Association between albumin-corrected anion gap and delirium in acute pancreatitis: insights from the MIMIC-IV database

Abstract

Background: Delirium frequently occurs as a severe complication among patients with acute pancreatitis (AP), contributing to extended hospital stays, higher mortality rates, and lasting cognitive deficits. The pathogenesis of delirium in this setting is strongly influenced by metabolic abnormalities, including disturbances in electrolyte balance and widespread inflammation. Although the albumin-corrected anion gap (ACAG) is a recognized indicator of metabolic dysfunction, its relevance to delirium in AP patients has not been adequately investigated.

Methods: This study utilized patient records from the MIMIC-IV database to investigate how ACAG relates to the onset of delirium in individuals with acute pancreatitis. Analytical approaches included the use of summary statistics, Kaplan-Meier survival analyses, receiver operating characteristic (ROC) curve evaluation, and both univariable and multivariable Cox proportional hazards models. To capture potential nonlinear effects, restricted cubic spline (RCS) modeling was implemented. Subgroup analyses were conducted to examine possible demographic and clinical effect modifiers. Additionally, several machine learning algorithms-such as the Random Forest-were employed to further evaluate the predictive power of ACAG.

Results: Elevated levels of ACAG were independently linked to an increased likelihood of developing delirium during both the 28-day hospitalization period and throughout the ICU stay. Results from the multivariable Cox proportional hazards analysis indicated that each incremental rise in ACAG was associated with a greater risk of delirium (hazard ratio: 1.06, 95% confidence interval: 1.02-1.10, p < 0.001). The application of restricted cubic spline modeling verified the linear nature of this association. Among the machine learning models, the Random Forest achieved superior predictive accuracy (AUC = 0.81), and SHAP analysis highlighted ACAG as a primary determinant in model prediction.

Conclusions: The ACAG emerged as an independent predictor of delirium among individuals with acute pancreatitis, displaying a linear association with the risk of delirium onset. When compared to other commonly used biomarkers, ACAG exhibited enhanced predictive capacity for identifying patients at risk. These findings suggest that ACAG could serve as a practical clinical marker for the early detection and prompt management of delirium in this patient population.

Keywords: Acute pancreatitis; Albumin-Corrected anion gap; Anion gap; Delirium; Machine learning; Risk prediction.

Fig. 1

Patient inclusion flowchart

Fig. 2

Cumulative incidence curves: a incidence of delirium during the 28-day hospitalization; b incidence of delirium during the 28-day ICU stay

Fig. 3

ROC curve analysis of predictive performance for delirium: a during 28-day hospitalization; b during 28-day ICU stay

Fig. 4

Restricted cubic spline analysis in patients with acute pancreatitis: a delirium incidence risk during the 28-day hospitalization; b delirium incidence risk during the 28-day ICU stay

Fig. 5

Subgroup analysis of delirium risk in acute pancreatitis patients: a risk during 28-day hospitalization; b risk during 28-day ICU stay

Fig. 6

Important predictors of delirium in acute pancreatitis patients identified by boruta feature selection, with green variables representing the most significant predictors

Fig. 7

ROC a and PR b curves of multiple models on the validation set

Fig. 8

Variable importance based on SHAP values in the random forest model for predicting delirium