164 Am J C lin Nutr 1996;63:l64-9. Printed in USA.

l996American Society for Clinical Nutrition

Measuring the thermic effect of food 3
George W Reed and James 0 Hill
ABSTRACT The thermic effect of food (TEF), defined as the
increase in metabolic rate after ingestion of a meal, has been
studied extensively, but its role in body weight regulation is
controversial. W e analyzed 1 3 1 TEF tests from a wide range of
subjects ingesting meals of varying sizes and compositions. Each
test lasted 6 h. Of the total 6-h TEF, 60% of the total had been
measured after 3 h, 78% after 4 h, and 91% after 5 h. W e
developed a three-parameter curve to fit the data, which reduced
noise and gave additional information about the TEF. The area
under this parametric curve was positively correlated with fat-free
mass (FFM ) and meal size (M S) and negatively correlated with
meal size squared (M S2) with an R2 of 0.35. The usual area under
a curve created by connecting the data points of a line was
correlated with the same factors but with an R2 of 0.28. The peak
of the parametric curve was positively correlated with FFM and
M S and negatively correlated with M S2. percent body fat, and
meal composition. The time at which the peak occurred correlated
positively with M S and percent fat in the meal. Our analysis
suggests that an inadequate measurement duration of the TEF
could lead to errors. In general, we recommend that the TEF be
measured for 5 h. Am J Clin Nutr 1996;63: 164-9.
KEY W ORDS Thermic effect of food, metabolic rate, en-
ergy expenditure, obesity
INTRODUCTION
In attempts to understand the regulation of energy balance,
investigators have studied all of the components of energy
expenditure (EE). EE can be divided into sleeping metabolic
rate (SM R), resting metabolic rate (RM R = SM R + arousal),
the thermic effect of exercise (TEE), and the thermic effect of
food (TEF) (1). The TEF, sometimes also called diet-induced
thermogenesis (DIT), is defined as the increase in RM R after
ingestion of a meal. Although this component of TEE may
account for a relatively small proportion of total EE (3-10% ),
it could play a role in the development and/or maintenance of
obesity (2) because small differences in the TEF over long
periods of time could account for large differences among
individuals.
Although the TEF has been investigated quite extensively,
its role in energy balance in man is still controversial. For
example, DAlessio et al (3) cite 15 papers reporting a reduced
TEF in obese subjects as compared with lean subjects and 12
papers reporting no difference in the TEF between the two
groups. Such discrepant results suggest that methodologic dif-
ferences between studies may be important. Except in the three
studies with 24-h measurements, the TEF was measured for
from 60 to 360 mm after the meal. Similarly, Kinabo and
Durnin (4) and W eststrate (5) point out numerous studies
showing discrepancies concerning the influence of various
factors (age, exercise, nutritional status, energy content of a
meal, and meal composition) on the TEF.
M any of these discrepancies may be the result of method-
ologic differences, particularly differences in the time taken to
measure TEF. W eststrate (5) suggests that the TEF can be
accurately assessed within 3 h for meals providing = 2508 kJ.
Kinabo and Durnin (4) showed that the TEF did not return to
baseline after 5 h for a 5016-Id meal and their data indicate that
the TEF did not return to baseline after 3 h for a 2508-lU meal.
Belko et al (6) showed that the TEF returned to baseline after
3 h for meals containing 15% of energy requirements but not
for meals containing 30% and 45% of energy requirements.
DAlessio et al (3) indicated that the TEF could last as long as
8 h after meals providing 6688 kJ and that the TEF did not
return to baseline after 4 h for meals providing 2090 kJ. W elle
et al (7) showed that the TEF did not return to baseline after
4 h for a 1672-id meal. Hill et al (8) showed that the TEF did
not return to baseline after 3 h for meals providing 2090, 4180,
and 6270 Id. Segal et al (9) showed that 70% of a 6-h TEF is
measured by 3 h for a 3010-id liquid meal and state, . . .even
a 6-hour period may not be sufficient to quantify the total
thermic response to a meal in all individuals. .
M ost studies of the TEF have used small numbers of sub-
jects, which likely contributes to the controversy because
W eststrate (5) noted that the high intraindividual variation in
the measurement of TEF allows for small power in studies with
sample sizes of < 10 subjects.
It appears that differences in the duration of the TEF mea-
surement and the amount of variation in measurements are
contributing factors to the controversy regarding the impor-
tanceof differences in the TEF to body weight regulation. In
the present study we used an extensive database of TEF tests to
evaluate the effect of the duration of measurement on the TEF.
A second goal was to develop a model of the TEF to reduce
I From the Clinical Nutrition Research Unit, Vanderbilt University,
Nashville, TN, and the Center for Human Nutrition, University of Cob-
rado. Denver, CO 80262.
2 Supported by NIH grants DK42549 and DK26615.
3 Address reprint requests to GW Reed, Department of Preventive M cd-
icine, Vanderbilt University School of M edicine, A- 1 1 24 M edical Center
North, Nashville, TN 37232-2637.
Received February 6, 1995.
Accepted for publication September 21, 1995.

