Developing and Comparing a BMI-Inclusive Energy Expenditure Algorithm for Wrist-Worn Wearables
In today’s fitness-driven world, the quest for accurate energy expenditure (EE) measurement has become a cornerstone in health technology. Here, we delve into the intricate journey of creating and evaluating a groundbreaking Body Mass Index (BMI)-inclusive energy expenditure algorithm, targeting wrist-worn devices. With a focus on user engagement and real-world applicability, this article aims to illuminate the innovative research conducted in this exhilarating field.
Recruitment of Participants
Dual-Phase Recruitment Strategy
Participants were recruited in two distinct phases for our in-lab and free-living studies. Using flyers, Craigslist, and ResearchMatch services, we ensured a diverse cohort that reflects the challenges faced by individuals with a higher BMI.
Our inclusion criteria was stringent—participants had to be 18 years or older with a BMI of 30 kg/m² or greater. All procedures were meticulously approved by the Institutional Review Board at Northwestern University (STU00208545), in accordance with the Declaration of Helsinki. Informed consent was rigorously obtained, establishing a strong ethical foundation for our study.
Accurate Measurement of BMI
During the initial visit, research staff conducted essential measurements of each participant’s height and weight using a Detecto mechanical eye-level physician scale to confirm eligibility based on BMI.
Innovative Study Devices
Cutting-Edge Wearable Technology
To capture the most accurate wrist activity, participants donned an LG Nexus 5 smartphone paired with a Fossil Sport Gen 4 smartwatch. This setup enabled the collection of raw accelerometer and gyroscopic data at a sampling rate of 20 Hz.
Participants wore the smartwatch on their dominant wrist, ensuring optimal skin contact to precisely measure energy expenditure during daily activities like typing, eating, or jogging. This approach intentionally aimed to maximize data accuracy by focusing on the fine motor control of their dominant hand.
Comprehensive Data Collection Tools
The study also enlisted an ActiGraph GT3X+ device worn on the hip, and participants utilized a wearable camera to document real-life activities. The camera, equipped with an OmniVision RGB camera and Panasonic infrared thermal camera, provided invaluable contextual data for validating participant behaviors.
Streamlined Data Collection Procedures
In-Lab Study Dynamics
After equipping participants with wearable devices, they engaged in a series of 12 meticulously designed activities, each lasting 5 minutes with brief rest intervals. The activities spanned various intensities and domains, including sedentary and vigorous tasks.
Using a Vmax 29n Encore metabolic cart, we collected breath-by-breath pulmonary gas exchange data, allowing for a precise calculation of energy expenditure through the Weir equation.
Free-Living Study Insights
For the free-living study, a different cohort was trained to wear wrist devices during their typical daily activities. This real-world setting enriched our data collection, providing insights into how energy expenditure unfolds in everyday environments.
Visual Activity Classification
To enhance data accuracy, we employed footage from the wearable camera to perform visual confirmation of the activities. This validation process allowed us to classify activities, ensuring that our data was both comprehensive and reliable.
Algorithm Development: A New Era in EE Estimation
Innovative Algorithm Pipeline
To develop our EE algorithm, we crafted a sophisticated pipeline designed to predict METs (Metabolic Equivalents) over various time durations. By analyzing accelerometer and gyroscope data, we synchronized and processed this information to create a robust estimation model.
Classifying Activities
Our algorithm’s first step involved binary classification of activities into sedentary and non-sedentary categories. We employed advanced machine learning models, with the XGBoost classifier emerging as the top performer for accurately distinguishing between activity types.
Regression Model for Enhanced Predictions
After classification, we extracted intensity-based features to be used in the regression model, ensuring that participant demographics were integrated into the analysis. This innovative approach allowed for nuanced energy expenditure predictions tailored to individual users.
Rigorous Evaluation of the Algorithm
Assessing Performance in Controlled and Real-World Settings
In the in-lab setting, we utilized leave-one-participant-out cross-validation (LOPO-CV) to rigorously evaluate our model’s accuracy against established methods. Our proposed algorithm showed statistically significant improvements in estimating METs, confirming its robustness and reliability.
Free-Living Contextual Validity
For the free-living assessment, we optimized our model based on in-lab results and compared it against established models. Using a statistical threshold to evaluate under- and overestimations allowed us to maintain rigor in our findings.
Conclusion: Shaping the Future of Wearable Technology
Our research has paved the way for a BMI-inclusive energy expenditure algorithm that not only enhances the accuracy of wearable technology but also improves user engagement and data applicability in real-life contexts. As wearables continue to evolve, studies like ours highlight the potential for personalized health monitoring and its impact on lifestyle choices.
For further reading and exploration of foundational studies, we invite you to check out these relevant sources on wearable technology and energy expenditure estimation.
Engage with us as we lead the charge toward a healthier tomorrow, one innovative algorithm at a time!
