gms | German Medical Science

Objective: Automated recommender systems, which are also referred to as digital nudges, are technology-based systems that generate recommendations or guide users to relevant information. This study aimed to describe what is known about the recommender systems targeting obesity prevention using a scoping review of reviews (protocol: [1]).

Methods: Eligibility was defined according to the PCC (Population: any human; Concept: recommender systems or digital nudging; Context: obesity prevention) framework. Ten reviews with systematic methodology were included from 148 studies identified as reviews in database (MEDLINE, PsycINFO, Web of Science, CINHAL, Scopus, ACM Digital Library, and IEEE Xplore) or internet searches. Data on bibliographic characteristics, PCC details, and review findings (e.g., aim, health focus, recommender system description, and evidence gaps) were charted as qualitative statements from reviews and processed into predefined categories (e.g., health domain: ‘nutrition’ or ‘physical activity’) or categories that inductively emerged from the data (e.g., evidence gaps). The processed data were synthesised using relative frequencies or described narratively. An overlap that occurs when the same primary studies are included in multiple reviews was assessed as the overall corrected covered area (CCA: low overlap of 0-5% to very high overlap of 15%) using the Graphical Representation of Overlap for OVErviews (GROOVE) tool.

Results: The reviews (n=8 systematic and n=2 scoping) were published in 2017-2023 and included 308 primary studies. The overlap in primary studies among the 10 reviews was low (CCA=1.29%). Overall 282 primary studies (92%) were included in n=1 review, 20 (6%) in n=2 reviews, and six (2%) in n=3 reviews. The health domains addressed in the reviews were nutrition (n=6), physical activity (n=1), or both (n=3). The reviews described recommender systems for any population (n=7) or people with diabetes (n=1) or assessed the implementation of digital nudging (n=2) in adults (at workplace or online grocery shopping settings). The topics addressed in reviews were (1) system technical properties (e.g., system types, platforms, techniques and data to generate recommendations) in n=9 reviews, (2) system health domains (e.g., health focus and content of recommendations, target health behaviour) in n=7 reviews, and (3) evaluation, including system evaluation (e.g., system strengths and weaknesses) and implementation (i.e., user application) in n=5 reviews. The evidence gaps included the need for evaluation (e.g., of user outcomes or system validation) in n=8 reviews, development of new systems (e.g., based on frameworks, open-source codes, and considering diversity to personalise recommendations) in n=8 reviews, and more diverse health applications (e.g., focus on healthy lifestyle and disease prevention within obesity prevention) in n=4 reviews.

Conclusion: Recommender systems are relatively new concepts in obesity prevention. Various technical elements of such systems are already described, including system and data types required for the development of recommendations. However, a low overlap in primary studies among 10 reviews on this topic indicates that more transdisciplinary collaborations between technical and health experts are required to advance this field. System validation and user outcome evaluation are needed to identify optimal parameters for any long-term behaviour change in recommender system users.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.