Accuracy of skinfold and bioelectrical impedance assessments of body fat percentage in ambulatory individuals with cerebral palsy

Transcription

1 DEVELOPMENTAL MEDICINE & CHILD NEUROLOGY ORIGINAL ARTICLE Accuracy of skinfold and bioelectrical impedance assessments of body fat percentage in ambulatory individuals with cerebral palsy DONNA J OEFFINGER 1 MATTHEW J GURKA 2 MICHELLE KUPERMINC 3,4 SAHAR HASSANI 5 NEELEY BUHR 1 CHESTER TYLKOWSKI 1 1 Medical Staff Research, Shriners Hospital for Children, Lexington, KY; 2 Department of Biostatistics, School of Public Health, West Virginia University, Morgantown, WV; 3 University of Virginia, Charlottesville, VA; 4 Naval Medical Center, Portsmouth, VA; 5 Research, Shriners Hospital for Children, Chicago, IL, USA. Corresponding to Donna J Oeffinger, Shriners Hospitals for Children, Lexington, 19 Richmond Road, Lexington, KY 452, USA. doeffinger@shrinenet.org. This article is commented on by Samson-Fang on page 416 of this issue. PUBLICATION DATA Accepted for publication th November 13. Published online 17th December 13. ABBREVIATIONS BIA Bioelectrical impedance analysis CCC Concordance correlation coefficient DXA Dual-energy X-ray absorptiometry AIM This study assessed the accuracy of measurements of body fat percentage in ambulatory individuals with cerebral palsy (CP) from bioelectrical impedance analysis (BIA) and skinfold equations. METHOD One hundred and twenty-eight individuals (65 males, 63 females; mean age 12y, SD 3, range 6 18y) with CP (Gross Motor Function Classification System [GMFCS] levels I (n=6), II (n=46), and III (n=19) participated. Body fat percentage was estimated from (1) BIA using standing height and estimated heights (knee height and tibial length) and (2) triceps and subscapular skinfolds using standard and CP-specific equations. All estimates of body fat percentage were compared with body fat percentage from dual-energy X-ray absorptiometry (DXA) scans. Differences between DXA, BIA, and skinfold body fat percentage were analyzed by comparing mean differences. Agreement was assessed by Bland Altman plots and concordance correlation coefficients (CCC). RESULTS BMI was moderately correlated with DXA (Pearson s r=.53). BIA body fat percentage was significantly different from DXA when using estimated heights (95% confidence intervals [CIs] do not contain ) but not standing height (95% CI 1.9 to.4). CCCs for all BIA comparisons indicated good to excellent agreement (.75.82) with DXA. Body fat percentage from skinfold measurements and CP-specific equations was not significantly different from DXA (mean.8%; SD 5.3%; 95% CI.2 to 1.7) and demonstrated strong agreement with DXA (CCC.86). INTERPRETATION Accurate measures of body fat percentage can be obtained using BIA and two skinfold measurements (CP-specific equations) in ambulatory individuals with CP. These findings should encourage assessments of body fat in clinical and research practices. Accurate assessment of body composition is essential in evaluating the health and nutritional status of children with cerebral palsy (CP). Anthropometric measurements, such as body mass index (BMI), weight for height, and percentage ideal body weight are frequently used to estimate body fat or nutritional status, yet are poor predictors of body fat percentage and have limited usefulness in guiding nutritional intervention. 1,2 Altered growth patterns and body posture in children with CP prevent clinicians from relying on height and weight norms to inform nutritional needs. 3 Difficulty in obtaining accurate height measurements from individuals with CP because of scoliosis or joint contractures 4 can reduce the accuracy of BMI. Additionally, BMI does not allow differentiation between lean and fat mass and has been shown to have only a low to moderate correlation with body fat in individuals with moderate to severe CP. 1 Despite this, BMI continues to be used to assess nutritional status and in weight management. Including an accurate estimate of body fat in nutritional need assessments can help determine the requirement for nutritional intervention; however, there is a lack of research demonstrating safe, non-invasive, cost effective, and valid methods to assess clinically the body composition of children with CP. Dual-energy X-ray absorptiometry (DXA) is often considered a criterion method for assessing body composition because of its accuracy and precision compared with in vivo or in vitro multiple component reference methods. 