Twenty-eight articles were included, of which 22 were statistically combined. The sample sizes were not stated.
Qualitative catheter segment culture (5 trials): the pooled sensitivity was 95% (39 out of 41 participants)and the specificity 75% (652 out of 867 participants); P-values for homogeneity were 0.07 (sensitivity) and <0.001 (specificity). In the ROC analysis the mean D-value was 3.30. The mean Youden index was 0.65.
Semi-quantitative catheter segment cultures (13 trials): the pooled sensitivity was 85% (109 out of 128 participants) and the specificity 85% (2,688 out of 3,152 participants); P-values for homogeneity were <0.001 (sensitivity) and <0.001 (specificity). In the ROC analysis the mean D-value was 3.61. The mean Youden index was 0.70.
Quantitative catheter segment cultures (5 trials): the pooled sensitivity was 94% (65 out of 69 participants) and the specificity 92% (620 out of 764 participants); P-values for homogeneity were 0.123 (sensitivity) and <0.03 (specificity). In the ROC analysis the mean D-value was 4.86. The mean Youden index was 0.85.
Unpaired qualitative catheter blood culture method (6 trials): the pooled sensitivity was 91% (54 out of 59 participants) and the specificity 86% (569 out of 660 participants); P-values for homogeneity were <0.001 (sensitivity) and <0.001 (specificity). In the ROC analysis the mean D-value was 3.86. The mean Youden index was 0.70.
Unpaired quantitative catheter blood culture method (5 trials): the pooled sensitivity was 78% (40 out of 51 participants) and the specificity 96% (395 out of 412 participants); P-values for homogeneity were 0.02 (sensitivity) and <0.001 (specificity). In the ROC analysis the mean D-value was 4.41. The mean Youden index was 0.67.
Paired quantitative blood culture methods (4 trials): the pooled sensitivity was 79% (58 out of 73 participants) and the specificity 94% (160 out of 170 participants); P-values for homogeneity were <0.001 (sensitivity) and <0.001 (specificity). In the ROC analysis the mean D-value was 4.98. The mean Youden index was 0.74.
The Karduan-Karduan test of homogeneity indicated a possible lack of fit of the summary ROC curve for unpaired quantitative blood culture (P=0.06) only; P-values for the other methods exceeded 0.03, indicating no detectable heterogeneity. A linear regression detected a significant increase in accuracy of catheter segment cultures with increasing quantitation (P=0.03), but not with blood culture methods (P=0.40).
For the 5 direct staining methods, the median sensitivity was 87% (range: 60 to 100) and the median specificity 88% (range: 64 to 94).