Objectives

Traditionally, Quantified Self technologies have been used by empowered health consumers, which might not be the more representative type of individuals in need of behavioral change support. In our study we wanted to explore the feasibility of capturing health-related information using quantified self-technologies by overweight children and their families.

Our study is based on the Technology Acceptance Model [5], but instead of focusing on ”Perceived Usefulness”,”Perceived Easy of Use”,”Attitude Towards Using” or ”Behavioral Intention to Use” we focused on ”Actual System Usage” with regards to the 360QS technologies that were deployed during the obesity trial. Combining the actual 360QS system usage with the demographic data of the users would allow us to better understand how 360QS could be used for this type of population. In order to evaluate ”Actual System Usage”, we can rely on the amount of data captured by individuals as an indicator of system usage.

Childhood Obesity

Obesity and the lack of physical activity are the major factors leading to the rise of non-communicable diseases such as diabetes. Childhood obesity is a global epidemic that affects both developing and developed countries [1, 2]. Obesity among children is especially worrying, since the impact on healthcare costs can be a major burden in the future. Furthermore, obesity has been found to increase the risk of a host of health conditions (e.g., sleep apnea, diabetes, cancer, or stroke) [2, 6–8] and also reduce mental well-being (e.g., leading to low self-esteem) [9].

Obesity is a highly complex condition where genetics, environment, education, socio-economic status, climate, schools, and many other factors play a role. Each of these factors can affect individuals differently. Ideally, any health intervention to tackle childhood obesity should take into account all aspects to maximize its chance of success. However, it is not always possible to obtain information about the environment of the affected youth. Furthermore, asking the patient is not always a feasible solution. For example, there is evidence that self-reported questionnaires for behaviors such as sleeping habits can be less reliable than data obtained from wearable devices [10].

For these reasons, designing interventions aimed at weight loss and physical activity promotion is a challenging task [11]. Many external factors (e.g., public transportation, diet, culture, family, schools) can positively or negatively affect the efficacy of public health interventions. Environmental contexts heavily affect decision making; for example social networks have been found to influence obesity [12] both positively and negatively. Another important factor to consider is the temporal aspect of when health decisions are made.

Demographics

Overweight and obese children from several male and female Qatari schools, in the age range of 9 to 12, participated in the interventions. The schools were randomized between intervention groups and control groups. There were 53 boys and 55 girls in the intervention group while the control group consisted of 60 boys and 59 girls.

Health intervention

The health intervention consisted of a set of health coaching and education actions as follows:

• Health camp: two-week weight management camp from January 24 to February 4, 2016, which included physical and social activities, lifestyle coaching and nutrition counseling.

• Weekend clubs: ten-week program from February 20, 2016, where every weekend for four hours, participants received further reinforcement to acquire healthy behaviors.

• Summer break: during the summer break (June - August 2016) a group of mothers participated in a mobile intervention using WhatsApp to reinforce healthy eating and behaviors.

The 360QS implementation

The 360QS implementation in the ICAN Study involved different technologies:

• Wearable devices: Wearable devices (actigraphy sensors) were used to monitor sleep and physical activity of the participants.

• Mobile Social Media (Photos and Messaging): We incorporated the use of social media for tracking dietary habits of participants (Instagram) and health education of mothers (WhatsApp).

• Health camp food Photos: We also captured photos of the food consumed by participants during the two-week health camp to gain insights on dietary preferences.

• Physiological Data: Physical health of the participants was measured periodically via height, weight, etc. to obtain an overall picture of the impact of the intervention on participants’ health.

The wearable devices and mobile social media are clear examples of QS technologies that can be used to capture data about individuals’ behaviors. The health camp food photos and physiological data was not captured by the individuals, but the research team. The food photos can be combined with the other data sources to gain more insights on the health behaviors of the individuals.

Figure 1 shows the data collected over time for the 2016 trial using each 360QS technology. It also indicates the time periods with data overlaps. For example, between January 31, 2016 and February 4, 2016, for about 10 participants, we have food photos collected during the camp, physiological data, as well as actigraphy sensor data. As explained below, we need to be aware that these technologies were not applied homogeneously in all the subjects. Not all the mothers accepted to participate in the WhatsApp intervention, and only a subsets of subjects received mobile phones for taking food photos using Instagram.

