595227
research-article2015
PENXXX10.1177/0148607115595227Journal of Parenteral and Enteral NutritionEarthman
Tutorial
Body Composition Tools for Assessment of Adult
Malnutrition at the Bedside: A Tutorial on Research
Considerations and Clinical Applications
Journal of Parenteral and Enteral
Nutrition
Volume 39 Number 7
September 2015 787­–822
© 2015 American Society
for Parenteral and Enteral Nutrition
DOI: 10.1177/0148607115595227
jpen.sagepub.com
hosted at
online.sagepub.com
Carrie P. Earthman, PhD, RD, LD1
Abstract
Because of the key role played by the body’s lean tissue reserves (of which skeletal muscle is a major component) in the response to
injury and illness, its maintenance is of central importance to nutrition status. With the recent development of the Academy of Nutrition
and Dietetics/American Society for Parenteral and Enteral Nutrition diagnostic framework for malnutrition, the loss of muscle mass has
been recognized as one of the defining criteria. Objective methods to evaluate muscle loss in individuals with acute and chronic illness
are needed. Bioimpedance and ultrasound techniques are currently the best options for the clinical setting; however, additional research
is needed to investigate how best to optimize measurements and minimize error and to establish if these techniques (and which specific
approaches) can uniquely contribute to the assessment of malnutrition, beyond more subjective evaluation methods. In this tutorial,
key concepts and statistical methods used in the validation of bedside methods to assess lean tissue compartments are discussed. Body
composition assessment methods that are most widely available for practice and research in the clinical setting are presented, and clinical
cases are used to illustrate how the clinician might use bioimpedance and/or ultrasound as a tool to assess nutrition status at the bedside.
Future research needs regarding malnutrition assessment are identified. (JPEN J Parenter Enteral Nutr. 2015;39:787-822)
Keywords
body cell mass; nutrition status; malnutrition; fat-free mass; lean body mass; muscle mass; muscle loss; skeletal muscle; intracellular
water; bioimpedance; spectroscopy; impedance ratio; phase angle
There has been long-standing interest in the assessment of lean
tissue as a key parameter of nutrition status. Specifically, lean
tissue mass (of which skeletal muscle is a major component)
plays a central role in the body’s ability to respond to acute and
chronic illness by serving as a vitally important reservoir of
amino acids that can be redirected to the tasks of injury repair
and the immune response when needed.1,2 Albeit highly functional, the process of protein catabolism under these circumstances leads to loss of lean tissue. The loss of lean tissue and
overt sarcopenia (defined as loss of muscle mass and strength3)
that often occur in the face of chronic and acute illness carry significant ramifications in terms of clinical outcomes, including
increased incidence of infections,4,5 increased length of stay,6 and
increased morbidity and mortality.7–9 Indeed, we now have published cutpoints defining sarcopenia by several body composition assessment methods (see Table 1)10–13; these are also used as
part of the defining characteristics of cachexia, a term used to
describe the significant weight loss, protein catabolism, and
muscle and fat tissue loss that occur due to underlying disease
processes.14 Furthermore, with the recent publication of the
Academy of Nutrition and Dietetics (AND)/American Society
for Parenteral and Enteral Nutrition (A.S.P.E.N.) consensus
statement15 defining diagnostic criteria for malnutrition, there
has been renewed focus on the assessment of muscle loss as a
key component of nutrition status. As the consensus
malnutrition diagnostic framework has been developed, much
attention has been paid to the assessment of muscle loss through
physical examination techniques16; however, the subjective
nature of physical examination is a potential limitation. One of
the major limitations of any subjective method is poor reliability
due to interobserver and intraobserver variability. Although
intensive and standardized training programs may be able to
minimize that variability, the potential for error remains high as
the technique is broadly applied across medical centers.
Furthermore, lean tissue loss can precede overt weight loss17 and
may be masked by excess extracellular water (ECW),18 thus
making it difficult to detect through visual techniques. A recent
From the 1Department of Food Science and Nutrition, University of
Minnesota–Twin Cities, St Paul, Minnesota.
Financial disclosure: C.P.E. received loaner bioimpedance devices and
a small gift grant in support of her research from Bodystat Ltd (UK), a
company that manufactures bioimpedance devices.
Received for publication February 6, 2015; accepted for publication June
17, 2015.
Corresponding Author:
Carrie P. Earthman, PhD, RD, LD, Professor, Department of Food
Science & Nutrition, University of Minnesota, Food Science & Nutrition
225, 1334 Eckles Ave, St Paul, MN 55108-6099, USA.
Email: [email protected]
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
788
Journal of Parenteral and Enteral Nutrition 39(7)
Table 1. Commonly Referenced Cutpoints Defining Sarcopenia by Various Body Composition Assessment Techniques.14,17,a
Method
Appendicular skeletal muscle index by DXA
(Baumgartner et al10),b kg/m2
Appendicular lean muscle index by DXA
(Delmonico et al90),c kg/m2
Lumbar skeletal muscle index by CT (Prado
et al11),d cm2/m2
Skeletal muscle index (%weight) by SF-BIA
(Janssen et al12),e % Skeletal muscle index (height) by SF-BIA
(Janssen et al13),f kg/m2 Reference Cutpoint Males
Reference Cutpoint Females
<7.26
<5.45
<7.25
<5.67
<52.4
<38.5
Class II: <31
Class I: 31–37
High risk: ≤8.50
Moderate risk: 8.51–10.75
Class II: <22
Class I: 22–28
High risk: ≤5.75
Moderate risk: 5.76–6.75
CT, computed tomography; DXA, dual-energy X-ray absorptiometry; SF-BIA, single-frequency bioelectrical impedance analysis.
a
This is not an exhaustive list of cutpoints from the literature.
b
Appendicular skeletal muscle index calculated as the sum of arm and leg lean soft tissue mass divided by height in meters squared, and cutpoints are 2
SD below the mean values from young adults in the Rosetta Study.
c
Appendicular lean muscle index calculated as the sum of arm and leg lean soft tissue mass divided by height in meters squared, and cutpoints are based
on sex-specific lowest 20% of study group in the Health, Aging, and Body Composition Study.
d
Lumbar 3 (L3) skeletal muscle index was calculated as the total area of L3 skeletal muscle in centimeters squared, divided by height in meters squared,
and cutpoints are based on mortality using optimum stratification.
e
Skeletal muscle index (%weight) calculated by the Janssen SF-BIA skeletal muscle mass equation developed from magnetic resonance imaging data137:
Skeletal muscle mass (kg) = [(Height2/R × 0.401) + (Sex × 3.825) + (Age × –0.071)] + 5.102, where height is in centimeters, R is resistance in ohms, and
for sex, male = 1 and female = 0, divided by weight in kilograms multiplied by 100. Cutpoints for class I sarcopenia are 1–2 SD below the mean skeletal
muscle index (SMI) percent values from young adults from the National Health and Nutrition Examination Survey III (NHANES III), and cutpoints for
class II sarcopenia are 2 SD below the mean SMI percent values from young adults.
f
Skeletal muscle index (height) calculated by the above Janssen skeletal muscle mass SF-BIA equation, divided by height in meters squared, and
cutpoints are based on high and moderate risk for physical disability in adults aged >60 years from the NHANES III (1988–1994).
article by Sheean et al19 would attest to this. They reported that
60% of individuals with respiratory failure who were deemed
normally nourished at intensive care unit (ICU) admission by
subjective global assessment20 (administered by experienced clinicians) were actually sarcopenic as defined by computerized
tomography (CT). Approximately 33% of individuals who were
misclassified were overweight or obese. Although there is no
clear consensus yet on how best to define and identify it,21,22 the
problem of sarcopenia in the presence of obesity has significant
clinical implications.7,9,23 For all these reasons, a method to
objectively and reliably evaluate lean tissue loss at the bedside is
highly desirable but has been somewhat elusive.
In the first part of this tutorial, basic terminology around the
assessment of lean tissue body composition is introduced. In
the second part, key concepts around validity of field (which
we will henceforward term bedside) techniques, as they compare with reference methods, are presented. Subsequently, the
statistical methods that are typically used in the validation of
bedside techniques are reviewed to gain perspective on the
interpretation and application of the research literature. In the
third and fourth parts, body composition assessment methods
that are most widely available for practice and research in the
clinical setting are subsequently discussed, including multiple
dilution, dual-energy X-ray absorptiometry (DXA), and CT
(reference methods), as well as bioimpedance and ultrasound
(bedside methods). In the fifth part, clinical cases are used to
illustrate how the clinician might ultimately use bioimpedance
and ultrasound to assess lean tissue and/or nutrition status at
the bedside. Finally, the sixth part addresses gaps in the
research literature around malnutrition assessment where additional clinical research studies are needed.
Introduction to Lean Tissue Terminology
Multiple terms to describe the lean tissue compartment of the
body, based on various conceptual models of body composition, have been well described elsewhere.24,25 At the most basic
level, the 2-component model of body composition describes
the body as the sum of fat mass and fat-free mass (FFM). FFM
thus is a rather broad term that includes skeletal muscle, nonskeletal muscle, organs, connective tissue, total body water
(TBW, including ECW and intracellular water [ICW]), and
bone. FFM can be estimated from TBW measurements using
the following equation:
TBW ( kg ) /0.73 = FFM ( kg ) ,based on the
standard hydration constant for FFM.26
This hydration constant has been shown to be remarkably stable
under euhydrated, healthy conditions26 but varies under clinical
conditions where hydration may be altered and has been shown
to be somewhat higher (>0.75) in individuals with extreme obesity.27,28 The lean body mass (LBM), which is more appropriately termed lean soft tissue (LST) when measured by DXA, is
a slightly more specific compartment that includes all of the
aforementioned components except for bone.29 The body cell
mass (BCM) is a term originally described by Moore and
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
789
Boyden30 as the total mass of cells in the body that consume
oxygen and produce work; it is the nonfat cellular portion of
tissues, of which the primary components are skeletal muscle,
organ tissue mass, blood, and the brain.31 As such, the BCM has
been considered by many to be the primary lean tissue compartment to assess nutrition status and to serve as a key target for
nutrition interventions.32,33
There is no direct way to quantify BCM, but there are several ways it can be estimated, including from total body nitrogen determined by neutron activation analysis (NAA), total
body potassium through total body potassium counting (TBK),
and ICW measured by multiple dilution.30,32,34–36 Given that
NAA exposes individuals to radiation, and both NAA and TBK
are highly technical and expensive, ICW through multiple dilution is one of the more accessible reference methods to clinical
researchers and depends on the underlying observations that a
large proportion of the BCM is represented by the water content of cells, and acute changes in body protein occur primarily
within cells, and these changes are generally accompanied by
changes in ICW.37,38 Nevertheless, despite differences in their
level of specificity to core functional processes, all of the
aforementioned lean tissue compartments, including FFM,
LST, and BCM (and/or ICW itself), may be measured as the
target variable in methods validation, as well as nutrition
assessment and intervention studies, depending on the available body composition assessment methods.
Validation of Bedside Techniques to
Assess Lean Tissue in Clinical Settings
Several excellent reviews describe the various methods that are
available to evaluate lean tissue compartments.29,39,40 Bedside
techniques need to be validated against more established reference techniques before they can be applied with confidence in
the clinical setting. It should be noted that all methods to evaluate body composition are indirect, requiring various assumptions that may or may not hold true in acute and/or chronic
illness, and none is completely free from error. Thus, the term
gold standard, which has been used in the past, has been
replaced with the term reference to describe the more established techniques against which bedside methods are validated.
For the purposes of this discussion, the focus is primarily on
bioimpedance and ultrasound techniques as 2 promising currently available bedside assessment options. The most common reference methods against which bioimpedance and
ultrasound have been compared in clinical studies include multiple dilution, DXA, and CT. Although magnetic resonance
imaging (MRI) is also an important reference method for validation studies, it is generally less widely available to clinician
researchers and thus will not be discussed in this tutorial.
However, the reader is referred to a recent review that includes
discussion of MRI.29 Before these methods are briefly
described, however, it is important to understand the key terminology and statistical approaches used in methods validation.
Basic Terminology Regarding Validity
There are several important concepts to keep in mind as we
talk about the validation of bedside methods. At the heart of it,
the effort to prove the validity of a bedside method is undertaken to establish that method as an acceptable alternative to a
more expensive and/or technical and, usually, less widely
available technique. Theoretically speaking, the definition of
validity is the extent to which a measurement actually measures the “true” value of a particular body composition parameter; however, given that the “true” quantification of body
composition components may only be known through cadaver
analysis, and there are no completely error-free “gold standards” among body composition assessment methods applied
to living humans, it is virtually impossible to determine that
“true” value. The best we can do is to compare the bedside
method with the best reference method that is available to measure the compartment of interest, and if the methods produce
sufficiently comparable measures with acceptable measurement error, we can then call the bedside method valid. Thus, in
most of the body composition literature, the term validity is
used more generally and encompasses such concepts as precision, bias, and accuracy, among others. In general, measurement error can be conceived of as falling into 1 of 2 categories:
(1) random error due to biological variation and other factors
and (2) systematic error that occurs from constant fixed and/or
proportional bias. From a practical standpoint, investigators
aiming to validate a device or technique should strive to evaluate (1) how close do the values obtained by the bedside method
agree with the values obtained by the reference method, (2)
how often are the values within an acceptable range of difference, and (3) whether there is a consistent tendency for the
bedside method to over- or underestimate the body composition compartment of interest compared with the reference
method. To answer these questions, it is important to know the
precision of the reference method and to evaluate the precision,
accuracy, and bias of the bedside method.
Precision and reliability. The term precision as it applies to
body composition measurement is typically used to describe
the degree of agreement among repeated measurements by a
particular assessment method (ie, how well does a particular
method produce the same result on multiple occasions). There
is another use of the word that is sometimes seen in the literature, although it deviates from the true definition; precision is
sometimes loosely used to describe how well a bedside technique produces measurements sufficiently close to those produced by a reference method in a group of individuals (ie, how
variable are individual measures between methods, defined in
part by the limits of agreement or 95% confidence interval,
which will be discussed later) but not necessarily taking into
consideration inter- or intraobserver errors that would typically
be used to describe precision in terms of repeatability and
reproducibility. We will assume the first definition when precision is discussed and will indicate when the alternative
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
790
Journal of Parenteral and Enteral Nutrition 39(7)
meaning is intended by using “precision.” Reliability is a term
that is used to describe a method that is considered to have high
precision. Random error will affect the precision and reliability
of a method.41 For a method to be considered reliable, it should
have high repeatability and reproducibility. Repeatability is a
term that refers to the variability between repeated, independent measurements of a particular variable (eg, FFM) that are
made by the same operator or observer using a single device, in
the same individual, under the same conditions. Reproducibility, on the other hand, can be used to describe the variability in
measurements of a particular variable taken by that particular
device operated by different observers in the same institution
or in different institutions. Imprecision has been defined as the
variability of repeated measurements due to intra- and interobserver measurement differences.42 Interobserver measurement
error (ME) refers to the variability in a particular variable by
the same method when measurements are made by more than
one operator. Intraobserver ME refers to the variability in a
particular variable by the same method when more than one
measurement is made by the same operator. ME is typically
referred to as technical error in the anthropometric literature.42
To calculate intraobserver ME for 2 measurements and interobserver ME for 2 operators, the same equation may be used42:
ME = √ ( ∑ D 2 ) / 2n , where D is the difference between
the paired measurements and n is the number
of individuals measuredd.
Because ME can sometimes be positively associated with the
size of measurement, it is advantageous to convert the absolute
ME to a relative ME term, which can be done by calculating
the %ME using the following formula42:
( ME / mean of measurements ) ×100.
Calculating the %ME in this way is a variation on the measurement of the coefficient of variation (CV), which is typically
calculated by dividing the standard deviation (SD) by the mean
of the set of measurements and then converting to a percent (ie,
CV = SD/mean × 100). In fact, it could be said that all expressions of error are ideally described as a percentage relative to
the actual size of the compartment being measured, although it
is important to remember that there will be differences in relative magnitude between various error terms. A method that is
subject to high inter- and intraobserver errors is judged to have
low precision. Simple anthropometry using skinfolds and circumferences is an example of a method that has been shown to
have low precision, with %ME approaching 8% and CVs as
high as 45% for some skinfold measurements.42–44
To summarize, random error (caused by biological variability, as well as errors in the measurement procedure caused by
protocol deviations, random instrument error, operator error,
and environment, among other factors) as it relates to precision
can be expressed in several different ways, but the bottom line
is we are attempting to capture the degree of variability in measurements not due to fixed systematic error. You will see derivations of the same concept applied to calculate other error
terms as we go along, with minor adjustments to the formulae
depending on the statistical approach. Statistical terms that are
commonly used to describe reliability and precision include
test-retest correlation coefficients (same day or between day),
intraclass correlation coefficient (ICC), SD, and CV for a given
set of measurements. ICC values approaching 1.0 would indicate low variability between repeated measures of the same
subject (ie, low ME). Root mean squared error (RMSE) can
also be calculated to estimate random error in a bedside
method, and although there are slight deviations that might be
applied when using multiple regression analysis,45 RMSE
(sometimes termed pure error) can be calculated as follows46:
RMSE = √ [(∑ (Y − X ) ) / n ], where Y is the reference
2
measure, X is the bedside measure, and n is the number
of subjects measured.
Accuracy and bias. The term accuracy is used to indicate the
closeness of agreement in a particular variable between 2
assessment methods (ie, how close are the values of a particular
variable by a bedside method to those generated by an accepted
reference method). The term bias is a general term to describe
the systematic error in a method (ie, the difference between the
measurements made by the bedside method and those made by
the reference method). Bias is typically calculated as the mean
of the differences in a particular measurement variable between
the 2 methods being compared. Fixed (or constant) bias refers
to the type of systematic error that occurs when a method yields
measurements of a particular variable that are consistently
higher or lower than those taken by the reference method. The
method can be highly precise in terms of repeatability and
reproducibility yet still yields values that are predictably and
consistently lower or higher than those made by a reference
method (ie, low accuracy). For example, a single-frequency
bioelectrical impedance analysis (SF-BIA) device shown to
have high precision can be used to generate data applied to an
equation that might have been developed from deuterium dilution data to predict TBW; when that equation-generated TBW
value is converted to FFM using standard assumptions (eg,
TBW/0.73 = FFM) and those values are subsequently compared with FFM values generated by DXA, it can lead to a type
of fixed constant bias (ie, systematic error) termed scaling
error. In this case of high precision but low accuracy, it is possible that the fixed bias might be addressed through the use of a
correction factor. Unfortunately, errors between methods are
rarely that simple to address. Proportional or positional bias
refers to errors that are proportional to the value of the variable
being measured. An example of this type of error is the proportional bias observed in bioimpedance measurements of lean tissue in individuals with extreme obesity. The errors in ICW and
other lean tissue compartments by all bioimpedance techniques
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
791
have been well demonstrated to increase with increasing adiposity.47–49 Thus, the accuracy of a single measurement in a
single individual is impossible to ascertain with 100% certainty,
but it may be inferred from estimates of the precision of the
method as well as from an understanding of the systematic error
of the method.
In methods validation, there are 2 primary ways of thinking
about accuracy, in part based on the statistical methods used to
evaluate it. The first school of thought would argue that a bedside method is accurate if the average of all measurements
taken by that method is close to the average of all measurements taken by the reference method (ie, the mean difference
between methods is close to 0); the closeness of agreement
between methods at the individual level is not taken into consideration. In this line of reasoning, the primary concern is that
a method has low bias (ie, a method could have low precision
and individual values by the 2 methods could be quite different, and the method could still be called accurate as long as low
bias at the mean level is evident). An example of when this
definition of accuracy is most commonly (and probably appropriately) applied is in large-scale epidemiologic studies that
compare measurements of a variable (eg, FFM) by a field
method such as bioimpedance with a reference method such as
DXA, in order to establish bioimpedance as an acceptable
method for describing population-level mean differences in
FFM across different ethnic groups. The statistical approaches
used to establish accuracy based on this definition include correlation and paired t test statistics. We will revisit these methods in the following sections.
The second school of thought would argue that a technique
cannot be considered accurate unless measurements of a particular variable by 2 methods agree closely with one another
across all individuals. The statistical approach typically used to
establish accuracy based on this definition is Bland-Altman
analysis, including calculations of various error terms and limits of agreement (defined as mean difference between methods
±1.96 SD) that can help us to determine if a method is truly
acceptable to replace another more established method for the
individual at the bedside. There are no clear guidelines on what
should be considered an acceptable range for the limits of
agreement to define a method as valid. It is up to the individual
conducting the study to determine what defines acceptable in
terms of the width of the limits of agreement. Therein lay one
of several difficulties in the interpretation of validation studies.
We will get back to Bland-Altman analysis and its application
and interpretation in a subsequent section.
Sensitivity and specificity. Although not typically used to
describe methods comparisons, for our purposes, it is useful to
review what is intended by the terms sensitivity and specificity,
particularly as they pertain to the identification of malnutrition
by a particular technique. Sensitivity can be defined as the percentage of truly malnourished individuals who are correctly
identified as such, calculated as follows:
Sensitivity = ∑ True Positives / (∑ True Positives +
∑ False Negatives) ×100.
Specificity, on the other hand, can be defined as the percentage
of well-nourished individuals correctly identified as not being
malnourished, calculated as follows:
Specificity = ∑ True Negatives / (∑ False Positives +
∑ True Negatives) × 100.50
If a body composition assessment technique is to be useful to
identify individuals with malnutrition, it will need to have high
sensitivity and specificity (>90%, ideally—particularly for
sensitivity).51
Statistical Techniques Used to Evaluate
Agreement Between Methods in Validation
Studies
Multiple statistical approaches can be applied to the problem
of establishing the validity of a particular bedside method compared with a reference method. It is highly recommended that
a combination of approaches be used to evaluate both meanlevel and individual-level agreement. Regression and correlation analyses and paired t tests are the most commonly applied
statistical approaches used to evaluate agreement between 2
body composition assessment methods in terms of mean-level
agreement.
Regression analysis. Simple linear regression and correlation analyses are statistical approaches that have traditionally
been used to evaluate the relationship between 2 different
(eg, bedside and reference) body composition assessment
methods as a way of validating the bedside technique. While
correlation analysis simply tells us the strength of the interrelatedness of 2 sets of data, linear regression evaluates the
nature of the relationship.52 A strong linear relationship and
high correlation between 2 sets of data may be interpreted to
mean “good agreement” between methods, although this conclusion may be faulty. To use linear regression analysis on a
set of data, several underlying assumptions should be met,
including normal distribution of the continuous variable data,
among others. Typically, measurements of a body composition compartment (eg, TBW) by a bedside method (eg, an
SF-BIA equation) are regressed against those measured by a
reference method (eg, deuterium dilution) to create a least
squares regression line describing the relationship between
the 2 methods, reflected by the general equation y = mx + c,
where m is the slope of the line and c is the intercept (ie, the
value of y when x is 0). The 45° line of identity (or line of
equality) is used to represent perfect unity and can be represented by the equation y = x, where the slope is 1 and the line
goes through the origin.
