In Reply: We thank Hurley for his comments on our meta-analysis.1 However, we strongly dispute that our analytical technique is flawed, and argue that we have been extremely cautious in drawing our conclusions. Hurley's contention is that hospital length of stay before bacteraemia (LOS) is a surrogate for severity of underlying disease and risk for colonisation with methicillin-resistant Staphylococcus aureus (MRSA), and that these factors explain the higher mortality in patients with MRSA.

We agree that LOS may be a confounder. It may be an effect modifier, whereby patients in hospital for longer may be more ill, and therefore more susceptible to infection with and death from MRSA. Both are intuitive and biologically plausible conclusions. In fact, we referred to these possibilities in our Discussion, writing that "patients who ultimately become infected with MRSA are more seriously ill than those who become infected with MSSA [methicillin-sensitive S. aureus]" and "separating the effect of the bacteraemia per se from the effects of patients' underlying disease and treatment is a major problem when comparing outcomes". We also cautioned readers that available published data on mortality made it impossible for us to adjust for numerous potential confounders, including LOS, as the information given did not link these potential confounders with the outcome in individual patients.

Hurley has not, as he suggests, undertaken an analysis that would allow him to control correctly for the potential confounder, LOS. He, like us, used "group-as-a-unit" data, but, although the groups are homogeneous for MRSA or MSSA, they are heterogeneous for LOS. Adequate examination of and control for potential confounders requires either individual patient data or data from homogeneous groups. Hurley has attempted to use analysis normally reserved for individual data.4 His analysis was analogous to treating the data as though from an ecological study, a design in which control of confounding is difficult,5 and thus does not permit him to draw his conclusions.

Our analysis (not presented in our original article) of only those studies where the authors attributed mortality to bacteraemia6-8 found that the magnitude of effect remained (fixed-effect relative risk, 2.27; 95% CI, 1.75–2.96; P < 0.001; test for heterogeneity, χ2 = 6.14, df = 4, P = 0.19).

As MRSA bacteraemia is a rare event and published studies are small, the statistical ability to control for confounding and effect modification is limited. Until sufficient suitable data for individual patients are available for analysis, we have remained restrained in our assessment. Mindful that MRSA bacteraemia is associated with increased mortality, regardless of the cause, we hold with our original conclusion that "our findings justify ongoing surveillance and proactive management of MRSA in healthcare facilities".