Skip to main content

Probabilistic prediction of segmental body composition in Iranian children and adolescents

Abstract

Background

Adolescents' body composition is considered an important measure to evaluate health status. An examination of any of the segmental compartments by anthropometric indices is a more usable method than direct methods.

Objectives

To propose a method based on the network approach for predicting segmental body composition components in adolescent boys and girls using anthropometric measurements.

Methods

A dual-energy X-ray absorptiometry (DXA) dataset in the south of Iran, including 476 adolescents (235 girls and 241 boys) with a range of 9–18 years, was obtained. Several anthropometric prediction models based on the network approach were fitted to the training dataset (TRD 80%) using bnlearn, an R add-in package. The best fitted models were applied to the validation dataset (VAD 20%) to assess the prediction accuracy.

Results

Present equations consisting of age, weight, height, body mass index (BMI), and hip circumference accounted for 0.85 (P < 0.001) of the variability of DXA values in the corresponding age groups of boys. Similarly, reasonable estimates of DXA values could be obtained from age, weight, height, and BMI in girls over 13 years, and from age, weight, height, BMI, and waist circumference in girls under 13 years, respectively, of 0.77 and 0.83 (P < 0.001). Correlations between robust Gaussian Bayesian network (RGBN) predictions and DXA measurements were highly significant, averaging 0.87 for boys and 0.82 for girls (P < 0.001).

Conclusions

The results revealed that, based on the present study’s predictive models, adolescents' body composition might be estimated by input anthropometric information. Given the flexibility and modeling of the present method to test different motivated hypotheses, its application to body compositional data is highly appealing.

Peer Review reports

Introduction

Obesity and overweight in adolescents are serious risk factors for the development of obesity in adulthood. Several clinical studies have discovered strong links between segmental fat mass and health risk, particularly the relationship between trunk fat mass and insulin resistance, dyslipidemia [1], and cardiovascular risk [2] in adolescents, as well as between lean mass and cardio metabolic risk [3]. Adolescence is also a critical period in human life for laying down bone minerals, which increase exponentially in both sexes at this time [4]. Human segmental body composition, which is the distribution of the three parts of body weight (bone, fat, and lean) over the trunk, legs, and arms, is a way to describe what the body is made of and is considered an important measure to evaluate the health status of the human body [5]. Thus, an examination of any of the segmental compartments yields interesting results. Direct body composition measurement methods such as underwater weighing (UWW) and dual-energy X-ray absorptiometry (DXA), though they are accurate and widely acceptable as gold standards in epidemiological studies, are impractical, time-consuming, and expensive in large studies [6, 7]. Moreover, there are physical restrictions on body weight, length, thickness, and width, as well as the type of DXA device. Most obese adults and many obese children are often too wide and too heavy to receive a full-body DXA scan [8]. In clinical and epidemiological studies, simple and practical techniques such as anthropometric-based measurements, which include waist circumference, body mass index (BMI), and skinfold thickness, are the cheapest and most commonly used methods for the estimation of body composition [9]. However, there are some significant drawbacks to each, for instance, BMI tends to overestimate body fat levels in individuals with high muscle mass [10]. For more details, refer to Chung et al. [11]. Statistical methods that use predictive algorithms for assessing body composition have also been developed, which are thought to be more applicable in large studies in non-laboratory settings [5]. Numerous studies have been conducted throughout the world on this topic [5, 12,13,14,15,16,17,18,19,20,21,22]. The literature review revealed that multivariate regression models [12,13,14,15,16], receiver operating characteristic (ROC) curves [17,18,19], neural networks [21], and Bayesian networks (BN) [5, 22] are among the methods used to predict body composition. In a study to predict body composition, Bayesian networks performed better than linear regression models, and predicted distributions based on this method were close to their published counterparts [22]. In another study, it was suggested that, under certain anthropometric constraints, BN predictions might be used to provide a complementary body composition analysis for large populations [22]. Because of BN’s ability to deal with parametric dimension reduction and uncertainty, it forms the basis of modern methods for assessing many real-world problems, including medical diagnosis and prediction. In the present study, our goal is to predict body compositional measures in adolescent boys and girls in the second-largest country in the Middle East, Iran, using several types of anthropometric data; age, weight, height, waist circumference, hip circumference, and BMI; using a robust Gaussian Bayesian network (RGBN). To our knowledge, until now, no prediction model has been proposed for simultaneously assessing the body composition compartment in the Iranian adolescent population, this research gap led to the present study.

Methods

The present cross-sectional study was carried out in Kavar, an urban community east of Shiraz, the capital of Fars province in southern Iran, during 2012–2013 [23].

