The HIV epidemic in Russia is still ongoing; the most common causes of death in these patients are secondary infections. We do not expect any decrease in the number of HIV-infected patients requiring special medical care. Objective. To develop a prognostic model aimed to evaluate the risk of death in HIV-infected patients using multivariate analysis. Materials and methods. We performed retrospective analysis of 1,440 cases of severe HIV infection to develop and train a new prognostic model. Using logistic regression, we identified the most important variables in the model. Receiver operating characteristic (ROC) analysis was used to test the model. We also created a special MS Excel calculator to predict the disease outcome. Results. Twelve clinical and laboratory parameters were included into the model. The area under the ROC-curve (AUC) was 0.769; specificity of the model (proportion of correctly classified patients in the group with lethal outcome) was 0.92 (92%); sensitivity of the model (proportion of correctly classified patients in the group with improvement) was 0.45 (45%); overall accuracy of the model (overall proportion of correctly classified patients) was 0.802 (80.2%). Our model proved to be highly effective: 83.9% of the deceased patients had a probability of death between 71% and 92%, whereas 75% of patients who showed an improvement had a probability of death between 8% and 42%. Conclusion. Our model predicting lethal outcome in patients with severe HIV infection can be used in routine clinical practice, especially at the pre-hospital stage and shortly after hospitalization. It allows a doctor to make a correct decision on patient's admission to a proper department. The program is registered with the Federal Service for Intellectual Property on November 18, 2022, No. 2022682077. © 2022, Dynasty Publishing House. All rights reserved.