| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
LETTER TO THE EDITOR |
Department of Biological Sciences
Ohio University
Athens, OH 45701
E-mail: loucks{at}ohio.edu
Dear Sir:
de Jonge et al (1) investigated the validity of estimates of energy intake (EI) during calorie restriction (CR) determined by the combination of doubly labeled water (DLW) and changes in body composition determined by dual-energy X-ray absorptiometry (DXA). They concluded that DLW+DXA overestimated EI by 8.7 ± 36.7% and, therefore, that "EI cannot be assessed accurately during CR by a combination of DLW and DXA" over brief (ie, practical) periods of time, because of large interindividual variability caused by random errors in the measurement of changes in fat mass and fat-free mass by DXA. This disappointing conclusion may have been premature.
For 7 d, de Jonge et al measured the energy expenditure of 10 subjects by DLW in a room calorimeter while the subjects were fed a 30% CR diet prepared in the authors' metabolic kitchen. The Bland-Altman plot in the authors' Figure 2B
provided the proper (2) comparison of their measurements of energy expenditure by DLW and calorimetry. This plot illustrated the authors' encouraging finding that the 2 measurements differed by an average of only 1.3 ± 8.9%. However, the plot may also be interpreted as showing that the random variation of these differences increases with a CV of 30% of energy expenditure in excess of 1600 kcal/d. This is not encouraging.
|
) was constructed from the data in the authors' Figure 3A. Taken at face value, this plot appears to indicate that most of the differences between the 2 calculations of EI are not random, as the authors had concluded. Regression analysis suggests that the differences are strongly proportional (R2 = 0.82, P < 0.005) and parallel (slope – 1 = 0.07 ± 0.18, P = 0.7) to EI above and below 1400 kcal/d. This proportionality appears to imply that an unidentified error in one calculation or the other could be corrected to reduce differences between the methods by 82%, which leaves a residual random error with an SEE of only 225 kcal/d (15%).
|
|
) shows that the regression analysis may not have faithfully represented the true relation between the 2 calculations. The highest and the 2 lowest data points do not fall on the same straight line as do the other normally distributed data points. What is worse, these extreme outliers with standardized residuals of –8.6, –7.5, and 16.6 from a regression line obtained without them have the strongest leverage on the regression analysis.
A second Bland-Altman plot constructed without these data points (Figure 3) shows that they accounted entirely for the apparent linear relation between the 2 calculations of EI. Neither the slope (P = 0.65) nor the intercept (P = 0.85) found by the second regression analysis is significant, and the residuals are normally distributed (plot not shown).
The mean (± SD) of the 7 remaining data points appears to indicate that DLW+DXA calculations yield estimates of EI that are 310 ± 116 kcal/d higher than the calculations made by food preparers. Per this analysis, DLW+DXA estimates of EI during CR appear to be less accurate and more precise than the estimates made by de Jonge et al (1). Compared with the average EI calculated by both methods, DLW+DXA appears to overestimate EI by 100 x [(310 kcal/1511 kcal) ± (116 kcal/1511 kcal)] = 21 ± 8%. Whether the 21% bias resides in the DLW+DXA technique or in the authors' metabolic kitchen cannot be determined from the authors' data.
Thus, the Bland-Altman and regression analyses indicate that the authors' conclusion about DLW+DXA not being suitable for estimating EI during short-term CR studies may have been premature. Their experiment appears to have generated 3 widely spaced data points precisely suggesting one relation between the 2 calculations of EI (a large difference proportional to EI) and 7 closely spaced data points precisely suggesting a completely different relation (a large constant difference).
This data set is too small and heterogeneous to warrant any further conclusion about how accurate and precise DLW+DXA may be, but accurate estimation of EI is of fundamental importance for everyone interested in CR. Therefore, investigators should be encouraged to collect and analyze more experimental data of this kind to search for potentially correctable sources of error in both calculations of EI and in the experimental methods that generate the data. If such sources are found and corrected, differences between the 2 calculations may be reduced to a random error of 8%.
ACKNOWLEDGMENTS
One of the authors of the de Jonge et al article (LM Redman) was a postdoctoral scholar in the laboratory of ABL. LB and ABL have no financial or personal relationships with any research sponsor that would constitute a conflict of interest.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
