Abstract
OBJECTIVES
We investigated which variables independently associated with protection against or development of post-endoscopic retrograde cholangiopancreatography (ERCP) pancreatitis (PEP) and severity of PEP. Subsequently we derived predictive risk models for PEP.
METHODS
In a case-control design, 5254 patients had 8264 ERCPs, 211 had PEP and 22 had severe PEP. We randomly selected 348 non-PEP controls. We examined seven established- and nine investigational variables.
RESULTS
In univariate analysis, seven variables predicted PEP: younger age, female gender, suspected sphincter of Oddi dysfunction (SOD), pancreatic sphincterotomy, moderate-difficult cannulation (MDC), pancreatic stent placement, and lower Charlson score. Protective variables were: current smoking, former drinking, diabetes and chronic liver disease (CLD, biliary/transplant complications). Multivariate analysis identified seven independent variables for PEP, three protective (current smoking, CLD-biliary, CLD-transplant/hepatectomy complications) and four predictive (younger age, suspected SOD, pancreatic sphincterotomy, MDC). Pre- and post-ERCP risk models of seven variables have a C-statistic=0.74. Removing age (7th variable) did not significantly affect the predictive value (C-statistic=0.73) and reduced model complexity. Severity of PEP did not associate with any variables by multivariate analysis.
CONCLUSIONS
By using the newly identified protective variables with three predictive variables we derived two risk models with a higher predictive value for PEP compared to prior studies.
Keywords: acute pancreatitis, cigarette smoking, chronic liver disease, ERCP
INTRODUCTION
Pancreatitis is a serious potential complication of endoscopic retrograde cholangiopancreatography (ERCP). Post-ERCP pancreatitis (PEP) occurs in 3.5% of all ERCPs1 and up to 40–60% of high risk ERCPs.2,3 Measures that reduce PEP risk in high risk patients are prophylactic stenting of the pancreatic duct 4 and possibly rectal indomethacin.5–7
Counseling of patients about ERCP and PEP necessitates stratifying patients into low and high risk categories. Previous studies developed risk stratification schemes with two different combinations of patient and procedural predictors.2,3 One scheme included women, normal bilirubin, suspected sphincter of Oddi dysfunction (SOD) and difficult cannulation2 and the other included pain during procedure, pancreatic duct cannulation, prior history of PEP and number of cannulation attempts.3 The observations draw attention to patient predictors for PEP, but do not establish which variables are predictive for PEP.
There is some consensus which variables are predictors for PEP based on consistent reporting of specific predictors by at least two of four large multicenter prospective studies.2,8–10 Patient predictors include women,2,9,10 age ≤ 60 years,8–10 prior history of PEP,2,8 and suspected SOD.2,8 Procedural predictors include pancreatic sphincterotomy,2,10 moderate/difficult cannulation (MDC)2,10 and ≥ 2 pancreatic duct contrast injections.2,8,9 A limitation is that these predictors may be confounded by the effects of unknown predictors or those with a doubtful or unclear association to PEP.
Predictors for severity of PEP are not well established, primarily due to limited data. Three of the four large, prospective multicenter studies reported a total of only 18 cases of severe PEP out of 415 cases of PEP.2,8,10 Most patients were young women with suspected SOD, MDC and normal bilirubin.
Based on our recent observations11 we hypothesized that cardiovascular disease (CVD) associates with PEP and severity of PEP. We previously reported that current alcohol and former cigarette-smoking are potential predictors for PEP and those former smokers were older and had a higher frequency of CVD.11 In support of our hypothesis, CVD is a major risk factor for pancreatic atheroma12,13 and pancreatic hypoperfusion with resulting ischemia is an early event in acute pancreatitis.14
To both test our hypothesis and to develop risk models that predict PEP, we used a retrospective case-control design. In contrast to our prior study,11 we collected all new data, increased the study span by 3.7 years to increase the PEP sample size and statistical power, and included additional variables.