5 However, owing to cost constraints and availability, DXA assessments are not always feasible to obtain. Simpler, less expensive methods to assess body composition include bioelectric impedance analysis (BIA) and skinfold measurement. Research comparing BIA and skinfold assessments to a criterion standard has been limited and has primarily focused on moderate to severely impaired individuals with CP (Gross Motor Function Classification System [GMFCS] levels III V). 6,7 13 Mac Keith Press DOI:.1111/dmcn

2 BIA uses electrical current to assess total body water and provides a quick, simple estimate of body fat percentage. BIA depends on accurate height measurements, which may compromise its accuracy for individuals with CP. The height of individuals with CP can be predicted from knee height or tibial length using equations developed by Stevenson 4 and Chumlea. 8 Although height estimates based on those equations have been used in previous research, 7,9 their effect on the accuracy of estimates of body fat percentage assessed by BIA in ambulatory individuals with CP has not been assessed. Body fat percentage is frequently estimated using a simple method of two skinfold measurements and equations developed by Slaughter et al. Gurka et al. 11 found the Slaughter equations consistently underestimated body fat in children with CP. They proposed new equations specific to the population with CP, which included a correction factor specific to sex and pubertal status. 11 These equations substantially improved estimates of body fat percentage and were recommended for children with CP, 11 but they have not been externally validated. Neither the Slaughter nor Gurka equations have been assessed exclusively in children with CP with mild to moderate impairment (GMFCS levels I III). Further validation of the Gurka equations, in an independent group of children with CP, is needed before they can be confidently recommended for clinical use. A review of previous studies 6 that examined the accuracy of body fat estimates from skinfold measurements or BIA in children with severe impairment due to CP concluded that small sample sizes, lack of methodological quality, and use of inappropriate analytical methods hampered reliable conclusions. Future trials to investigate these body fat assessment methods in a larger population using Bland Altman limits of agreement analysis were suggested. The current study addresses these concerns and assesses the accuracy of BIA and skinfold equations for estimating body fat percentage in ambulatory individuals with CP, using DXA as a criterion standard. The study aims were (1) to determine the reliability and precision of BIA when using standing height and estimated heights from segmental measurements and (2) to validate the equations of Gurka et al. 11 for estimating body fat percentage in children with CP. METHOD Cross-sectional data from a prospective cohort study were analyzed. Institutional review board approval was obtained by the three participating pediatric orthopedic facilities. Informed consent was obtained from the children s legal guardians, and the children s assent was obtained where possible. Participants Individuals with a diagnosis of CP classified in GMFCS levels I to III, aged 6 to 18 years, were eligible for study inclusion. Children with a pacemaker, baclofen pump, What this paper adds BIA is an accurate assessment of body fat in ambulatory individuals with CP. Use of estimated heights in BIA analysis only slightly reduces the accuracy. CP-specific skinfold equations are validated. CP-specific equations have improved accuracy over the traditional Slaughter equations. Body fat can be accurately assessed using relatively inexpensive portable equipment. metal implant, cast, significant loss or gain of body weight in the past year, any known metabolic disorder or medical condition known to affect growth, or a chance of