Predictive index for optimizing empiric treatment of gram-negative bacteremia.

In a survey of 296 episodes of gram-negative bacteremia in 286 patients (aged 13-99 years), four clinical variables were found to predict both significantly and independently the subsequent isolation of a multiresistant strain; hospital acquisition of the infection, antibiotic treatment before the bacteremic episode, endotracheal intubation, and thermal trauma as the cause of hospitalization. These variables were combined in an index that served to classify the patients into four groups with an increasing prevalence of multiresistant strains, Pseudomonas isolates, and isolates resistant to each of the antibiotic drugs in common use. For example, the percentage of isolates susceptible to cefuroxime in the four groups were 79%, 56%, 34% and 25%, and to gentamicin, 89%, 79%, 46%, and 33% (P less than .001 for both comparisons). The performance of the index was validated in a second group of 144 episodes of gram-negative bacteremia. The index kept its discriminative power. Compared with the prescriptions of the attending physicians, the index could probably have improved empiric antibiotic treatment in 24% of patients.