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

0
0
0
4 ) -
0 -
C
C
E
:
. # {1 4 9 }\\/\
\
M EA URI NOTH TK M I C O P0011 1
some of the measurement variation and thus improve our
understanding of the factors that influence TEF.
SUBJ E C T S AND M E T H O DS
The data consist of 131 independent baseline meal tests
conducted in the energy metabolism laboratory at V anderbilt
University over 5 y (1988-1992). The TEF was measured
during several series of studies (10-17), although the TEF was
only reported in two studies (10, 14). Subjects participated in
various studies, with various interventions, but all tests used in
these analyses were baseline tests, before any intervention. I f a
subject was in two separate studies, only the first baseline meal
test was used so that the 131 tests represent 131 separate
individuals.
All subjects came to the laboratory early in the morning
between 0700 and 0800 after an overnight fast. Subjects lay
quietly for 45 mm, after which their RM R was determined for
30 mm with a ventilated-hood indirect calorimetry system
(SensorM edics 2900; SensorM edics Corp. Anaheim, CA).
Subjects then consumed the test meal within 15 mm. The test
meal size and composition varied for different studies. I n some
studies, a fixed meal size and composition were given [eg,
4180 I d and 40% of energy as fat (10)] and in others the size
of the meal was related to the subject s usual intake based on
diet records [25% of usual intake (14)]. All meals contained
15% of energy as protein. RM R was then measured for 10 mm
every 30 mm for the subsequent 6 h, during which time the
subject remained lying down but awake. Energy expenditure
was calculated from the amount of oxygen consumed and
carbon dioxide produced (18). The TEF was calculated as the
area under the response curve minus the 6-h RM R (extrapo-
lated from the baseline measurement). F igu r e 1 illustrates a
sample TEF curve that was adjusted for RM R by subtracting
the RM R from the energy expenditure measured. The TEF is
the area under this curve, calculated by summing each trape-
zoidal region formed with a base at zero.
There were 77 females (59%) and 54 males (41 %). T a b le 1
I 1 1 1 I J I
0 1 2 3 4 5 6
Time in hours
FI GURE 1. A sample of a TEF (thermic effect of food) curve that was
adjusted for resting metabolic rate (RM R) by subtracting the RM R from
the energy expenditure measured at each 30-ruin time point. The trapezoid
area of the TEF was measured as the area under this curve.
T ABL E 1
Subject and meal characteristics
Va l u e
W eight (kg) 88.6 (49.8-130.2)
Fat-free mass (kg) 56.5 (37.6-82.8)
Percent body fat (%) 35.4 (8.7-5 1.3)
Age (y) 38.1 (18-65)
M eal size (U) 3953 (2721-5831)
M eal composition
(% of energy as fat) 36.6 (24-65)
(% o f energy as carbohydrate) 48.4 (20-69)
(% of energy as protein)2 15
, j; range in parentheses. n = 54 males and 77 females.
2 Fixed amount for all studies.
lists the subject and meal characteristics and the range of
values. Body composition was estimated from body density by
using underwater weighing to determine body volume (19).
Residual lung volume was determined simultaneously with the
underwater weight by using the closed-circuit nitrogen-dilution
technique (19). Percent body fat was estimated from body
density by using the revised equation of Brozek Ct al (20).
SAS software, version 6.03 (21), on a digital vaxstation 3100
with a V M S (a standard operating system) operating system
was used for all statistical analyses and nonlinear curve fitting
described in the results. A nonparametric sign test was used to
test whether medians were different from zero; t tests were
used to test whether means were different from zero. M ultiple-
regression analysis was used to compare the TEF area and
model parameters with subject and meal characteristics.
R E SUL T S
H ow lon g sh ou ld t h e T E F b e m ea su r ed ?
To determine the optimum time that it takes to measure the
TEF it was necessary to ask two questions. First, was there a
positive component to TEF between N hours and 6 h, where
N = 3, 4, and 5? And second, what was the correlation between
theTEFat6handtheTEFat3,4,and5h?
EE at the end of the TEF test (ie, at 6 h) was compared with
RM R (measured before the test meal). The mean difference
was 15.9 kJ/h and the median difference was 12.5 U/h; both
differences were significantly different from zero, P < 0.0001.
Seventy percent of the subjects had positive differences at 6 h.
Next, the TEF was computed at 3, 4, 5, and 6 h and
expressed as a percentage of the meal size. The distributions of
the difference between the TEF over 6 h and the TEF over 3,
4, and 5 h are illustrated in Figure 2. The distribution means
are all significantly different from zero (P < 0.0001) and the
variation becomes smaller as the time approaches 6 h. A
one-sample sign test in each case indicates that the median
difference is significantly different from zero (P < 0.0001).
T a b le 2 lists the median differences and the 25th percentiles of
the difference distributions to indicate that even between 5 and
6 h > 75% of the subjects had a positive component of the
TEF. Assuming that the TEF at 6 h indicates the total TEF, we
computed the percentage of the total TEF that had not been
measured if the measurement were stopped at 3, 4, or 5 h. The
median percentage of TEF missed at each stopping point is

