Cross-Validation of a New General Population Resting Metabolic Rate Prediction Equation Based on Body Composition

• Published in: Nutrients Vol. 15; no. 4; p. 805
• Main Authors: Kfir, Aviv, Lahav, Yair, Gepner, Yftach
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 01.02.2023; MDPI
• Subjects: Adipose tissues; Agreements; Basal metabolism; Bias; Body composition; Body fat; Calibration; Calorimetry; Dual energy X-ray absorptiometry; Energy; equation; Gastrointestinal surgery; Health aspects; Integrated software; Measurement; Metabolic rate; Metabolism; Nutrition; Physiological aspects; Population; prediction; Prediction models; resting metabolic rate; Variables; Israel; prediction; body composition; resting metabolic rate; equation
• ISSN: 2072-6643; 2072-6643
• DOI: 10.3390/nu15040805

Current prediction equations for resting metabolic rate (RMR) were validated in a relatively small sample with high-individual variance. This study determined the accuracy of five common RMR equations and proposed a novel prediction equation, including body composition. A total of 3001 participants (41 ± 13 years; BMI 28.5 ± 5.5 kg/m ; 48% males) from nutrition clinics in Israel were measured by indirect calorimetry to assess RMR. Dual-energy X-ray absorptiometry were used to evaluate fat mass (FM) and free-fat mass (FFM). Accuracy and mean bias were compared between the measured RMR and the prediction equations. A random training set (75%, = 2251) and a validation set (25%, = 750) were used to develop a new prediction model. All the prediction equations underestimated RMR. The Cunningham equation obtained the largest mean deviation [-16.6%; 95% level of agreement (LOA) 1.9, -35.1], followed by the Owen (-15.4%; 95% LOA 4.2, -22.6), Mifflin-St. Jeor (-12.6; 95% LOA 5.8, -26.5), Harris-Benedict (-8.2; 95% LOA 11.1, -27.7), and the WHO/FAO/UAU (-2.1; 95% LOA 22.3, -26.5) equations. Our new proposed model includes sex, age, FM, and FFM and successfully predicted 73.5% of the explained variation, with a bias of 0.7% (95% LOA -18.6, 19.7). This study demonstrates a large discrepancy between the common prediction equations and measured RMR and suggests a new accurate equation that includes both FM and FFM.