High-standard predictive equations for estimating body composition using bioelectrical impedance analysis: a systematic review

The appropriate use of predictive equations in estimating body composition through bioelectrical impedance analysis (BIA) depends on the device used and the subject’s age, geographical ancestry, healthy status, physical activity level and sex. However, the presence of many isolated predictive equations in the literature makes the correct choice challenging, since the user may not distinguish its appropriateness. Therefore, the present systematic review aimed to classify each predictive equation in accordance with the independent parameters used. Sixty-four studies published between 1988 and 2023 were identified through a systematic search of international electronic databases. We included studies providing predictive equations derived from criterion methods, such as multi-compartment models for fat, fat-free and lean soft mass, dilution techniques for total-body water and extracellular water, total-body potassium for body cell mass, and magnetic resonance imaging or computerized tomography for skeletal muscle mass. The studies were excluded if non-criterion methods were employed or if the developed predictive equations involved mixed populations without specific codes or variables in the regression model. A total of 106 predictive equations were retrieved; 86 predictive equations were based on foot-to-hand and 20 on segmental technology, with no equations used the hand-to-hand and leg-to-leg. Classifying the subject’s characteristics, 19 were for underaged, 26 for adults, 19 for athletes, 26 for elderly and 16 for individuals with diseases, encompassing both sexes. Practitioners now have an updated list of predictive equations for assessing body composition using BIA. Researchers are encouraged to generate novel predictive equations for scenarios not covered by the current literature. Registration code in PROSPERO: CRD42023467894. Supplementary Information The online version contains supplementary material available at 10.1186/s12967-024-05272-x.


Introduction
The determination of body composition is a common practice for evaluating health and nutritional status, as well as monitoring the effects of training or diet strategies [1,2].While a simplistic approach involves assessing body composition based on body mass and its changes over time, breaking it down into different components enables a more meticulous and in-depth evaluation [3,4].At the molecular level of body composition analysis [4] initial attempts to break down body mass were based on a two-component model, categorizing it into fat (FM) and fat-free mass (FFM), also known as lean mass [5].It is worth noting that FFM is often also referred to by synonyms such as lean mass or lean body mass [6].However, monitoring FFM alone can obscure meaningful changes in body composition since its composition comprises several components (e.g., water, protein, and minerals) not considered in the assessment [5,6].
To overcome the limitations of the two-component model and enhance the accuracy of body composition analysis, various multi-component models have been developed over time [2].These models necessitate the use of one or more methods, including dual-energy X-ray absorptiometry (DXA) for bone mineral assessment, dilution techniques for total-body water, hydrostatic weighing or air displacement plethysmography for body volume, total-body potassium for body cell mass, and magnetic resonance imaging (MRI) and computed tomography for skeletal muscle or other tissues and organs [5].For example, Wang et al. [3] presented a procedure based on a four-component model that is currently considered the state-of-art method for determining FFM and FM.Additionally, the use of multi-components models allows for the breakdown of FFM into different parts such as total body water (TBW), intra (ICW) and extracellular water (ECW), body cell mass (BCM), lean soft mass (LSM), skeletal muscle mass (SMM), and bone mass [7,8].With these procedures available to researchers, predicting body mass components by assuming constant hydration of FFM (i.e., TBW/FFM = 0.73) or neglecting the potential variation in FFM and FM density [6] cannot be considered as a valid approach for assessing body composition.Indeed, several dated studies have employed coefficients such as this to derive FFM and subsequently FM based on TBW predictions, or they have relied on density assumptions to derive FFM and FM using only densitometry techniques (i.e., hydrostatic weighing or air displacement plethysmography) [9][10][11].In particular, the practice of deriving FFM by assuming that TBW represents a constant fraction of it has led to BIA being incorrectly identified as a hydration-dependent method.Therefore, a comprehensive determination of body mass components using multi-components models enables a valid assessment of body composition in different populations, especially those with non-conventional hydration or density properties (e.g., children, athletes, or older individuals) [6,12].
Considering the limitations of densitometric, hydrometric, or imaging techniques for quantifying body composition due to high costs, lengthy procedures, and non-user-friendly processes, alternative methods such as bioelectrical impedance analysis (BIA) or surface anthropometry are often employed for routine assessments [13][14][15][16].The theoretical basis of BIA revolves around the conductivity properties of biological tissues, quantified as bioelectrical resistance and reactance, deriving by conductor volumes (i.e., lean soft components) [17].Bioelectrical resistance represents the opposition offered by the body to the flow of an alternating electrical current and is inversely related to the water and electrolyte content of tissues [17].Bioelectrical reactance is related to the capacitance properties of the cell membrane and variations that can occur depending on its integrity, function, and composition [17].Notably, these properties can be attributed to theoretical models, according to which biological tissues are traditionally conceived as electrical circuits arranged in series with each other [18].Starting from the relationships between resistance and reactance with TBW [19], numerous BIA-based predictive equations have been developed over the years [20].Previous studies have demonstrated that the use of different BIA technologies (i.e., hand to hand, leg to leg, foot to hand, and segmental) and sampling frequencies results in different outputs, so that these equations cannot be interchangeable between different devices [21][22][23][24].Additionally, the choice of the equation should be made considering the subjects' characteristics, such as chronological and biological age, geographical provenance, sex, health status, and level of physical activity [25][26][27].Therefore, the choice of the appropriate predictive equation is crucial to ensure the validity of the body composition estimation.A further question is that various studies including those providing reference data based on BIA do not disclose the procedures used [24,28].Many of these studies merely mention the type of software employed, making it impossible to discern which formulas were used to convert raw bioelectrical parameters into components of body mass.Furthermore, such data may no longer be representative over time because companies producing such software can alter the equations without notifying users [29].While this approach is undoubtedly quicker than a systematic and accurate selection of the most appropriate equation among those available in the literature, does not guarantee a high-standard validity.
The literature is now replete with predictive equations developed for estimating body composition using BIA.However, several limitations may question the validity of some equations.First, the use of the two-component model instead of a multi-component model approach.Second, predicting FFM and FM starting from hydration or density assumptions.Third, the development of predictive models without using criterion methods.Fourth, mixing different populations without including specific factors (e.g., age, maturity status, sex, physical activity level) as independent variables in the predictive models [30].Therefore, the present review aimed to examine the relevant literature to extract all the predictive equations currently available, and list only those free of the aforementioned limitations.In addition, we provided a clear organization of the predictive equations based on the BIA technology, sampling frequency, and population characteristics.Such a list will help scientists and practitioners select the most appropriate predictive equations in accordance with the BIA devices and population peculiarities.Summarizing the state-of-the-art of BIA-based prediction of body composition will also be helpful to optimize the development of new predictive equations considering what is currently lacking in the literature.Obtaining accurate results would also allow for utilizing the data to evaluate other health-related parameters through estimating basal metabolism or the quantity of macronutrients based on body composition data [31].Figure 1 depicts the premise and the problem at the basis of the present study, presenting the aims and future perspectives.

