Precise Prediction of Total Body Lean and Fat Mass From Anthropometric and Demographic Data: Development and Validation of Neural Network Models

Abstract

Background:

Estimating body composition is relevant in diabetes disease management, such as drug administration and risk assessment of morbidity/mortality. It is unclear how machine learning algorithms could improve easily obtainable body muscle and fat estimates. The objective was to develop and validate machine learning algorithms (neural networks) for precise prediction of body composition based on anthropometric and demographic data.

Methods:

Cross-sectional cohort study of 18 430 adults and children from the US population. Participants were examined with whole-body dual X-ray absorptiometry (DXA) scans, anthropometric assessment, and answered a demographic questionnaire. The primary outcomes were predicted total lean body mass (_predLBM), total body fat mass (_predFM), and trunk fat mass (_predTFM) compared with reference values from DXA scans.

Results:

Participants were randomly partitioned into 70% training (12 901) data and 30% validation (5529) data. The prediction model for _predLBM compared with lean body mass measured by DXA (_DXALBM) had a Pearson’s correlation coefficient of R = 0.99 with a standard error of estimate (SEE) = 1.88 kg (P < .001). The prediction model for _predFM compared with fat mass measured by DXA (_DXAFM) had a Pearson’s coefficient of R = 0.98 with a SEE = 1.91 kg (P < .001). The prediction model for _predTFM compared with DXA measured trunk fat mass (_DXAFM) had a Pearson’s coefficient of R = 0.98 with a SEE = 1.13 kg (P < .001).

Conclusions:

In this study, neural network models based on anthropometric and demographic data could precisely predict body muscle and fat composition. Precise body estimations are relevant in a broad range of clinical diabetes applications, prevention, and epidemiological research.

Keywords

diabetes neural network prediction lean body mass fat mass body composition

Introduction

Muscle and fat are important compartments of the human body composition, as both a shortage and an excess of these can impact health and physical function.^1,2 Body mass index (BMI) is a method to distinguish those with an increased risk of morbidity such as diabetes mellitus and cardiovascular disease, as the World Health Organization defines overweight as a BMI above 25 and obesity as BMI above 30.^3-5 A BMI above 30 is a substantial risk factor for increased mortality and morbidity.^6,7 It is a common measurement of obesity and widely used in clinics and research purposes. It reflects the total body mass, which consists of fat mass (FM) and lean body mass (LBM). The FM can further be divided into subcutaneous fat and visceral adipose tissue (VAT), the latter being categorized as the dangerous fat due to an association with an elevated risk of morbidity.^8-10 Epidemiological studies have shown that the distribution of fat and the amount of visceral fat is related to increased disease risk in both children and adolescents.¹¹ The LBM is identified to have a significant influence on various physiological pathways, and people with reduced LBM may have negative health consequences.^12-14 These parameters are dynamic and change due to aging, illness, and physical performance. Therefore, BMI does not accurately define the relationship between FM, LBM, and increased morbidity incidence.^15,16 Moreover, the body composition is an important factor in drug dose calculations and administration, especially in obese patients with diabetes.⁶

Even though several advanced methods are available to get accurate estimations of the FM and LBM, such as dual X-ray absorptiometry (DXA) or imaging techniques (ie, magnetic resonance imaging),¹⁷ these approaches are expensive and often unfitting for daily medical diabetes management and analysis of broad populations. Therefore, several studies have been proposing estimation models based on anthropometric measurements as an alternative to BMI.^18,19,20-29 However, these models are focused on simple statistical models usability such as estimations calculated by hand in clinical settings rather than optimizing the performance of the estimates.¹⁵ Also, many studies have been targeting specific subpopulation such as young athletes, nonobese subjects, or elderly patients, and many of these studies included a low number of participants.³⁰ These circumstances limit the transferability.

With the increased availability of computers in the clinical setting, tasks as calculating estimates can easily be delegated to a computer or a smartphone. Moreover, methods from machine learning have shown the potential to make accurate predictions of clinical events.^31-35 It is of high value to investigate how a state-of-art machine learning algorithm could perform in estimating body composition using simple patients’ anthropometric measurements.

This research had the aim of exploring the potential of developing a machine learning model to predict accurate estimates of LBM and FM from simple anthropometric measurements in a wide sample of the general population which could be useful for patients with both type 1 and type 2 diabetes, people at risk, and young adults.