For the purpose of illustration, Figure 1 represents various
scenarios resulting from linear regression analyses applied to
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
792
Journal of Parenteral and Enteral Nutrition 39(7)
Figure 1. Interpretation of linear regression for evaluation of methods agreement. (A) Regression line suggesting good agreement
in hypothetical fat-free mass (FFM) measures by single-frequency bioelectrical impedance analysis (SF-BIA) and dual-energy X-ray
absorptiometry (DXA), with low bias and high precision. (B) Regression line suggesting poor agreement between SF-BIA and DXA
methods, with high fixed bias and high precision. (C) Regression line suggesting poor agreement between SF-BIA and DXA methods,
with low bias and low precision. (D) Regression line suggesting poor agreement between SF-BIA and DXA methods, with high bias
and low precision. Figure reproduced with permission from David Frankenfield.
body composition data. Plots A and C of Figure 1 show the
regression line to be very close to the line of identity, when a
bedside method is either highly precise and unbiased (A) or
not very precise but unbiased (C). However, you would not
usually expect the regression line (Y on X) to go through the
origin and have a slope of 1; in fact, more commonly, measurement errors in X will reduce the slope of the line, such that
the slope will be <1, and the intercept will be >0. The correlation coefficient (eg, the Pearson product-moment correlation
coefficient), referred to as r (along with the P value indicating
statistical significance), is used to reflect the strength of the
association between the 2 methods compared with the line of
identity.41 Interpretation varies depending on the application53; the closer the r value is to 1.0 (+1 or –1), the stronger
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
793
the association (positive or negative, respectively), with values in the range of 0.9–1.0 typically interpreted as strong and
values <0.50 interpreted as weak associations between body
composition methods. Squaring r to get the coefficient of
determination (R2) is useful, in that it can be interpreted
directly as the percentage of variation in the measurements by
one method that is related to the variation in the other.53,54 For
example, an r of 0.99 would yield an R2 value of 0.98, which
could be interpreted as “98% of the variation in TBW predicted by the SF-BIA equation is related to the variation in
TBW measured by deuterium dilution”; such a high value for
r and R2 could further be interpreted to suggest that the BIA
approach might be considered a valid way to estimate TBW,
although other considerations should be made before that conclusion is drawn. For example, the degree of error in the measurement needs to be evaluated. How far the measurements
plotted on the y-axis vary around the regression line is typically described as the degree of error and may be represented
by the term standard error of the estimate (SEE). SEE can be
calculated by taking the square root of the sum of the squared
differences between measured and predicted values divided
by n, the number of pairs of scores, minus 2; this is represented by a very similar equation to that used to compute
RMSE55:
SEE = √ [(∑ (Y − X ) ) / n − 2].
2
It is also useful to evaluate SEE for a bedside method in
terms of percentage error compared with the mean body composition compartment measurement by the reference method;
for example, if we were comparing FFM measured by an
SF-BIA approach against DXA, we would calculate
SEE/ ( mean FFM by DXA ) × 100.
Linear regression analysis has been criticized as a poor
approach to establish the validity of a body composition assessment method, due to several inherent methodological limitations, and if used as the sole method to evaluate agreement
between methods, it will more often than not lead to faulty
conclusions. For example, if you have a high degree of heterogeneity and low precision in your sample, it is quite easy to
obtain a high value of r.56 On the other hand, highly precise,
homogeneous data generated by a bedside method can also
yield a high value of r, despite having substantial systematic
error (eg, high bias).57 The reader is again referred to Figure 1,
which can further illustrate some of these points, based on
hypothetical FFM data generated by a bioimpedance technique
and a DXA instrument. If the bioimpedance method is highly
precise and unbiased, the regression line for the relationship
between that method and DXA might appear as shown in
Figure 1A, with a statistically significant (P < .05) and very
high value for r and R2. On the other hand, the bioimpedance
method might be highly precise but significantly biased (see
Figure 1B), and you could still have a strong correlation signified by a high value for r and R2 with a high degree of statistical significance. Such a situation can easily arise, as you can
imagine when you look at 2 data sets that differ in value by
50% (eg, {1,2,3,4,5,6} and {2,4,6,8,10,12}); comparison of
these data by linear regression would yield a perfect correlation of 1 and an SEE of 0, despite the significant bias present.
Figure 1C illustrates a highly imprecise method that produces
values that on average are unbiased compared with the average
of the values produced by DXA. This kind of result would
meet our first definition of accuracy based on mean-level
agreement but, due to the imprecision of the method, would not
meet our second definition of accuracy based on individuallevel agreement. Yet another possibility, illustrated in Figure
1D, can be seen when a bedside method is both imprecise and
highly biased; in this case, the poor agreement between methods would be reflected by a low correlation coefficient.
The commonly applied least squares linear regression
approach is partly based on the assumption that the reference
method has minimal error compared with the range of the
measurements. Given that there is error in both reference and
bedside methods, this assumption may be faulty. Orthogonal
least squares regression (eg, Deming regression58,59 or
Passing-Bablok regression59–63) assumes measurement errors
are produced by both methods and are independent and normally distributed; this approach has been advocated as an
alternative58,64,65 but has not yet been widely applied in the
body composition assessment field. Finally, concordance correlation coefficient (CCC) and ICC (which is sometimes
applied to evaluate precision, as mentioned earlier) analyses
have been advocated as superior to simple linear regression
analyses, because they yield a value approaching 1.0 only if
there is minimal bias and the paired measurements are in close
agreement. Thus, the CCC not only provides a measure of
association between method variables but also indicates how
close values are to the line of identity. Although CCC has been
advocated to be superior to Pearson’s product-moment correlation because it considers both systematic and random
error,66–68 it has also been criticized for relying on similar
assumptions to those underlying Pearson’s correlation, such
that highly heterogeneous and low precision data can lead to
high r values and erroneous interpretation as good agreement
between methods.56
That said, there may be occasions when some form of
regression analysis is the only procedure that can reasonably
be applied to 2 sets of data. The particular example that comes
to mind is when one is attempting to validate a method that
measures a distinctly different lean tissue compartment from
that measured by a reference method (eg, ultrasound measures
of the upper quadriceps muscle compared with LST measures
by DXA). It would be impossible to compare these directly,
given the magnitude of difference between them, but a strong
correlation would suggest that there might be utility in the
ultrasound measures, particularly if the SEE were relatively
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
794
Journal of Parenteral and Enteral Nutrition 39(7)
small. In this validation scenario, it would be recommended to
take longitudinal measures to ascertain measures of change in
the compartment of interest, assuming that relatively high precision and reliability (ie, minimal random error) in the bedside
assessment method could be achieved. Strong correlation
between measures of change in this case could be interpreted
as good agreement between the techniques. Another situation
that frequently arises in the nutrition assessment field is the
evaluation of a variable or set of variables for the ability to
identify malnutrition. For example, in the bioimpedance field,
select raw data parameters have been evaluated for their ability
to identify malnutrition. The only way that this kind of relationship can be established is through the use of regression
analysis, most commonly taking a multivariate approach.
Nevertheless, in most instances, it is not recommended to rely
solely on the evaluation of the linear association between
methods but rather to use a combination of techniques to evaluate both mean-level and individual-level agreement. Although
analysis of variance (ANOVA) techniques would be most
appropriate when comparing more than 2 methods, our focus
here is on the comparison of two, a bedside method to a reference method, and thus, ANOVA will not be discussed. Paired t
test is another procedure that is frequently used to determine
mean-level agreement between a bedside and reference
method.
Paired t test. The paired t test procedure can be used to test the
difference between the means of 2 sets of measurements on
each of a group of individuals. Put simply, the difference
between each set of measurements is computed, and the test is
applied to determine whether the mean difference is significantly different from zero. Because you are essentially testing
the null hypothesis that the 2 sets of data produce equal means,
achieving statistical significance for this test is interpreted as
“the 2 methods do not produce equivalent values.” If, on the
other hand, statistical significance is not reached, and thus the
null hypothesis is not rejected, the conclusion cannot be
assumed to be that there is no difference between the methods,
only that no difference could be found. The statistical significance level (α) is typically set at 0.05, and the 95% confidence
interval of the differences between the methods (defined as the
mean difference ±1.96 SD) is calculated in the procedure. If
the 95% confidence interval of the differences excludes zero as
a possibility, then we conclude that bias is present in the
method, and the null hypothesis is rejected (P < .05). On the
other hand, if zero is included in the 95% confidence interval,
the null hypothesis is not rejected (P > .05), and it might be
concluded that the methods are not different, although this
could be an erroneous conclusion.
A major limitation of using the paired t test to compare a
bedside method with a reference method for agreement is that
the results are quite difficult to interpret. For example, if statistical significance is reached when a paired t test is conducted, the
finding of difference between methods could be due to random
and/or systematic error. It would be impossible to ascertain
which of these (and to what magnitude) is primarily in evidence. Furthermore, a nonsignificant result from a paired t test
can easily occur when there is equally distributed imprecision
in a set of measurements, with both negative and positive intermethod differences; these differences will cancel each other out
when averages are compared, and the mean will appear to be
close to zero. In fact, one conservative approach to method
comparisons, particularly with small sample sizes, is to set statistical significance (α) at .10 rather than the typical .05. One
might choose to do this to minimize the possibility of making a
type II error (ie, to avoid the error of falsely accepting the null
hypothesis of no difference between the 2 methods).49
One reasonable approach to minimize erroneous conclusions from a paired t test is to apply it in combination with
linear regression and correlation analyses; a high degree of
correlation and a highly nonsignificant P value for the paired t
test would suggest good agreement (at least at the mean level)
between methods. However, to ascertain fixed and proportional bias (ie, systematic error), as well as the level of agreement on the individual level, it is advisable to take additional
steps, including analysis of the data as recommended by Bland
and Altman.57,69,70
Bland-Altman analysis. Although it is not the only methodologic approach that may be taken and is not universally
advocated,71,72 Bland and Altman’s seminal paper in 198670
proposing their method of plotting the differences between
values generated by 2 methods of measurement on the y-axis
against the average of the values produced by the 2 methods on
the x-axis is considered by many to be important to the body
composition assessment field. They did not invent the method,
but they advocated its application to the comparison of medical
devices, laboratory tests, and other clinical techniques to ascertain bias in one method compared with another. One of the
strengths of the Bland-Altman approach is that by calculating
the mean of values measured by the 2 methods, there is recognition that there is random error associated with both methods
(ie, the reference method is not infallible and thus is not a true
“gold standard”). The Bland-Altman method can also be used
to evaluate agreement between replicate measurements by one
method, but this will not be discussed here. When using BlandAltman plots to evaluate agreement between 2 methods, the
mean values by the 2 methods are usually plotted on the x-axis.
On the y-axis, the values for the difference between methods in
the body composition parameter are typically plotted, but difference may also be expressed as a percentage of the mean
body composition parameter values (either the mean value
from both methods or the mean value by the reference method).
Figure 2 provides a series of Bland-Altman plots that might
be constructed based on different hypothetical sets of FFM data
generated by an SF-BIA equation and DXA. The first step in the
process is to construct a scatterplot of the differences between
methods (y) against the averages between methods (x). The
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
795
Figure 2. Interpretation of Bland-Altman Plots for evaluation of methods agreement. (A) Bland-Altman plot showing good agreement
in hypothetical fat-free mass (FFM) measures by single-frequency bioelectrical impedance analysis (SF-BIA) and dual-energy X-ray
absorptiometry (DXA). (B) Bland-Altman plot showing poor agreement in hypothetical FFM measures by SF-BIA and DXA. (C)
Bland-Altman plot showing proportional bias in hypothetical FFM measures by SF-BIA and DXA. (D) Bland-Altman plot showing
fixed bias in hypothetical FFM measures by SF-BIA and DXA. Figure reproduced with permission from David Frankenfield.
next step is to take the mean difference (ie, bias) between methods and calculate the limits of agreement around the bias. This
is done by calculating the mean ± 1.96 SD for the differences
between the methods and drawing a horizontal line corresponding to the mean, to the value at the mean + 1.96 SD, and to the
value at the mean – 1.96 SD. Assuming a normal distribution,
the limits of agreement should encompass 95% of all measured
values; the width of the limits of agreement has been described
by the word “precision” based on the alternative definition of
the word discussed earlier.73 Bland and Altman talked about
how to calculate confidence intervals and SD for the limits of
agreement as a way of evaluating the “precision” of those estimates, but that is beyond our discussion here.
A major criticism of the Bland-Altman technique is that the
decision to accept the new technique, or, in our case, the bedside method as an acceptable alternative to the reference technique, is left entirely up to the subjective judgment of the
evaluator. The ideal bedside method would demonstrate
narrow limits of agreement, around a mean bias of zero. This is
demonstrated in Figure 2A. Far more common, particularly
when it comes to bioimpedance techniques compared with
dilution reference methods, is to observe wide limits of agreement (see Figure 2B) when applying the methods in various
clinical populations. Another important step is to determine if
the differences are significantly correlated with the averages. If
there is a significant correlation, then there is proportional bias.
This can be seen in the example shown in Figure 2C. If, on the
other hand, there is no significant correlation but the mean difference between methods is consistently and significantly less
than or greater than zero, suggesting that the bedside method
consistently produces underestimates or overestimates of the
measured variable compared with the reference method, then
fixed bias is evident (see Figure 2D).
The Bland-Altman method has recently been criticized, and
suggestions have been made on how to optimize its application
and interpretation.74–77 Ludbrook74 proposed that after plotting
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
796
Journal of Parenteral and Enteral Nutrition 39(7)
the differences against the averages, the next step is to evaluate
if the differences are correlated with the averages and, if the P
value is not statistically significant for r, then proceed with
calculating the standard 95% limits of agreement. If, on the
other hand, the averages and differences are significantly correlated, then proportional bias is evident and hyperbolic 95%
prediction limits can be calculated. These recommendations
hold if the scatter of differences is uniformly distributed across
the range of averages. The reader is referred to the article by
Ludbrook74 for additional information on what to do if the scatter of differences increases with the values of the averages and
other contingencies.
One of the major problems in method comparisons is that
both the bedside and reference techniques have a certain degree
of random error associated with them. Even the most precise
methods have some degree of error. The Bland-Altman
approach is good, in that calculating the average between the 2
methods allows for relatively equal weighting between the 2
methods. However, as stated previously, it is up to the clinician
to judge whether the limits of agreement are sufficiently narrow to allow for the bedside method to replace the reference
method. To truly answer that question, it is important to consider the precision of both the reference and the bedside
method. Bland and Altman57 described the importance of having at least 2 measurements by a particular method on each
subject, so that the repeatability coefficient (ie, precision) for
the method could be calculated from the SD of the differences
between pairs of those repeated measurements, reflected by the
equation 2 × SD. They described the repeatability coefficient
as the “difference that will be exceeded by only 5% of pairs of
measurements on the same subject.”57 This would allow for
evaluating not only the limits of agreement between the 2
methods but also the repeatability (ie, precision) for each
method separately. Repeated measures are not always obtained
in studies; Ward75 recently provided a thought-provoking illustration of how a modified version of that approach might be
used when repeated measures are not available to better define
the validity of a method, building from a method described for
techniques to measure cardiac output.76,77 These authors advocated for the calculation of the percentage error (PE) of the
limits of agreement compared with the mean of the measurements as a way of defining a cutoff for acceptability of a
method. In this approach, the “precision” of a method is
defined to be 1.96 (or for simplicity, rounded to 2) times the
CV for that method. Note that the use of the word “precision”
here is not the traditional definition based on CV for repeatability and reproducibility but rather the alternate definition
based on the spread of the individual data from the BlandAltman analysis. Using the Bland-Altman plot shown in Figure
2B, we can see how we might reevaluate a hypothetical set of
validation data. To do this, we will need to make an assumption
that the precision of the DXA was actually determined from
repeat measures to get a CV of 2%. We will also assume that
we do not have repeat measures by our SF-BIA method, and
thus, we can follow the recommended steps to calculate the
“precision” of the SF-BIA equation as follows75:
Precision DXA = 2CVDXA ; this we know and
can calculate as 2 × 2 = 4.
"Precision"SF-BIA = 2CVSF-BIA ; this we do not know
and will try to estimate in a moment.
PE DXA − SF-BIA = 2CVDXA − SF-BIA × 100; this is the percent
error of the difference between methods, which
we can calcuulate from Figure 2B using the limits of
agreement of 5.8//mean of FFM by DXA, which we
will assume was determined to be 32 kg.
This means our PE DXA −SF-BIA = 5.8/32 × 100 = 18%.
Next, we can think about the PEDXA – SF-BIA in terms of each
technique’s CV (with CV based on actual repeatability for the
DXA method and CV based on methods comparison data from
the Bland-Altman for the SF-BIA method) as outlined by
Cecconi et al76:
CVDXA − SF- BIA = √ [(CVDXA ) 2 + (CVSF- BIA ) 2 ],
which can be rearranged to yield
PE DXA − SF- BIA = √ [( Precision DXA ) 2 + (" Precision "SF- BIA ) 2 ],
which can be rearranged to yield
( PE DXA − SF- BIA )
2
= (Precision DXA ) 2 + (" Precision "SF- BIA ) 2
" Precision "SF- BIA = √ [(PE DXA − SF- BIA ) 2 − (Precision DXA ) 2 ],
which can be solved as "Precision "SF- BIA = √ [(18) 2 − (4) 2 ]
= 17%, which compared with the DXA precision of
2% is rather high and would be considered unacceptable.
Of course, had we not gone through these steps, we might have
come to the same conclusion—namely, that the wide limits of
agreement reflected too great a variability at the individual
level; however, if our reference method had worse precision,
say, a CV of 10%, then we may have deemed a 17% “precision” in our bedside method to be acceptable. If, on the other
hand, we look at the data in Figure 2A, making the same
assumptions about the precision of the DXA reference (still 4)
and the mean of the FFM values by the 2 methods (still 32 kg),
then using the limits of agreement of 1.0, we can calculate the
PE to be 1.8/32 = 5.6%, and then “precision” of the SF-BIA in
this case would be √(5.6) 2 – (4)2 = 3.9%; this would obviously
be considered quite good. To reiterate Ward’s point, this is not
to say that this approach should be adopted as standard but
rather to suggest another way to look at limits of agreement in
light of the reference method precision.75 Furthermore, the
decision as to what degree of “precision” can be considered clinically acceptable is likely to vary depending on the application.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
797
A 10% CV may be acceptable for a population-based study but
may not be acceptable for individual bedside monitoring; there
is not yet consensus regarding this point.
Hopefully, it is clear from these illustrations that it is important to keep in mind the precision of the reference method
against which your bedside method is being compared. It is
advantageous for the reference technique to be as accurate and
precise as possible and for the precision of the reference technique to be measured within the same study, although logistical
constraints may make that difficult to achieve, depending on
the method. Although there is no consensus on this, it may be
suggested that for clinical utility for bedside assessment and
monitoring, a PE between reference and bedside methods of
10%, with an estimated “precision” for the bedside method that
comes within 10% of the precision of the reference technique,
could be considered acceptable for absolute whole-body estimates. If, on the other hand, we are attempting to measure longitudinal changes in whole-body estimates, to have confidence
in our measures, we would want the percent error in our bedside (and reference) methods to be less than the expected percent change in the body composition variable.
Measuring Longitudinal Changes as a Way
of Validating a Bedside Method
In many ways, for a method to be useful in the clinical setting,
it is most critical to establish a method’s ability to detect
changes over time. In fact, the evaluation of a bedside method’s ability to measure longitudinal changes in individuals who
are undergoing a stimulus to induce anabolic or catabolic
changes in lean tissue is an important way to validate it; furthermore, the measurement of changes can help to mitigate the
problem of scaling differences (ie, fixed constant bias or systematic error) between methods. Two excellent examples of
opportunities to do this include individuals undergoing cancer
treatment who are likely to exhibit lean tissue wasting or individuals undergoing anabolic therapy to treat wasting in human
immunodeficiency virus (HIV) infection. Obviously, the ability to measure changes in lean tissue is of fundamental importance to the clinician wishing to monitor a patient’s response to
a nutrition intervention or simply to monitor nutrition status
while the patient undergoes various medical therapies during a
hospital stay. The clinician wants to have confidence that the
bedside method is sufficiently sensitive to provide meaningful
information about lean tissue changes. To achieve that, measures of change by the bedside method must be sufficiently
close in value to those measured by the reference method to
establish the method’s validity. If the reference method against
which you are comparing your bedside method is sufficiently
precise (ie, has high reproducibility and repeatability), and the
expected level of change in the body composition compartment is sufficiently high to be detected by that method, then
one can evaluate the ability of a bedside method to measure
that change. The minimal detectable change (MDC) is an
important concept to keep in mind, as it is defined as the minimum change that must occur in a body compartment to achieve
statistical significance at the group mean level; it is determined
by the precision of the method, and it is expressed as a percentage (so it can be related to the size of the compartment being
measured). The %MDC can be calculated as follows78:
%MDC = 1.96 X √ 2 ( which comes to 2.83) ×
%precision , so essentially, 2.83 × precision
expressed as %CV for repeat measurements;
this assumes a desired significance level of 5%.
When comparing a bedside method with a reference method
for its ability to measure change, it is important to know the
precision and MDC of the method. If a bedside method’s MDC
is 6% for FFM, a measured change in FFM <6% could be erroneous (ie, it may or may not reflect a real change in body composition because it is below the threshold for detection). MDC
will be discussed with regard to individual methods in subsequent sections. Another consideration for measuring body
composition changes over time is the safety and efficacy of
conducting repeat measurements by a method. Whereas multiple dilution, DXA, and CT would not be advisable to repeat
over short-term follow-up due to logistical (dilution) and safety
(DXA and CT) reasons, bioimpedance and ultrasound measures can be repeated frequently without concern. A review of
the aforementioned reference methods will be presented next,
followed by discussion of bedside techniques.
Reference Methods for Lean Tissue
Assessment in Clinical Research: Multiple
Dilution, DXA, and CT
Multiple Dilution
In the context of this discussion, multiple dilution techniques
are commonly used to produce reference values for BCM and
the fluid compartments to validate bioimpedance techniques.
