Prediction equations are commonly used to estimate body fat from anthropometric measurements, but are population specific. We aimed to establish and validate a body composition prediction formula for Asian newborns, and compared the performance of this formula with that of a published equation.
Among 262 neonates (174 from day 0, 88 from days 1–3 post delivery) from a prospective cohort study, body composition was measured using air-displacement plethysmography (PEA POD), with standard anthropometric measurements, including triceps and subscapular skinfolds. Using fat mass measurement by PEA POD as a reference, stepwise linear regression was utilized to develop a prediction equation in a randomly selected subgroup of 62 infants measured on days 1–3, which was then validated in another subgroup of 200 infants measured on days 0–3.
Regression analyses revealed subscapular skinfolds, weight, gender and gestational age were significant predictors of neonatal fat mass, explaining 81.1% of the variance, but not triceps skinfold or ethnicity. By Bland–Altman analyses, our prediction equation revealed a non-significant bias with limits of agreement (LOA) similar to those of a published equation for infants measured on days 1–3 (95% LOA: (−0.25, 0.26) kg vs (−0.23, 0.21) kg) and on day 0 (95% LOA: (−0.19, 0.17) kg vs (−0.17, 0.18) kg). The published equation, however, exhibited a systematic bias in our sample.
Our equation requires only one skinfold site measurement, which can significantly reduce time and effort. It does not require the input of ethnicity and, thus, aid its application to other Asian neonatal populations.
Excess adiposity is a major risk factor for adverse health outcomes and chronic diseases.1 Body fat assessment in infants is important not only as an indicator of nutritional status, but also because of the increasing evidence of its role in the developmental origins of health and disease later in life. Techniques such as dual-energy X-ray absorptiometry (DXA) and quantitative nuclear magnetic resonance have been used in research studies, which provide non-invasive, accurate and precise measurements of body composition, but both techniques are generally unsuitable for large-scale pediatric use. DXA often requires infants to lie still during scanning, thus making the implementation of the technique challenging. In addition, estimates of infant body composition measured by quantitative nuclear magnetic resonance without appropriate mathematical adjustment fares poorly when compared with a four-compartment model, deuterium oxide dilution (D2O) technique and air-displacement plethysmography (ADP) for infants.2 The infant-sized ADP instrument, PEA POD, provided a reliable and accurate assessment of body fat in infants.3, 4, 5 It has been shown to provide better estimates than other techniques, such as DXA,6 and considered as a principal method for body composition assessment in children. The measurements are easy to perform and are validated against the four-compartment model in other studies.7, 8 Ellis et al.7 showed that measurements by PEA POD can be used as a reliable and accurate reference in healthy infants. Their study on 49 healthy infants demonstrated no significant difference in the mean percent body fat (%BF) obtained from PEA POD and the four-compartment model, and the regression between %BF obtained by both did not deviate significantly from the line of identity (R2=0.73, standard error of the estimate=3.7 %BF). In addition, PEA POD takes into account that the hydration status of neonates differs from adults and that hydration of fat-free mass decreases with age,9, 10 unlike other reference methods, such as DXA, which assumes constant hydration. The machine, however, is bulky and expensive, and its use is restricted only to fixed locations, such as hospital settings. Therefore, it would be desirable to have a prediction equation for estimation of total body fat in infants using a combination of anthropometric variables. This would allow for quick estimations of body composition without the need for specialized laboratories or expensive equipment.
Skinfold thickness (SFT) measurements provide estimates of subcutaneous fat layer,11 which can be easily converted into values of %BF or fat mass via prediction formulas.12, 13 SFT measurements are fast and non-invasive bedside methods, which can be performed with high reproducibility, so long as great care is taken with fieldworker training and quality control. SFT measurements for body composition assessment in adolescents and children have been widely used in clinical research and epidemiological settings.14, 15 Studies have shown that SFT measurements from single-site skinfolds were highly correlated with total fat mass in infant subjects.16 In older children, SFT measurements have been shown to correlate with body fat measured by DXA.17 Questions have been raised, however, on their validity in infancy, given the age-related variability in hydration status as well as variability in skinfold compressibility among neonates.18, 19 Generalized skinfold prediction equations, such as those by Slaughter et al.,20 for estimating body composition in infants have been developed, although these equations are population specific,21 and, thus, may not be suitable for Asian infants.
