Using explainable machine learning and fitbit data to investigate predictors of adolescent obesity

Area of Science:

• Adolescent health
• Obesity research
• Wearable technology applications

Background:

• Early adolescence is a critical period for obesity development.
• Sociodemographic and lifestyle factors influence obesity risk.
• Predictive modeling can identify at-risk individuals.

Purpose of the Study:

• To identify key predictors of obesity in early adolescence.
• To analyze the role of sleep, physical activity, and sociodemographics.
• To assess the utility of wearable devices in obesity risk assessment.

Main Methods:

• Analysis of data from 2971 adolescents (mean age 11.94 years) from the ABCD Study.
• Utilized Fitbit Charge HR 2 devices for objective sleep and activity monitoring.
• Employed glass box machine learning models to identify obesity predictors.

Main Results:

• Key obesity predictors included non-White race, low income, later bedtime, short/variable sleep, low daily steps, and high heart rates.
• Machine learning model achieved an AUC of 0.726.
• Wearable data provided insights into sleep, cardiovascular fitness, and activity levels.

Conclusions:

• Inadequate sleep, physical inactivity, and socioeconomic disparities are significant contributors to adolescent obesity risk.
• Wearable technology offers a viable tool for continuous monitoring of adolescent health metrics.
• Understanding predictor tipping points can guide effective obesity interventions.