Multiple dilution typically involves the use of tritium (3H),
deuterium (2H), or 18O as a tracer for TBW determination, in
combination with a tracer such as bromide for ECW determination. By subtraction, ICW can be calculated, and from ICW,
an estimate of BCM can be determined. The use of 3H is far
less common, due to its radioactive nature, and 18O is typically
only used when given in combination with 2H as doubly labeled
water for total energy expenditure determination. In one
approach, multiple dilution entails taking a predose blood or
urine sample and then allowing a 3- to 5-hour equilibration
period following dosing with deuterium oxide and sodium bromide solutions (with longer periods for individuals with
expanded ECW, including those with extreme obesity, particularly for ECW determination by bromide dilution79) before a
final postdose blood or urine sample is collected. Deuterium
enrichment of the biological sample may be determined
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
798
Journal of Parenteral and Enteral Nutrition 39(7)
by isotope ratio mass spectrometry80–82 or Fourier transform
infrared spectrometry,83 and bromide enrichment can be determined by high-performance liquid chromatography84,85 or nondestructive liquid X-ray fluorescence.86 The reader is referred
to the International Atomic Energy Agency website resources87
and the excellent chapter by Schoeller88 on this topic.
Deuterium dilution for the determination of TBW has a
reported precision of 1%–2%.78,80,81,88,89 The precision of ECW
by bromide dilution is less well characterized but has been estimated to be attainable at 1%–3%.78,88 The determination of
ICW by the multiple dilution technique has a somewhat lower
precision, because it is derived by subtraction (TBW – ECW =
ICW), and thus the errors in TBW and ECW lead to a propagation of error in the determination of ICW. Thus, it has been
estimated that the ICW calculation by this approach has a relative precision of approximately 2.5%.88 It should be noted,
however, that most research studies report precision estimates
that are typically derived from repeated analytical measures of
the final tracer concentration in biological samples, rather than
repeated measures of TBW or ECW within a particular subject.
This is because of the time required for clearance of tracer concentrations from the body, which makes it difficult to conduct
repeated measures over the short term (ie, days apart), although
separation by 2 or more weeks in a physiologically stable person and/or the use of 2 different tracers for the same compartment used simultaneously provide other ways of estimating
precision.80,81,88,89 When precision can be determined by using
2 different tracers or by taking multiple measures to get a total
CV, the contribution of biological variation can be calculated
by subtracting the analytic error CV (from repeat analyses of
the same sample or multiple samples drawn at the same time
point for a subject) from the total precision CV as80
Total CV 2 − Analytical CV 2 = Physiological Error CV 2 .
It should be noted that Schoeller’s group80 applied this method
to calculate the precision of total energy expenditure estimates
using doubly labeled water; however, the concept can be
applied to the determination of TBW by dilution methods as
well. They reported a precision of 1.8% in TBW determined by
18
O and 2H within the same subject; thus, it is certainly possible
to achieve excellent precision with dilution techniques (and to
define it by using 2 different tracers for the same compartment).80 This raises another concern with dilution techniques,
in that the use of different tracers (or the same tracer but applied
using different protocols) can lead to differences in the dilution
space that is used to estimate the fluid compartment; in addition, when a bedside method (eg, bioimpedance) developed
using one tracer protocol is compared with a different one,
large scaling differences may occur that could be interpreted as
error in the bedside method when in fact it may be a systematic
error in the reference. Schoeller et al81 have reported a systematic error of 3% in the TBW measured by 18O compared with
2
H. However, this is not likely to be a prominent concern in the
validation literature; most studies of body composition will not
use both tracers simultaneously (ie, doubly labeled water),
unless they are also estimating total energy expenditure. These
types of issues are not always easy to discern. If one is measuring fluid volume changes, at a minimum, it is important to keep
in mind that the estimated %MDC78 based on the aforementioned precision estimates would be ~3%–5% for TBW by
deuterium dilution and ~3%–8.5% for ECW by bromide dilution of the total volume. Similarly, the %MDC for ICW by
multiple dilution would be at ~7%. True precision estimates
based on repeated measures are not typically done in most
research studies using dilution, so the %MDC for these compartments could be higher. To minimize error when a multiple
dilution approach is applied in a clinical research study, it is
advisable to adhere closely to standardized protocols88 and to
obtain sample analyses of deuterium and bromide enrichment
through collaboration and/or contract with established investigators and facilities. With that approach, multiple dilution can
be a relatively affordable technique for the clinical researcher,
compared with DXA or CT scanning; however, every effort
should be made to minimize the risk of error as the study protocol is developed.
DXA
DXA scanners are becoming fairly commonplace in medical
centers across the United States. DXA is commonly used as a
reference method for the validation of bioimpedance and other
bedside techniques to quantify FFM and LST. DXA is also
used to generate a reference value for skeletal muscle mass; a
value for appendicular skeletal muscle mass (ASM) is generated by summing the LST of the arms and legs. The ASM
index can then be calculated by dividing the arm and leg LST
by height squared. DXA-generated ASM and ASM index values have been used to define sarcopenia (see Table 1).10,90
Briefly, the DXA method requires an individual to lie down on
a scanning table, and low-dose X-rays of 2 different energies
pass through the body; a whole-body scan typically takes 10–
20 minutes. An image is created as the photon detector measures the differential attenuation (or absorption) of the low and
high X-ray energy by the soft tissue and bone. Soft tissue is
further delineated into fat mass and LST, thus resulting in 3
body compartments: fat mass, LST, and bone mass (and density). Traditionally, weight and size limits constrained the use
of DXA instruments in individuals with extreme obesity, but
newer instruments allow for larger and heavier individuals to
be scanned. DXA has a high degree of precision, with reports
of 1%–2% CV for LST.40,91,92 The MDC78 for DXA-measured
LTM based on these precision estimates would be 3%–6%.
CT
CT is another imaging technique that has been used to estimate
lean tissue in the clinical research setting. It involves exposure
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
799
to high-dose radiation and yields a highly accurate quantitative
and qualitative image of skeletal muscle tissue, as well as total
and regional adipose tissue, visceral organs, and bone from the
detection of different X-ray attenuation.29 Given the high-dose
radiation exposure and the high cost, CT as a method for determining skeletal muscle is generally limited to those patients for
whom a CT is ordered for diagnostic and other clinical reasons.
Thus, one of the primary clinical populations in whom CT has
been successfully used for lean tissue assessment and monitoring has been individuals with cancer.93,94 The precision of CT
has been reported to be 1% or less for the evaluation of skeletal
muscle,95 which translates to an MDC of <3%. The reader is
referred to the excellent tutorial by Prado and Heymsfield29 in
the November 2014 issue of the Journal of Parenteral and
Enteral Nutrition for a more complete discussion of the advantages and limitations of DXA and CT as reference techniques.
Bedside Methods for Lean Tissue
Assessment: Bioimpedance and
Ultrasound
Bioimpedance
Bioimpedance is the bedside approach that has been most
widely investigated in clinical research and used by clinicians
for assessment purposes in Europe and elsewhere around the
globe, in large part due to the affordability, portability, and ease
of use of bioimpedance devices.96 A bioimpedance measurement takes less than 15 minutes and is completely noninvasive,
making it advantageous for repeat measurements. Like most
body composition methods, bioimpedance devices do not
directly measure body composition; they provide indirect estimates from the measurement of resistance of body tissues to an
electric current. All bioimpedance techniques involve the
application of a weak, alternating current at one or more frequencies, through leads attached to the body through current
injection electrodes or through direct contact with electrodes in
the case of stand-on scale devices. The differential flow of the
current varies depending on the body composition; electrolyterich components, including blood and muscle, easily conduct
the current, whereas fat and bone do not. The drop in voltage as
the current passes through the body (ie, impedance) is detected
through the voltage detection electrodes, and the impedance
data (resistance, R; reactance, X; impedance, Z; and phase
angle, PA) are recorded by the bioimpedance device.97 Some
devices may only measure Z; it should be noted that R and Z
are not interchangeable in BIA equations. Furthermore, different devices have different electronic circuitry, and the raw data
generated cannot be considered interchangeable, although
interdevice differences have not been well explored in the literature. Although no difference was detected between R and X
measured by the RJL (RJL Systems, Clinton Township, MI)
SF-BIA and Xitron 4000B and Hydra 4200 (Xitron
Technologies, San Diego, CA) BIS devices,46 Chumlea et al98
reported a difference of ~1%–1.5% in R values measured by
the Valhalla (Valhalla Scientific, San Diego, CA) and RJL
SF-BIA instruments that they corrected for in their FFM estimates from the National Health and Nutrition Examination
Survey III (NHANES III) data. Although interdevice differences may not always be substantial, one should be wary of
applying bioimpedance data from a particular device to an
equation that was developed using a different device or a different approach (eg, whole body vs segmental).
The extrapolation from raw impedance data to volume or
mass depends on several key assumptions, including that the
body comprises 5 cylinders of uniform cross-sectional area,
and height is an acceptable surrogate for the length of the conductor (ie, the distance between electrodes), among others.
Body geometry assumptions are violated in individuals with
obesity and in those with longer or shorter than average limbs.
Basic concepts about bioimpedance are presented in an excellent online tutorial through the University of Vermont.99 There
are many different ways that bioimpedance can be applied to
estimate whole-body lean tissue (FFM or BCM) or fluid volume (TBW, ECW, or ICW) estimates using data generated by
SF-BIA or multifrequency BIA (MF-BIA) or bioimpedance
spectroscopy (BIS) devices. The reader is referred to several
recent reviews that describe the different ways that bioimpedance may be used and the differences in technological applications and underlying assumptions between the 3 primary
categories of bioimpedance techniques,96,97,100–104 as well as
currently available devices on the market, which range in cost
from ~$500 to $20,000.97
In general, the level of precision (ie, repeatability) produced
by SF-BIA and MF-BIA devices is typically very good, with
1%–2% variability between repeat measures being reported in
the literature.92,101,105 BIS devices may be somewhat more variable, due to the technical difficulties in producing stable measurements at the extremes of frequencies, but even most BIS
devices have been reported to produce fairly precise measurements (2%–3%).46,106 Based on these precision estimates, the
sensitivity of BIA/BIS techniques to measure whole-body
composition changes can be calculated; for example, the MDC
in FFM by SF- and MF-BIA has been estimated from 3%–6%78
and in TBW by BIS was observed to be 5%–8%.106 On the
other hand, the level of accuracy in whole-body measures produced by various bioimpedance techniques, particularly in
clinical populations with fluid overload, has been observed to
be somewhat more variable.
Precision and accuracy of bioimpedance techniques are
influenced by a number of factors, including patient (eg, degree
of adiposity, fluid and electrolyte status, skin temperature) and
environmental factors (ambient temperature, proximity to metal
surfaces and electronic devices), the assumptions underlying
prediction (SF-BIA or MF-BIA) or modeling (BIS) equations,
instrumentation factors, and variations in measurement protocol. Close adherence to recommended measurement protocols
for body weight107 (see Figure 3), height107 (see Figures 4 and 5),
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
800
Journal of Parenteral and Enteral Nutrition 39(7)
•
•
•
•
•
•
•
Test the accuracy of your measuring scale by using standardized weights. It should have accuracy to 0.1 kg.
If it is a portable scale, ensure that it is on a level surface and horizontal, using a portable spirit level.
Ensure that the individual being weighed has voided the bladder prior to measurement, and that all external clothing (e.g.
coats, sweaters, hat, shoes etc.), heavy jewelry, watches, and pocket contents have been removed.
Ensure that the scale reads ‘0’ prior to asking the individual to step on it.
Ensure that the individual being weighed is standing on the center of the scale platform without touching anything or anyone.
Take two sequential measurements. Ask the individual being weighed to step off the scale and re-zero it in between each
measurement.
If the readings are not ±0.1 kg of each other, repeat the process until you have two measurements that are, and then record the
mean of the two.
Figure 3. Recommendations for optimizing the standing weight measurement in an adult.107
••
••
••
••
••
••
••
••
••
••
Test the accuracy of your stadiometer using a standardized pole of known height.
Ensure that the stadiometer is vertical, using a portable spirit level.
Ensure that the individual being measured has removed shoes, hats, and heavy external clothing prior to measurement.
Ask that the individual remove any hair clips or headbands, and undo braids from the top of the head.
Ask that the individual back up to the stadiometer facing out, with the heels touching the base.
Ensure that the individual’s spine at the level of the pelvis and at the shoulder is just touching the stadiometer, but that the individual is not leaning on the stadiometer.
Check the position of the head. In an adult, position the head in the Frankfort plane (see Figure 5).
Ask the individual to take a deep breath and stand tall just prior to the measurement and hold until the measurement is
complete.
Gently lower the head plate down onto the crown of the individual’s head, pressing down the hair, and read the measurement at
eye level.
Repeat the measurement process two times. If the two measurements are not within 2 mm of each other, repeat the process until
you achieve two consistent measurements. Record the mean of the 2 measurements.
Figure 4. Recommendations for optimizing the standing height measurement in an adult.107
Figure 5. Frankfort plane for positioning the head for a standing
height measurement in an adult. To optimize the accuracy of
a standing height measurement, the head should be placed in
the correct position, as determined by the Frankfort plane. The
Frankfort plane is defined by a horizontal line drawn from the
lower margin of the eye socket to the notch above the tragus of
the ear and perpendicular to the vertical backboard.107
and the bioimpedance96,108,109 measurement itself (see Table 2)
is important to optimize the results of bioimpedance measurements and to minimize potential errors.
For example, how electrodes are placed is quite important.
Depending on the device and whether the aim is to obtain
whole-body or segmental measurements, adhesive electrodes
may be placed in a standard wrist-ankle tetrapolar arrangement
on the hand and the foot of one side of the body, on both sides
of the body on both limbs using an 8-electrode arrangement, or
on different segments of the body.97,110,111 Stand-on devices do
not involve use of adhesive electrodes but rather direct contact
with the electrodes at the feet and/or hands, depending on the
device. Different devices and electrode placement techniques
(eg, for whole-body and various segmental measurements)
should not be considered interchangeable. For example, raw
data obtained by a stand-on device should not be applied to an
equation developed from data derived using a wrist-ankle tetrapolar electrode approach and vice versa, as to do so would
introduce error into the estimate. The most important consideration is the consistent placement of the voltage detection
(proximal) electrodes at standardized anatomical sites; for
example, for wrist-ankle tetrapolar placement, the proximal
electrodes are placed at specific sites on the dorsal surface of
the wrist (eg, between the styloid processes of the ulna and
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
801
Table 2. Recommendations for Optimizing Whole-Body Bioimpedance Measurements in an Adult.96,108,109
Protocol Parameter
Preparing for the
measurement
Food/beverage and
activity
Void bladder
Clean skin surface
Device calibration
Height and weight
Testing conditions
and considerations
Device placement
Ambient
temperature
Electrodes and leads
Electrode placement
Side of body
Body positioning
and limb separation
Fluid and
electrolyte status
Menstrual cycle in
women
Timing of
measurement
Repeat measurements
Recommendation
Individual should fast (nil per os except water) and avoid alcohol, caffeine, and exercise at least 8 hours
prior to measurement in the morning (research settings); shorter time frames and other times of day may
be acceptable in the clinical setting—note time of day for consistency in follow-up measures.
Individual should void bladder prior to measurement.
Clean skin surface well with alcohol; individual should not use lotion or oils on the skin prior to
measurement; avoid placing electrodes on broken skin.
Calibrate the bioimpedance device according to the manufacturer’s recommendations prior to
measurement.
Obtain an accurate measure of height and weight (refer to Figures 3, 4, and 5).
Place device on a nonmetal surface, at least 1 m away from electronic or magnetic devices.
Avoid excessively warm or cool ambient temperature.
Use electrodes with sufficient surface area (≥4 cm2); store electrodes in sealed bag away from heat; use
device-specific leads provided by manufacturer.
Place electrodes at least 5 cm apart, if possible; proximal electrodes should never be moved from standard
anatomical site placement; if necessary, the distal electrodes may be moved to achieve at least 3 cm
of separation; the most important thing is to measure and record distance between electrodes to ensure
placement consistency for follow-up measurements.
If using standard tetrapolar placement of electrodes, measure on the same side of the body as previous
measures; in individuals with amputations, muscle atrophy, or other abnormal conditions, use
the nonaffected side, if possible; be consistent on side of measurement for follow-up. Right-side
measurements are commonly used in the literature.
Body position should be supine, except for stand-on scale devices, with arms separated ≥30° from the
trunk and legs separated by ~45°; in individuals with overweight and obesity, separate arms from trunk
and legs from each other using rolled cotton towels/blankets.
Note if serum electrolytes are abnormal; it is best to conduct bioimpedance measurements only when
serum electrolytes are normal.
Note if edema is present; edema causes lower resistance values.
Note menstrual cycle; be consistent in terms of timing for follow-up measurements.
If individual is ambulatory, individual should assume a supine position for ~5–10 minutes; standardize
the timing for measurements by noting the time when the individual assumes the supine position and the
time when you take the measurement (eg, at 10 minutes), and ensure consistency of timing for all followup measurements. Note if individual is confined to bed.
Repeat measures recommended for research studies.
radius, ie, midline between the wrist bones) and anterior surface of the ankle (eg, between the medial malleolus formed by
the lower end of the tibia and the lateral malleolus formed by
the lower end of the fibula, ie, between the bony protuberances
on the inner and outer sides of the ankle). Although it is common to have the electrode dissect the anatomical mark midline,
different device manufacturers may suggest slight variations
(eg, leading edge, midline, or trailing edge at the anatomical
mark). Small variations in the placement of proximal electrodes can introduce error to impedance measurements (by
~2%).109 When applying data to an equation or making comparisons to reference data, one should follow the same procedure for electrode placement to optimize accuracy.109 Although
it is often recommended that electrodes be placed at least 5 cm
apart in adults, this may not always be possible; it has been
suggested to move the distal electrode to create sufficient separation or to allow a 3-cm separation in children.96 The most
important point is to keep the proximal electrodes at the anatomical marks and standardize the distance for follow-up measurements106; fixed-distance (5-cm) electrodes have been
shown to improve reproducibility in follow-up measurements,
particularly in men.106 The placement of electrodes for segmental measurements to obtain consistency in the anatomical
location and distance between them is also important to minimize potential error, particularly for longitudinal measurements, and to optimize accuracy when applying data to
previously established equations and comparing data with reference data generated from those equations.102
Protocol deviations (eg, failure to abduct the limbs) can contribute substantially to the error in measurements (eg, can affect
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
802
Journal of Parenteral and Enteral Nutrition 39(7)
resistance measurements by 2%–3% for minor deviations109 and
18% and 43% for skin-to-skin contact with crossed legs and
contact of hands to the waist, respectively108). Body positioning
is also quite important; after a person assumes a supine position,
there is an immediate increase in impedance that occurs within
the first 10 minutes and then a more progressive increase in
impedance over the next several hours, reflecting the movement
in fluid distribution toward the trunk.108 The greatest increase in
impedance occurs at lower frequencies, reflecting the decrease
in ECW. BIS-measured ECW decreased by approximately 3%
and ICW increased by just less than 5% after moving from a
semi-upright to a supine position for 30 minutes.112 Those who
resume an upright position for 5 minutes return to presupine
measurements.112 Because most bioimpedance reference data
are measured within 5–10 minutes after patients assume the
supine position, it is not recommended to apply such reference
data to whole-body measurements taken from individuals who
are restricted to bed rest, as errors in TBW and, in particular,
ECW will be observed.108 Similarly, many investigators take
measurements on the right side of the body, which may or may
not be the dominant side; thus, much of the available reference
data have been generated from right-sided measurements. It is
estimated that impedance measurements can vary by approximately 2% between the dominant and nondominant side of the
body109; thus, it is recommended to follow the same protocol
that was used to generate the equation and/or reference data that
you intend to use and to maintain consistency in the side of
body for longitudinal measurements.96 Increasing skin temperature by 6.5°C (11.7°F) also has a significant inverse effect on
impedance, increasing 50-kHz estimates of TBW by approximately 13%; it is recommended that ambient temperature be
maintained between 22.3 and 27.7°C (72.1–81.9°F) to minimize these effects.109,113 Other deviations (eg, failure to clean
skin with alcohol or failure to remove metallic objects from the
body prior to measurement) are associated with smaller errors
(<1% effect on resistance).109 The reader is referred to several
excellent sources that describe errors associated with protocol
violations.106,108,109,113
In general, the application of all varieties of bioimpedance
techniques for the estimation of whole-body compartments has
been accepted as reasonably accurate for the assessment of
healthy nonobese individuals and for large-scale epidemiologic studies (ie, mean-level accuracy), but they have typically
been questioned in terms of their applicability to the clinical
setting due to variability at the individual level. That variability
has been attributed to random and/or systematic errors that are
sufficiently problematic to raise doubts about the capacity of
bioimpedance methods to accurately quantify and monitor
changes in whole-body volumes and masses, and the question
has been raised if bioimpedance techniques can offer any
advantages beyond simple anthropometrics to the assessment
of nutrition status.114 On the other hand, bioimpedance data are
typically highly correlated with lean tissue and fluid volumes
measured by reference techniques, as well as with other indices
of nutrition status, and so the primary problem seems to be the
degree of variability in individual measures when comparing
reference and bioimpedance methods. It has been proposed
that the failure to account for precision error in the reference
techniques (particularly multiple dilution) used to validate bioimpedance methods may have contributed (at least to some
degree) to erroneous conclusions about the capacity of bioimpedance techniques to estimate body composition.75,115
Indeed, one recent study reported that bioimpedance techniques had errors of a similar magnitude to those observed in
the reference (dilution) techniques and underscored the importance of reference method precision when interpreting validation data.115 Although bioimpedance techniques have been
criticized for their limitations, particularly when applied to
individuals with acute and chronic disease as well as individuals with obesity due to the potential violation of underlying
assumptions,97 bioimpedance remains one of the few inexpensive, noninvasive, and portable bedside options currently
available to clinicians.
It is not unreasonable to suggest that bioimpedance techniques might be able to contribute important objective information that may help the clinician to identify sarcopenia and
malnutrition, whether or not a truly accurate measure of lean
tissue or a fluid compartment volume can be obtained. Where
bioimpedance techniques (particularly MF-BIA and BIS) may
play an important role is in those individuals who have muscle
loss that is not evident upon visual examination (including
individuals with sarcopenic obesity) and those with excess
ECW. Furthermore, bioimpedance techniques may be useful in
monitoring the early changes in lean tissue that can happen
with inflammation associated with chronic disease conditions
(eg, cancer and HIV infection).116,117 An objective method for
monitoring anabolic changes in response to nutrition and other
interventions would also be highly advantageous to the clinician. Of note, bioimpedance is being used in dialysis centers
around the world, based on a growing body of literature demonstrating that bioimpedance techniques (segmental MF-BIA
and, to an even larger extent, wrist-ankle BIS) can be used to
monitor fluid status and assess dry weight in individuals on
dialysis.118–124 Finally, although much focus has been paid to
evaluating the accuracy of bioimpedance techniques to quantify lean tissue and/or fluid volumes in absolute terms with
variable success, it has been suggested by many that its true
benefit may be in different applications (eg, by getting away
from whole-body volume or mass estimates and using the data
as markers to reflect lean tissue changes, nutrition status, and/
or clinical outcomes).125,126 These and other novel applications
of bioimpedance techniques are described in several recent
reviews97,100,127 and are an active area of research.