Search strategy and eligibility criteria
The present study was carried out following the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines [32].The two independent researchers (F.C. and Z.N.) conducted systematic searches in Scopus, and PubMed on December 22th, 2023, identifying potential eligible studies without any restriction related to year of publication.A search query included combinations of at least one of the terms identifying BIA, with at least one of the terms regarding the study design, a term regarding reference techniques, and a term referring to body composition.We composed different query strings depending on the variability of the databases functioning, using the following terms: (bioelectrical impedance analysis) OR The inclusion criteria were as follows: -Peer-reviewed articles that developed at least one predictive equation for estimating body mass compo-

Study selection
A total of 106 BIA-based predictive equations resulted from the 64 studies included in the review.The PRISMA flow chart is shown in Fig. 2.

Bioelectrical impedance-based predictive models according to gender, populations, and technologies
Figure 4 schematizes the number of equations available for each BIA technology (e.g., hand to hand, leg to leg, foot to hand, and segmental) according to sex and population.
The number of the predictive equations for each estimable body mass component, according to sex and population are shown in Figs. 5 and 6, for the foot to hand and segmental technologies, respectively.

Bioelectrical impedance-based equations for under 18 years old people
Table 1 reports the characteristics of the 19 predictive equations developed in 14 studies on subjects < 18 years old from the general populations.Fifteen predictive equations from 11 studies were developed with a SF-BIA, while four predictive equations from three studies were developed with MF-BIA using bioelectrical parameters measured at 50 kHz.

Bioelectrical impedance-based equations for adults from the general population
Table 2 reports the characteristics of the 26 predictive equations developed in 14 studies on subjects from the general populations.Twenty-two predictive equations from 12 studies were developed with a SF-BIA, while four predictive equations from two studies were developed with MF-BIA using bioelectrical parameters measured at 5, 50, and 250 kHz.Only one study provided a predictive equation using BIS at a frequency of 50 kHz.

