Comparing ecological momentary assessment to sensor-based approaches in predicting dietary lapse

Area of Science:

• Behavioral science
• Digital health
• Weight management

Background:

• Ecological momentary assessment (EMA) shows promise in predicting dietary lapse for weight loss interventions.
• Passive sensors offer objective data collection but have limitations in measuring lapse predictors.
• Comparing the accuracy and burden of EMA versus sensors is crucial for optimizing digital health tools.

Purpose of the Study:

• To preliminarily compare the participant burden and accuracy of commercially available sensors versus EMA in predicting dietary lapse.
• To evaluate the utility of sensor-derived data (physical activity, sleep, geolocation) for lapse prediction.

Main Methods:

• Twenty-three adults with overweight/obesity used a weight loss app, wore a Fitbit, and enabled GPS tracking for 6 weeks.
• Participants completed EMA surveys, and sensor data (233 features) and EMA data (19 risk factors) were collected.
• Two LASSO classification models were developed to predict dietary lapse using sensor data and EMA data.

Main Results:

• The EMA model demonstrated higher accuracy (59% sensitivity, 72% specificity) than the sensor model (63% sensitivity, 60% specificity).
• Participant-reported burden was significantly higher for EMA (M=2.96) compared to sensors (M=1.50).
• Despite lower accuracy, sensors offered a less burdensome data collection method.

Conclusions:

• While EMA models currently offer superior accuracy for dietary lapse prediction, passive sensors present a promising, less burdensome alternative.
• Future digital health interventions should consider the trade-off between prediction accuracy and participant burden when selecting monitoring methods.
• Combining EMA and sensor data may offer a balanced approach for comprehensive lapse prediction in weight management.