It is important to understand how the different bioimpedance approaches (ie, SF-BIA, MF-BIA, and BIS) use the bioimpedance data to generate whole-body lean tissue mass (eg,
FFM, LBM, or BCM) or fluid volumes (eg, TBW, ECW, or
ICW), because they vary in terms of underlying assumptions
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
803
and thus their theoretical applicability to different clinical conditions. For example, applying an SF-BIA equation to estimate
ICW in an edematous patient is likely to produce substantial
error because of the limitations of using just one frequency,
whereas an MF-BIA or BIS device would offer at least the
theoretical potential of differentiating ICW from ECW by measuring impedance at low frequencies. Furthermore, assumptions regarding normal body geometry and consistent,
predictable hydration of FFM, as well as fluid distribution
between the intra- and extracellular compartments, are frequently violated in the setting of acute and chronic illness. The
generation of whole-body lean tissue estimates by SF- and
MF-BIA requires the utilization of empirically derived prediction formulas that are often erroneous when applied to populations that differ fundamentally from those in which the formula
was developed. The reader is referred to several recent reviews
on bioimpedance97,100,102,103,128 for more complete descriptions
of each of the major bioimpedance approaches and for summaries of the validation literature.33,96,129–131 A basic description of each of the major approaches to the use of bioimpedance,
including novel applications of the raw data, will be presented
next.
SF-BIA. SF-BIA devices measure impedance variables (ie, R,
X, Z, and PA) at a single frequency, typically 50 kHz. To obtain
an estimate of a body compartment, it is necessary to apply one
or more of the impedance variables (most commonly Ht2/R,
which relates the length of the conductor [ie, height of the
body] to volume [eg, TBW]) to a prediction equation. Because
components of the body (eg, TBW, ICW, FFM, BCM, and
LST) are all intercorrelated, impedance can be used to predict
any body composition compartment, if calibrated sufficiently
to a reference method.132 Many prediction equations have been
published in the literature, almost as many as there are studies
that have been conducted to evaluate this approach. SF-BIA
equations are typically developed by regressing impedance
data against a reference method for some body composition
variable (eg, FFM), perhaps using DXA or a 4-compartment
model as the reference, in a homogeneous population sample.
Ideally, the population-specific equation that gets developed is
subsequently cross-validated in a separate sample (and if not
cross-validated, the equation should not be used). It is important to realize that the measurement of impedance at only one
frequency (ie, 50 kHz) is theoretically unable to differentiate
between ECW and ICW (or BCM); in fact, in the clinical setting, it is not likely to be sufficiently high to quantify the entire
TBW. This will become more clear once we have discussed
BIS. Indeed, many SF-BIA devices produce elaborate output
consisting of ECW, ICW, BCM, FFM, and more; however,
they are simply predicting these compartments based on
assumptions of static relationships between the body compartments from normative data; these assumptions may hold sufficiently true in healthy individuals. SEE for predictions of
FFM by various SF-BIA equations have been observed to
range from 1.8–4 kg in healthy normal-weight adults and from
1.6–3.4 kg in in healthy elderly individuals and as high as 8.8
kg in overweight women, compared with DXA, 4-compartment models, and/or densitometry.104 Although they reported
%body fat (%BF), rather than FFM, it is interesting to note that
in women with extreme obesity (mean %BF 51.4% ± 3.6%),
Das et al27 observed limits of agreement of ±5.1% and ±5.8%
for 2 SF-BIA equations, as well as ±2.2% for 2H dilution for
%BF measures, compared with %BF by the reference 3-compartment model. After gastric bypass-induced weight loss (and
–16.8 ± 8.5 %BF), the methods agreement worsened considerably, with limits of agreement for the measures of change in
%BF compared with a reference of ±8.5% (estimated PE calculated as per Ward75 of 51%) and ±9.6% (estimated PE 57%)
for 2 SF-BIA approaches, compared with ±5.7% (estimated PE
of 33.9%) for 2H dilution, underscoring the particular difficulties (by any method) when measuring changes in individuals
with extreme obesity undergoing massive weight loss. In clinical populations, similar or higher errors have typically been
observed for FFM and SM estimates.133,134 For a clinician who
wants to evaluate lean tissue using an SF-BIA device, he or she
is faced with the dilemma of choosing a prediction equation
(from the abundant literature on the topic) that ideally matches
the patient that he or she wishes to assess. The difficulties of
making such a selection cannot be overemphasized. This is one
of several limitations to the application of SF-BIA in the clinical setting, regardless of questions about potential accuracy for
whole-body lean tissue estimates.
It has been suggested by a growing number of researchers that
a potentially more useful approach to the use of SF-BIA might be
to apply the raw data to a published equation that has established
normative reference values12,13,135–137 or to generate the PA and
compare that with published normative reference values.138,139
Although many BIA equations have been published, several different applications of 50-kHz data for estimating lean tissue have
been primarily advocated, based on the concept of normalizing
lean tissue data to height to create a standardized measure that
might be used to indicate nutrition status.140 One is to apply the
50-kHz data to a validated multiethnic prediction equation for
skeletal muscle mass (SM) and then to calculate the skeletal muscle index (SMI)12; this has been incorporated into the definition
of sarcopenia (relabeled as “whole-body fat-free mass index
without bone” by Fearon et al14) as part of the international consensus definition of cachexia,14 based on evidence that it correlated well with MRI137 and predicted functional status in older
adults.12,13 This SF-BIA SM equation is as follows137:
Skeletal muscle mass (kg ) = [(Height 2 / R × 0.401) +
(Sex × 3.825) + (Age × − 0.071) ] + 5.102,
where height is in cm, R is resistance in ohms,
and for sex , male = 1 and female = 0.
This equation, first developed from bioimpedance data generated by an SF-BIA device (RJL Systems, Detroit, MI) in 269
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
804
Journal of Parenteral and Enteral Nutrition 39(7)
white men and women (mean age, 42.5 years [range, 18–86
years]; mean BMI, 28.9 kg/m2 [range, 16–48 kg/m2]), was subsequently cross-validated against MRI measurements in
Hispanic (n = 26), African American (n = 53), and Asian (n =
40) men and women. The equation underpredicted skeletal
muscle in Asian individuals (mean BMI, 22 kg/m2) but performed reasonably well in the other groups who had an average
BMI similar to the white group, with an overall R2 and SEE of
0.86 and 2.7 kg (9%), respectively.137 Individual-level agreement was not evaluated, and thus, it is not known how well this
equation can predict SMI for bedside assessment. The first version of the SMI using this equation was calculated by dividing
the skeletal muscle mass by body weight in kilograms and multiplying by 100.12 Subsequently, these investigators calculated
SMI by dividing the skeletal muscle mass by height in meters
squared and identified cutpoints to define sarcopenia based on
physical disability13; these cutpoints have been published as
potential defining cutpoints (among others from the literature) in
the European consensus paper on the definition of sarcopenia.17
Reference values for sarcopenia cutpoints based on the above
BIA-derived SMI estimates are presented in Table 1.
Another proposed way to apply 50-kHz data is to calculate
FFM using a validated SF-BIA equation developed from
healthy reference population data.98,135,141 Kyle et al142 have
taken the lead in investigating this concept for its application in
the clinical setting; they recommend calculating FFM by their
validated equation (coined the Geneva equation) generated
from a Swiss population sample using a BIS (Xitron 4000B,
then owned by Xitron Technologies) device and converting it to
an index (fat-free mass index [FFMI]) by dividing it by the
squared height; values can then be compared with reference
data to evaluate the individual compared with healthy reference
norms.133,136,142 The Geneva FFM calculation is as follows135:
FFM (kg ) = 4.104 + (0.518 × Height 2 / R ) +
(0.231 × Weight ) + (0.130 × Reactance ) +
(4.229 × Sex ), where height is in cm, R is resistance
in ohms, and weight is kg, and for sex , male = 1
and female = 0; convert to FFMI by dividing by height
in meters squared.
FFM by this equation has been validated against DXA and has
been shown to have acceptable error for healthy individuals
(CV of 3.6%, SEE of 1.72 kg, and estimated PE75 of 6.3% in a
large sample of 343 Swiss whites spanning ages 20–94 years
and BMI from 17–33.8 kg/m2).135 Note that the authors did not
calculate PE, but it was estimated from their data based on the
calculation of 2 CV for the reported difference between methods, which would be 1.7 SD × 2/mean FFM of 54 = 6.3% error.
Chumlea et al98 also developed and cross-validated FFM
SF-BIA prediction equations for men and women (ethnicity
combined) based on the NHANES III data; however, FFMI
values were not calculated in their study, and only the mean
and SD for SF-BIA–predicted FFM in various age groups were
presented rather than percentile data.141 Means and SD for
FFM are inherently limited and should be interpreted with caution. It would be advantageous to have reference data by percentiles for FFM (and FFMI) from the NHANES III equations,
if they are to be useful as a way to identify individuals with low
muscle mass or at increased nutrition risk. Schutz et al136 have
proposed that FFMI values below the fifth percentile cutpoint
from healthy reference norms could be interpreted to suggest
compromised nutrition status, although this is not well established, and fifth percentile cutpoints for FFMI are only available based on Swiss white population data, and thus, their
application to other ethnicities and population samples may be
limited. Nevertheless, the fifth percentile reference cutpoints
defining low FFMI calculated by the Geneva equation,135,136
based on a Swiss white population sample, range from <12.9
and <16.6 kg/m2 for women and men, respectively, older than
75 years, to <14.4 and <17.2 kg/m2 for women and men,
respectively, 35–54 years in age.
Some studies reported that a low FFMI in individuals (calculated by the Geneva equation135 but defined by slightly
higher cutpoints142 than previously reported136) was significantly associated with increased length of stay compared with
those with normal or better values.6 Another recent study143 in
a Brazilian population sample reported that low FFMI (defined
by the Geneva equation135 and Kyle cutpoints136) was correlated with malnutrition diagnosed by subjective global assessment. Low FFMI (defined by the Kyle cutpoints136) but with
FFM estimated by a BIS device with the device software (i.e.,
not by the Geneva equation) before surgery was shown to be
independently associated with a higher prevalence of postoperative infections and increased length of stay in a Dutch ICU
study sample.144 Another group recently reported145 that discrepancies between 2 different bioimpedance devices (an
SF-BIA device and a BIS device) were evident for FFMI values that resulted in significant differences in the number of
individuals identified at nutrition risk; however, it should be
clarified that the authors did not simply use the raw bioimpedance data generated by the 2 devices to calculate the Geneva
equation FFMI (which would be a more appropriate way to
make that comparison). Rather, they took the FFM values generated by the proprietary software in the SF-BIA device and the
FFM value generated by the modeling and mixture-based
equations programmed into the BIS device. This important
detail was overlooked in a recent review by this author.97 Thus,
it is not clear that important differences in actual impedance
values were evident in that study. Furthermore, it should be
emphasized that it is not recommended to use the cutpoints
generated by one method (ie, using the Geneva SF-BIA FFM
equation) to interpret a measurement made by a completely
different method (ie, FFM generated by device software). The
only way that it could be advocated to use the Kyle reference
cutpoints136 for FFMI would be to use the raw bioimpedance
data to calculate FFM by the Geneva equation135 and then
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
805
Reactance (X), in Ohms
Frequency (kHz) increasing
Impedance (Z), in Ohms
Characterisc Frequency (fc)
Phase
Angle
R
Resistance (R), in Ohms
Key Cole model terms:
R0: Represents ECW resistance and is renamed RE
R : Represents TBW resistance
RI: Represents ICW resistance*
Cm: Cell membrane capacitance
Exponent
R0 = RE
RI and fc can be computed:
1/ RI = 1/ R - 1/ R E
fc = 1/[2 Cm(RE + RI)]
Figure 6. Cole plot. The RI term is mostly representative of the
intracellular water (ICW) resistance, but ICW resistance is also
affected by cell membrane capacitance. Exponent α accounts
for the distribution effects (including cell size and shape) that
cause suppression of the semi-circle below the x-axis.102 Adapted
from Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical
impedance analysis—part I: review of principles and methods.
Clin Nutr. 2004;23(5):1226–1243. Reprinted with permission
from Elsevier.
convert to FFMI. That said, significant differences between
devices might yield different values for the raw bioimpedance
data that would affect the end FFMI or SMI values, as has been
observed with reference data for PA, which will be discussed
next.
PA is a simple bioimpedance parameter generated from the
arctangent of the ratio of reactance to resistance at 50 kHz (see
Figure 6). It can be calculated from the following equation126:
PA = arctangent (X / R ) × 180° / π.
Although it has been purported to relate to cell biology—
namely, membrane integrity, permeability, size, hydration, and
capacitance—it is not entirely clear that there is a physiologic
foundation for the observed statistical associations.
Nevertheless, a low PA has been reported to be an independent
prognostic indicator of disease and/or nutrition risk/poor nutrition status in a variety of clinical populations, including HIV
infection,146,147 cancer,148–151 chronic heart failure,152 cirrhosis,153 stage 5 chronic kidney disease,154 and hospitalized
elderly.155 Normative values for PA measured at 50 kHz have
been published based on healthy German,138,156 Swiss,157 and
American139 populations; however, fifth percentile cutpoints
are available only from German138 and American139 reference
data. These fifth percentile cutpoints are presented in Table 3.
The mean reference data138,139,156,157 (see Table 4) can be used
to calculate the standardized PA (SPA), which involves creating a z score by using the following equation139:
SPA = (Observed PA − Mean PA) / SD , where
the mean and SD are from reference values.
The SPA was reported to be a significant independent predictor
of malnutrition and decreased functional status, as well as
6-month survival.158 A standardized PA below –1.65 has been
suggested as a cutoff to indicate malnutrition in individuals
with cancer, representing the 5th percentile.159 Age, sex, and
BMI are major factors that may affect the PA; Bosy-Westphal
et al138 have published age-, sex-, and BMI-specific reference
cutpoints. Others139,156,157 have presented age- and sex-specific
reference data without delineating BMI categories. Observed
differences between published reference cutpoints may be
attributed to device-, BMI-, and/or population-specific differences. These issues are addressed quite well in the recent
review by Norman et al127; in addition to discussing the clinical
use of 50-kHz PA and SPA, they also discuss the applications
of 50-kHz bioelectrical impedance vector analysis (BIVA)
originally developed by Piccoli et al160 for the assessment of
hydration status. Whether or not PA or SPA can be used to
identify individuals with muscle loss (compared with reference
methods) or to uniquely contribute to the diagnosis of malnutrition beyond other diagnostic criteria has yet to be
established.
MF-BIA. MF-BIA approaches may be taken using data collected by an MF-BIA device or by a BIS device, although the
data are handled quite differently by the BIS technique, as will
be discussed later. MF-BIA is quite similar to SF-BIA, but
instead of just one frequency, MF-BIA devices measure impedance at 2 or more frequencies, typically at 4 or 5 frequencies,
including at least 1 low (most commonly 5 kHz) and several
higher ones (typically 50, 100, 200, and 500 kHz). As with SFBIA, the MF-BIA approach to estimate whole-body volumes or
lean tissue masses typically involves the use of linear regression–derived, population-specific equations. However, it has
the advantage of taking measurements at a very low frequency
and one or more high frequencies and thus theoretically allows
for the estimation of both ECW and ICW. Separate equations
are typically developed to predict ECW and TBW based on reference values derived from bromide dilution and deuterium (or
tritium) dilution, respectively, in a particular sample population.
Subsequently, ICW can be estimated through subtraction (ie,
TBW – ECW = ICW). Alternatively, total body potassium reference values may be used to develop prediction equations for
ICW and BCM. DXA, 4-compartment models, densitometry, or
deuterium dilution data may be used to develop prediction
equations for FFM or other lean compartments. Another less
conventional approach to the use of MF-BIA data, if measurements are taken at 3 or more (and ideally 6 or 7) frequencies, is
to model the data through nonlinear least squares curve-fitting
techniques similar to those applied in BIS,161 although this
approach has not been well studied and is likely to be less robust
than the BIS approach of using spectral data measured at 50 or
more frequencies, which will be discussed later.
Despite the added advantage of measuring at multiple frequencies, the more typical MF-BIA prediction equation approach
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
806
Journal of Parenteral and Enteral Nutrition 39(7)
Table 3. Cutpoints Defining Low Phase Anglea From German138 and American139 Reference Data.
Age Group, y
BMI
20–29
30–39
40–49
50–59
60–69
≥70
5.5
5.9
5.8
5.7
5.4
5.4
5.6
5.7
5.5
5.2
5.2
5.3
5.3
5.1
5.0
4.2
4.9
4.9
4.7
4.5
3.8
4.2
4.3
4.2
3.6
6.6
6.5
6.1
5.4
4.8
5.1
5.2
5.3
5.3
5.2
5.0
5.1
5.2
5.1
5.0
4.7
4.9
4.9
4.8
4.8
4.4
4.6
4.6
4.5
4.4
3.9
4.2
4.2
4.1
4.0
5.6
5.6
5.5
4.7
4.1
138
Bosy-Westphal et al : Males (n = 30,572)
5.8
18.5–25
25.1–30
5.8
5.9
30.1–35
35.1–40
5.7
40.1–50
5.6
Barbosa-Silva et al139: Males (n = 832)
Mean 25.6
6.8
138
Bosy-Westphal et al : Females (n = 183,176)
5.0
18.5–25
25.1–30
5.1
5.2
30.1–35
5.2
35.1–40
5.1
40.1–50
Barbosa-Silva et al139: Females (n = 1135)
Mean 26.0
5.6
a
138
Low phase angle defined as the fifth percentile value based on healthy reference data; the Bosy-Westphal et al study used the BIA 2000 Data Input
(Frankfurt, Germany) single-frequency bioelectrical impedance analysis (SF-BIA) device; the Barbosa-Silva et al139 study used the Model 101 RJL
Systems (Mt Clemens, MI) SF-BIA device. All values presented as means rounded to the nearest tenth to simplify the presentation. BMI, body mass
index.
is limited for whole-body estimates by the same complaints that
may be made about the SF-BIA prediction equation approach,
particularly in the clinical setting and in individuals with obesity,49 where underlying assumptions may be violated. Far fewer
MF-BIA equations than SF-BIA equations have been published
in the literature, but some have been specifically developed for
lean tissue and/or fluid volumes (TBW, ECW, and/or ICW) in
healthy individuals,162–167 and some have been developed for
clinical assessment of fluid volumes (eg, in surgical patients168,169
and individuals with HIV infection146). Although MF-BIA equations may do reasonably well in healthy individuals, reports of
errors are quite variable in clinical populations, with reported
SEE of 2.0 for ECW and 4.0 for TBW in surgical patients,168
limits of agreement of ±9 and ±14 L for TBW and ECW measures in individuals with cirrhosis,170 and limits of agreement of
±18% for ICW estimates in individuals with HIV infection.146
Although clinicians may search the literature and identify appropriate equations with which to apply MF-BIA impedance data to
estimate whole-body volumes or masses in their particular
patient, more often than not, time will be limited, and the only
alternative will be to accept the output generated by the device,
the underlying equations for which may not be readily available,
being considered “proprietary” by the manufacturer. The raw
data generated by an MF-BIA device could have utility independent of whole-body volumes/masses by allowing a clinician to
evaluate the 50-kHz PA as well as a high to low frequency (eg,
200/5-kHz) impedance ratio, as will be discussed next.
Unfortunately, some MF-BIA devices do not provide the raw
bioimpedance data but rather just produce whole-body volumes
or masses with no indication to the clinician which equations
and/or specific variables were used to generate them. Any bioimpedance device that does not provide the raw data and/or uses
“black box” proprietary equations is not as useful as it could be.
These issues are major barriers facing the further development
of bioimpedance for clinical use.
Given the concerns about the accuracy of whole-body measures in the clinic setting, there has been significant interest in the
utilization of a simple ratio of impedance measured at 200 kHz to
impedance measured at 5 kHz. Although it is possible to create
ratios from impedance measured at other frequencies, the 200/5kHz impedance ratio has been the most commonly investigated as
a potential marker for nutrition status and/or disease severity, with
worse outcomes the closer to 1.0 the ratio is. For example, higher
200/5-kHz impedance ratios have been observed in individuals
with poor nutrition status,171,172 postoperative edema,173 and
impaired renal and cardiac function.174,175 Values >1.0 suggest
device error.
In a study that included 151 healthy volunteers, normal
values for this impedance ratio were ≤0.78 in men and ≤0.82
in women.171 Values above these cutpoints were associated
with more than a 4-fold higher risk of malnutrition, identified by low total body protein measured by neutron activation analysis.171 Importantly, a high impedance ratio was
reported to have a sensitivity of 79% and a specificity of
53% for identifying malnutrition.171 Similar to the concept of
using PA as a predictive variable, the simplicity of using a
ratio to evaluate nutrition status has appeal, given that it
requires no population-specific regression equations and
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
807
Table 4. Mean Values for Phase Anglea From German,138 Swiss,157 and American139 Reference Data.
