## Abstract

Critical illness affects body composition profoundly, especially body cell mass (BCM). BCM loss reflects lean tissue wasting and could be a nutritional marker in critically ill patients. However, BCM assessment with usual isotopic or tracer methods is impractical in intensive care units (ICUs). We aimed to modelize the BCM of critically ill patients using variables available at bedside. Fat-free mass (FFM), bone mineral (Mo), and extracellular water (ECW) of 49 critically ill patients were measured prospectively by dual-energy X-ray absorptiometry and multifrequency bioimpedance. BCM was estimated according to the four-compartment cellular level: BCM = FFM − (ECW/0.98) − (0.73 × Mo). Variables that might influence the BCM were assessed, and multivariable analysis using fractional polynomials was conducted to determine the relations between BCM and these data. Bootstrap resampling was then used to estimate the most stable model predicting BCM. BCM was 22.7 ± 5.4 kg. The most frequent model included height (cm), leg circumference (cm), weight shift (Δ) between ICU admission and body composition assessment (kg), and trunk length (cm) as a linear function: BCM (kg) = 0.266 × height + 0.287 × leg circumference + 0.305 × Δweight − 0.406 × trunk length − 13.52. The fraction of variance explained by this model (adjusted *r*^{2}) was 46%. Including bioelectrical impedance analysis variables in the model did not improve BCM prediction. In summary, our results suggest that BCM can be estimated at bedside, with an error lower than ±20% in 90% subjects, on the basis of static (height, trunk length), less stable (leg circumference), and dynamic biometric variables (Δweight) for critically ill patients.

- critical illness
- body composition
- bioelectrical impedance
- dual-energy X-ray absorptiometry

other than the basic two-compartment [fat and fat-free mass (FFM)] model, the composition of the human body can be described by multicompartment models from the atomic to the functional level (11). Body cell mass (BCM) is the metabolically active compartment of FFM that reflects the body's cellular components involved in oxygen consumption, carbon dioxide production, and resting metabolism (11, 19, 20). Thus, physiological modeling of body composition at the cellular level can be separated into different compartments associated with functions: energy store in fat mass (FM), energy expenditure and metabolism by BCM, exchanges in extracellular water (ECW), and support by bone mineral (Mo).

In healthy adults, homeostasis maintains a stable body composition. Conversely, body composition is profoundly modified in acutely or chronically ill patients (28). BCM is altered by changes of nutritional status and the catabolic effects of disease (11, 39, 47). In this context, BCM could be a marker of malnutrition or prognosis of chronically or critically ill patients (14, 37). BCM is associated with resting energy expenditure (REE), and it has been shown that assessment of energy balance may have particular significance for the nutritional management of the most critically ill patients (8, 13, 44). Therefore, development of clinical tools for evaluating body composition, especially BCM/weight ratio (14), could help demonstrate the relevance of the concept of energy balance by patients hospitalized in intensive care units (ICUs).

In healthy subjects, total body potassium, determined by means of ^{40}K and NaBr dilution, is considered the gold standard for BCM and ECW measurements, respectively (39, 48). However, tracer dilution methods require steady-state H_{2}O turnover and constant body water pool size during the measurement period (25, 40). Because of massive shifts in body water and poor renal and intestinal function during critical illness, the isotope and tracer dilution methods are impractical in critically ill patients during the first or second week after their admission (7, 11, 25, 40, 48). Furthermore, radioactive tracer studies may be unethical in some countries and impracticable.

For the past two decades, bioelectrical impedance analysis (BIA), an easy and noninvasive technology available at bedside, has been widely used to assess water compartments or FFM in patients without marked fluid distribution disturbances (7, 21, 42). However, regression equations for FFM or BCM estimation derived from single- or multiple-frequency BIA are population specific and yield misleading results for patients with fluid overload (17, 19, 21, 32, 39). In addition, the use of single-frequency, 50-kHz BIA is not appropriate to estimate body water or FFM when critical illness is causing compartmental flux (17, 26). Multiple-frequency BIA, although less accurate in critical illness, remains the only bedside technique for estimating water compartments (26, 32, 43, 48). Therefore, combining multiple-frequency BIA with FM, FFM, and Mo determinations may be an alternative method of estimating BCM in critically ill patients. Dual-energy X-ray absorptiometry (DEXA) is becoming one of the reference methods for body composition assessment (11, 27, 33) and can be used in these settings. However, DEXA, like computed tomography, requires transport out of the ICU. We prospectively investigated the body composition of critically ill patients using multifrequency BIA and DEXA to derive a simple method for calculating BCM at bedside for the most severely ill patients.