Subjects

The participants were girls and boys aged 9–18 years, inhabitants of Kawar, and who were attending elementary to secondary school. An age-stratified systematic random sample of 7.5% was applied, and 500 participants (250 girls and 250 boys) were selected for this study. At each school and for each age group, an alphabetized list of student names was used and a random starting point was picked. Then, every fifth student was selected to take a survey. From them, each participant with a major chronic disorder was excluded, and eventually 476 participants (235 girls and 241 boys) who completed the study were included in the statistical analysis. The participants were divided according to the average age of puberty. As children go through puberty, there are significant effects on body composition. Girls acquire a greater amount of fat mass, and boys gain significantly more fat-free mass, and both genders gain bone mineral density at the highest rate [24, 25]. The cut-off age for girls is ≥ 13 years (88 participants) and for boys it is ≥ 14 years (136 participants). The flowchart of recruitment and research is shown in Fig. 1.

Fig. 1
figure 1

The flowchart of recruitment and research

Anthropometric measures

Weight and height were measured by a training nurse, weight using the standard scale closest to 0.1 kg (Seca, Germany) and height at the nearest 0.5 cm using a wall-mounted meter with the participant wearing light clothing and no shoes. BMI was calculated according to this formula: BMI (kg/m2) = weight (kg) / (height (m))2. Also, we measured the waist circumference to the nearest 0.1 cm just above the patient’s uppermost lateral border of the right ilium.

Body composition measures

The Hologic system (Discovery QDR, USA) was used to measure BMC (in gram), bone area (in square centimeter), BMD (in gram per square centimeter), lean mass, and fat mass (in gram). Bone mineral density was measured in the total body (without the head), lumbar spine, and left femoral neck by a single experienced technician. Lean mass and fat mass were measured in the total body as, well as in specific body compartments, such as the trunk, legs, and arms. Densitometric studies were done with the participant wearing special clothing and no footwear. To eliminate physiological lumbar lordosis during measurement of the lumbar spine, we elevated the participants’ knees while they were in a supine position. In accordance with international standards, all measurements of the femur were done on the left femur at the position of internal rotation. During DXA measurements, the position, size, and location of the region of interest were the same for all participants. Scanner stability was checked throughout the course of the study with plots of daily spine phantom scans. Based on preliminary measurements in ten children, the coefficients of variation in our laboratory were 1% for the total body, 0.51% for the lumbar spine, and 2.4% for the femoral neck [23].

Statistical analysis

The statistical method that we used in the present study was based on BNs [26,27,28] which is a graph-based method that represents complex probabilistic interactions and causality relationships between variables [29]. In plain language, RGBN was used to predict segmental body composition compartments by simple anthropometric indices according to the method provided by Kenett et al. [30]. According to the cut-off age, there are 4 groups under study: girls over 13 years (147 participants); girls under 13 years (88 participants); boys over 14 years (136 participants); and boys under 14 years (105 participants). Each group was divided into a training dataset (TRD: 80%) and a validation dataset (VAD: 20%). For each group, different subsets of predictor variables were considered to construct the Gaussian Bayesian network (GBN) [31, 32], which was based on the assumption that the variables followed a normal distribution. To learn the network structure related to each subset of predictor variables, 12 structure-learning algorithms (4 constraint-based algorithms [33], two score-based algorithms [34] and two hybrid algorithms [35] with different parameters) that limited the network arcs according to prior knowledge were used. To run the model, an add-in package, “bnlearn in R [36] was used. After identifying the network structure with the stated algorithms, to create an RGBN from all possible edges between the predictor variables and the response variables, only the edges that appeared in at least seven algorithms were present in the RGBN. The Bootstrap method was used to evaluate the robustness of the created structures. In this way, by generating 500 random subsets of data related to each group and each with 500 observations, the occurrence ratio of each edge in the bootstrap iterations was measured. Those that did not have the minimum occurrence rate (70%) were removed from the network. The quality of RGBNs made for each subset of predictors variables was evaluated with four criteria:\(\left|Bias\right|={\sum }_{v=1}^{p}{\sum }_{i=1}^{n}{B}_{{i}^{v}}\),\(SD={\sum }_{v=1}^{p}{\sigma }^{v}\),\(SEP= {\sum }_{v=1}^{p}({\sum }_{i=1}^{n}\sqrt{{\left|{B}_{{i}^{v}}\right|}^{2}+{({\sigma }^{v})}^{2}})\), and the number of edges. Following the selection of the RGBNs, the linear model corresponding to each response variable was extracted from the networks. Prediction of body composition should be done hierarchically. The variable that was the offspringFootnote 1 node of anthropometric indices was predicted first, and then the offspring of this variable were estimated using the predicted values. Thus, all body composition variables were predicted. The normal distribution of data was checked using the Kolmogorov–Smirnov test and confirmed by a visual inspection of the histograms and Q-Q plots. Skewed variables were log transformed. Student t-tests were used to compare sex differences in all variables' means. Results with a p-value < 0.05 were considered statistically significant. The statistical power of the study was assessed by the "pwr" package in R [37] and for all of the analyses, it was greater than 0.95.

Results

Except for weight and body mass index, all the anthropometric variables were significantly different for girls compared with boys (Table 1). In addition, for DXA body composition compartments, except for the trunk BMC, all the variables differed significantly between boys and girls.