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

REED AND HILL
T E F Diffe re n c e : 6 h rs - 3 h rs
----I-
- 0 . 0 2 0.0 0.02 0.04 0.06
Difference
T E F Diffe re n c e : 6 hrs - 4 hrs
166
0
C.)
C
0
C)
0
0
0
C.)
C
0
U
0
0
0
C )
C
0
C.)
0
0
-0.02 0.0 0.02 0.04 0.06
Di f f e r e nc e
T E F Diffe re n c e : 6 h rs - 5 h rs
1
0.0 0.02 0.04 0.06
TEF o v er 3 h o u r s
I 0 9
0.0 0.02 0.04 0.06 0.08 0.10
T E F o v er 4 h o u r s
I
-0.02 0.0 0.02 0.04 0.06 0.08 0.10 0.12
T E F o v er 5 h o u r s
FIGURE 3. Regression line for TEF6 (thermic effect of food measured
for 6 h) compared with TEF3, TEF4, and TEF5 with the surrounding lines
showing the 95% CIs for individual measures. The variance explained (R2
value) is noted in the corner of each graph.
-I
-0.02 0.0 0.02 0.04 0.06
Difference
FIGURE 2. Histograms of the difference in the TEF (thermic effect of
food) measured for 6 h and measured at 3, 4, and 5 h. The variation
decreases but the mean and median are significantly greater than zero even
a t 5 h.
A measure of the TEF
listed in Table 2. At 4 h, > 20% of the TEF is not measured for
one-half of the subjects.
The TEF is related to meal size, but even for meals providing
3344 Id (n = 39) the mean and median differences in the
TEF at 6 and 5 h are significantly positive and the 25th
percentile is positive. For these smaller meals 10% of the
total TEF was not measured at 4 h.
Figure 3 illustrates the correlation between the TEF at 6 h
and the TEF at 3, 4, and 5 h. The variance explained is noted
on each graph. At 4 h 90% of the variance of TEF6 is ex-
plained. There is still variation and hence information about
differences among individuals in the TEF from 4 to 6 h.
TABLE 2
Thermic effect of food (TEF) missed by short-term measurements
M edian 25th Percentile M edian percent of TEF6
TEF63 0.0252 0.0125 40.0
TEF64 0.0128 0.0057 22.5
TEF65 0.0054 0.0021 9.1
The median and 25th percentile of the difference between TEF mea-
sured at 6 h and that at 3 (TEF63), 4 (TEF64), and 5 (TEF65) h. The
difference is also expressed as a percent of TEF at 6 h. TEF is measured
as a proportion of meal size.
To reduce the noise and further describe the TEF, a three-
parameter curve was used to describe each TEF curve over the
6-h period. Figure 4 illustrates the parametrized TEF model
curve. Each TEF curve was assumed to be of this form, the
equation for the curve being
A + Bte t/C
(1)
where t is the time in hours. The three parameters have the
following interpretation: A is the intercept, a term used for
adjustment for error in the measurement of RM R; BCe is the
maximum value or peak value on the curve; and C is the
maximum time or the time at which the peak value occurs.
For a given subject s TEF values, the parametric curve was
fit as illustrated in Figure 5. SAS proc NLIN was used to fit
each curve. Because the fit of the curve in some cases could be
sensitive to the initial values, a matrix of initial values was
considered, and the set that minimized the mean squared error
was used. (A was set at 0, B went from 0 to 40 in steps of 5,
and C went from 0.5 to 4.5 in steps of 1.) This automated
curve-fitting method was used on all TEF curves. Of the 131
subject curves, 128 were fit by the automated program. The
three curves that did not converge or converged to a degenerate
curve were essentially flat curves. W hen the initial values were
adjusted and the program was rerun these curves could be fit
but this was felt to bias the fitting process. The analysis was

