(OR21-05-24) Testing the Validity of the PortionSize™ Smartphone App for Estimating Children’s Dietary Intake in Free-Living Conditions

Objectives: The PortionSize™ (PS) smartphone app provides users with real-time feedback on food selection and intake, which may assist in modifying children’s dietary intake to meet recommendations. This free-living study tested the validity of dietary intake estimates from the PS app compared to intake determined by trained raters who used the validated Digital Photography of Foods Method in a pediatric population.

Methods: During two consecutive free-living days, parents (n=28) recorded their children's (aged 7-12 y) food intake using the PS app for at least one meal and one snack per day. Parents measured portion size of foods and drinks by using embedded portion templates in the PS app. The app then automatically estimated dietary intake using the templates and a built-in nutrient database (i.e., a subset of foods from the FNDDS). The app also transmitted participants’ food images to trained raters who quantified food intake using Digital Photography (DP), which served as the criterion measure. Equivalence was evaluated with two one-sided t-tests (± 25% bounds), and Bland-Altman analysis determined if error variance differed over levels of intake. Mean percent error was calculated at group level.

Results: Most children were boys (57%), 25% had overweight or obesity, mean age was 9.6 ± 1.8 y and mean BMI percentile was 56.2 ± 31.1. Parental estimates of children's food intake using the PS app (mean ± SD, 529± 304 g/day) were not statistically equivalent to the estimated intake by raters using DP (488 ± 230 g/day; p=0.06). Similarly, PS and DP estimates of energy intake (872 ± 410 vs. 748 ± 275 kcal/day) were not equivalent (p=0.24). While dietary estimates were not statistically equivalent, the mean percent error was marginal, with PS overestimating food intake by 9% and energy by 17%. Based on Bland–Altman analysis, measurement error of PS in estimating food intake was consistent (R²=0.10, p=0.09); however, error in estimating energy intake had a slightly positive trend over different levels of intake (R²=0.15, p=0.05).

Conclusions: In free-living conditions, the PS app did not produce equivalent dietary estimates compared to DP. The magnitude of error, however, was modest for an app that automatically estimates food selection and intake, which shows promise for real-time estimation of children’s dietary intake.

Funding Sources: Blue Cross and Blue Shield of Louisiana Foundation