MATERIALS AND METHODS
Selection of Case and Control Samples
This study was approved by the University of Michigan Institutional Review Board. We performed an IDX cross-sectional search of our health system’s databases and identified 6,505 unique patients who had a total of 8,264 ERCPs performed between 1/1/1997 and 3/31/2009. 1469 patients had an ICD-9 code (577.0) for acute pancreatitis and 211 patients met criteria by Cotton et al15 for PEP, and had no exclusion criteria (see below). We used a random number generator program to screen the remaining 5036 potential controls for exclusion criteria to select 1.5 control patients for each PEP case for a total of 348 controls. This ratio was selected because initial power calculations of this proportion were similar to higher ratios, assuming a fixed number of PEP cases. To develop a risk model that included age and gender as variables and controlled for changes in technical/operator and endoscopy-related variables over time, we matched controls to PEP cases by two-year increments of time. Exclusion criteria were identical to a previous study.11
Variable Selection and Definitions
Data was collected for patient variables we hypothesized may influence PEP risk, including alcohol-use, cigarette-smoking, Charlson Comorbidity Score16 and individual comorbidities, presence/risk of CVD (see review by Anderson et al17), hypertension, BMI and aspirin. We included NSAID use as a possible patient variable that may influence PEP risk,5,6 descriptive data for ERCP indications, and three of the four established patient variables consistently reported to influence PEP risk by at least two of the four large, multicenter prospective studies.2,8–10 These included age, gender and suspected SOD but not prior history of PEP, the latter because selection bias would have occurred in the process of choosing retrospectively one ERCP among all ERCPs performed for an individual patient. For this reason we extracted data from ERCP reports associated with the first-episode of PEP at the University of Michigan. Finally, we collected data for prophylactic pancreatic stent placement,4 and three established procedural variables reported to predict PEP by at least two of the four large, multicenter prospective studies,2,8–10 including pancreatic sphincterotomy, MDC and ≥ 2 pancreatic duct contrast injections.
We defined cigarette-smoking and alcohol-use as current, former or never.11 Quantitative data was frequently absent for cigarette-smoking exposure and alcohol-use and inadequate for analysis. We used previous/common definitions for suspected SOD, pancreatic sphincterotomy, pancreatic injections, MDC and pancreatic stent placement.11
Chronic liver disease (CLD) is one of 19 variables contributing to a Charlson Comorbidity Score16 (see Table 1) and is defined by having a >6-month elevation in liver transaminases, alkaline phosphatase and/or bilirubin. We defined four categories (and 19 subcategories) of CLD (Table 2) and defined severity (from the Charlson Comorbidity Score16) as mild (for non-cirrhotic or cirrhotic disease and a Child-Turcotte-Pugh [CTP]18 score < 7) or moderate-severe (cirrhosis and a CTP score > 7).
Table 1.
Charlson comorbidity score - PEP
| Variable | Cases (%) | Control (%) | P-value |
|---|---|---|---|
| 1 Point | |||
| Previous myocardial infarction | 16/211 (7.6) | 36/348 (10.3) | 0.297 |
| Congestive heart failure | 6/211 (2.8) | 13/348 (3.7) | 0.638 |
| Peripheral vascular disease | 5/211 (2.4) | 5/348 (1.4) | 0.514 |
| Cerebrovascular disease | 4/211 (1.9) | 13/348 (3.7) | 0.311 |
| Dementia | 1/211 (0.5) | 3/348 (0.9) | 1.000 |
| COPD | 8/211 (3.8) | 20/348 (5.7) | 0.424 |
| Connective tissue disease | 8/211 (3.8) | 6/348 (1.7) | 0.164 |
| Ulcer disease | 20/211 (9.5) | 42/348 (12.1) | 0.405 |
| Mild liver disease | 32/211 (15.1) | 131/348 (37.6) | <0.0001 |
| Diabetes mellitus (mild) | 17/211 (8.1) | 56/348 (16.1) | 0.006 |
| 2 Points | |||
| Hemiplegia | 0/211 (0) | 0/348 (0) | N/A |
| Renal disease | 16/211 (7.6) | 33/348 (9.5) | 0.538 |
| Diabetes mellitus + organ damage | 3/211 (1.4) | 13/348 (3.7) | 0.125 |
| Any tumor | 29/211 (13.7) | 40/348 (11.5) | 0.430 |
| Leukemia | 1/211 (0.5) | 1/348 (0.3) | 1.000 |
| Lymphoma | 1/211 (0.5) | 5/348 (1.4) | 0.417 |
| 3 Points | |||
| Moderate/severe liver disease | 5/211 (2.4) | 14/348 (4.0) | 0.345 |
| 6 Points | |||
| Metastatic solid tumor | 10/211 (4.7) | 15/348 (4.3) | 0.835 |
| AIDS | 0/211 (0) | 1/348 (0.3) | N/A |
| *MEDIAN TOTAL SCORE (IQR) | 0 (0–2) | 1 (0–3) | <0.0001 |
Statistics:
Wilcoxon rank sum test, otherwise Fisher’s exact test.
