1478

LETTERS TO THE EDITOR

Erratum
Hall KD, Jordan PN. Modeling weight-loss maintenance to help prevent body weight regain. Am J Clin Nutr 2008;88:1495503. On page 1497, Equation 10 contains an error in the exponential term. The correct equation appears below. DEI BWinit Dd=1 b C 2 cF F M cF M F Minit =1 b C 2 DBW cF F M dinit Dd=1 b W fF Minit C expC 1 F Minit DBW g 3 C cF F M cF M =1 b
doi: 10.3945/ajcn.2009.27585.

10

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

Erratum
Landberg R, Aman P, Friberg LE, Vessby B, Adlercreutz H, Kamal-Eldin A. Dose response of whole-grain biomarkers: alkylresorcinols in human plasma and their metabolites in urine in relation to intake. Am J Clin Nutr 2009,89:2906. The rst paragraph of Results on page 292 should read: All subjects except for one completed the study, and one was excluded because of noncompliance with the advised intake (tick-off list) and unsatisfactory urine collections [.100% within-subject CV in creatinine excretion (26)]. Blood sampling was completed by 16 subjects (samples from 15 subjects were included in the statistical analysis), and 90% of urine collections were reported as complete. In Figure 1 on page 294, (n 16 3 3) should be replaced with (n 15 3 3). Figure 4 on page 295 is incorrect and should be replaced with the gure below.

doi: 10.3945/ajcn.2009.27667.

Modeling weight-loss maintenance to help prevent body weight regain13

Kevin D Hall and Peter N Jordan
ABSTRACT Background: Lifestyle intervention can successfully induce weight loss in obese persons, at least temporarily. However, there currently is no way to quantitatively estimate the changes of diet or physical activity required to prevent weight regain. Such a tool would be helpful for goal-setting, because obese patients and their physicians could assess at the outset of an intervention whether long-term adherence to the calculated lifestyle change is realistic. Objective: We aimed to calculate the expected change of steadystate body weight arising from a given change in dietary energy intake and, conversely, to calculate the modification of energy intake required to maintain a particular body-weight change. Design: We developed a mathematical model using data from 8 longitudinal weight-loss studies representing 157 subjects with initial body weights ranging from 68 to 160 kg and stable weight losses between 7 and 54 kg. Results: Model calculations closely matched the change data (R2 0.83, 2 2.1, P 0.01 for weight changes; R2 0.91, 2 0.87, P 0.0004 for energy intake changes). Our model performed significantly better than the previous models for which 2 values were 10-fold those of our model. The model also accurately predicted the proportion of weight change resulting from the loss of body fat (R2 0.90). Conclusions: Our model provides realistic calculations of bodyweight change and of the dietary modifications required for weightloss maintenance. Because the model was implemented by using standard spreadsheet software, it can be widely used by physicians and weight-management professionals. Am J Clin Nutr 2008; 88:1495503.

amount ( BW), how would his or her diet or physical activity have to change to maintain the goal weight? A quantitative answer to this question would be helpful for goal-setting, because both patient and physician could assess whether long-term adherence to the calculated lifestyle change is a realistic proposition. Such a calculation is not currently possible. In this report, we propose a mathematical model for calculating the changes in dietary intake and physical activity required to maintain a given body-weight change and to prevent weight regain. To facilitate use of the model by physicians and weightmanagement professionals, the model was implemented by using standard spreadsheet software that can be downloaded (see Spreadsheet files under Supplemental data in the current online issue; the spreadsheet files and an online version of the model are available at http://www2.niddk.nih.gov/NIDDKLabs/LBM/ lbmHall.htm). We developed our model by using longitudinal weight-change data from studies that measured energy expenditure (EE) and body composition during periods of weight stability both before and after weight loss (8 15). We compared the performance of our model with that of previous mathematical models (16 18) regarding the capacity to match the steady-state weight-change data, and we found that our model was superior and could provide realistic estimates of both the magnitude of the weight change and the changes in dietary energy intake (EI) required to prevent weight regain.
MATERIALS AND METHODS

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

Proposed model of steady-state body-weight change We developed a simple model of the steady-state EE rate of the body as a function of body composition, EI, and physical activity:

INTRODUCTION

The increasing prevalence of obesity poses a serious health concern (1). Whereas lifestyle interventions can result in significant weight loss in obese patients, weight regain is very common (2, 3). Therefore, rather than focusing exclusively on weight loss, several investigators have emphasized the importance of maintaining body weight at a lower level and preventing weight regain (4 7). Data from the National Weight Control Registry have been used to glean useful insights regarding the strategies used by persons who have successfully maintained significant weight change over extended periods (6, 7). However, there currently are no available quantitative tools to estimate, at the outset of obesity treatment, the lifestyle changes required to maintain a specific weight-loss goal. In other words, the question remains: If a patient wishes to change his or her body weight by a certain

EE

EI

FFM

FFM

FM

FM

FFM
1 1

FM

(1)

where K is a constant, FFM 22 kcal kg d , and FM 3.2 kcal kg 1 d 1 are the regression coefficients for resting metabolic rate versus fat-free mass (FFM) and fat mass (FM),
From the Laboratory of Biological Modeling, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, MD. 2 Supported by the Intramural Research Program of the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health. 3 Reprints not available. Address correspondence to KD Hall, NIDDK/ NIH, 12A South Drive, Room 4007, Bethesda, MD 20892-5621. E-mail: kevinh@niddk.nih.gov. Received April 28, 2008. Accepted for publication August 28, 2008. doi: 10.3945/ajcn.2008.26333.
1

Am J Clin Nutr 2008;88:1495503. Printed in USA. 2008 American Society for Nutrition

1495

Supplemental Material can be found at: http://www.ajcn.org/cgi/content/full/88/6/1495/DC1

