In a previous blog we explored how different techniques measure body composition. Here we are going to look at how valid and reliable these measures actually are. It is worth noting that the main limitation of all body composition assessments is that they are based on assumptions. The only truly accurate way to assess body composition is cadaver analysis (i.e. dissection). In this article we will focus on the 3 most used methods to measure body composition: Dual X-ray absorptiometry (DXA), skinfolds and bio electrical impedance measurements (BIA). Air displacement measurements (Bodpod is only mentioned in the infographic).
Methodological issues of measuring body composition
The measurement of body composition can be influenced by many factors including prior physical activity, the technician, fasted/fed state, hydration status etc. It is therefore very important measurements are standardised as much as possible. This will improve the accuracy of measures, as well as the reliability if you are taking repeated measures over time. This depends upon the method used, but some standardisation techniques that are easy to employ are:
Perform in the morning, before the athlete has eaten anything or performed any exercise
If you are not performing the test in the morning, then make sure you always take the measurements at the same time of day
Ensure the athlete is well hydrated
Use the same machine/technician when you are taking repeated measurements over time (3-9% variability can be attributed to differences between investigators)
How valid and reliable are the different measures?
Whilst there has been significant progress in the techniques used to estimate body composition, a gold standard method does not exist. No techniques have an accuracy better than 1% (1) and this means that, at best, we can estimate body fat to within 700 grams in a 70kg person and in reality, it is usually closer to 1000 grams. This means that a percentage body fat of 15% could really be anything between 9.5kg (20.9lbs) and 11.5kg (25.4lbs) for a 70kg (164lbs) athlete, and body fat would thus be between 13.6% and 16.4%: obviously a wide range. For a 50kg person, this range would be even greater: 13.0% and 17.0%. This is hardly ever taken into account when interpreting the data and we are thus assuming and working with a pseudo accuracy. This is even worse if the measurements are variable from day to day (poor reproducibility or reliability). Whilst measuring more compartments may provide more accurate estimates, they require multiple body composition techniques to be used to measure each of the compartments. This increases expense and time, making it unpractical to be used in a sport setting.
No techniques have an accuracy better than 1%
DEXA, skinfolds and BIA are the most widely used techniques in a practical (i.e. athlete) context. And so, we will now compare those methods in further detail…
Whilst DEXA is considered the laboratory reference method, it still has errors within the measurement. Caution should be taken when repeating DEXA measurements, and how often these are made, because it will not pick up on small changes in body composition. The estimated error for prediction of body fat % is between 2-3% (1). An IOC consensus regarding body composition methods concludes for DEXA that whilst it is relatively precise for whole-body estimates of body composition, it is less reliable in producing accurate fat estimates of lean athletes (1). The assessment of total and regional fat free mass (FFM) is generally acceptable if total scanned mass equates to scale mass.
Over 100 equations have been created to predict body fat % from skinfold equations, however this introduces an additional level of error. Consequently, a wide range of estimates of body fat % can be produced from one set of skinfold measurements. Using an equation specific to the population you are working with improves the validity of the prediction, however it is still only an estimate. When generalised equations are used, the error can be up to ~5%. Skinfold measures are often used in their raw value (millimetres skinfold), as opposed to converting to body fat % using an equation. By doing this, we accept that translating the number to body fat is too inaccurate, but we can still track changes in skinfold thickness over time within one person.
Skinfold measures can be looked at individually at each site, or values can be summed together (e.g. sum of skinfold thickness at 7-sites), which allows changes in total and individual sites to be tracked over time. In terms of reliability, if the same measurer is used each time and a standard protocol is adhered by, the results can be reliable. Adequate training, and experience is essential to get reliable results.
BIA has poor accuracy in estimating body fatness or body water content, with studies reporting large errors in estimates of up to ~5%. If there is a ±5% error from a ‘true’ body fat value of 8%, then body fat will sit somewhere in the region of 3-11%. This is obviously a very large range. Estimates are largely influenced by hydration status, because the scales measure the resistance to the electrical current (better hydrated = less resistant). If measures are taken repeatedly over time, then changes in estimates may be due to a change in hydration status as opposed to a change in body composition. Similar to skinfolds, BIA scales use equations to predict fat mass and fat free mass. The accuracy of the estimations will be improved by matching the population used to create the equation with the athlete being measured. Many modern scales will ask to enter details of the type of athlete on the scales, but if this sort of input is required to get more accurate numbers, one should question the underlying measurements.
How long between measurements to see a change?
Measurements are often taken regularly and frequently (daily or weekly). This helps to see trends over time. Even methods that are less reliable and less accurate can be used to see such trends. However, due to the errors within measures, a significant amount of time (around 2-3 months) should be left in between measurements. It takes time to respond to interventions and to see significant changes in body composition. Drawing conclusions when measuring too frequently and looking at the individual values is dangerous. Due to the inherent errors within measurements, especially if a single body composition method is used (and not a multi-compartment model), a specific value should not be used as a cut-off/threshold for selection. For example if body composition is measured once a month and one measurement is suddenly higher, it would be wrong to conclude that body composition is changing. It might be, or it might not be.
So… which method should I use?
The most appropriate method for the specific context should be used. It is about getting a balance between accuracy, reliability and practicality. If you have an understanding of the reason you are assessing body composition, then this will help when deciding which technique to use. If you are taking a one-time measurement, and really want to get an absolute number for body composition, then you will want a more valid measure that you know will give accurate results (in this case, you may want to use a multi-compartment model). A DEXA measurement could also do the job, although even a DEXA measurement would have to be interpreted with caution. However, if you are monitoring body composition and taking repeated measures over time, then you will want a more reliable measure that is sensitive to change between measures (for example DEXA or skinfolds). You may not be able to say anything about the absolute values for body composition but you will be able to look at trend lines. Conclusions based on one or two measurements should be avoided with these methods.
Ackland TR, Lohman TG, Sundgot-Borgen J, et al. Current status of body composition assessment in sport: review and position statement on behalf of the ad hoc research working group on body composition health and performance, under the auspices of the I.O.C. Medical Commission. Sports Med. 2012;42(3):227-249.