Abstract

A survey of 208 elderly subjects living in four long term care institutions was undertaken over three months to identify nutritional and other variables that could be used to predict mortality during the subsequent three months. There were 58 men (mean age (SD) 75.6 (9.6) years) and 150 women (79.5 (8.4) years). Twenty nine subjects died (12 men and 17 women) within three months of completing the study. Twenty eight out of 57 variables differed significantly between those who died and those who survived. Subjects who died had lower systolic blood pressure, poorer intake of protein calories, lower concentrations of haemoglobin, plasma retinol, zinc, total cholesterol, and higher albumin adjusted plasma calcium concentrations. Stepwise regression analysis identified five variables that predicted mortality: plasma fructosamine; transferrin; glycosylated haemoglobin; prealbumin; and haemoglobin. The sensitivity, specificity, and predictive values of the discriminant function score using 0 as the demarcation between survivors and non-survivors were 75%, 97%, and 95%, respectively. This score could therefore be used to identify those most in need of nutritional support.