FACTORS ASSOCIATED WITH BACTEREMIA IN FEBRILE, GRANULOCYTOPENIC CANCER-PATIENTS

The objective of this investigation was to determine factors predictiv e of bacteraemia at presentation in febrile, granulocytopenic cancer p atients in order to estimate the probability of bacteraemia in each pa tient, and to compare factors associated with a diagnosis of gram-posi tive or gram-negative bacteraemia. Retrospective analysis of two sets of data (derivation and validation sets) randomly obtained from a larg e prospective study was conducted in a multicentre study of febrile, g ranulocytopenic cancer patients admitted for empiric antibacterial the rapy. Within the derivation set, prognostic factors (clinical and labo ratory data) likely to be associated with a generic diagnosis of bacte raemia and with a specific diagnosis of gram-positive or gram-negative bacteraemia were analysed by means of three backward, stepwise, logis tic regression analyses. The predictive probability of bacteraemia was calculated using the logistic equation. The discriminating ability of the model in predicting bacteraemia was evaluated in the derivation a nd validation sets using receiver-operating characteristic curves. The predictive probability of gram-positive or gram-negative bacteraemia was not calculated. In the derivation set, 157 of 558 episodes (28%) w ere microbiologically documented bacteraemias. Predicting factors were antifungal prophylaxis, duration of granulocytopenia before fever, pl atelet count, highest fever, shock and presence and location of initia l signs of infection. The variables institution, antibacterial prophyl axis and underlying disease showed borderline associations with bacter aemia. Shock was associated with gram-negative bacteraemia, while sign s of infection at catheter site were predictive of gram-positive bacte raemia. Quinolone prophylaxis was negatively associated with gram-nega tive bacteraemia. When tested in the validation set, the model was poo rly predictive, although a small subgroup of episodes (representing on ly 16% of the total sample size) with low risk of bacteraemia was iden tified. Factors predictive of bacteraemia can be identified, with disc rimination between gram-positive and gram-negative aetiology. Further studies are warranted in order to improve the discriminant ability of the model.