- Email: [email protected]
Reply to Thibault & Genton:
Clinical Nutrition 33 (2014) 1158e1159
Contents lists available at ScienceDirect
Clinical Nutrition journal homepage: http://www.elsevier.com/locate/clnu
Reply
Reply to Thibault & Genton: We thank Thibault and Genton for their interest in our paper [1], in which we compared measurements of skeletal muscle mass (SMM) by DXA and single-frequency bioimpedance (BIA), where SMM by BIA was calculated according to three previously published equations. We found that one BIA equation (Tengvall) [2] yielded an accurate estimate of DXA-derived SMM, meaning that the group mean value was not significantly different from the DXA SMM estimate. The two other BIA equations [3,4] over-estimated SMM compared to DXA (both p < 0.001), but all equations were highly correlated. We concluded that the BIA equation used, developed in a different healthy elderly population, gave an accurate estimate of DXAderived SMM in a population with various clinical disorders; and that BIA appears potentially capable to estimate SMM in clinical disorders, but the optimal approach to its use for this purpose requires further investigation. Thibault and Genton argue that the presented data do not support the part of the conclusion that the Tengvall equation was the most accurate, and present three arguments for this: First, that the mean difference was only slightly better when using the Tengvall equation than the Kyle equation (0.15 ± 2.36 vs 0.90 ± 2.08 kg). We agree that the difference was small, but nonetheless statistically significant, and thus we do not think this argument refutes the conclusion. The second argument has, as far as we can see, three components: a/ That the BlandeAltman graph shows that the difference using the Tengvall equation has a higher 95% confidence interval than with the Kyle equation: This is correct, as reflected in the small difference in SD (2.36 vs 2.08 kg), but does not contradict the conclusion about accuracy. b/ That there is a slope in the linear fit of the data in the BlandeAltman graphs, that is present in the Tengvall comparison but almost absent in the Kyle comparison: This is correct, but the determination coefficients for the linear fit, as presented in the graphs, are very small (0.008 for Kyle and 0.046 for Tengvall), and was largest for the third comparison (0.148 for the Janssen equation). c/ That the Kyle equation yielded more outlying values, but looking at the graphs, an equal number of observations outside the 95% CI are found: As stated in the paper [1] (Results section, second paragraph), we identified four outliers each for the Tengvall and Janssen comparisons, while the Kyle comparison showed eight outliers. Outliers were removed before the final
DOI of original article: http://dx.doi.org/10.1016/j.clnu.2014.08.004.
analysis, and thus do not show in the graphs. Statistically one would expect the same number of observations outside the 95% CI, which was also the case for the Tengvall and Kyle comparisons. Third, looking at the regressions, the determination coefficient is slightly better with the Kyle equation (0.939) than the Tengvall equation (0.915); visually, the Kyle formula has a lower standard error estimate; and the intercept seems closer to zero with the Kyle and Janssen equations: The differences in determination coefficients and SEE are small and related to the differences in SDs discussed above. We agree that the larger intercept in the Tengvall equation is a concern, and we discussed a number of factors in the equations that might affect performance, summarizing that “… all three equations seem to have advantages and disadvantages, but in most aspects the Kyle and Tengvall equations seemed to perform better in this study population, and the Tengvall equation alone produced an unbiased estimate of mean DXA-derived SMM.” Finally, as we conclude, we think BIA appears potentially capable to estimate SMM in clinical disorders, but the optimal approach to its use for this purpose requires further investigation. Equations could probably be somewhat improved, but we think it is unlikely that major improvements in this area could be expected. Thus, clinical use of BIA for SMM assessment should recognize and take into account the methodological limitations. We thank Thibault and Genton for their comments on these issues.
References [1] Bosaeus I, Wilcox G, Rothenberg E, Strauss BJ. Skeletal muscle mass in hospitalized elderly patients: comparison of measurements by single-frequency BIA and DXA. Clin Nutr 2014;33(3):426e31. [2] Tengvall M, Ellegard L, Malmros V, Bosaeus N, Lissner L, Bosaeus I. Body composition in the elderly: reference values and bioelectrical impedance spectroscopy to predict total body skeletal muscle mass. Clin Nutr 2009;28:52e8. [3] Janssen I, Heymsfield SB, Baumgartner RN, Ross R. Estimation of skeletal muscle mass by bioelectrical impedance analysis. J Appl Physiol 2000;89(2):465e71. [4] Kyle UG, Genton L, Hans D, Pichard C. Validation of a bioelectrical impedance analysis equation to predict appendicular skeletal muscle mass (ASMM). Clin Nutr 2003;22(6):537e43.
Ingvar Bosaeus* Clinical Nutrition Unit, Sahlgrenska University Hospital, Per Dubbsgatan 14, S-413 45 Gothenburg, Sweden Gisela Wilcox Dept. of Medicine, Southern Clinical School, Monash University, Clayton, Victoria, Australia
http://dx.doi.org/10.1016/j.clnu.2014.08.008 0261-5614/© 2014 Elsevier Ltd and European Society for Clinical Nutrition and Metabolism. All rights reserved.