Table 1 Summary statistics for anthropometric indices and body composition variables, including mean, skewness, standard deviation (SD), and quartiles, as well as their units, are presented by gender

Among all the fitted models, including the multivariate regression model with all mentioned anthropometric indices and high parametric dimension, we selected those models that had a small number of parameters with the lowest SEP in the validation data set; the best models were as follows: for boys over 14 years with bias = 4.09, SD = 4.46 and SEP = 6.38; for boys under 14 years with bias = 3.29, SD = 4.63 and SEP = 6.02; for girls over 13 years with bias = 4.02, SD = 5.66 and SEP = 7.35; for girls under 13 years old with bias = 4.95, SD = 5.14 and SEP = 7.76 (Table 2). All direct connections in the final structure of RGBNs were significant (P < 0.001) (Fig. 2).

Table 2 Some RGBNs we explored by applying RGBNs to all groups under study
Fig. 2
figure 2

Top: Structures of the chosen RGBNs for boys over 14 years (left) and boys under 14 (right). Bottom: Structures of the chosen RGBNs for girls over 13 years (left) and girls under 13 years (right). W: weight, H: height, WC: waist circumference, HC: hip circumference, TF: trunk fat mass, TL: trunk lean mass, TB: trunk BMC, LF: leg fat mass, LL: leg lean mass, LB: leg BMC, AF: arm fat mass, AL: arm lean mass, AB: arm BMC

All estimated regression models with standardized coefficients were extracted from RGBNs by considering the parents of each node as predictors of these linear models. In addition, the proportion of variance explained by these models is given for each variable. Predictive models were able to describe the variations in each response variable, which averaged 0.85 for all groups (Table 3).

Table 3 Prediction models and proportion of variance explained for each of the nine response variables by the respective parents within the chosen RGBNs

The validation scores for the prediction models as derived from the RGBN method. For the prediction of body composition variables, the maximum SEP value was 1.28 kg for both boys and girls. The corresponding values varied from 0.65 to 0.93 for both sexes. According to the MAPE score, it was possible to estimate the segmental fat-free mass components (lean and BMC) with a maximum difference of ± 9.48% and ± 14.97% of DXA values for boys and girls, respectively. Segmental fat mass could be estimated with less accuracy than other components, with a maximum difference of ± 19.43% and ± 18.97% of DXA values for boys and girls, respectively (Table 4). Segmental body compositions were well predicted overall, even if some bias arose in particular areas, especially for fat masses. Generally, RGBN’s prediction had a fairly strong correlation with DXA measurements (Fig. 3).

Table 4 Quality of fit between nine segmental compartments as derived by DXA with their counterparts predicted by the RGBN
Fig. 3
figure 3

Scatter plots of RGBN prediction model for nine segmental compartments in relation to their DXA observation counterparts

The validity of RGBN performance to predict total fat mass and fat-free mass (lean and bone mass) is unknown. SEP values of less than 1.28 kg were found for boys and girls. The corresponding values were significantly high for all boys and girls. Fat mass could be estimated with a maximum difference of ± 9.98% of actual values. Additionally, the fat-free mass prediction had more accuracy than fat mass, with a maximum difference of ± 3.56% of actual values for boys and girls (Table 5). The predicted values of fat and fat-free mass versus the observed values, all of the correlations were significant (Fig. 4) (P < 0.001).

Table 5 Quality of fit between fat free mass and fat mass as derived by DXA with their counterparts predicted by the RGBN
Fig. 4
figure 4

Quality of fit between fat-free mass and fat mass as derived by DXA with their counterparts predicted by the RGBN

Discussion

The importance of estimation of segmental body composition and finding interrelationships between them has been recognized by specialists and experts [38]. In the present cross-sectional study of Iranian children and adolescents, the segmental body composition components are estimated by using easy-to-measure anthropometric indices such as age, weight, height, BMI, arm circumference, and hip circumference, which have a high correlation with body composition components.

The trunk lean mass has been mentioned in the literature as an important measurement for overcoming musculoskeletal diseases [39, 40], as well as a critical supporting component of various spinal pathologies [41, 42]. Therefore, its evaluation in adolescence can improve the management and prevention of such diseases. To the best of our knowledge, the literature has focused more on predicting adolescents’ total fat-free mass [43] and anthropometric prediction equations were not available for predicting adolescents’ trunk lean mass. Accordingly, in the current study, this network method enables us to do this. Our findings revealed that the trunk lean mass had a highly significant association with boys’ height, weight, and hip circumference. Nevertheless, in girls, it was only significantly related to height and weight. Due to its causative role in several body composition components, the trunk lean mass is an essential variable in the hierarchical prediction approach, and some other segments’ estimation depends on its prediction.

Assessing appendicular (legs and arms) lean mass during early growth is especially important because of its association with health issues [44,45,46,47,48,49]. Maintaining optimal appendicular lean mass in adolescents may improve peak lean mass and bone strength, as well as have beneficial effects on adult cardiovascular health [50]. Moreover, low lean mass is associated with metabolic risk, and muscular strength is positively related to higher insulin sensitivity in children [51]. In our results, leg lean mass was associated with weight and height in boys, but in the age group under 14 years, it also had a significant relationship with hip circumference. In girls, leg lean mass was associated just with weight. The arm lean mass had a significant relationship with age in boys over 14 years old and with weight in girls over 13 years old. A cross-sectional study on adolescents found a high correlation with appendicular lean mass and weight, height, and age in both sexes and no mean differences between predicted and measured values [52].

