JMIR Publications

ABSTRACT

Background:

The presence and availability of food has been shown to influence eating. Knowing the presence of food in the environment may enable mHealth apps to use geo-fencing techniques to determine the most appropriate time to proactively deliver interventions. To date, however, studies on eating typically rely on self-reports of environmental context, which may not be accurate or feasible for issuing mHealth interventions.

Objective:

Here, we compare subjective and Geographic Information System (GIS) assessments of the momentary food environment to explore the feasibility of using GIS data to predict eating behaviour and inform geo-fenced interventions.

Methods:

72 participants recorded their food intake in real-time over 14 days using Ecological Momentary Assessment methods. Participants logged their food intake and responded to ~5 randomly-timed assessments each day. During each assessment, participants reported the number and type of food outlets nearby. Their electronic diaries simultaneously recorded their GPS coordinates. GPS data was later overlayed with a GIS map of food outlets to produce an objective count of the number of food outlets within 50m of the participant.

Results:

Correlations between self-reported and GIS counts of food outlets within 50m were only of small size (r= .17, p<.001). Logistic regression analyses revealed that the GIS count significantly predicted eating similar to the self-reported counts (AUC-ROC self-report= 0.53, SE= 0.00 vs. AUC-ROC 50m GIS= 0.53, SE= 0.00, p=.41). However, there was a significant difference between both GIS-derived and self-reported counts of food outlets and the self-reported type of food outlets (AUC-ROC self-reported outlet type= 0.56, SE= 0.01, p <.001).

Conclusions:

Subjective food environment appears to predict eating better than objectively measured food environment via GIS. mHealth apps may need to consider the type of food outlets, rather than the raw number of outlets in an individual’s environment.