Validation of bioimpedance body composition assessment by TANITA BC-418 in 7 years- old children vs. Dual X-ray absorptiometry Veronica Luque on behalf to URV Team 18th November 2012 Milano, 10th Bi-annual Nutrimenthe meeting WG on Anthropometry and Physical Activity
Aim Th i f th t k i t lid t TANITA BC 418 f The aim of the present work is to validate TANITA BC-418 for its clinical and epidemiological use to assess body composition in children, having DXA as the reference.
Methods Design: Cross sectional validation study Subjects: All children from the Spanish subsample of the EU Childhood Obesity Project, who took in part in the study at 7 years of age, from who both measures of bioimpedance and DXA were available. Outcome measures were: 1. DEXA 2. BIA (BIAoutputs) 3. New predictive equations from BIA (BIAregressions) Total fat & lean mass Trunk fat & lean mass Left arm and left leg fat and lean masses [Kg]. FMI and LMI [Kg/m 2 ]
Statistics T-test for repeated measures to test for differences between methods Linear regression models Boodstrap samples to validate new predictive equations Reliability: ICC Consistency: Cronbach alfa Agreement in quartile classification: Kappa Sensibility & specificity to assess children in the highest FMI quartile The agreement between methods was assessed through Bland and Altman plots.
Results: Description of body composition (BIA outputs vs DXA) N= 171 (84 boys, 87 girls). DXA median (IQR) BIA outputs median (IQR) Difference [%] median (IQR) Total Fat mass [Kg] 4.92 (3.43, 7.28) 5.60 (4.60, 7.05) 13.16 (-4.47, 37.82) TtlL Total Lean mass 19.53 (18.39, 19.50 (17.70, 70 [Kg] 21.16) 21.40) -1.02 (-4.76, 4.29) Trunk Fat mass [Kg] 30.04 (6.11, 192(128 1.92 (1.28, 294) 2.94) 240(205 2.40 (2.05, 325) 3.25) 66.67) Trunk Lean mass [Kg] 8.95 (8.30, 9.62) 12.45 (11.6, 41.07 (32.54, 13.35) 35) 46.74) Left arm, Fat mass 32.45 (-3.85, 0.28 (0.16, 0.47) 0.40 (0.30, 0.50) [Kg] 96.08) Left arm, Lean p<0.001 mass [Kg] vs DXA 0.88 (0.78, 0.98) 0.70 (0.60, 0.80) -19.29 (-25.35, - 12.60)
Results: Linear regression models with DXA results as dependent variables Depende Independe Β R 2 (*) Β (IC 95%) p- R 2 ( nt nt (IC Independe valu au % variables variables 95 nt e ) %) Variables Total Fat Total fat 1.343 94. Total body imp 0.011 (0.009, 0.012) 0.00 95.7 Mass mass (1.293, 3 [Ω] 1 % [Kg] (DXA) [Kg] from BIA outputs 1.393) % Weight [Kg] 0.873 (0.829, 0.919) 0.00 1 Height [cm] -0.231 (-0.270, - 0.00 0.192) 1 Gender 0.841 (0.623, 1.095) 0.00 1 Total Total lean 0.752 77. Total body imp -0.010 010 (-0.012, 012-0.0000 88.3 Lean mass [Kg] (0.691, 6 [Ω] 0.008) 1 % Mass from 0.813) % Weight [Kg] 0.110 (0.069, 0.151) 0.00 [Kg] BIA output 1 (DXA) s Height [cm] 0.251 (0.208, 0.292) 0.00 1 Gender -1.020 (-1.304, - 0.752) 0.00 1 Trunk Trunk Fat 1.241 86. Total body imp 0.005 (0.003, 0.006) 0.00 93.1 (*) FatBootstrapped Mass [Kg] regressions (1.168, 8 [Ω] 1 % Mass from 1.314) % Weight [Kg] 0.333 (0.279, 0.415) 0.00
Results: Description of body composition (BIA regressions vs DXA) N= 171 (84 boys, 87 girls). Difference vs. DXA BIA regressions DXA [%] median (IQR) median (IQR) median (IQR) Ttlft Total fat mass [Kg] [K] 4.92 (3.43, 5.51 (3.89, 4.62 (-2.61, 7.28) 7.36) 15.64) Total Lean Mass 19.53 (18.39, 19.54 (18.37, -0.21 (-2.67, [Kg] 21.16) 21.04) 2.09) Trunk Fat mass 1.92 (1.28, 2.39 (1.69, 20.79 (4.25, [Kg] 2.94) 3.36) 36) 37.88) Trunk Lean mass 8.95 (8.30, 9.08 (8.50, 1.54 (-1.63, [Kg] 9.62) 9.75) 4.47) 47) p<0.001 Left arm vs Fat DXA mass [Kg] 0.28 (0.16, 0.47) 0.22 (0.11, -20.83 (-45.75, 0.41) 4.72)