A significant association was found between trunk fat mass and BMI and height in boys; in addition to BMI, this variable also had a significant relationship with weight in girls over 13 years of age and waist circumference in girls under 13 years of age. The literature review showed that BMI and waist circumference had a significant role in predicting fat mass [12, 15, 16]. A cross-sectional study in Brazilian adolescents reported that body circumferences, including waist circumference, had high adjusted coefficients of determination to estimate fat mass in girls [15]. Another study in schoolchildren and adolescents found that BMI and waist circumference are among the best predictors of body fat percentage [16].

About appendicular fat mass, leg fat mass was associated with age in boys over 14 years old and with height in girls over 13 years old. A significant relationship was found between leg fat mass and weight in both sexes in the younger age group. Also, arm fat mass had a significant association just with height in boys and girls. In both boys and girls, absolute and adjusted values of lean mass for body size increased with age, although to a higher amount in males. In teenagers, there was also a difference in body composition between the sexes [25]. This pattern of lean mass quantity and sexual dimorphism throughout adolescence is consistent with earlier results across ethnic groups and can be explained by hormonal effects. During pubertal growth spurts, growth hormone and sex steroids are dramatically activated, resulting in significant gains in skeletal muscle and, as a result, lean mass [53].

Regarding the relationship between body composition measurements, there was no significant association between segmental fat mass and BMC conditionally by segmental lean mass in any group. According to a research on Iranian children and adolescents, total body BMD and total body lean mass had the most correlation, whereas total body fat percentage had the least, and BMD distribution was affected by age and total body lean mass, which were also independent factors [54]. In another study on adolescents and young adults, whereas Pearson correlations between DXA measures of fat mass and bone parameters (BMC and BMD) were generally positive, the result showed that after accounting for lean mass and some other indices, fat mass did not correlate with DXA values for bone [55].

Based on these estimated body composition measurements, several indicators that are linked to human health concerns could be extracted, such as the body fat to lean mass ratio, which is a valuable metric for predicting and managing cardiometabolic events in the early stages [56]; the proportion of the trunk to appendicular fat mass increase is associated with an increase in blood pressure [57]; appendicular lean mass is an index to define sarcopenia, which is one component of malnutrition [45]; elevated values of this index may increase sensitivity to insulin [58] and development in bone health [46], in opposition, low values relate to metabolic risk factors and resistance to insulin [47, 48, 59], and including the risk of osteoporosis [49].

According to our findings, RGBN was an excellent way to predict segmental body composition compared to the multivariate regression model due to the considerable reduction in the parametric dimension and acceptable accuracy. Predictions using the present equations to estimate body composition components were reliable, given high coefficients of determination, which averaged 85% for boys and 79% for girls. The high correlation of body composition components from RGBN and DXA values (averaged R2: 0.87 for boys and 0.82 for girls) supports using RGBN equations to estimate body composition in children and adolescents.

These results are consistent with other studies based on the BN method that have indicated that anthropometric variables are good predictors of body components. L. Mioche et al. [22] used a non-parametric equation derived from a BN including sex, age, weight, and height to predict the fat-free mass. Their results indicated that correlations between BN predictions and DXA measurements are significant for fat-free mass. Furthermore, they confirmed that BN predictions are more accurate for fat-free mass and fat mass than those obtained from linear models. L. Mioche et al. [22] in another study assessed the robustness of the BN predictions and determined that predicted fat-free mass distributions are close to their published counterparts for both sexes between 20 and 79 years old; they suggested that, under certain anthropometric constraints, BN predictions might be used as a complementary body composition analysis for large populations.

An essential feature of our proposed model is that it can simultaneously predict several segmental compartments to better assess individuals’ health status or metabolic risks. To our knowledge, this is the first proposal for a network approach for the Iranian adolescent population. Although the direct method (DXA) and the estimation approach, including regression equations, Bayesian networks, or other appropriate methods based on anthropometric variables, in most cases show slight differences, the present equations are a safe alternative for implementation in non-clinical settings where there are no complex laboratories and equipment for evaluating body composition.

Limitation

This study was limited in some aspects. The main limitation was its cross-sectional design. It may not be possible to generalize these findings to other populations. Furthermore, the results of the present study are based solely on a statistical model, and future prospective studies are required to accredit our observations. Biological plausibility has not been the objective of the present study and requires comprehensive discussion in another study.

Conclusion

Based on the present study’s predictive models, adolescents' body composition might be estimated by using anthropometric measurements for a large population in non-clinical settings. Additionally, a few significant health risk factors that are involved with body composition components can be identified, such as excess fat mass, peak bone mass, and inadequate lean body mass of an adolescent. Given the flexibility and modeling of the present method to test different motivated hypotheses, its application to body compositional data and screening large populations is highly appealing.