1496

HALL AND JORDAN

respectively (19). We assumed that the energy cost of most physical activities is proportional to body weight and is thus specified by the parameter . The parameter accounts for the thermic effect of feeding as well as any adaptations of the EE rate beyond that predicted by body-composition change alone (20). Because there is significant debate as to whether such adaptations occur with weight loss (20, 21), the numerical value of the parameter was determined by the best fit to the weight-change data (described in the final paragraph of the Materials and Methods section). When the body weight is maintained at a constant reduced level, the EE rate equals the EI rate. Therefore, if the dietary intake changes by EI, and the physical activity changes by , the following equation is satisfied at the new steady state:

recognize that most readers cannot easily calculate expected weight change results by using equation 5 because of the appearance of the nonelementary Lambert W function. To address this issue, we have provided standard spreadsheet files that will allow readers to examine the predicted changes in steady-state body weight as a function of the model parameters (see Spreadsheet files under Supplemental data in the current online issue; the spreadsheet files and an online version of the model are available at http:// www2.niddk.nih.gov/NIDDKLabs/LBM/lbmHall.htm). Previous models of steady-state body-weight change Previous mathematical models have been proposed to calculate BW resulting from a specified change of diet or physical activity (16 18). The first model, by Christiansen et al (16), was given by the following equation:

EI

EI

FFM FM

init init

FFM FM BWinit (2)

BW

1 k

EI PAL

(6)

where BWinit is the initial body weight, and FM and FFM are the changes in body fat and FFM, respectively. (Note that the constant, K, no longer appears in equation 2, so we need not specify its value.) Using the fact that BW FFM FM, the following equation holds for the change in FFM:

where k 13.5 kcal kg 1 d 1 for men and k 11.6 1 d 1 for women, and PAL the physical activity kcal kg level, defined as the ratio of total EE rate to the resting metabolic rate. The second model, by Kozusko (18), expressed the predicted body-weight change according to the following equation:

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

FFM

EI
FM

BWinit
init

BW BW /
FFM FM

EI/EIinit 1 BWinit/2 (7)

(3)

Our previously published modification of the classic Forbes equation of body-composition change (22) predicts that the change in FFM is given by the following nonlinear function:

where

was given by the following equation:

FFM

BW

FMinit W

C FMinit FMinit exp C C BW (4)

3 1 1 tanh 2 2 2

FMinit BWinit

FMinit 1 BWinit

FMinit BWinit 1 2 (8)

where FMinit is the initial body FM, C 10.4 kg is the constant providing the best fit to the empirical Forbes body-composition curve (23), and W is the Lambert W function (24). Equations 3 and 4 can be solved for the expected change in steady-state body weight, as shown in the following equation:

BW

EI

BWinit
FM init

FFM

FFM

FMinit

In the third model under consideration, Heymsfield et al (17) used the Institute of Medicine of the National Academies of Science (IOM-NAS) equations to estimate the expected weight change for a given dietary intervention, but those investigators did not allow for changes in physical activity. We accounted for physical activity changes by solving the IOM-NAS equations for the expected body-weight change as follows:

C
FM FM

BW
FFM init init FFM init init FM

EI/ K1PAfinal

BWinit

K2h/K1

PA/PAfinal
1 1

9)

W exp

FMinit FMinit
init

C
FM

where h is height in meters, K1 14.2 kcal kg d and K2 503 kcal m 1 d 1 for men, and K1 10.9 kcal kg 1 d 1 and K2 660.7 kcal m 1 d 1 for women (25). PA stands for a dimensionless physical activity parameter related to PAL, but it takes on discrete values within a range of PAL values (17, 25). Proposed model of diet and physical activity changes required to prevent weight regain

FFM

exp

1 C
FFM

EI

BWinit
init

(5)

Equation 5 allows us to calculate the change in steady-state body weight, given information about the initial body weight and FM and about the changes in dietary EI and physical activity. We

Our main goal was to calculate the changes in dietary intake and physical activity required to maintain a specified bodyweight change. The following rearrangement of equation 5 expresses the EI change required for a specified body weight change and physical activity change:

MODELING WEIGHT-LOSS MAINTENANCE

FFM init FM

1497
BWinit, initial body weight; FMinit, initial fat; init, initial body composition and initial physical activity parameter; , change; EE, total energy expenditure; RMR, resting metabolic rate; FFM and FM, regression coefficients for resting metabolic rate versus fat-free mass (FFM) and fat mass (FM), respectively; C, the constant providing the best fit to the empirical Forbes body-composition curve. Physical activity was (0.9 EE RMR)/BW, and all models used the following parameters: FFM 22 kcal kg 1 d 1, FM 3.2 kcal kg 1 d 1, C 10.4 kg, and calculated by using 0.24 2 x SD (all such values). 33.7 29 17.5 11.7 6.5 4.8 17.8 477 21.4 609 8.6 522

Westerterp et al (15)

EI

BWinit

/1 C2 BW
FFM

C2

FMinit/ 1

/1 BW]
FM

W FMinitC exp

1 C

FMinit
FFM

/1

(10)
Weinsier et al (14) van Gemert et al (13) Martin et al (12)

TABLE 1 Data and parameters used for model comparisons in Figures 1 and 21

35.3 23

yi
i 1

mi

(11)

Das et al (9)

N 2 i 1

yi

m i 2/

(12)

where i was the SD of the measurement. The chi-square value was our primary measure of model fit to the data. Using the incomplete gamma function, we computed the probability that the chi-square for a correct model should be less than the chisquare calculated for the model (26). Finally, we calculated the coefficient of determination according to the following equation:

Subjects (n) BWinit (kg) Finit (kg) kg 1 init (kcal BW (kg) EI (kcal/d) (kcal kg 1

18 83.7 8.52 34.3 3.3 11.2 22 10 231 668 3.5

Amatruda et al (8)

where yi was the model calculation corresponding to the measured value mi for each group and the brackets ( and ) denote the mean value. Second, we calculated the chi-square value, which is a weighted measure of the distance between the model calculations and the data according to the following equation:

