Oral Presentation 49th Nutrition Society of Australia Annual Scientific Meeting 2025

Are Diet Apps Accurate Enough for Type 1 Diabetes? Assessing the Accuracy of Commercial Nutrition Apps for Carbohydrate Estimation (129965)

Naufal Nurdin 1 2 , Chiara Murgia 1 , Robyn Larsen 1
  1. School of Agriculture, Food and Ecosystem Sciences (SAFES), Faculty of Science, University of Melbourne, Parkville, VICTORIA, Australia
  2. Department of Community Nutrition, Faculty of Human Ecology, IPB University, Bogor, West Java, Indonesia

Carbohydrate counting is crucial in the management of type 1 diabetes for determining meal time insulin and maintaining postprandial euglycemia. The use of automated dietary tracking apps may aid carbohydrate counting in Type 1 Diabetes (T1D) management. However, the accuracy of these apps in detecting carbohydrate-containing foods might depend on their presentation. Segmenting or separating carbohydrate-rich foods on a plate could be a strategy to improve carbohydrate estimation accuracy with these tools. This study aimed to compare the performance of commercially available dietary apps in assessing the carbohydrate content of reference meals served as segmented vs. mixed meals.Under standard laboratory conditions, we analysed 30 meals of known composition using four commercially available apps: LoseIT, SnapCalorie, MyFitnessPal, and MacroFactor. Meals were presented with the carbohydrate dense food separated from other plate elements (segmented) (i.e bolognaise sauce and pasta) and again with all elements combined (mixed). Mann-Whitney U tests were used to compare mean absolute errors (MAEs) for mixed vs. segmented meals across the different apps. Bland-Altman plots and 95% limits of agreement (95% LoA) were used to assess systematic bias and clinical acceptability. Differences of <20 g from known carbohydrate amounts was used as the clinically acceptable limit1. We found no significant difference in the MAEs of carbohydrate amounts assessed in segmented vs mixed meals (median: 27.8 g vs. 29.8 g; P = 0.192). This finding did not vary according to the type of diet tracking app used. All apps showed a bias towards underestimating carbohydrate amounts, with mean bias (difference from known amounts) ranging from -5.6 g to -20.9 g. MacroFactor had the lowest mean bias (-5.6 g) and was the only diet tracking app with 95% LoAs within acceptable clinical limits for carbohydrate determination (-16.9 g, 5.7 g). The LoAs exceeded clinically acceptable thresholds for LoseIT (95% LoA: -29.0 g, -7.4 g), MyFitnessPal (95% LoA: -31.2 g, -10.6 g), and SnapCalorie (95% LoA: -24.9 g, 3.6 g). In conclusion segregating carbohydrate foods on a plate did not appear to improve carbohydrate estimation when using photo-assisted diet tracking apps. Current commercially available tools tended to underestimate carbohydrate amounts, and their accuracy varied between apps, suggesting that not all apps may be suitable for T1D management.