Age Group, y
BMI
20–29
30–39
40–49
50–59
60–69
≥70
6.7 ± 0.70
6.9 ± 0.69
6.9 ± 0.72
6.9 ± 0.69
6.7 ± 0.76
6.5 ± 0.70
6.7 ± 0.70
6.8 ± 0.68
6.6 ± 0.74
6.4 ± 0.77
6.2 ± 0.66
6.4 ± 0.72
6.4 ± 0.70
6.4 ± 0.76
6.2 ± 0.77
5.8 ± 0.82
6.0 ± 0.75
6.0 ± 0.76
6.0 ± 0.85
5.8 ± 0.86
5.1 ± 0.86
5.4 ± 0.77
5.5 ± 0.76
5.4 ± 0.73
5.0 ± 0.87
8.0 ± 0.85
7.8 ± 0.85
7.3 ± 0.89
7.0 ± 1.1
6.2 ± 0.97
6.0 ± 0.68
6.1 ± 0.67
6.2 ± 0.69
6.2 ± 0.70
6.1 ± 0.72
5.7 ± 0.68
5.9 ± 0.70
5.9 ± 0.70
5.9 ± 0.72
5.8 ± 0.70
5.5 ± 0.78
5.6 ± 0.72
5.6 ± 0.73
5.6 ± 0.75
5.5 ± 0.77
5.1 ± 0.84
5.3 ± 0.78
5.3 ± 0.75
5.3 ± 0.84
5.1 ± 0.72
6.9 ± 0.85
6.6 ± 0.87
6.0 ± 0.83
5.6 ± 1.02
138
Bosy-Westphal et al : Males (n = 30,572)
6.9 ± 0.72
18.5–25
25.1–30
7.0 ± 0.72
7.0 ± 0.71
30.1–35
35.1–40
6.9 ± 0.74
40.1–50 6.7 ± 0.69
Barbosa-Silva et al139: Males (n = 832)
Mean BMI 25.6 ± 4.2
8.0 ± 0.75
138
Bosy-Westphal et al : Females (n = 183,176)
6.0 ± 0.68
6.0 ± 0.67
18.5–25
25.1–30
6.1 ± 0.68
6.2 ± 0.67
6.2 ± 0.68
6.3 ± 0.67
30.1–35
6.2 ± 0.68
6.2 ± 0.66
35.1–40
6.2 ± 0.66
6.2 ± 0.71
40.1–50
Barbosa-Silva et al139: Females (n = 1135)
Mean BMI 26.0 ± 6.4
7.0 ± 0.92
6.9 ± 0.84
Age Group, y
Kyle et al157,b
Males (n = 2735)
Mean BMI
Females (n = 2490)
Mean BMI
15–24
25–34
35–44
45–54
55–64
65–74
75–84
>85
7.3 ± 0.8
22.2
6.6 ± 0.8
21.0
7.5 ± 0.8
23.5
6.6 ± 0.9
21.4
7.2 ± 0.9
24.1
6.7 ± 0.8
21.9
7.2 ± 0.9
24.4
6.5 ± 0.9
22.7
6.6 ± 0.9
25.0
6.0 ± 0.8
24.3
6.1 ± 0.9
25. 7
5.4 ± 0.9
25.9
5.3 ± 0.9
24.9
4.8 ± 0.9
25.3
4.6 ± 0.8
26.2
4.6 ± 1.2
25.2
a
The Bosy-Westphal et al138 study used the BIA 2000 Data Input (Frankfurt, Germany) single-frequency bioelectrical impedance analysis (SF-BIA)
device; the Barbosa-Silva et al139 study used the Model 101 RJL Systems (Mt Clemens, MI) SF-BIA device; the Kyle et al157 study used a combination of
Model 109 and 101 RJL Systems SF-BIA devices and the Xitron 4000B (Xitron Technologies, San Diego, CA) bioimpedance spectroscopy device. All
values presented as means rounded to the nearest tenth to simplify the presentation. BMI, body mass index.
b
BMI listed for each age group represents the mean group BMI for that category.
gets away from the potential errors associated with wholebody estimates; however, for it to be clinically useful, reference cutpoints need to be established. Furthermore, just as is
true with the PA, the use of the impedance ratio as a marker
relies on a statistical relationship, such that other variables
that are equally good at predicting nutrition status could be
easier to measure. The question remains to be answered
regarding whether the measurement of the 200/5-kHz or any
other high- to low-frequency impedance ratio can improve
upon the diagnosis of malnutrition beyond physical examination and other clinical data.
BIS. In contrast to SF-BIA and MF-BIA approaches, which
require the use of a regression-derived equation to predict
whole-body volumes or masses from raw data assuming a linear relationship, BIS devices use a more scientific approach
based on biophysical modeling.102,176 Impedance spectroscopy
is a well-known analytical technique used in materials and biological research; technological advances in the early 1990s
facilitated the application of BIS for the assessment of in vivo
body composition for the first time.102 In contrast to SF-BIA
and MF-BIA devices, BIS devices apply the electrical current
(typically ≤800 µA) over a range of frequencies, from very low
(eg, 1 or 5 kHz) to very high (eg, 1000–1200 kHz), measuring
impedance data (ie, R, X, Z, and PA) at 50 or more frequencies.
The application of a variable-frequency current to biological
tissue yields a response that gives rise to a semicircular graphical relationship between R and X due to the cell membrane
capacitance (Cm); this relationship is best represented by the
standard and classic Cole model of biological tissue.177 Nonlinear least squares curve-fitting techniques are applied to the
spectral data, generating a semicircular Cole plot and Cole
model terms RE, RI, Cm, and exponent α (see Figure 6). These
terms can then be used to provide estimates of ECW and ICW
through algorithms that will be discussed later.
Theoretically, at an exact very low frequency, the ECW is
measured independent of ICW, and at an exact very high frequency, the ICW is fully measured. However, it is not possible
to measure at frequencies >10 MHz or <1 kHz, and thus, using
nonlinear least squares curve fitting, data are fit to the
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
808
Journal of Parenteral and Enteral Nutrition 39(7)
semicircle and extrapolated to the exact point where there is
only ECW and no ICW conduction (ie, R0 or RE) and where
there is complete conduction through both ECW and ICW (ie,
R∞). From RE and R∞, RI and Cm can be derived; the exponent
α term that is also a Cole model parameter has been discussed
elsewhere.176 To best solve for the primary components of the
Cole model, sufficient data points are required; thus, impedance data are typically measured at 50 frequencies.
The characteristic frequency (fc) is also computed from
RE, RI, and Cm; it is defined as the point of maximal reactance
or the frequency where the measurement is halfway between
zero and infinity.102 As you can see from Figure 6, it is graphically represented as the top of the semicircular Cole plot.
Theoretically speaking, at any frequency other than zero and
infinity, the proportion of ICW measured varies with increasing frequency and when the tissue changes, causing a change
in fc.102 The BIS modeling approach is the only bioimpedance
approach that actually calculates the fc. Although fc may in
time have its own important clinical applications, for this discussion, it is useful to understand fc because it underscores an
important advantage of the BIS modeling approach.102 For
example, at 10 kHz, there is more ICW measured than at 5
kHz, and at 50 kHz, there is less ICW measured if fc is 100
kHz than when fc is 30 kHz. Although 50 kHz is used to predict TBW, it is typically >20:1 away from where ICW is fully
measured (ie, at infinitely high frequency). One of the
assumptions made by SF-BIA is that 50 kHz is sufficiently
high to predict TBW, but that is likely to only hold true in
healthy people (for whom the fc is typically around 40 kHz
and underlying relationships between compartments are not
altered).102 That will almost certainly not hold true in many
individuals with acute and chronic disease, who may have
significantly greater fc (eg, >200 kHz in individuals on dialysis176). For these individuals, much higher frequencies would
likely be needed. For more in-depth insight into this topic, the
reader is referred to the excellent review by Matthie.102
Interestingly, it has been suggested that the reason 50 kHz PA
has been shown to predict nutrition status is because, for most
individuals, 50 kHz exceeds their fc102; it is quite possible that
the application of PA as a marker of malnutrition might be
improved by using the PA measured at the fc measured by a
BIS device, but this remains to be investigated. Nevertheless,
it is important to realize that no single frequency (5, 50, fc,
100, 200, 500, or even infinite kHz) is a measure of TBW but
rather a mixture of ICW and ECW, which have substantially
different (ie, 5:1) resistivities.176 The only way to accurately
estimate ECW and ICW is to independently evaluate them,
and BIS is theoretically ideal to do this.
Typically, BIS devices are programmed with software that
first models the data, using nonlinear least squares curve fitting
and assuming that the fit is sufficiently good, then applies
parameters (RE, RI, Cm) generated by the modeling procedure
to calculate fluid volumes and other body composition compartments using a particular algorithm. The Cole model
parameters can be applied to equations that were originally
developed based on Hanai’s mixture theory,178 which, as it
applies to humans, assumes that the electrical properties of tissues are affected by the mixture of conducting (eg, body water,
electrolytes, muscle and other lean tissue) and nonconducting
(eg, bone, fat) body components. The BIS approach involves
several other key assumptions, including constants for estimating the specific resistivity of the ICW and ECW compartments
(termed apparent resistivities or resistivity coefficients) and
constants for body density and shape; these are likely sources
of error in individuals with obesity and edema, who may vary
from normal-weight shape and density. ICW and ECW resistivity coefficients used in device software algorithms may differ, depending on the manufacturer’s decision to use published
values or to generate its own from reference (eg, multiple dilution) data. Most commonly applied are the coefficients published by De Lorenzo et al,176 which were derived from TBW
and ECW determined by deuterium and bromide dilution,
respectively, and by the Xitron 4000B (Xitron Technologies)
BIS device in healthy normal-weight individuals, but coefficients have been generated in other populations, including
individuals with extreme obesity.47 The use of different dilution techniques and different BIS devices will yield different
constants; this is a potential source of error in BIS measurements, although it has not been well studied. The original mixture theory-based equations for determining fluid volumes
from BIS spectral data were first developed by Xitron
Technologies in the early 1990s. Improvements were made
after recognizing how the resistive properties of the ICW compartment were affected by adiposity, and those second-generation equations were published.176,179
Much of the published literature on BIS has evaluated the
original Xitron-Hanai–based mixure equation approach using
resistivity coefficients generated by the device manufacturer or
those published by De Lorenzo et al176 to estimate fluid volumes
and/or FFM or BCM in various populations. In general, a similar
pattern of errors in whole-body estimates has been observed
with BIS as with other bioimpedance approaches—namely,
good mean-level agreement but sometimes large variability at
the individual level. For example, focusing on ICW given its
relevance to lean tissue, relatively acceptable mean-level errors
(SEE of ~1.5–2.5 L in ICW, approximately 8%–14% error) have
been observed in healthy individuals.180 BIS using the XitronHanai–based mixture equation approach was also shown in individuals with HIV infection undergoing treatment with
oxandrolone to produce similar SEE estimates for ICW in absolute terms before (SEE 9%) and after (SEE 6%) treatment and to
track changes in ICW (BCM) better than an HIV-specific
SF-BIA equation,116 but variability in individual measurements
was in some cases quite large.46 Wide limits of agreement (±5 L,
compared with the mean change of ~4.4 L) were observed in
measures of ECW changes in ICU patients with major trauma
and sepsis.181 Substantially greater error (with fixed and proportional bias, as well as wide limits of agreement) was observed
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
809
for all fluid volume measures in individuals with extreme obesity.47–49 Since then, it was determined that a significant portion
of the error observed in BIS measurements came from the
impact that adiposity had on the ICW estimates, and new equations that incorporated a correction for BMI were advocated and
tested.182 These revised equations, with the correction for BMI,
are as follows182:
2
2
ECW = k ECW × [(Height × √ Weight ) / R 0 ] 3 , and
2
ICW = k ICW × [(Height 2 × √ Weight ) / R I ] , where
3
height is in cm, weight is in kg, R I is the intracellular
resistance ( value generated from the Cole modeling
procedure), and k ECW and k ICW are functions of BMI,
calculated as follows :
k ECW = (a / BMI) + b, and k ICW = (c / BMI) + d,
where a = 0.188, b = 0.2883, c = 5.8758, and
d = 0.4194, as determined from reference data
and later cross-validated.
The BMI correction reduced, but did not eliminate, the individual errors associated with BIS ICW measurements in individuals on dialysis and in healthy individuals, particularly in
those at the extremes of BMI.182
This same group has incorporated a new conceptual model
(which will be referred to henceforth as the Chamney model)
regarding the hydration of adipose and lean tissue based on
cadaver data183 into the BIS approach and is applying the technology to the problem of fluid management in individuals on
dialysis.184 In the most recently published version of the model,
it is conceptually possible to quantify excess fluid and thus more
accurately assess the lean tissue compartment, which they term
“normally hydrated lean tissue.” The ECW and ICW volumes
calculated from the BMI-corrected mixture based equations
could be applied to the Chamney model equations183:
M ExF = (1.136 × ECW ) − (0.430 × ICW ) −
(0.114 × Weight),
M NH _ LT = (2.725 × ICW ) + (0.191 × M ExF ) −
( 0.191 × Weight ),
M NH _ AT = Wt − M ExF − M NH _ LT ,
M Fat = 0.753 × M NH_AT , where M ExF is the excess
fluid mass (kg ), M NH_LT is the normally hydrated
lean tissue mass (kg ), M NH_AT is the normally
hydrated adipose tissue mass, M Fat is the fat
mass (kg ), ECW is the extracellular water
mass (k
kg ), ICW is the intracellular water
mass ( kg ) , and weight is in kg.
The validation of this approach to quantify lean tissue is limited by the fact that there is no clear reference method for
specifically measuring the “normally hydrated lean tissue”
compartment. However, it does appear that their approach, or
some derivation, is being successfully applied to evaluate dry
weight and the adequacy of dialysis in individuals with renal
failure.183,184 It may be surmised that the better their modelderived estimates of excess fluid become, the more refined
will be the estimates of lean tissue. The refinement of the
apparent resistivity constants, among other factors, will
almost certainly be a necessary part of that process. A recent
report provides compelling evidence of the efficacy of this
approach for lean tissue assessment.185 Marcelli et al185
reported that individuals on hemodialysis who had NH_LT
values below the 10th percentile from healthy reference data
had significantly higher mortality than those with higher values, suggesting that it could be indicative of malnutrition and
poor clinical status. This is an exciting and promising area of
bioimpedance application, and additional research is warranted to see if BIS-measured “normally hydrated lean tissue” can be used to evaluate muscle loss for diagnosing
malnutrition and to track changes in response to nutrition
interventions.
Ultrasound Measurements of the Quadriceps
Muscle
Ultrasonography is a readily available, portable technique used
for diagnostic and monitoring purposes in clinical settings.
Good-quality ultrasound devices may cost $30,000–$50,000,
making them slightly more expensive than the top-end BIS
devices but far less expensive than a DXA or CT scanner. The
application of ultrasound to measuring body composition is not
entirely new; it has historically been used for assessing the
abdominal adipose compartment to distinguish visceral and
subcutaneous fat depots.186 Much of the research that has been
done on ultrasound applications in muscle tissue has come
from the sport injury and neuromuscular disease fields.187,188
More recently, there has been growing interest in the use of
ultrasound to quantify lean tissue at the bedside, particularly in
the intensive care setting.189–192 For example, one recent study
tracked loss of muscle mass in critically ill patients using ultrasound measures of the quadriceps muscle thickness.192 Muscle
thickness can be measured at one or more anatomical sites, and
these values may be extrapolated to whole-body lean tissue
estimates,193 although there are only a few equations that have
been proposed to predict FFM194,195 or skeletal muscle
mass196,197 using the sum of thicknesses from 3–9 anatomical
sites in healthy individuals. For a more complete discussion
and review of the available validation literature on ultrasound
for lean tissue assessment, the reader is referred elsewhere.198
Generally speaking, ultrasound involves the transduction
of high-frequency sound waves through the skin; the ultrasound beam is partially reflected back (as an echo) to the
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
810
Journal of Parenteral and Enteral Nutrition 39(7)
transducer from the interface of different underlying tissues
(eg, subcutaneous adipose and skeletal muscle).29,186 How
much the tissue reflects is expressed as acoustic impedance,
and this ranges from the lowest acoustic impedance levels
associated with air to the highest acoustic impedance values
associated with bone tissue.29 The best image quality is
observed with the highest acoustic impedance values.186 As
the transducer receives the reflected beam, the echo is converted into electric signals and a 2-dimensional image is
formed.29 The scanning procedure itself is deceptively simple;
the difficulties arise when it comes to interpreting the image to
determine muscle thickness at the site(s) of measurement. The
interface between adipose tissue and muscle can be difficult to
differentiate, because the acoustic impedance of these tissues
is somewhat similar, and they produce a weaker echo compared with one that would include interface with bone tissue.29
One must identify the tissue boundaries and then measure the
thickness of muscle using the electronic calipers that are standard with most systems; boundary identification is somewhat
subjective and can lead to interrater errors.
In addition, ultrasound measurements of muscle may be made
with the muscle in the contracted or relaxed state, in the standing
or supine position.198,199 For most clinical settings, the individual
will likely be in the supine position and the muscle will be relaxed;
in the relaxed state, muscle tissue is compressible.199 There is not
yet consensus on the optimal protocol to follow in terms of the
degree of pressure to apply when taking a measurement in the
supine, relaxed position. Most published studies have used no
or minimal191,193,195–197,200–204 tissue compression, but one recent
report used maximal205 compression. Although maximal pressure has been advocated as a way to counter the problem of
edema interfering with measurements in clinical populations,205
it has also been suggested that minimal or no compression may
better allow for the identification and perhaps even quantification of edema, as has been observed in measures of the anterior
thigh muscle thickness.189 However, the application of no/minimal transducer compression on the muscle tissue may actually
reduce the ability to detect small lean tissue changes in the presence of edema and/or adipose stranding. Furthermore, the no/
minimal pressure approach in individuals with obesity will
quite likely be ineffectual in visualizing full tissue thickness
and reference bone because of shear limb thickness exceeding
ultrasound depth capabilities. On the other hand, with maximal
compression, the degree of muscle compressibility and presence of edema may contribute to error,29,199 and the degree of
pressure placed on the transducer against the skin is difficult to
control to standardize measurements, making reproducibility
difficult to achieve for longitudinal measurements. Maximal vs
no/minimal transducer force will yield quite different muscle
thickness measures, although this has not yet been well studied.
In the adipose ultrasound literature, it was observed that the use
of maximal transducer force reduced the measurements of subcutaneous adipose thickness by 24%–37% compared with minimal transducer force.206 These are vitally important issues that
need to be resolved so that consensus on protocol can be reached
and appropriate reference data can be generated for use in clinical populations.
A limited number of studies that have investigated the
validity of ultrasound for quantifying muscle mass are relevant
to the clinical setting. Ultrasound measurements of the biceps,
forearm, and midthigh muscles (degree of transducer force was
not described) were shown to correlate reasonably well (r =
.872; P < .001) with LST measured by DXA in individuals
with multiple-organ failure.189 The intraobserver variability for
midthigh measurements was observed to be 2.3% (CV), and
the interobserver CV for the midthigh was observed to be
4.3%. In another study, the thickness of the quadriceps femoris
muscle was negatively correlated with length of stay in individuals in the ICU; interrater CV was observed to be 1.3%
using no/minimal compression.190 Ultrasound of the quadriceps muscle using no/minimal compression has been shown to
correlate well with quadriceps muscle strength in individuals
with chronic obstructive pulmonary disease203 and has also
been shown to correlate reasonably well (r = 0.78; no P value
provided) with MRI measurements in healthy individuals
undergoing a bedrest study.200
Two recent studies have evaluated the reliability and reproducibility of ultrasound measurements of the quadriceps muscle thickness in the setting of critical illness.191,205 Following a
maximal compression protocol207 (see Figures 7 and 8),
Tillquist et al205 reported good intrarater (average ICC = 0.98
with between-subject variance of 0.26 and within-subject variance of .005) and interrater (average ICC = 0.95 with betweensubject variance of 0.26 and within-subject variance of 0.015)
reliability, with average muscle thicknesses for the left and
right leg of 2.01 ± 0.52 cm for trainer 1, 2.0 ± 0.50 cm for
trainer 2, and 2.09 ± 0.52 cm for the trainee. More recently,
following a minimal compression protocol, another group
reported excellent intrarater CVs of 1.91% and 1.32% with
absolute variability expressed as the limits of agreement of
0.33 and 0.12 cm for operators 1 and 2, respectively. The absolute median interrater variability was 0.05 cm, with an ICC of
0.99 for single measurements; the mean muscle thickness over
86 duplicate measurements for the 14 participants was 2.10 ±
0.85 cm.
Ultrasound holds great promise, given its portability, relatively low cost, noninvasiveness, and clinical availability; ultrasound devices are prolifically abundant in ICUs and other units
of most hospitals, and it takes less than 15 minutes to complete a
measurement. Further research is needed to develop consensus
on the optimal way to conduct ultrasound measurements to
assess lean tissue with minimal error in clinical populations
where edema, adipose stranding within muscle, and obesity in
general are important considerations. The optimal site(s) for
ultrasound measurement, the level of compression force to minimize error (ie, maximal vs no or minimal compression), and the
optimal application of the muscle thickness measures for lean
tissue assessment have yet to be established; furthermore, the
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
••
••
••
••
••
••
••
••
••
••
••
••
••
••
••
811
Please see the excellent training video on how to conduct ultrasound using maximal compression, created by Dr. Daren
Heyland’s research group (Clinical Evaluation Research Unit at the Kingston General Hospital, Kingston, Ontario, Canada):206
http://www.criticalcarenutrition.com/index.php?option=com_content&view=article&id=182&Itemid=79
Obtain access to a portable ultrasound machine that uses a high frequency linear transducer.
Obtain access to a flexible tape measure.
Have the patient lay supine with the knees extended and relaxed.
Locate the top of the patella and the anterior superior iliac spine (ASIS).
Identify two points for measurements of the quadriceps femoris muscle (see Figure 8A):
o Reading #1 will be taken at the border between the lower third and upper two thirds of the ASIS and the upper
pole of the patella.
o Reading #2 will be taken at the midpoint between the ASIS and the upper pole of the patella.
o Note: In individuals with obesity, it may be very difficult to locate the ASIS. In such cases, note the distance from
the upper border of the patella for all subsequent measures to be sure you are consistently coming back to the
same place.
Apply a generous amount of ultrasound gel to the area of the thigh that is to be evaluated.
Use a large (5 mHz) ultrasound probe and hold the probe perpendicular to the skin.
Start at maximum depth to identify the femur. Use the depth button to set the electronic focus depth at the shallowest depth
allowable to see the femur.
Set the frequency at maximum allowable frequency.
For taking a measurement using maximal compression: Compress the probe maximally to measure muscle thickness directly
anterior to the shaft of the femur.
For taking a measurement using no compression: Position the scanning probe on the skin surface at the area of the thigh
to be evaluated, and visualize the femur without compressing the skin. In individuals with excessive adiposity, it may
not be possible to use no compression.195 In that event, attempt to compress the tissue only as needed to visualize the
femur, and gradually reduce the pressure until the outer margins of the femur are no longer visible before taking the
measurement.
Muscle measurement should include the area anterior to the femur and proximal to the adipose tissue.
Take readings at the two points as described above (see Figure 8B):
o Measurement (roller ball or cursor); highlight caliper on screen.
o Mark depth @ surface of femur (press Set).
 Make sure measurement line is perpendicular.
o Mark proximal depth of muscle (press Set) up the fascia below the adipose tissue.
o Check top left of screen for measurements.
Repeat the measurement for the other thigh, and then repeat the entire process on both sides; average the measurements and
record for longitudinal tracking.
Figure 7. Sample protocol for taking ultrasound measurements of the quadriceps muscle.205
use of simple muscle thickness measurements alone or the application of these measurements to equations to generate wholebody lean tissue values will certainly require the generation of
population-specific reference data for their interpretation.
Finally, once consensus on optimal protocol is reached and reference data are generated, additional research is needed to evaluate the role of ultrasound in bedside assessment of lean tissue
and diagnosis of malnutrition in the hospital setting.
