Prediction Models of Mortality in Acute Pancreatitis in Adults A Systematic Review

Background: Acute pancreatitis (AP) varies in severity, prompting development of systems aimed at predicting prognosis to help guide therapy. Although several prediction approaches are available, their test characteristics and clinical utility are not completely understood. Purpose: To evaluate the test characteristics (prognostic accuracy, incremental predictive value) and clinical utility (effect on patient outcomes) of severity scores for predicting mortality in AP. Data Sources: Ovid MEDLINE and EMBASE (inception to 3 May 2016). Study Selection: Longitudinal studies, in any language, that evaluated the prognostic value of at least 1 clinical severity score in AP. Data Extraction: Dual data extraction and quality assessment. Data Synthesis: Of 4039 citations screened, 94 unique studies evaluating 18 scores in 53 547 patients met the inclusion criteria. All studies provided data on prognostic accuracy, whereas 6 provided data on incremental predictive values. Most scores demonstrated low prognostic accuracy. The Acute Physiology and Chronic Health Evaluation (APACHE) II score and the Ranson criteria were studied most extensively. The median sensitivity and specificity of APACHE II at a threshold of 7 were 100% (range, 68% to 100%) and 63% (range, 21% to 96%), respectively, and those of the Ranson criteria at a threshold of 2 were 90% (range, 0% to 100%) and 67% (range, 14% to 97%), respectively. Estimates of sensitivity were based on relatively few patients. Evidence was limited regarding the incremental predictive value of the scoring systems or their effect on patient outcomes. Limitation: Substantial clinical heterogeneity and inadequate methodological and reporting quality precluded a meta-analysis. Conclusion: The test characteristics and clinical utility of AP severity scores remain uncertain. Additional studies with improved methodological rigor are needed, and the development of new scoring systems may be justified.