pregnancy were excluded. Study assessments Data were collected by study coordinators who participated in a 2-day onsite training session. Participants were assessed to determine GMFCS level. The history of the patient was collected from a questionnaire administered to a legal guardian. Anthropometry Standing height and weight were used to calculate BMI. Using a knee height caliper (Weigh and Measure, LLC, Olney, MD, USA), two knee height measurements from the left leg were taken and averaged. Height was estimated from Stevenson 4 and Chumlea et al. 8 equations for knee height. Tibial length was measured using a segmometer (Rosscraft, Vancouver, British Columbia, Canada) and height calculated using Stevenson equations. 8 Stevenson equations are recommended for children of 12 years and under but were applied to the entire population in this study to allow the inclusion of all participants in analyses assessing the influence of estimated heights on BIA accuracy. A single individual at each site collected triceps and subscapular skinfold measurements following Cameron. 12 Holtain calipers were used (Holtain, Crymych, UK) and calibrated before each assessment. Data were recorded to the nearest.2mm. Two measurements that varied by.2mm or less, from the least involved side for those with unilateral involvement or the right side for those with bilateral involvement, were averaged. Body fat was estimated from the equations of Slaughter et al. and Gurka et al. 11 Pubertal status was determined using self-assessment of the Tanner method. 13 To use the equations, Tanner stages 1 and 2 were considered prepubescent, Tanner 3 pubescent, and Tanner 4 and 5 postpubescent. Biolectrical impedance analysis Body fat percentage as estimated using a QuadScan 4 bioelectrical impedance analyzer (Bodystat Ltd, Isle of Man, UK). Data collection followed the recommended protocol and the child regression equation from the Quad- Scan software used. Owing to the asymmetry present in individuals with CP, data were collected twice bilaterally and all trials were averaged to determine an overall body 476 Developmental Medicine & Child Neurology 14, 56:

3 fat percentage. Standing height was entered into the BIA during data collection. The original four trials were reprocessed using the three additional height estimates by staff at Bodystat. Dual energy X-ray absorptiometry Body fat percentage was measured using a total body DXA bone densitometer (Lunar [DPX-IQ software version 4.7, Lunar, Madison, WI, USA] or Hologic Discovery W [QDR APEX software version 2.3.1, Hologic, Boston, MA, USA]). Whole-body body fat percentage without the head was used. Pediatric settings were not used according to the recommendation of the manufacturers. To compare the effect of body fat percentage on the accuracy of the BIA and skinfold measurements, the participants were grouped into one of three weight status categories: low ( % males, 15% females), adequate (% 25% males, 15% 3% females), or excess (>25% males, >3% females) using the DXA body fat percentage. These categories were suggested by Lohman 14 and mirror the work of Kuperminc et al.; 1 they are not suggested as nutritional guidelines for this population. Statistical analyses Statistical analyses were performed using SAS software, version 9.3 (SAS Institute, Cary, NC, USA). Pearson s correlations were calculated between BMI and DXA body fat percentage. To assess body fat percentage agreement between DXA and the aforementioned methods, an initial graphical examination was performed, plotting the individual pairs of estimates and visually assessing the overall proximity to the identity line. Bland Altman 15 figures were created in which differences between the two estimates were plotted against the values of the DXA estimates. To gauge the overall bias and precision of body fat percentage for each method, means and standard deviations of the difference values were computed overall and for various subgroups (sex, GMFCS level, and body fat). Ninety-five percent confidence intervals (95% CIs) were also computed for differences between body fat percentage estimates and DXA, with 95% CIs not containing indicating statistically significant differences at a=.5. Finally, to get an overall statistic quantifying agreement (i.e. proximity to the identity line), the concordance correlation coefficient (CCC) was computed. 