BACKGROUND: Severity of illness is an important confounder in outcome studies involving infectious diseases. However, it is unclear whether the time at which severity of illness is measured is important. METHODS: We performed a retrospective study of 328 episodes of gram-negative bacteremia in adult patients to assess the impact of the time of measurement of severity of illness on the association between empirical antimicrobial therapy received and in-hospital mortality. Using a modified Acute Physiology Score (APS), severity of illness was measured at 2 time points: (1) hospital admission and (2) 24 hours before the first culture-positive blood sample was collected. Multivariate logistic regression was used to estimate the impact of adjusting for the APS on the relationship between empirical therapy received (ie, the exposure) and in-hospital mortality (ie, the outcome). RESULTS: The mean APS (+/- standard deviation) of patients with bacteremia increased during their hospital stay (from 19.2 +/- 11.6 at admission to 24.2 +/- 13.6 at the second time point; P < .01). When examining the association between empirical antimicrobial therapy received and in-hospital mortality, and controlling for the APS, there was a trend toward a decreased impact of appropriate therapy received on in-hospital mortality. The unadjusted odds ratio (OR) for the association between appropriate therapy received and in-hospital mortality was 0.83 (95% confidence interval [CI], 0.51-1.34). After controlling for the APS at admission, this association was attenuated (OR, 0.94 [95% CI, 0.57-1.55]), and when a change in the APS was also included in the multivariate logistic regression model, the association was further attenuated (OR, 0.99 [95% CI, 0.58-1.69]). CONCLUSIONS: The magnitude of the association between appropriate antimicrobial therapy received and in-hospital mortality among patients with gram-negative bacteremia was sensitive to the timing of adjustment for severity of illness.