Current issues of ACP Journal Club are published in Annals of Internal Medicine


Prediction of bacteremia in febrile patients

ACP J Club. 1992 Jan-Feb;116:17. doi:10.7326/ACPJC-1992-116-1-017

Source Citation

Leibovici L, Greenshtain S, Cohen O, Mor F, Wysenbeek AJ. Bacteremia in febrile patients. A clinical model for diagnosis. Arch Intern Med. 1991 Sep;151:1801-6.



To determine which clinical and laboratory features available within 24 hours of hospital admission predict bacteremia in febrile patients and to validate a clinical model that can be used for early diagnosis of bacteremia.


Cohort study of consecutive febrile patients, with both a derivation and validation set.


University medical center in Israel.


244 consecutive febrile patients (52% male; mean age, 71 years) admitted to an internal medicine ward provided data for a diagnostic model (derivation set). 257 additional patients (37% male; mean age, 71 years) were used to validate the model (validation set). Patients were included if they had had a febrile illness (oral temperature ≥ 38.0°C or rectal temperature ≥ 38.2°C) for < 2 weeks.

Description of tests and diagnostic standard

Within 24 hours of admission, the patient history, chest roentgenogram, blood and urinalysis results, and a clinician's prediction of bacteremia were obtained. The diagnostic standard was the subsequent isolation of a microorganism, not considered a contaminant, from a blood culture.

Main results

In the derivation set, 52 patients (21%) were bacteremic; mortality was higher for bacteremic patients (29% vs. 5%; P < 0.001). In a stepwise regression analysis, lowered albumin levels, presence of chills, low premorbid performance status, suspected urinary tract infection at admission, and renal failure (serum creatinine ≥ 177 mmol/L) were each independently associated with bacteremia (P < 0.03). The diagnostic model to predict bacteremia divided patients into 3 groups based on probabilities for bacteremia (group 1 < 20%; group 2, 20% to 80%; group 3 > 80%).

In the validation set, 36 patients (14%) were bacteremic and 6 (20%) died compared with 8 deaths (4%) in patients without bacteremia. By the predictive rule, group-1 patients had 1% bacteremia, group 2 had 23%, and group 3 had 65%. The physician prediction of bacteremia had a sensitivity of 53% and specificity of 85%. Application of the diagnostic rule improved physician identification of group-1 patients by 5% and of group-3 patients by 18%.


Lowered albumin levels, presence of chills, low previous functional status, suspected urinary tract infection at admission, and renal failure are clinical and laboratory features known within 24 hours of hospital admission that predicted bacteremia in febrile patients.

Source of funding: Not stated.

Address for article reprint: Dr. L. Leibovici, Department of Medicine B, Beilinson Medical Center, 49 100 Petah Tiqva, Israel.


Serious bacterial infections are not always accompanied by bacteremia. In sepsis, septic shock, and refractory septic shock, terms that have been proposed for classifying severe infections, blood cultures may be either positive or negative (1). In a recent study on the use of antiendotoxin antibody in the treatment of septic shock, investigators entered 543 patients who had a clinical diagnosis of septic shock, but only 200 were bacteremic (2). Interestingly, this remarkably expensive therapy could only be shown to be effective in those patients who had documented gram-negative bacillary bacteremia. It is likely, however, that further studies will show that antiendotoxin antibody and other immunotherapies (e.g., monoclonal antibodies against tumor necrosis factor) will be beneficial in nonbacteremic patients as well.

The abstracted study addressed a wider population: all those with a febrile disease. In this group of patients, however, it is probably more important to predict a treatable bacterial infection than bacteremia. In a study of febrile adults with no obvious focus of infection, Mellors and colleagues identified factors predictive of bacterial infection (3). Their study, like others, supported bedside clinical judgment (4).

The clinical model developed by Leibovici and colleagues can predict bacteremia, but it will probably be most useful in helping physicians decide more confidently which febrile patients' blood cultures are unlikely to be positive and may not need to be done.

Stephen R. Jones, MD
Good Samaritan Hospital & Medical Center Portland, Oregon