## MATERIALS AND METHODS

#### Study setting and patient sample.

This prospective study was conducted over a 5-mo period in the medical and surgical ICUs of our hospital. We evaluated adults who were intubated and mechanically ventilated for >24 h and ≤7 days. The local Ethics Board approved the study design. Patients or immediate family members gave written informed consent or assent, respectively. Patients were included when their condition was compatible with intrahospital transport and met the following criteria: *1*) hemodynamic: absence of fluid loading and no introduction or dose modification of inotropic drugs during the preceding 4 h; *2*) respiratory: FiO_{2} <60%, no ventilator adjustments during the preceding 4 h, and no signs of hyperventilation (respiratory rate >35/min) or respiratory weakness during pressure-support ventilation; and *3*) no agitation. Exclusion criteria were pregnancy, any chronic pathology with a life expectancy <3 mo, and any clinical conditions responsible for erroneous measured DEXA or multifrequency BIA values (1, 2, 26, 27, 42): *1*) pacemaker or implanted cardiac defibrillator, *2*) amputated limb, *3*) orthopedic prosthesis/implants (metal); *4*) body mass index (BMI) >40 or <15 kg/m^{2}, *4*) abnormal body geometry (scoliosis or atrophy), *5*) ascites, *6*) skin lesions at the site where BIA electrode should be placed, *7*) use of contrast agents for diagnostic procedures during the 7 days preceding DEXA, and *8*) high-volume hemodiafiltration or hemodialysis during the preceding 4 h.

#### Anthropometry.

Body weight was measured at ICU admission and the day of body composition assessment by means of an electronic scale (ARJO, Gloucester, UK) with an accuracy of ±100 g. Body weight was corrected according to the weight of monitoring and support devices that could interfere with DEXA measurements (see below). We also calculated body weight shift between ICU admission and the day of body composition assessment as follows: Δweight (kg) = body weight at the day of body composition − body weight at ICU admission. Using a measuring tape, height was determined with the patient lying in a supine position, as were right arm (A), trunk (T), and right leg (L) lengths and the mean circumferences of each, as described previously (48).

#### BIA.

We used the multiple-frequency bioimpedance analyzer SFB7 (Impedimed, Eight Mile Plains, Australia), which has a 200-μA alternating current at 4–1,000 kHz. The impedance range is 10–1,100 Ω, with an accuracy of ±1% (50–1,100 Ω) to ±5% (<50 Ω). The device has a tetrapolar set of leads, which are attached to self-adhesive skin electrodes placed on the right hand (wrist next and dorsal surface, 1 cm proximal to the middle knuckle) and foot (ankle at the level of the protruding bones on the sides of the ankle and at the base of the toes, 1 cm proximal to the joint of the 2nd toe). Before measurement, patients were in a fully supine position, with their arms lying relaxed at their sides but not touching the body and thighs separated. The resistance R and the reactance Xc were measured directly in Ω at 5 kHz and 1 MHz. The SFB7 device is coupled with an onboard computer utilizing automatic time delay (an error caused by the speed at which electrical information is transferred through a conductor) corrections and rejection limits for resistance R and reactance Xc. Only ECW is conductive at low frequency (5 kHz), whereas high frequency (1 MHz) allows the electric current to pass through the ECW and intracellular water (ICW) compartment (4, 11, 43). Three consecutive measurements were obtained, and the R and Xc means were computed. The impedance Z was calculated as follows:

