Volume 10, Number 9—September 2004
Research
Computer Algorithms To Detect Bloodstream Infections
, Brandon M. Zagorski†, Jerome I. Tokars*, Michael O. Vernon†, Sharon F. Welbel†‡§, Mary F. Wisniewski†‡, Chesley Richards*, and Robert A. Weinstein†‡§
Table 3
Comparing alternative methods for determining if positive blood cultures represented a hospital-acquired, primary, central-venous catheter–associated bloodstream infection, Cook County and Provident Hospitals, Chicago, Illinoisa
| Method | % sensitivity | % specificity | % PVP | % PVN | κ (95% CI) |
|---|---|---|---|---|---|
| Cook County Hospital (n = 104) | |||||
| Investigator review (reference method) | – | – | – | – | – |
| Infection control professional review | 67 | 75 | 62 | 79 | 0.41 (0.24–0.59) |
| Positive blood culture + CVC determinationb | 100 | 55 | 57 | 100 | 0.48 (0.35–0.62) |
| Worst computer algorithm (rules A, B1, C1, D)c | 72 | 74 | 62 | 81 | 0.44 (0.27–0.62) |
| Best computer algorithm (rules A, B2, C2, D)d | 79 | 72 | 63 | 85 | 0.49 (0.33–0.66) |
| Computer algorithm + CVC determinationb | 79 | 88 | 79 | 88 | 0.67 (0.52–0.82)e |
| Provident Hospital (n = 31) | |||||
| Investigator review (reference method) | – | – | – | – | – |
| Infection control professional review | 56 | 68 | 42 | 79 | 0.22 (–0.13–0.56) |
| Positive blood culture + CVC determinationb | 100 | 59 | 50 | 100 | 0.46 (0.20–0.70) |
| Worst computer algorithm (rules A, B1, C1, D)c | 78 | 64 | 53 | 88 | 0.35 (0.04–0.65) |
| Best computer algorithm (rules A, B2, C2, D)d | 89 | 68 | 53 | 94 | 0.48 (0.19–0.76) |
| Computer algorithm + CVC determinationb | 89 | 95 | 89 | 95 | 0.84 (0.63–1.0)e |
| Summary for both hospitals (n = 135) | |||||
| Investigator review (reference method) | – | – | – | – | – |
| Infection control professional review | 65 | 74 | 57 | 79 | 0.37 (0.21–0.53)e |
| Positive blood culture + CVC determinationb | 100 | 56 | 56 | 100 | 0.48 (0.36–0.60) |
| Worst computer algorithm (rules A, B1, C1, D)c | 72 | 74 | 62 | 81 | 0.42 (0.27–0.57) |
| Best computer algorithm (rules A, B2, C2, D)d | 81 | 72 | 62 | 87 | 0.49 (0.35–0.63) |
| Computer algorithm + CVC determinationb | 81 | 90 | 81 | 90 | 0.73 (0.61–0.85)e |
aPVP, predictive value positive; PVN, predictive value negative, CI, confidence interval.
bPresence of a central-venous catheter (CVC) determined by investigator medical record review.
cThe computer algorithm with the worst agreement, which used only microbiology data for the determination of infection vs. contaminant (rule B1, Table 1), and an abbreviated time period for the determination of primary vs. secondary (rule C1, Table 1).
dThe computer algorithm with the best agreement, which used the microbiology and pharmacy data for the determination of infection vs. contaminant (rule B2, Table 1), and the entire length of stay for the determination of primary vs. secondary (rule C2, Table 1).
eAgreement between investigator review and the best performing computer algorithm plus CVC determination was significantly better than between investigator and infection control professional reviews, i.e., p value < 0.05.