Assessment of body composition: Intrinsic methodological limitations and statistical pitfalls

Evaluation of body composition (BC) is crucial for an adequate assessment of nutritional status and its alterations, to ensure the optimal tailoring of nutritional therapies during several pathologic conditions. The need for feasible and reliable methods for BC measurement, which could be applied either in healthcare across the lifespan as well as in clinical research and epidemiologic studies, has led to the development of various techniques. Unfortunately, they have not always produced equivalent results due to the fact that they are based on completely different principles or suffer intrinsic biases related to specific conditions. Furthermore, different population and clinical settings (ethnicity, age, type of disease) may interfere, thereby leading to dissimilar results. Finally, the need to compare the data obtained by new techniques to a reference standard has produced a further bias, due to a systematic misinterpretation of the statistical methods in the attempt to correlate the various techniques. In this context, the most used statistical methods for the comparison between different techniques have been Pearson's correlation test, the more recent intraclass correlation coefficient, Lin's concordance correlation coefficient method, and the Bland–Altman analysis. The aim of this review was to offer a summary of the methods that are mostly used in clinical practice to measure BC with the intent to give appropriate suggestions when statistical methods are used to interpret data, and underline pitfalls and limitations.