Reply to Nannini and to a previous letter by Hurley.

Impact of van-comycin resistance on mortality among patients with neutropenia and enterococcal bloodstream infection. Epidemiology and mortality risk of vanco-mycin-resistant enterococcal bloodstream infections .terococcal bacteremia: risk factors for vanco-mycin resistance and predictors of mortality.tant and vancomycin-susceptible enterococcal bacteremia: comparison of clinical features and outcomes. Comparison of risk factors and outcome in patients with Enterococcus faecalis vs. Entero-coccus faecium bacteraemia. A simple model host for identifying gram-positive virulence factors. To the Editor—In recent correspondence , Hurley [1] performed an ecologic analysis that suggests that length of hospital stay (LOS) before the onset of bac-teremia may have confounded the association between vancomycin resistance and mortality observed in our meta-analysis [2]. We believe his conclusions are misleading. Ecologic analyses are subject to the so-called ecologic fallacy [3], in which associations observed using group average data are not necessarily a valid reflection of associations seen at the individual patient level. There is evidence from patient-level multivariate analyses that the observed association between vancomycin resistance and mortality cannot be explained by differences in LOS. Of the 6 studies cited by Hurley [1] in which multivariate analyses controlling for vancomycin resistance, LOS, and severity of illness were performed, none found an independent association between LOS and mortality [4–9]. Although we acknowledge the limitations of ecologic studies, we repeated Hur-ley's [1] analytic approach, but controlled for vancomycin resistance. We excluded 2 studies cited by Hurley [1] for which insufficient data were provided [10, 11] and 1 [12] of 2 studies [12, 13] that contained duplicate data. In a multivariate linear regression model with 2 exposure variables (LOS and vancomycin resistance) and 1 outcome variable (mortality), the association between LOS and mortality disappears (), whereas vancomycin reP p .91 sistance shows an independent and significant association with death (P p). .0048 Therefore, multivariate analyses of individual patient data and multivariate linear regression using group average data fail to support Hurley's [1] suggestion that the association between vancomycin resistance and death described in our meta-analysis [2] could be explained by a difference in LOS between patients with bacteremia due to vancomycin-resistant enterococci (VRE) and patients with bac-teremia due to vancomycin-sensitive en-terococci (VSE). Nannini [14] suggests that 4 studies included in our meta-analysis failed to show a statistically significant association,because there was similarity in the frequency of appropriate antibiotic administrationbetween the comparison groups [5–7, 12]. We believe that this claim is not well supported. Only 3 of the studies included …