0748 Using daily wearable cardio fitness and sleep data to predict obesity in early adolescence
Orsolya Kiss,Natalia Smith,Erin Dooley,Massimiliano de Zambotti,Fiona C Baker,Kelley Pettee Gabriel,Jason Nagata
DOI: https://doi.org/10.1093/sleep/zsad077.0748
IF: 6.313
2023-05-01
SLEEP
Abstract:Abstract Introduction Obesity, and its co-morbidities, is a serious health risk currently facing adolescents in the US, with increased risk for adult obesity and cardiovascular health problems. Sleep is crucial for energy restorative processes and insufficient sleep influences food intake and appetite regulation. Adolescents are at risk for poor/insufficient sleep, which could, therefore, contribute to obesity risk. Given multifactorial risk factors, there is a need to identify the most salient predictors of adolescent obesity. Predicting adolescent obesity using objective Fitbit data has not previously been investigated and is the aim of the current study. Methods We used data collected with the Fitbit Charge 2 device from 2491 participants (Year 2, Mean-age=11.95 years, SD=0.65, 49.5% female) in the Adolescent Brain Cognitive Development (ABCD) Study®. We trained machine learning models (Gradient Boosted Trees and Regularized Logistic Regression) to identify the most important measures that predict obesity (>95 BMI percentile). In addition to sleep (timing, efficiency, duration, and regularity across the week) and cardio fitness (step counts, resting heart rate) Fitbit measures, averaged across seven days, we included socio-demographic characteristics. We ran a grid search embedded in a 2x5 fold cross-validation and we evaluated the performance based on the area under the curve-AUC and accuracy-ACC metrics. Results 15% percent of the sample were classified as having obesity. Both classification models performed acceptably high, with the LogReg having better performance (AUC-LogReg=0.76, ACC=85%) than the GBT model (AUC-GBT=0.75, ACC=0.69). Shorter sleep duration, more day-to-day variability in sleep measures, low number of steps with low variability over the week, and higher resting heart rate, with high day-to-day variability, were important predictors. Sociodemographic factors (Hispanic/Latinx ethnicity, African American race, low access to food) were also important predictors. Conclusion Results show that there are socio-economic disparities in obesity risk and that shorter sleep duration and more variable sleep patterns, as detected with a wearable, are important predictors of obesity risk. Results highlight the clinical utility of wearable measures and the importance of continuous monitoring of sleep, physical activity, and cardio fitness in adolescents. Support (if any) NIH U01DA041022
neurosciences,clinical neurology