for the participants and educate them regarding good eating habits. These robots can be implemented with adaptive and real-time feedback techniques, making it easier to monitor the progress in real-time from home. Most of the studies included in the present review showed positive statistical outcomes. Exergaming is the most popular strategy among the literature published in the past 10 years. Thus, our review indicates that technology-based interventions have great potential in terms of efficacy in monitoring and managing obesity in children and adolescents, but its potential has not been reached. This area of research still lacks AI and robot-based interventions, which could be combined with exergaming and ML techniques to maximize their effectiveness and establish a strategic synergy to promote children’s health. As for the future direction, we are currently at the stage of introducing an adaptable intelligent robot that can personalize the extent of interaction based on the child’s personal interest and level of obesity. Such personalization could enhance the benefits of using robots in the management of childhood obesity.