AimsAcute‐on‐chronic liver failure (ACLF) is defined as acute liver injury superimposed on chronic liver disease, resulting in a significantly increased short‐term mortality rate. Serum lipids are associated with the severity of liver disease and contribute to the prognostic assessment for cirrhosis and liver failure. However, the presentation of serum lipids varies in different studies, making it difficult to draw definitive conclusions. This study aimed to investigate the predictive value of serum lipids for short‐term prognosis in patients with ACLF.MethodsThis retrospective analysis was conducted using the clinical data of patients hospitalized for ACLF between January 2018 and December 2021. The collected data were subjected to Least Absolute Shrinkage and Selection Operator and logistic regression analyses to identify the independent predictors of 28‐day and 90‐day mortality. Separate regression models were developed for each identified serum lipid. These models were combined with the Model for End‐Stage Liver Disease (MELD), MELD combined with serum sodium concentration (MELD‐Na), Chronic Liver Failure Consortium Acute‐on‐chronic Liver Failure (CLIF‐C ACLF), and Chinese Group on the Study of Severe Hepatitis B‐ACLF (COSSH‐ACLF) scores to construct new models. The performance of these models was evaluated using the area under the receiver operating characteristic curve (AUC), Nagelkerke R2, Brier score, net reclassification index (NRI), and integrated discrimination improvement (IDI) to assess the prognostic ability of serum lipids.ResultsThis study included 266 patients with ACLF, of whom 25.9% (69/266) died within 28 days and 38.5% (102/265, with 1 lost to follow‐up) died within 90 days. The independent predictors of 28‐day mortality were age, the presence of overt hepatic encephalopathy at admission, high‐density lipoprotein cholesterol (HDL‐C) levels, and international normalized ratio (INR). The independent predictors of 90‐day mortality were age, albumin levels, serum creatinine levels, and INR. The comparative analysis revealed that the AUC of all other models exceeded that of HDL‐C (p < 0.001). The COSSH‐ACLF model demonstrated superior predictive ability compared with the MELD and HDL‐MELD models (p = 0.013 and 0.017, respectively). Furthermore, the HDL‐COSSH‐ACLF model exhibited a similar trend in predictive ability compared with the latter two models (p = 0.009). No significant differences were observed in the AUC performances of the other models. A comparison of NRI calculated using the cutoff values assigned by equal assignment revealed that the addition of HDL‐C improved the predictive ability of the MELD score by 4.6% (NRI = 0.0460, p = 0.036). The MELD‐Na score exhibited an 8.4% improvement (NRI = 0.0844, p = 0.047), while the COSSH‐ACLF score demonstrated a 14.4% improvement (NRI = 0.1443, p = 0.015). However, these changes did not significantly impact the capacity of the CLIF‐C ACLF score (NRI = 0.0350, p = 0.141). No significant differences were observed in the IDI comparisons between the models with different cutoff values (all p > 0.050).ConclusionsSerum HDL‐C is an independent predictor of 28‐day mortality in patients with ACLF. Although it has inferior predictive ability when considered alone, HDL‐C enhances the predictive capacity of MELD, MELD‐Na, and COSSH‐ACLF when incorporated into these models.