Risk score to predict inpatient mortality of acute pancreatitis patients admitted to the intensive care unit.

Background/objectives: Predicting inpatient mortality for acute pancreatitis (AP) patients in the ICU is crucial for optimal treatment planning. This study aims to develop a concise risk score model for this purpose, enhancing the predictability and management of AP in ICU settings.

Methods: We included 380 patients in our training set. Twenty-seven variables were retrospectively collected, and predictive variables were selected using LASSO penalized regression and refined through backward elimination multivariate models. Effect sizes were used to create the final model to predict 7 and 30-day mortality among AP patients admitted to the ICU.

Results: Of 380 patients, the mortality rate was 23.2 %. The final model included five predictors: INR, Albumin, Lactic Acid, BUN, and Bilirubin. The 5-fold cross-validated mean AUC was 0.93 (SD: 0.048) for 7-day mortality and 0.84 (SD: 0.033) for 30-day mortality, with a sensitivity of 77 % and specificity of 74 %. The risk score outperformed BISAP (AUC: 0.60) and APACHE-II (AUC: 0.76) in predicting mortality.

Conclusion: Our model offers a convenient tool using commonly available laboratory results to predict mortality among AP patients, with potential applicability in both ICU settings.