To assess the association between inclusion of a macrolide in a beta-lactam-based empirical antibiotic regimen and mortality among patients with bacteremic pneumococcal pneumonia, 10 years of data from a database were analyzed. The total available set of putative prognostic factors was subjected to stepwise logistic regression, with in-hospital death as the dependent variable. Of the 409 patients analyzed, 238 (58%) received a beta-lactam plus a macrolide and 171 (42%) received a beta-lactam without a macrolide. Multivariate analysis revealed 4 variables to be independently associated with death: shock (P<.0001), age of >or=65 years (P=.02), infections with pathogens that have resistance to both penicillin and erythromycin (P=.04), and no inclusion of a macrolide in the initial antibiotic regimen (P=.03). For patients with bacteremic pneumococcal pneumonia, not adding a macrolide to a beta-lactam-based initial antibiotic regimen is an independent predictor of in-hospital mortality. However, only a randomized study can definitively determine whether this association is due to a real effect of macrolides.