Estimating the composition of tissue gained or lost from measurements of elementary body composition

Studies of relationships between changes in body constituents resulting from disease or treatment can, in principle, give information about the composition of the tissues gained or lost. Frequently, however, the random errors of measurement are of similar magnitude to the changes measured, so that standard statistical methods such as linear regression analysis give biased estimates of the parameters in the relationships. Possible ways of obtaining better estimates are discussed and models suitable for two types of experiment are described. In the first, patients are measured before and after the changes; in the second, patients are measured only after the changes have occurred and their results compared with those for a group of control subjects assumed to be representative of the patients before they became ill. The models have been used to analyse losses of total body nitrogen and potassium as a result of surgery and of disease. The results suggest that patients who underwent surgical operation lost 1.7 mmol K per g N. Those who suffered from disease lost 2.0 mmol K per g N, but also lost, on average, 234 mmol K independently of N. Conventional linear regression analysis would have led to very different conclusions, emphasising the importance of applying appropriate models to the analysis of such data.