The birth cohort study, ‘Growing Up in Singapore Towards Healthy Outcomes’ (GUSTO), is designed to capture both prenatal and infant growth predictors in a longitudinal fashion in Singapore. GUSTO was designed to test specific hypotheses related to the developmental pathways to obesity and cardiometabolic disorders in Chinese, Malay and Indian children. We collected %BF and fat mass data as measured by PEA POD from a subgroup of GUSTO infant subjects. The aims of this study are to first establish and validate a prediction formula that is specific for the GUSTO cohort during the early postnatal period, using PEA POD measurements as reference, and second, to compare the performance of our prediction equation with that of Slaughter et al.,20 for estimating neonatal fat mass in our cohort.
Materials and Methods
GUSTO birth cohort
The Singapore birth cohort study, GUSTO, is designed to examine developmental origins of obesity and associated disorders, and one of its aims is to determine developmental factors that can predict patterns of growth and body composition in infancy and childhood. Pregnant women (age >18 years) in their first trimester were recruited from the two major public hospitals with obstetric service in Singapore, KK Women’s and Children’s Hospital and National University Hospital. Those on chemotherapy, psychotropic drugs or type I diabetes mellitus were excluded from the study. A total of 1163 women were recruited for the study from June 2009 till September 2010. Informed written consent was obtained from each participant either on the day the study staff first approached the subject, or a few days later after consultation with her spouse.
The offspring were born at National University Hospital and KK Women’s and Children’s Hospital, between November 2009 and May 2011. For our analysis, only healthy, singleton term infants in gestational age (GA) between 37 and 40 weeks were considered. A subgroup of babies born in KK Women’s and Children’s Hospital, whose parents consented, had PEA POD measurements. We excluded data from infants with birth weight below 2.5 kg, or had %BF measured by PEA POD below 5%. After exclusion, a total of 262 infants, from day 0 to 3 after birth, were analyzed in this study.
Weight was recorded to the nearest gram using a calibrated scale (SECA Corp. Hamburg, Germany). For reliability, measurements were taken in duplicate. Abdominal circumferences were recorded to the nearest 0.1 cm, using a non-stretchable measuring tape (SECA 212 Measuring Tape, SECA Corp.). Two skinfolds (triceps and subscapular) were measured in triplicates using Holtain skinfold calipers (Holtain Ltd, Crymych, UK) on the right side of the body, recorded to the nearest 0.2 mm. The following published equation by Slaughter et al.20 was used to estimate body fat from SFT:
where ∑SFT=triceps+subscapular skinfolds
%BF and fat mass was measured using PEA POD (Life Measurement Inc., Concord, CA, USA),22 which measured body volume, and coupled with body weight, was used to calculate body density. %BF can then be calculated from the body density, assuming that the body consists of two components, fat mass and fat-free mass, each with a known density.
For purposes of model derivation, infants whose measurements were taken at day 0 (<24 h after delivery) were excluded; this was based on a recent study that had suggested there was a significant weight loss during the time period of less than 24 h after delivery.23 Hence, in deriving the prediction model for neonatal fat mass, only infants whose measurements were taken from day 1–3 were considered (n=88). The subjects were divided into two groups using the SPSS random number generator. The ‘derivation group’ included two-thirds of the subjects and was used to derive the prediction equation for neonatal fat mass. The ‘validation group’ consisted of the remaining subjects. Stepwise linear regression was utilized to derive the best model to predict neonatal fat mass from the ‘derivation group’. The starting maximum model included all independent variables. It is well-documented that body composition differs significantly between male and female infants;24, 25 thus, infant gender was included in the equation. The dependent variable was fat mass (in kg) as measured by PEA POD. The independent variables used for development of the prediction equation were gender, ethnicity, weight, abdominal circumference, tricep skinfolds, subscapular skinfolds, GA and day of measurement post delivery.
