To explore the usefulness of Bioelectrical Impedance Analysis (BIA) for general use by identifying best-evidenced formulae to calculate lean and fat mass, comparing these to historical gold standard data and comparing these results with machine-generated output. In addition, we explored how to best to adjust lean and fat estimates for height and how these overlapped with body mass index (BMI).

Cross-sectional observational study within population representative cohort study.

Urban community, North East England

Sample of 506 mothers of children aged 7–8 years, mean age 36.3 years.

Participants were measured at a home visit using a portable height measure and leg-to-leg BIA machine (Tanita TBF-300MA).

Height, weight, bioelectrical impedance (BIA).

Lean and fat mass calculated using best-evidenced published formulae as well as machine-calculated lean and fat mass data.

Estimates of lean mass were similar to historical results using gold standard methods. When compared with the machine-generated values, there were wide limits of agreement for fat mass and a large relative bias for lean that varied with size. Lean and fat residuals adjusted for height differed little from indices of lean (or fat)/height^{2}. Of 112 women with BMI >30 kg/m^{2}, 100 (91%) also had high fat, but of the 16 with low BMI (<19 kg/m^{2}) only 5 (31%) also had low fat.

Lean and fat mass calculated from BIA using published formulae produces plausible values and demonstrate good concordance between high BMI and high fat, but these differ substantially from the machine-generated values. Bioelectrical impedance can supply a robust and useful field measure of body composition, so long as the machine-generated output is not used.

Population-based cohort, but female only and restricted to parents.

Explicit, well-evidenced computational approach.

Validated against published gold standard data.

Compared with widely used commercial methods.

The WHO defines obesity as “the disease in which excess body fat has accumulated to such an extent that health may be adversely affected”.^{2}) is only an indirect measure of fatness, so reliable methods of assessing body composition are also needed. Hydrodensitometry is usually regarded as the nearest to a gold standard,^{2}/Z, lean mass (LM) and total body water (TBW) can then be estimated, from which fat mass (FM) can be calculated.

Although BIA is already widely used in practice and some body composition research, there remain doubts about its accuracy and precision.

A further problem is that lean and FM values are difficult to interpret in isolation, as they differ systematically depending on the participant’s height,

Finally, it is widely believed in the lay population that BMI is a poor predictor of actually fatness. Published information on this suggests generally that BMI has high specificity, but low sensitivity to identify high %fat,

We thus set out to:

Identify best-evidenced formulae to calculate lean and FM and compare these to historical gold standard data;

Compare these results with machine-generated output;

Explore how to best adjust estimates for height and how these overlap with BMI.

The impedance data were obtained from mothers of participants in the Gateshead Millennium Study (GMS).

The data were collected on the children's parents at a home visit. While it was possible to study mothers at most of these visits, participation by fathers was minimal, so the paternal data were not used further. Impedance was measured using a single frequency (50 kHz) leg-to-leg BIA machine (Tanita TBF-300MA, Tokyo, Japan). The participants were measured wearing light clothing and bare feet after being asked to empty their bladders. The raw impedance and the machine calculated values for LM, FM and %fat were recorded. Height was measured without shoes and socks using a portable scale (Leicester height measure) to 0.1 cm with the head in the Frankfort plane. Weight was measured to 0.1 kg using the Tanita TBF-300MA. BMI was calculated as weight (kg)/height (m)^{2}.

The analysis was carried out using the software package R (V.2.2.0). We used the measured impedance to arrive at our own estimates of TBW and thus lean and FM using best published estimates of various constants. We assumed the hydration constant to be equal to 0.732 in adults, supported by previous studies^{2}/Z). Combining these two formulas, we obtained the following, simple prediction equation for adult women: LM=0.66/0.732 (height^{2}/Z) or LM=0.898 (height^{2}/Z). FM was then obtained as weight minus LM. To check whether the values we obtained for TBW, LM and FM using this approach were reasonable, we compared them to reference values from the two previous studies which had used gold standard measurement methods and published separate values for women.^{2}H_{2}O dilution, body density using underwater weighting and a three-component model to estimate %fat and LM. The second^{2}H_{2}O or ^{3}H_{2}O dilution and FM and LM using dual-energy X-ray absorptiometry (DXA).

