### INTRODUCTION

^{1}. The total energy expenditure in 24 hours consists of RMR, physical activity energy expenditure (PEE), and diet-induced thermogenesis (DIT). The RMR represents approximately 60-75% of daily energy expenditure (DEE) in a 70 kg person

^{2,3}, accounting for the largest contribution to the 24-hour energy expenditure. It consequently has a large impact on the regulation of body composition and energy balance. An abnormally high RMR is associated with pathologic and inflammatory conditions. It also tends to decrease with aging, with a low RMR playing an important role in the pathogenesis of obesity and age related chronic diseases

^{4-6}. Therefore, an accurate measurement of the RMR is very important.

^{7-9}. However, both the methods require tedious procedures and are time- and cost- consuming. Thus, many researchers have developed various regression models for RMR estimation

^{2,9-14}. Determining the contribution of the RMR to the DEE is an important calculation for understanding, developing, and executing body weight- related interventions

^{3,4}. For example, RMR estimation regression model is applied to determine the target energy intake in weight loss programs, develop dynamic prediction models of weight gain and loss, identify the patients with potential metabolic abnormalities, design public health programs promoting obesity prevention in diverse populations, and assess the potential energy deficits in metabolically stressed patients

^{4}.

^{10}was the most commonly used, but only 50-75% of the RMR variability could be explained by this equation. Its major disadvantage was overestimating RMR by at least 5%. Additionally, RMR was estimated only by age, height, and weight and the regression rate, R

^{2}value, did not exceed 0.7. Previous attempts to overcome these shortcomings failed to show high regression rates

^{11-14}, as they also estimated RMR with only age, height, weight, and lean body mass. Therefore, developing a regression model, with a higher regression rate using various dependent variables, is important for accurate RMR measurements.

### METHODS

### Subjects

^{7}. All subjects were of Korean origin, with a stable weight for at least 3 months prior to the measurements, and without a history of thyroid disease, diabetes mellitus-I or II, cardiovascular disease, or severe hypertension in the past 6 months. There was also no history of orthopedic disease or other medical issues over the past year in the pre-screening surveys. All the subjects received a medical clearance for their participation and were explained about the purpose, procedures, and potential risks of the study. All proceedings of the study were approved by the Institutional Review Board of Konkuk University (7001355-201903-HR-305) in Korea and were conducted according to the Declaration of Helsinki. All subjects arrived at the laboratory early in the morning (8:00 AM) after overnight fasting (≥ 8 hour) and rested for 30 minutes, after which their blood pressure, body composition, and resting HR was measured, followed by the RMR measurement. All the subjects were instructed to sleep for at least 8 hours before the RMR measurement and to stay awake during the process. If they fell asleep, their shoes or toes were squeezed to keep them awake.

##### Table 1.

### Height and body composition

### Blood pressure and resting heart rate

### Resting metabolic rate

_{2}and 5% CO

_{2}) before the measurement. All the RMR testing procedures were performed in a 9 m (width) × 7 m (length) × 3 m (height) chamber with a temperature of 23 ± 1 °C and a humidity of 50 ± 5%, regulated by the environmental control chamber (NCTC-1, Nara Controls, Seoul, Korea). Subjects were asked to limit their physical activity and abstain from alcohol intake one day before the measurement. The subjects took a rest for 30 min, prior to the measurement. The RMR was measured in a supine position for 30 min, and the average value of the last 25 min was used for the analysis.

### Statistical Analysis

*β*-value). Regression analysis using the stepwise method was used to predict the RMR from age, height, BMI, FFM, fat mass, % body fat, SBP, DBP, MAP, PP, and HR. A two-tailed student’s paired t-test was used to detect differences between measured and predicted RMR. Bias was calculated as the difference between measured and predicted RMR. The authors rigorously conformed to the basic assumptions of a regression model (linearity, independency, continuity, normality, homoscedasticity, autocorrelation, and outlier). Statistical Package for the Social Sciences (SPSS) version 24.0 (IBM Corporation, Armonk, NY, USA) was used for the statistical analysis and the level of significance (

*p value*) was set at 0.05.

### RESULTS

### Check outlier data

##### Table 2.

RMR (kcal/day) | ||||
---|---|---|---|---|

Both (n=53) |
Males (n=23) |
Females (n=30) |
||

Age (yrs) |
Correlationp-value |
-.284* .039 |
-.214 .327 |
-.212 .262 |

Body height (cm) |
Correlationp-value |
735* .000 |
.171 .434 |
.200 .289 |

Body weight (kg) |
Correlationp-value |
.743* .000 |
.293 .176 |
-.065 .732 |

Body mass index (kg/m ^{2}) |
Correlationp-value |
.536* .000 |
.275 .204 |
-.231 .218 |

Fat-free mass (kg) |
Correlationp-value |
.817* .000 |
.292 .176 |
.168 .375 |

Fat mass (kg) |
Correlationp-value |
-.191 .171 |
.142 .517 |
-.233 .215 |

Percent body fat (%) |
Correlationp-value |
-.631* .000 |
.036 .870 |
-.259 .167 |

Systolic blood pressure (mmHg) |
Correlationp-value |
.408* .002 |
.236 .278 |
-.250 .183 |

Diastolic blood pressure (mmHg) |
Correlationp-value |
.279* .043 |
.295 .172 |
-.220 .242 |

Mean arterial pressure (mmHg) |
Correlationp-value |
.346* .011 |
.292 .176 |
-.246 .191 |

Pulse pressure (mmHg) |
Correlationp-value |
.277* .045 |
-.099 .653 |
-.097 .610 |

Heart rate (beat/min) |
Correlationp-value |
-.149 .288 |
.004 .986 |
-.060 .751 |

### Significance of regression models and the independent variables

*t*-test to verify the significance of the regression coefficients of the independent variables.

