To predict the risk of hospital deaths in patients with hospital-acquired pneumonia (HAP) caused by multidrug-resistant Acinetobacter baumannii (MDR-AB) infection. A total of 366 patients who were diagnosed with HAP caused by MDR-AB infection were enrolled between January 2013 and December 2016. The sociological characteristics and clinical data of these cases were collected. Univariate and multivariate logistic analyses were used to explore the risk factors of hospital deaths before medication and after drug withdrawal. The receiver operating characteristic (ROC) curve and the area under the curve (AUC) were utilized to assess the predictive effectiveness of the models with or without the adjustment. Hospital deaths occurred in 142 cases (38.80%). The results showed that acute physiology and chronic health evaluation II (APACHE II) and sequential organ failure assessment (SOFA) scores before medication and after drug withdrawal were associated with the risk of hospital deaths. Adjusting the covariants including the age, autoimmune disease, venous cannula, transfer of patients from other hospitals, and APACHE II score at admission, then no differences were discovered in predicting the hospital deaths between adjusted APACHE II and adjusted SOFA scores before medication (AUC: 0.808 vs 0.803, P =0.614) and after drug withdrawal (AUC: 0.876 vs 0.878, P =0.789). Before medication or after drug withdrawal, the adjusted APACHE II and adjusted SOFA scores all performed well in determining the predictive effectiveness of the hospital deaths in patients with HAP caused by MDR-AB infection, indicating that the appropriate infection control may reduce the occurrence of nosocomial deaths and improve the prognosis. © 2020 Shu et al.