30 139.5 71.6 6.9 53.4 862 1.0

22.2 598

12 67.5 5.1 9 4.7 24.5 15.8 6.1 2018 252 16.5

We compared the various model calculations for BW with longitudinal weight-change data that were achieved over periods of weight loss ranging from 1 to 14 mo. The whole-body EE rates were accurately measured during periods of weight stability both before and after weight loss either by using the doubly labeled water method (8, 9, 1215) or by titrating the food intake in an in-patient setting to achieve weight stability for 2 wk (10, 11). Because the measurements were made during periods of weight stability, we assumed that the EE rate equaled the dietary EI rate. Therefore, the measured changes of EE after weight loss gave an accurate estimate of the dietary EI changes required to maintain the measured body-weight change and prevent regain. To provide all of the parameter values used in the above calculations, we required measurements of the initial body FM and the PAL. We found 8 longitudinal weight-loss studies, representing 157 patients, that satisfied these criteria, in which a wide range of weight losses were induced either by bariatric surgery (9, 13, 15) or by diet restriction (8, 10 12, 14). The parameters for each study are shown in Table 1. We evaluated each model in comparison to the weight-change data by using the following measures. First, we calculated the Pearson correlation coefficient (r) between the model calculations and the data according to the following equation:

12 82.3 6.6 26.1 3.1 12.7 12 2.1 417 687 1.3

Comparison of model results with longitudinal weightchange data

8 130.1 68.3 7.0 46.3 861 1.2

We have provided a spreadsheet file for this calculation (see Spreadsheet files under Supplemental data in the current online issue; the spreadsheet files and an online version of the model are available at http://www2.niddk.nih.gov/NIDDKLabs/LBM/ lbmHall.htm). We compared the predicted EI change from equation 10 with the change in the steady-state total EE rate measured in the longitudinal weight-loss studies described below.

32 78.8 30.7 6.3 12.9 114 2.0

5 158.6 74 6.1 53.9 908 2.4

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

Leibel et al (11) Keys (10)

11 70.5 11.7 17.5 12.6 9.0 6.8 15.4 428 664 1.7

9 132.1 26.9 68 19.8 5.5 17.8 34.4 551 853 1.0

10 124.8 29.6 64.4 24.8 6.7 29.2 37.1 886 891 2.1

1498
R2 1
i

HALL AND JORDAN

(yi

m i) 2
i

(mi

m )2

(13)

0 Amatruda Das -10 Keys Leibel -20 Martin van Gemert Weinsier -30 Westerterp Identity -40

Because the models are generally nonlinear, the coefficient of determination (R2) is not identical to the correlation coefficient squared. The model parameter was determined by using a weighted least-squares optimization procedure to find the best fit value of that corresponded to the weight-change data listed in Table 1 (26). To calculate the uncertainty of our estimate of , we performed a Monte-Carlo analysis in which we re-fit with the use of 5000 sets of synthesized data with the same statistics as the real data. In other words, each synthesized data-point was randomly selected from a normal distribution with a mean and SD corresponding to a real data-point. The uncertainty of the model parameter was then calculated as the SD of the 5000 best-fit parameters from the Monte-Carlo simulations. All model calculations and statistical comparisons were performed with the use of MATLAB software (version R2008a; The MathWorks Inc, Natick, MA).

Measured BW (kg)

-50

-60

-70 -70 -60 -50 -40 -30 -20 -10 0

Predicted BW (kg)

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

RESULTS

0 Amatruda Das Keys -500 Leibel Martin van Gemert -1000 Weinsier Westerterp Identity -1500

Comparison of results from the proposed model with longitudinal weight-change data A comparison of results from our model with the measured changes of steady-state body weight given the measured changes of EI and physical activity are plotted in Figure 1. For our proposed model, the best-fit value for the parameter was 0.24 0.13. Statistical evaluation of our model showed a Pearson correlation coefficient of 0.983 between our models bodyweight calculations and the measured values. The coefficient of determination was 0.83, which indicated that our model described 83% of the data variability. The chi-square value was 2.08 and, according to an evaluation of the incomplete gamma function, the probability was 0.01 that the chi-square for a correct model should be less than the chi-square calculated for our model (26). Our model results are compared (Figure 1B) with the measured changes in EI rate required to maintain the average measured weight loss observed in each study (R2 0.91). In this case, the chi-square value was 0.87, and the incomplete gamma function gave a probability of 3.1 10 4. Therefore, our model provided an excellent match to the data over a wide range of changes in body weight (Figure 1A) and dietary intake (Figure 1B), which suggests that our model successfully captured the salient physiologic changes. The comparison of the predicted and the measured fat fraction of the weight loss is plotted in Figure 2; and (R2 0.90) indicates that our model accurately described the observed bodycomposition changes. There was a wide range of observed bodycomposition changes, which were due to the different initial body compositions and the range of weight losses in the various studies (22). For example, the initially lean subjects from the Minnesota starvation experiment of Keys et al (10) mostly lost FFM, and only 34% of their weight loss was accounted for by a loss of body fat. In contrast, body fat loss accounted for 84% of the weight loss in the in the obese subjects studied by Leibel et al (11) who reduced their weight by 10%.
Measured EI (kcal/d)

-2000

-2500 -2500

-2000

-1500

-1000

-500

Predicted EI (kcal/d)
FIGURE 1. A: Predicted versus measured changes in steady-state body weight ( BW) resulting from the measured changes in energy intake and physical activity (R2 0.83). B: Predicted versus measured changes in dietary energy intake ( EI) required to achieve the observed changes in steady-state body weight (R2 0.91). The dotted lines are lines of identity, and error bars are SEs. BW, body weight; , change.

