A new in-hospital mortality prediction nomogram for intensive care unit patients with acute pancreatitis

Abstract

Introduction: Acute pancreatitis (AP) is a prevalent inflammatory disease that can lead to severe abdominal pain and multiple organ failure, potentially resulting in pancreatic necrosis and persistent dysfunction. A nomogram prediction model was developed to accurately evaluate the prognosis and provide therapy guidance to AP patients.

Material and methods: Retrospective data extraction was performed using MIMIC-IV, an open-source clinical database, to obtain 1344 AP patient records, of which the primary dataset included 1030 patients after the removal of repeated hospitalizations. The prediction of in-hospital mortality (IHM) used the least absolute shrinkage and selection operator (LASSO) regression model to optimize feature selection. A multivariate logistic regression analysis was used to build a prediction model incorporating the selected features, and the C-index, calibration plot, and decision curve analysis (DCA) were utilized to evaluate the discrimination, calibration, and clinical applicability of the prediction model.

Results: The nomogram utilized a combination of indicators, including the SAPS II score, RDW, MBP, RR, PTT, and fluid-electrolyte disorders. Impressively, the model exhibited a satisfactory diagnostic performance, with area under the curve values of 0.892 and 0.856 for the training cohort and internal validation, respectively. Moreover, the calibration plots and the Hosmer-Lemeshow goodness-of-fit (HL) test revealed a strong correlation between the predicted and actual outcomes (p = 0.73), further confirming the reliability of our model. Notably, the results of the decision curve analysis (DCA) highlighted the superiority of our model over previously described scoring methods in terms of net clinical benefit, solidifying its value in clinical applications.

Conclusions: Our novel nomogram is a simple tool for accurately predicting IHM in ICU patients with AP. Treatment methods that enhance the factors involved in the model may contribute to increased in-hospital survival for these ICU patients.

Keywords: Medical Information Mart for Intensive Care IV (MIMIC-IV); acute pancreatitis; in-hospital mortality; nomogram; prediction model.

Figure 1

Flow diagram of study design

Figure 2

Clinical variables were selected using the lasso logistic regression model. A – Optimal parameter (l) selection in the LASSO model used fivefold cross-validation via minimum criteria. The partial likelihood deviance (binomial deviance) curve was plotted versus log(l). Dotted vertical lines were drawn at the optimal values by using the 1 SE of the minimum criteria (the 1-SE criteria). B – LASSO coefficient profiles of the 47 features. A coefficient profile plot was produced against the log(l) sequence. A vertical line was drawn at the value selected using tenfold cross-validation, where optimal l resulted in five features with nonzero coefficients

Figure 3

A – Risk factors for SAPS II, RDW, MBP, RR, and PTT for the nomogram prediction model. B – Dynamic nomogram used as an example. The significance of the asterisks beside each variable in part B represents the importance of all the risk factors

Figure 4

ROC curve and AUROC of the nomogram, SAPS II, RDW, MBP, RR, and PTT in the training set (A) and validation set (B). The AUROC of the nomogram was larger than that of SAPS II, RDW, MBP, RR, and PTT in the training set. The AUROC of the nomogram was larger than that of RDW, MBP, RR, and PTT in the validation set but similar to that of SAPS II

Figure 5

Calibration curves constructed by the bootstrap approach in the training set (A) and validation set (B). In both sets, the apparent curve and bias-corrected curve slightly deviated from the reference line, but good conformity between the observation and prediction was observed

Figure 6

DCA curve of medical intervention in patients with the nomogram, SAPS II, RDW, MBP, RR, and PTT in the training (A) and validation sets (B)