A true positive blood culture in hospitalized patients can be differentiated from a contaminant by type of organism, presence of multiple positive cultures, and clinical risk score
ACP J Club. 1992 Sept-Oct;117:51. doi:10.7326/ACPJC-1992-117-2-051
Bates DW, Lee TH. Rapid classification of positive blood cultures. Prospective validation of a multivariate algorithm. JAMA. 1992 Apr 8;267:1962-6.
To develop and validate a decision rule predicting whether a positive blood culture represents a true positive or a contaminant in hospitalized patients.
Cohort study, with independent derivation and validation cohorts.
Urban tertiary care hospital in Boston.
219 inpatients who had ≥ 1 positive blood culture provided data for derivation of a predictive model, and a second group of 129 patients with a positive blood culture provided data for validation of the model.
Description of test and diagnostic standard
Potential predictors included shaking chills, hypotension, evidence of intravenous drug abuse, evidence of an acute abdomen, major comorbidity, maximum temperature ≥ 38.3°C, evidence of a rapidly or ultimately fatal disease, leukocyte count, bands, clinical risk score class, organism type, days until the blood culture became positive, and number of positive cultures. The clinical risk score class divided patients into 4 risk groups based on the presence of clinical predictors derived from a previous study (body temperature ≥ 38.3oC, fatal disease, shaking chills, intravenous drug use, acute abdomen, and major comorbidity). The diagnostic standard was the determination of true bacteremia by independent reviewers blinded to the potential predictors, with any disagreements resolved by 3 infectious disease specialists.
In the derivation cohort, 115 (53%) of the episodes were classified as true positives. In a multivariate analysis, organism type (odds ratio [OR] for the 3 highest catagories of probable bacteremia 21.9, CI 5% CI 3.3 to 145; 40.5, CI 8 to 205; and 465, CI 37 to 5861), days until the blood culture became positive (with ≥ 4 days as the lowest risk group [OR 2.2, CI 1.3 to 3.5]), multiple positive cultures (OR 19.0, CI 4.6 to 77), and clinical risk score class (OR 1.5, CI 1.0 to 2.3) were each independently associated with bacteremia. The derived model divided patients into 4 risk groups with 8% (6 of 71), 20% (7 of 35), 67% (16 of 24), and 97% (86 of 89) true-positive blood cultures. In the validation cohort, the corresponding percentages of true-positive blood cultures were 14% (8 of 59), 19% (3 of 16), 70% (7 of 10), and 89% (39 of 44). The area under the receiver operating characteristic curve was 0.93 for the derivation set and 0.86 for the derivation set.
Type of organism, days until the first blood culture became positive, presence of multiple positive cultures, and clinical risk score class were independent predictors of bacteremia in hospitalized patients.
Sources of funding: Agency for Health Care Policy and Research and the American Heart Association.
Address for article reprint: Dr. D.W. Bates, Division of General Medicine, Department of Medicine, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115, USA.
You are on morning rounds when a call is received from the bacteriology laboratory. Your patient, Mr. S, has a positive blood culture: gram-positive cocci (probably staphylococci) in 1 of 3 samples. Last night Mr. S presented with fever and chills. He has a history of intravenous drug use but states that he has not "shot up" for the past 6 months. You withheld antibiotics overnight to obtain blood cultures. You now have some decisions to make. You know that staphylococcus is a common cause of bacteremia but that it is also a common blood culture contaminant. Should you start treatment on the basis of this preliminary report? With what? For how long?
Bates and Lee propose a practical model for analyzing such data. In developing the model, they used true bacteremia, as determined by independent observers, as the gold standard. Their model incorporates the clinical questions that define the prior probabilities, including a previously validated clinical risk score and key bacteriologic data (number of positive cultures, length of time for cultures to become positive, type of organism). Any clinician with access to the Bates algorithm and the requisite clinical information can make a reasonable preliminary decision without depending on consultants.
Because early treatment decisions often remain in effect for the duration of a therapeutic course (often despite the emergence of data indicating alternative treatments), such prediction rules help optimize antibiotic therapy (1, 2).
Thomas A. Parrino, MD
Providence Veterans Affairs Medical CenterProvidence, Rhode Island, USA