Methods and Materials

Data Source

The cohort was derived from individuals who were enrolled in the cross-sectional study: National Health and Nutrition Examination Survey 1999-2006 (NHANES). The study includes a stratified multistage probability sample of the noninstitutionalized US population. The survey comprised of home interviews of participants followed by physical examinations and laboratory measurements. The Centers for Disease Control and Prevention National Center for Health Statistics institutional review board approved the original survey protocol, and all participants provided written informed consent. Further information about the NHANES can be found online.³⁶ This research meets the recommendations standards described by the “Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis” (TRIPOD).³⁷

Study Sample

The study sample included participants aged 8-69 years who had a whole-body DXA scan examination, had a body measurement assessment (anthropometric assessment), and answered the demographic questionnaire. Exclusion from DXA assessment was pregnancy, self-reported history of radiographic contrast material in recent past, and reported weight over 300 pounds or height over 65″ (DXA table limitations). The DXA examinations were assessed with a Hologic QDR-4500A fan-beam densitometer (Hologic Inc., Massachusetts), and the body measures were obtained by trained specialists during the physical examinations. Only participants with complete data on the DXA scan, body measurements, and demographics were included in the cohort. A total of 18 430 individuals were qualified to enroll in the cohort.

Predictive Features

The study dataset included two classes of predictors (ie, features) that are easily obtainable in a clinical setting: (1) demographic predictors were age, sex, and ethnicity; (2) anthropometric measurements were height, weight, BMI, upper leg length, maximal calf circumference, upper arm length, arm circumference, waist circumference, thigh circumference, triceps skinfold, and subscapular skinfold. A total of 14 features were investigated for their potential to predict the study outcome.

Outcome

The primary outcomes were LBM in kg, body FM (FM) in kg, and TFM in kg (as an indicator of VAT^11,38). The predicted _predLBM, _predFM, and _predTFM were based on machine learning with a combination of the features described. The predicted estimates were compared with the DXA scan results of _DXALBM, _DXAFM, and _DXATFM in order to assess the accuracy of the predicted values.

Prediction Approach

The cohort was randomly divided into the training dataset, of which the prediction models were developed, and the validation dataset, to obtain an unbiased estimate of the model’s performance. The dataset used for training included 70% of the study sample, and the dataset used for validation consisted of the remaining 30%. We split our study sample randomly so that participants could not be in both datasets. The procedure is illustrated in Figure 1. We derived three machine learning models from the training data: all based on artificial neural network regression algorithms (ANN) using Levenberg-Marquardt as a training function with ten hidden nodes; one model for predicting _predLBM, one for prediction of _predTFM, and another model for predicting _predFM. We chose ANN for deriving the prediction models as they have several advantages, that is, flexible input layer, capacity to indirectly identify dynamic nonlinear interactions between dependent and independent variables, capacity to identify all potential interactions between predictor variables, and have proven good performance in the medical field of prediction.^39,40

Figure 1.

Illustrating the procedure for selecting the training dataset, validation dataset, and training the neural network prediction models using the training dataset. The training dataset is further divided into an actual training dataset and a test dataset to minimize the potential of overfitting the models.

To avoid the potential of overfitting the model to the training data, which would lead to decreased performance in the validation dataset (decreased generalizability), the training dataset was further split into 70%/30% for training and testing during the optimization of the models. This procedure minimizes the potential of overfitting the models. For transparency of potential overfitting, both results from the training dataset and the validation dataset are presented in the results. The ANN was implemented in Matlab R2016b (The Mathworks Inc., Natick, Massachusetts) using the Levenberg-Marquardt algorithm and with a hidden layer size of ten. The layer size was optimized using the training dataset. To investigate the performance of the features, the features were added to the model using forward selection and evaluated using the impact on total performance in the training dataset. Our study did not recalibrate the ANN models post the validation procedure.

Statistical Assessment

Statistical assessments were used to compare the predicted _predLBM, _predTFM, and _predFM to the values obtained by the whole-body DXA scans. We calculated the Pearson correlation coefficient R for relations between _predLBM/_DXALBM, _predTFM/_DXATFM, and _predFM/_DXAFM, as well as P-values for testing the hypothesis that there is no relationship between the observed. Moreover, the standard error of estimate (SEE) was used to assess statistical uncertainty. To investigate the performance in subgroups, the Pearson’s correlation coefficient was also calculated for participants below the age of 18, with a BMI above 30, and for men and women.

We also constructed a regression plot of the model predictions and DXA scan measurements. All analyses were implemented using the Matlab (vR2016b) (The MathWorks Inc., Natick, Massachusetts) software.