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

M EASURING THE THERM IC EFFECT OF FOOD
0
0#{149}
0
L O
0
C
C
a)
>
2
Ca
.
I 67
illustrates whether the relation was positive or negative. The
rc u1t for bothtlictrapezoidarcaan the niooth darea w r
similar. The TEF was related to meal size in a quadratic fashion
and was related to the subjects FFM ; however, the R2 value
was 25% higher for the model area. This indicates a reduc-
tion in noise when the TEF model is used.
T ab le 4 sh ow s the characteristics that were significantly
related to the parameters of the smooth curve that describes
TEF. The intercept was not correlated with any of the factors.
This was expected because the intercept represents an adjust-
ment for error in the initial RM R measurement. The peak EE
was related to meal size and FFM just as the smoothed area was
related to these factors; however, EE also correlated negatively
with the subjects percentage body fat. There was a trend
(P = 0.06) for the peak to decrease with increases in the
relative fat content of meals. The time of the peak was related
to meal size in a positive linear fashion and to amount of body
- I I I I I I I fat (kg) of the subject in a positive fashion. Figure 6 illustrates
0 1 2 3 4 5 6 how the TEF curve shifts depending on the meal and subject
characteristics.
Ti m e in hours
0
0
C
C
g L C)
2
Ca
E
0
Ti m e in hours
FI G UR E 5. An example of one subjects actual TEF (thermic effect of
food) data and the corresponding TEF curve that was fit to the data.
Equation of the TEF curve: - 10.28 + 175.9 X T X e T /I .3, where T
is time and e is the base of the natural logarithm. RM R, resting metabolic
rate.
0 1 2 3 4 5 6
FI G UR E 4. An example of the form of the parametrized curve used to fit
the TEF(thermic effect offood)data. Equation: 175.9 X T X e T II 3 where
T is time and e is the base of the natural logarithm. RM R, resting metabolic
rate.
done using the 128 fitted curves. The average mean squared
error of the fitted curves was 336. The curve depicted in Figure
S has a mean squared error of 546.
Using the fitted curves, TEF could be computed by finding
the area under the smooth curve (smoothed area). The
smoothed area, the trapezoid area (original TEF), and the
smooth-curve parameters were compared with subject (FFM ,
fat mass, percentage fat, age, and sex) and meal characteristics
(size and content) by using multiple-regression analysis.
T ab le 3 sh ow s the characteristics that were significantly
related to the trapezoid area and the smoothed area. It also
DI S C US S I O N
Analysis of this database of meal tests strongly indicates that
the TEF lasts beyond 6 h for the majority of subjects. The TEF
is commonly measured for periods as short as 3 h. A 3-h
TEF measurement can miss > 40% of the total TEF, and a
4-h TEF measurement can miss 22.5% of the total TEF. Even
for meals providing 3344 kJ, 10% of the 6-h TEF is left
between 4 and 6 h. M ost studies indicate that a positive
component exists beyond 3 h. Segal et al (9) indicate that
60-70% of the measured TEF is sufficient for comparison of
the TEF across subjects. W e feel that there is important infor-
mation contained in the remaining unmeasured TEF. Although
the 3-h and 4-h TEFs are significantly correlated with the total
6-h TEF, our curve fitting indicates that many factors can
produce differences in the shape of the TEF curve. Thus, by not
measuring to 5 h it is possible that important effects could
either be masked or misinterpreted. For example, our new
summarization of the TEF indicates that nonobese subjects
show an earlier and higher peak TEF than do obese subjects.
Our analysis suggests that the shorter the duration of measure-
ment, the more likely it is that the total TEF will differ between
T A B L E 3
Regression analysis of trapezoid area and smoothed area TEF
Variable P Relation
Trapezoid area (R2 = 0.28)
M eal size 0.02 +
M eal size squared 0.03 -
FFM 0.0001 +
Smoothed area (R2 = 0.35)
Meal size 0.0001 +
M eal size squared 0.0002 -
FFM 0.0001 +
Results of multiple-regression analysis using the usual trapezoid area
measure of the TEF (thermic effect of food) and the area under the
smoothed curve. The R2 values indicate that the relation of the smoothed-
area TEF is stronger with increased meal size and fat-free mass (FFM ) than
the usual trapezoid area TEF measure.

