Standardized Methods for Evaluating Physical and Eating Behaviors: The WEALTH Cross-Sectional Study Protocol.

The WEALTH study enrolled 627 participants across 5 European research centers in the Czech Republic, France, Germany, and Ireland.

• 44% (n=275) of participants were male
• Mean age was 32.7 (SD 13.3) years
• Mean body mass index was 24.5 (SD 4.0) kg/m²
• Data collection took place from spring 2023 to spring 2024

The study employed a multi-device wearable sensor protocol combining two research-grade and two consumer-grade devices per participant.

• Devices were worn at both wrist and hip locations to capture accelerometer data
• Participants first completed a standardized semistructured lab-based activity protocol designed to simulate common physical and eating behaviors typical of a daily routine
• Following the lab session, participants underwent a 9-day free-living data collection period
• The lab protocol was specifically designed to collect labeled data for machine learning model development

The study used a multi-modal ecological momentary assessment (EMA) approach combining time-based, event-based, and self-initiated surveys during the free-living period.

• EMA surveys were conducted over the 9-day free-living data collection period
• EMA methods were designed to evaluate interactions between physical behaviors and eating behaviors
• EMA surveys were complemented by three 24-hour dietary recalls using validated web-based programs
• Feasibility of the procedures was assessed via a questionnaire completed upon survey protocol completion

Participants completed an in-person lab visit that included anthropometric measurements, handgrip strength assessment, and an online questionnaire.

• Measures collected at the lab visit included anthropometry and handgrip strength
• Participants were fitted with wearable devices during the lab visit
• The lab-based activity protocol was semistructured and designed to replicate common daily physical and eating behaviors
• The protocol was designed to generate labeled data suitable for training machine learning classifiers

The WEALTH project is designed to produce a publicly available toolbox including labeled datasets and machine learning models for physical and eating behavior classification.

• Outputs will include a repository of publicly available labeled data
• Machine learning models for behavior classification from accelerometer data will be made available
• A methodology for simultaneously capturing eating behaviors and physical behaviors will be included
• Data processing and machine learning model development were underway at time of publication, with primary results expected in 2026

The accurate measurement of physical behaviors and eating behaviors is identified as critical for designing, monitoring, and implementing public health guidelines and intervention strategies.

• Existing methods lacked standardization for identifying daily physical and eating behaviors from wearable sensors
• The WEALTH project specifically aimed to evaluate the interaction and contexts of physical and eating behaviors
• Both research-grade and consumer-grade sensors were included to broaden applicability of the resulting methods
• The study protocol was designed to address gaps in simultaneously capturing physical behavior and eating behavior data