The reliability of the newly derived model and the published equation were assessed on the ‘validation’ group and the subgroup of day 0 infants; differences between measured and predicted values were tested for significance from zero using one-sample paired Student’s t-tests. Technical errors and intraclass correlations were computed to evaluate reproducibility of SFT measurements.26, 27 %BF values predicted from Slaughter’s equation were converted to fat mass using the following equation:
Fat massSlaughter=(%BFSlaughter/100) × weight
Bland and Altman28 analysis was used to compare fat mass prediction from Slaughter’s equation and the newly derived model with measurements obtained from PEA POD, by determining the bias and limits of agreement (LOA). Bias was defined as (Fat Massprediction−Fat MassPEAPOD). LOA was determined by mean bias ±1.96 s.d. to indicate the possible extent of variation between predicted fat mass and PEA POD measurement for any subject. All analysis was performed using IBM SPSS Statistic for Windows Version 19 (Armonk, NY, USA).
Baseline characteristics of the infants at day 0 and days 1–3 are illustrated in Table 1. No significant differences were observed in the anthropometric measurements between derivation and validation groups. The reproducibility of SFT measurements in our study are illustrated in Supplementary Table 1. We noted small mean differences between minimum and maximum measurements for triceps (0.16 mm for boys, 0.15 mm for girls) and subscapular skinfolds (0.15 mm for boys, 0.14 mm for girls). Intraclass correlation for triceps skinfolds was high in both boys (r=0.994) and girls (r=0.997), and the same observation was noted for subscapular skinfolds (r=0.996 for boys, r=0.997 for girls). Technical error for triceps (0.19 mm for boys, 0.15 mm for girls) and subscapular skinfolds (0.15 mm for boys, 0.14 mm for girls) were also quite small in our study, indicating high reproducibility of SFT measurements.
Stepwise linear regression analysis identified GA, weight, subscapular skinfolds and gender to be significant predictors of neonatal fat mass (Table 2), which explained 81.1% of the variance in neonatal fat mass (R2=0.811). Day of measurement post delivery, abdominal circumference, triceps skinfold and ethnicity were not significant predictors of neonatal fat mass. The final GUSTO equation, which was used in subsequent analyses, was:
Fat MassGUSTO=−0.022+(0.307 × weight)−(0.077 × Gender)+(0.028 × subscapular skinfolds)−(0.019 × GA),
where gender=1 for male, 0 for female
Fat mass predicted using the GUSTO equation exhibited a moderately strong correlation with Fat massPEAPOD (r=0.567, P=0.003), which is similar compared with fat mass predicted using Slaughter’s equations (r=0.570, P=0.002).
The accuracy of each prediction equation for the ‘validation group’ was assessed using Bland–Altman plots (Figures 1a and b), which revealed the bias and LOA for body fat predicted by GUSTO and Slaughter’s equation. The mean bias for GUSTO equation is 0.003 kg (P>0.05), similar with the mean bias for Slaughter’s equation, at −0.01 kg (P>0.05). The LOA for GUSTO equation was (−0.25, 0.26) kg, which is also similar with Slaughter’s equation (−0.23, 0.21) kg. There was no significant relationship between the mean and difference of the measured and predicted values, but the relationship approached significance for Slaughter’s equation (r=0.05, P=0.802 for GUSTO equation; r=−0.40, P=0.063 for Slaughter’s equation).