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

M eal and subject characteristics w ere signif icantly related to the pa-
ram eters of the T EF (therm ic ef f ect of f ood) curv e, by m ultiple-regression
analy sis.
FFM
Me a l S iz e
1
Meal Si ze
Fat
0
0
0
0
C
C
a
>
0
C a
Me a l C o m p
0 1 2 3 4 5 6
Ti m e in hours
168 R EED A N D HIL L
TABLE 4
Characteristics related to peak energy ex penditure and tim e of peak
Va r i a b l e P Re l a t i o n
Peak energy ex penditure (m ax im um ; R2 = 0.39)
M eal siz e 0.0001 +
M eal siz e squared 0.0001 -
FFM 0 . 0 0 1 +
Percentage body f at 0.004 -
M eal com position 0.06 -
T im e of peak (m ax im um tim e; R2 = 0.09)
M eal siz e 0.003 +
Fa t 0.004 +
obese and nonobese subjects. A m easurem ent lasting an addi-
tional 2 h m ay w ipe out that dif f erence. It is of interest to note
that if the sm ooth curv e f or the T EF is projected to 8 h, the total
T EF w ould increase an av erage of 7% and that the increase is
correlated w ith m eal siz e (r = 0.28, P < 0.002).
W e used our database to show how a three-param eter
curv e f it to T EF data w ill giv e a m ore detailed sum m ary of
these data points. T he curv e is not intended to predict the
indiv idual T EF based on subject and m eal characteristics
any m ore so than subject and m eal characteristics can predict
T EF m easured by the trapez oid area under the curv e. Each
indiv idual w ith dif f erent m eal tests w ould hav e dif f erent
param eters in the sam e w ay that they w ould hav e dif f erent
areas. T he purpose of the curv e f itting is to giv e a m ore
detailed description of the m easure of the T EF. T he usual
area under the curv e can be thought of as a single-param eter
description of the T EF. T he new curv e can be thought of as
I I I I I I I
FIGURE 6. A n illustration of how the T EF (therm ic ef f ect of f ood)
curv e shif ts depending on subject and m eal characteristics. M eal siz e and
f at-f ree m ass (FFM ) tend to increase the peak ; subjects body f at and m eal
siz e squared (M S 2) tend to low er the peak ; m eal siz e and subjects body f at
tend to m ov e the tim e of the peak f urther out. T here is som e ev idence that
increased f at in the m eal m ay decrease the peak . Illustrated curv e equation:
175.9 x T X e T /I .3 w here T is tim e and e is the base of the natural
logarithm . R M R , resting m etabolic rate.
distilling the curv e to tw o param eters (plus one param eter
f or noise reduction). W e believ e tw o indiv iduals can hav e
the sam e area under the curv e, but drastically dif f erent T EF
curv es. T he usual trapez oid area m ethod of m easuring the
T EF w ould not pick this up. T he new m ethod can pick this
up. T he data illustrate that the shape of the T EF curv e m ay
dif f er betw een lean and obese indiv iduals. If the shape of the curv e
dif f ers, but the total T EF does not, this m ight ex plain w hy som e
studies f ind dif f erences in the total T EF betw een lean and obese
indiv iduals and others do not. W hen the T EF is m easured to 3 h,
