Prediction of resting energy expenditure in Italian older adults with severe obesity

Background In the last decade a large number of studies proposed and/or validated equations to estimate the Resting Energy Expenditure (REE) in adults and/or older adults, however, no equation currently available showed good accuracy for older adults with severe obesity. Thus, this study aimed to develop and validate new predictive equations for REE, based on data from the indirect calorimetry, in Italian older adults with severe obesity. Methods A retrospective study was as conducted with 764 Caucasian older adults with severe obesity (age range: 60-74 years and BMI ≥ 35 kg/m/²). Four models were used to test the accuracy of anthropometry and body composition variables in multivariable prediction of REE. All models were derived by stepwise multiple regression analysis using a calibration group of 382 subjects [295 females and 87 males] and the equations were cross-validated in the remaining 382 subjects [295 females and 87 males] as validation group. The new prediction equations and the other published equations were tested using the Bland-Altman method. Prediction accuracy was defined as the percentage of subjects whose REE was predicted within ± 10% of measured REE. Results All the equations analyzed predicted higher energy requirements for males than females, and most of them underestimated the energy requirement values of our sample. The highest accuracy values were observed in the new equations, with 62% in the anthropometric model and 63% in the body composition model. Conclusion Although the accuracy of our equations was slightly higher in comparison with the other taken into consideration, they cannot be considered completely satisfactory for predicting REE in Italians older adults with severe obesity. When predicting equations cannot guarantee precise or acceptable values of REE, the use of indirect calorimetry (if available) should be always recommended, especially in clinical practice.

Background: In the last decade a large number of studies proposed and/or validated equations to estimate the Resting Energy Expenditure (REE) in adults and/or older adults, however, no equation currently available showed good accuracy for older adults with severe obesity.Thus, this study aimed to develop and validate new predictive equations for REE, based on data from the indirect calorimetry, in Italian older adults with severe obesity.
Methods: A retrospective study was as conducted with 764 Caucasian older adults with severe obesity (age range: 60-74 years and BMI ≥ 35 kg/m/²).Four models were used to test the accuracy of anthropometry and body composition variables in multivariable prediction of REE.All models were derived by stepwise multiple regression analysis using a calibration group of 382 subjects [295 females and 87 males] and the equations were cross-validated in the remaining 382 subjects [295 females and 87 males] as validation group.The new prediction equations and the other published equations were tested using the Bland-Altman method.Prediction accuracy was defined as the percentage of subjects whose REE was predicted within ± 10% of measured REE.
Results: All the equations analyzed predicted higher energy requirements for males than females, and most of them underestimated the energy requirement values of our sample.The highest accuracy values were observed in the new equations, with 62% in the anthropometric model and 63% in the body composition model.

Conclusion:
Although the accuracy of our equations was slightly higher in comparison with the other taken into consideration, they cannot be