We also assessed GUSTO and Slaughter’s equation with infants who had their measurements taken on day 0, as illustrated in Figures 2a and b. The mean bias for GUSTO equation is −0.01 kg (P>0.05), similar with the mean bias for Slaughter’s equation, at 0.002 kg (P>0.05). The LOA for GUSTO equation was (−0.19, 0.17) kg; again, this is similar with that of Slaughter’s equation (−0.17, 0.18) kg. There was no significant relationship between the mean and difference of the measured and predicted values for the GUSTO equation (r=0.102, P=0.126), but the relationship was significant for Slaughter’s equation (r=−0.252, P=0.001). Again, this is indicative that Slaughter’s equation has a tendency to underestimate fat mass as body fatness increased, and overestimate as body fat decreases.
We have developed a new fat mass prediction equation for neonates and compared it with the equation of Slaughter et al.20 and with measurements obtained by PEA POD. To our knowledge, this study is one of the few that incorporates PEA POD as a reference method to cross-validate fat mass prediction equations. Our study utilized absolute fat mass rather than relative body fat (that is, %BF) as the outcome variable, as early studies have identified absolute fat mass as a more desirable outcome when estimating body composition from anthropometric variables in infants and children.29 Recent studies had also identified poor predictability when attempting to correlate anthropometric measurements with relative body fat.30 SFT measurements have also been identified to be more useful in estimating fat mass rather than relative body fat.16 Consistent with other studies, we found that weight, gender and GA were significant contributors in estimating neonatal fat mass.30, 31 Our study found that subscapular SFT improved the prediction of fat mass. SFT has been widely accepted as a predictor of body fat32 and can be measured directly using well-calibrated callipers. As with all quantitative biological measures, it is important to minimize error. In our study, we observed a high reproducibility of SFT measurements consistent with other studies conducted in children,33 which reflected that our observers were well-versed and trained in SFT measurement.
In our stepwise multiple regression analysis, it was somewhat surprising that triceps skinfold was not a significant predictor, but subscapular skinfold is. This might be due to greater within-subject variability for triceps SFT measurements as observed in our study group. Subcutaneous fat is also known to be unevenly distributed around the circumference of the limbs, which may explain why triceps SFT is not predictive of fat mass. Early studies have also highlighted that subscapular skinfolds are more predictive of fat mass than other single-skinfold sites for infants.16 Given the difficulty in performing skinfold measurements in newborns, and wide inter- and intra-individual variability in such measurements, our equation can significantly reduce time and effort in large birth studies. Ethnicity was also not a significant contributor in predicting neonatal fat mass. The lack of contribution of ethnicity in the prediction of body fat is also consistent with earlier literature on adult Singaporean Chinese, Malays and Indians, which had noted a similar observation.34, 35 This allowed us to derive a prediction model, which does not require input of ethnicity for fat mass prediction, and thus may be helpful in its application to other Asian neonatal populations. We still believe there is significant difference in body composition between Caucasian and Asian babies, but our study appeared to suggest that the difference between Asian babies of different ethnicity is much less.
Our model was derived from infants whose measurements were taken on days 1–3, and excluded infants whose measurements were taken on day 0. This was based on a recent report that had identified significant initial weight loss measured at day 0,23 due to differences in hydration status. It is well-documented that neonates become lighter than they were at birth because of change in hydration status.9, 10, 36 We chose to exclude infants with anthropometric measurements taken on day 0 as the difference in hydration status may influence body fat estimates made by PEA POD, which may in turn impact the ability of our model to estimate neonatal fat mass if day 0 infants were included. We also showed that our model, derived from infants at days 1–3 post delivery, could still estimate neonatal fat mass of infants at day 0 with a small mean bias and no significant systematic bias, unlike Slaughter’s equation (Figures 2a and b). This illustrates the applicability of our model for other newborns in the GUSTO cohort, who did not have PEA POD measurements.