Results

Cohort and Study Population Characteristics

There were a total of 18 430 participants included in this study. The training and validation dataset contained 12 901 (70%) and 5529 (30%) participants.

The characteristics of the study population for the training and validation dataset are presented in Table 1. No substantial differences were seen between the training and validation dataset. A slight overrepresentation of males was included in the study cohort (57%). 62% of the cohort were adults (age above 18 years), and the composition of ethnicity was 40% white (non-Hispanic), 29% Mexican-American, 23%, black (non-Hispanic), and 8% other (including multiracial).

Tabel 1.

Characteristics of the Participants in the Training and Validation Dataset.

	Training	Validation
n	12 901	5529
Males, %	57	56
Age, years	32 (21.1)	32.2 (21.2)
Weight, kg	66.2 (18.8)	66.3 (18.9)
Height, cm	164.1 (13.2)	164.1 (13.3)
BMI, w/h²	24.2 (5)	24.2 (5)
Ethnicity, %
Mexican-American	29	27
Other Hispanic	4	4
Non-Hispanic White	40	41
Non-Hispanic Black	23	24
Other race, including multiracial	4	4

Presented as mean (STD) or as percentage of the populations.

Model Performance

In the validation dataset, the prediction model for lean body mass (_predLBM) compared with LBM measured by DXA (_DXALBM) had a Pearson’s correlation coefficient of R = 0.99 with a SEE = 1.88 kg. These performance measurements were close to the model performance in the training dataset which had an R = 0.99 and SEE = 1.84 kg. The regression plot for the validation dataset is presented in Figure 2, and performances are reported in Table 2. In the subgroup analysis, among participants below the age of 18 and men, the correlations were the same as for the whole dataset (R = 0.99) and slightly reduced in participants with high BMI and in women (R = 0.97).

Figure 2.

Regression plot of dual X-ray absorptiometry measured lean body mass and predicted lean body mass.

Tabel 2.

Performance of the Models in the Validation Dataset Among Subgroups in the Dataset.

Model	R value
Model	All	Age <18	BMI >30	Men	Women
_predLBM	0.99	0.99	0.97	0.99	0.97
_predFM	0.98	0.98	0.91	0.98	0.98
_predTFM	0.98	0.98	0.9	0.98	0.97

BMI, body mass index; _predFM, predicted fat mass; _predLBM; predicted lean body mass; _predTFM, predicted trunk fat mass.

In the validation dataset, the _predFM compared with FM measured by DXA (_DXAFM) had a Pearson’s correlation coefficient of R = 0.98 with a SEE = 1.91 kg. These performance measures were also close to the model performance in the training dataset which had an R = 0.99 and SEE = 1.87 kg. The regression plot for the validation dataset is presented in Figure 3, and performances are reported in Table 2. In the subgroup analysis with participants below the age of 18, among women and men, the correlations were the same as for the whole dataset (R = 0.98) and reduced in participants with high BMI and in women (R = 0.91).

Figure 3.

Regression plot of dual X-ray absorptiometry measured fat mass and predicted fat mass.

Moreover, in the validation dataset, the _predTFM compared with TFM measured by DXA (_DXAFM) had a Pearson’s correlation coefficient of R = 0.98 with a SEE = 1.13 kg. These performance measures were also close to the model performance in the training dataset which had an R = 0.98 and SEE = 1.11 kg. The regression plot for the validation dataset is presented in the supplement (eTable 1 in the Supplement) and performances are reported in Table 2. Due to the large data material all P values were small (P < .001)

Algorithm Variable Importance

The importance of adding each feature into the model are presented in the supplement (eTable 2 in the Supplement) for predicting _predLBM, in the supplement (eTable 3 in the Supplement) for predicting _predFM, and in the supplement (eTable 4 in the Supplement) for _predTFM.

For the model predicting _predLBM, the features with the most influence on the SEE are: body weight, triceps skinfold, waist circumference, and gender—moreover, age, height, subscapular skinfold, BMI, arm circumference, ethnicity, and leg length also add to improve the prediction in a lower scale.

For the model predicting _predFM, the features with the most influence on the SEE are BMI, triceps, waist circumference, weight, gender, age, and ethnicity. Adding calf circumference, thigh circumference, subscapular skinfold, arm length, arm circumference, and height further adds little predictive information.

Discussion

The present study was designed to develop whole-body LBM, FM, and TFM prediction models based on anthropometric dimensions and demographics information. With the use of whole-body DXA scanning as the reference, two prediction models were developed and then validated in an extensive and heterogeneous subject sample. Furthermore, the model’s performance was tested in subgroups of young people, men, women, and obese individuals. In general, the developed models had accurate predictive capability to assess LBM and FM. In obese participants, the prediction of FM was considerably reduced; this might be due to an increased interpersonal variation of fat distribution and difficulties in the accuracy of obtaining anthropometric measurements in obese people.⁴¹

Several research studies have proposed anthropometric prediction models for the assessment of LBM and FM.^18-29 However, most of these studies have been focusing on developing oversimplified prediction equations, have been targeting specific subpopulations, or have low numbers of participants. Kulkarni et al.²⁹ investigated a wide sample of Indian adult individuals (n = 2220) to develop LBM prediction equations from anthropometric measurements using DXA as a reference. Their results showed promising R values in the range of 0.95-0.97 and low SEEs in the range of 1.5-1.9 kg. Wen et al.⁴² developed an equation based on anthropometric for LBM estimation in 763 Chinese participants. The estimation was based on height, weight, and limb circumferences. The R values were reported between 0.95 and 0.96 with low SEE ~1.5 kg. However, these results have limited generalizability to other ethnic populations.

A new study by Lee et al.¹⁵ had the aim to develop simple anthropometric measurements to use calculation for prediction of LBM and FM in adults based on anthropometric from the NHANES 1999-2006. The proposed calculations included age, ethnicity, height, weight, and waist circumference. The results for _predLBM were for men: R² = .91 (SEE = 2.6 kg), for women: R² = 0.85 (SEE = 2.4 kg), and results for _predFM were for men: R² = .90 (SEE = 2.6 kg) and for women: R² = .93 (SEE = 2.4 kg). This study is comparable with the presented study as the cohort was also derived from the NHANES population. The main differences are the focus by Lee et al. on developing practical prediction equations and the inclusion of only adults. Comparing results between this study by Lee et al. and our study supports the hypothesis that developing simplified and easy to calculate hand-prediction models could come at a cost of accuracy. Our results, in general, showed higher R (and R²) values with a lower SEE. Furthermore, Woolcott and Bergman¹⁹ presented a simple prediction equation (relative FM) with the aim of predicting whole-body fat percentage more accurately than BMI. They also used a large cohort derived from the NHANES. The results showed R² ranging from 0.69 to 0.84.

Study Strengths and Limitations

We demonstrated how a machine learning approach could be applied in predicting body composition using measures, which could easily be obtained in a clinical setting for the management of diabetes or an epidemiological study. The models were developed and validated from a wide representative sample of children, teens, and adults; multiple ethnic groups; and participants with large anthropometric diversity. Our separation of the data material into a training and validation dataset ensured that the potential of overfitting was minimized—this was further confirmed in the performance analysis between the training and validation dataset, which showed similar results.

However, there are several limitations to this analysis.

First, our data material was produced with a precise protocol for measuring participants’ anthropometrics. However, the bias in obtaining such measures is unavoidable even in a strictly and protocolized environment as the NHANES. Therefore, it is unexplored how this could influence the performance of the models when used in other settings.

Second, the reference used for LBM, FM, and TFM were measured using one type of DXA device—when developing our machine learning models, this was used as the ground truth. This means the models developed were produced with the aim of increasing accuracy compared with this ground truth. Interunit variability among DXA devices exists due to different causes, including the calibration procedure, calculation software, and variations in photon source intensities.⁴³ This could potentially influence the generalizability of our results (and others) to different settings and needs to be investigated in the future.

Third, our sample included multiple ethnic groups. However, some groups such as Asians and multiethnic groups were not well represented in our cohort.

Conclusion

This study describes the development and subsequent validation of three models for predicting the LBM, FM, and TFM from anthropometric and demographics information using a large representative sample. The approach was based on machine learning and with a focus on optimizing prediction accuracy. The results demonstrate that this approach could provide accurate estimates of the whole-body composition of FM and LBM in a heterogeneous population. This could potentially be a useful tool for the management of patients with both type 1 and type 2 diabetes as well as people at risk.

Supplemental Material

Supplement – Supplemental material for Precise Prediction of Total Body Lean and Fat Mass From Anthropometric and Demographic Data: Development and Validation of Neural Network Models

Supplemental material, Supplement for Precise Prediction of Total Body Lean and Fat Mass From Anthropometric and Demographic Data: Development and Validation of Neural Network Models by Simon Lebech Cichosz, Nicklas Højgaard Rasmussen, Peter Vestergaard and Ole Hejlesen in Journal of Diabetes Science and Technology

Footnotes

Author Contributions

Professor OH and SLC had access to all of the data analyzed in this study. OH and SLC take responsibility for the integrity and the accuracy of the study data analysis and results. SLC, NHR, PV, and OH were involved in the study design, concept, analysis, and interpretation of data. SLC drafted the manuscript and performed the statistical analysis. NHR, PV, and OH were involved in critical revision of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Simon Lebech Cichosz .

Supplemental Material

Supplemental material for this article is available online.

References

Ferrucci

Cooper

Shardell

, et al. Assessing daily physical activity in older adults: unraveling the complexity of monitors, measures, and methods. J Gerontol A Biol Sci Med Sci. 2010;11(6):391-396.

Ebeling

Scott

Body composition and falls risk in older adults. Curr Geriatr Reports. 2019;8(3):210-222.

Hars

Trombetti

Body composition assessment in the prediction of osteoporotic fractures. Curr Opin Rheumatol. 2017;29(4):394-401.

Lee

Giovannucci

EL.

Body composition and mortality in the general population: a review of epidemiologic studies. Exp Biol Med. 2018;243(17-18):1275-1285.

World Health Organization. Obesity and overweight. Available at: https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight. Accessed February 24, 2020.

Renehan

Tyson

Egger

Heller

Zwahlen

Body-mass index and incidence of cancer: a systematic review and meta-analysis of prospective observational studies. Lancet. 2008;371(9612):569-578.

Kopelman

Health risks associated with overweight and obesity. Obes Rev. 2007;8(s1):13-17.

Miazgowski

Krzyżanowska-Świniarska

Dziwura-Ogonowska

Widecka

The associations between cardiometabolic risk factors and visceral fat measured by a new dual-energy X-ray absorptiometry-derived method in lean healthy Caucasian women. Endocrine. 2014;47(2):500-505.

Sasai

Brychta

Wood

, et al. Does visceral fat estimated by dual-energy X-ray absorptiometry independently predict cardiometabolic risks in adults? J Diabetes Sci Technol. 2015;9(4):917-924.

10.

Kaul

Rothney

Peters

, et al. Dual-energy X-ray absorptiometry for quantification of visceral fat. Obesity. 2012;20(6):1313-1318.

11.

Goran

Gower

Treuth

Nagy

. Prediction of intra-abdominal and subcutaneous abdominal adipose tissue in healthy pre-pubertal children. Available at: http://www.stockton-press.co.uk/ijo. Accessed February 24, 2020.

12.

Ruiz

Sui

Lobelo

, et al. Association between muscular strength and mortality in men: Prospective cohort study. BMJ. 2008;337(7661):92-95.

13.

Janssen

Heymsfield

Ross

Low relative skeletal muscle mass (sarcopenia) in older persons is associated with functional impairment and physical disability. J Am Geriatr Soc. 2002;50(5):889-896.

14.

Wolfe

RR.

The underappreciated role of muscle in health and disease. Am J Clin Nutr. 2006;84(3):475-482.

15.

Lee

Keum

, et al. Development and validation of anthropometric prediction equations for lean body mass, fat mass and percent fat in adults using the National Health and Nutrition Examination Survey (NHANES) 1999-2006. Br J Nutr. 2017;118(10):858-866.

16.

Lee

Keum

, et al. Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study. BMJ. 2018;362:k2575.

17.

Lee

Gallagher

Assessment methods in human body composition. Curr Opin Clin Nutr Metab Care. 2008;11(5):566-572.

18.

Valensise

Andreoli

Lello

Magnani

Romanini

De Lorenzo

Total-body skeletal muscle mass: development and cross-validation of anthropometric prediction models. Am J Clin Nutr. 2000;72(3):796-803.

19.

Woolcott

Bergman

RN.

Relative fat mass (RFM) as a new estimator of whole-body fat percentage ─ A cross-sectional study in American adult individuals. Sci Rep. 2018;8(1):1-11.

20.

Newton

Alfonso

York-Crowe

, et al. Comparison of body composition methods in obese African-American women. Obesity. 2006;14(3):415-422.

21.

Jackson

Stanforth

Gagnon

, et al. The effect of sex, age and race on estimating percentage body fat from body mass index: the heritage family study. Int J Obes. 2002;26(6):789-796.

22.