Introduction
The aging population is increasing worldwide and, in parallel, it is also observed an increase in the prevalence of older people with obesity (1), which is caused by a chronic energy imbalance wherein energy intake regularly exceeds energy expenditure (2).
An important factor to be considered in older adults with obesity is the pronounced body composition changes, with greater deposition of fat mass in visceral organs, at the expense of the amount of fat-free mass, particularly muscle mass (3).These age-related changes are due to the interplay between aging, inflammation, bad eating habits, metabolic disorders, and oxidative stress (4).
It is well established that both aging and obesity, contribute to altering the Resting Energy Expenditure (REE), which represents in most individuals around 50-70% of the total daily energy expenditure (5), which also includes energy consumed during food thermogenesis and physical activity.REE is considered an important parameter to provide a proper caloric intake essential to maintain the ideal energy balance and/or to determine weight loss in subjects with obesity (6).
The most common method of measuring REE in clinical is Indirect Calorimetry (IC) (7) which measures oxygen and carbon dioxide concentrations in the expired air, and then calculates the energy expenditure with equations based on the concentration values of these gases (8).Although the IC is a non-invasive and reliable method, its application is limited because it requires specialized equipment and trained personnel, which can be costly in daily clinical practice (9).
Thus, several predictive equations developed to calculate the REE have already been created in the last 10 years, and a large number of studies proposed new equations and/or validated the existing ones.However, they included data analysis considering adults with obesity without age differentiation (10-13), normal weight and/or older adults (10, 14-16).Other equations were proposed considering groups of people classified by gender (17)(18)(19), ethnicity (20,21), hospitalization (22) and the presence of chronic diseases (23).Despite being widely referenced, these equations were designed for the general population, without considering existing divergences related to age groups and other epidemiological and body composition variables.
In a recent systematic review, Cioffi et al. (24) evaluated the prediction accuracy of REE in healthy older adults according to the group level, and after analyzing 14 included studies, they concluded that, to date, no equation could be recommended because of the criteria limitation.To the best of our knowledge, to date, two equations for older adults with obesity have been validated.The study performed by Noreik et al. (5) found the best accuracy for estimating REE in the equation proposed by Lührmann et al. (3) (0.876, p = 0.0001), which however was elaborated only with body weight and did not take into account body composition variables.In another recent validation study performed by Griffith et al. (6), the best accuracy (59%) for older adults with obesity was observed in the equation proposed by FAO/WHO/ONU (1985), which however was not considered completely satisfactory.
Given the above findings, it remains important to investigate new accurate predictive equations to determine the REE in older adults with obesity, considering their differences according to subgroups (age, gender, ethnicity, and body composition), since these results could help to improve the health strategies in this population at-risk.Thus, this study aimed to develop and validate new predictive equations for REE, based on data from indirect calorimetry, in Italian older adults with severe obesity.
2 Subjects and methods

Study sample and design
A retrospective study was performed on 764 Caucasian older adults with severe obesity (590 females and 174 males) hospitalized between January 2016 and January 2019 at the Division of Metabolic Diseases, Italian Institute for Auxology, IRCCS, Piancavallo (VB), Italy, for a 3-week multidisciplinary integrated body weight reduction program (BWRP).Briefly, the BWRP entailed an energy-restricted diet in combination with physical rehabilitation (moderate aerobic activity), psychological counseling, and nutritional education (25).
The inclusion criteria were: 1) age between 65 and 74 years; and 2) Body Mass Index (BMI) calculated as weight (kg)/height² (m) ≥ 35 kg/m 2 , according to the cut-off for obesity in older people recommended by the Pan American Health Organization (2001).Individuals who had overt metabolic and/or endocrine diseases (i.e.diabetes, hypothyroidism, hypertension, amenorrhea), acute or chronic kidney/liver diseases, acute or chronic infection or inflammatory conditions, autoimmune diseases, malignant diseases, neurodegenerative diseases, hematological and/or oncological disorders, and those taking medications regularly or using any drugs known to influence energy metabolism were excluded.For each participant, anthropometric and instrumental measurements, such as the evaluation of body composition by bioimpedance analysis (see below for details), were collected.
This study was approved by the Ethical Committee of Istituto Auxologico Italiano, IRCCS, Milan Italy (code: 2023_03_21_06; research code: 01C311, acronym: EQUASTIMET60).Written informed consent was signed by all participants at the admission to the Hospital.

Anthropometric measurements and body composition
All the anthropometric measurements, REE, and body composition analysis were taken before the beginning of the BWRP by the same trained operators, according to the Anthropometric Standardization Manual (26).
Body composition was measured by using a multifrequency tetrapolar impedancemeter (BIA, Human-IM Scan, DS-Medigroup, Milan, Italy) with a delivered current of 800 A at a frequency of 50 kHz.To reduce errors of measurement, attention was paid to the standardization of the variables that affect measurement validity, reproducibility, and precision.Measurements were performed according to the method of Lukaski et al. (27) after 20 minutes of rest in a supine position with relaxed arms and legs without any contact with other body parts.Fat-free mass (FFM) was estimated using the prediction equation for adults developed by Bedogni et al. (28).Fat Mass (FM) was obtained by subtracting FFM from body weight (BW) and % FM as (FM/BW) × 100.

