Validation of computational models to characterize cumulative intake curves from video-coded meals.

Introduction: Observational coding of eating behaviors (e.g., bites, eating rate) captures behavioral characteristics but is limited in its ability to capture dynamic patterns (e.g., temporal changes) across a meal. While the Universal Eating Monitor captures dynamic patterns of eating through cumulative intake curves, it is not commonly used in children due to strict behavioral protocols. Therefore, the objective of this study was to test the ability of computational models to characterize cumulative intake curves from video-coded meals without the use of continuous meal weight measurement. Methods: Cumulative intake curves were estimated using Kisslieff's Quadratic model and Thomas's logistic ordinary differential equation (LODE) model. To test if cumulative intake curves could be characterized from video-coded meals, three different types of data were simulated: (1) Constant Bite: simplified cumulative intake data; (2) Variable Bite: continuously measured meal weight data; and (3) Bite Measurement Error: video-coded meals that require the use of average bite size rather than measured bite size. Results: Performance did not differ by condition, which was assessed by examining model parameter recovery, goodness of fit, and prediction error. Therefore, the additional error incurred by using average bite size as one would with video-coded meals did not impact the ability to accurately estimate cumulative intake curves. While the Quadratic and LODE models were comparable in their ability to characterize cumulative intake curves, the LODE model parameters were more distinct than the Quadradic model. Greater distinctness suggests the LODE model may be more sensitive to individual differences in cumulative intake curves. Discussion: Characterizing cumulative intake curves from video-coded meals expands our ability to capture dynamic patterns of eating behaviors in populations that are less amenable to strict protocols such as children and individuals with disordered eating. This will improve our ability to identify patterns of eating behavior associated with overconsumption and provide new opportunities for treatment.