IQR, interquartile range.
Table 2.
Categories of chronic liver disease - PEP
| Variable | Cases (%) | Control (%) | Total |
|---|---|---|---|
| Liver transplant/hepatectomy complications | 14/211 (6.6) | 71/348 (20.4) | 85/559 (15.2) |
| Biliary stricture | 9/211 (4.3) | 33/348 (9.5) | 42/559 (7.5) |
| Bile leak | 3/211 (1.4) | 18/348 (5.2) | 21/559 (3.8) |
| Cholangitis | 1/211 (0.5) | 6/348 (1.7) | 7/559 (1.3) |
| Abnormal liver tests | 1/211 (0.5) | 14/348 (4.0) | 15/559 (2.7) |
| Biliary disease/cancer (extra-hepatic) | 11/211 (5.2) | 45/348 (12.9) | 56/559 (10.0) |
| Primary sclerosing cholangitis | 7/211 (3.3) | 22/348 (6.3) | 29/559 (5.2) |
| Ampullary stenosis | 2/211 (0.9) | 7/348 (2.0) | 9/559 (1.6) |
| Biliary stricture | 1/211 (0.5) | 7/348 (2.0) | 8/559 (1.4) |
| Choledocholithiasis | 1/211 (0.5) | 4/348 (1.1) | 5/559 (0.9) |
| Cholangiocarcinoma | 0/211 (0) | 3/348 (0.9) | 3/559 (0.5) |
| Biliary metastases | 0/211 (0) | 2/348 (0.6) | 2/559 (0.4) |
| Non-biliary cancer | 3/211 (1.4) | 7/348 (2.0) | 10/559 (1.8) |
| Metastases | 2/211 (0.9) | 7/348 (2.0) | 9/559 (1.6) |
| Hematologic malignancy | 1/211 (0.5) | 0/348 (0) | 1/559 (0.2) |
| Other | 9/211 (4.3) | 22/348 (6.3) | 31/559 (5.5) |
| Non-alcoholic fatty liver disease | 2/211 (0.9) | 3/348 (0.9) | 5/559 (0.9) |
| Viral hepatitis | 4/211 (1.9) | 3/348 (0.9) | 7/559 (1.3) |
| Drug induced hepatitis | 0/211 (0) | 2/348 (0.6) | 2/559 (0.4) |
| Idiopathic | 2/211 (0.9) | 9/348 (2.6) | 11/559 (2.0) |
| Primary biliary cirrhosis | 0/211 (0) | 4/348 (1.1) | 4/559 (0.7) |
| Hepatic cysts/PCKD* | 0/211 (0) | 1/348 (0.3) | 1/559 (0.2) |
| Autoimmune hepatitis | 1/211 (0.5) | 0/348 (0) | 1/559 (0.2) |
PCKD, poly cystic kidney disease
Data Collection
We used methods identical to our previous study11 to pool and mask the case and control electronic medical records, to search the entire electronic medical record for data of interest, to reconcile variations in the data recorded in the electronic medical record, and to collect and tabulate data.
Statistical Analyses
The primary outcomes analyzed were PEP and severity of PEP, the latter graded as mild, moderate and severe according to the criteria by Cotton et al.15 Data was analyzed with Stata 12 statistical software (StataCorp, College Station, TX). Descriptive statistics were compiled to evaluate variable distribution prior to modeling. Univariate analyses were performed on all variables. Multivariate logistic regression analysis was also performed to identify independent predictors of PEP, using a stepwise forward technique with inclusion of variables with P<0.1 in the final model. Categorical and dichotomous variables were assessed by χ2 tests or Fischer’s exact test when necessary. Continuous variables were assessed using either two sample t-tests for normally distributed variables or the Wilcoxon rank sum test for non-normal variables.
RESULTS
Clinical characteristics of PEP and control groups
The most common indications for ERCP in the cohort were choledocholithiasis (24.3%), suspected SOD (22.9%), biliary stricture (9.7%), cyst/neoplasm (9.7%), abnormal liver function tests (8.8%), cholangitis (5.9%), bile leak (5.7%) and other categories (all < 5% each). The PEP case sample had a lower median Charlson score (Table 1) compared to the control sample, attributable to lower frequencies of diabetes mellitus and chronic liver disease (CLD). The two most common etiologic/anatomic categories of CLD (Table 2) were liver transplant or hepatectomy complication (CLD-transplant/hepatectomy complications) and extra-hepatic biliary tract disease (CLD-biliary).