16 A value of 1 indicates perfect agreement; values greater than.7 were considered to be in excellent agreement, CCC values between.5 and.7 in moderate agreement, and CCC values less than.5 in poor agreement. RESULTS Participants One hundred and twenty-eight children (65 males, 63 females) with a mean age of 12 years (SD 3y 5mo, range 6 18y) completed DXA assessments. There were more participants in GMFCS levels I and II than level III (I, 63; II, 46; III, 19) owing to greater difficulty in obtaining accurate DXA data with increasing impairment. Feeding difficulties were minimal, with no study participants fed via gastronomy tube. Full characteristics of participants are shown in Table SI (online supporting information). Body mass index BMI calculated using standing height (Table SI) was moderately correlated with DXA (Pearson s r=.53). Body fat percentage DXA mean body fat percentage was 28.% (SD 11.4). Sixty-three individuals (49%) were categorized as having adequate body fat and 6 (47%) as having excess body fat. Five individuals were categorized as having low fat and were excluded from further subgroup analyses. Mean differences between BIA and DXA body fat percentage were calculated for the entire group and by the subgroups of sex, GMFCS level, and weight status category (low, adequate, excess; Table I). A negative number represents an underestimation and a positive number an overestimate of body fat by BIA relative to DXA. Regardless of the method to determine height, BIA body fat percentage differed from DXA on average by less than 2%. Mean differences ranged from underestimating body fat by 1.8% (SD 7.4; 95% CI 3.1 to.5; Stevenson height) to overestimating by 1.7% (SD 6.6%; 95% CI.5 2.9; Chumlea height; Table I). BIA body fat percentage using standing height was not significantly different from DXA (mean.7%; SD 6.1; 95% CI 1.9 to.4). Results using estimated heights were statistically different from DXA based on 95% CI (Table I), but the small differences in magnitude observed were not considered clinically meaningful. The CCC for all BIA comparisons ranged from.75 to.82 (Table I), indicating good to excellent agreement between BIA and DXA. The best agreement was for BIA standing height (CCC.82). Bland Altman graphs showed most differences between BIA and DXA were within two standard deviations (Fig. 1). There is indication of a trend in the nature of these differences. BIA estimates tend more often to underestimate body fat percentage for larger individuals. Sex comparisons BIA accuracy did not significantly differ by sex when standing height was used (neither clinically nor statistically: t-test p=.36) but the CCC was slightly better for males. Findings were inconsistent when estimated heights were used. BIA body fat percentage using height based on knee height (Chumlea) and tibial length was better for females; results for Stevenson-estimated height were better for males (Table I). CCCs were all good to excellent. GMFCS level comparisons BIA accuracy varied by GMFCS level. For GMFCS level I, BIA was best when knee height (Chumlea; mean.; SD 6.7; 95% CI 1.7 to 1.7) and tibial length (mean.8; Assessments of Body Fat in Ambulatory CP Donna J Oeffinger et al. 477

4 Table I: Overall error of predicted body fat percentage for bioelectrical impedance analysis (BIA) using various height estimates Difference a Standing height Chumlea height Stevenson height Tibial height Mean (SD) b 95% CI CCC Mean (SD) b 95% CI CCC Mean (SD) b 95% CI CCC Mean (SD) b 95% CI CCC n Overall (6.1) ( 1.9 to.4) (6.6) (.5 to 2.9) (7.4) ( 3.1 to.5) (7.3) (.2 to 2.8).76 Sex Male 65.2 (6.1) ( 1.8 to 1.4) (6.6) (1. to 4.3) (7.3) ( 3.2 to.4) (7.3) (.7 to 4.4).74 Female (6.3) ( 2.9 to.4).76.8 (6.6) (.9 to 2.5) (7.7) ( 4.2 to.3).65.5 (7.1) ( 1.4 to 2.3).72 GMFCS level I (6.2) ( 3.6 to.5).84. (6.7) ( 1.7 to 1.7) (7.8) ( 6. to 2.1).73.8 (6.7) ( 2.5 to.9).83 II (5.7) (.8 to 2.8) (5.7) (2.1 to 5.6) (5.9) (.8 to 2.8) (6.6) (2. to 6.1).67 III 19.4 (6.5) ( 4. to 4.8) (7.3) ( 1.1 to 6.2).7.8 (7.6) ( 4.6 to 