Resistance R and impedance Z refer to the opposition of an object to direct and alternating current, respectively. In addition, the reactance caused by the resistive effect due to the capacitance produced by tissue interfaces and cell membranes becomes negligible at 5 KHz and 1 MHz, and Z becomes equal to R (11). Accordingly, R_{1} and R_{2} were obtained at low frequency (5 KHz) and high frequency (1 MHz), respectively. We normalized R_{1} and R_{2} by the square of the body height (H) because H^{2}/R_{1} is a linear function of ECW and H^{2}/R_{2} is correlated with FFM and BCM (4, 18, 43). ECW was calculated as follows (5):
*K*_{ECW} is a factor related to body geometry, density, and resistivity as follows:
^{−3} kg/cm^{3} that is 1.05 kg/l), *ρ*_{ECW} is ECW resistivity (40.5 Ω/cm for men and 30 Ω/cm for women), and *K*_{B} is a coefficient accounting for body height-to-limb geometry (5, 12). *K*_{B} was calculated as follows:
_{A}, C_{T}, and C_{L} represent segmental arm, trunk, and leg circumferences (cm), respectively, and L_{A}, L_{T}, and L_{L} are the lengths of those segments (cm) (5). ICW was calculated as follows (5, 29):
*ρ*_{TBW} is total body water resistivity, *ρ*_{ECW} is ECW resistivity, *ρ*_{ICW} is ICW resistivity (273.9 Ω/cm for men and 264.9 Ω/cm for women), and R_{ICW} is intracellular resistance (Ω). R_{ICW} was calculated as follows (29):

#### Dual-energy X-ray absorptiometry.

Body composition was measured with a Hologic QDR 4500W apparatus (Waltham, MA). This device was calibrated with an external dedicated phantom before each patient's measurement. Patients were monitored with nonmagnetic EKG skin electrodes. The whole body was then scanned twice consecutive times. Briefly, this device measures the differential attenuation of 2 different energy-level X-rays as they pass through the body, enabling determination of Mo content and soft tissue mass. Then, the calibration procedure allows partition of soft tissue mass into fat and nonfat lean body mass. Since the monitoring and support devices used for ICU patients could interfere with DEXA measurements, we conducted a preliminary study to evaluate their effects by using a calibrated whole body phantom (weight: 28 kg). Ten DEXA measurements were repeated without and with a typical set of monitoring and support devices (endotracheal and nasogastric tubes, respiratory circuit, perfusion, intravascular catheters, urinary catheter, nonmagnetic EKG skin electrodes) placed on the calibrated phantom. These devices significantly increased nonfat lean body mass by 1.0 ± 0.2 kg (14.6 ± 0.2 vs. 13.6 ± 0.2 kg, *P* < 0.0001 by repeated ANOVA) and decreased FM by 0.7 ± 0.2 kg (13.1 ± 0.2 vs. 13.8 ± 0.2 kg, *P* < 0.0001). The same analysis conducted with a Mo phantom showed that monitoring and support devices did not significantly alter DEXA determination of Mo (0.7 ± 0.01 kg without vs. 0.7 ± 0.02 kg with devices, *P* = 0.09). Thus, DEXA-derived FM and FFM (non-fat lean body mass + Mo) measurements were corrected accordingly for each patient.

#### BCM calculation, hydration of FFM, and BCM/FFM estimation.

Cohn's model (11, 46, 47) expresses FFM as the sum of BCM, extracellular fluid (ECF), and extracellular solids (ECS) at the cellular level: FFM = BCM + ECF + ECS. Assuming ECS = 0.73 × Mo and ECF = ECW/0.98 in a developed four-compartment cellular level DEXA model (39), BCM can be calculated with the equation BCM = FFM − (ECW/0.98) − (0.73 × Mo), where FFM and Mo were assessed by DEXA and ECW was assessed by BIA at 5 kHz. Because several pathological factors that affect water distribution may influence the metabolically active portion of FFM, we calculated the hydration of FFM (∼73% in healthy subjects) as follows (45): (ECW + ICW)/FFM. We also calculated the BCM/FFM ratio (a determinant of REE) for comparison with the simplified model established by Wang et al. (46) in healthy adults: BCM/FFM = 1.429/[1.569 + 1.16 × (ECW/ICW)]*.*

#### Patient data.

We recorded the following. Nutritional status at admission was considered to be malnourished when BMI was <19 kg/m^{2} or weight was <90% ideal body weight (30, 44) [at 24 h post-ICU admission: simple acute physiology score (SAPS) II; at inclusion: sequential organ failure assessment (SOFA) score and patient characteristics]. Before BIA and DEXA measurements, two metabolic parameters, body temperature, and minute ventilation were also recorded. Body temperature (eardrum) was measured electronically, and minute ventilation was assessed with the Evita 4 respiratory device (Dräger Medical, Antony, France).

