Predictive ability of various nutritional variables for mortality in elderly people

Nutritional indices (percentage ideal body weight [IBW], serum albumin, serum transferrin, total lymphocyte count [TLC] and delayed cutaneous hypersensitivity [DH] response) were assessed in 80 consecutive patients (aged 85-100 y) within 24 h of admission to determine their predictive value for mortality. Nine patients died. Pearson correlation analysis demonstrated that death was significantly (p less than 0.05 to less than 0.01) associated with sepsis, serum albumin less than 30 g/L, TLC less than or equal to 1500 cells/mm3, and percentage IBW less than or equal to 90%. However, when serum albumin was controlled for, logit regression analyses demonstrated that the impact of other nutritional indices on death was insignificant. The effect of serum albumin remained significant (p less than 0.05 to less than 0.01) even when age and physician's diagnosis were held constant. With the logit model, serum albumin greater than or equal to 30 g/L had a sensitivity of 0.33, specificity of 0.99, and overall predictive power of 0.91. Serum albumin is thus the simplest and best single predictor of mortality and can provide early identification of elderly people at increased risk of death.