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


Diagnosis

A clinical model predicted bacteremia but not antibiotic-resistant uropathogens in patients with urinary tract infections

ACP J Club. 1993 May-June;118:89. doi:10.7326/ACPJC-1993-118-3-089


Source Citation

Leibovici L, Greenshtain S, Cohen O, Wysenbeek AJ. Toward improved empiric management of moderate to severe urinary tract infections. Arch Intern Med. 1992 Dec;152: 2481-6.


Abstract

Objective

To develop and validate clinical models that predict the likelihood of bacteremia and drug-resistant uropathogens in patients hospitalized for urinary tract infection (UTI).

Design

Cohort study of consecutive patients with UTI, with derivation and validation sets.

Setting

University medical center in Israel.

Patients

247 consecutive patients (median age 75 y) hospitalized with proven UTIs were used to derive the clinical models. 144 patients (median age 73 y) hospitalized because of suspected UTI were used to validate the models.

Description of tests and diagnostic standard

Medical history, physical examination, blood cell count, blood chemistry, and urinalysis results were all obtained within 24 hours of admission. The diagnostic standards were the results of urine and blood cultures for the resistant uropathogen and bacteremia models, respectively.

Main outcome measures

Predictors of bacteremia and drug-resistant uropathogens.

Main results

80 patients (32%) were bacteremic in the derivation set. On logistic regression analysis, 5 factors were independently associated with bacteremia: serum creatinine level, leukocyte count, temperature, low albumin level, and diabetes mellitus. The logistic regression model was used to divide the patients into 3 groups of increasing predicted probability of bacteremia (< 0.2, ≥ 0.2 and < 0.5, and ≥ 0.5). The actual prevalence of bacteremia within the 3 groups was 6%, 39%, and 69%, respectively. {At a cut-off level of ≥ 0.5, the sensitivity of the model was 49% (95% CI 38% to 61%), the specificity was 90% (CI 84% to 94%), the likelihood ratio for a positive result was 4.70 (CI 2.87 to 7.77), and the likelihood ratio for a negative test result was 0.57 (CI 0.44 to 0.69)}.* Of the 247 patients, 116 (47%) had a resistant microorganism. 3 factors—use of antibiotics before admission, advanced age, and male gender—were independently associated with a resistant strain. A regression model was used to divide patients into 2 groups. For patients predicted by the model to have < 0.5 chance of a resistant strain (or ≥ 0.5 chance), 9% (and 28%) had isolates resistant to cefuroxime, 7% (and 20%) had isolates resistant to gentamicin, and 30% (and 55%) had isolates resistant to sulfamethoxazole-trimethoprim. The accuracy of the bacteremia model was maintained in the validation set of patients, but was not maintained in the resistant uropathogen model.

Conclusion

A clinical model derived using serum creatinine level, leukocyte count, temperature, diabetes mellitus, and low serum albumin level predicted bacteremia in patients with suspected urinary tract infection.

Source of funding: W. Schreiber Foundation for Medical Research.

For article reprint: Dr. L. Leibovici, Department of Medicine B, Beilinson Medical Center, Petah Tiqva 49100, Israel. FAX 972-3-9376505.

*Calculated from data in article.


Commentary

This study by Leibovici and colleagues found a number of clinical features that helped to predict bacteremia in patients hospitalized for UTI. It seems unlikely, however, that you would stop to calculate the probability of bacteremia using the complex mathematical rules they derived, particularly when the answer might not be accurate for your patient or clinical setting. If so, of what use to the clinician is the prediction rule of this study? The immediate value is that the authors have identified the clinical variables that independently increased the chances that their patients would have bacteremia. Many possible signs and symptoms must be evaluated, and it is important to know which ones require attention. The authors found that an elevated serum creatinine level, elevated leukocyte count, fever, low albumin, and diabetes each increased the likelihood of bacteremia in their patients. They showed that these findings worked just as well in a second group of patients and that their results, therefore, were not simply chance associations in the original data. The authors also derived a model for predicting whether the organism cultured would be resistant to antibiotics, but that model was less useful because it was not a strong predictor and would not help select which antibiotic to start before the culture comes back.

The rule may not work as well in other settings. In addition, because the study population was heterogeneous, better rules for patient subsets such as women or elderly persons may be found. Meanwhile, learning which clinical features make bacteremia more likely helps us to make decisions about patients hospitalized with UTI.

Robert S. Wigton, MD
University of Nebraska Medical CenterOmaha, Nebraska, USA