Initial Validation of the PATCH: an Open-Source Device Designed to Measure Energy Expenditure among Children Aged 3-8 Years

ABSTRACT Background Wearable devices that measure energy expenditure are not designed for children. Therefore, we developed the PATCH (Platform for Accurate Tracking of Children’s Health), an open-source device to measure children’s energy expenditure using heartrate (HR) and acceleration. This study examines three models to estimate children’s oxygen consumption using HR and acceleration compared to a criterion of indirect calorimetry. Methods Fifty-two children aged 3-8 yrs (mean age 6.4 ± 1.7, 42% female, 73% White) completed a semi-structured protocol ranging in intensity from inactive (e.g., using iPad) to vigorous (e.g., running). The PATCH was attached to the chest, and measured HR (photoplethysmography) and acceleration (3-axis accelerometer, ±16 g). The criterion (Cosmed K5) measured breath by breath oxygen uptake (VO 2 ml/kg/min). We used cross-sectional time series (CSTS) models, generalized additive mixed models (GAMM) and random forest (RF) to predict oxygen consumption from a combination of HR, acceleration and participant characteristics (biological sex, age, weight, height). We used 10-fold cross-validation, testing each fold and training on the rest, repeated for robustness. Model fit was assessed using mean bias, mean absolute error (MAE), mean absolute percent error (MAPE) and variance explained (R 2 ). We reported out-of-sample R-squared values without subject-specific random effects to ensure broad applicability. Results Mean bias for CSTS, GAMM, and RF was -0.01, 0.01, and 0.08 ml/kg/min, respectively. MAE values were 1.54, 1.56, and 1.99 ml/kg/min; MAPE was 10% for CSTS and GAMM, and 13% for RF. The CSTS explained 86% (SD 5%) of variance, GAMM 86% (SD 5%), and RF 81% (SD 6%) in oxygen consumption. Conclusions CSTS, GAMM, and RF models provide similarly accurate estimates of children’s oxygen consumption using the PATCH device compared to indirect calorimetry. Further validation in larger, free-living samples is needed.