Identification of early predictors for infected necrosis in acute pancreatitis

Acute pancreatitis is the most frequent non-malignant gastroenterological disorder leading to hospitalization in Western countries. It accounts for almost 280,000 hospital admissions in the US [1] and around 55,000 in Germany per year [2]. While the majority of patients suffers from a mild disease with an uneventful recovery, there is a severe course of acute pancreatitis in 10 to 15% of cases leading to organ or even multi-organ failure, necessity for intensive care therapy and a high mortality [3]. Besides organ failure, approximately 5 to 20% of patients develop necrotizing pancreatitis, involving the pancreas, the surrounding fatty tissue or both [4]. Necroses may cause further local complications such as compression of adjacent organs, increase of intraabdominal pressure or gastric outlet obstruction. Secondary infection of the necrotic tissue is a severe condition with increased morbidity and mortality [5] requiring antibiotic treatment or even invasive interventions [6, 7].

Diagnosis of an infected necrosis is still challenging and often it needs to be confirmed ultimately by microbiological analysis after fine-needle aspiration or even drainage, measures that have to be carried out judiciously because they also encompass a periprocedural risk [8]. Established multiparameter scores such as the Acute Physiology and Chronic Health Evaluation II (APACHE II) score [9] and the Ranson Score [10] have been used for grading disease severity and prediction of mortality. However, they are cumbersome to calculate as they need a large number of parameters requiring different time points and their predictive accuracy for infected necrosis is unclear. Several routine laboratory parameters, for instance, markers for inflammation, kidney function or hematocrit [11], have been attempted for accurate prediction of severe acute pancreatitis, development of necrosis and mortality. Despite promising potential, these measurements have to be repeated at later time points and their usability is limited when using every single parameter alone. So far predictive factors of infected pancreatic necrosis, allowing the initiation of an early and preemptive therapy to improve the outcome of acute necrotizing pancreatitis, have not been established.

Infected necrosis is a severe complication of acute pancreatitis that usually arises during the later phase of pancreatitis. In this study, we identified parameters associated with infection of necrosis in acute pancreatitis. Based on these findings, we developed a logistic regression model based on blood levels of creatinine, albumin, and CRP, as well as alcoholic etiology that predicts infection with higher accuracy than any individual laboratory parameter or the APACHE II score.

The parameters we finally included in our prediction model are coherent with existing literature on prediction of the course and complications in acute pancreatitis. For instance, CRP, an acute-phase reactant, has been shown repeatedly to predict severity of acute pancreatitis—although there has been debate about the optimal time point and cut-offs [18, 19]. Moreover, CRP had a good prognostic accuracy not only for severe acute pancreatitis but also pancreatic necrosis and in-hospital mortality [20]. Prognostic value has also been found specifically regarding development of secondary infections in acute pancreatitis [21].

Likewise, regarding creatinine, there is evidence that elevated levels in early disease can predict pancreatic necrosis [22, 23]. It is conclusive that creatinine also predicts secondary infection of pancreatic necrosis as it indicates impaired renal function and potential subsequent renal failure, which we found to be associated with infected necrosis.

Inclusion of albumin in the final prediction model was not an unexpected finding. Serum albumin has been found predictive of persistent organ failure in acute pancreatitis in multiple studies [24, 25]; and in our analysis it was linked to infected pancreatic necrosis as well. Being independently associated with both inflammation and a compromised nutritional status in acute conditions [26], hypoalbuminemia—by the very same mechanisms—could not only predispose to organ failure but also infection of pancreatic necrosis.

Although the role of etiology on course and progression of acute pancreatitis has been discussed controversially for a long time, recent findings support the relevance of alcoholic etiology for prediction of infected pancreatic necrosis. A recent meta-analysis found necrosis to be more common in alcoholic than biliary pancreatitis [27]—the two most common etiologies in acute pancreatitis. Additionally, evidence has accumulated from experimental studies that alcohol increases intestinal permeability and thus facilitates translocation of both bacteria and bacterial products [28] that could elicit infection of pancreatic necrosis.

A number of multiparameter predictors have been evaluated for prediction of adverse outcome in acute pancreatitis [29]. The APACHE II system is one of the most widely used severity scores for critically ill patients, which incorporates both markers of patient physiology recorded immediately or shortly after hospital admission and chronic comorbidity categories. Due to these known relations, we evaluated this score regarding a potential link to infected necrosis as well. There was an association of APACHE II score with infected pancreatic necrosis. However, our model outperformed it. Considering that the APACHE II score is not specific for acute pancreatitis and requires multiple items that in part are laborious to record, such as parameters for blood oxygenation, its usefulness for prediction of infected necrosis seems to be limited in clinical practice [28].