Resting energy expenditure
REE was determined in a 22 to 25°C room in the morning (between 08:00 and 10:00 a.m.) and after an overnight fast using an open-circuit, indirect computerized calorimetry with a rigid, transparent, ventilated canopy (Vmax 29, Sensor Medics, Yorba Linda, CA), periodically undergone to quality control tests to ensure the reliability of the measurements.Before each test, the gas analyzers were calibrated using a reference gas mixture (15.00% O 2 and 5.00% CO 2 ).The subjects were fasting from at least 8 hours, were not smoking for at least 1 hour, and waited 30 minutes in a sitting position before undergoing REE measurement.After achieving a steady state in the lying position, the REE of individuals was measured for 30 minutes.The data relative to the acclimation period were discarded.The steady state was defined as at least 5 minutes< 5% variation in the respiratory quotient,< 10% variation in O 2, and< 10% variation in minute ventilation (29).Oxygen consumption and carbon dioxide production, standardized for temperature, barometric pressure and, humidity, were recorded at 1-min intervals for a minimum of 30 min and averaged over the whole measurement period.Energy expenditure was derived from the measured oxygen uptake and carbon dioxide output according to the equation of Weir (30).

Statistical analysis
Values of continuous variables are given as mean and standard deviation or median and interquartile range (IQR), according to the data distribution.The data were analyzed using the Kolmogorov-Smirnov test to verify the normal distribution.Student's t-test or Mann-Whitney U test was used to compare continuous variables between groups, as appropriate.For the development and crossvalidation of new equations for predicting REE, the data set of 764 measured REE values was randomly split into two groups: a calibration group (n: 382, 295 females and 87 males) for the development of predictive equations, and a cross-validation group (n: 382, 295 females and 87 males) for the validation of the predictive equations.In the calibration group, the univariable relationships between REE and continuous predictors (age, body weight, height, FFM, and FM) were first studied using scatterplots and non-parametric regression plots.A first-degree linear model was as accurate as more complex models to describe all the REEpredictor relationships and was thus chosen as the reference model for all univariable analyses.Four pre-specified models were used to test the accuracy of anthropometry and body composition in multivariable prediction of REE.Model 1 was based on body weight, age, and gender; Model 2 added height to the predictors of Model 1; Model 3 was based on age, gender and FFM; Model 4 added FM to the predictors of Model 3. Standard diagnostic plots were used to test univariable and multivariable model fit.The adjusted coefficient of determination (R 2 adj ) and the root mean squared error of the estimate (RMSE) were used as measures of model fit.
The new prediction equations and the other published equations were tested in the cross-validation group by using the Bland-Altman method (31) to ascertain the degrees of systematic and magnitude of bias when predicted REE was compared with measured REE.In addition, prediction accuracy was defined as the percentage of subjects whose REE was predicted to be within ±10% of measured REE.This error limit on prediction accuracy was accepted empirically as being consistent with calorimetry measurement errors of 5% or less (32).Differences between predicted REE and measured REE higher than ±10% were considered prediction errors and are reported as percentages of subjects whose REE was overestimated or underestimated.
Statistical analysis was performed using STATA 18.0 (STATA Corp, College Station, TX, USA).