In order to produce estimates of FM and LM adjusted for height, a regression method to obtain lean and fat residuals for children^{2}).

The Bland-Altman method

When the cohort was formed in 1999–2000, 1009 (81%) eligible mothers agreed to join the study and impedance and growth data were collected on 498 mothers in 2007, with mean (SD) age 36.3 (5.6) years (age range 23.6–53.1 years). Sixteen (3.2%) women were underweight (BMI <19), 141 women (28%) were overweight (BMI 25–30) and 112 (22%) were obese (BMI >30).

Descriptive statistics for the anthropometric measurements are summarised in

Descriptive statistics for the anthropometric measurements

Median | IQR | |
---|---|---|

Age (years) | 36.63 | 32.25, 40.17 |

Height (cm) | 163.00 | 158.90, 167.40 |

Weight (kg) | 67.50 | 59.38, 78.20 |

Waist circumference (cm) | 80.40 | 74.20, 90.55 |

Hip circumference (cm) | 102.70 | 96.90, 111.30 |

BMI (kg/m^{2}) | 25.06 | 22.64, 29.32 |

Impedance (ohms) | 554 | 511, 600 |

Generated using published constants | Tanita-generated data | Mean difference | p Value* | |||
---|---|---|---|---|---|---|

Median | IQR | Median | IQR | |||

TBW (L) | 31.8 | 29.0, 34.9 | NA | NA | ||

LM (kg) | 43.4 | 39.6, 47.6 | 44.3 | 42.3, 47.4 | 1.19 | <0.001 |

FM (kg) | 24.3 | 18.0, 33.5 | 23.0 | 17.1, 31.0 | −1.20 | <0.001 |

%Fat | 36.0 | 30.4, 43.0 | 34.5 | 28.6, 39.9 | 1.97 | <0.001 |

*One sample t test.

%Fat, percentage fat; BMI, body mass index; FM, fat mass; LM, lean mass; NA, not available; TBW, total body water.

Our results are compared with the results from the two historical papers in

Values (only females) reported by Hewitt 1993 and Chumlea 2001 (means±SD) compared with GMS values

Study | Hewitt | Chumlea | GMS | Chumlea | GMS | Chumlea | GMS |
---|---|---|---|---|---|---|---|

Age(year) | 32.6±6.0 | 20–29 | 23–29 | 30–39 | 30–39 | 40–49 | 40–50 |

Year | 1993 | 2001 | 2007 | 2001 | 2007 | 2001 | 2007 |

N | 19 | 124 | 75 | 130 | 292 | 104 | 128 |

Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

Weight (kg) | 59.6 | 8.0 | 62.4 | 12.4 | 68.7 | 16.9 | 63.6 | 13.7 | 71.9 | 17.2 | 68.5 | 15.5 | 69.7 | 13.0 |

LM (kg) | 43.9 | 4.2 | 44.1 | 6.2 | 42.5 | 7.4 | 43.1 | 5.3 | 44.2 | 7.1 | 43.5 | 6.6 | 45.1 | 6.5 |

%Fat | 26.0 | 5.4 | 28.5 | 8.8 | 36.6 | 9.1 | 30.4 | 8.2 | 37.1 | 9.3 | 35.0 | 8.9 | 34.2 | 9.0 |

FM (kg) | * | * | 18.4 | 8.8 | 26.2 | 11.8 | 19.9 | 9.3 | 27.8 | 12.8 | 24.8 | 10.9 | 24.6 | 10.1 |

BMI | * | * | 22.6 | 4.2 | 26.0 | 5.9 | 23.4 | 4.8 | 27.1 | 6.3 | 25.2 | 5.5 | 26.0 | 4.6 |

*Not described in that paper.

