Bias Characterization of Resting Energy Expenditure Prediction Equations in Japanese Subjects: A Pilot Exploratory Study

Background: Accurate estimation of energy requirements is essential for appropriate nutritional therapy. Resting energy expenditure (REE) predictive equations are derived from specific populations and may therefore contain systematic individual-level biases when applied in different clinical settings. In this pilot exploratory study, we compared REE measured by portable indirect calorimetry with estimates obtained using multiple predictive methods in Japanese adults and characterized their association, absolute agreement, and error structure. Methods: Thirty-six university staff members and students were enrolled. The association and absolute agreement between REE measured by portable indirect calorimetry and estimates obtained using four predictive methods (Harris–Benedict, DRIs, Cunningham, and Ganpule) were evaluated using Spearman’s rank correlation coefficients, ICC(2,1), and Bland–Altman analyses. For equations demonstrating significant proportional bias, multivariate linear regression analyses were additionally performed using residual estimation error, calculated as (predicted REE-measured REE)/measured REE × 100, as the dependent variable. Results: Participants included 16 men and 20 women, with a mean age of 28.9 ± 10.3 years and BMI of 22.1 ± 3.1 kg/m2. Spearman’s rho ranged from 0.750 to 0.800, and ICC(2,1) estimates ranged from 0.736 to 0.766. Mean biases (predicted minus measured REE) were 50.7 kcal/day (95% CI, −20.1 to 121.4) for Harris–Benedict, −73.1 kcal/day (95% CI, −148.8 to 2.6) for DRIs, −94.7 kcal/day (95% CI, −161.7 to −27.8) for Cunningham, and −90.6 kcal/day (95% CI, −161.1 to −20.2) for Ganpule. The corresponding proportions of predictions within ±10% of measured REE were 41.7% (95% CI, 27.1–57.8%), 50.0% (95% CI, 34.5–65.5%), 55.6% (95% CI, 39.6–70.5%), and 52.8% (95% CI, 37.0–68.0%), respectively. Multivariate analyses suggested greater underestimation in females and a shift toward overestimation with higher BMI. Conclusions: Despite moderate-to-strong rank associations with measured REE, Bland–Altman analysis revealed proportional bias in three of the four predictive methods that was not apparent from the correlation coefficients alone. Exploratory analyses suggested that sex and BMI should be considered when interpreting the direction of prediction error. Because the precision of the limits of agreement was restricted by the small sample size, larger studies are needed to estimate individual-level agreement more precisely and to examine these error patterns in older adults and individuals with acute or chronic diseases.