Accuracy of Nutrient Calculations Using the Consumer-Focused Online App MyFitnessPal: Validation Study

Abstract

Background Digital food registration via online platforms that are coupled to large food databases obviates the need for manual processing of dietary data. The reliability of such platforms depends on the quality of the associated food database. Objective In this study, we validate the database of MyFitnessPal versus the Belgian food composition database, Nubel. Methods After carefully given instructions, 50 participants used MyFitnessPal to each complete a 4-day dietary record 2 times (T1 and T2), with 1 month in between T1 and T2. Nutrient intake values were calculated either manually, using the food composition database Nubel, or automatically, using the database coupled to MyFitnessPal. First, nutrient values from T1 were used as a training set to develop an algorithm that defined upper limit values for energy intake, carbohydrates, fat, protein, fiber, sugar, cholesterol, and sodium. These limits were applied to the MyFitnessPal dataset extracted at T2 to remove extremely high and likely erroneous values. Original and cleaned T2 values were correlated with the Nubel calculated values. Bias was estimated using Bland-Altman plots. Finally, we simulated the impact of using MyFitnessPal for nutrient analysis instead of Nubel on the power of a study design that correlates nutrient intake to a chosen outcome variable. Results Per food portion, the following upper limits were defined: 1500 kilocalories for total energy intake, 95 grams (g) for carbohydrates, 92 g for fat, 52 g for protein, 22 g for fiber, 70 g for sugar, 600 mg for cholesterol, and 3600 mg for sodium. Cleaning the dataset extracted at T2 resulted in a 2.8% rejection. Cleaned MyFitnessPal values demonstrated strong correlations with Nubel for energy intake (r=0.96), carbohydrates (r=0.90), fat (r=0.90), protein (r=0.90), fiber (r=0.80), and sugar (r=0.79), but weak correlations for cholesterol (ρ=0.51) and sodium (ρ=0.53); all P values were ≤.001. No bias was found between both methods, except for a fixed bias for fiber and a proportional bias for cholesterol. A 5-10% power loss should be taken into account when correlating energy intake and macronutrients obtained with MyFitnessPal to an outcome variable, compared to Nubel. Conclusions Dietary analysis with MyFitnessPal is accurate and efficient for total energy intake, macronutrients, sugar, and fiber, but not for cholesterol and sodium.