Development and validation of a multivariate prediction model for patients with acute pancreatitis in Intensive Care Medicine.

Background/objectives: Development and validation of a multivariate prediction model for patients with acute pancreatitis (AP) admitted in Intensive Care Units (ICU).

Methods: A prospective multicenter observational study, in 1 year period, in 46 international ICUs (EPAMI study).

Patients: adults admitted to an ICU with AP and at least one organ failure.

Interventions: Development of a multivariate prediction model, using the worst data of the stay in ICU, based in multivariate analysis, simple imputation in a development cohort. The model was validated in another cohort.

Results: 374 patients were included (mortality of 28.9%). Variables with statistical significance in multivariate analysis were age, no alcoholic and no biliary etiology, development of shock, development of respiratory failure, need of continuous renal replacement therapy, and intra-abdominal pressure. The model created with these variables presented an AUC of ROC curve of 0.90 (CI 95% 0.81-0.94) in the validation cohort. We developed a multivariable prediction model, and AP cases could be classified as low mortality risk (between 2 and 9.5 points, mortality of 1.35%), moderate mortality risk (between 10 and 12.5 points, 28.92% of mortality), and high mortality risk (13 points of more, mortality of 88.37%). Our model presented better AUC of ROC curve than APACHE II (0.91 vs 0.80) and SOFA in the first 24 h (0.91 vs 0.79).

Conclusions: We developed and validated a multivariate prediction model, which can be applied in any moment of the stay in ICU, with better discriminatory power than APACHE II and SOFA in the first 24 h.