Sepsis is a major cause of morbidity and mortality worldwide. When selecting empiric antibiotics for sepsis, clinicians are encouraged to use local resistance rates, but their impact on individual outcomes is unknown. Improved methods to predict outcomes are needed to optimize treatment selection and improve antibiotic stewardship. We expanded on a previously developed theoretical model to estimate the excess risk of death in gram-negative bacilli (GNB) sepsis due to discordant antibiotics using 3 factors: the prevalence of GNB in sepsis, the rate of antibiotic resistance in GNB, and the mortality difference between discordant and concordant antibiotic treatments. We focused on ceftriaxone, cefepime, and meropenem as the anti-GNB treatment backbone in sepsis, pneumonia, and urinary tract infections. We analyzed both publicly available data and data from a large urban hospital. Publicly available data were weighted toward culture-positive cases. Excess risk of death with discordant antibiotics was highest in septic shock and pneumonia. In septic shock, excess risk of death was 4.53% (95% confidence interval [CI], 4.04%-5.01%), 0.6% (95% CI, .55%-.66%), and 0.19% (95% CI, .16%-.21%) when considering resistance to ceftriaxone, cefepime, and meropenem, respectively. Results were similar in pneumonia. Local data, which included culture-negative cases, showed an excess risk of death in septic shock of 0.75% (95% CI, .57%-.93%) for treatment with discordant antibiotics in ceftriaxone-resistant infections and 0.18% (95% CI, .16%-.21%) for cefepime-resistant infections. Estimating the excess risk of death for specific sepsis phenotypes in the context of local resistance rates, rather than relying on population resistance data, may be more informative in deciding empiric antibiotics in GNB infections.