503 Parallel Matched Cohort Design Has Advantages over a Standard Cohort Design when Estimating the Burden of Methicillin Resistant S. aureus Blood Stream Infections

Saturday, March 20, 2010
Grand Hall (Hyatt Regency Atlanta)
Marlieke E.A. de Kraker, MSc , National Institute for Public Health and the Environment (RIVM), Bilthoven, Netherlands
Peter Davey, Professor , Division of Community and Population Sciences and Education, Dundee, United Kingdom
Hajo Grundmann, Professor , National Institute for Public Health, Bilthoven, Netherlands

Background: To determine the burden of antimicrobial resistance most often patients with an infection caused by a resistant pathogen are compared to patients with the susceptible type of the pathogen (‘standard' cohort). However, these patient groups often have a very different clinical picture. It is also possible to construct two parallel cohorts, comparing patients with an infection caused by either the susceptible or the resistant pathogen to patients without the infection. In the end the outcomes of the parallel cohort studies can be compared. This design could reduce the residual bias present in the standard design.

Objective: In this study we determined the performance of a parallel matched cohort design versus a ‘standard' cohort design to establish the excess mortality related to methicillin resistance in S. aureus blood stream infections (BSI).

Methods: In the parallel matched cohort design patients with a S. aureus BSI were compared to patients without such a BSI (controls). Two parallel cohorts were constructed, comparing either methicillin resistant (MRSA) or methicillin susceptible S. aureus (MSSA) exposure. Matching was based on duration of admission prior to enrolment. In the standard cohort design patients with a BSI caused by MRSA were directly compared to patients with a BSI caused by MSSA. In thirteen European hospitals patients were routinely sampled and all patients above 18 years with a S. aureus BSI were included from July 2007 to July 2008. Multivariate logistic regression was used to determine the excess mortality 30 days after infection/enrolment.

Results: In the parallel matched cohort design patients with a BSI caused by MRSA had an increased risk of dying compared to controls without a S. aureus BSI (odds ratio (OR) 4.4, confidence interval (CI) 2.8-7.0). For MSSA patients a smaller effect was seen (OR 2.4, CI 1.7-3.3). Combining the two cohorts, the OR for mortality associated with methicillin resistance was 1.8 (CI 1.04-3.2). In the standard cohort the odds for dying for MRSA patients was increased by 1.2 (CI 0.85-1.7) compared to MSSA patients. In the standard design more confounders were identified than in the parallel cohort design. The ROC curves from the three regression models showed that the parallel cohort design gave a better discrimination than the standard cohort design, especially for MRSA parallel cohort.

Conclusions: In this case the estimates for 30 day mortality associated with methicillin resistance differed only a little for the two designs. Nevertheless, we believe a parallel matched cohort design is a more valid approach, as the discriminative power of the regression models was higher and less confounders were identified. Next to that, it gives information about the single impact of methicillin resistance, but also about the impact of a MSSA or MRSA BSI compared to patients without this type of infection.