Greater steady-state weight change is predicted for persons with higher initial body fat The fact that our model accounts for variations in bodycomposition change leads to an interesting result when comparing the calculated steady-state weight changes in people with different initial body weights. The predicted steady-state bodyweight changes as a function of the diet change for 2 example subjects are illustrated in Figure 3: the solid line represents the first subject, who corresponds to an average participant from a study by Martin et al (12), and the dashed line represents the second obese subject, who corresponds to an average participant

MODELING WEIGHT-LOSS MAINTENANCE

1 Amatruda Das Keys 0.8 Leibel Martin van Gemert 0.6 Weinsier Westerterp Identity

1499

can also be used to investigate the sensitivity of the model calculations to changes in parameters, much as were illustrated in Figure 3, where we examined a range of values for the parameter . Comparison with previous steady-state weight-change models The residuals between the various model calculations and the measured body-weight changes for each group of subjects are compared in Figure 4. Our proposed model performed better than the other models because the residual body-weight changes were closer to zero and less spread out than were those in the other models. Whereas the model by Christiansen et al (16) resulted in a respectable correlation coefficient of 0.855 and an R2 of 0.64, it had a high chi-square value of 20.9 and a resulting high probability of 0.98 that the chi-square for a correct model would be less than the chi-square calculated for their model. The model by Kozusko (18) resulted in a correlation coefficient of 0.722 but had a poor R2 (ie, 0.067) and a high chi-square value of 19.1, which gave a probability of 0.98 that the chi-square for a correct model would be less than the chi-square calculated for his model. The predictions based on the IOM-NAS equations (25) resulted in a correlation coefficient of 0.60, but a very poor R2 (ie, 1.79), which indicated that the mean of the data provided a better description of the body weight change than the predictions from the IOM-NAS equations. Correspondingly, the chi-square value of 184 gave a probability of 1 that the chi-square for a correct model would be less than the chi-square calculated for the IOM-NAS equations.
DISCUSSION

Measured FM/ BW

0.4

0.2

0 0 0.2 0.4 0.6 0.8 1

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

Predicted FM/BW
FIGURE 2. Predicted versus measured changes ( ) in the fat fraction (FM) of body weight (BW) loss ( FM/ BW) (R2 0.90). The dotted lines are lines of identity, and error bars are SEs.

in a study by Das et al (9). For the obese subject starting at a body weight of 139.5 kg, a 400-kcal/d reduction in dietary intake (with no change of physical activity) resulted in a predicted bodyweight loss of 23.4 kg, whereas the overweight subject whose initial weight was 82.3 kg lost only 13.7 kg for the same reduction in dietary EI. The dotted curves on either side of each line indicate the range of predicted weight changes when the model parameter was swept by 1 SD. Practical calculation of dietary changes required to prevent weight regain Our proposed model calculations for sedentary 40-y-old men and women are shown in Table 2 and Table 3, respectively, under the assumption that these persons physical activity level 0). Separate calcudoes not change with weight loss (ie, lations were required for men and women because women have a higher amount of body fat than do men of similar weight. Thus, whereas sex was not an explicit variable in our equations, specification of the initial body composition at a given body weight differed between men and women. The initial body composition and the initial physical activity parameter ( init) were estimated by using the regression equations of Jackson et al (27) and Mifflin et al (28), respectively, with an assumed initial PAL of 1.4. Whereas these calculations provided a useful estimate of the expected steady-state weight change as a function of EI changes for sedentary people, we have also provided spreadsheet files for calculations of changes in physical activity along with changes in EI (for these files, see Spreadsheet files under Supplemental data in the current online issue; the spreadsheet files and an online version of the model are available at http://www2.niddk. nih.gov/NIDDKLabs/LBM/lbmHall.htm). These spreadsheet files

Current methods estimate body weight loss on the basis of simplified rules such as 3500 kcal 1 pound (5, 29). Because such rules do not account for changes in EE with weight loss, they do not allow for stabilization of body weight at a new steady state despite continued adherence to a lifestyle intervention. Unfortunately, this crude approximation currently is the only widespread
0 -5 -10 -15

BW (kg)

-20 -25 -30 -35 -40 -45 -50 -800

-700

-600

-500

-400

-300

-200

-100

EI ( kcal/d)
FIGURE 3. Predicted diet-induced weight losses ( BW) as a function of dietary intake changes ( EI) for 2 subjects with very different initial body weights corresponding to average data from Martin et al (12) (, initial BW 82.3 kg) and Das et al (9) (- - -, initial BW 139.5 kg). , change; BW, body weight. The dotted lines indicate the sensitivity of the model predictions when the parameter was changed by 1 SD.

1500

HALL AND JORDAN

1.4) of average height (1.77 m) assuming no change of physical activity ( 0)1

TABLE 2 Predicted body weight (BW) changes in 40-y-old sedentary men (PAL

Change in energy intake (kcal/d) BWinit (kg) 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160 165 170 175 180 185 190 195 100 4.7 5.0 5.3 5.5 5.8 6.0 6.2 6.4 6.6 6.8 7.0 7.1 7.3 7.5 7.6 7.8 7.9 8.0 8.2 8.3 8.4 8.5 8.6 8.7 200 9.2 9.8 10.3 10.8 11.3 11.7 12.2 12.6 13.0 13.4 13.7 14.1 14.4 14.7 15.1 15.3 15.6 15.9 16.2 16.4 16.6 16.9 17.1 17.3 300 13.5 14.3 15.1 15.8 16.5 17.2 17.9 18.5 19.1 19.7 20.3 20.8 21.3 21.8 22.3 22.7 23.2 23.6 24.0 24.4 24.7 25.1 25.4 25.7 400 17.6 18.6 19.6 20.6 21.5 22.4 23.3 24.2 25.0 25.8 26.6 27.3 28.0 28.7 29.3 29.9 30.5 31.1 31.6 32.2 32.7 33.1 33.6 34.0 500 21.4 22.7 23.9 25.1 26.3 27.4 28.5 29.6 30.7 31.7 32.6 33.6 34.4 35.3 36.1 36.9 37.7 38.4 39.1 39.8 40.4 41.0 41.6 42.2 600 700 800 kg 26.5 28.0 29.4 30.8 32.1 33.5 34.7 36.0 37.2 38.4 39.5 40.6 41.7 42.7 43.6 44.6 45.5 46.3 47.2 47.9 48.7 49.4 50.1 900 1000 1100 1200 1300 1400 1500