Bioelectrical impedance-based equations for athletes
Table 3 reports the characteristics of the 19 predictive equations developed in 11 studies on athletes.Fifteen predictive equations from 10 studies were developed with a SF-BIA, while four predictive equations from one study were developed with MF-BIA using bioelectrical parameters measured at 50 kHz.

Bioelectrical impedance-based equations for elderly
Table 4 reports the characteristics of the 26 predictive equations developed in 17 studies on elderly subjects.Seventeen predictive equations from 11 studies were developed with a SF-BIA, while eight predictive equations from seven studies were developed with MF-BIA using bioelectrical parameters measured at 2, 50, and

Bioelectrical impedance-based equations for people with diseases
Table 5 reports the characteristics of the 16 predictive equations developed in 8 studies on subjects with diseases.Three studies presented seven predictive equations using SF-BIA, while six predictive equations from two studies were developed with MF-BIA using bioelectrical parameters measured at 50 and 100 kHz.Three studies developed three predictive equation using BIS at a frequency of 50 kHz.

Discussion
The present systematic review was designed to compile a list of BIA-based predictive equations for estimating body mass components in target populations.Specifically, we included only articles that developed new predictive models and excluded experimental designs that validated existing equations, mixed different populations, or used the two-component model as a reference for developing predictive equations.This resulted in 106 predictive equations that met high-standard procedures, organized based on the four BIA technologies (i.e., handto-hand, leg-to-leg, foot-to-hand, and segmental), the sampling frequency (i.e., single and multi-frequency) and five population categories (i.e., under 18 years old, adults, athletes, elderly, and people with diseases).Additionally, the geographical areas of the participants were identified.The present findings provide an updated starting point for researchers to identify potential gaps in the literature and develop further predictive equations.Moreover, this updated list facilitates the easy identification of the most accurate equations for specific populations and technologies, aiding practitioners in implementing best practices.

Bioelectrical impedance-based equations for athletes
Nineteen predictive equations developed in 11 studies on athletes were included.Four studies involved Portuguese male and female participants from different sport disciplines [39,69,84,87], while two studies recruited male Brazilian army cadets [85] considered as very active individuals [103] and adolescents from various sport disciplines [104].One study was conducted on USA collegiate female athletes from different sports [62], one on Greek elite female dancers [51], and two studies on elite male soccer players in Taiwan [88] and Italy [31].One study con considered Italian paddle players [48].Out of the 19 available equations, 15 were developed with foot to hand technology [31,39,48,51,62,69,84,85,87,104], while four equations were developed using segmental technology [88].Considering the foot to hand technology, 12 equations are suitable for male [31,39,48,69,84,85,87,104] and eight for female people only [39,51,62,69,87,104].As for segmental technology, four equations are specific for males [88], and no equation is available for females.Regarding foot to hand technology, we found an equation for estimating FM in males [48], five equations for estimating FFM in males [31,69,84,85,104], and four in females only [51,62,69,104].Moreover, we retrieved one equation for estimating TBW [39], one for estimating ECW [39] and five for estimating LSM [31,87], where three of them are suitable for both sexes [39,87], and two in male only [31].As for segmental technology, four equations for estimating whole-body, upper-limb,        lower-limb, and trunk FFM are available for males only [88].All these equations were developed at a single frequency of 50 kHz, even those developed using a multifrequency device [88].

Bioelectrical impedance-based equations for people with diseases
Sixteen predictive equations developed in eight studies on people with diseases were included.The selected studies included European [35,40,56,64,65,98] and Asian [41,99] male and female participants.Out of the 16 available equations, 14 were developed with foot to hand technology [40,64,98] and eight using segmental technology [40,41,56,64,65,98,99].Regarding the foot to hand technology, there is one equation suitable for assessing FFM in overweight and obese postmenopausal females [98], four for appendicular FM and FFM in functionally limited male and female individuals [40], four for assessing FFM in amputee males [41], one for predicting TBW in anorexic females [56], and three for appendicular LSM in frail [64], chronic kidney [65], or hemodialysis male and female subjects [99].Seven equations were developed using single frequency devices at 50 kHz [35,40,98], while 9 equations were made with multifrequency devices involving measures obtained at 50 [41,64,65,99] and 100 kHz [56].However, the use of BIA in individuals with underlying medical conditions necessitates a thorough consideration of the reliability features inherent in the measurements.This is imperative due to the heightened likelihood of fluctuations in water content, which may manifest more frequently in this cohort compared to other populations [105].

