A Unified Framework for Predicting Bacteremia and Interpreting Blood Culture Results

The Problem

Bacteremia is common, with an incidence of approximately 1% in hospitalized patients; it is also deadly, with mortality rates as high as 37% [1]. The gold standard for diagnosing bacteremia remains blood cultures. Yet this status-quo has at least three significant limitations. First, bacteremia may go undetected if patients at high risk are not identified. For example, central line-associated bloodstream infection (CLABSI) is an important cause of bacteremia, which carries a high mortality and morbidity, and is an NHSN reportable condition directly tied to hospital quality metrics and CMS reimbursement. According to national safety organizations, DUHS is rated as below average with regards to CLABSI [2]. Second, inappropriate test utilization can cause delays in hospital discharge, as well as false positive and false negative results [3]. In July 2019, just 4.75% (n=285) of 6004 blood cultures from DUHS yielded a new first positive result while the cost of performing these blood cultures and related rapid identification and susceptibility tests likely exceeded $500,000 for just one month. Similarly, a retrospective assessment of blood culture use in a single DUH ICU in 2017 (7 West Cardio-Thoracic ICU), showed that 255 of 2106 blood cultures collected were positive (12%) during 1230 encounters (3.9%). Of these, most positive cultures were false positives and 24 were CLABSIs. These data suggest that blood cultures were routinely ordered when the likelihood of true bacteremia and/or central line-associated bloodstream infection was low. Moreover, investigators at Johns Hopkins recently reported that use of a manual checklist to systematize blood culture ordering in hospitalized children resulted in a 46% reduction in the rate of blood culture collections with no increase in in-hospital morbidity or readmission [4]. Finally, difficulties in identifying false positive and false negative results can lead to misdiagnosis, unnecessary antibiotic treatment, prolonged hospitalization and poor performance on hospital quality metrics [5]. It has been estimated that 4% of blood cultures are falsely positive and that patients with a false-positive blood culture have a mean increase length of stay of 2.35 days and an increase of over 50% in hospital costs [6].

While these three challenges are typically studied in isolation, they can all be framed as resulting from a lack of tools for assessing the pre-test and post-test probability of bacteremia. The framework of pre-test/post-test probability has proven useful in a variety of clinical settings, often echoing clinicians natural thought-processes [7]. This framework relies on three things: (1) an estimated probability of disease (bacteremia) prior to observing test results (blood cultures) — the pretest probability; (2) a model reflecting the likelihood of disease given the observed test results — the likelihood model; (3) a Bayes formula for integrating the pre-test probability and likelihood model to calculate the overall probability of a patient truly having a disease given the observed test results. A clinical decision support tool that integrates these concepts can be used to identify patients with in-place central venous catheters who have a high pre-test probability of bacteremia. Similarly, this tool can improve test utilization by reducing blood cultures that provide minimal updating of the pre-test probability of bacteremia. Lastly, by combining the pre-test probability of disease with a likelihood model of blood culture results, we will calculate a post-test probability of true bacteremia to help front-line clinicians interpret blood culture results.

Yet, appropriately quantifying the pre-test probability of bacteremia and likelihood of blood culture results is challenging due to the need for substantial personalization. For bacteremia, assessing the pre-test probability is exceedingly complex requiring physicians to consider a multitude of factors including vital signs, immune status, previous antibiotic use, intra-venous foreign body, other sites of infection and cultures, age, and gender. The likelihood of blood culture results is modified by many of the same factors plus others such as the number of blood cultures collected and number positive, the site of collections, the time-to-blood culture-positivity, and the identified organism. Furthermore, while many of these factors are known to be associated with both the pre-test probability of bacteremia and the likelihood of bacteremia given observed blood culture results, the relationship between these factors and quantities remains poorly understood.

Our Solution

References

1. Coburn, B., et al., Does this adult patient with suspected bacteremia require blood cultures? JAMA, 2012. 308(5): p. 502-11.
2. The Leapfrog Group. Leapfrog Hospital Safety Grade. 2019 2019]; Available from: https://www.hospitalsafetygrade.org/h/duke-university-hospital.
3. Tabriz, M.S., et al., Repeating blood cultures during hospital stay: practice pattern at a teaching hospital and a proposal for guidelines. Clin Microbiol Infect, 2004. 10(7): p. 624-7.
4. Woods-Hill, C.Z., et al., Association of a Clinical Practice Guideline With Blood Culture Use in Critically Ill Children. JAMA Pediatr, 2017. 171(2): p. 157-164.
5. Bates, D.W., L. Goldman, and T.H. Lee, Contaminant blood cultures and resource utilization. The true consequences of false-positive results. JAMA, 1991. 265(3): p. 365-9.
6. Geisler, B.P., et al., Model to evaluate the impact of hospital-based interventions targeting false-positive blood cultures on economic and clinical outcomes. J Hosp Infect, 2019. 102(4): p. 438-444.
7. Gill, C.J., L. Sabin, and C.H. Schmid, Why clinicians are natural bayesians. BMJ, 2005. 330(7499): p. 1080-3.