31.8 33.4 35.0 36.6 38.1 39.6 41.1 42.5 43.9 45.2 46.5 47.8 49.0 50.1 51.2 52.3 53.3 54.3 55.3 56.2 57.0 57.9

39.0 40.8 42.5 44.2 45.9 47.5 49.1 50.6 52.1 53.6 55.0 56.3 57.6 58.9 60.1 61.2 62.3 63.4 64.4 65.4

44.8 46.7 48.5 50.4 52.2 54.0 55.7 57.4 59.1 60.7 62.2 63.7 65.1 66.5 67.9 69.2 70.4 71.6 72.8

50.6 52.6 54.7 56.6 58.6 60.5 62.4 64.3 66.1 67.8 69.5 71.1 72.7 74.2 75.7 77.1 78.5 79.8

56.5 58.7 60.8 63.0 65.1 67.1 69.2 71.2 73.1 75.0 76.8 78.6 80.3 82.0 83.6 85.1 86.6

64.8 67.1 69.4 71.6 73.8 76.0 78.1 80.1 82.2 84.1 86.0 87.9 89.7 91.4 93.1

71.0 73.4 75.8 78.1 80.5 82.8 85.0 87.2 89.4 91.5 93.5 95.5 97.4 99.3

79.7 82.2 84.7 87.2 89.6 92.0 94.3 96.6 98.8 101.0 103.1 105.2

86.1 88.7 91.3 93.9 96.4 98.9 101.4 103.8 106.2 108.5 110.7

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

1 Blank cells indicate that the prescribed intake changes would result in a BMI (in kg/m2) 18.5. PAL, physical activity level; , physical activity parameter; BWinit, initial BW.

clinical tool available for predicting weight change, and any improvement represents a significant step forward. We addressed this problem by using the available longitudinal weight-loss data to develop a mathematical model that accounts for reduced energy requirements at the lower steady-state body weight. Mathematical modeling of human body-weight change has been attempted many times and involves solving differential equations that model the rate of body-weight change (16, 18, 30 43). However, these mathematical models typically require specialized software to numerically approximate the equation solutions via computer simulation. In contrast, we solved our nonlinear model equations at the new steady state and provided standard spreadsheet files so that readers can easily perform the calculations for themselves (see Spreadsheet files under Supplemental data in the current online issue; the spreadsheet files and an online version of the model are available at http://www2. niddk.nih.gov/NIDDKLabs/LBM/lbmHall.htm). Previous mathematical models of human body-weight change have been used to explicitly calculate the steady-state bodyweight change for prescribed changes in diet or physical activity (16 18). In the present study, we showed that, whereas 2 of the previous steady-state models provided a reasonably good correlation with the data (16, 18), our model performed significantly better with a chi-square value one-tenth that of the previous models. The predictions based on the IOM-NAS EE equations were surprisingly poor, and they indicated that these cross-sectional equations should not be used to predict weight change. This conclusion is in contrast to the recent study by Heymsfield et al

(17) that reported good agreement between the IOM-NAS equations and the measured EE rate of formerly obese subjects. However, most of the analysis performed in that study involved crosssectional comparisons. Only 3 longitudinal data sets were used by Heymsfield et al, and their analysis pooled these data and did not report information about variability of the results or model statistics. Our analysis included these same 3 longitudinal data sets, but the data were not pooled, and we included 5 additional longitudinal weight-loss studies. Therefore, we believe that the existing longitudinal weight-loss data do not support the use of the cross-sectional IOM-NAS equations to predict weight change. Our model was also used to calculate the dietary EI changes required to maintain a given amount of weight loss and to prevent weight regain. This important question has been previously addressed by Heymsfield et al (44) in another study using the IOM-NAS cross-sectional EE equations. However, given the poor performance of these equations in predicting steady-state weight change, we do not recommend their use for predicting changes of EI. Furthermore, the assumed linear dependence of EE on body weight implies that a specified decrement in dietary intake leads to the same weight change regardless of the initial body weight or body composition. This result ignores the facts that body-composition changes are likely to be nonlinear functions of the initial fat mass and the magnitude of weight loss (22, 23) and that FFM contributes to EE to a greater degree than does body fat (19, 45). Our proposed model incorporates nonlinear changes in body composition with weight loss, and Figure 2

MODELING WEIGHT-LOSS MAINTENANCE

TABLE 3 Predicted long-term body weight (BW) changes in 40-y-old sedentary women (PAL activity ( 0)1

1501

1.4) of average height (1.63 m) assuming no change in physical

Change in energy intake (kcal/d) BWinit (kg) 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160 165 170 175 180 185 190 195
1

100 5.9 6.2 6.4 6.7 6.9 7.1 7.3 7.5 7.7 7.8 8.0 8.2 8.3 8.4 8.6 8.7 8.8 8.9 9.0 9.2 9.3 9.3 9.4 9.5

200 11.5 12.1 12.6 13.1 13.5 14.0 14.4 14.8 15.2 15.5 15.8 16.2 16.5 16.7 17.0 17.3 17.5 17.8 18.0 18.2 18.4 18.6 18.8 19.0

300 16.9 17.7 18.5 19.2 19.9 20.6 21.3 21.9 22.4 23.0 23.5 24.0 24.4 24.9 25.3 25.7 26.1 26.4 26.8 27.1 27.4 27.7 28.0 28.3

400 22.0 23.1 24.1 25.1 26.1 27.0 27.9 28.7 29.5 30.2 30.9 31.6 32.2 32.8 33.4 34.0 34.5 35.0 35.4 35.9 36.3 36.7 37.1 37.5

500 26.9 28.2 29.5 30.8 32.0 33.1 34.2 35.3 36.3 37.2 38.1 39.0 39.8 40.6 41.3 42.0 42.7 43.3 43.9 44.5 45.1 45.6 46.1 46.6

600

700

800 kg

900

1000

1100

1200

1300

1400

1500

33.0 34.6 36.1 37.5 38.9 40.3 41.6 42.8 44.0 45.1 46.1 47.1 48.1 49.0 49.9 50.7 51.5 52.3 53.0 53.7 54.3 55.0 55.5

39.4 41.1 42.8 44.5 46.1 47.6 49.0 50.4 51.8 53.0 54.2 55.4 56.5 57.6 58.6 59.5 60.4 61.3 62.1 62.9 63.6 64.4

47.8 49.7 51.5 53.3 55.0 56.6 58.2 59.6 61.1 62.4 63.7 65.0 66.1 67.3 68.3 69.4 70.3 71.3 72.2 73.0

54.6 56.7 58.7 60.6 62.5 64.3 66.0 67.6 69.2 70.7 72.1 73.5 74.8 76.1 77.2 78.4 79.5 80.5 81.5

63.8 65.9 68.0 70.0 72.0 73.8 75.6 77.4 79.0 80.6 82.1 83.5 84.9 86.2 87.4 88.6 89.8

73.3 75.5 77.7 79.8 81.8 83.7 85.6 87.4 89.1 90.7 92.3 93.8 95.2 96.5 97.8

83.0 85.3 87.6 89.8 91.8 93.8 95.8 97.6 99.4 101.1 102.7 104.2 105.7

90.6 93.1 95.5 97.8 100.0 102.1 104.2 106.2 108.1 109.9 111.6 113.3

103.4 105.8 108.2 110.5 112.6 114.8 116.8 118.7 120.6

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010

113.9 116.4 118.8 121.1 123.4 125.5 127.6

Blank cells indicate that the prescribed intake changes would result in a BMI (in kg/m2) 18.5. PAL, physical activity level; , physical activity parameter; BWinit, initial BW.

shows that the resulting model predictions for body-composition change match the data quite well. Further illustrating the importance of this effect, Figure 3 showed that different initial body compositions can lead to different degrees of weight change for the same reduction in dietary intake. In our model, higher initial body fat leads to greater predicted weight change for an equal decrement in dietary EI.
Proposed Model IOM - NAS equations Christiansen
20

This can be shown mathematically by considering the simplified case in which the fraction of weight lost as body fat is specified by a parameter ( ). Equation 3 can then be rearranged to obtain the following equation for the expected weight change:

BW
FM

1
init

EI

BWinit
FFM FM

(14)

40

Assuming that there is no change in physical activity (ie, 0), the predicted steady-state weight change per unit reduction of EI varies with the parameter according to the following equation:

Predicted Measured BW (kg)

Kozusko

BW EI

FFM FFM init

FM

1
FM 2

(15)

FFM

-20

-40

-60

-80

FIGURE 4. The difference between the predicted and measured weight losses ( BW) are plotted for each mathematical model. , change; BW, body weight. Each data-point compares the model prediction with the mean value for each group of subjects.

This quantity is greater than zero for FFM FM, so the expected weight loss is enhanced as the fraction of weight lost as body fat, , increases. Because the Forbes body-composition theory states that is an increasing function of the initial body fat, persons with higher initial body fat will eventually lose more body weight for a specified decrement in dietary EI. At first glance, this result appears to be contradictory to our previous study showing that the required energy deficit per unit of weight loss is higher in obese than in lean persons (29). The physiologic explanation derives from the fact that persons with higher initial body fat lose a smaller proportion of their metabolically expensive FFM and are thereby better able to preserve their total EE rate during weight loss. In contrast, an initially lean

1502

HALL AND JORDAN

individual will lose a greater amount of FFM, which concomitantly decreases the EE rate to a greater degree. Although a greater cumulative energy deficit is required per unit of body weight lost by persons with higher initial body fat (29), this energy deficit will be more readily achieved by persons with higher initial body fat, because the metabolically expensive FFM is spared, whereas the metabolically more inert but energy-dense FM is lost. Some studies suggested that, after the steady-state body weight has been reached, no adaptation of EE occurs beyond that predicted by body composition change alone (8, 9, 14, 46 52). Others maintained that changes in EE after weight loss cannot be accounted for by body composition changes (1113, 15, 20, 5357). Many of these previous studies evaluated resting metabolic rate in the overnight fasted state by using indirect calorimetry, which accounts for only a fraction of the total daily EE, or 24-h EE in a metabolic chamber, which does not represent the free-living situation. In contrast, the present study used the available doubly labeled water data (8, 9, 1215)the gold standard for measuring freeliving total EE (58)to model changes of total free-living EE after a new steady-state body weight was achieved. The best fit value of the unknown parameter (ie, 0.24 0.13) was clearly greater than zero and also 0.1, which is the typical value used to represent the thermic effect of feeding (11). This indicates that some degree of adaptation of total EE beyond that expected from body weight change alone was required to explain the experimental observations. We do not propose a physiologic mechanism for such an adaptation. But when using our simplified model of EE rate, with body FM and FFM as the only independent variables, an additional term related to the EI change was required to adequately represent the data. We do not want to leave the reader with the impression that our model can make precise calculations of weight change or the required lifestyle changes to prevent weight regain by individual subjects. The variability in the weight loss data would suggest that precise calculations would be very difficult with the use of a general model whose parameters are not adjusted for individual subjects. Prospective evaluation of our model using individual subjects, with the possibility of developing personalized models of weight change, will be the subject of future research. Nevertheless, given the current inability to make any quantitative estimates regarding the expected level of body-weight stabilization or the required lifestyle changes to prevent weight regain, our model represents a significant step forward that we believe is useful for setting goals before an obesity intervention. Given a weight-change goal, BW, equation 10 allows for testing of various lifestyle intervention scenarios by specifying the prescribed physical activity change ( ) and then calculating the dietary EI change ( EI) that would be required to maintain weight loss and prevent weight regain. The physician and patient can then evaluate whether long-term adherence to such an intervention is a realistic possibility. For example, consider a 150-kg woman who wants to lose 50% of her body weight and prevent regain. Our model (Table 3) suggests that this would require a permanent reduction of 1000 kcal/d from the diet to prevent weight regain, a goal that may not be realistic in the long term. However, by increasing her physical activity, she will be able to cut back less on her diet. As mentioned earlier, we have provided spreadsheet files to facilitate such a calculation (see Spreadsheet files under Supplemental data in the current online issue; the

spreadsheet files and an online version of the model are available at http://www2.niddk.nih.gov/NIDDKLabs/LBM/lbmHall.htm). It is important to emphasize that the present study has not dealt with the time-course of weight loss. If the calculated intervention for maintaining a desired weight change was implemented at the onset of obesity treatment, it would likely take several years for the body weight to reach the steady state (59). Therefore, it may be beneficial to partition an obesity intervention into a weightloss phase followed by a weight-maintenance phase. The model proposed here would be most useful for helping define the dietary and physical activity changes required for the weight maintenance phasea problem for which no clinical tool is currently available.
We thank SR Smith for suggesting the creation and inclusion of Tables 2 and 3. The authors responsibilities were as follows both authors contributed to the design of the study, the analysis of the results, and the writing of the manuscript. Neither author had a personal or financial conflict of interest.

REFERENCES
Downloaded from www.ajcn.org at Guadalajara on April 13, 2010 1. Must A, Spadano J, Coakley EH, Field AE, Colditz G, Dietz WH. The disease burden associated with overweight and obesity. JAMA 1999; 282:15239. 2. Stunkard A, McLaren-Hume M. The results of treatment for obesity: a review of the literature and report of a series. Arch Intern Med 1959; 103:79 85. 3. Svetkey LP, Stevens VJ, Brantley PJ, et al. Comparison of strategies for sustaining weight loss: the weight loss maintenance randomized controlled trial. JAMA 2008;299:1139 48. 4. Hill JO. Understanding and addressing the epidemic of obesity: an energy balance perspective. Endocr Rev 2006;27:750 61. 5. NHLBI Obesity Education Initiative Expert Panel on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. Practical guide to the identification, evaluation, and treatment of overweight and obesity in adults. Bethesda, MD: National Heart, Lung, and Blood Institute, 2000. 6. Wing RR, Hill JO. Successful weight loss maintenance. Annu Rev Nutr 2001;21:323 41. 7. Wing RR, Phelan S. Long-term weight loss maintenance. Am J Clin Nutr 2005;82(suppl):222S5S. 8. Amatruda JM, Statt MC, Welle SL. Total and resting energy expenditure in obese women reduced to ideal body weight. J Clin Invest 1993;92: 1236 42. 9. Das SK, Roberts SB, McCrory MA, et al. Long-term changes in energy expenditure and body composition after massive weight loss induced by gastric bypass surgery. Am J Clin Nutr 2003;78:2230. 10. Keys A. The biology of human starvation. Minneapolis, MN: University of Minnesota Press, 1950; 11. Leibel RL, Rosenbaum M, Hirsch J. Changes in energy expenditure resulting from altered body weight. N Engl J Med 1995;332:621 8. 12. Martin CK, Heilbronn LK, de Jonge L, et al. Effect of calorie restriction on resting metabolic rate and spontaneous physical activity. Obesity (Silver Spring) 2007;15:2964 73. 13. van Gemert WG, Westerterp KR, van Acker BA, et al. Energy, substrate and protein metabolism in morbid obesity before, during and after massive weight loss. Int J Obes Relat Metab Disord 2000;24:711 8. 14. Weinsier RL, Hunter GR, Zuckerman PA, et al. Energy expenditure and free-living physical activity in black and white women: comparison before and after weight loss. Am J Clin Nutr 2000;71:1138 46. 15. Westerterp KR, Saris WH, Soeters PB, ten Hoor F. Determinants of weight loss after vertical banded gastroplasty. Int J Obes 1991;15:529 34. 16. Christiansen E, Garby L. Prediction of body weight changes caused by changes in energy balance. Eur J Clin Invest 2002;32:826 30. 17. Heymsfield SB, Harp JB, Reitman ML, et al. Why do obese patients not lose more weight when treated with low-calorie diets? A mechanistic perspective. Am J Clin Nutr 2007;85:346 54. 18. Kozusko FP. The effects of body composition on setpoint-based weight loss. Math Comput Model 2002;29:17.