there m ay be dif f erences in the T EF curv e betw een lean and obese
subjects; how ev er, w hen the T EF is m easured to 6 h, these
dif f erences m ay disappear.
Our param etric curv e f itting reduces som e of the v ariation
in T EF m easurem ents and opens a new av enue of inv esti-
gation of the T EF. Our m ethod dem onstrates the sam e
relation betw een f actors such as FFM and m eal siz e and T EF
as the conv entional m easurem ent, but the R2 v alue increases.
If the new m odel w as not a closer approx im ation to T EF, it
is possible that the v ariation could increase. T hus, our
sm ooth curv e increases the pow er to identif y f actors that
inf luence the T EF. B oth the trapez oid area and sm oothed
area are positiv ely related to a subjects FFM and to m eal
siz e. How ev er, the sm oothed area identif ies a negativ e qua-
dratic relation betw een T EF and m eal siz e. T his positiv e
linear and negativ e quadratic ef f ect w as reported prev iously
(6, 8, 22) and could be an indication of a m eal-siz e threshold
ef f ect. Once the m eal reaches a certain siz e, no increase in
m eal siz e w ill increase the T EF. It is also possible that this
resulted because the T EF w as only m easured f or 6 h and that
the total T EF w as underestim ated f or large m eals as
suggested by DA lessio et al (3). How ev er, the sm all posi-
tiv e dif f erence betw een the T EF at 6 h and R M R w as not
correlated w ith m eal siz e (P - 0.43, r = 0.07) w hereas the
positiv e dif f erence betw een the T EF at 5 h and the R M R
w as signif icantly correlated w ith m eal siz e (P = 0.002,
r 0.26).
W e present an alternativ e to the standard w ay of assessing
the area under the T EF curv e. Our three-param eter curv e
sum m ariz es other inf orm ation in addition to total T EF and
thus prov ides a m eans of better assessing the w ay s in w hich
characteristics of the subject and of the diet ef f ect the T EF.
For ex am ple, w e f ound that w hereas total T EF m ay not be
inf luenced by the subjects body f at content, the shape of the
T EF curv e m ay dif f er by the subjects body f at content. A s
seen in Figure 6, the peak of the T EF curv e m ov es up as
m eal siz e and the subjects FFM increase. T he peak is
related positiv ely to m eal siz e and negativ ely to m eal siz e
squared in the sam e w ay as f or the total T EF. T his nonlinear
relation to m eal siz e indicates that the peak increases w ith
m eal siz e, then reaches a saturation lev el. T he new f actor
here is that peak T EF is inv ersely related to percent body f at
of the subject.
Using this three-param eter curv e, w e can begin to under-
stand how characteristics of the m eal and of the subject inf lu-
ence the T EF. A s m eal siz e increases, the peak T EF increases
and occurs at a later tim e af ter m eal ingestion, and total T EF
increases. T he peak (and possibly the total T EF) appears to
reach a threshold once the m eal reaches a certain siz e. T here is
a direct positiv e relation betw een the am ount of FFM of the
subject and both the peak and total T EF. T he body f at content