Applications to Clinical Case Scenarios
Scenario 1
A 67-year-old African American man is admitted to the hospital after experiencing a fall at home. He is diagnosed with a hip
fracture. He admits to not eating very well over the past 6
months but is not sure if he has lost any weight. He has noticed
that he is having difficulty taking care of chores around the
house and is slower getting out of his chair than usual. No visible signs of muscle or fat loss are evident upon examination.
Height: 188 cm; current weight: 128.6 kg; BMI: 39.0 kg/m2
How might a bioimpedance and/or ultrasound measurement
help us to evaluate this patient’s nutrition status? Although it
is not yet standard practice, and there is a need for additional
research to clearly establish how bioimpedance and/or ultrasound measurements might uniquely contribute to the assessment of nutrition status in the hospital setting, particularly in
terms of what they tell us about muscle mass, we can
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
812
Journal of Parenteral and Enteral Nutrition 39(7)
Figure 8. Ultrasound of the quadriceps muscle. (A) Quadriceps muscle layer thickness measurements. Image illustrates the location
for ultrasound readings. Reprinted with permission from Tillquist M, Kutsogiannis DJ, Wischmeyer PE, et al. Bedside ultrasound
is a practical and reliable measurement tool for assessing quadriceps muscle layer thickness. JPEN J Parenter Enteral Nutr.
2013;38(7):886-890. (B) Correct and incorrect positioning of the calipers for a quadriceps muscle layer thickness measurement. Image
reprinted from the VALIDUM study’s ultrasound protocol with permission from Daren Heyland’s research group at the Clinical
Evaluation Research Unit.207
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
813
hypothetically consider how these data might be used in a particular clinical case.
difficulty with activities of daily living, this information
could be considered corroborative of some degree of muscle
loss and/or nutrition compromise.
Using bioimpedance
A. With any type of bioimpedance device (SF-BIA,
MF-BIA, BIS), we could use the 50-kHz impedance (R
and X) data to calculate Janssen’s SMI equation to evaluate for sarcopenia, although it is important to understand that the use of this equation for bedside assessment
in this way has not been validated. In addition, we should
recognize that the equation was developed from a fairly
small population sample, using a particular SF-BIA
device (ie, RJL), and if we use a different device, some
error may result. Although the equation and its reference
cutpoints have been published as diagnostic for sarcopenia, its use for bedside assessment has not been established and would need to be further evaluated before it
can be recommended as part of standard nutrition assessment practice. As an exercise, we can consider how this
might be used as one component of a comprehensive
evaluation of nutrition status. In addition, we could use
our 50-kHz data to evaluate his PA and SPA.
From the 50-kHz measurement, we have the following:
R = 412 ohms
X = 29.6 ohms
Z = 413 ohms
PA = 4.1°
1. Calculate SMI using the Janssen equation12:
Skeletal muscle mass ( kg ) = [( Height 2 / R × 0.401) +
(Sex × 3.825) + (Age × − 0.071)] + 5.102
SM = [(1882 / 412 × 0.401) + (1 × 3.825) +
(67 × − 0.071)] + 5.102 = 38.6 kg.
SMI=
(height )
3=
8.6 / 3.5
11 kg/m 2 .
SMI (% weight ) = 38.6 / 128.6 × 100 = 30%.
Interpretation: Referring to Table 1, we can compare our calculated values with the SMI by published SF-BIA cutpoints,
recognizing the limitations of applying it in this way.
Assuming for the moment that this has been validated for
assessing muscle loss at the bedside, based on the SMI
(height) cutpoint, his value of 11 kg/m2 would not suggest
that he is sarcopenic, although he is not far from the 10.75
cutpoint for moderate risk of disability based on that method.
Based on the SMI (%weight) cutpoint, his value of 30% does
suggest that he may be sarcopenic, given that values <31%
indicate class II sarcopenia. Given that he has been having
2. Evaluate his PA:
Interpretation: To interpret PA, it is most appropriate to use reference data generated from a population that most closely resembles
your patient in terms of ethnic background and other attributes.
Ideally, we would like to have PA reference cutpoints generated by
the same device that we are using and from an American reference
population such as NHANES. Given that we do not currently
have NHANES PA data available, the best option we have is the
reference data generated in American individuals published by
Barbosa-Silva et al.139 Referring to Table 3, we can compare his
value of 4.1 with the non-BMI-specific cutpoint of 5.4, corresponding to the fifth percentile value for 60- to 69-year-olds (who
were overweight on average). By these criteria, he has a low PA,
which could indicate malnutrition. For the purposes of this hypothetical exercise, given the very high BMI of our patient, we might
also look at the Bosy-Westphal et al138 BMI-specific reference
cutpoint representing the fifth percentile value for 60- to 69-yearolds, with BMI between 35.1 and 40, which is 4.7. Based on this
cutpoint, we would also conclude that he has a low PA and may
have some degree of malnutrition, although it may not be appropriate to apply the German white population cutpoints to our
African American patient. We can also calculate his SPA.
3. Calculate SPA using the following equation:
SPA = (Observed PA − Mean PA) / SD,
where the mean and SD are from reference values.
Referring to Table 4, we can see from the Barbosa-Silva
et al139 reference data for 60- to 69-year-olds that
SPA (Barbosa-Silva et al139 ) = (4.1 − 7.0) / 1.1 = − 2.45.
If we calculate the SPA using the Bosy-Westphal et al138 reference data for 60- to 69-year-old men with a BMI between 35.1
and 40 (mean PA is 6.0 and the SD is 0.85),
SPA (Bosy-Westphalet al138 ) = (4.1 − 6.0) / 0.85
= − 2.24.
Interpretation: Given that SPA values below –1.65 are suggested to indicate malnutrition, both of our calculated SPA values, regardless of the minor differences between them, suggest
that he may be malnourished.
B. If we have an MF-BIA or BIS device, we can then also
evaluate the impedance ratio at 200/5 kHz.
From the MF-BIA or BIS measurement, we have the
following:
Z at 200 kHz = 383
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
814
Journal of Parenteral and Enteral Nutrition 39(7)
Z at 5 kHz = 453
Impedance ratio = 0.85
Interpretation: We have less to go on in terms of reference
data for the 200/5-kHz impedance ratio; however, values ≤0.78
have been observed in healthy men.171 Furthermore, in men
from various research studies, mean values of 0.81 have been
associated with the development of postoperative edema,208
values >0.85 with worsening renal function in individuals with
heart failure,174 and values >0.78 with increasing odds of having malnutrition.171 In that light, we might reasonably interpret
this value as corroborative of the other bioimpedance data that
we have, in that it could indicate nutrition compromise and/or
altered fluid status that could occur with lean tissue loss.
Putting all the bioimpedance data together, along with this
patient’s history of altered functional status, we might conclude that he has at least some degree of nutrition compromise
and may in fact have sarcopenic obesity.
Using ultrasound
Quadriceps muscle thickness measured with maximal compression: 2.05 cm
Quadriceps muscle thickness measured with minimal compression: 4.2 cm
Given that there is currently no consensus on the optimal protocol to follow to take an ultrasound measurement, and thus we do
not yet have any reference data to draw from, at this point in time
it is difficult to know how we might best use an ultrasound measurement in our patient. In the interest of this hypothetical case,
however, we might consider using ultrasound measurements to
track changes in lean tissue (through the measurement of quadriceps muscle thickness, following a consistent protocol) over time,
after taking a baseline measurement upon admission. For the purposes of illustration, you can see that when we used a maximal
compression protocol, his quadriceps muscle thickness was 2.05
cm, compared with 4.2 cm when we applied minimal compression. Thus, it is vitally important that we use the same protocol for
follow-up measurements if we want to track muscle changes. At
this point, our measurements cannot be easily interpreted because
we have no reference data for either method, but we will remeasure the quadriceps muscle thickness every week while he is in the
hospital to monitor, following consistent protocol.
necessary to get by. She has lost weight since starting treatment,
although she is not certain how much. She appears to have some
loss of fat from the orbital area but does not exhibit any other
notable signs of muscle or fat loss upon physical examination.
Height: 160 cm; current weight: 70.8 kg; BMI: 28 kg/m2
How might a bioimpedance and/or ultrasound measurement
help us to evaluate this patient’s nutrition status?
Using bioimpedance A. With any type of bioimpedance device (SF-BIA,
MF-BIA, BIS), we could use the 50-kHz impedance (R
and X) data to calculate Janssen’s SMI equation (and the
aforementioned caveats apply), and we might also look
at her PA and SPA.
From the 50-kHz measurement, we have the following:
R = 447 ohms
X = 17 ohms
Z = 447 ohms
PA = 2.2°
1. Calculate SMI using the Janssen equation12:
Skeletal muscle mass ( kg ) = [(Height 2 / R × 0.401) +
(Sex × 3..825) + (Age × − 0.071)] + 5.102.
SM = [(1602 / 447 × 0.401) + (0 × 3.825)
+ (51 × − 0.071)] + 5.102 = 24.5 kg.
SMI (=
height ) 24
=
.5 / 2.6
9.4 kg/m 2 .
SMI (% weight ) = 24.5 / 70.8 × 100 = 34.6%.
Interpretation: Referring to Table 1, we can compare our calculated values with the SMI by SF-BIA cutpoints, with the same reservations as previously discussed. For the purposes of this exercise,
based on the SMI (height) cutpoint, her value of 9.4 kg/m2 does not
suggest that she is sarcopenic. Based on the SMI (%weight) cutpoint, her value of 34.6% also does not suggest sarcopenia.
2. Evaluate her PA:
Scenario 2 A 51-year-old white woman of European descent is receiving
outpatient chemotherapy in combination with radiation therapy
for the treatment of throat cancer. She is in her fourth week of
treatment. Her intake has been quite poor, and she complains of
sores in her mouth. She is generally fatigued and tires easily on
exertion. She has stopped her daily walks (since the first week of
treatment) and now tries to keep activity to the minimum
Interpretation: Referring to Table 3, we can compare her value
of 2.2 with the non-BMI-specific cutpoint of 5.5, corresponding to the fifth percentile value for 50- to 59-year-olds (who
were overweight on average). By these criteria, she has a low
PA, which could indicate malnutrition. For the purposes of this
hypothetical exercise, given the BMI and white ethnicity of our
patient, we might also look at the Bosy-Westphal et al138 BMIspecific reference cutpoint representing the fifth percentile
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
815
value for 50- to 59-year-olds, with BMI between 25.1 and 30,
which is 4.9. Based on this cutpoint, we would also conclude
that she has a low PA and may have some degree of malnutrition. We can also calculate her SPA.
3. Calculate SPA using the following equation:
SPA = (Observed PA − Mean PA) / SD,
where the mean and SD are from reference values.
She is 51 years old, with a BMI of 28.7, so referring to Table 4,
we can see from the Barbosa-Silva et al139 reference data for
50- to 59-year-olds that
SPA (Barbosa-Silva et al139 ) = (2.2 − 6.6) / 0.87 = − 5.06.
If we calculate the SPA using the Bosy-Westphal et al138 reference data for 50- to 59-year-old women with a BMI between
25.1 and 30 (mean PA is 5.9 and the SD is 0.70):
SPA (Bosy-Westphalet al138 ) = (2.2 − 5.9) / 0.70
= − 5.29.
Interpretation: Given that SPA values below –1.65 have been
suggested to indicate malnutrition, both of our calculated SPA
values, regardless of the minor differences between them, suggest that she may be malnourished.
B. If we have an MF-BIA or BIS device, we can then also
evaluate the impedance ratio at 200/5 kHz.
From the MF-BIA or BIS measurement, we have the
following:
Z at 200 kHz = 431
Z at 5 kHz = 468
Impedance ratio = 0.92
As stated earlier, we do not have large sample healthy reference data for the 200/5-kHz impedance ratio; however, values
≤0.82 have been observed in healthy women.171 Given what
has been observed in the literature with impedance ratios above
this cutpoint, our patient’s value of 0.92 could be interpreted as
suggesting altered nutrition status and/or altered hydration
status.
Taken together, the bioimpedance data could be interpreted
to indicate that this patient has at least some degree of muscle
loss and/or nutrition compromise.
Using ultrasound
Quadriceps muscle thickness measured with maximal compression: 1.7 cm
Quadriceps muscle thickness measured with minimal compression: 2.8 cm
In this individual, we measured a quadriceps muscle thickness using maximal compression of 1.7 cm and 2.8 cm using
minimal compression. As mentioned in the first case scenario,
we do not currently have appropriate reference data to interpret
this cross-sectional measurement at this point in time, but we
could monitor our patient weekly to observe for any changes,
following a consistent protocol.
What about using bioimpedance or ultrasound to generate
whole-body lean tissue estimates at the bedside? Generally
speaking, it is not easy to know how best to obtain an accurate
measure of whole-body FFM, BCM, or LST. At present, we do
not have validated equations for predicting whole-body lean
mass from ultrasound-measured quadriceps thickness. With
regard to bioimpedance techniques, if we have access to a BIS
device, we could consider generating the “normally hydrated
lean tissue” (assuming we have a device that has appropriate
software to model the data and calculate it from ECW and ICW
values generated from the BMI-modified Xitron-Hanai–based
mixture equations). Absolute accuracy of whole-body lean tissue measurements is less critical if we are interested in monitoring changes over time, although small changes may not be
detectable given conservative estimates of 3%–8% MDC for
bioimpedance techniques. It is important to recognize that all
body composition measurement involves some degree of error
due to biological variation and measurement error caused by
protocol violations and human error, as well as device- and
equation-specific error. If we wish to use our SF-BIA or MFBIA device to evaluate whole-body lean tissue, we can refer to
published sources78,96,104 and, if needed, search the literature to
find a population-specific, validated equation appropriate for the
individual we wish to assess, to calculate FFM, BCM, or LST;
however, this is not always an easy task, and whole-body estimates are difficult to interpret, given potential violations of
underlying assumptions, particularly in acute illness. If we
choose to evaluate whole-body measures, our best option is to
use the same device with the same predictive equation and
standardize the testing conditions as much as possible for longitudinal measures (see Table 2) to minimize error.
Caution regarding interpretation
•• W
e must exercise extreme caution when interpreting
whole-body values coming from a device that does not
provide the source of the equation used to predict the
values or from a known equation that does not well
match the characteristics of our patient.
•• Note that we do not apply reference cutpoints generated
by one bioimpedance or reference method to interpret a
body composition compartment or bioimpedance
parameter measured by a completely different method
(eg, do not use the Geneva equation-generated FFMI
reference cutpoints to evaluate your patient’s segmental
MF-BIA-generated FFM value).
•• We should understand that the interpretation of these
various bioimpedance parameters, including PA,
impedance ratio, SMI, and FFMI, is subject to the limitations of available reference data, the potential error
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
816
Journal of Parenteral and Enteral Nutrition 39(7)
caused by device-specific and population variable-specific differences between the various research laboratories producing them, and the lack of clarity over exactly
what is being measured by these parameters. We must
interpret findings in light of other clinical data at our
disposal to build the most complete picture that we can
of our patient’s nutrition status.
Summary and Call to Action: Future
Research Needs Because of the key role that the body’s lean tissue reserves play
in the response to injury and illness, it is central to the maintenance of good nutrition status; this concept has been incorporated into the diagnostic criteria for malnutrition developed by
AND and A.S.P.E.N.15 Clearly, given the implications that the
loss of lean tissue has on clinical outcomes in acute and chronic
illness, its assessment is of critical importance to the nutrition
field. Our ability to measure lean tissue at the bedside has been
limited by the lack of available methods that can be considered
sufficiently accurate for whole-body lean tissue estimates in
the clinical setting, although absolute accuracy may not be the
only consideration in the application of bedside techniques as
monitoring tools. Many have advocated the use of subjective
physical examination techniques to evaluate muscle loss in the
clinical setting. The application of these more subjective techniques is likely to be limited by lack of precision and sensitivity, although research is needed to validate nutrition-focused
physical examination techniques (applied using standardized
training protocols) against more objective measures. To identify muscle loss in a severely cachectic individual through
nutrition-focused physical examination techniques16 is likely
to be far less challenging than it would be in an overweight
individual dealing with acute or chronic illness. Indeed, the
detection of sarcopenia in the presence of overweight or obesity presents a significant and important challenge to the
assessment of malnutrition today and is likely to require objective measurement techniques (eg, CT, as observed in the recent
study by Sheean et al19). Whether or not existing bedside techniques (eg, bioimpedance and ultrasound) can be used to routinely evaluate lean tissue loss (and thus nutrition status) in the
clinical setting remains to be determined.
It is hoped that through this tutorial, the reader has gained an
appreciation for the complexities involved in the validation of
bedside body composition assessment methods and for the different ways that bioimpedance data are generated and applied.
There is substantial and growing interest worldwide in the
potential utility of simple bioimpedance variables to identify
malnutrition, apart from the estimation of whole-body lean tissue compartments that have sometimes yielded wide variation in
individual measures compared with reference methods, particularly in individuals with extreme obesity and other populations
where assumptions may be violated. Although these violations
are likely to yield some degree of error in the bioimpedance
measurements, the precision (ie, from repeated measurements)
of the reference method is not always evaluated in research studies, particularly those using dilution methods; lower precision in
the reference method could contribute error to the methods comparison that may be inadvertently interpreted as solely coming
from the bioimpedance method. A consensus on how best to
evaluate the validity of a bedside method in terms of limits of
agreement PE and measurement error in light of reference
method precision (among other possible criteria) needs to be
reached. Further refinements of the BIS technique as applied in
the dialysis population184 to quantify “excess fluid” may yield
improvements in the “normally hydrated lean tissue” wholebody estimates; in that event, reference data would need to be
generated and made available. SMI by the Janssen equation,137
FFMI by the Geneva equation,135 and FFM by the Chumlea
NHANES III equation98 are perhaps the best candidates for
equations that could be cross-validated from the most recent
NHANES data; reference data could be generated from
NHANES for those equations proven to be most valid to provide
useful cutpoints for sarcopenia based on American population
data, ideally by BMI, age, and ethnicity. Similarly, reference
data for the 50-kHz PA and 200/5-kHz (and also Z∞/Z0 if available) impedance ratio could be generated from the most recent
NHANES data set. For raw bioimpedance parameters to be useful at the bedside, a consensus on normal reference values and
cutpoints to define sarcopenia and/or malnutrition is needed to
facilitate interpretation; this may require bioimpedance device
companies to come together to somehow cross-calibrate and/or
streamline technologies. In addition, recent developments in the
ultrasound field are making that method an attractive option for
bedside nutrition assessment. It will need additional refinement
in terms of standardizing measurement protocols to minimize
inter- and intraobserver errors so that good precision can be consistently achieved. In addition, reference values for normal muscle thickness as well as cutpoints to define sarcopenia will need
to be established. How well ultrasound and/or any of the available bioimpedance techniques can contribute to nutrition assessment at the bedside (to produce whole-body estimates of lean
tissue, to identify muscle loss [ie, sarcopenia], and/or to produce
useful markers of malnutrition) is a vitally important question
that remains to be answered.
Call to Action The global clinical nutrition community needs to work
together to come to consensus on the optimal tool(s) to use
to assess nutrition status at the bedside. Clinician researchers
who are interested in validating the AND/A.S.P.E.N. consensus malnutrition diagnostic criteria should strive to include
at least one reference technique for lean tissue (eg, CT and/
or DXA) to validate the subjective nutrition-focused physical examination method of evaluating muscle loss. In addition, at least one and ideally both of the currently available
bedside techniques, ultrasound and bioimpedance (eg, an
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
817
MF-BIA and/or BIS device can produce PA and an impedance ratio; a BIS device appropriate to provide whole-body
“normally hydrated lean tissue” and other whole-body estimates in addition to the simpler raw BIA parameters would
have added benefit) should be used in the same study to
evaluate their ability to contribute uniquely to the identification of lean tissue loss and malnutrition. Collaboration and
open sharing of data among researchers and cooperation
between device manufacturers to streamline technologies
and facilitate the production of normal reference data would
allow for greater progress.
Two key questions that must be answered are:
1. Can a standardized subjective nutrition-focused physical examination method accurately detect muscle loss
in individuals across the acute/chronic illness and
weight spectrum compared with objective reference
measures (eg, CT, DXA)?
2. In these same clinical populations, can ultrasound or a
particular bioimpedance approach help us to identify
and monitor subtle changes in lean tissue (and thus
nutrition status) compared with reference measures
that would not have otherwise been possible through
other, more subjective assessment methods?
Acknowledgments
The author thanks David Frankenfield for sharing his insights on
precision, accuracy, and bias as they relate to the comparison of
energy expenditure methodologies, as well as how to discuss the
interpretation of Bland-Altman and linear regression plots in terms
of precision and bias. Figures 1 and 2 were developed from his
unpublished figures, with his permission. The author also recognizes Daren Heyland, Marina Mourtzakis, and Michael Paris for
their input on ultrasound techniques and Adam Kuchnia, Urvashi
Mulasi, Levi Teigen, and Abigail Cole for their input and reflection on the presentation of key concepts. In addition, the author
thanks the reviewers who gave their time and care in providing
feedback and input on this manuscript.
Statement of Authorship
C. P. Earthman was responsible for the conception/design of the
research; was responsible for the acquisition, analysis, and interpretation of the data; drafted the manuscript; critically revised the
manuscript; agrees to be fully accountable for ensuring the integrity and accuracy of the work; and read and approved the final
manuscript.
Glossary
Accuracy: refers to the closeness of agreement in a particular variable
between 2 assessment methods.
BCM: body cell mass.
Bias: refers to the systematicerror, determined by the average differences in
a particular measurement variable between the bedside and reference
method.
BIS: bioimpedance spectroscopy.
Cachexia: among many definitions: significant weight loss, protein catabolism, and muscle and fat tissue loss that occur due to underlying disease processes.
ECW: extracellular water.
FFM: fat-free mass.
FFMI: fat-free mass index.
Fixed bias: refers to the type of systematic error that occurs when a method
yields measurements of a particular variable that are consistently
higher or lower than those taken by the reference method.
ICW: intracellular water.
Impedance ratio: most commonly studied is the ratio of impedance measured at 200 kHz to 5 kHz; other high- to low-frequency impedance
ratios may be useful, including Z∞/Z0 from BIS measurements.
LST: lean soft tissue (term used in the DXA literature for the bone-free
FFM, which is also called lean body mass).
MF-BIA: multiple-frequency bioelectrical impedance analysis.
PA: phase angle (most commonly studied is the PA measured at 50 kHz).
Precision: refers to the degree of agreement among repeated measurements
by a method, determined by repeatability and reproducibility.
Proportional bias: refers to the type of systematic error that is proportional
to the value of the variable being measured.