Results
The main characteristics of 764 older adults with severe obesity (590 females and 174 males) are shown in Table 1.Female and male subjects with obesity had a comparable age between 65 and 74 years and showed a mean BMI of 40.3 kg/m2 [37.4 -44.8] for females and 38.5 kg/m2 [35.5 -41.5] for males (p-value< 0.001).Male subjects with obesity were significantly taller (+8.9%, p-value< 0.001) and had the highest FFM values compared to females (+39.5%,p-value< 0.001), while female subjects with obesity showed a higher amount of FM (+19.7%, p-value< 0.001).The REE obtained with indirect computerized calorimetry was significantly higher in males (+20.8%,p-value< 0.001) compared to female patients, but REE normalized for FFM was significantly lower in males than females (-13.7%,p-value< 0.001).
The association between the obtained REE and the physical variables was analyzed through univariate regression analysis (Table 2).A significant association was observed between the REE and all the tested variables: male gender, age, height, BW, BMI, FFM and FM.
Table 3 shows that all models of multivariable regression analysis had similar accuracy for predicting REE in the analyzed sample (R 2 adj ~0.46; p< 0.001).The insertion of the height variable (Model 2) did not significantly modify the model of simple anthropometric measurements based on BW, age and gender (Model 1).Similarly, the insertion of the FM variable (Model 4) did not substantially change the fit value of the model for body composition based on age, gender and FFM (Model 3).
All equations, with both anthropometric and body composition measurements, predicted higher energy requirements for males than females (p<0.001)(Table 4).
Comparing the REE estimated by indirect calorimetry and the prediction equations analyzed based on anthropometric measurements, it was found that only the new equation proposed and the equation proposed by FAO/WHO/UNU (33) did not show a significant difference (Table 5).All other equations were significantly different, with most of them underestimating the energy requirement.Significant overestimations were observed in the equations proposed by Luhrmann et al. ( 3) and Fredrix et al. (37).
Considering prediction equations based on body composition measurements, the new equation and equations proposed by Cunningham (38) 5).
The new proposed equations had accuracies of 62% and 63%, considering anthropometric and body composition models, respectively.Considering the other equations analyzed, the most accurate was the equation proposed by  and Schofield (36) with 60%, followed by Pavlidou et al. (15) with 59% (Table 5).
Most of the analyzed equations presented a systematic bias for underestimation of energy requirements, as observed in Figure 1.The smallest values of relative bias were found in the new equation, in the FAO/WHO/UNU (33) (1.2%) and in the Harris-Benedict (34) (-1.9%) considering anthropometric models, and in the Korth et al. (39) (-0.8%) and the Cunningham (38) (-1.2%) considering body composition models.

