Acute pancreatitis is a common (incidence among the VA population is estimated at 1,100 per 100,000) and potentially lethal inflammatory process with a highly variable clinical course that progresses to persistent organ failure in 10-20% of patients. Since Ranson's attempt to predict the clinical course of acute pancreatitis, similar scoring systems and individual markers have been examined for this purpose. However, these tools have suboptimal accuracy (often d80%) and have not been subjected to comparative effectiveness research. Further, the outcome predicted in previous studies that assessed these tools varies and does not focus on a critical indicator of severe acute pancreatitis: persistent organ failure, which is defined as persisting for 48 or more hours and involves the cardiovascular system, the pulmonary system, and/or the kidneys. The time threshold is clinically relevant as organ failure lasting at least 48 hours directly affects the intensity and duration of hospital care, and death from acute pancreatitis predominantly occurs in patients with persistent organ failure. In our pilot study, a panel of 5-biomarkers predicted persistent organ failure in a subset of 25 patients for whom we had early serum samples with an accuracy of 96%. We now seek to use the 5-biomarker panel data to fit diverse statistical models and rank their ability to predict risk of persistent organ failure through cross- validation. We will also directly compare our models based on the 5-marker panel data with established individual prognostic markers and scoring systems. We will prospectively enroll a cohort of 180 patients admitted to the hospital with acute pancreatitis. We will obtain serum at admission (primary end-point) and at 24 and 48 hours from admission to measure the 5 biomarkers (tumor necrosis factor-a receptor-1, interleukin-8, hepatocyte growth factor, angiopoietin-2, resistin) using Luminex technology. For our primary analysis, we will use the 5-biomarker panel to fit and test via cross-validation five models that represent a variety of linear and non-linear techniques: Naove Bayes, logistic regression model, support vector machines, artificial neural networks, and decision trees. Using data from this same cohort, we will calculate the accuracy of 5 single markers (glucose, hematocrit, blood urea nitrogen, creatinine, and C-reactive protein) and 7 scoring systems (Ranson's, acute physiology and chronic health examination-II or APACHE-II, Glasgow, systemic inflammatory response syndrome, bedside index for severity in acute pancreatitis, Panc 3, and harmless acute pancreatitis score) with regard to predicting persistent organ failure. We will use areas under receiver operating characteristic curves (AUROC) to compare and rank the performance of our 5 models based on the 5-biomarker panel, the 5 individual markers, and the 7 scoring systems. The proposed study will improve our ability to reliably identify patients with acute pancreatitis who are at risk of developing persistent organ failure and determine, based on direct comparison of AUROC, which predictive approach should be applied in a clinical setting such as the Veterans Affairs Health System. The most accurate prognostic tool can then be applied to determine appropriate levels of clinical monitoring for hospitalized patients. The resulting comprehensive dataset will be available for subsequent secondary analyses to generate hypotheses regarding pathogenic mechanisms and to identify appropriate cohorts of high-risk patients for investigational clinical trials, such as administration of immunomodulating agents, which have been successfully tried in animal models of acute pancreatitis, i.e. tumor necrosis factor-a inhibitors.
Acute pancreatitis is a major health care issue in the US, including the VA Health System. Locally (VA Pittsburgh Healthcare System), we admit approximately 60 Veterans per year with acute pancreatitis. In a study of almost 60,000 Veterans, the prevalence of acute pancreatitis was about 1.1%. In the US population, certain groups reflective of the veteran population (blacks and older patients) have disproportionately high hospitalization rates for acute pancreatitis, and overall mortality is 2%. Acute pancreatitis hospitalizations cost more than $2 billion annually. The subgroup of patients with severe disease accounts for most of this expense. Therefore, a high priority exists to improve our understanding of acute pancreatitis, particularly those patients at high risk for a severe (and expensive) disease course. Our proposed work to identify the best method for predicting disease outcome will have a significant impact on patient management and hospital costs and may suggest new approaches to treatment.
