Meta-analysis: methods for diagnosing intravascular device-related bloodstream infection

No consensus exists on the best methods for diagnosis of intravascular device (IVD)-related bloodstream infection.

To identify the most accurate methods for diagnosis of IVD-related bloodstream infection.

51 English-language studies published from 1966 to 31 July 2004.

Studies of diagnostic tests for IVD-related bloodstream infection that described a reference standard and provided sufficient data to calculate sensitivity and specificity.

Study quality, diagnostic tests examined, patient characteristics, prevalence, sensitivity, and specificity.

Pooled sensitivity and specificity were calculated for 8 diagnostic methods. Summary measures of accuracy were Q* (the upper leftmost point on the summary receiver-operating characteristic curve) and mean D (a log odds ratio). Subgroup analyses were used to assess heterogeneity. Overall, the most accurate test was paired quantitative blood culture (Q* = 0.94 [95% CI, 0.88 to 1.0]), followed by IVD-drawn quantitative [corrected] blood culture (Q* = 0.89 [CI, 0.79 to 0.99]) and the acridine orange leukocyte cytospin test (Q* = 0.89 [CI, 0.79 to 0.91]). The most accurate catheter segment culture test was quantitative culture (Q* = 0.87 [CI, 0.81 to 0.93]), followed by semi-quantitative culture (Q* = 0.84 [CI, 0.80 to 0.88]). Significant heterogeneity in pooled sensitivity and specificity was observed across all test categories.

The limited number of studies of some of the diagnostic methods precludes precise estimates of accuracy.

Paired quantitative blood culture is the most accurate test for diagnosis of IVD-related bloodstream infection. However, most other methods studied showed acceptable sensitivity and specificity (both >0.75) and negative predictive value (>99%). The positive predictive value of all tests increased greatly with high pretest clinical probability. Catheters should not be cultured routinely but rather only if IVD-related bloodstream infection is suspected clinically.