Lean

MEJ

Han

Deurenberg

. Predicting body composition by densitometry from simple anthropometric measurements. Am J Clin Nutr. 1996;63(1):4-14.

23.

Friedl

Westphal

Marchitelli

Patton

Chumlea

Guo

SS.

Evaluation of anthropometric equations to assess body-composition changes in young women. Am J Clin Nutr. 2001;73(2):268-275.

24.

Cui

Truesdale

Cai

Stevens

Evaluation of anthropometric equations to assess body fat in adults: NHANES 1999-2004. Med Sci Sports Exerc. 2014;46(6):1147-1158.

25.

Gallagher

Heymsfield

Heo

Jebb

Murgatroyd

Sakamoto

Healthy percentage body fat ranges: an approach for developing guidelines based on body mass index. Am J Clin Nutr. 2000;72(3):694-701.

26.

Gómez-Ambrosi

Silva

Catalán

, et al. Clinical usefulness of a new equation for estimating body fat. Diabetes Care. 2012;35(2):383-388.

27.

Stevens

Cai

Heymsfield

Truesdale

KP.

Prediction of percent body fat measurements in Americans 8 years and older. Int J Obes. 2016;40(4):587-594.

28.

Withers

Norton

Craig

Hartland

Venables

The relative body fat and anthropometric prediction of body density of South Australian females aged 17-35 years. Eur J Appl Physiol Occup Physiol. 1987;56(2):181-190.

29.

Kulkarni

Kuper

Taylor

, et al. Development and validation of anthropometric prediction equations for estimation of lean body mass and appendicular lean soft tissue in Indian men and women. J Appl Physiol. 2013;115(8):1156-1162.

30.

Al-Gindan

Hankey

Leslie

Govan

Lean

MEJ

. Predicting muscle mass from anthropometry using magnetic resonance imaging as reference: a systematic review. Nutr Rev. 2014;72(2):113-126.

31.

Arcadu

Benmansour

Maunz

Willis

Haskova

Prunotto

Deep learning algorithm predicts diabetic retinopathy progression in individual patients. npj Digit Med. 2019;2(1).

32.

Parikh

Manz

Chivers

, et al. Machine learning approaches to predict 6-month mortality among patients with cancer. JAMA Netw Open. 2019;2(10):e1915997.

33.

Jensen

Cichosz

Dinesen

Hejlesen

OK.

Moving prediction of exacerbation in chronic obstructive pulmonary disease for patients in telecare. J Telemed Telecare. 2012;18(2):99-103.

34.

Cichosz

Frystyk

Hejlesen

Tarnow

Fleischer

A novel algorithm for prediction and detection of hypoglycemia based on continuous glucose monitoring and heart rate variability in patients with type 1 diabetes. J Diabetes Sci Technol. 2014;8(4):731-737.

35.

Cichosz

Voelsang

Tarnow

Hasenkam

Fleischer

Prediction of in-hospital pressure ulcer development. Adv Wound Care. 2019;8(1):1-6.

36.

NHANES 1999-2006. Centers for Disease Control and Prevention (CDC). National Center for Health Statistics (NCHS). National Health and Nutrition Examination Survey Data. Hyattsville, MD: U.S. Department of Health and Human Services, Centers for Disease Con. Available at: https://wwwn.cdc.gov/nchs/nhanes/Default.aspx. Accessed December 1, 2019.

37.

Moons

KGM

Altman

Reitsma

, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162(1):W1-W73.

38.

Clasey

Bouchard

Teates

, et al. The use of anthropometric and dual-energy X-ray absorptiometry (DXA) measures to estimate total abdominal and abdominal visceral fat in men and women. Obes Res. 1999;7(3):256-264.

39.

Dreiseitl

Ohno-Machado

Logistic regression and artificial neural network classification models: a methodology review. J Biomed Inform. 2002;35(5-6):352-359.

40.

J V.

Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. J Clin Epidemiol. 1996;49(11):1225-1231.

41.

Bray

Bouchard

James

Handbook of Obesity. New York: Marcel Dekker, Inc, 1998.

42.

Wen

Wang

Jiang

Zhang

YM.

Anthropometric equation for estimation of appendicular skeletal muscle mass in Chinese adults. Asia Pac J Clin Nutr. 2011;20(4):551-556.

43.

Trevisan

Gandolini

Sibilla

Penotti

Caraceni

Ortolani

Bone mass measurement by DXA: influence of analysis procedures and interunit variation. J Bone Miner Res. 1992;7(12):1373-1382.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.58 MB