Background A model which can early and sensitively identify poor clinical outcome in short-term and long-term could be a useful tool to help physicians to assess the severity of the dis-ease and early onset of therapeutic measures would be implemented in order to improve the prognosis of sepsis patients. This present study aimed to develop early predictive models for clinical outcomes based on a public database. Methods In the Medical Information Mart for Intensive Care-Ⅲ (MIMIC-Ⅲ)database,patients with severe sepsis or septic shock were included. Clinical variables were com-pared between survivor group and non-survivor group. Risk factors were identified by logistic regres-sion model. Results A total of 2057 patients with severe sepsis or septic shock were finally enrolled. Mortality in 30-day and 180-day were 35.39％ and 48.47％,respectively. Four independent factors including age,RDW,lactate and albumin for 30-day and 180-day mortality were identified in multiva-riate analysis. The accuracy of 30-day mortality model and 180-day mortality model were 0.702 and 0.716,respectively. The area under the receiver operating characteristic curves (AUCs)of two mod-els were 0.711 and 0.722,respectively.Conclusions In our study,a predictive model with four inde-pendent factors including age,RDW,lactate and albumin was performed by logistic regression,which could be applied for early identification in both 30-day and 180-day mortality in severe sepsis or septic shock.