Discussion
In the present study we proposed new equations to estimate REE in older adults with severe obesity based on data of indirect calorimetry and also, we validated them and other main equations already proposed in the literature.In summary, there was a variation in the previously proposed equations to predict REE in our sample, in which most of them underestimated the energy requirement and had poor accuracy.The best accuracies were found in the new proposed equations, with 62% and 63%, respectively, considering the anthropometric and body composition in the models.
Our results showed that only the new equations and the equation proposed by FAO/WHO/UNU (33) (for anthropometric models) and by Cunningham (38) and Korth et al. (39) (for body composition models) did not show a significant difference between the estimated values and those obtained by using the indirect calorimetry of REE.This evidence contradicts that found by Noreik et al. (5) in which a better accuracy for REE estimation in older adults with obesity was observed in the equation proposed by Lührmann et al. (3).However, this equation was tested in a sample of not hospitalized patients with a mean BMI value of 26 kg/m², which is far lower than the one considered for our sample of severely obese.Moreover, other studies that analyzed samples of old people (3,14,15,36,37) also included non-obese individuals, making extremely difficult the comparison of the results.
It is important to highlight that advanced aging alters the body composition, with an increase in adipose tissue and a decrease in lean tissues (42), and changes in both the quantity and quality of lean mass which are key factors to determine the REE precision of prediction equations.Due to their greater body mass, individuals with severe obesity require a higher REE than eutrophic people, to maintain their essential body functions (17).However, the weight excess in individuals with obesity has a significant impact on the age-related declines in terms of REE, with a mean decline almost 1.5 times greater than normal weight individuals (43).
Recently, Griffith et al. (6) validated four equations commonly used to guide weight loss interventions in a sample of older adults with obesity (≥ 65 years and BMI ≥ 30kg/m²) and found that FAO/ WHO/UNU (33) also had the best accuracy for REE estimation.However, none of them obtained values above 60%.In the present study, although the highest accurate values were observed in the models with anthropometric variables, only the new equation with body composition variables achieved an accuracy of 62%.Other body composition-based prediction models compared in our study had a lower agreement, including the study by Batista et al. ( 14) with 40%, which was the only one that analyzed a sample of older adults, however non-obese.This finding is in agreement with other studies where FFM values were more pertinent than isolated anthropometric variables during energy requirements estimations (38,39,41).
In the present study, the REE obtained by indirect calorimetry was higher in men than in women, and in the same way, all equations predicted higher energy requirements for men.However, when REE values were normalized to the FFM, higher REE values were observed in women.The greater amount of fat mass in older women than in men could lead to a lower basal metabolic rate and, consequently, lower energy (18), thus reinforcing the importance of considering FFM values during REE estimations.Also, it is important to note that the inclusion of the FM variable in our prediction model did not substantially change the predictive estimation of REE calculated only with the FFM.Although our model explained less than 50% of the variance, it is in agreement with other studies that found FFM as the major factor related to REE (38,44).
Most of the body composition equations tested in our study presented a tendency to underestimate the REE values.As previously mentioned, body composition, especially the FFM values of the sample and other measures of nutritional status are the main factors that influence the REE.In addition to the differences in clinical characteristics of the samples, the methods used in each study to calculate body composition should be considered to explain these variations.While we used BIA measurements, other tested equations were created based on skinfold measurements (38,40) or densitometry methods (i.e.DXA) (14).It is also important to mention that other variables not included in our analysis may contribute to the variations observed in the REE of older adults with severe obesity.According to Martin et al. (45), the race can significantly affect REE variability.In fact, these authors demonstrated that black individuals typically have lower REE than white participants.Another anthropometric characteristic described in the literature is waist circumference, which has been shown to tend to increase in women with aging (46).
Some limitations of our study should be considered while interpreting the results.The new equations might not be entirely representative of European older adults with obesity being the average BMI of our sample considered representative of a severe obesity condition.Also, the sample was recruited in a hospital setting, and patients with serious illnesses were excluded.Therefore, body composition variables were measured by BIA, while DXA is still considered the gold standard method for the estimation of fat mass and fat-free mass.
Nevertheless, REE was measured by indirect calorimetry with a highly accurate method.To the best of our knowledge, our study was the first to propose specific equations for European older adults with severe obesity, with a mean BMI of 40.3 kg/m² for women and 38.5 kg/ m² for men.Furthermore, our equation based on body composition measurements achieved an accuracy greater than 60%, which had not been found in previous studies.The validation analysis considered all the main equations used in the last decades, highlighting the importance of using specific equations for each population.
In conclusion, although the accuracy of our equations was slightly higher (62% and 63% considering anthropometric and body composition models, respectively) in comparison with the other available equations taken into consideration, they cannot be considered completely satisfactory for predicting REE in Italians older adults with severe obesity.To date, we cannot exclude that the use of these equations in other different populations might provide more accurate results than those obtained in the Italian population from which formulas have been developed.When predicting equations cannot guarantee precise or acceptable values of REE, the use of indirect calorimetry (if available) should be always recommended, especially in clinical practice.
and Korth et al. (39) did not show a significant difference from REE.However, the equations proposed by Mifflin et al. (35), Owen (1988), Weigle et al. (41) and Batista et al. (14) showed a significant underestimation of REE (Table

FIGURE 1 Bland-
FIGURE 1Bland-Altman plots of energy expenditure measured by indirect calorimetry and predicted by fifteen tested equations.The dashed lines represent the mean difference between Resting Energy Expenditure (REE) predicted by each of the tested equations and REE measured by indirect calorimetry and ±1.96 standard deviations (SD) of the mean (limits of agreement), with a 95% confidence interval (dotted lines).

TABLE 2
Univariable regression analysis between anthropometric and body composition variables and Resting Energy Expenditure (REE) in the calibration group (n: 382, 295 females and 87 males).

TABLE 1
Physical characteristics of the sample (n:764).
All results are median (IQR) BMI, Body Mass Index; FFM, Fat-Free Mass; FM, Fat Mass; REE, Resting Energy Expenditure p-value for independent Student t-test for comparison between gender.

TABLE 4
Resting Energy Expenditure (REE) measured by calorimetry and estimated by predictive equations in the validation group (n: 382).
All results are mean ± SD p-value for independent Student t-test for comparison between genders.

TABLE 3
Multivariable regression analysis between anthropometric and body composition variables and Resting Energy Expenditure (REE) in the calibration group (n: 382, 295 females and 87 males).

TABLE 5
Accuracy of predictive equations for resting energy estimation compared to Resting Energy Expenditure by indirect calorimetry in the validation group (n: 382).
p-value: Paired t-test comparing predictive equations with the reference method (indirect calorimetry).RMSE, Root mean square error.