%Fat, percentage fat; BMI, body mass index; FM, fat mass; GMS, Gateshead Millennium Study; LM, lean mass.

Using our method, 22 (4%) women had fat <20%, and 180 (36%) had fat >40%. A majority of women with BMI >30 kg/m^{2} (88, 79%) also had greater than 40% fat, but only a minority of women with BMI <19 kg/m^{2} (4, 25%) had less than 20% fat.

The machine-calculated values were also available for all but eight mothers. The sample mean of the Tanita LM values was lower than our calculated values (mean (SD) difference −1.19 (3.33) kg, 95% CI −1.49 to −0.90) while they were higher for FM and %fat (1.19 (3.33) kg, 95% CI 0.90 to 1.49 and 1.97 (4.86) %, 95% CI 1.54% to 2.40%, respectively). The two sets of results were compared using the Bland-Altman method,

Bland-Altman plots for (A) lean mass (LM), (B) fat mass (FM) and (C) percentage fat (%fat) comparing our own calculated values to the machine output values (Tanita).

In order to achieve approximate normality and constant variance of errors, LM was inverse-transformed before being regressed on height. The resulting equation was obtained:

Similarly, FM was log-transformed and regressed on height. The resulting equation was obtained:

These residuals were normally distributed with mean 0 and variance 1. Fourteen women (2.4%) had fat residuals <−2 SD (roughly the 2.5th centile for the normal distribution) and 124 (25%) had fat residuals >0.68 SD (roughly the 75th centile) as expected.

As would be expected, there was no association between height and the lean and fat residuals (Spearman correlation (95% CI) of height with lean residual −0.02 (−0.11 to 0.07); with fat residual 0.01 (−0.07 to 0.10)), but nor was there any significant correlation of height with the lean index (LM/height^{2}: −0.05 (−0.14 to 0.04)) or the fat index (FM/height^{2}: −0.03 (−0.12 to 0.06)). Of the 112 women with BMI >30 kg/m^{2}, 100 (91%) also had fat residuals >75th centile, while a BMI of >30 kg/m^{2} identified 81% of all with high fat residual. In contrast, of the 16 with BMI <19 kg/m^{2} only 5 (31%) also had fat residual <2nd centile (

Scatter plot of lean mass adjusted for height (lean mass residual) against fat mass adjusted for height (fat mass residual) per body mass index (BMI) category (underweight (<19) and obese (≥30). The vertical lines denote the cut-off for low (<2nd centile) and high (>75th centile) fat residual.

In this analysis, we set out first to identify the best published constants to use for estimating lean and FM from BIA. The use of different devices and methods, under different conditions and on different populations, can make it difficult to extrapolate formulas from one study to another, but when we compared our estimated values for FM, LM and TBW to historical data, these revealed that results for LM were similar, while in contrast there were striking increases in average fat for the youngest, though not in the oldest category, who were already relatively more adipose even in the earlier cohorts.

Although based on a simplified mathematical model of the human body’s shape and composition, BIA has been shown to be a reliable method in population studies, though likely to have less accuracy in individuals.^{2}/Z with no involvement of other variables.

There are limitations to the study. We were not able to directly compare the results to a gold standard method and had to rely instead on published data. However, we were able to show how similar our results were to these, when using this simple parsimonious computational approach. The age range of the women was relatively narrow, but while there are major changes in body composition, hydration and body proportions during infancy and again in old age,

The results we obtained were very different from those automatically produced by the Tanita machine. It is important to understand the distinction between these essential mathematical transformation and factors that then correlate with or influence LM and FM, such as weight, sex and age. Manufacturers may seek to include these other variables in their output to contextualise their final estimates of adiposity. However, this then is no longer the true estimate of actual LM for that individual, derived solely from the impedance reading.