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

I Le a n
L- O b e s e
0
0
0
0
C
C
a)
0
. 0
Ca
E
.
M EASURI NG THE THERM I C EFFECT OF FOOD 169
0 1 2 3 4 5 6
Ti m e in hours
FI GURE 7. Predicted TEF (thermic effect of food) curves for a lean
individual with a higher peak and a shorter time of occurrence of peak
(dotted line) and for an obese individual (solid line). Equation for lean
subjects: 175.9 X T X e 771.3; Equation for obese subjects: 113.6
x T X e_T/I85 where T is time and e is the base of the natural logarithm.
RM R, resting metabolic rate.
of the subjects (either amount or percent) appears to effect the
shape of the TEF response but not the total amount of the TEF.
As the subject s body fat content increases, the peak TEF is
lowered and occurs at a later time after the meal. Figu r e 7
illustrates the possible difference in the TEF due to differences
in body fat content. The dotted line shows the response of the
theoretical lean individual, who will have a high early peak that
then drops off. Conversely, the theoretical obese individual
(solid line) will have a lower flat curve. The total area under
both curves would be the same.
M easuring the TEF for 6 h and using the parametric curve
will not resolve all differences. For example, Segal et al (9)
measured the TEF for 6 h and found differences in the TEF
between lean and obese subjects. I n general, we did not find
such a difference across a variety of conditions. However, com-
pared with our study, Segal et al used meals that were lower in fat,
meals that had a lower energy content, and a liquid diet.
I t is likely that the pattern of the TEF is more complicated
than is the three-parameter curve presented here; however, a
higher-order model is more difficult to fit given 12 data points.
Our studies show that a three-parameter curve reduces noise
better than does a two-parameter curve (excluding the intercept
term) and that a four-parameter curve fits fewer TEF curves
(101 of 131).
I n summary, our results indicate that the TEF is a response
lasting 6 h in most people and we recommend that measure-
ments last 5 h. Additionally, we demonstrated that variation in
the TEF can be reduced by fining the data to a three-parameter
curve. This alternative summary ofTEF data provides information
not only about total TEF but also about the time course of the TEF
and will be useful in understanding how characteristics of the diet
and of the subject influence the TEF. U
R E FE R E N C E S
1. Hill 10. Exercise, energy expenditure and fat oxidation. I n: Bray GA,
Ryan DH, eds. The science of food regulation. Baton Rouge: LSU
Pr e s s , 1992:67-84.
2. Schutz Y , Bressard T, iequier E. Diet-induced thermogenesis mea-
sured over a whole day in obese and nonobese women. Am I Clin Nutr
1984:40:542-52.
3. D Alessio DA, Kavle EC, M ozzoli M A, et al. Thermic effect of food
in lean and obese men. I Clin I nvest 1988;81:1781-9.
4. Kinabo I L, Durnin JV GA. Thermic effect of food in man: effect of
meal composition, and energy content. Br I Nutr 1990;64:37-44.
5. W eststrate I A. Resting metabolic rate and diet-induced thermogenesis:
a methodological reappraisal. Am I Clin Nuts l993;58:592-601.
6. Belko AZ, Barbien TF, W ong EC. Effect of energy and protein intake
and exercise intensity on the thermic effect of food. Am I Clin Nutr
1986:43:863-9.
7. W elle S. Lilavivat U, Hana R, Campbell RG. Thermic effect of feeding
in man. I ncreasing plasma norepinephrine levels following glucose but
not protein or fat consumption. M etab Clin Exp 1981 ;30:953-8.
8. Hill 10, Heymsfield SB, M cM anus I I I CB, DiGirolamo M . M eal size
and thermic response to food in male subjects as a function of maxi-
mum aerobic capacity. M etabolism 1984;33:743-9.
9. Segal KR, Eda#{ 241} oA, Tomas M B. Thermic effect of a meal over 3 and
6 hours in lean and obese men. M etabolism 1990;39:985-92.
10. Hill JO, Peters I C, Y ang D, et al. Thermogenesis in man during over-
feeding with medium chain triglycerides. M etabolism 1989;38:641-8.
1 1 . Hill 10, Schlundt DG, Sbrocco T, et al. Evaluation of an alternating
calorie diet with and without exercise in the treatment of obesity. Am
I Clin Nutr l989;50:248-54.
12. Hill JO, Peters I C, Reed GW , Schlundt DG, Sharp T, Green HL.
Nutrient balance in man: effects of diet composition. Am I Clin Nutr
199 l;54:lO-17.
I 3. Schlundt DG, Hill 10, Sbrocco T, Pope-Cordle I , Sharp T. The role of
breakfast in the treatment of obesity: a randomized clinical trial. Am I
Clin Nutr 1992; 55: 645-5l .
14. Bennett C, Reed GW , Peters I C, Abumrad NN, Sun M , Hill 10. The
short-term effects of dietary fat ingestion on energy expenditure and
nutrient balance. Am I Clin Nutr 1992:55:1071-7.
15. Thomas CD, Peters I C, Reed GW , Abumrad NN, Sun M , Hill 10.
Nutrient balance and energy expenditure during ad libitum feeding of
high-fat and high-carbohydrate diets in humans. Am I Clin Nutr
1992;55:934-42.
16. Drougas HI , Reed GW , Hill 10. Comparison of dietary self-reports
with energy expenditure measured using a whole-room indirect cab-
rimeter. I Am Diet Assoc 1992;92: 1073-7.
17. Sharp T, Reed GW , Abumrad NN, Sun M , Hill 10. Relationship
between aerobic fitness level and daily energy expenditure in weight-
stable humans. Am I Physiol l992;263:El 21-8.
18. W eir lB. New methods for calculating metabolic rate with special
reference to protein metabolism. I Physiol 1946;l09:l-9.
19. Goldman RF, Buskirk ER. Body volume measurement by underwater
weighing: description of a method. I n: Brozek I , ed. Techniques for
measuring body composition. W ashington, DC: National Academy of
Sciences, 1961:78-9.
20. Brozek I , Grande F, Anderson I T, Keys A. Densitometric analysis of
body composition: revision of some quantitative assumptions. Ann N
Y Ac a d S c i l 9 6 3 ; 1 l 0 : l 1 3 - 4 0 .
21. SAS I nstitute, I nc. SAS/STAT user s guide, version 6. 4th ed. V ol 2.
Cary, NC: SAS I nstitute I nc. 1990.
22. Bagatell CI , Heymsfield SB. Effect of meal size on myocardial oxygen
requirements: implications for postmyocardial infarction diet. Am I
Clin Nutr 1984:39:421-6.

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

6
,

2
0
1
4
a
j
c
n
.
n
u
t
r
i
t
i
o
n
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m