Limitations of the review and future perspectives
One limitation of the current manuscript may concern the grouping of predictive equations based on the BIA technology, since different devices using the same technology may yield different outcomes due to potential between-device inconsistencies [106], even though agreement has been observed [21].Therefore, new comparisons between different devices of the same technology should be conducted.Other considerations could be made, not merely as limitations of the present manuscript but intrinsic to BIA.First, considering the lack of agreement between bioelectrical values obtained at different frequencies [21,107], the sampling frequency information has been reported for each of the included studies.In this regard, future studies should clarify the advantages of using multi-frequency devices given that most predictive equations predominantly include measures obtained at a single frequency of 50 kHz.Second, the procedures for BIA were updated in 2004 [15] and further in 2020 for athletes [2].Therefore, inconsistencies in methodological procedures (e.g., the standardization of food and beverage intake before the assessment, electrode placement, skin cleansing) may have impacted the final outcomes for each study [108].Third, the standard error is often reported as absolute values of body mass, and this does not take into account how much each error is related to the total body mass.For example, a standard error of 1.5 kg in a female population with a mean FFM equal to 50 kg is less precise than in men with a mean FFM equal to 80 kg.Fourth, even though some variables are not directly estimated by the BIA (e.g., FM and ICW), these can be easily derived as the difference between the body mass and the FFM (for FM) or the total water and the ECW (for ICW).Lastly, the utilization of multicomponent models to estimate body composition may be susceptible to error propagation, stemming from the use of multiple methods for reference body composition values.
A procedure for identifying the most valid equation for predicting body composition is illustrated in Fig. 7.There is reason to believe that such procedures will be considered in the development of AI-based software systems in the future.

Conclusions
Numerous predictive equations for quantifying body composition with BIA have been developed using highstandard procedures.However, upon categorizing these equations based on BIA technology, target population, and estimable body mass components, the overall availability of predictive equations seems limited.Notably, no predictive equations for hand to hand and leg to leg technology are included here, implying that these BIA devices predominantly employ procedures of low quality, often unknown, and lacking scrutiny, with equations typically owned by manufacturers.In contrast, there are several predictive equations available for foot to hand and segmental technology.However, the accuracy of estimating certain body components is compromised due to the absence of population-specific predictive equations.Given the necessity for reference values and BIA-based predictive equations to be population-and technologyspecific, the current findings underscore the urgency of developing new predictive equations tailored to specific BIA technology, population characteristics, and body mass components.Nevertheless, the foot to hand technology stands out with the highest number of predictive equations, offering the most accurate estimation of body composition.That said, the predictive equations available so far do not cover all possible combinations of technology, population and body mass component.Hence, the present manuscript may be helpful to pinpoint what has been done so far and what is currently lacking, so to generate novel predictive equations in cases not covered by the present literature.This may help to increase the trust of the practitioners in the BIA, often doubtful about the goodness of the results provided by inaccurate equations included in the device.Practitioners now possess an updated list of predictive equations for assessing body composition.

Fig. 1 A
Fig. 1 A schematization of the background and objectives set for the current study

Fig. 2
Fig. 2 PRISMA Flow chart of the studies' selection

Fig. 3
Fig. 3 World map with number of included studies for under 18 years old subjects (A), adults (B), athletes (C), elderly (D), and people with disease (E)

Fig. 4
Fig. 4 Number of predictive equations available for each BIA technology (e.g., hand to hand, leg to leg, foot to hand, and segmental) according to sex and population

Study selection and data processing
nents (i.e., FM, FFM, TBW, BCM, LSM, and SMM) using BIA.-Accessible in English in full text.

Table 1
Predictive equations (N = 19) from studies (N = 14) on under 18 years people from the general population Xc reactance, Z impedance, 4C 4 compartmental model BIS bioelectrical impedance spectroscopy, DXA dual-energy X-ray absorptiometry, FFM fat-free mass, H height (cm), LSM lean soft mass, MF-BIA multifrequency bioelectrical impedance analysis, R resistance, R 2 coefficient of determination, SEE standard error of estimation, SF-BIA single-frequency bioelectrical impedance analysis, TBW total body water, Wt body mass,

Table 2
Predictive equations (N

Table 4
Predictive equations (N