MODELING WEIGHT-LOSS MAINTENANCE

19. Nelson KM, Weinsier RL, Long CL, Schutz Y. Prediction of resting energy expenditure from fat-free mass and fat mass. Am J Clin Nutr 1992;56:848 56. 20. Major GC, Doucet E, Trayhurn P, Astrup A, Tremblay A. Clinical significance of adaptive thermogenesis. Int J Obes (Lond) 2007;31:204 12. 21. Flatt JP. Exaggerated claim about adaptive thermogenesis. Int J Obes (Lond) 2007;31:1626(letter). 22. Hall KD. Body fat and fat-free mass inter-relationships: Forbess theory revisited. Br J Nutr 2007;97:1059 63. 23. Forbes GB. Lean body mass-body fat interrelationships in humans. Nutr Rev 1987;45:22531. 24. Corless R, Gonnet G, Hare D, Jeffery D, Knuth D. On the Lambert W function. Adv Computat Math 1996;5:329 59. 25. Dietary reference intakes for energy, carbohydrate, fiber, fat, fatty acids, cholesterol, protein, and amino acids. Washington, DC: The National Academies Press, 2002. 26. Press W, Flannery B, Teukolsky S, Vetterling W. Numerical recipes: the art of scientific computing. New York, NY: Cambridge University Press, 1986; 27. Jackson AS, Stanforth PR, Gagnon J, et al The effect of sex, age and race on estimating percentage body fat from body mass index: The Heritage Family Study. Int J Obes Relat Metab Disord 2002;26:789 96. 28. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr 1990;51:2417. 29. Hall KD. What is the required energy deficit per unit weight loss? Int J Obes (Lond) 2008;32:573 6. 30. Alpert SS. A two-reservoir energy model of the human body. Am J Clin Nutr 1979;32:1710 8. 31. Alpert SS. Growth, thermogenesis, and hyperphagia. Am J Clin Nutr 1990;52:784 92. 32. Alpert SS. A limit on the energy transfer rate from the human fat store in hypophagia. J Theor Biol 2005;233:113. 33. Antonetti VW. The equations governing weight change in human beings. Am J Clin Nutr 1973;26:64 71. 34. Christiansen E, Garby L, Sorensen TI. Quantitative analysis of the energy requirements for development of obesity. J Theor Biol 2005;234: 99 106. 35. Flatt JP. Carbohydrate-fat interactions and obesity examined by a twocompartment computer model. Obes Res 2004;12:201322. 36. Girardier L. [Autoregulation of body weight and body composition in man. A systematic approach through models and simulation]. Arch Int Physiol Biochim Biophys 1994;102:A2335 (in French). 37. Hall KD. Computational model of in vivo human energy metabolism during semistarvation and refeeding. Am J Physiol Endocrinol Metab 2006;291:E2337. 38. Kozusko FP. Body weight setpoint, metabolic adaption and human starvation. Bull Math Biol 2001;63:393 403. 39. Payne PR, Dugdale AE. A model for the prediction of energy balance and body weight. Ann Hum Biol 1977;4:52535. 40. Payne PR, Dugdale AE. Mechanisms for the control of body-weight. Lancet 1977;1:583 6. 41. Song B, Thomas DM. Dynamics of starvation in humans. J Math Biol 2007;54:27 43. 42. Weinsier RL, Bracco D, Schutz Y. Predicted effects of small decreases

1503

43.

44.

45.

46.

47.

48. 49.

50.

51.

52.

53.

54. 55.

56.

57.

58. 59.

in energy expenditure on weight gain in adult women. Int J Obes Relat Metab Disord 1993;17:693700. Westerterp KR, Donkers JH, Fredrix EW, Boekhoudt P. Energy intake, physical activity and body weight: a simulation model. Br J Nutr 1995; 73:337 47. Heymsfield SB, Harp JB, Rowell PN, Nguyen AM, Pietrobelli A. How much may I eat? Calorie estimates based upon energy expenditure prediction equations. Obes Rev 2006;7:36170. Weyer C, Snitker S, Rising R, Bogardus C, Ravussin E. Determinants of energy expenditure and fuel utilization in man: effects of body composition, age, sex, ethnicity and glucose tolerance in 916 subjects. Int J Obes Relat Metab Disord 1999;23:71522. Bessard T, Schutz Y, Jequier E. Energy expenditure and postprandial thermogenesis in obese women before and after weight loss. Am J Clin Nutr 1983;38:680 93. Buemann B, Astrup A, Madsen J, Christensen NJ. A 24-h energy expenditure study on reduced-obese and nonobese women: effect of betablockade. Am J Clin Nutr 1992;56:66270. Larson DE, Ferraro RT, Robertson DS, Ravussin E. Energy metabolism in weight-stable postobese individuals. Am J Clin Nutr 1995;62:7359. Nelson KM, Weinsier RL, James LD, Darnell B, Hunter G, Long CL. Effect of weight reduction on resting energy expenditure, substrate utilization, and the thermic effect of food in moderately obese women. Am J Clin Nutr 1992;55:924 33. Weinsier RL, Nagy TR, Hunter GR, Darnell BE, Hensrud DD, Weiss HL. Do adaptive changes in metabolic rate favor weight regain in weight-reduced individuals? An examination of the set-point theory. Am J Clin Nutr 2000;72:1088 94. Weinsier RL, Nelson KM, Hensrud DD, Darnell BE, Hunter GR, Schutz Y. Metabolic predictors of obesity. Contribution of resting energy expenditure, thermic effect of food, and fuel utilization to four-year weight gain of post-obese and never-obese women. J Clin Invest 1995;95: 980 5. Wyatt HR, Grunwald GK, Seagle HM, et al. Resting energy expenditure in reduced-obese subjects in the National Weight Control Registry. Am J Clin Nutr 1999;69:1189 93. Doucet E, St-Pierre S, Almeras N, Despres JP, Bouchard C, Tremblay A. Evidence for the existence of adaptive thermogenesis during weight loss. Br J Nutr 2001;85:71523. Leibel RL, Hirsch J. Diminished energy requirements in reduced-obese patients. Metabolism 1984;33:164 70. Vettor R, Mingrone G, Manco M, et al. Reduced expression of uncoupling proteins-2 and -3 in adipose tissue in post-obese patients submitted to biliopancreatic diversion. Eur J Endocrinol 2003;148:54350. Weigle DS, Sande KJ, Iverius PH, Monsen ER, Brunzell JD. Weight loss leads to a marked decrease in nonresting energy expenditure in ambulatory human subjects. Metabolism 1988;37:930 6. Weyer C, Pratley RE, Salbe AD, Bogardus C, Ravussin E, Tataranni PA. Energy expenditure, fat oxidation, and body weight regulation: a study of metabolic adaptation to long-term weight change. J Clin Endocrinol Metab 2000;85:108794. Schoeller DA. Measurement of energy expenditure in free-living humans by using doubly labeled water. J Nutr 1988;118:1278 89. Chow CC, Hall KD. The dynamics of human body weight change. PLoS Comput Biol 2008;4:e1000045.

Downloaded from www.ajcn.org at Guadalajara on April 13, 2010