### Performance evaluation of regression models and regression equations

^{2}), adjusted coefficients of determination (adjusted R

^{2}), and standard errors of estimates (SEE) were calculated for the regression model. The mean explanatory power of RMR

_{1}regression models estimated only by FFM was 66.7% (R

^{2}) and 66.0% (adjusted R

^{2}), while the mean SEE was 219.85 kcal/day <Table 4>. The mean explanatory power of RMR

_{2}regression models developed by FFM and age were 70.0% (R

^{2}) and 68.8% (adjusted R

^{2}), while the mean SEE was 210.64 kcal/day <Table 4>.

### Difference between measured and predicted RMR of young and middle-aged Korean adults

_{1}and RMR

_{2}equations. The mean bias between the measured RMR and the predicted RMR

_{1}and RMR

_{2}equations was + 0.02 kcal/day and -0.01 kcal/day, respectively <Table 5>. The measured and the predicted RMR showed a similar average value, and their correlation coefficients also showed a significant correlation (measured RMR and predicted RMR

_{1}: R = 0.817,

*p*= 0.000, measured RMR and predicted RMR

_{2}: R = 0.837,

*p*= 0.000) <Figure 1>.

### DISCUSSION

_{1}= 24.383 × FFM + 634.310, RMR

_{2}= 23.691 × FFM - 5.745 × age + 852.341).

^{15}. No outliers were observed in this study. This finding demonstrated a clear linearity between the independent and the dependent variables.

_{1}equation) and 70.0% (RMR

_{2}equation) of the variance in the criterion variable of RMR was attributable to the variance of the combined predictor or independent variables.

^{10}developed separate RMR estimation models (male = 66.5 + 13.75 × weight + 5.003 × height - 6.775 × age, R

^{2}= 0.64, female = 655.1 + 9.563 × weight + 1.850 × height - 4.676 × age, R

^{2}= 0.36) for males (n = 136) and females (n = 103). Only 50-75% of the RMR variability could be explained by this equation, with the disadvantage of overestimating the RMR by at least 5%. Further, the RMR was estimated only by age, height, and weight and the regression rate, R

^{2}value, did not exceed 0.7. Schofield

^{12}developed an RMR estimation model for Italian males based on multiple sample sizes (n = 2879, RMR

_{1}= 63.0 × weight + 2896, RMR

_{2}= 63.0 × weight - 0.42 × height + 2953). However, the samples had a disproportionate number of Italian military cadets, soldiers, workers, and miners, who did not represent the typical Italian population. These constituted 56% of the 18 - 30 years old male cohort

^{2}. Additionally, the regression rate of the RMR estimation model was low (RMR

_{1}: R

^{2}= 0.423, RMR

_{2}: R

^{2}= 0.423). Hayter & Henry

^{16}developed an RMR estimation model using Schofieldʼs database, which predicted the RMR in Northern Europeans and Americans, except Italians. However, this model also had a regression rate similar to that of Schofield (n = 478, RMR = 51.0 × weight + 3500, R

^{2}= 0.449)

^{11}re-evaluated the Harris-Benedict equation using the age, height, and weight used in the latter’s estimation model using normal subject data (n = 239) from the study and additional data, which were obtained from the subjects spanning a wider age range (n = 98). As a result, a model with a relatively higher regression rate (male: 88.362 + 13.397 × weight + 4.799 × height - 5.677 × age, R

^{2}= 0.77, female: 447.593 + 9.247 × weight + 3.098 × height - 4.330 × age, R

^{2}= 0.69) was developed. Also, Mifflin et al.

^{13}developed a predictive equation for RMR from the data of 498 healthy subjects, including females (n = 247) and males (n = 251), aged 19-78 yrs (45 ± 14 yrs). Normal weight (n = 264) and obese (n = 234) individuals were studied and the RMR was measured by indirect calorimetry. This model showed high regression rates for both males and females (male: 10 × weight + 6.25 × height - 5 × age + 5 kcal/day, R

^{2}= 0.71, female: 10 × weight + 6.25 × height - 5 × age - 161 kcal/day).

^{11}and Mifflin et al.

^{13}using FFM and age, with a smaller sample size. This was due to the strict adherence to the basic assumptions of a linear regression model. Compared to a study by Piers et al.

^{9}, which developed an RMR estimation model using similar sample sizes and independent variables, the present model showed a higher regression rate.

_{1}= 24.383 × FFM + 634.310, RMR

_{2}= 23.691 × FFM - 5.745 × age + 852.341. The bias (RMR

_{1}= 0.02, RMR

_{2}= -0.01) and correlation (RMR

_{1}: R = 0.817, RMR

_{2}: R = 0.837) between the estimated RMR and the measured RMR were reasonable.