#### Statistical analysis.

Results are expressed as numbers (%), means ± SD, or medians (ranges) for data with nonnormal distributions. *P* < 0.05 was considered significant. SAS statistical software (release 9.1; SAS Institute, Cary, NC) and R program (release 2.10.1, 2009; The R Foundation for Statistical Computing, Vienna, Austria) were used for all analyses. Because the BCM value of critically ill patients and its predictors were not known, we applied Maxwell's significance rule (23), which states that a sample size close to 50 patients is required to achieve acceptable statistical power. The repeatability of the DEXA and BIA measurements was examined on Bland-Altman plots (not shown), and the measurement error (S_{w}) was quantified in terms of its standard deviation, which was estimated from replicates, as within-subjects standard deviation (square root of the mean within-subjects variance). According to Bland and Altman (3), the difference between a subject's measurement and the true value is expected to be <1.96 S_{w}, and the difference between two measurements for the same subject is expected to be <2.77 S_{w} for 95% of paired observations. Agreement between measured (electronic scale) and DEXA-calculated (FFM + FM) body weight was also assessed by Bland-Altman analysis. The mean bias in the Bland-Altman analysis represents the degree of systematic difference between measurement methods. A bias of zero would represent perfect agreement between methods. The limits of agreement between methods are defined as the mean difference ± 2 SD. To derive the best-fitting multiple regression equation to predict BCM, we determined the functional form of the relationship between continuous covariates and BCM using fractional polynomials (FPs) (36, 38) and performed bootstrap resampling with 1,000 replications to obtain the most stable model (10, 36). Briefly, the FPs are functions with one or two terms of the form X^{p}. FPs of degree 1 (FP1) are defined as β_{0} + β_{1}X^{p}, and FPs of degree 2 (FP2) are defined as β_{0} + β_{1}X^{p1} + β_{2}X^{p2}*.* The best power transformation X^{p} for FP1 (X^{p1} and X^{p2} for FP2) is found with the power *p* (the powers *p*_{1} and *p*_{2}) chosen among the set (−2, −1, 0.5, 0, 0.5, 1, 2, and 3), where X^{0} denotes ln(X). For example, FP1 with *p* = 1 corresponds to the simple linear model β_{0} + β_{1}X, whereas FP2 with *p*_{1} = 3 and *p*_{2} = 0 corresponds to the model β_{0} + β_{1}X^{3} + β_{2}ln(X). If *p*_{2} =*p*_{1}, the repeated-powers model is defined as β_{0} + β_{1}X^{p1} + β_{2}X^{p1}ln(X). We used the mfp package in R (R Development Core Team, 2004) to build multivariable fractional polynomial (MFP) models. The model selection procedure is a form of backward elimination that starts from a most complex permitted FP model and attempts to simplify it by reducing the degrees of freedom [“RA2” procedure proposed by Sauerbrei et al. (38)]. When the association between the covariate X and the outcome is not significant, X is omitted in the model. Bootstrapping is a resampling technique to derive robust estimates of standard error of an estimator (10). The original data set is randomly sampled with replacement many times, with each sample (replication) containing the same number of observations as the original data. We used 1,000 replications for reliability and performed all analyses using these 1,000 bootstrap samples. For univariable analysis, a MFP model was performed for each variable of clinical interest on each of the 1,000 bootstrap samples. For the multivariable analysis, we considered the variables selected in >20% of the replications in univariable analysis and the trunk length and sex, which were considered physiologically important. To avoid multicollinearity and to take into account the small size of the study sample, highly correlated covariates were excluded, and the degree of power of the FPs was reduced to the degree selected most in univariable analysis. We chose the final clinical model as the most frequently selected model (same variables and same functional form for the 1,000 bootstrap replications) considered as the most robust model. The goodness of fit was assessed using standard diagnostic plots, and the adjusted *r*^{2} was calculated for the original sample data. Once that step was completed, the BIA covariates were added into the best clinical model, and the goodness of fit was again checked to evaluate the improvement of fit. The 95% confidence intervals (95% CI) of the regression coefficients and the adjusted *r*^{2} of the final model were calculated using the 1,000 bootstrap replications of subjects with bias correction.

