Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

High School Poster

HS Poster

Poster 153 - Uncovering Transparency in Black Box Metrics Provided by Wearable Devices: A Feasibility Study on Elite Warfighters

Saturday, October 26, 2024
10:00 AM - 12:00 PM EST
Location: Exhibit Hall E, F & G

Introduction:

Introduction: Wearable technologies track physiological metrics while worn on the body. Many wearable companies are suited to improving human performance, providing instantaneous data during exercise and long-term sleep and health reports. To aid in this endeavor, wearable companies create proprietary algorithms to simplify a user’s view of their current health. The Body Battery metric by Garmin is one example. Garmin claims that body battery is derived through a combination of heart rate (HR), heart rate variability (HRV), and movement data. While studies have validated Garmin’s ability to track real metrics, it remains an unmet engineering need to assess the validity of body battery because of its unknown formula (e.g. often referred to as black box analytics in the digital health community). A second drawback of a body battery is its dependence on compliance from the user. Inconsistent readings are reported if users aren’t constantly wearing the watch. In addition, small inconsistencies in readings could drastically affect the score if they are heavily weighted in the formula.

Materials and

Methods: Materials and

Methods: Five participants of an elite warfighter team were tracked using Garmin’s Fenix 7 throughout twelve weeks of training from April 1 to June 21. Each week consisted of training from Monday through Friday with a specialized activity designated for that week. The analysis began on May 20th with four participants recorded. Data collection was conducted using the Garmin Clipboard application (Garmin Ltd., Version 3.4.3) and extracted into CSV files. Data cleaning consisted of replacing missing values with a random sampling of the mean and standard deviation of that column. A visual analysis was conducted with the creation of a descriptive statistics table, histograms, correlation matrices, and single regressions, all sorted by week and variable. Finally, a statistical analysis was completed implementing a principal component analysis (PCA) to reduce dimensionality. The factor loadings from the PCA were used in a regression analysis to deconstruct the body battery. The entire analysis was done in Python (Version 3.11.7) and Microsoft Excel (Version 2406).

Results, Conclusions, and Discussions: Results and

Discussion: Our equation combines each coefficient and multiples them by their corresponding factor loading to predict Body Battery (Average). However, with an average r-squared score of 0.112, our linear model struggled to accurately predict body battery scores. In addition, only two of our five p-values were < 0.05, so we cannot assume our data was statistically significant. Our model performed poorly in a sensitivity and specificity analysis, scoring 0.00758 and 0.000, respectively.

Conclusions: With an average r-squared score of only 0.112, our model showed a poor correlation between Garmin metrics and body battery. This result can be credited to two deficiencies of Garmin. First, incompliance with the military combined with long tracking periods left many data points missing. After five weeks of training with the watches, only 16.66% of the body battery points were recorded. Second, users may be unaware of how Garmin is calculating their body battery scores. With little to no correlation between variables, Garmin’s body battery cannot be verified to be formulated through the metrics they claim or be a valid tracker of daily health. These results suggest further testing is required to fully understand the body battery metric and what Garmin is doing to assure its accuracy.

Acknowledgements (Optional):

References (Optional):