Development of a Nomogram to Predict In-ICU Mortality of Elderly Patients with Sepsis-associated liver injury: An Analysis of the MIMIC-IV

Xuemei Hu ^1,2 Jianbao Wang ^1,2 Susu Cao ^1,2 Aolin Xia ^1,2 Xiaocong Jiang ^2,3 Tianfeng Hua ^1,2 Min Yang ^1,2*

• ¹ Department of Intensive Care Unit, Second Affiliated Hospital of Anhui Medical University, Hefei, Anhui Province, China
• ² The Laboratory of Cardiopulmonary Resuscitation and Critical Care Medicine, The Second Affiliated Hospital of Anhui Medical University, hefei, China
• ³ Department of Emergency, The Second Hospital of Anhui Medical University, Hefei, Anhui Province, China

SALI is a frequent and lethal complication among critically ill patients in the intensive care unit (ICU). Despite its significance, there has been a notable lack of specialized tools for evaluating the in-ICU mortality risk in these patients. This study seeks to address this gap by developing a practical nomogram to predict risk factors associated with in-ICU mortality in patients suffering from SALI. Data were extracted from the MIMIC-IV database, a Critical Care Public Medical Information Mart. The diagnostic criteria for sepsis adhered to the Sepsis 3.0 guidelines, requiring a SOFA score of ≥2. SALI was defined as total bilirubin (TBIL) levels > 2 mg/dL in patients with sepsis and an International Normalized Ratio (INR) > 1.5. Lasso regression analyses were conducted on the training set (n=653) to develop a predictive nomogram model. Receiver Operating Characteristic (ROC) curves were generated to evaluate model discrimination. Model calibration was assessed through calibration curves and Hosmer-Lemeshow goodness-of-fit tests. Clinical decision curves were plotted to analyze the net benefit of the model and evaluate its clinical applicability. A total of 934 elderly patients with SALI were included in the study. Random seeds were allocated in a 7:3 ratio, resulting in training and validation sets comprising 653 and 281 patients, respectively. Variables were selected using lasso regression, culminating in the inclusion of six final variables within the model. The nomogram was evaluated against standard ICU scoring systems, specifically SAPS II and SOFA scores, yielding AUROC values of 0.814, 0.798, and 0.634 for the training set, respectively. Conversely, the validation set demonstrated AUROC values of 0.809, 0.791, and 0.596. The nomogram exhibited strong predictive performance for in-ICU outcomes. P-values from the Hosmer-Lemeshow goodness-of-fit test for both training and validation sets were recorded at 0.627 and 0.486 respectively, indicating good fit quality. DCA revealed that the nomogram consistently provides greater net benefits compared to SAPS II and SOFA scores. A prediction model of in-ICU mortality in SALI elderly patients was established by screening variables through lasso regression. Nomgram was the best predictor of in-ICU mortality in SALI patients, which has a high reference value and clinical application.