## RESULTS

Over a 5-mo period, the sample population consisted of 50 patients intubated and mechanically ventilated for >24 h and ≤7 days and who could be transported to the DEXA room. Among the 50 included patients, DEXA measurements failed in one patient because of technical problems. Demographic and descriptive data of the 49 patients eligible for analysis are summarized in Table 1.

#### Body composition.

Anthropometrics, BIA, and DEXA measurements are listed in Table 2. Measurement error, accuracy, and repeatability of R, Xc, FM, and FFM are given in Table 3. The differences between two measurements for the same subject were estimated to be <5.9 Ω for R_{1}, 13.3 Ω for R_{2}, 11.8 Ω for Xc_{1}, 8.2 Ω for Xc_{2}, and 1.1 kg for FM and FFM. Bland-Altman analysis showed a good agreement between measured (electronic scale) and DEXA-calculated body weight with a mean bias of 1.1 kg and clinically relevant limits of agreement (Fig. 1, *A* and *B*), indicating acceptable accuracy of the measurements. The coefficient linking body height to limb geometry (*K*_{B}) was 6.9 ± 1.7, and the factor accounting for body geometry, density, and resistivity (*K*_{ECW}) was 0.39 ± 0.08; resulting ECW and ICW values were 24.8 ± 8.0 and 19.3 ± 10.1 kg, respectively, for the 49 studied patients. Hydration of FFM was 89.4 ± 23.3%. BCM calculated according to the four-compartment cellular level was 22.8 ± 5.4 kg. Figure 2 depicts the proportion of FFM as BCM according to the ECW/ICW ratio and the predicted values for healthy subjects from the nonlinear equation developed by Wang et al. (46). Median difference between predicted values for healthy subjects and observed values of BCM/FFM in critically ill patients was −0.014 (1st quartile to 3rd quartile: −0.096 to 0.034).

#### Clinical regression model.

Univariable FP analysis to predict BCM (Table 4) indicated that leg circumference, arm circumference, height, and weight were selected in >80% of the bootstrap replications, whereas SOFA, Δweight, trunk length, and sex were selected at 47, 35, 19, and 17%, respectively. The most common relationship with BCM was linear except for SOFA (1st degree) and trunk length (2nd degree). Arm circumference was highly correlated with leg circumference (Spearman correlation coefficient *ρ* = 0.59), and weight was highly correlated with most of the studied covariates (*ρ* >0.50). Consequently, only leg circumference, height, SOFA, Δweight, trunk length, and sex were considered for the multivariable analysis. MFP analysis to predict BCM indicated that leg circumference and height were included simultaneously in >70% of the replications. Trunk length, Δweight, SOFA, and sex were selected with these two variables in 45, 41, 29, and 7% of replications, respectively. The final model (the most frequent model considered as the most robust) included linear relationship between the variable leg circumference, height, trunk length, and Δweight as follows: BCM (kg) _{(95% CI)} = 0.266_{(0.162, 0.373)} × height + 0.287_{(0.167, 0.501)} × leg circumference +0.305_{(0.183, 0.458)} × Δweight − 0.406_{(−0.574, 0.373)} × trunk length − 13.52_{(−28.00, 2.42)}. This model explained 46% of the BCM variance of the 49 studied patients (adjusted *r*^{2}). Figure 3*A* depicted predicted vs. observed BCM values. The difference between the BCM predicted with our clinical model and the BCM calculated according to the four-compartment cellular level (absolute error) varied from −6.92 to 8.13 kg, with a mean absolute error of 0.21 kg (95% CI: −0.86, 1.29) (Fig. 3*B*). The limits of agreement within which 95% of absolute differences were expected to lie were −7.28 and 7.71 kg. In our sample, 40 patients (81.6%) had an absolute error lower than ±5 kg. The absolute error divided by the BCM calculated according to the four-compartment cellular level (relative error) varied from −26.54 to 69.85% with a mean relative error of 3.82% (95% CI: −1.88%, 9.52%). The limits of agreement within which 95% of relative differences were expected to lie were −35.89 and 43.53%. In our sample, 37 (75.5%) and 44 patients (89.8%) had a relative error lower than ±10 and ±20%, respectively.