Repeatability: refers to the degree of agreement between repeated, independent measurements of particular variable made by the same operator using a single method in the same individual under the same
conditions; is affected by intraobserver variability, among other
factors.
Reproducibility: refers to the degree of agreement in measurements of a
particular variable taken by a single method but by different operators; is affected by interobserver variability, among other factors.
Sarcopenia: loss of muscle mass with loss of muscle strength and/or
function.
SF-BIA: single-frequency bioelectrical impedance analysis.
SMI: skeletal muscle mass index.
TBW: total body water.
Reading List
1.Ellis KJ. Human body composition: in vivo methods. Physiol Rev.
2000;80(2):649-680.
2. Baracos V, Caserotti P, Earthman CP, et al. Advances in the science and
application of body composition measurement. JPEN J Parenter Enteral
Nutr. 2012;36(1):96-107.
3. Prado CMM, Heymsfield SB. Lean tissue imaging: a new era for nutritional
assessment and intervention. JPEN Parenter Enteral Nutr. 2014;38:940-953.
4. Mulasi U, Kuchnia AJ, Cole AJ, Earthman CP. Bioimpedance at the bedside: current applications, limitations, and opportunities. Nutr Clin Pract.
2015;30(2):180-193.
5. Matthie JR. Bioimpedance measurements of human body composition:
critical analysis and outlook. Expert Rev Med Devices. 2008;5(2):239-261.
6. Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical impedance analysis—part I: review of principles and methods. Clin Nutr. 2004;23(5):12261243.
7. Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical impedance analysis—part II: utilization in clinical practice. Clin Nutr. 2004;23(6):14301453.
8. Lukaski HC. Evolution of bioimpedance: a circuitous journey from estimation of physiological function to assessment of body composition and a
return to clinical research. Eur J Clin Nutr. 2013;67(suppl 1):S2-S9.
9. Kushner RF, Gudivaka R, Schoeller DA. Clinical characteristics influencing biolectrical impedance analysis measurements. Am J Clin Nutr.
1996;64(suppl):423S-427S.
10. Norman K, Stobäus N, Pirlich M, Bosy-Westphal A. Bioelectrical phase
angle and impedance vector analysis: clinical relevance and applicability
of impedance parameters. Clin Nutr. 2012;31(6):854-861.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
818
Journal of Parenteral and Enteral Nutrition 39(7)
References
1. Hoffer LJ, Bistrian BR. Appropriate protein provision in critical illness : a
systematic and narrative review. Am J Clin Nutr. 2012;96:591-600.
2. Biolo G, Toigo G, Ciocchi B, et al. Metabolic response to injury and sepsis: changes in protein metabolism. Nutrition. 1997;13(9):52S-57S.
3. Muscaritoli M, Anker SD, Argilés J, et al. Consensus definition of sarcopenia, cachexia and pre-cachexia: joint document elaborated by Special
Interest Groups (SIG) “cachexia-anorexia in chronic wasting diseases”
and “nutrition in geriatrics.” Clin Nutr. 2010;29(2):154-159.
4.Cosquéric G, Sebag A, Ducolombier C, Thomas C, Piette F, WeillEngerer S. Sarcopenia is predictive of nosocomial infection in care of the
elderly. Br J Nutr. 2006;96:895-901.
5.Lieffers JR, Bathe OF, Fassbender K, Winget M, Baracos VE.
Sarcopenia is associated with postoperative infection and delayed
recovery from colorectal cancer resection surgery. Br J Cancer.
2012;107(6):931-936.
6.Pichard C, Kyle UG, Morabia A, Perrier A, Vermeulen B, Unger P.
Nutritional assessment: lean body mass depletion at hospital admission is associated with an increased length of stay. Am J Clin Nutr.
2004;79(4):613-618.
7. Moisey LL, Mourtzakis M, Cotton B, et al. Skeletal muscle predicts ventilator-free days, ICU-free days, and mortality in elderly ICU patients. Crit
Care. 2013;17(5):R206.
8.Montano-Loza AJ, Meza-Junco J, Prado CMM, et al. Muscle wasting is
associated with mortality in patients with cirrhosis. Clin Gastroenterol
Hepatol. 2012;10(2):166-173.
9.Parsons HA, Baracos VE, Dhillon N, Hong DS, Kurzrock R. Body composition, symptoms, and survival in advanced cancer patients referred to a
phase I service. PLoS One. 2012;7(1):e29330.
10.Baumgartner RN, Koehler KM, Gallagher D, et al. Epidemiology
of sarcopenia among the elderly in New Mexico. Am J Epidemiol.
1998;147(8):755-763.
11. Prado CM, Lieffers JR, McCargar LJ, et al. Prevalence and clinical implications of sarcopenic obesity in patients with solid tumours of the respiratory and gastrointestinal tracts: a population-based study. Lancet Oncol.
2008;9:629-635.
12. Janssen I, Heymsfield SB, Ross R. Low relative skeletal muscle mass (sarcopenia) in older persons is associated with functional impairment and
physical disability. J Am Geriatr Soc. 2002;50(5):889-896.
13. Janssen I, Baumgartner RN, Ross R, Rosenberg IH, Roubenoff R. Skeletal
muscle cutpoints associated with elevated physical disability risk in older
men and women. Am J Epidemiol. 2004;159(4):413-421.
14. Fearon K, Strasser F, Anker SD, et al. Definition and classification of cancer cachexia: an international consensus. Lancet Oncol. 2011;12(5):489495.
15.White J, Guenter P, Jensen G, Malone A, Schofield M. Consensus statement: Academy of Nutrition and Dietetics and American Society for
Parenteral and Enteral Nutrition: characteristics recommended for the
identification and documentation of adult malnutrition (undernutrition).
JPEN J Parenter Enteral Nutr. 2012;36(3):275-283.
16.Malone A, Hamilton C. The Academy of Nutrition and Dietetics/
the American Society for Parenteral and Enteral Nutrition consensus
malnutrition characteristics: application in practice. Nutr Clin Pract.
2013;28(6):639-650.
17. Cruz-Jentoft AJ, Baeyens JP, Bauer JM, et al. Sarcopenia: European consensus on definition and diagnosis. Age Ageing. 2010;39(4):423.
18.Yamada Y, Schoeller D, Nakamura E, Morimoto T, Kimura M, Oda
S. Extracellular water may mask actual muscle atrophy during aging. J
Gerontol A Biol Sci Med Sci. 2010;65(5):510-516.
19.Sheean PM, Peterson SJ, Gomez Perez S, et al. The prevalence of sarcopenia in patients with respiratory failure classified as normally nourished
using computed tomography and subjective global assessment. JPEN J
Parenter Enteral Nutr. 2014;38(7):873-879.
20.Detsky A, McLaughlin JR, Baker J, et al. What is subjective global
assessment of nutritional status? JPEN J Parenter Enteral Nutr.
1987;11:8-13.
21.Prado CMM, Wells JCK, Smith SR, Stephan BCM, Siervo M.
Sarcopenic obesity: a critical appraisal of the current evidence. Clin Nutr.
2012;31(5):583-601.
22. Gallager D, DeLegge M. Body composition (sarcopenia) in obese patients:
implications for care in the intensive care unit. JPEN J Parenter Enteral
Nutr. 2011;35(1):21S-28S.
23.Martin L, Birdsell L, Macdonald N, et al. Cancer cachexia in the age of
obesity: skeletal muscle depletion is a powerful prognostic factor, independent of body mass index. J Clin Oncol. 2013;31(12):1539-1547.
24.Heymsfield SB, Wang Z, Baumgartner RN, Ross R. Human body composition: advances in models and methods. Annu Rev Nutr. 1997;17:
527-558.
25. Pietrobelli A, Heymsfield SB, Wang ZM, Gallagher D. Multi-component
body composition models: recent advances and future directions. Eur J
Clin Nutr. 2001;55:69-75.
26.Wang Z, Deurenberg P, Wang W, Pietrobelli A, Baumgartner RN,
Heymsfield SB. Hydration of fat-free body mass: new physiological modeling approach. Am J Physiol. 1999;276(6, pt 1):E995-E1003.
27.Das SK, Roberts SB, Kehayias JJ, et al. Body composition assessment in extreme obesity and after massive weight loss induced by
gastric bypass surgery. Am J Physiol Endocrinol Metab. 2003;284(6):
E1080-E1088.
28. Leone PA, Gallagher D, Wang J, Heymsfield SB. Relative overhydration
of fat-free mass in postobese versus never-obese subjects. Ann N Y Acad
Sci. 2000;904:514-519.
29.Prado CMM, Heymsfield SB. Lean tissue imaging: a new era for nutritional assessment and intervention. JPEN J Parenter Enteral Nutr.
2014;38:940-953.
30. Moore FD, Boyden CM. Body cell mass and limits of hydration of the fatfree body: their relation to estimated skeletal weight. Ann N Y Acad Sci.
1963;110:62-71.
31. Wang Z, St-Onge M-P, Lecumberri B, et al. Body cell mass: model development and validation at the cellular level of body composition. Am J
Physiol Endocrinol Metab. 2004;286(1):E123-E128.
32. Shizgal HM. Nutritional assessment with body composition measurements
by multiple isotope dilution. JPEN J Parenter Enteral Nutr. 1990;17(3)
(suppl):9-17.
33. Earthman C, Traughber D, Dobratz J, Howell W. Bioimpedance spectroscopy for clinical assessment of fluid distribution and body cell mass. Nutr
Clin Pract. 2007;22(4):389-405.
34.Gallagher D, Visser M, Wang Z, Harris T, Pierson RN, Heymsfield SB.
Metabolically active component of fat-free body mass: influences of age,
adiposity, and gender. Metabolism. 1996;45(8):992-997.
35.Pierson RN, Wang J, Thornton J, Heymsfield S. The quality of the
body cell mass—1996. Are we ready to measure it? Appl Radiat Isot.
1998;49(5):429-435.
36.Shizgal HM. Nutritional assessment with body composition measurements. JPEN J Parenter Enteral Nutr. 1987;11(5)(suppl):42S-47S.
37. Beddoe A, Streat S, Hill G. Hydration of fat-free body in protein-depleted
patients. Am J Physiol. 1985;249(2, pt 1):E227-E233.
38. James HM, Dabek JT, Chettle DR, et al. Whole body cellular and collagen
nitrogen in healthy and wasted man. Clin Sci (Lond). 1984;67:73-82.
39.Lee SY, Gallagher D. Assessment methods in human body composition.
Curr Opin Clin Nutr Metab Care. 2008;11(5):566-572.
40.Baracos V, Caserotti P, Earthman CP, et al. Advances in the science and
application of body composition measurement. JPEN J Parenter Enteral
Nutr. 2012;36(1):96-107.
41.Ludbrook J. Comparing methods of measurement. Clin Exp Pharmacol
Physiol. 1997;24(2):193-203.
42. Ulijaszek SJ, Kerr DA. Anthropometric measurement error and the assessment of nutritional status. Br J Nutr. 1999;82:165.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
819
43. Lohman T, Pollock M, Slaughter M, Brandon L, Boileau R. Methodological
factors and the prediction of body fat in female athletes. Med Sci Sports
Exerc. 1984;16(1):92-96.
44. Jackson AS. Research design and analysis of data procedures for predicting body density. Med Sci Sports Exerc. 1984;16(6):616-620.
45.Bosy-Westphal A, Schautz B, Later W, Kehayias J, Gallagher D, Müller
M. What makes a BIA equation unique? Validity of eight-electrode multifrequency BIA to estimate body composition in a healthy adult population. Eur J Clin Nutr. 2013;67(suppl 1):S14-S21.
46. Earthman CP, Matthie JR, Reid PM, Harper IT, Ravussin E, Howell WH.
A comparison of bioimpedance methods for detection of body cell mass
change in HIV infection. J Appl Physiol. 2000;88(3):944-956.
47.Cox-Reijven PL, Soeters PB. Validation of bio-impedance spectroscopy:
effects of degree of obesity and ways of calculating volumes from measured
resistance values. Int J Obes Relat Metab Disord. 2000;24(3):271-280.
48.Cox-Reijven PLM, van Kreel B, Soeters PB. Accuracy of bioelectrical
impedance spectroscopy in measuring changes in body composition during
severe weight loss. JPEN J Parenter Enteral Nutr. 2002;26(2):120-127.
49.Mager JR, Sibley SD, Beckman TR, Kellogg TA, Earthman CP.
Multifrequency bioelectrical impedance analysis and bioimpedance spectroscopy for monitoring fluid and body cell mass changes after gastric
bypass surgery. Clin Nutr. 2008;27(6):832-841.
50.Skipper A, Ferguson M, Thompson K, Castellanos VH, Porcari J.
Nutrition screening tools: an analysis of the evidence. JPEN J Parenter
Enteral Nutr. 2012;36(3):292-298.
51.Tappenden KA, Quatrara B, Parkhurst ML, Malone AM, Fanjiang G,
Ziegler TR. Critical role of nutrition in improving quality of care: an interdisciplinary call to action to address adult hospital malnutrition. JPEN J
Parenter Enteral Nutr. 2013;37(4):482-497.
52.Zou KH, Tuncali K, Silverman SG. Correlation and simple linear regression. Radiology. 2003;227:617-628.
53.Taylor R. Interpretation of the correlation coefficient: a basic review. J
Diagnostic Med Sonogr. 1990;6(1):35-39.
54. Saracino G, Jennings LW, Hasse JM. Basic statistical concepts in nutrition
research. Nutr Clin Pract. 2013;28(2):182-193.
55.Lane DM. Online estatsbook. http://onlinestatbook.com/2/regression/
accuracy.html. Accessed January 27, 2015.
56. Atkinson G, Nevill A. Comment on the use of concordance correlation to assess
the agreement between two variables. Biometrics. 1997;53(2):775-777.
57. Bland JM, Altman DG. Applying the right statistics: analyses of measurement studies. Ultrasound Obstet Gynecol. 2003;22:85-93.
58.Martin RF. General Deming regression for estimating systematic bias
and its confidence interval in method-comparison studies. Clin Chem.
2000;46(1):100-104.
59.Stöckl D, Dewitte K, Thienpont LM. Validity of linear regression in
method comparison studies: is it limited by the statistical model or the
quality of the analytical input data? Clin Chem. 1998;44(11):2340-2346.
60. Passing H, Bablok W. A new biometrical procedure for testing the equality of measurements from two different analytical methods: application
of linear regression procedures for method comparison studies in clinical
chemistry, part I. J Clin Chem Clin Biochem. 1983;21(11):709-720.
61.Passing H, Bablok W. Comparison of several regression procedures for
method comparison studies and determination of sample sizes: application
of linear regression procedures for method comparison studies in clinical
chemistry, part II. J Clin Chem Clin Biochem. 1984;22(6):431-445.
62. Bilic-Zulle L. Lessons in biostatistics comparison of methods: Passing and
Bablok regression. Biochem Med. 2011;21(3):49-52.
63.Bablok W, Passing H. Application of statistical procedures in analytical
instrument testing. J Automat Chem. 1985;7(2):74-79.
64.Linnet K. Evaluation of regression procedures for methods comparison
studies. Clin Chem. 1993;39(3):424-432.
65. Cornbleet PJ, Gochman N. Incorrect least-squares regression coefficients
in method-comparison analysis. Clin Chem. 1979;25(3):432-438.
66. Lin LI. A concordance correlation coefficient to evaluate reproducibility.
Biometrics. 1989;45(1):255-268.
67.Chinchilli VM, Martel JK, Kumanyika S, Lloyd T. A weighted concordance correlation coefficient for repeated measurement designs.
Biometrics. 1996;52(1):341-353.
68.King TS, Chinchilli VM, Carrasco JL. A repeated measures concordance
correlation coefficient. Stat Med. 2007;26(16):3095-3113.
69. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res. 1999;8(99):135-160.
70.Bland JM, Altman DG. Statistical methods for assessing agreement
between two methods of clinical measurement. Lancet. 1986;1(8476):307310.
71.Hopkins WG, Marshall SW, Batterham AM, Hanin J. Progressive statistics for studies in sports medicine and exercise science. Med Sci Sports
Exerc. 2009;41(1):3-12.
72.Hopkins WG. Bias in Bland-Altman but not regression validity analyses.
Sportscience. 2004;8:42-26.
73. Hanneman SK. Design, analysis and interpretation of method-comparison
studies. AACN Adv Crit Care. 2008;19(2):223-234.
74. Ludbrook J. Confidence in Altman-Bland plots: a critical review of the
method of differences. Clin Exp Pharmacol Physiol. 2010;37(2):143149.
75. Ward LC. Bioelectrical impedance validation studies: alternative approaches
to their interpretation. Eur J Clin Nutr. 2013;67(suppl 1):S10-S13.
76.Cecconi M, Rhodes A, Poloniecki J, Della Rocca G, Grounds RM.
Bench-to-bedside review: the importance of the precision of the reference
technique in method comparison studies—with specific reference to the
measurement of cardiac output. Crit Care. 2009;13:201.
77. Critchley LA, Critchley JA. A meta-analysis of studies using bias and precision statistics to compare cardiac output measurement techniques. J Clin
Monit Comput. 1999;15:85-91.
78.Ellis KJ. Human body composition: in vivo methods. Physiol Rev.
2000;80(2):649-680.
79. Schlosser K, Kehayias J, Tegenkamp M, Schoeller D. In-vivo and in-vitro
bromide equilibration time course in adults and sample handling effect. Int
J Body Compos Res. 2009;7(4):141-146.
80. Trabulsi J, Troiano R, Subar A, et al. Precision of the doubly labeled water
method in a large-scale application: evaluation of a streamlined-dosing
protocol in the Observing Protein and Energy Nutrition (OPEN) study.
Eur J Clin Nutr. 2003;57(11):1370-1377.
81. Schoeller D, van Santen E, Peterson D, Dietz W, Jaspan J, Klein P. Total
body water measurement in humans with 0 and H labeled water. Am J Clin
Nutr. 1980;33(12):2686-2693.
82.International Atomic Energy Agency. Introduction to Body Composition
Assessment Using the Deuterium Dilution Technique With Analysis of
Urine Samples by Isotope Ratio Mass Spectrometry. Vienna, Austria:
Author; 2010. IAEA Human Health Series 13.
83.International Atomic Energy Agency. Introduction to Body Composition
Assessment Using the Deuterium Dilution Technique With Analysis of
Saliva Samples by Fourier Transform Infrared Spectrometry. Vienna,
Austria: Author; 2010. IAEA Human Health Series 12.
84. Miller E. Bromide space determination using anion-exchange chromatography for measurement of bromide. Am J Clin Nutr. 1989;50(1):168-171.
85.Miller ME, Cappon CJ. Anion-exchange chromatographic determination
of bromide in serum. Clin Chem. 1984;30(5):781-783.
86. Kehayias J, Ribeiro S, Skahan A, et al. Water homeostasis, frailty and cognitive function in the nursing home. J Nutr Heal Aging. 2012;16(1):35-39.
87.Nutritional and Health-Related Environmental Studies. International
Atomic Energy Agency, Division of Human Health. Hum Heal Ser. 2014.
http://www-naweb.iaea.org/nahu/NAHRES/publication.html. Accessed
April 19, 2015.
88.Schoeller DA. Hydrometry. In: Heymsfield SB, Lohman TG, Wang
Z, Going SG, eds. Human Body Composition. 2nd ed. Champaign, IL:
Human Kinetics; 2005:35-49.
89.Schoeller DA, Hnilicka JM. Reliability of the doubly labeled water
method for the measurement of total daily energy expenditure in freeliving subjects. J Nutr. 1996;126(1):348S-354S.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
820
Journal of Parenteral and Enteral Nutrition 39(7)
90.Delmonico MJ, Harris TB, Lee JS, et al. Alternative definitions of sarcopenia, lower extremity performance, and functional impairment with
aging in older men and women. J Am Geriatr Soc. 2007;55(5):769-774.
91.Leonard CM, Roza MA, Barr RD, Webber CE. Reproducibility of DXA
measurements of bone mineral density and body composition in children.
Pediatr Radiol. 2009;39:148-154.
92.Jensen M, Hermann A, Hessov I, Mosekilde L. Components of variance
when assessing the reproducibility of body composition measurements
using bio-impedance and the Hologic QDR-2000 DXA scanner. Clin
Nutr. 1997;16(2):61-65.
93. Dodson S, Baracos VE, Jatoi A, et al. Muscle wasting in cancer cachexia:
clinical implications, diagnosis, and emerging treatment strategies. Annu
Rev Med. 2011;62:265-279.
94. Mourtzakis M, Prado CMM, Lieffers JR, Reiman T, McCargar LJ, Baracos
VE. A practical and precise approach to quantification of body composition in cancer patients using computed tomography images acquired during routine care. Appl Physiol Nutr Metab. 2008;33(5):997-1006.
95. Ross R, Janssen I. Computed tomography and magnetic resonance imaging.
In: Heymsfield SB, Lohman TG, Wang Z, Going SB, eds. Human Body
Composition. 2nd ed. Champaign, IL: Human Kinetics; 2005:89-108.
96. Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical impedance analysis—
part II: utilization in clinical practice. Clin Nutr. 2004;23(6):1430-1453.
97. Mulasi U, Kuchnia AJ, Cole AJ, Earthman CP. Bioimpedance at the bedside: current applications, limitations, and opportunities. Nutr Clin Pract.
2015;30(2):180-193.
98. Chumlea WC, Guo SS, Kuczmarski RJ, et al. Body composition estimates
from NHANES III bioelectrical impedance data. Int J Obes Relat Metab
Disord. 2002;26(12):1596-1609.
99.Buzzell PR, Pintauro SJ. BIA Interactive Tutorial—Department of
Nutrition and Food Sciences, University of Vermont. 2002. nutrition.uvm.
edu/bodycomp/bia. Accessed April 23, 2015.
100.Lukaski HC. Evolution of bioimpedance: a circuitous journey from estimation of physiological function to assessment of body composition and a
return to clinical research. Eur J Clin Nutr. 2013;67(suppl 1):S2-S9.
101.Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical imped
ance analysis—part I: review of principles and methods. Clin Nutr.
2004;23(5):1226-1243.
102.Matthie JR. Bioimpedance measurements of human body composition:
critical analysis and outlook. Expert Rev Med Devices. 2008;5(2):239-261.
103.Jaffrin MY, Morel H. Body fluid volumes measurements by impedance:
a review of bioimpedance spectroscopy (BIS) and bioimpedance analysis
(BIA) methods. Med Eng Phys. 2008;30(10):1257-1269.
104.Mialich MS, Maria J, Sicchieri F, Afonso A, Junior J. Analysis of body
composition: a critical review of the use of bioelectrical impedance analysis. Int J Clin Nutr. 2014;2(1):1-10.