Univariate Analysis of PEP
The PEP case sample differed from the control sample (Table 3) by having a lower mean age, more women and a higher frequency of variables known to be associated with PEP – suspected SOD, ≥ 2 pancreatic injections, pancreatic sphincterotomy and MDC. The PEP case sample also had a higher frequency of pancreatic stent placement, a lower median Charlson score and a lower frequency of former drinkers, current smokers, diabetes and CLD. The latter finding was attributable to a lower frequency of CLD-transplant/hepatectomy complications and CLD-biliary groups (Table 3).
Table 3.
Univariate Analysis - PEP
| Variables | Cases (%) | Control (%) | P-value |
|---|---|---|---|
| * Mean age, y (SD) | 47.5 (15.4) | 52.1 (16.1) | 0.0008 |
| Female | 158/211 (74.9) | 200/348 (57.5) | <0.0001 |
| Suspected SOD | 77/211 (36.5) | 51/348 (14.7) | <0.0001 |
| ≥ 2 Pancreatic injections | 10/211 (4.7) | 3/348 (0.01) | 0.006 |
| Pancreatic sphincterotomy | 41/211 (19.4) | 19/348 (5.5) | <0.0001 |
| Mod./diff. cannulation | 65/211 (30.8) | 36/348 (10.3) | <0.0001 |
| Pancreatic stent placement | 64/211 (30.3) | 52/348 (14.9) | <0.0001 |
| Alcohol-use status | |||
| Never drinker | 82/211 (38.9) | 125/348 (35.9) | Reference |
| Current drinker | 81/211 (38.4) | 106/348 (30.5) | 0.456 |
| Former drinker | 48/211 (22.7) | 117/348 (33.6) | 0.035 |
| Cigarette-smoking status | |||
| Never smoking | 131/211 (62.1) | 184/348 (52.9) | Reference |
| Current smoking | 23/211 (10.9) | 60/348 (17.2) | 0.022 |
| Former smoking | 57/211 (27.0) | 104/348 (29.9) | 0.192 |
| ** Median Charlson Score (IQR) | 0 (0–2) | 1 (0–3) | <0.0001 |
| Cardiovascular disease (Any below) | 22/211 (10.4) | 54/348 (15.5) | 0.099 |
| Cerebral vascular accident | 4/211 (3.3) | 13/348 (3.7) | 0.311 |
| Transient ischemic attack | 3/211 (1.4) | 7/348 (2.0) | 0.750 |
| Previous myocardial infarction | 16/211 (7.6) | 36/348 (10.3) | 0.297 |
| Angina | 2/211 (0.9) | 1/348 (0.3) | 0.560 |
| Congestive heart failure | 6/211 (2.8) | 13/348 (3.7) | 0.638 |
| Claudication | 3/211 (1.4) | 5/348 (1.4) | 1.000 |
| Coronary stent/angioplasty | 4/211 (1.9) | 10/348 (2.9) | 0.584 |
| Coronary bypass surgery | 6/211 (2.8) | 10/348 (2.9) | 1.000 |
| Hypertension | 47/211 (22.3) | 92/346 (26.6) | 0.268 |
| Diabetes mellitus | 20/211 (9.5) | 69/348 (19.8) | 0.001 |
| * Mean BMI (SD) | 28.2 (6.8) | 27.7 (6.6) | 0.411 |
| Aspirin use | 22/211 (10.4) | 51/348 (14.7) | 0.157 |
| NSAIDS | 10/211 (4.7) | 14/348 (4.0) | 0.673 |
| Chronic liver disease (CLD) | 37/211 (17.5) | 145/348 (41.7) | <0.0001 |
| Etiology of CLD | <0.0001 | ||
| No CLD | 174/211 (82.5) | 203/348 (58.3) | Reference |
| Transplant/hepatectomy complications | 14/211 (6.6) | 71/348 (20.4) | <0.001 |
| Biliary (extra-hepatic) | 11/211 (5.2) | 45/348 (12.9) | <0.001 |
| Non-biliary cancer | 3/211 (1.4) | 7/348 (2.0) | 0.162 |
| Other | 9/211 (4.3) | 22/348 (6.3) | 0.070 |
Statistics:
one way ANOVA,
Wilcoxon rank sum test, otherwise Fisher’s exact test.
