Clinical and laboratory characteristics associated with liver failure in hospitalized dengue patients: insights from a non-endemic region

BACKGROUND: Dengue fever is an emerging public health issue expanding into non-endemic regions like Changsha, China. Liver involvement is a critical complication, and early identification of patients at risk of liver failure remains challenging. This study aimed to characterize hospitalized dengue patients, identify routine laboratory parameters significantly associated with liver failure, and utilize a machine learning model to corroborate the importance of these parameters in this non-endemic setting. METHODS: A retrospective chart review was conducted on 86 laboratory-confirmed dengue patients hospitalized at the First Hospital of Changsha between 2023 and 2024. Routine demographic and laboratory data were collected. Statistical analyses were performed to identify parameters associated with liver failure within this cohort. Separately, a random forest model was developed using an independent dataset of dengue patients and healthy controls to rank feature importance for distinguishing dengue from a healthy baseline. RESULTS: Among 86 dengue patients, mean age was 38.8 years, with 48.8% male. Cases peaked in summer and autumn. Common abnormalities included leukopenia (83.7%), thrombocytopenia (47.7%), elevated ALT (52.3%), AST (48.8%), and LDH (46.9%). After excluding patients with pre-existing liver disease or co-infections, 79 patients were analyzed for liver failure risk factors. Compared with patients without liver failure (n = 40), those with liver failure (n = 39) had significantly lower platelet counts (P = 0.003) and higher ALT (P = 0.005), AST (P = 0.045), and LDH (P = 0.041). A random forest model ranked AST, platelets, and ALT among its top features for distinguishing dengue from healthy status. CONCLUSION: Platelet count, ALT, AST, and LDH were identified as laboratory parameters associated with dengue-related liver failure in hospitalized patients. A complementary machine learning-based feature importance analysis showed consistent ranking of key variables in an independent dataset. These findings provide potential biomarkers for risk stratification and support future prospective validation studies.