The equations used by different manufacturers and for different models are not made generally available, but have been published for a machine similar to the one used in our study.^{2}/Z and that the relative contribution of impedance to the final value is tiny relative to that of weight. For example, within our study, a decrease in impedance by 1SD (75 Ω) changes the Tanita-generated FFM estimate by <1 g while an increase in weight by 1SD (16 kg) increases it by 10% (2.8 kg). Thus, the machine estimate of FFM at least is actually largely based on weight rather than impedance.

Our results are in substantial agreement with the findings of Jebb ^{7} that Tanita underestimates LM in adult women by between 1 and 2 kg on average, and adds the new conclusion that the relative bias in the Tanita estimate of LM varies with size. The size of the positive biases in FM and %fat obtained using Tanita, relative to our method, are very similar to the bias relative to the four-compartment model reported in Jebb ^{23} and our results also confirm poor agreement in individual cases.

Meanwhile, BIA technology has been moving on and there are now multifrequency devices and eight electrode techniques which aim to estimate different body segments and intracellular and extracellular fluids separately. However, the underpinning assumption and prediction formulae for these machines are likely to be even more complex and difficult to assess objectively.

We also considered the most robust way to adjust measures of fat and lean for height. A method that expresses lean and FM separately adjusted for height is much more informative than raw LM and FM estimates. We have shown previously in children that lean and fat residuals are effective in fully adjusting for height as well as allowing the data to be expressed as SD scores compared with a reference population.^{2} also fully adjusted for height, suggesting that this would be equally valid and simpler. This adds further weight to Well's proposal^{2}=(H^{2}/Z)/Ht^{2}=1/Z. Ideally, any reference should be validated against a more direct measure of body composition, but such studies seem only to have been done in children.

We have also shown that, as in children,^{2} showing 90% specific and 80% sensitivity for fat index above the internal 75th centile. This is generally a much better correspondence than was found in a systematic review of the use of various BMI thresholds to detect high %fat measured, using different methods.

In conclusion, these data demonstrate that using BIA in models with published constants produces estimates of LM that are, on average, very similar to earlier studies using more direct methods, while the larger FM values are entirely plausible given the secular trends in obesity. These suggest that the physical measurement of impedance can produce useful estimates when appropriately transformed. However, the machine-generated estimates are likely to vary between machines and manufacturers and usually do not only reflect the physical measurement of impedance. They cannot therefore be used to validate or verify other measures of adiposity such as BMI. We would recommend that researchers using BIA in future should not rely on machine-generated estimates and should instead express lean and fat indices, divided by height^{2} in order to adjust for height.

The authors are grateful for the participation and advice of all those involved with The Gateshead Millennium Study—the research team, the families and children who took part, the External Reference Group, Gateshead Health NHS Foundation Trust, Gateshead Education Authority and local schools.

Gateshead Millennium Study core team: Ashley Adamson, Anne Dale, Robert Drewett, Ann Le Couteur, Paul McArdle, Kathryn Parkinson, John J Reilly.

MSP and the Gateshead Millennium Study core team designed the research and supervised the data collection and data entry. MF-V analysed the data, performed the statistical analysis and initially drafted the paper. JHM and AS supervised the analysis and commented on successive drafts of the paper. CMW designed the research study, supervised the analysis, edited the paper and has primary responsibility for final content. All authors read and approved the final manuscript.

Gateshead Millennium Study was first established with funding from the Henry Smith Charity and Sport Aiding Research in Kids (SPARKS) and followed up with grants from Gateshead NHS Trust R&D, Northern and Yorkshire NHS R&D, and Northumberland, Tyne and Wear NHS Trust. This study wave was supported by a grant from the National Prevention Research Initiative (incorporating funding from British Heart Foundation; Cancer Research UK; Department of Health; Diabetes UK; Economic and Social Research Council; Food Standards Agency; Medical Research Council; Research and Development Office for the Northern Ireland Health and Social Services; Chief Scientist Office, Scottish Government Health Directorates; Welsh Assembly Government and World Cancer Research Fund).

None declared.

Gateshead Local Research Ethics Committee (LREC).

Not commissioned; externally peer reviewed.

No additional data are available.