What’s Ahead? Prediction Models of Mortality in Acute Pancreatitis

Acute pancreatitis (AP) is one of the most common causes of hospitalization for a gastrointestinal condition, and its incidence continues to increase in the U.S. and across the world.1 With improvements in general clinical care, the overall mortality of AP in the U.S. is now less than 2 percent. Despite this seemingly low mortality, all clinicians know that some patients will develop organ system failure or pancreatic necrosis, leading to ICU care and increased risk of death. There is clinical value in identifying these patients destined for poor outcomes. Accurate predictive models could allow more appropriate and early use of intensive care, more aggressive or specific early therapy, and prevention of serious complications and mortality. Accurate predictions have proven exceedingly difficult to achieve.

Most patients with AP have a short-lived illness and promptly recover. About one-fifth of patients with AP have a more severe disease course. Severe disease is characterized by persistent organ system(s) failure, by pancreatic and/or peripancreatic necrosis, and, most importantly, by infected necrosis.2,3 In patients with severe AP, mortality rates are as high as 20 percent. These patients also experience longer hospital stays, more hospital-acquired infections, more ICU days, more need for intervention, and much higher health-care costs and utilization. The ability to identify these patients prior to organ system failure and pancreatic necrosis might allow targeted therapy to prevent these complications. It could reduce many of the important clinical outcomes, including mortality. Conversely, early prediction might have little clinical impact if therapies to prevent these outcomes are not available or if predictive models do not change therapeutic decisions. Currently no specific treatment modality, other than perhaps fluid resuscitation, exists for treating AP. Therefore, early prediction is not now linked to specific effective treatment.


More predictive individualized biomarkers, along with randomized trials documenting improvements in meaningful clinical outcomes, are needed.


Clinicians routinely make estimates and predictions about individual patients. The prediction of severe AP is often based on observations from an experienced clinician who utilizes patient and disease-specific factors. Patients who are older, are obese, have multiple medical comorbidities or abuse alcohol are particularly likely to die from AP.4 Similarly, findings such as the presence of systemic inflammatory response syndrome (SIRS), evidence of substantial third-space loss and intravascular volume contraction (elevations in hematocrit, BUN or creatinine), or crosssectional imaging showing severe necrosis might serve as clinical predictors of severe AP. These are largely qualitative estimates of prognosis. Various multiple-factor scoring systems assign a score to each patient to develop a more quantitative measure of prognosis. These systems have been developed to aid experienced clinical judgment. The Ranson’s criteria are the most commonly used numerical model to predict AP mortality on initial and 48-hour lab values; however, many others exist. These prediction models are of variable complexity in the number of incorporated factors and in the cutoff chosen to differentiate severe from mild AP. In many of these models, factors are assessed over two to three days, which does not predict severe disease but might allow severe disease to be identified as it develops.

A recent systematic review5 analyzed 94 studies and utilized 18 different multiple-factor scoring systems available for predicting mortality from AP. These include not only the Ranson’s criteria (now of mainly historical interest) but also APACHE-II and numerous others. Some of these studies were carried out more than 40 years ago, and all studies were of low methodological quality with moderate or high risk of bias. This systematic review assessed the clinical value of these systems using their prognostic accuracy and the incremental predictive value (the additive accuracy of combining systems). It was noted the scoring systems provided generally high sensitivity (identified most patients likely to die) but very poor specificity (many patients with scores above the cutoff did not die). This poor specificity is predictable, due to the relatively low prevalence of mortality. Analysis of overall accuracy, using ROC curves, demonstrated consistently poor accuracy for all systems. There was no evidence of incremental accuracy by combining systems. There was no evidence from any study demonstrating clinical utility, impact on patient outcomes or cost-effectiveness.

Currently available multiple-factor scoring systems are inadequate to guide clinical judgment.4,6 Development of newer multiplefactor scoring systems, by themselves, are not likely to increase the prognostic accuracy. More predictive individualized biomarkers, along with randomized trials documenting improvements in meaningful clinical outcomes, are needed. Most importantly, specific therapies that might prevent complications or mortality would make prediction much more valuable and are sorely needed. For the moment, the best prediction remains ongoing clinical surveillance and experienced judgment informed by patient- and disease-related factors (age, obesity, comorbid conditions, the presence of SIRS, and simple laboratory tests, such as BUN, hematocrit and creatinine).

Dr. Gupte has no conflicts to disclose. Dr. Forsmark serves on the Boehringer Ingelheim Data Safety Monitoring Board, and as a consultant for Akcea Therapeutics, Aerial BioPharma and Sun Biopharma.

References
1.Yadav, D., Lowenfels, A.B. The Epidemiology of Pancreatitis and Pancreatic Cancer. Gastroenterology. 2013;144(6):1252–1261.
2.Banks, P.A., Bollen, T.L., Dervenis, C., Gooszen, H.G., Johnson, C.D., Sarr, M.G., et al, Acute Pancreatitis Classification Working Group. Classification of Acute Pancreatitis — 2012: Revision of the Atlanta Classification and Definitions by International Consensus. Gut.2013;62(1):102–11.
3.Dellinger, E.P., Forsmark, C.E., Layer, P., Levy, P., Maravi-Poma, E., Petrov, M.S., et al. Pancreatitis Across Nations Clinical Research and Education Alliance (PANCREA). Determinant-Based Classification of Acute Pancreatitis Severity: An International Multidisciplinary Consultation. Ann Surg. 2012;256(6):875–80.
4.Forsmark, C.E., Vege, S.S., Wilcox, C.M. Acute Pancreatitis. N Engl J Med. 2016;375:1972–1981.
5.Di, M.Y., Liu, H., Yang, Z.Y., Bonis, P.A. L., Tang, J.L., Lau, J. Prediction Models of Mortality in Acute Pancreatitis in Adults. A Systematic Review. Ann Intern Med. 2016;165(7):482–490.
6.Forsmark, C. E., Yadav, D. Predicting the Prognosis of Acute Pancreatitis. Ann Intern Med. 2016;165(7):523–524.

Join the discussion

Your email address will not be published. Required fields are marked *