3.) (8.1) (.2 to 7.9).63 Weight status c Adequate (5.1) (.7 to 3.3) (5.7) (3.6 to 6.6) (6.5) (.4 to 3.7) (7.2) (2.7 to 6.4).34 Excess (5.1) ( 5.9 to 3.) (5.2) ( 3.6 to.9) (6.3) ( 7.7 to 4.4) (5.4) ( 3.5 to.6).75 a Individual difference between BIA body fat percentage and dual-energy X-ray absorptiometry (DXA) body fat percentage. b The mean difference provides an estimate of the bias (systematic error) of the equation-predicted body fat percentage, whereas the SD of this difference provides a measure of the precision of the prediction. c Weight status categories based on DXA body fat percentage (only five in total were categorized as low and were not included in the analysis): males: low, <% body fat; adequate, % 25% body fat; excess, >25% body fat; females: low, <15% body fat; adequate, 15% 3% body fat; excess, >3% body fat. CI, confidence interval; CCC, concordance correlation coefficient; GMFCS, Gross Motor Function Classification System. SD 6.7; 95% CI 2.5 to.9) equations were used. Body fat was consistently overestimated using BIA for GMFCS level II. BIA body fat percentage for participants in GMFCS level III was not significantly different from DXA, regardless of the height method (Table I). Weight status category Consistent with Bland Altman graphs (Fig. 1), BIA underestimated fat for those in the excess category (range 2.1% to 4.4%, depending on height measurement used) and overestimated for those in the adequate category (.42% 5.1%, depending on height measurement used) regardless of the height method (Table II). The magnitudes of the underestimation and overestimation are unlikely to affect clinical decision-making. CCC values ranged from.34 to.59 for the adequate group and from.59 to.77 for the excess group. Skinfold measurements: Slaughter and Gurka equations One hundred and eighteen of the 128 participants completed skinfold and DXA measurements. Differences between body fat percentages calculated using skinfold measurements (Slaughter and Gurka equations) and DXA measurements, along with BIA estimates (using standing height) as a reference, are reported in Table II. Consistent with previous findings, 11 the Slaughter equations significantly underestimated (mean difference 7.5%; SD 4.7; 95% CI 8.3 to 6.6) body fat. Bland Altman graphs demonstrated this consistent underestimation with the Slaughter equations, with more substantial underestimation for larger participants (Fig. 2). Improvement with the Gurka equations exhibited by the mean difference estimates and the CCCs in Table II are verified with the Bland Altman figure of the differences as a function of the magnitude of body fat percentage. Body fat estimates using Gurka equations resulted in a.8% mean difference (SD 5.3; 95% CI.2 to 1.7) that was not significantly different from DXA. Body fat percentage from Slaughter equations using skinfold measurements had only moderate agreement with DXA (Slaughter CCC.69). Body fat estimates from Gurka equations were comparable to BIA and had slightly better agreement with DXA (CCC.86) than BIA (CCC.81; Table II). For all subgroups, Gurka equations performed better than Slaughter equations. Because the Slaughter equations and subsequently the Gurka corrected equations were developed for a population from 8 to 18 years old, participants between 6 and 7 years old were grouped and compared with the 8- to 18-year-old group. Gurka equations performed better for the 8- to 18-year-olds; despite this, agreement (CCC) between Gurka and DXA body fat percentage in the 6- to 7-year-old group was no worse. DISCUSSION This study examined the accuracy of BIA and skinfold measurement for assessing body fat in ambulatory individuals with CP. These less expensive, portable methods were 478 Developmental Medicine & Child Neurology 14, 56:

