Objective To explore the relationship between C-reactive protein/albumin ratio (CAR) and the disease severity in patients with severe pneumonia, and its predictive value for 28-day mortality risk. Methods A retrospective analysis was conducted on 152 patients with severe pneumonia admitted to Fuyang People's Hospital from January 2020 to January 2022. They were divided into non-critical illness group (n=51), critical illness group (n=63), and extremely critical illness group (n=38) based on the disease severity. The clinical data such as age and gender of patients was collected, and Pearson correlation analysis was used to explore the correlation between CAR and the severity of illness [determined by Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score]. Multivariate logistic regression was employed to identify independent influencing factors of the severity of illness. According to the survival status of patients after 28 days of treatment, they were divided into survival group (n=107) and death group (n=45). CAR was categorized into quintiles (Q₁-Q₅), and multivariate logistic regression analysis was conducted to explore the correlation between CAR and 28-day mortality risk in severe pneumonia patients. A restricted cubic spline (RCS) model was used to analyze the dose-response relationship between CAR and mortality risk. The predictive value of CAR and related indicators for patient mortality risk was evaluated using the receiver operating characteristic curve (ROC). Results CAR was significantly positively correlated with the severity of the disease (APACHE Ⅱ score) (r=0.716, P<0.05). Neutrophil/lymphocyte ratio (NLR), blood lactate (Lac), and high CAR were independent risk factors for the disease severity in patients with severe pneumonia (P<0.05). After adjusting for confounding factors, the mortality risk increased with the increase of CAR (P<0.05). Subgroup analysis of the screened confounding factors revealed that the correlation between CAR and 28-day mortality risk in severe pneumonia patients remained stable across different APACHE Ⅱ scores, GCS scores, SOFA scores, white blood cell counts (WBC), neutrophils (NEU), red cell volume distribution width (RDW), procalcitonin (PCT), and Lac, with interactions observed between NLR and Lac subgroups (P<0.05). The RCS model indicated that there was no non-linear dose-response relationship between CAR and 28-day mortality risk in patients with severe pneumonia of different genders. ROC curve analysis showed that CAR, Lac, and NLR had good predictive value for 28-day mortality in severe pneumonia patients, with the combined predictive efficacy being significantly higher than that of individual indicators. Conclusion There is a close relationship between CAR and the progression and prognosis of severe pneumonia, making it a new approach to assessing the severity of illness and predicting mortality risk in patients.