Results: reliability and consistency Cronbac h α BIA outputs ICC (CI 95%) Cronbach α BIA regressions ICC (CI 95%) Total fat mass [Kg] 0.960 0.911 (0.855, 0.989 0.975 (0.955, 0.942) 0.984) Ttll Total lean mass [Kg] 0.931 0.871 (0.829, 0.969 0.939 (0.919, 0.903) 0.955) Trunk fat mass [Kg] 0.945 0.839 (0.514, 0.982 0.936 (0.658, 0.927) 0.976) Trunk lean mass [Kg] 0.881 0.141 (-0.021, 0.949 0.897 (0.856, 0.443) 0.926) Left arm, fat mass 0.891 0.775 (0.650, 0.960 0.898 (0.761, [Kg] 0.850) 0.946) Left arm, Lean mass p<0.001 [Kg] 0.912 0.501 (-0.082, 0.945 0.863 (0.692, 0.808) 0.926)
Results: FMI quartiles QUARTILES CLASSIFICATION BY DXA QUARTILES CLASSIFICATIO N BY DXA Tot al Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Tot al QUARTILE S CLASSIFIC ATION BY BIAoutput s Q1 33 9 0 0 42 Q2 10 23 11 0 44 Q3 0 10 28 4 42 Q1 35 7 0 0 42 QUARTILES CLASSIFIC ATION BY Q2 8 28 8 0 44 PREDICTED Q3 VARIABLES 0 7 33 3 43 Q4 0 0 5 38 43 Q4 0 0 3 39 42 Total 43 42 44 42 171 Total 43 42 44 42 171 Κ= 0,618, p<0,001; Correct classification κ= 71.3% ***, p<0,001; Correct classification
Results: sensibility & specificity to classify in 4th FMI quartile 4th FMI quartile from BIAoutputs t 4th FMI quartile from predictive equations % (95 CI min, max) % (95 CI min, max) Sensitivity 88,4% (74.1, 95.6) 92.9% (79.4, 98.1) Specificity 96,9% (91.7, 99.0) 97.6% (92.8, 99.4) Positive predictive value 90.5% (76.4, 96.9) 92.9% (79.4, 98.1) Negative predictive value 96.1% (90.7, 98.6) 97.6% (92.8, 99.4)
Results: Bland & Altman plots. Agreement between techniques
Results: Bland & Altman plots. Agreement between techniques
Summary & Conclusions: Whole body composition predictions made by internal algorithms of TANITA BC-418, reveal that these are valid for epidemiological use. High reliability and consistency of measures support its use. The prediction of segmental composition is poorer. The use of own predictive equations (including segmental specific anthropometrical t measures) improves significantly ifi the precision of the method (which is strongly supported by the Bland & Altman analyses). Bioimpedance could be a useful technique at the clinical setting to support obesity diagnose and follow up (as shown by the high sensibility and specificity to properly classify children in the 4th FMI quartile)
Future plans: BIA vs. DXA paper ready to be submitted in December
Validation of anthropometry based body composition assessment in 7 years-old children vs. Dual X-ray absorptiometry Veronica Luque on behalf to URV Team 18th November 2012 Milano, 10th Bi-annual Nutrimenthe meeting WG on Anthropometry and Physical Activity
Aim The aim of the present work is to validate body composition The aim of the present work is to validate body composition assessed by anthropometrical methods in children, having DXA as the reference.
Methods Design: Cross sectional validation study Subjects: All children from the Spanish subsample of the EU Childhood Obesity Project, who took in part in the study at 7 years of age, from who both measures of anthropometrical evaluation and DXA were available.
Methods Outcome measures were: FM (kg) assessed by DXA FM (kg) assessed by anthropometry following three methods, Siri (a) and Deurenberg (b) equations as: Fat mass (kg) = [%Fat mas x weight (kg)] / 100 a) %Fat mass Siri = (4.95 D - 4.5) x 100 D = body density: D = 1.1533 (0.0643 x Log 4 skindfolds) for boys under 17y D = 1.1369 (0.0598 x Log 4 skindfolds) for girls under 17y b) %Fat mass Deurenberg = [562-4.2(age-2)] / D [525-4.7(age- 2)] D = body density: Males: D=1.1133-0.0561 x Log 4 skindfolds + 1.7 (age x 10-3 ) Females: D=1.1187 1187 0.063063 x Log 4 skindfolds + 1.9 (age x 10-3 ) c) FM (Kg) predicted by boodstraped linear regression model
Statistics T-test for repeated measures to test for differences between methods Linear regression models Boodstrap samples to validate new predictive equations Reliability: ICC Consistency: Cronbach alfa Agreement in quartile classification: Kappa Sensibility & specificity to assess children in the highest FM quartile The agreement between methods was assessed through Bland and Altman plots.
Results: Description of FM (anthropometry vs DXA) N= 169 (84 boys, 85 girls). DXA median (IQR) Anthropometry Siri equation median (IQR) Difference [%] median (IQR) TtlFt Total Fat mass [Kg] 4.91 (3.36, 7.26) 3.49 (2.58, 4.86) -28.39 (-34.47, -21.08) DXA median (IQR) Anthropometry Deurenberg equation median (IQR) Difference [%] median (IQR) Total Fat mass [Kg] 4.91 (3.36, 7.26) 439(332 4.39 (3.32, 598) 5.98) -11.77 (-19.17, 0.00) DXA Anthropometry t Bootstrap t regresions Difference [%] median (IQR) median (IQR) median (IQR) Total Fat mass [Kg] 4.91 (3.36, 7.26) 4.75 (3.56, 7.30) 0.01 (-0.56, 0.63) DXA: Dual X-Ray measurements; Difference between methods [%] was calculated as (Anthrooutput - DXA)*100/DXA. p<0.001 vs DXA
Results: reliability and consistency Cronbach α ICC (CI 95%) Total fat mass [Kg] Siri equation 0.931 0.838 (0.018, 0.945) 09 Total lean mass [Kg] 0.959 0.940 (0.819, Deurenberg equation 0.972) Total lean mass[kg] p<0.001 0.974 0.974 (0.964, Bootstrap regresion 0.981)
Results QUARTILES CLASSIFICATION BY DXA Q1 Q2 Q3 Q4 Total QUARTILES CLASSIFICATION BY ANTHROPOMETRY Q1 35 7 0 0 42 Q2 7 29 6 0 42 Deurenberg Q3 0 7 34 2 43 Q4 0 0 2 40 42 Total 42 43 42 42 169 Κ= 0,755, p<0,001; Correct classification 81.7%
Results QUARTILES CLASSIFICATION BY DXA Q1 Q2 Q3 Q4 Total QUARTILES CLASSIFICATION BY ANTHROPOMETRY Q1 34 8 0 0 42 Q2 8 29 6 0 42 Siri Q3 0 6 33 3 43 Q4 0 0 3 39 42 Total 42 43 42 42 169 Κ= 0,732, p<0,001; Correct classification 79.9%
Results QUARTILES CLASSIFICATION BY DXA Q1 Q2 Q3 Q4 Total QUARTILES CLASSIFICATION BY ANTHROPOMETRY Q1 33 9 0 0 42 Q2 8 25 9 0 42 Bootstrap regresion Q3 1 8 29 5 43 Q4 0 1 4 37 42 Total 42 43 42 42 169 Κ= 0,645, p<0,001; Correct classification 73.4%
Results: sensibility & specificity to classify in 4th quartile 4th FM quartile from Deurenberg 4th FM quartile 4th FM quartile from equati from Siri on equation Bootstrap regresion % (95 CI min, max) % (95 CI min, max) % (95 CI min, max) Sensibili ty 95.2 (82.6, 99.2) 92.9 (79.5, 98.1) 88.1 (73.6, 95.5) Specifici ty 98.4 (93.6, 99.7) 97.6 (92.7, 99.4) 96.1 (90.6, 98.5) PPV 95.2 (82.6, 99.2) 92.9 (79.5, 98.1) 88.1 (73.6, 95.5) NPV 98.4 (93.6, 99.7) 97.6 (92.7, 99.4) 96.1 (90.6, 98.5)
Results: Bland & Altman plots. Agreement between techniques
Conclusions: The use of skinfolds is consistent and reliable as compared to DXA to assess fat mass The use of our predictive equation may reduce systematic error
Future plans: To perform predictive equations of lean mass and of trunk fat and lean masses
Strengths & Limitations: Possible strength Could the predictive equations obtained in the Spanish subsample applied in the future to the whole sample to predict fat mass? Would the use of the new predictive equations help to find differences in fat mass between feeding groups? Limitation: New predictive equations only useful for 7 y-old children
Thank you!