#### BIA measurements and BCM prediction.

Univariable FP analysis to predict BCM (Table 4) indicated that the BIA covariates height^{2}/R1 and height^{2}/R_{2} were selected in 23 and 41% of the bootstrap replications, respectively, and the most common relationship was linear. When these two covariates were candidates in the multivariable clinical model, both of them were excluded in 77% of the replications. The model, including clinical and both BIA covariates (equation not shown), was retained in 13% of replications and did not explain the BCM variance more than the model, including only clinical covariates (adjusted *r*^{2} = 0.44).

## DISCUSSION

The BCM of 49 critically ill patients can be estimated from a few bedside variables with an error lower than ±20% in 90% of patients: height, trunk length, leg circumference, and Δweight, which reflects the shift in body water following disease-associated injury. Simple and regular BCM estimation might improve efficacy of nutrition intervention in the most severely ill patients. Indeed, BCM may represent a valuable biomarker of nutritional status impacting on the outcome of severe disease states (14, 37, 47). To our knowledge, this is the first study to develop BCM estimation from DEXA and BIA measurements in critically ill patients. To remain within the scope of this study, we paid particular attention to applying multiple exclusion criteria to avoid erroneous DEXA and BIA values. Consequently, some of the patients with abnormal body geometry or outliers in BMI were excluded, but the study group still represented severe illness as measured SAPS II and/or SOFA score. As expected, our patients experienced dramatic increases of hydration of their FFM since their (ECW + ICW)/FFM ratios were as high as ∼90%, whereas their ICW decreased by ∼15% compared with healthy subjects (11, 46). Similar body water redistribution was established by Finn et al. (15) in 20 septic and trauma ICU patients, confirming the accuracy of our BIA measurements. Nevertheless, body weight can be estimated accurately by DEXA in healthy humans with a precision of 0.5 kg, whereas we showed in our critically ill patients a mean bias of 1.1 kg between weight measured by electronic scale and weight calculated with the body compartments. This discrepancy probably reflects the combination of dramatic fluid inflation and patient repositioning and movements (<30% were under sedatives or morphine) during the DEXA measurements, which underestimates determination of Mo content and FM (34, 41).

Few authors have investigated BCM in ICU, predominantly in surgical patients and by using specific population regression models derived from BIA measurements with inconsistent precision (9, 19). Indeed, BIA measurements are based on the geometry and the specific resistivity of the conductor. In accord with that assumption, the arms and legs contribute to 47 and 50%, respectively, of whole body resistance, whereas they contribute only 4 and 17% to body weight (6, 17, 27). It was also estimated that skeletal muscles consumed 16–30% of basal oxygen consumption (37). In contrast, the trunk contains 50% of the body mass and is responsible for 70% of the REE due to its visceral components, but it contributes only 5–12% of whole body resistance (6, 17, 37). Because BCM is the metabolically active component of the body, these findings may explain part of BIA-associated misestimation. Moreover, excess ECW is preferentially contained in the trunk and worsens overestimation of total body water, FFM, and BCM (16, 17). Therefore, BIA alone should be restricted to ECW and ICW estimations, but when combined with DEXA body composition it can be estimated in fluid overload (27). Herein, we showed a clinically acceptable agreement between measured and DEXA-calculated body weight, confirming that DEXA is accurate for estimating body composition in critically ill patients. In healthy individuals, trunk lean tissue mass is a better predictor of REE than peripheral lean tissue mass (24). In agreement with previous data (24, 31, 37), our BCM prediction was dependent on trunk length, thereby emphasizing the contribution of the visceral component to the BCM.

