Momentary dietary lapse prediction for obesity management: Developing the Eating Behaviour Lapse Inventory Survey...

As the obesity prevalence continues to rise, effective interventions that promote dietary adherence and address the intricate array of factors contributing to dietary lapses are warranted.

This study introduces the Eating Behaviour Lapse Inventory Survey Singapore (eBLISS), a novel Ecological Momentary Assessment (EMA) tool, of which findings were used to build a machine learning algorithm aimed at predicting self-reported dietary overconsumption. eBLISS was meticulously crafted through an exploratory mixed-methods approach, deriving dietary lapse triggers from a rich synthesis of literature and in-depth thematic analysis of interviews with individuals with overweight/obesity. The tool's content validity was rigorously affirmed through both qualitative and quantitative means, with iterative refinement by a multi-disciplinary expert panel via the Delphi method. Data harvested from eBLISS were used to develop the eating Trigger Response Inhibition Program (eTRIP) an AI-powered smartphone application designed for weight management. The app enhances user engagement in tracking eating behaviours and identifying lapse triggers. Through a thorough process involving data cleaning, normalisation, and feature selection via Recursive Feature Elimination with cross-validation, the models' predictive powers were evaluated.

The content validity of eBLISS was deemed satisfactory with an item-content validity index (I-CVI) and scale CVI (S-CVI) of >0.79 ≥ 0.80 respectively. Among various machine learning techniques, gradient boosting was chosen as the best technique used to develop a dietary lapse prediction model (sensitivity=0.72, specificity=0.85, accuracy=0.79, F1 score=0.76, AUC score=0.86).

An eating behaviour lapse survey and a dietary prediction model was developed and tested to be valid for use in obesity management.