5 (a) Standing height (b) Height estimated by Chumlea equations Mean diff. =.7 Mean diff. = (c) Height estimated by Stevenson equations (d) Height estimated by Tibial length Mean diff. = 1.8 Mean diff. = Figure 1: Bland Altman plots of bioelectrical impedance analysis (BIA) body fat percentage estimates compared with criterion dual-energy X-ray absorptiometry (DXA). (a) Standing height; (b) height estimated by Chumlea equations; (c) height estimated by Stevenson equations; (d) height estimated by tibial length. Points represent the difference between the BIA estimate of body fat percentage and the DXA estimate (y-axis) versus the mean of those two estimates (x-axis). The dashed horizontal lines represent (no difference) and 2SDs of the difference values. Mean differences are presented as well as regression lines of the differences versus the means. BF, body fat; Mean diff., mean difference. Table II: Overall error of predicted body fat percentage for bioelectrical impedance analysis (BIA) comparing skinfold equations and BIA a Difference a Slaughter equations Gurka equations BIA (standing height) n Mean (SD) b 95% CI CCC Mean (SD) b 95% CI CCC Mean (SD) b 95% CI CCC Overall (4.7) ( 8.3 to 6.6).69.8 (5.3) (.2 to 1.7) (6.1) ( 2.2 to.1).81 Sex Male (4.4) ( 7.3 to 5.).77.7 (5.1) ( 2. to.7).87.7 (6.1) ( 2.4 to 1.).82 Female (4.6) ( 9.9 to 7.5) (5.3) (.7 to 3.4) (6.2) ( 3. to.3).73 GMFCS (5.2) ( 8.7 to 6.).71.1 (5.7) ( 1.4 to 1.6) (6.) ( 4. to.8) (4.) ( 8.4 to 5.9).65.4 (5.) ( 1.2 to 2.).82.6 (5.7) ( 1.2 to 2.5) (4.5) ( 11. to 6.5) (3.8) (1.8 to 5.6).87.2 (6.8) ( 4.7 to 5.1).75 Weight status c Adequate (3.4) ( 6.7 to 4.9) (3.9) (1.5 to 3.5) (5.2) (.4 to 3.2).59 Excess (4.2) (.9 to 8.7) (5.1) ( 3. to.3) (5.1) ( 5.8 to 3.).68 Age group 6 7y (4.5) ( 9.7 to 4.7) (4.6) (1.7 to 6.8) (3.4) (1.4 to 5.7) y (3.4) (1.4 to 5.7) (6.1) ( 2.9 to.4) (6.1) ( 2.9 to.4).8 a Individual difference between estimated body fat percentage (Slaughter, Gurka, and BIA) and dual-energy X-ray absorptiometry (DXA) body fat percentage. b The mean difference provides an estimate of the bias (systematic error) of the equation-predicted body fat percentage, whereas the SD of this difference provides a measure of the precision of the prediction. c Weight status categories based on DXA body fat percentage (only five in total were categorized as low and were not included in the analysis): males: low, <% body fat; adequate, 25% body fat; excess, >25% body fat; females: low <15% body fat; adequate=15% 3% body fat; excess, >3% body fat. CI, confidence interval; CCC, concordance correlation coefficient; GMFCS, Gross Motor Function Classification System. Assessments of Body Fat in Ambulatory CP Donna J Oeffinger et al. 479

6 (a) Slaughter %BF - DXA %BF (b) Gurka %BF - DXA %BF Slaughter equations Mean (Slaughter and DXA %BF) Gurka equations Mean (Gurka and DXA %BF) Mean diff. = 7.3 Mean diff. =.9 Figure 2: Bland Altman plots of skinfold body fat percentage estimates compared with criterion dual-energy X-ray absorptiometry (DXA) body fat percentage. (a) Slaughter equations; (b) Gurka equations. Points represent the difference between the skinfold equation estimate of body fat percentage and the DXA estimate (y-axis) versus the mean of those two estimates (x-axis). The dashed horizontal lines represent (no difference) and 2SDs of the difference values. Mean differences are presented as well as regression lines of the differences versus the means. BF, body fat; Mean diff., mean difference. compared with DXA, a criterion standard. The study improved upon previous research by using a larger study population and by examining accuracy within subgroups including GMFCS levels. Assessing the three measurements of body fat during the same visit allowed direct comparison of all methods. BMI had only a moderate correlation with body fat percentage in ambulatory individuals with CP, which is consistent with previous findings. 1 Estimating body fat percentage using BIA or skinfold methods with Gurka equations provides better assessments of body composition than BMI. The time it takes to administer these assessments is minimal and provides more information to the healthcare provider. These findings should encourage clinicians and researchers to supplement BMI data with body fat assessments in their clinical and research practices. Although not assessed in this study, one would expect that weight for height and percentage ideal body weight would have similar issues as BMI, as these weight-for-height measures do not accurately distinguish between adiposity and muscularity. 