A prediction model for 'ICU mortality or prolonged ICU stay' in critically unwell patients with acute pancreatitis: insights from a 2003-2020 cohort analysis using the ANZICS-CORE database

Objective: Critically unwell patients with acute pancreatitis (AP) are at increased risk of mortality and prolonged ICU length of stay (LOS). We quantified the frequency, risk factors and complications of prolonged ICU LOS in a large cohort of critically unwell adult patients with AP and developed a model to predict a low-risk trajectory 'survived ICU with ICU LOS ≤7 days' versus a high-risk trajectory 'ICU mortality or ICU LOS > 7 days'. 

Methods: A retrospective cohort analysis of adult patients admitted to Australian and New Zealand ICUs with AP between 2003 and 2020 was conducted using the Australian and New Zealand Intensive Care Society Centre for Outcome Reporting and Evaluation database. Data was censored to December 2020 in order to pre-date the COVID-19 pandemic. The incidence, risk factors and outcomes related to prolonged ICU LOS in AP patients was reported. Multivariate logistic regression was used to build a prediction model for a low-risk versus high-risk outcome. Discrimination was performed with 10-fold cross validation and calibration plot analysis was reported. 

Main results: 13,275 patients met inclusion criteria; 60% were male, with a mean age 59±18, mean APACHE III 56±26. 2860 (21.6%) had an ICU LOS > 7 days, 1022 (7.7%) died in ICU, and 3557 (26.8%) had a high-risk trajectory. Prolonged ICU LOS was associated with increased ICU mortality (OR 1.57 95% CI 1.43-1.73 p < 0.001), hospital mortality (OR 1.69 95% CI 1.56-1.83 p < 0.001), and resource use: mechanical ventilation (OR 5.99 95% CI 5.21-6.90 p < 0.001), inotrope/vasopressor support (OR 3.27 95% CI 2.82-3.79 p < 0.001) and dialysis (OR 4.12 95% CI 3.63-4.68 p < 0.001). Model accuracy was 79.5%, Cohen K = 0.49 and AUROC 0.827. For a high-risk trajectory, sensitivity was 0.54 and specificity 0.916. APACHE III, PaO2:FiO2 ratio and early mechanical ventilation were the most influential covariates. Prolonged ICU LOS was associated with increased rate of hospital discharge to rehabilitation or a nursing home. 

Conclusions: More than a quarter of ICU patients with AP have a high-risk trajectory. Prolonged ICU admissions are associated with significantly worse mortality and hospital outcomes, and increase resource use. Our prediction model, if confirmed in future studies, may present an opportunity for prognostic enrichment in patients with more severe disease.