105.Kushner RF, de Vries PM, Gudivaka R. Use of bioelectrical impedance
analysis measurements in the clinical management of patients undergoing
dialysis. Am J Clin Nutr. 1996;64(3)(suppl):503S-509S.
106. Moon JR, Stout JR, Smith AE, et al. Reproducibility and validity of bioimpedance spectroscopy for tracking changes in total body water: implications for repeated measurements. Br J Nutr. 2010;104(9):1384-1394.
107. Centers for Disease Control and Prevention. National Health and Nutrition
Examination Survey (NHANES) Anthropometry Procedures Manual. 2013.
www.cdc.gov/nchs/data/nhanes/nhanes_13_14/2013_Anthropometry.pdf
108.Kushner RF, Gudivaka R, Schoeller DA. Clinical characteristics influencing bioelectrical impedance analysis measurements. Am J Clin Nutr.
1996;64(suppl):423S-427S.
109. Gonzalez C, Evans J, Smye S, Holland P. Total body water measurement
using bioelectrical impedance analysis, isotope dilution and total body
potassium: a scoring system to facilitate intercomparison. Eur J Clin Nutr.
2002;56(4):326-337.
110.Organ L, Bradham G, Gore D, Lozier S. Segmental bioelectrical impedance analysis: theory and application of a new technique. J Appl Physiol.
1994;77(1):98-112.
111.Ward LC. Segmental bioelectrical impedance analysis. Curr Opin Clin
Nutr Metab Care. 2012;15(5):424-429.
112.Scharfetter H, Monif M, László Z, Lambauer T, Hutten H, HinghoferSzalkay H. Effect of postural changes on the reliability of volume estimations
from bioimpedance spectroscopy data. Kidney Int. 1997;51:1078-1087.
113.Gudivaka R, Schoeller D, Kushner RF. Effect of skin temperature
on multifrequency bioelectrical impedance analysis. J Appl Physiol.
1996;81(2):838-845.
114.Elia M. Body composition by whole-body bioelectrical impedance and
prediction of clinically relevant outcomes: overvalued or underused? Eur
J Clin Nutr. 2013;67(suppl 1):S60-S70.
115. Raimann JG, Zhu F, Wang J, et al. Comparison of fluid volume estimates
in chronic hemodialysis patients by bioimpedance, direct isotopic, and
dilution methods. Kidney Int. 2014;85(4):898-908.
116. Kotler D, Burastero S, Wang J, Pierson R. Prediction of body cell mass, fatfree mass, and total body water with bioelectrical impedance analysis: effects
of race, sex, and disease. Am J Clin Nutr. 1996;64(suppl):489S-497S.
117.Kotler DP, Tierney AR, Wang J, Pierson RN. Magnitude of body-cellmass depletion and the timing of death from wasting in AIDS. Am J Clin
Nutr. 1989;50(3):444-447.
118.Kuhlmann MK, Zhu F, Seibert E, Levin NW. Bioimpedance, dry weight
and blood pressure control: new methods and consequences. Curr Opin
Nephrol Hypertens. 2005;14(6):543-549.
119.Zhou Y-L, Liu J, Sun F, et al. Calf bioimpedance ratio improves dry
weight assessment and blood pressure control in hemodialysis patients.
Am J Nephrol. 2010;32(2):109-116.
120.Zhu F, Kuhlmann MK, Sarkar SKC, et al. Adjustment of dry weight in
hemodialysis patients using intradialytic continuous multifrequency bioimpedance of the calf. Int J Artif Organs. 2004;27(2):104-109.
121. Wizemann V, Wabel P, Chamney P, et al. The mortality risk of overhydration in haemodialysis patients. Nephrol Dial Transplant. 2009;24(5):15741579.
122.Wabel P, Moissl U, Chamney P, et al. Towards improved cardiovascular
management: the necessity of combining blood pressure and fluid overload. Nephrol Dial Transplant. 2008;23(9):2965-2971.
123.O’Lone EL, Visser A, Finney H, Fan SL. Clinical significance of multifrequency bioimpedance spectroscopy in peritoneal dialysis patients:
independent predictor of patient survival. Nephrol Dial Transplant.
2014;29(7):1430-1437.
124.Onofriescu M, Hogas S, Voroneanu L, et al. Bioimpedance-guided fluid
management in maintenance hemodialysis: a pilot randomized controlled
trial. Am J Kidney Dis. 2014;64(1):111-118.
125. Barbosa-Silva MCG, Barros AJD, Post CLA, Waitzberg DL, Heymsfield
SB. Can bioelectrical impedance analysis identify malnutrition in preoperative nutrition assessment? Nutrition. 2003;19(5):422-426.
126.Barbosa-Silva MCG, Barros AJD. Bioelectrical impedance analysis in
clinical practice: a new perspective on its use beyond body composition
equations. Curr Opin Clin Nutr Metab Care. 2005;8(3):311-317.
127.Norman K, Stobäus N, Pirlich M, Bosy-Westphal A. Bioelectrical phase
angle and impedance vector analysis: clinical relevance and applicability
of impedance parameters. Clin Nutr. 2012;31(6):854-861.
128.Ellis KJ, Bell SJ, Chertow GM, et al. Bioelectrical impedance methods
in clinical research: a follow-up to the NIH Technology Assessment
Conference. Nutrition. 1999;15(11-12):874-880.
129.Haverkort EB, Reijven PLM, Binnekade JM, et al. Bioelectrical impedance analysis to estimate body composition in surgical and oncological
patients: a systematic review. Eur J Clin Nutr. 2015;69(1):3-13.
130.Lingwood BE. Bioelectrical impedance analysis for assessment of fluid
status and body composition in neonates—the good, the bad and the
unknown. Eur J Clin Nutr. 2013;67(suppl 1):S28-S33.
131.Buchholz AC, Bartok C, Schoeller DA. The validity of bioelec
trical impedance models in clinical populations. Nutr Clin Pract.
2004;19(5):433-446.
132.Schoeller DA. Bioelectrical impedance analysis: what does it measure?
Ann N Y Acad Sci. 2000;904:159-162.
133. Kyle UG, Genton L, Gremion G, Slosman DO, Pichard C. Aging, physical
activity and height-normalized body composition parameters. Clin Nutr.
2004;23(1):79-88.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
Earthman
821
134.Bosaeus I, Wilcox G, Rothenberg E, Strauss BJ. Skeletal muscle mass
in hospitalized elderly patients: comparison of measurements by singlefrequency BIA and DXA. Clin Nutr. 2014;33(3):426-431.
135. Kyle UG, Genton L, Karsegard L, Slosman DO, Pichard C. Single prediction equation for bioelectrical impedance analysis in adults aged 20-94
years. Nutrition. 2001;17(3):248-253.
136. Schutz Y, Kyle UUG, Pichard C. Fat-free mass index and fat mass index
percentiles in Caucasians aged 18-98 y. Int J Obes Relat Metab Disord.
2002;26(7):953-960.
137.Janssen I, Heymsfield SB, Baumgartner RN, et al. Estimation of skeletal muscle mass by bioelectrical impedance analysis. J Appl Physiol.
2000;89:465-471.
138. Bosy-Westphal A, Danielzik S, Dorhofer R-P, Later W, Wiese S, Muller
M. Phase angle from bioelectrical impedance analysis: population reference values by age, sex, and body mass index. JPEN J Parenter Enteral
Nutr. 2006;30(4):309-316.
139.Barbosa-Silva MCG, Barros AJD, Wang J, Heymsfield SB, Pierson RN.
Bioelectrical impedance analysis: population reference values for phase
angle by age and sex. Am J Clin Nutr. 2005;82(1):49-52.
140.VanItallie T, Yang M-U, Heymsfield S, Funk R, Boileau R. Heightnormalized indices of the body’s fat-free mass and fat mass: potentially
useful indicators of nutritional status. Am J Clin Nutr. 1990;52:953-959.
141. Sun SS, Chumlea WC, Heymsfield SB, et al. Development of bioelectrical
impedance analysis prediction equations for body composition with the
use of a multicomponent model for use in epidemiologic surveys. Am J
Clin Nutr. 2003;77(2):331-340.
142.Kyle UG, Schutz Y, Dupertuis YM, Pichard C. Body composition interpretation: contributions of the fat-free mass index and the body fat mass
index. Nutrition. 2003;19(7-8):597-604.
143.Meireles MS, Wazlawik E, Bastos JL, Garcia MF. Comparison between
nutritional risk tools and parameters derived from bioelectrical impedance analysis with subjective global assessment. J Acad Nutr Diet.
2012;112(10):1543-1549.
144.Van Venrooij LM, de Vos R, Zijlstra E, Borgmeijer-Hoelen MM, van
Leeuwen PA, de Mol BA. The impact of low preoperative fat-free body
mass on infections and length of stay after cardiac surgery: a prospective
cohort study. J Thorac Cardiovasc Surg. 2011;142(5):1263-1269.
145.Haverkort EB, Binnekade JM, de van der Schueren MAE, Gouma DJ,
de Haan RJ. Estimation of body composition depends on applied device
in patients undergoing major abdominal surgery. Nutr Clin Pract.
2015;30(2):249-256.
146. Schwenk A, Eschner W, Kremer G, Ward LC. Assessment of intracellular
water by whole body bioelectrical impedance and total body potassium in
HIV-positive patients. Clin Nutr. 2000;19(2):109-113.
147.Ott M, Fischer H, Polat H, et al. Bioelectrical impedance analysis
as a predictor of survival in patients with human immunodeficiency
virus infection. J Acquir Immune Defic Syndr Hum Retrovirol. 1995;9:
20-25.
148.Gupta D, Lammersfeld CA, Vashi PG, et al. Bioelectrical impedance
phase angle as a prognostic indicator in breast cancer. BMC Cancer.
2008;8:249.
149.Gupta D, Lammersfeld CA, Vashi PG, et al. Bioelectrical impedance
phase angle in clinical practice: implications for prognosis in stage IIIB
and IV non–small cell lung cancer. BMC Cancer. 2009;9:37.
150.Gupta D, Lis CG, Dahlk SL, Vashi PG, Grutsch JF, Lammersfeld CA.
Bioelectrical impedance phase angle as a prognostic indicator in advanced
pancreatic cancer. Br J Nutr. 2004;92(6):957-962.
151.Gupta D, Lis CG, Dahlk SL, et al. The relationship between bioelectrical impedance phase angle and subjective global assessment in advanced
colorectal cancer. Nutr J. 2008;7(1):7-19.
152. Colín-Ramírez E, Castillo-Martínez L, Orea-Tejeda A, Vázquez-Durán M,
Rodríguez AE, Keirns-Davis C. Bioelectrical impedance phase angle as a
prognostic marker in chronic heart failure. Nutrition. 2012;28:901-905.
153. Selberg O, Selberg D. Norms and correlates of bioimpedance phase angle
in healthy human subjects, hospitalized patients, and patients with liver
cirrhosis. Eur J Appl Physiol. 2002;86:509-516.
154.Maggiore Q, Nigrelli S, Ciccarelli C, Grimaldi C, Rossi GA, Michelassi
C. Nutritional and prognostic correlates of bioimpedance indexes in
hemodialysis patients. Kidney Int. 1996;50(6):2103-2108.
155.Wirth R, Volkert D, Rösler A, Sieber CC, Bauer JM. Bioelectric impedance phase angle is associated with hospital mortality of geriatric patients.
Arch Gerontol Geriatr. 2010;51:290-294.
156.Dittmar M. Reliability and variability of bioimpedance measures in normal adults: effects of age, gender, and body mass. Am J Phys Anthropol.
2003;122(4):361-370.
157.Kyle UG, Genton L, Slosman DO, Pichard C. Fat-free and fat mass percentiles in 5225 healthy subjects aged 15 to 98 years. Nutrition. 2001;17(78):534-541.
158.Norman K, Stoba N, Zocher D, et al. Cutoff percentiles of bioelectrical
phase angle predict functionality, quality of life, and mortality in patients
with cancer. J Clin Nutr. 2010;92(3):612-619.
159.Paiva SI, Borges LR, Halpern-Silveira D, Assunção MCF, Barros AJD,
Gonzalez MC. Standardized phase angle from bioelectrical impedance
analysis as prognostic factor for survival in patients with cancer. Support
Care Cancer. 2011;19(2):187-192.
160. Piccoli A, Rossi B, Pillon L, Bucciante G. A new method for monitoring
body fluid variation by bioimpedance analysis: the RXc graph. Kidney Int.
1994;46:534-539.
161.Cornish BH, Thomas BJ, Ward LC. Improved prediction of extracellular
and total body water using impedance loci generated by multiple frequency
bioelectrical impedance analysis. Phys Med Biol. 1993;38:337-346.
162. Sun G, French CR, Martin GR, et al. Comparison of multifrequency bioelectrical impedance analysis with dual-energy X-ray absorptiometry for
assessment of percentage body fat in a large, healthy population. Am J
Clin Nutr. 2005;81(1):74-78.
163.Olde Rikkert MG, Deurenberg P, Jansen RW, van’tHof MA, Hoefnagels
WH. Validation of multifrequency bioelectrical impedance analysis in
monitoring fluid balance in healthy elderly subjects. J Gerontol A Biol Sci
Med Sci. 1997;52(3):M137-M141.
164.Kim M, Kim H. Accuracy of segmental multi-frequency bioelectrical
impedance analysis for assessing whole-body and appendicular fat mass
and lean soft tissue mass in frail women aged 75 years and older. Eur J
Clin Nutr. 2013;67(4):395-400.
165. Dittmar M, Reber H. Evaluation of different methods for assessing intracellular fluid in healthy older people: a cross-validation study. J Am
Geriatr Soc. 2002;50(1):104-110.
166.Deurenberg P, Tagliabue A, Schouten FJ. Multi-frequency impedance
for the prediction of extracellular water and total body water. Br J Nutr.
1995;73(3):349-358.
167. Anderson LJ, Erceg DN, Schroeder ET. Utility of multifrequency bioelectrical impedance compared with dual-energy X-ray absorptiometry for
assessment of total and regional body composition varies between men
and women. Nutr Res. 2012;32(7):479-485.
168. Hannan WJ, Cowen SJ, Plester CE, Fearon KCH, DeBeau A. Comparison
of bio-impedance spectroscopy and multi-frequency bio-impedance analysis for the assessment of extracellular and total body water in surgical
patients. Clin Sci. 1995;89(6):651-658.
169. Hannan WJ, Cowen SJ, Plester C, Fearon KC. Proximal and distal measurements of extracellular and total body water by multi-frequency bio-impedance analysis in surgical patients. Appl Radiat Isot. 1998;49(5-6):621-622.
170.Lehnert ME, Clarke DD, Gibbons JG, et al. Estimation of body water
compartments in cirrhosis by multiple-frequency bioelectrical-impedance
analysis. Nutrition. 2001;17(1):31-34.
171.Plank LD, Li A. Bioimpedance illness marker compared to phase
angle as a predictor of malnutrition in hospitalised patients. Clin Nutr.
2013;32(suppl 1):S85.
172. Earthman CP, Kruizenga HM, Weijs PJM. Impedance ratio Z200/Z5 compared to phase angle at 50 kHz better predicts nutritional status and length
of stay in hospitalized patients. Int J Obes. 2011;35(2):S58.
173.Itobi E, Stroud M, Elia M. Impact of oedema on recovery after major
abdominal surgery and potential value of multifrequency bioimpedance
measurements. Br J Surg. 2006;93(3):354-361.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016
822
Journal of Parenteral and Enteral Nutrition 39(7)
174. Valdespino-Trejo A, Orea-Tejeda A, Castillo-Martínez L, et al. Low albumin levels and high impedance ratio as risk factors for worsening kidney
function during hospitalization of decompensated heart failure patients.
Exp Clin Cardiol. 2013;18(2):113-117.
175. Castillo Martínez L, Colín Ramírez E, Orea Tejeda A, et al. Bioelectrical
impedance and strength measurements in patients with heart failure: comparison with functional class. Nutrition. 2007;23(5):412-418.
176.De Lorenzo A, Andreoli A, Matthie J, Withers P. Predicting body cell
mass with bioimpedance by using theoretical methods: a technological
review. J Appl Physiol. 1997;82(5):1542-1558.
177. Cole K. Membranes, Ions and Impulses: A Chapter of Classical
Biophysics. Berkeley: University of California Press; 1972.
178. Hanai T. Electrical properties of emulsions. In: Sherman PH, ed. Emulsion
Science. London, UK: Academic Press; 1968:354-377.
179.Matthie JR. Second generation mixture theory equation for estimating
intracellular water using bioimpedance spectroscopy. J Appl Physiol.
2005;99(2):780-781.
180. Ellis KJ, Wong WW. Human hydrometry: comparison of multifrequency
bioelectrical impedance with 2H2O and bromine dilution. J Appl Physiol.
1998;85(3):1056-1062.
181.Plank LD, Monk DN, Woollard G, Hill GL. Evaluation of mul
tifrequency bioimpedance spectroscopy for measurement of the
extracellular water space in critically ill patients. Appl Radiat Isot.
1998;49(5-6):481-483.
182.Moissl UM, Wabel P, Chamney PW, et al. Body fluid volume determination via body composition spectroscopy in health and disease. Physiol
Meas. 2006;27(9):921-933.
183.Chamney PW, Wabel P, Moissl UM, et al. A whole-body model to distinguish excess fluid from the hydration of major body tissues. Am J Clin
Nutr. 2007;85(1):80-89.
184.Wabel P, Chamney P, Moissl U, Jirka T. Importance of whole-body
bioimpedance spectroscopy for the management of fluid balance. Blood
Purif. 2009;27(1):75-80.
185. Marcelli D, Usvyat L a., Kotanko P, et al. Body composition and survival
in dialysis patients: results from an international cohort study [published
online April 21, 2015]. Clin J Am Soc Nephrol.
186.Bellisari A, Roche AF. Anthropometry and ultrasound. In: Heymsfield
SB, Lohman TG, Wang Z, Going SB, eds. Human Body Composition. 2nd
ed. Champaign, IL: Human Kinetics; 2005:109-127.
187. Peetrons P. Ultrasound of muscles. Eur Radiol. 2002;12:35-43.
188.Ohata K, Tsuboyama T, Ichihashi N, Minami S. Measurement of muscle
thickness as quantitative muscle evaluation for adults with severe cerebral
palsy. Phys Ther. 2006;86(9):1231-1239.
189.Campbell IT, Watt T, Withers D, et al. Muscle thickness, measured
with ultrasound, may be an indicator of lean tissue wasting in multiple organ failure in the presence of edema. Am J Clin Nutr. 1995;62:
533-539.
190.Gruther W, Benesch T, Zorn C, et al. Muscle wasting in intensive care
patients: ultrasound observation of the M. quadriceps femoris muscle
layer. J Rehabil Med. 2008;40(3):185-189.
191.Fivez T, Hendrickx a, Van Herpe T, et al. An analysis of reliability and
accuracy of muscle thickness ultrasonography in critically ill children and
adults [published online March 9, 2015]. JPEN J Parenter Enteral Nutr.
192.Puthucheary Z, Montgomery H, Moxham J, Harridge S, Hart N.
Structure to function: muscle failure in critically ill patients. J Physiol.
2010;588:4641-4648.
193.Abe T, Loenneke JP, Young KC, et al. Validity of ultrasound prediction
equations for total and regional muscularity in middle-aged and older men
and women. Ultrasound Med Biol. 2015;41(2):557-564.
194.Abe T, Kondo M, Kawakami Y, Fukunaga T. Prediction equations for
body composition of Japanese adults by B-mode ultrasound. Am J Hum
Biol. 1994;6:161-170.
195. Takai Y, Ohta M, Akagi R, et al. Applicability of ultrasound muscle thickness measurements for predicting fat-free mass in elderly population. J
Nutr Heal Aging. 2014;18(6):579-585.
196. Sanada K, Kearns CF, Midorikawa T, Abe T. Prediction and validation of
total and regional skeletal muscle mass by ultrasound in Japanese adults.
Eur J Appl Physiol. 2006;96(1):24-31.
197. Abe T, Dabbs NC, Nahar VK, Ford MA, Bass MA, Loftin M. Relationship
between dual-relationship between dual-energy X-ray absorptiometryderived appendicular lean tissue mass and total body skeletal muscle mass
estimated by ultrasound. Int J Clin Med. 2013;4:283-286.
198.Mourtzakis M, Wischmeyer P. Bedside ultrasound measurement of skeletal muscle. Curr Opin Clin Nutr Metab Care. 2014;17(5):389-395.
199. Mayans D, Cartwright MS, Walker FO. Neuromuscular ultrasonography:
quantifying muscle and nerve measurements. Phys Med Rehabil Clin N
Am. 2012;23:133-148.
200.Arbeille P, Kerbeci P, Capri A, Dannaud C, Trappe SW, Trappe TA.
Quantification of muscle volume by echography: comparison with
MRI data on subjects in long-term bed rest. Ultrasound Med Biol.
2009;35(7):1092-1097.
201.Abe T, Loenneke JP, Thiebaud RS. Ultrasound assessment of hamstring
muscle size using posterior thigh muscle thickness [published online
November 3, 2014]. Clin Physiol Funct Imaging.
202. Abe T, Kawakami Y, Kondo M, Fukunaga T. Comparison of ultrasoundmeasured age-related, site-specific muscle loss between healthy Japanese
and German men. Clin Physiol Funct Imaging. 2011;31:320-325.
203. Seymour JM, Ward K, Sidhu PS, et al. Ultrasound measurement of rectus
femoris cross-sectional area and the relationship with quadriceps strength
in COPD. Thorax. 2009;64(5):418-423.
204. Takai Y, Katsumata Y, Kawakami Y, Kanehisa H, Fukunaga T. Ultrasound
method for estimating the cross-sectional area of the psoas major muscle.
Med Sci Sports Exerc. 2011;43:2000-2004.
205.Tillquist M, Kutsogiannis DJ, Wischmeyer PE, et al. Bedside ultrasound is a practical and reliable measurement tool for assessing
quadriceps muscle layer thickness. JPEN J Parenter Enteral Nutr.
2013;38(7):886-890.
206.Toomey C, McCreesh K, Leahy S, Jakeman P. Technical considerations
for accurate measurement of subcutaneous adipose tissue thickness using
B-mode ultrasound. Ultrasound. 2011;19(2):91-96.
207.Heyland D. Clinical evaluation research unit. 2015. http://www.criticalcarenutrition.com/index.php?option=com_content&view=article&id=182
&Itemid=79. Accessed April 23, 2015.
208.Itobi E, Stroud M, Elia M. Value of multifrequency bioelectrical impedance in assessing fluid disturbances in patients undergoing major abdominal surgery. Eur J Gastroenterol Hepatol. 1997;9(8):9-10.
Downloaded from pen.sagepub.com at Vrije Universiteit Bibliotheek on March 23, 2016