2 Skinfold body fat estimates using Gurka equations performed well in a new independent group of children with CP that included an expanded age range and a less involved group of individuals than previously studied. 11 These findings further support the recommendation to use the CP-specific Gurka equations for clinical assessments and research. 1,11 To obtain the best accuracy, high-end medical-grade calipers should be used and those administering the test should receive proper training. BIA is an accurate method to assess body fat for individuals with CP when using standing or estimated height. Standing height proved to be the most accurate in the studied population. Height estimate equations were developed from a more involved group of individuals than the participants in the current study. 4,8 Application of the equations in the less involved ambulatory population may have led to increased variability and decreased accuracy of height estimates, decreasing the BIA accuracy slightly for GMFCS levels I and II. There was no significant difference between the BIA and DXA measurements for individuals in GMFCS level III when the estimated heights were used. This suggests that BIA could be used to obtain an accurate estimate of body fat in more involved individuals where standing height is difficult to obtain accurately. These findings should guide clinical and research practices when using BIA to assess body fat in individuals with CP. The findings from the current study differ from those reported by Rieken et al.; 17 however, the population and methods differed significantly. Rieken et al. 17 focused on children with severe neurological impairment and intellectual disability, whereas our study focused on children with milder CP. The study sample of Rieken et al. 17 had a large proportion of children with high body fat percentage, a subgroup of individuals where skinfold equations tend to perform poorly. 11,17 BIA measurements differed between the two studies, as well as the equipment used to collect data and the standard with which these measurements were compared. The present study found better agreement between DXA and skinfold and BIA measurements than those of Liu et al. 7 They reported a moderate correlation (r=.515) between body fat percentage from DXA and BIA using height estimated from knee height. They reported a better correlation between two skinfold measurements and DXA than with BIA. However, comparing two estimates of the same measure warrants the use of agreement statistics, 15 not correlation statistics such as Spearman s rho, as done by Liu et al. 7 Two measures can be strongly correlated, with one measure consistently under- or overestimating the other. The study results are limited to GMFCS levels I to III because DXA data were too difficult to obtain from individuals with more severe impairments. BIA and DXA cannot be used for individuals with metal implants, which limits the versatility of the methods. The use of skinfold measurements 48 Developmental Medicine & Child Neurology 14, 56:

7 with the Gurka equations is an excellent alternative for this population. BIA can be affected by hydration, medications, fluid deficits, and movement artifact. 7 These factors were not controlled for but care was taken to have participants in a typical hydrated state and to limit movement during data collection. Multiple measurements were taken to help ensure consistency of data and eliminate any outliers. Because BIA assessments would typically be done during a clinic visit, it is unlikely that hydration and medications could be controlled; therefore, the data reported are more likely to be reflective of what would be seen in a clinical setting. The generalizability of the study findings may be limited to the Quadscan 4 BIA devices and software used in this study. Although DXA is often considered a criterion standard for body composition, it is important that it has not been validated specifically for the studied population of individuals with mild to moderate CP. Owing to the availability of testing equipment, two different manufacturer brands of DXA were used. This could have introduced additional variability and explain the small difference seen between sites. Further work