2.4. Interpretation of bacterial culture results
Positive culture was defined as growth of
≥1 colony of
the same species in both studies. The second of the duplicate
samples that were collected in the Dutch study was
ignored in order to have the same definition of positive culture
in both studies. In both datasets, growth of
≥3 types
of bacteria in a sample was considered to be attributable to
contamination and these results were excluded from analysis.
Culture results were classified as a binary variable:
culture of a major pathogen (S. aureus, non-staphylococcus
gram-positive cocci, gram-negatives and yeasts) or no culture
of a major pathogen (culture-negative samples or
samples infected with coagulase-negative staphylococci,
Corynebacterium bovis and Bacillus species).
2.5. Statistical analysis
2.5.1. Risk factor model
The effect of the possible RF on the odds of IMI caused
by a major pathogen was estimated with logistic regression
models. Because bacteriological culture and SCC are both
imperfect tests for diagnosing IMI with a major pathogen
(Koop et al., 2011), we used a combination of culture and
SCC as the dependent variable for the logistic regression
model. This was done in a Bayesian setting using Open-
BUGS, version 3.0.8 (Thomas et al., 2006). The 4 possible
outcomes of the joint test results (BC+SCC+, BC+SCC−, BCSCC+,
and BC-SCC−) were modelled as a 4-state discrete
distribution and were linked to the latent IMI status in
terms of Se and Sp: