The study examined patients with severe community-acquired pneumonia (CAP) who were on mechanical ventilation and had difficulty weaning from the ventilator in a respiratory care center. A retrospective cohort analysis included 353 patients with CAP and hospital-acquired pneumonia (HAP) admitted in 2015–2017, while the main analysis focused on 270 patients with CAP. The authors compared the commonly used CURB-65 score with new models to predict survival during hospitalization in a respiratory care center. CURB-65 and its modification CUB-65 achieved an area under the ROC curve (AUROC) of approximately 65%. The new three models included age ≥ 65 years, low blood pressure, elevated BUN (> 19 mg/dl), ventilator type, low GCS consciousness score (≤ 8), and in model 1 or 2, white blood cell count or hemoglobin level. In the training sample, the new models had higher AUROC than CURB-65, with statistically significant improvements for models 1 (p = 0.0354) and 3 (p = 0.0383). In a validation sample of patients with CAP, all three new models achieved AUROC ≥ 80%, indicating good ability to predict survival during the respiratory care center stay. The authors conclude that these new models more accurately predict survival and may serve as a better tool for clinicians in the care of patients with CAP and difficult ventilator weaning.