Introduction: Multiple culture reports from a patient during a hospital stay can inform us of new infections. However, it is also possible that the same infection is being monitored and repeated results could bias an analysis that has other aims than immediate patient care. Simple techniques such as dropping repeat positives are in common use but discards precious data. We present a method to weight successive positive cultures in the same patient without the drawback of complete information loss. The weighting approach is compared to two alternatives: an all-inclusive analysis and the more conservative approach of discarding data.

Methods: All final microbiology culture results from surgical intensive care patients in five isolation room beds over a 53 month period were collected and stratified by bed occupied. We considered two common cases: (1) repeated, duplicate findings within N days of one another and (2) a non-specific finding followed by a specific one. The following scheme was used to reduce the impact of repeat cultures, while retaining evidence of their existence :
1. Repeated findings of a particular organism, spanning D days, in which there was no time gap between findings longer than N days, were ordered chronologically.
2. For each sequence we assign a weight of 1 to the first observation. Each subsequent observation is weighted by a monotone function fw(d) that satisfies fw(0)=0 and fw(N)=1, where d is the number of days to the preceding observation. For the purpose of this study we used N=7 days, and a nonlinear fw(d)=(Exp(d2/N2)-1)/(Exp(1)-1). We then renormalize the weights so that they sum to (D+N)/N. If N=7 days and D=25 days then we allow a total of 4.57 observations within this period.

Results: Out of 904 unique patients, 397 had one or more positive cultures.

Conclusion: Whether or not repeat cultures are taken into account can lead to different interpretations of historical microbiology culture data. We have described a straightforward method to model the degree to which a subsequent positive culture represents an independent result, without discarding data. Our method leads to increased utility of clinical culture datasets for research and quality-improvement efforts.