A linear correlation exists between oxygen consumption and BCM in healthy and critically ill surgical patients (19). However, there is a paradoxical rise in the REE/BCM ratio in malnourished patients, resulting in a curvilinear relationship between the two parameters (35, 37). This is due to preferential wasting of skeletal rather than visceral BCM (36). The relative increase in REE may accentuate the energy deficit and ICU morbidity in malnourished patients (13, 37). Herein, we observed no relationships between denutrition at ICU admission and the size of BCM. In addition, our patients' observed BCM were somewhat similar to those of healthy adults in the same age group (11). No major BCM modification could be explained by the relatively short hospital and ICU stays of our patients.

BCM and FFM are considered metabolically active. Energy production rates per unit of metabolically active tissue are not constant but vary as a function of body size (20). Therefore, patients weighing less have higher REE per kilogram of BCM and FFM, and their REE/FFM or REE/BCM ratios are higher than those of heavier patients (20). For healthy subjects, the BCM/FFM ratio is one of the major determinants of REE and is dependent on water distribution, especially ECW/ICW (46). Several other factors may influence BCM/FFM, such as sex, age, and adiposity (46). In critically ill patients, we found that BCM/FFM differed from the cellular body composition level modeling developed by Wang et al. (46) for healthy subjects, which showed that higher ECW/ICW is associated with smaller BCM/FFM. ECW inflation occurring in critically ill patients might also explain the discrepancy between BCM/FFM in healthy subjects and BCM/FFM observed in our study. In critically ill patients, ECW inflation may worsen the heterogeneity of the metabolically active tissues in FFM, thereby increasing REE variability expressed as a function of active tissue mass (22, 24). Finally, BCM should be considered a better indicator of the metabolically active tissues and a better nutritional biomarker than FFM in critically ill patients because of the body water redistribution following disease-associated injury. In this context, regular BCM/weight estimation at bedside might improve efficacy of nutrition intervention in the most severely ill patients. Indeed, BCM interacts with energy stores, energy intake, and disease progression throughout the ICU stay (11, 15, 19).

According to our selection criteria, normal body geometry is required to calculate BCM with this predictive method. Indeed, an overestimation of the quality of the prediction is possible because the equations were developed in a specific population, therefore limiting its generalization to other populations. By contrast with the simplicity of the proposed method compared with the available ones, another possible limitation of our predictive method is its moderate accuracy at the individual level. Finally, although our results suggest that BCM can be predicted from a few bedside variables in the most critically ill patients, further prospective validation is required in a larger, more varied critically ill population to confirm that BCM can be estimated from static (height, trunk length), less stable (leg circumference), and dynamic biometric variables (Δweight).

## DISCLOSURES

The authors have no potential conflicts of interest, financial or otherwise, to disclose.

## AUTHOR CONTRIBUTIONS

M.S., G.M., P.H., J.-Y.F., and C.F. did the conception and design of the research; M.S. and C.F. performed the experiments; M.S., F.G., and C.F. analyzed the data; M.S., E.P., F.B., and C.F. interpreted the results of the experiments; M.S. and C.F. prepared the figures; M.S. and C.F. drafted the manuscript; M.S., F.G., G.M., E.P., F.B., P.H., J.-Y.F., and C.F. edited and revised the manuscript; M.S., F.G., G.M., E.P., F.B., P.H., J.-Y.F., and C.F. approved the final version of the manuscript.

## ACKOWLEDGMENTS

We thank Noël Lucas, Pascaline Aucouturier, Ludovic Trinquard, and Jean-François Leforestier (Clinical Research Unit, European Georges Pompidou Hospital) for coordination of the study and data management, Nicolas Weiss, Jean-Marc Tadie, and Audrey Imbert (Medical Intensive Care Unit, European Georges Pompidou) for assistance during the patients' inclusion. We also thank Juliette Djadi-Prat (Clinical Research Unit, European Georges Pompidou Hospital) for the help in R programming. We appreciate the diligent revision of the manuscript by Dr. Stephen J. Taylor of the Department of Nutrition and Dietetics, Frenchay Hospital, Bristol, UK.

- Copyright © 2012 the American Physiological Society