(58) THE ACCURACY OF BASAL ENERGY EXPENDITURE ESTIMATION BASED ON HEAT FLUX AND MACHINE LEARNING AS BODY COMPOSITION CHANGES

The basal energy expenditure (BEE) provides more than half of our total daily energy consumption. However, most sports enthusiasts monitor BEE estimates using portable monitors based on background information (age, gender, height, weight) and the user's heart rate. It is well known that changes in body composition affect to BEE [1]. The 1^st law of thermodynamics proves that all chemical processes (e.g. metabolism) in our body produce heat. A novel heat flux sensor solution measures amount of heat released from the skin surface [2] combined with a machine learning algorithm.

Purpose: The purpose of this study was to compare the accuracies of these two different BEE assessment-methods with indirect calorimetry of respiratory gases before and after preparing (23 weeks) and recovering (23 weeks) phases of fitness and body building athletes.

Methods: A total of 58 physique athletes volunteered to participate in this study, out of which 27 athletes competed during the research period. In addition, 31 athletes were the control group and they performed fitness-related strength training with no goals in competition and body fat reduction during the study period. BEE measurements were taken simultaneously to measure respiratory gases, heart rate and heat flux, and body composition was analyzed at the same visit. BEE was calculated using Weir’s equation [4] as the gold standard and a trained algorithm with heat flux. Data were collected at three measurement points that differed in terms of body composition targets set for competitive athletes.

Results: After 23 weeks of preparation, competitors had significantly lower body fat percentages compared to control athletes (women 14.1 ± 7.4% and 25.3 ± 6.8%, p< 0.05; men 5.6 ± 2.1% and 18 ± 7.2%, p< 0.001). Results showed that BEE estimates by the heat flux method (1656 ± 410 kcal/d) were not statistically different from indirect calorimetry (1657 ± 305 kcal/d), in contrast to background information estimations (1718 ± 270 kcal/d, p< 0.001). Furthermore, EE values from the heat flux method and indirect calorimetry were positively correlated (r=0.35, p< 0.001), whereas resting heart rates and indirect calorimetry values were not correlated (p >0.05). The configuration combining background information and heat flux gave the lowest mean error (6.73%) for BEE. The accuracy of heat flux estimates did not differ between the competition and control groups. However, there was a statistical difference between sexes (p< 0.05), with higher errors in estimates for females.

Conclusions: The heat flux method using machine learning appears to be promising tool for more accurate estimation of basal energy expenditure. Overall, the average error, ~7%, is much lower than errors of consumer devices documented in previous studies [3]. Further studies should also focus on testing heat flux and machine learning in different populations as their body composition changes. PRACTICAL APPLICATION: These data can be used to develop more accurate EE assessments and thus more effective applications, for example to improve body composition for different purposes.

< !1. Isola et al. (2023) 2. Levikari et al. (2021) 3. Shcherbina et al. (2017) 4.Weir (1949)

Acknowledgements: None