Reply / Clinical Nutrition 33 (2014) 1158e1159
Elisabet Rothenberg Dept. of Food and Meal Science, Kristianstad University, SE-291 88 Kristianstad, Sweden Boyd Strauss Dept. of Medicine, Southern Clinical School, Monash University, Clayton, Victoria, Australia
1159 *
Corresponding author. Tel.: þ46 706 291 542; fax: þ46 31 342 3185. E-mail address: [email protected] (I. Bosaeus). 4 August 2014
Contents lists available at ScienceDirect
Clinical Nutrition journal homepage: http://www.elsevier.com/locate/clnu
Reply
Reply to Thibault & Genton: We thank Thibault and Genton for their interest in our paper [1], in which we compared measurements of skeletal muscle mass (SMM) by DXA and single-frequency bioimpedance (BIA), where SMM by BIA was calculated according to three previously published equations. We found that one BIA equation (Tengvall) [2] yielded an accurate estimate of DXA-derived SMM, meaning that the group mean value was not significantly different from the DXA SMM estimate. The two other BIA equations [3,4] over-estimated SMM compared to DXA (both p < 0.001), but all equations were highly correlated. We concluded that the BIA equation used, developed in a different healthy elderly population, gave an accurate estimate of DXAderived SMM in a population with various clinical disorders; and that BIA appears potentially capable to estimate SMM in clinical disorders, but the optimal approach to its use for this purpose requires further investigation. Thibault and Genton argue that the presented data do not support the part of the conclusion that the Tengvall equation was the most accurate, and present three arguments for this: First, that the mean difference was only slightly better when using the Tengvall equation than the Kyle equation (0.15 ± 2.36 vs 0.90 ± 2.08 kg). We agree that the difference was small, but nonetheless statistically significant, and thus we do not think this argument refutes the conclusion. The second argument has, as far as we can see, three components: a/ That the BlandeAltman graph shows that the difference using the Tengvall equation has a higher 95% confidence interval than with the Kyle equation: This is correct, as reflected in the small difference in SD (2.36 vs 2.08 kg), but does not contradict the conclusion about accuracy. b/ That there is a slope in the linear fit of the data in the BlandeAltman graphs, that is present in the Tengvall comparison but almost absent in the Kyle comparison: This is correct, but the determination coefficients for the linear fit, as presented in the graphs, are very small (0.008 for Kyle and 0.046 for Tengvall), and was largest for the third comparison (0.148 for the Janssen equation). c/ That the Kyle equation yielded more outlying values, but looking at the graphs, an equal number of observations outside the 95% CI are found: As stated in the paper [1] (Results section, second paragraph), we identified four outliers each for the Tengvall and Janssen comparisons, while the Kyle comparison showed eight outliers. Outliers were removed before the final
DOI of original article: http://dx.doi.org/10.1016/j.clnu.2014.08.004.
analysis, and thus do not show in the graphs. Statistically one would expect the same number of observations outside the 95% CI, which was also the case for the Tengvall and Kyle comparisons. Third, looking at the regressions, the determination coefficient is slightly better with the Kyle equation (0.939) than the Tengvall equation (0.915); visually, the Kyle formula has a lower standard error estimate; and the intercept seems closer to zero with the Kyle and Janssen equations: The differences in determination coefficients and SEE are small and related to the differences in SDs discussed above. We agree that the larger intercept in the Tengvall equation is a concern, and we discussed a number of factors in the equations that might affect performance, summarizing that “… all three equations seem to have advantages and disadvantages, but in most aspects the Kyle and Tengvall equations seemed to perform better in this study population, and the Tengvall equation alone produced an unbiased estimate of mean DXA-derived SMM.” Finally, as we conclude, we think BIA appears potentially capable to estimate SMM in clinical disorders, but the optimal approach to its use for this purpose requires further investigation. Equations could probably be somewhat improved, but we think it is unlikely that major improvements in this area could be expected. Thus, clinical use of BIA for SMM assessment should recognize and take into account the methodological limitations. We thank Thibault and Genton for their comments on these issues.
References [1] Bosaeus I, Wilcox G, Rothenberg E, Strauss BJ. Skeletal muscle mass in hospitalized elderly patients: comparison of measurements by single-frequency BIA and DXA. Clin Nutr 2014;33(3):426e31. [2] Tengvall M, Ellegard L, Malmros V, Bosaeus N, Lissner L, Bosaeus I. Body composition in the elderly: reference values and bioelectrical impedance spectroscopy to predict total body skeletal muscle mass. Clin Nutr 2009;28:52e8. [3] Janssen I, Heymsfield SB, Baumgartner RN, Ross R. Estimation of skeletal muscle mass by bioelectrical impedance analysis. J Appl Physiol 2000;89(2):465e71. [4] Kyle UG, Genton L, Hans D, Pichard C. Validation of a bioelectrical impedance analysis equation to predict appendicular skeletal muscle mass (ASMM). Clin Nutr 2003;22(6):537e43.
Ingvar Bosaeus* Clinical Nutrition Unit, Sahlgrenska University Hospital, Per Dubbsgatan 14, S-413 45 Gothenburg, Sweden Gisela Wilcox Dept. of Medicine, Southern Clinical School, Monash University, Clayton, Victoria, Australia
http://dx.doi.org/10.1016/j.clnu.2014.08.008 0261-5614/© 2014 Elsevier Ltd and European Society for Clinical Nutrition and Metabolism. All rights reserved.
Reply / Clinical Nutrition 33 (2014) 1158e1159
Elisabet Rothenberg Dept. of Food and Meal Science, Kristianstad University, SE-291 88 Kristianstad, Sweden Boyd Strauss Dept. of Medicine, Southern Clinical School, Monash University, Clayton, Victoria, Australia
1159 *
Corresponding author. Tel.: þ46 706 291 542; fax: þ46 31 342 3185. E-mail address: [email protected] (I. Bosaeus). 4 August 2014