In an earlier study, Chen et al. [30] used a similar approach to develop a prediction model for infection of pancreatic necrosis. Their final model included different parameters than ours. However, these aberrant findings do not necessarily contradict our results. First, etiologies of acute pancreatitis differed in the two cohorts. We had more alcoholic than biliary pancreatitis, which was the most common cause apart from hyperlipidemia in the study by Chen and co-workers. In addition, the studies were conducted in two different countries and findings in Asian populations cannot be transferred unrestrainedly to Western populations and vice versa [31]. We also included patients presenting with acute on chronic pancreatitis, which were excluded in the other study. Further, we investigated a wider a range of clinical and laboratory parameters and included, for instance, albumin, which we found to be an independent predictor of infected necrosis.

One must also consider parameters that have been suggested as predictors of infected necrosis before but did not contribute to prediction in the current study. For instance, higher median procalcitonin (PCT) concentrations have been found in patients with infected necrosis and a complicated course of acute pancreatitis resulting in death [32]. In our patients, overall mortality was as low as 6.7% percent, which could explain why we did not find an association. Besides, earlier findings suggest that PCT is not a specific marker of infected necrosis as it is also elevated in septic patients without pancreatitis [33]. Moreover, it has been hypothesized that PCT levels in acute pancreatitis are elevated as part of the systemic inflammatory response and therefore not necessarily indicate infection [34].

Blood urea nitrogen (BUN) has been reported with alleged predictive value as a rise in blood urea nitrogen within 48 h was associated with a risk for the development of infected pancreatic necrosis [35]. Although we found an association between BUN and infected pancreatic necrosis as well, the association was weaker than with other parameters and inclusion of BUN did not further improve the prediction model. Besides analyzing BUN at a single time point, a high correlation with creatinine, another indicator of renal function and the strongest single predictor of infected necrosis in our study, could explain why BUN was not included in our final prediction model.

There are limitations to our study. These are partly owed to its retrospective and monocentric design, including incomplete patient data and blood values as well as assessment of blood parameters only at time of admission. Therefore, there is a residual chance that we missed relevant parameters, especially those that show a dynamic during the course of diseases. On the contrary, our results realistically reflect the situation in clinical practice. It can be cumbersome and costly to monitor the course of multiple, potentially not routine blood parameters over a longer time. Hence, our prediction model likely presents a more feasible approach. However, it needs to be emphasized that its predictive performance has not been validated prospectively so far. A prospective trial will be necessary to confirm the validity of our model developed from the retrospectively collected data. Another limitation of our analysis is that we also included patients transferred from external hospitals. This may include that treatment of acute pancreatitis at least during the early phase was not uniform in all cases because local expertise varies in smaller district hospitals. In addition, time between actual onset of pain and hospital admission could vary leading to an inhomogeneous patient cohort regarding stage of pancreatitis. Although only individuals with microbiologically proven infection were included in our study there is a risk of false positive or negative results even after microbiologic analysis of the necrotic material which have been reported in up to 15% and 25% of cases, respectively [36]. In addition, the number of actually infected necroses might be lower as secondary infections might occur not only after percutaneous but even after endoscopic guided drainages of pancreatic necrotic collections and repeated necrosectomies. For further clarification of microbial transmissions rates into drained necroses additional studies will be necessary. The putative low number of patients with sterile necrosis (n = 30) in this investigation resulted from the fact, that only individuals with proven negative results on microbial culture were selected, even after repeated biopsies. Due to the selection of patients who have undergone intervention we observed a larger proportion of individuals with infected necrosis than reported in previous studies [37]. Under some circumstances a primarily conservative therapeutic strategy based on solely antibiotic treatment and drainage only if unavoidable, can be as effective as an immediate drainage therapy in terms of mortality [38]. Because suspected infected necroses could not be captured by ICD-10 codes, we have potentially missed patients with infected pancreatic necrosis who neither underwent EUS-FNA nor drainage for our model. Last, some patients may have responded to prophylactic antibiotic treatment that was given empirically without prior microbial confirmation and therefore did not develop infected necrosis. Nevertheless, the chance that predictive performance of our model was hampered by such treatment response is rather low as an infected necrosis was detected in almost 90% of patients receiving antibiotics.

We could develop a prediction model for identification of infected necrosis in acute pancreatitis. It might help to avoid overhasty interventions on pancreatic necrosis in situations when infections are suspected. Including only four parameters, already assessable in early disease, our model could facilitate clinical decision-making in treatment of acute pancreatitis. We therefore encourage use of this model in future prospective studies to validate its clinical relevance.