Earlier studies34, 35 revealed that existing prediction formulas for body fat37 were not applicable for Singaporean adults and adolescents, because of the populations in which these equations were developed in, which were mainly Caucasian. Our study revealed that Slaughter’s equation may not be as applicable in Singaporean infant population as evident by the systematic bias exhibited by Slaughter’s equation. This is indicative that prediction formulas for body fatness are population specific, and existing equations may not be entirely useful for multi-ethnic Asian populations. The GUSTO equation had no significant systematic bias when estimating neonatal fat mass, indicating that the equation is in agreement with PEA POD-derived fat mass and a general applicability of the equation to other newborns in the GUSTO cohort, who did not have PEA POD measurements. Given that the GUSTO equation was developed in a largely Asian cohort, it could also be applied to estimate body fat of Asian newborns in other studies.
Our model has its limitations; it was based on healthy, term infants and, thus, not representative of small (that is, preterm, low birth weight) infants. Our study sample had a wide distribution of fat mass, ranging from 0.13 to 0.86 kg with majority of the infants having fat mass ranging from 0.20 to 0.50 kg. Our model estimated neonatal fat mass with a coefficient of determination (R2) of 81.1%, which is slightly lower than R2 values reported in other studies.31, 32 This might be explained due to differences in methodologies (such as availability of SFT sites) used in other studies; our study lacked suprailiac and biceps SFT, which are also surrogate measures of central and peripheral adiposity, respectively.38 The addition of these SFT sites to our model might improve the prediction of neonatal fat mass. In addition, though we demonstrated high technical precision of SFT measurements, our model also showed somewhat broad limits of agreement. While this suggests that error in individual measurements could be large, the small mean bias observed in our equation suggests that our model is suitable for comparisons between groups of infants.
In conclusion, we have developed a new fat mass prediction equation for use in Asian neonates. This equation can be used as a non-invasive method to obtain an in vivo estimate of fat mass in groups of infant subjects, but would be of almost no use in any individual infant. In order to obtain more accurate assessments of body composition of an individual infant, prediction estimates should be followed up with more sophisticated techniques of body composition measurement such as PEA POD. Given that our equations were developed in a largely Asian cohort, it can be extrapolated to estimate body fat of Asian newborns in other studies.
We acknowledge our fellow investigators of the GUSTO study group: Dennis Bier, Arijit Biswas, Cai Shirong, Helen Chan, Jerry Chan, Yiong Huak Chan, Cornelia Chee, Audrey Chia, Chiang Wen Chin, Chng Chai Kiat, Mary Chong, Chong Shang Chee, Chua Mei Chien, Wayne Cutfield, Mary Daniel, Ding Chun Ming, Anne Ferguson-Smith, Eric Andrew Finkelstein, Marielle Fortier, Doris Fok, Anne Goh, Daniel Goh, Joshua J Gooley, Han Wee Meng, Mark Hanson, Mikael Hartman, Michael Heymann, Stephen Hsu Chin-Ying, Hazel Inskip, Jeevesh Kapur, Joanna Holbrook, Lee Bee Wah, BFP Leutscher-Broekman, Lim Sok Bee, Loh Seong Feei, Low Yen Ling, Iliana Magiati, Susan Morton, Krishnamoorthy N, Cheryl Ngo, Pang Wei Wei, Prathiba Agarwal, Qiu Anqi, Quah Boon Long, Victor S Rajadurai, Jen Richmond, Anne Rifkin-Graboi, Allan Sheppard, Lynette Pei-Chi Shek, Borys Shuter, Leher Singh, So Wing Chee, Walter Stunkel, Su Lin Lin, Tan Kok Hian, Tan Soek Hui, Teoh Oon Hoe, Terry Yoke Yin Tong, Hugo Van Bever, Rob Van Dam, Sudhakar Venkatesh, Helena Marieke Verkooijen, Inez By Wong, PC Wong and George SH Yeo. This study is under the Translational Clinical Research (TCR) Flagship Programme on Developmental Pathways to Metabolic Disease, NMRC/TCR/004-NUS/2008, funded by the National Research Foundation (NRF) and administered by the National Medical Research Council (NMRC), Singapore.
About this article
Supplementary Information accompanies this paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)