Predicting Mortality Based on Body Composition Analysis

Abstract

The role of the Nae/Ke ratio (the ratio of exchangeable sodium to exchangeable potassium) was examined as a nutritional marker in surgical patients in relation to anthropometrical and biochemical indexes by its ability to identify patients at risk for mortality after hospitalization. In 73 patients with sepsis and malnutrition (Training Group, Madrid) the following were determined: percentage of recent weight loss, triceps skin fold, midarm muscle circumference, serum albumin, serum transferrin, delayed hypersensitivity skin test response, total lymphocytes, and Nae/Ke ratio by multiple isotope dilution. The predictive power of Nae/Ke ratio was so strong (F = 105.1; p less than 0.00001) that it displaced anthropometric, biochemical, and immunologic variables from the linear equation derived from stepwise discriminant analysis using hospital mortality as the dependent variable. A theoretical curve of expected deaths was developed, based on an equation obtained by logistic regression analysis: Pr/death/ = 1/(1 + e[11.8-5.2 Nae/Ke]). Pre- and post-test probabilities on that curve allowed us to determine two cut-off values, Nae/Ke ratios of 1.5 and 2.5, which were markers for nonrisk and mortality, respectively. The model was tested in a heterogeneous data base of surgical patients (n = 417) in another hospital (Validation Group, Montreal). For patients exhibiting an abnormal Nae/Ke ratio (greater than 1.2) and a greater than 10% of probability of death, 54 deaths were expected and 53 observed (X2 = 1.8 NS). Two tests confirmed the basic agreement between the model and its performance, a G statistic of -0.704 and the area beneath the "receiver-operating-characteristic" (ROC) curve (Az = 0.904 + 0.0516 for the Madrid group vs. Az = 0.915 + 0.0349 for the Montreal group, NS). It was concluded from this analysis that, compared with the usual anthropometric measurements, the Nae/Ke ratio, if available, is the best method for identifying malnourished patients at risk of dying.