comparing the two DXA systems is underway. CONCLUSIONS Body composition of ambulatory individuals with CP can be accurately assessed using relatively inexpensive portable equipment such as BIA and two skinfold measurements, and by applying the Gurka equations. Using standing height for ambulatory individuals with CP in the BIA produces accurate body fat results. The use of height estimates from segmental measurements had only a minimal effect on the accuracy of BIA. Therefore, if standing height is too difficult to obtain, one can feel confident in the BIA results when using estimated heights. These findings should encourage clinicians to move beyond anthropometric measurements, such as BMI, weight for height, and percentage ideal body weight, to assess body composition in clinical and research settings. Future work includes assessing the ability of BIA and skinfold measurements (Gurka equations) to detect changes in body fat accurately. The BIA data will be further analyzed to determine if correction factors could improve the accuracy for those with excess body fat. ACKNOWLEDGEMENTS We acknowledge funding by Kosair Charities, Inc., grant number 7BH. This agency had no involvement in the study design, data collection, data analysis, manuscript preparation, and/or publication decisions. We also thank the research coordinators and staff at all participating facilities (Shriners Hospitals for Children, Lexington and Chicago, and University of Virginia) for their support and roles in data collection, and the participants and their families for their time and participation. The authors have stated that they had no interests that might be perceived as posing a conflict or bias. SUPPORTING INFORMATION The following additional information can be found online: Table SI: Characteristics of participants. REFERENCES 1. Kuperminc MN, Gurka MJ, Bennis JA, et al. Anthropometric measures: poor predictors of body fat in children with moderate to severe cerebral palsy. Dev Med Child Neurol ; 52: Zemel BS, Riley EM, Stallings VA. Evaluation of methodology for nutritional assessment in children: anthropometry, body composition, and energy expenditure. Annu Rev Nutr 1997; 17: Samson-Fang L, Stevenson RD. Linear growth velocity in children with cerebral palsy. Dev Med Child Neurol 1998; 4: Stevenson RD. Use of segmental measures to estimate stature in children with cerebral palsy. Arch Pediatr Adolesc Med 1995; 149: Erselcan T, Candan F, Saruhan S, Ayca T. Comparison of body composition analysis methods in clinical routine. Ann Nutr Metab ; 44: Rieken R, Calis EA, Tibboel D, Evenhuis HM, Penning C. Validation of skinfold measurements and bioelectrical impedance analysis in children with severe cerebral palsy: a review. Clin Nutr ; 29: Liu LF, Roberts R, Moyer-Mileur L, Samson-Fang L. Determination of body composition in children with cerebral palsy: bioelectrical impedance analysis and anthropometry vs dual-energy x-ray absorptiometry. J Am Diet Assoc 5; 5: Chumlea WC, Guo SS, Steinbaugh ML. Prediction of stature from knee height for black and white adults and children with application to mobility-impaired or handicapped persons. J Am Diet Assoc 1994; 94: , Tomoum HY, Badawy NB, Hassan NE, Alian KM. Anthropometry and body composition analysis in children with cerebral palsy. Clin Nutr ; 29: Slaughter MH, Lohman TG, Boileau RA, et al. Skinfold equations for estimation of body fatness in children and youth. Hum Biol 1988; 6: Gurka MJ, Kuperminc MN, Busby MG, et al. Assessment and correction of skinfold thickness equations in estimating body fat in children with cerebral palsy. Dev Med Child Neurol ; 52: e Cameron N. The Methods of Auxological Anthropology. New York: Plenum Press, Tanner J. Growth at Adolescence. London: Blackwell Science, Lohman TG. The use of skinfold to estimate body fatness in children and youth. J Phys Educ Rec 1987; 58: Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics 1989; 45: Rieken R, van Goudoever JB, Schierbeek H, et al. Measuring body composition and energy expenditure in children with severe neurologic impairment and intellectual disability. Am J Clin Nutr 11; 94: Assessments of Body Fat in Ambulatory CP Donna J Oeffinger et al. 481