Open in a separate window

Fig. 1

Datasets overview for the 360QS implementation - Volume of collected data from different sources per day. The volume is given in terms of number of participants from which data have been collected for a given day. Each row shows the volume from unique data sources (top five rows) and from combinations of two or three data sources (bottom rows). The combinations of sources are useful to show those data sources for which data is available for the same participants so the complementarity between various data sources can be analyzed in these cases

All the data collected during the trial was integrated using anonymized participant identifiers and stored in a secure database. This integration facilitated the study and analysis of the data.

Physiological data

As part of the clinical research protocol, all the participants’ physiological parameters were measured over time. That includes weight, height, body mass index (BMI), body fat percentage, waist circumference and blood pressure. These parameters were taken during recruitment, start and end of the health camp, during the weekend clubs and also at the end of the program.

Health camp: food photos

During the two-week health camp, before and after photos of food trays (breakfast and lunch) were manually taken for all participants for ten days, i.e. twenty meals, as shown in Fig. 2. The goal of this data collection was to learn food preferences of participants, which could in turn assist the nutrition professionals in designing diet plans for weight loss, and also give calories estimates of the food consumed during the camp.

Open in a separate window

Fig. 2

Example of food tray before and after meal

Weekend clubs: actigraphy sensor

In order to study the physical activity and sleep patterns, we used actigraphy. Actigraphy sensors were given at the end of the health camp to be used during the weekend clubs. An actigraphy device is a wearable medical device, which captures motion using accelerometers and other sensors to identify patterns in physical activity and sleep. We used the device Axivity AX3 (see product website: http://axivity.com/product/ax3) that has been used before in clinical studies [25]. The Axivity device is waterproof and was given to participants to be worn at all times as a wristband. The device was given to 67 participants; 16 in the control group (8 boys and 8 girls) and 51 in the intervention group (26 boys and 25 girls). Since the battery life of the device as well as its data storage memory is about four weeks, we recharged the devices and downloaded the data for participants in the intervention group multiple times during the weekend clubs.

Weekend clubs: social media - instagram

A total of 50 mobile phones with a pre-configured Instagram account were given to parents of participants (30 boys and 20 girls) at the end of the health camp, to take photos of the food eaten by their children in the remaining duration of the program, and upload to the Instagram account. This was done in order to monitor impact of the health camp on the participants’ dietary habits and also capture dietary preferences of the families. Figure 3 shows one of the participants’ Instagram photo.

Open in a separate window

Fig. 3

Instagram Photo uploaded by a participant

Weekend clubs: actigraphy sensor

The actigraphy sensor was given to 51 participants. Normally these sensors are given for periods of one week to do an assessment of the levels of physical activity and sleep patterns. In our study we opted to ask the participants to wear it for a longer period (a total of 12 weeks) in order to be able to monitor changes in physical activity and sleep during the health interventions. As expected the adherence declined overtime, and 21 devices got lost.

Figure 7 shows the visual analytic tool we developed to explore the actigraphy sensor data. It shows the level of physical activity recorded by the actigraphy device through time for all children. It also shows the total activity for boys and girls separately and allows getting detailed activity averaged per day in a user-selected time window.

Open in a separate window

Fig. 7

User interface of the visual analytic tool for actigraphy sensor data The interface allows comparing two participants’ activity level (left side) for the full time period and on daily-based average during week and weekend days (right side). Specific clinical measurements like BMI or body fat can also be compared during the same time period (line charts)

Relating the activity levels to the change in the child’s BMI during the study period, we compute Pearson product–moment correlation coefficient r at different levels. The raw average number of minutes logged per day did not have a statistically significant relationship with the change in BMI. However, the number of active days, defined as days in which active minutes is non-zero, correlates with BMI change at r=−0.263 (p=0.101), meaning the more days are recorded as active, the more negative the BMI change is. The relationship becomes less strong as we define active days more strictly: more than 30 active minutes per day results in r=−0.201 (p=0.214), and more than 4 h r=−0.140 (p=0.377). This suggests that mere adherence to wearing of the device correlates with beneficial health outcomes. This however might be caused by increased engagement of wearable device users, not necessarily implying causality.

Weekend clubs: social media - instagram

Of the 50 mobile phones with a pre-configured Instagram account, only 25 were used for such purpose. The number of photos acquired was 937 in total, but the top 3 contributors (1 girl and 2 boys) uploaded 70 percent of the photos, as shown in Fig. 8.

Open in a separate window

Fig. 8

Instagram posts uploaded by users

Correlating the extent of use of this Instagram account to the change of child’s BMI results in r=−0.179 (p=0.425). To overcome the low p-value due to sparsity, we employ a permutation test to test whether the relationship is significant compared to a randomly permuted outcome variable (here discretized to below/above average BMI change). This approach results in a permuted p-value=0.026. Again this shows that more active users tended to slightly have better health outcomes but that this is most likely related with a higher level of engagement in their health.