Availability of data and materials

The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.

Notes

  1. Since the present study was conducted on children, to avoid ambiguity, the word "offspring" was used instead of "child". In the context of Bayesian networks, child is a commonly-used word which means a node that is directly influenced by another node (called the parent).

Abbreviations

DXA:

Dual-energy x-ray absorptiometry

BMI:

Body mass index

UWW:

Underwater weighing

ROC:

Receiver operating characteristic

BN:

Bayesian networks

RGBN:

Robust gaussian bayesian network

BMC:

Bone mineral content

BMD:

Bone mineral density

TRD:

Training dataset

GBN:

Gaussian Bayesian network

SEP:

Standard error prediction

MAPE:

Mean absolute percentage error

References

  1. Niederauer CM, Binkley TL, Specker BL. Effect of truncal adiposity on plasma lipid and lipoprotein concentrations. J Nutr Health Aging. 2006;10(2):154–60 (PMID: 16554953).

    CAS  PubMed  Google Scholar 

  2. Berends A, Zillikens M, De Groot C, Rivadeneira F, Oostra B, Van Duijn C, et al. Body composition by dual-energy X-ray absorptiometry in women with previous pre-eclampsia or small-for-gestational-age offspring. BJOG. 2009;116(3):442–51. https://doi.org/10.1111/j.1471-0528.2008.02044.x (PMID: 19187378).

    Article  CAS  PubMed  Google Scholar 

  3. Burrows R, Correa-Burrows P, Reyes M, Blanco E, Albala C, Gahagan S. Low muscle mass is associated with cardiometabolic risk regardless of nutritional status in adolescents: A cross-sectional study in a Chilean birth cohort. Pediatr Diabetes. 2017;18(8):895–902. https://doi.org/10.1111/pedi.12505 (PMID: 28145023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  4. Jeddi M, Ardalan A, Heydari ST, Dabbaghmanesh MH. Non-linear association of body composition and its components with bone density in Iranian children and adolescents. Arch Osteoporos. 2021;16(1):77. https://doi.org/10.1007/s11657-021-00920-x (PMID: 33948735).

    Article  PubMed  Google Scholar 

  5. Tian S. Body composition prediction by locally weighted and Bayesian networks modeling: AgroParisTech; 2013.

  6. Cicek B, Ozturk A, Unalan D, Bayat M, Mazicioglu MM, Kurtoglu S. Four-site skinfolds and body fat percentage references in 6-to-17-year old Turkish children and adolescents. J Pak Med Assoc. 2014;64(10):1154–61 (PMID: 25823156).

    PubMed  Google Scholar 

  7. Hussain Z, Jafar T, uzZamanParveenSaeed MRF. Correlations of skin fold thickness and validation of prediction equations using DEXA as the gold standard for estimation of body fat composition in Pakistani children. BMJ Open. 2014;4(4):e004194. https://doi.org/10.1136/bmjopen-2013-004194 (PMID: 24755209).

    Article  PubMed  PubMed Central  Google Scholar 

  8. Tataranni PA, Ravussin E. Use of dual-energy X-ray absorptiometry in obese individuals. Am J Clin Nutr. 1995;62(4):730–4. https://doi.org/10.1093/ajcn/62.4.730 (PMID: 7572700).

    Article  CAS  PubMed  Google Scholar 

  9. Wells JC, Fewtrell MS. Measuring body composition. Arch Dis Child. 2006;91(7):612–7. https://doi.org/10.1136/adc.2005.085522 (PMID: 16790722).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Arumäe K, Mõttus R, Vainik U. Beyond BMI: Personality Traits’ Associations With Adiposity and Metabolic Rate. Physiology & behavior. 2022:113703.

  11. Chung S. Body composition analysis and references in children: clinical usefulness and limitations. Eur J Clin Nutr. 2019;73(2):236–42.

    Article  Google Scholar 

  12. Salamat MR, Shanei A, Salamat AH, Khoshhali M, Asgari M. Anthropometric predictive equations for estimating body composition. Adv Biomed Res. 2015;4:34. https://doi.org/10.4103/2277-9175.150429. PMID: 25709999.

  13. Martarelli D, Martarelli B, Pompei P. Body composition obtained from the body mass index. Eur J Nutr. 2008;47(8):409.

    Article  Google Scholar 

  14. Ripka WL, Ulbricht L, Gewehr PM. Body composition and prediction equations using skinfold thickness for body fat percentage in Southern Brazilian adolescents. PLoS ONE. 2017;12(9): e0184854. https://doi.org/10.1371/journal.pone.0184854 (PMID: 28910398).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Lyra CO, Lima SCVC, Lima KC, Arrais RF, Pedrosa LFC. Prediction equations for fat and fat-free body mass in adolescents, based on body circumferences. Ann Hum Biol. 2012;39(4):275–80. https://doi.org/10.3109/03014460.2012.685106 (PMID: 22594692).

    Article  PubMed  Google Scholar 

  16. Ortiz-Hernández L, López A, Ramos-Ibáñez N, Lara L, Gómez R, Pérez-Salgado D. Equations based on anthropometry to predict body fat measured by absorptiometry in schoolchildren and adolescents. J Pediatr (Rio J). 2017;93(4):365–73. https://doi.org/10.1016/j.jped.2016.08.008 (PMID: 28132762).

    Article  Google Scholar 

  17. Kelishadi R, Djalalinia S, Motlagh ME, Rahimi A, Bahreynian M, Arefirad T, et al. Association of neck circumference with general and abdominal obesity in children and adolescents: the weight disorders survey of the CASPIAN-IV study. BMJ Open. 2016;6(9): e011794. https://doi.org/10.1136/bmjopen-2016-011794 (PMID: 27694487).

    Article  PubMed  PubMed Central  Google Scholar 

  18. Ejtahed HS, Kelishadi R, Qorbani M, Motlagh ME, Hasani-Ranjbar S, Angoorani P, et al. Association of neck circumference with general and abdominal obesity in children and adolescents: the weight disorders survey of the CASPIAN-IV study. BMJ Open. 2016;6(9): e011794. https://doi.org/10.1136/bmjopen-2016-011794 (PMID: 27694487).

    Article  Google Scholar 

  19. Ehrampoush E, Arasteh P, Homayounfar R, Cheraghpour M, Alipour M, Naghizadeh MM, et al. New anthropometric indices or old ones: Which is the better predictor of body fat? Diabetes Metab Syndr. 2017;11(4):257–63. https://doi.org/10.1016/j.dsx.2016.08.027 (PMID: 27578617).

    Article  PubMed  Google Scholar 

  20. Kupusinac A, Stokić E, Doroslovački R. Predicting body fat percentage based on gender, age and BMI by using artificial neural networks. Comput Methods Programs Biomed. 2014;113(2):610–9. https://doi.org/10.1016/j.cmpb.2013.10.013 (PMID: 24275480).

    Article  PubMed  Google Scholar 

  21. Ferenci T, Kovacs L. Predicting body fat percentage from anthropometric and laboratory measurements using artificial neural networks. Appl Soft Comput. 2018;67:834–9. https://doi.org/10.1016/j.asoc.2017.05.063.

    Article  Google Scholar 

  22. Mioche L, Bidot C, Denis J-B. Body composition predicted with a Bayesian network from simple variables. Br J Nutr. 2011;105(8):1265–71. https://doi.org/10.1017/S0007114510004848 (PMID: 21144103).

    Article  PubMed  Google Scholar 

  23. Jeddi M, Roosta MJ, Dabbaghmanesh MH, Omrani GR, Ayatollahi SMT, Bagheri Z, et al. Normative data and percentile curves of bone mineral density in healthy Iranian children aged 9–18 years. Arch Osteoporos. 2013;8:114. https://doi.org/10.1007/s11657-012-0114-z (PMID: 23297104).

    Article  PubMed  Google Scholar 

  24. Loomba-Albrecht LA, Styne DM. Effect of puberty on body composition. Curr Opin Endocrinol Diabetes Obes. 2009;16(1):10–5. https://doi.org/10.1097/med.0b013e328320d54c (PMID: 19115520).

    Article  CAS  PubMed  Google Scholar 

  25. Ripka WL, Orsso CE, Haqq AM, Luz TG, Prado CM, Ulbricht L. Lean mass reference curves in adolescents using dual-energy x-ray absorptiometry (DXA). PLoS ONE. 2020;15(2): e0228646. https://doi.org/10.1371/journal.pone.0228646 (PMID: 32027713).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  26. Kim JH, Pearl J. Convince: A conversational inference consolidation engine. IEEE Trans Syst Man Cybern. 1987;17(2):120–32. https://doi.org/10.1109/TSMC.1987.4309025.

    Article  Google Scholar 

  27. Lauritzen SL, Spiegelhalter DJ. Local computations with probabilities on graphical structures and their application to expert systems. J R Stat Soc Series B Stat Methodol. 1988;50(2):157–94. https://doi.org/10.1111/j.2517-6161.1988.tb01721.x.

    Article  Google Scholar 

  28. Jensen FV. An introduction to Bayesian networks. London: UCL press; 1996.

  29. Korb KB, Nicholson AE. Bayesian artificial intelligence: CRC press; 2010.

  30. Kenett RS, Salini S. Modern analysis of customer surveys: with applications using R: John Wiley & Sons; 2011.

  31. Geiger D, Heckerman D. Learning gaussian networks. Uncertainty Proceedings 1994: Elsevier; 1994. p. 235–43. doi: https://doi.org/10.1016/B978-1-55860-332-5.50035-3

  32. Neapolitan RE. Learning bayesian networks: Pearson Prentice Hall Upper Saddle River, NJ; 2004.

  33. Kalisch M, Bühlman P. Estimating high-dimensional directed acyclic graphs with the PC-algorithm. J Mach Learn Res. 2007;8(3):613–36.

    Google Scholar 

  34. Yuan C, Malone B. Learning optimal Bayesian networks: A shortest path perspective. J Artif Intell. 2013;48:23–65.

    Google Scholar 

  35. Masegosa AR, Moral S. New skeleton-based approaches for Bayesian structure learning of Bayesian networks. Appl Soft Comput. 2013;13(2):1110–20.

    Article  Google Scholar 

  36. Scutari M. Learning Bayesian Networks with the bnlearn R Package. J Stat Softw. 2010;35(3):1–22. https://doi.org/10.18637/jss.v035.i03.

    Article  Google Scholar 

  37. Champely S, Ekstrom C, Dalgaard P, Gill J, Weibelzahl S, Anandkumar A, Ford C, Volcic R, De Rosario H, De Rosario MH. Package ‘pwr’. R package version. 2018;1(2).

  38. Holmes CJ, Racette SB. Holmes CJ, Racette SB. The Utility of Body Composition Assessment in Nutrition and Clinical Practice: An Overview of Current Methodology. Nutrients. 2021;13(8):2493. doi: https://doi.org/10.3390/nu13082493. PMID: 34444653.

  39. Hori Y, Hoshino M, Inage K, Miyagi M, Takahashi S, Ohyama S, et al. Gender-specific analysis for the association between trunk muscle mass and spinal pathologies. Sci Rep. 2021;11(1):7816. https://doi.org/10.1038/s41598-021-87334-4 (PMID: 33837250).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Hoy D, Bain C, Williams G, March L, Brooks P, Blyth F, et al. A systematic review of the global prevalence of low back pain. Arthritis Rheum. 2012;64(6):2028–37. https://doi.org/10.1002/art.34347 (PMID: 22231424).

    Article  PubMed  Google Scholar 

  41. Kjaer P, Bendix T, Sorensen JS, Korsholm L, Leboeuf-Yde C. Are MRI-defined fat infiltrations in the multifidus muscles associated with low back pain? BMC Med. 2007;5:2. https://doi.org/10.1186/1741-7015-5-2 (PMID: 17254322).

    Article  PubMed  PubMed Central  Google Scholar 

  42. Lorbergs AL, Allaire BT, Yang L, Kiel DP, Cupples LA, Jarraya M, et al. A Longitudinal Study of Trunk Muscle Properties and Severity of Thoracic Kyphosis in Women and Men: The Framingham Study. J Gerontol A Biol Sci Med Sci. 2019;74(3):420–7. https://doi.org/10.1093/gerona/gly056 (PMID: 29688268).

    Article  PubMed  Google Scholar 

  43. Cossio Bolaños MA, Andruske CL, De Arruda M, Sulla-Torres J, Urra-Albornoz C, Rivera-Portugal M, et al. Muscle Mass in Children and Adolescents: Proposed Equations and Reference Values for Assessment. Front Endocrinol (Lausanne). 2019;10:583. https://doi.org/10.3389/fendo.2019.00583 (PMID: 31555209).

    Article  Google Scholar 

  44. Shou J, Chen P-J, Xiao W-H. Mechanism of increased risk of insulin resistance in aging skeletal muscle. Diabetol Metab Syndr. 2020;12:14. https://doi.org/10.1186/s13098-020-0523-x (PMID: 32082422).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  45. Ooi PH, Thompson-Hodgetts S, Pritchard-Wiart L, Gilmour SM, Mager DR. Pediatric Sarcopenia: A Paradigm in the Overall Definition of Malnutrition in Children? JPEN J Parenter Enteral Nutr. 2020;44(3):407–18. https://doi.org/10.1002/jpen.1681 (PMID: 31328301).

    Article  PubMed  Google Scholar 

  46. Demontis F, Piccirillo R, Goldberg AL, Perrimon N. The influence of skeletal muscle on systemic aging and lifespan. Aging Cell. 2013;12(6):943–9. https://doi.org/10.1111/acel.12126 (Epub 2013 Jul 17 PMID: 23802635).

    Article  CAS  PubMed  Google Scholar 

  47. Artero EG, Ruiz JR, Ortega FB, España-Romero V, Vicente-Rodríguez G, Molnar D, et al. Muscular and cardiorespiratory fitness are independently associated with metabolic risk in adolescents: the HELENA study. Pediatr Diabetes. 2011;12(8):704–12. https://doi.org/10.1111/j.1399-5448.2011.00769.x (Epub 2011 Apr 6 PMID: 21470352).

    Article  PubMed  Google Scholar 

  48. Cohen DD, Gómez-Arbeláez D, Camacho PA, Pinzon S, Hormiga C, Trejos-Suarez J, et al. Low muscle strength is associated with metabolic risk factors in Colombian children: the ACFIES study. PLoS ONE. 2014;9(4): e93150. https://doi.org/10.1371/journal.pone.0093150 (PMID: 24714401).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Gómez-Campos R, Andruske CL, Arruda Md, Urra Albornoz C, Cossio-Bolaños M. Proposed equations and reference values for calculating bone health in children and adolescent based on age and sex. PLoS One. 2017;12(7):e0181918. doi: https://doi.org/10.1371/journal.pone.0181918. PMID: 28759569.

  50. Liu J, Yan Y, Xi B, Huang G, Mi J, Child C, et al. Skeletal muscle reference for Chinese children and adolescents. J Cachexia Sarcopenia Muscle. 2019;10(1):155–64. https://doi.org/10.1002/jcsm.12361 (PMID: 30499245).

    Article  PubMed  Google Scholar 

  51. Benson AC, Torode ME, Fiatarone Singh MA. Muscular strength and cardiorespiratory fitness is associated with higher insulin sensitivity in children and adolescents. Int J Pediatr Obes. 2006;1(4):222–31. https://doi.org/10.1080/17477160600962864 (PMID: 17907329).

    Article  PubMed  Google Scholar 

  52. Quiterio A, Carnero E, Silva A, Bright B, Sardinha L. Anthropometric models to predict appendicular lean soft tissue in adolescent athletes. Med Sci Sports Exerc. 2009;41(4):828–36. https://doi.org/10.1249/MSS.0b013e31818ffe4b (PMID: 19276850).

    Article  PubMed  Google Scholar 

  53. Cole TJ, Ahmed ML, Preece MA, Hindmarsh P, Dunger DB. The relationship between Insulin-like Growth Factor 1, sex steroids and timing of the pubertal growth spurt. Clin Endocrinol (Oxf). 2015;82(6):862–9. https://doi.org/10.1111/cen.12682 (PMID: 25418044).

    Article  CAS  Google Scholar 

  54. Jeddi M, Dabbaghmanesh MH, Omrani GR, Ayatollahi SMT, Bagheri Z, Bakhshayeshkaram M. Relative Importance of Lean and Fat Mass on Bone Mineral Density in Iranian Children and Adolescents. Int J Endocrinol Metab. 2015;13(3): e25542. https://doi.org/10.5812/ijem.25542v2 (PMID: 26401143).

    Article  PubMed  PubMed Central  Google Scholar 

  55. Janicka A, Wren TA, Sanchez MM, Dorey F, Kim PS, Mittelman SD, et al. Fat mass is not beneficial to bone in adolescents and young adults. J Clin Endocrinol Metab. 2007;92(1):143–7. https://doi.org/10.1210/jc.2006-0794 (PMID: 17047019).

    Article  CAS  PubMed  Google Scholar 

  56. Chen Y-Y, Fang W-H, Wang C-C, Kao T-W, Yang H-F, Wu C-J, et al. Fat-to-muscle ratio is a useful index for cardiometabolic risks: A population-based observational study. PLoS ONE. 2019;14(4): e0214994. https://doi.org/10.1371/journal.pone.0214994 (PMID: 30964893).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  57. Kouda K, Nakamura H, Fujita Y, Ohara K, Iki M. Increased ratio of trunk to appendicular fat and increased blood pressure: study of a general population of Hamamatsu children. Circ J. 2012;76(12):2848–54. https://doi.org/10.1253/circj.cj-12-0417 (PMID: 22893277).

    Article  PubMed  Google Scholar 

  58. Nam S, Kim K, Cha B, Song Y, Lim S, Lee H, et al. Low-dose growth hormone treatment combined with diet restriction decreases insulin resistance by reducing visceral fat and increasing muscle mass in obese type 2 diabetic patients. Int J Obes Relat Metab Disord. 2001;25(8):1101–7. https://doi.org/10.1038/sj.ijo.0801636 (PMID: 11477493).

    Article  CAS  PubMed  Google Scholar 

  59. Steene-Johannessen J, Anderssen SA, Kolle E, Andersen LB. Low muscle fitness is associated with metabolic risk in youth. Med Sci Sports Exerc. 2009;41(7):1361–7. https://doi.org/10.1249/MSS.0b013e31819aaae5 (PMID: 19516166).

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

The present study was supported by a grant from the Vice-chancellor for Research, Shiraz University of Medical Sciences, Shiraz, Iran.

Funding

The research grant provided by Research Deputy of Shiraz University of Medical Sciences (No. 20257). Funding body of the study did not play any role in the design of the study, collection, analysis, and interpretation of data and in writing the manuscript.

Author information

Authors and Affiliations

Authors

Contributions

AA, STH and MHD contributed in designed the study, analyzed the data, and interpreted the results. MR, MGZ and MJ contributed in analysis of data, interpretation the results and wrote the manuscript drafting. All authors have read and approved the manuscript.

Corresponding authors

Correspondence to Seyed Taghi Heydari or Mohammad Hossein Dabbaghmanesh.

Ethics declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of Shiraz University of Medical Sciences (IR.SUMS.REC.1399.676). All children gave oral consent and their parents gave written informed consent before participation in the study. All methods were carried out in accordance with relevant regulations and guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Rahmani, M., Ardalan, A., Ghaderi-Zefrehei, M. et al. Probabilistic prediction of segmental body composition in Iranian children and adolescents. BMC Pediatr 22, 524 (2022). https://doi.org/10.1186/s12887-022-03580-z

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12887-022-03580-z

Keywords