XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE Automation of Data Collection Techniques for Recording Food Intake: a Review of Publicly Available and Well-Adopted Diet Apps Klaus L. Fuchs D-MTEC ETH Zurich Zurich, Switzerland fuchsk@ethz.ch Mirella Haldimann ESS EAWAG Duebendorf, Switzerland mirella.haldimann@eawag.ch Alexander Ilic ITEM University of St. Gallen St. Gallen, Switzerland alexander.ilic@unisg.ch Denis Vuckovac D-MTEC ETH Zurich Zurich, Switzerland dvuckovac@ethz.ch Abstract — There has been a proliferation in the development of diet-related smartphone applications (mHealth) that support diet monitoring and can provide health-beneficial interventions. With these developments, the collection of accurate dietary consumption data is becoming an important field in mHealth, as less automated data collection techniques (DCT’s) are often associated with underreporting, ineffective or non-tailored interventions, and self-selection of motivated users and/or high attrition rates. Interventions that incorporate more automated or passive DCT’s have been linked with greater potential for user adoption and engagement, it remains unclear however what DCT’s exist and to what extent mHealth apps incorporate such techniques. As such, the purpose of this study is to investigate the presence of DCTs in well-adopted dietary apps and provide an overview of existing and emerging approaches. Keywords—uHealth, mHealth, data collection, nutrition, diet I. INTRODUCTION Lifestyle behaviours such as high calorific diets are mitigatable risk factors associated with many diet-related non-communicable diseases (NCDs). These account for 63 per cent of deaths worldwide [1], [2]. To date, many intervention programs targeting dietary changes have had only modest effects and their long-term effectiveness is not well established [3], [4]. Thus, public health researchers have begun to examine novel approaches to deliver behaviour change interventions. Mobile (mHealth) and ubiquitous health applications (uHealth) are a growing field in the prevention and management of NCDs and hold potential to deliver scalable, tailored health-related behaviour change interventions [5]. Mobile phone ownership has reached saturation in many industrial and developing countries with smartphone ownership rates of 90% and 70% respectively. As such diet-related mHealth hence promises inclusive and scalable means for health behavior change [6]. Despite the recent proliferation of apps to promote positive lifestyle change, there is a dearth of research evidence regarding their long-term effectiveness. While it has been acknowledged that novel technologies increase the likelihood of supporting behaviour change [7], it appears that current DCT approaches prove inadequate to fully complement mHealth interventions. For example, Mateo and colleagues [8] suggest that diet-related apps must become engaging in more relevant ways during usage and less effort- intensive. Further research has associated less automated DCTs with user attrition [9]–[12] as well as memory and recognition biases that reduce intervention effectivity of mHealth apps [12], [13]. In turn, further potentially discouraging the adoption of inexpensive diet-related health interventions. While studies have been conducted to determine the extent to which data collection techniques have been applied to app development (e.g. [14]), none have quantified the extent to which specific DCT’s are included. Moreover, we found that the few app reviews that exist focus on a selected subgroup of DCT’s. As such we lack a comprehensive over- view of DCT’s available today. Another limitation we see in this nascent field is that the aforementioned reviews - if not conceptual - have focused exclusively on apps that are not publicly available. However, without the use of qualitative or quantitative analysis of publicly available apps, on which many researchers, users and application developers rely on, we lack validation of DCT’s that are widely in use. To provide more insight regarding these gaps, we first developed an overview of existing DCT’s, discussed in the literature. In a next step this framework allowed us to con- duct a systematic determination of the presence/absence of DCT’s in a set of publicly available and well-adopted diet apps from the German speaking Google App Store, as the dominant mobile operating system in the country. We specifically selected Germany, as its ‘aging society’ is representative for the future of many developed countries, displaying high levels of NCD’s as well as high adoption of mobile devices and applications. We discuss the implications for theory and clinical practice, and identify research- practice gaps that hopefully stimulate the development of more sophisticated diet apps. II. DATA COLLECTION TECHNIQUES With the proliferation of improved technological means, the collection of accurate dietary consumption data is becoming an important field in mHealth as well as related fields such as nutritional epidemiology [15]. Accurate intake data allows to establish relationships between nutrition and health state. Traditionally implemented methods of food diaries usually involve the manual text entry of identifier and quantified amount of the consumed foods [16]. Such methods bear multiple issues most notably related to convenience and accuracy. The former involves a high burden that is placed upon respondents to record dietary information and to do so continuously, multiple times per day [10]. The latter involves memory biases leading to