Technical challenges

There are complex challenges in statistically analyzing temporal and heterogeneous data. Measuring causality from an intervention is not trivial, despite access to diverse heterogeneous data [26]. Moreover, integration of such datasets, is itself a complex challenge [27]. One of the main challenges is to aggregate highly heterogeneous data into a single platform.

We conducted meetings with the healthcare professionals involved in the trial in order to get their feedback about the utility of the different Quantified Self technologies tested in the study. They highlighted that to integrate such technologies in the practice it is very important to create tools to visualize trends and gain insights about the behaviors of the participants. In terms of visualization, it requires careful construction of the interface to focus on the user needs, providing only the information required for the current analytic task. Interaction is key for understanding, so graphical elements and codes must be linked throughout the interface in a coherent manner. To handle large-scale data, especially coming from actigraphy sensors, it is necessary to aggregate data by time periods like days or hours, to segment activity patterns using automatic techniques, and to give details only on demand as standard according to the visualization mantra “Overview first, zoom and filter, then details-on-demand.” [28].

Accessing the useful data for a given analysis is as important as analyzing the selected data themselves. So another challenge is to give a legible visual overview of the available data for all and each individual, in terms of the data type (photos, messages, sensor data, physiological data) and number of data records, throughout the entire time range. After providing an overview, the tool should allow for a selection of specific records given the type of analytic task requested by the user.

The collected data can be useful to healthcare providers to analyze children profiles and guide them to improve their health (for instance determining which food is preferred, or at which time a child is willing to be active and using this knowledge to orient the child’s behavior); but also for researchers to improve on the measurement protocols and techniques (for instance how to make children more diligent in wearing or using the devices? how to develop analysis techniques to extract more relevant information from the data?); and finally for the children to keep track of their progress in a specific attractive user interface (shall the children be put in competition with each other? or only aware of their own results? or be put in situation to help each other? what kind of information to show to children? and in which way to make it attractive for them?). Each of these needs require specific visual interfaces carefully designed in close collaboration with the targeted end-users and professional experts in children health, nutrition, pedagogy and psychology.

Comparison with prior work

To our knowledge this is the first reported exploration of the use of Quantified Self for personal health in an Arab country. Our study is highlighting many of the societal challenges that can be faced in introducing digital health interventions in the younger Arab population. There are previous studies in the region, but they have focused on a much narrower implementation of eHealth technologies [29] and specially focusing on diabetes and obesity in the adult population [30, 31]. Another novelty of our study is the combination of social media as an additional source of health quantified-self data. This is increasingly important since consumers are using social media for health purposes. There are other projects looking into the integration and visualization of Quantified Self health data [32, 33], but do not include the heterogeneous datasets as we have.

Apart from heterogeneity, our datasets have an added complexity of being sparse. Sparsity of the data poses significant challenges for visualization. This data sparsity can be attributed to human factors since we do not address visualization of data for people adherent to the ‘Quantified Self’ movement but rather individuals who are in need of a more active role in their health. This is a substantial difference if we consider that many Quantified Self users are highly motivated individuals [20] and that they do not represent the vast majority of people in need of support for behavioral change.

An additional challenge that we describe is how to integrate that information into meaningful visualizations that can help understand the health patterns of the individuals. This is not answered in our paper, but we provide some innovative examples of visualization of this type of data. There are other studies looking in the visualization of Electronic Health Records (EHR) for helping decision making [34], also including Quantified Self data [35–37] but in those studies the QS data is limited to wearable sensors data. Folter et al. studied the integration of health habit data (e.g., motion data for physical activity) to help decision making of healthcare professionals [36]. Ledesma et al. studied techniques for visualizing personal health data, but not including social media data. In our case we are looking into helping understand health behaviors (and not clinical aspects) in a more heterogeneous and sparse data set. Implementation Research methodologies [38–40] can provide a good framework for future work about the integration of 360QS in this type of health interventions.

We would like to acknowledge the contributions of all the members of the ICAN project. We greatly appreciate the work done by Dr. H Haddadi during the first steps of the 360° Quantified Self project.

The ICAN project was funded by the Qatar National Research Fund (a member of Qatar Foundation) under project number NPRP X- 036- 3–013 (Adapted Cognitive Behavioral Approach to Addressing Overweight and Obesity among Qatari Youth).