IQR, interquartile range.
Multivariate Analysis of PEP
Multivariate forward stepwise regression analysis identified three independent predictors of protection against PEP (Table 4), including current smoking, CLD-biliary and CLD-transplant/hepatectomy complications, and four independent predictors for PEP, including younger age, suspected SOD, pancreatic sphincterotomy and MDC.
Table 4.
Multivariate Analysis – PEP (C-statistic = 0.73)
| Variable | OR | 95% CI | P-value |
|---|---|---|---|
| Pre-Endoscopic Predictors‡ | |||
| Suspected SOD | 2.1 | 1.3–3.3 | 0.001 |
| Current smoking | 0.4 | 0.2–0.8 | 0.004 |
| CLD-Transplant/hepatectomy complication | 0.4 | 0.2–0.7 | 0.004 |
| CLD-Biliary | 0.4 | 0.2–0.9 | 0.024 |
| Endoscopic Predictors | |||
| Pancreatic sphincterotomy | 2.8 | 1.5–5.3 | 0.001 |
| Moderate/difficult cannulation (MDC) | 3.7 | 2.3–5.9 | <0.001 |
Removing a 7th variable from the model, younger age (OR=0.986, 95% CI 0.974–0.998, P=0.027), did not significantly affect the predictive value (C-statistic decreased from 0.74 to 0.73) and reduced the complexity of generating the risk score (Table 5).
CLD, chronic liver disease
C-statistic, discrimination value of the model (area under the receiver operating characteristic (ROC) curve, or AUC)
Risk Model Scoring System
We derived two risk models based on six of the seven independent variables for PEP: three protective (current smoking, CLD-biliary, CLD-transplant/hepatectomy complications) and three predictive (suspected SOD, pancreatic sphincterotomy, MDC). The seventh independent variable, younger age, we excluded from the model because it had little impact on predicting PEP (C-statistic increased from 0.73 to 0.74) and added significant complexity to generating a risk score. Six independent predictors were assigned points based on the beta coefficients of the regression analysis3 (Table 5, top panel). Cumulative risk scores (Table 5, bottom panel) were divided into quantiles, three for the pre-ERCP model (<0, 0, 1) and four for the post-ERCP model (<0, 0, 1, ≥ 2). Odds ratios for PEP in the pre-and post-ERCP models ranged from 0.4–2.8 and 0.4–7.3, respectively. Figure 1 shows the probability for PEP for a given patient, assuming a baseline PEP risk of 3.5%.1
Table 5.
Risk Model Variables and Risk Stratification by Risk Score Quantiles – PEP
| Risk Model Variables | Points* (Assigned by Beta Coefficient) |
|---|---|
| Pre-Endoscopic Variables‡ | |
| Suspected SOD | 1 |
| Current smoking | −1 |
| CLD-transplant/hepatectomy complications | −1 |
| CLD-biliary | −1 |
| Endoscopic Variables | |
| Pancreatic sphincterotomy | 1 |
| Moderate/difficult cannulation (MDC) | 1 |
| Risk Score Quantiles | Case (%) | Control (%) | Odds of PEP | 95% CI | P-value | ||
|---|---|---|---|---|---|---|---|
| Pre-ERCP (−2 to 1)* | |||||||
| < 0 | 40/211 | (19.0) | 150/348 | (43.1) | 0.4 | 0.27–0.64 | <0.001 |
| 0** | 102/211 | (48.3) | 159/348 | (45.7) | **1.0 (ref) | ||
| 1 | 69/211 | (32.7) | 39/348 | (11.2) | 2.8 | 1.73–4.39 | <0.001 |
| Post-ERCP (−2 to 3)* | |||||||
| < 0 | 27/211 | (12.8) | 134/348 | (38.1) | 0.4 | 0.25–0.69 | 0.001 |
| 0** | 71/211 | (33.6) | 148/348 | (42.5) | **1.0 (ref) | ||
| 1 | 71/211 | (33.6) | 54/348 | (15.5) | 2.7 | 1.74–4.31 | <0.001 |
| ≥2 | 42/211 | (19.9) | 12/348 | (3.4) | 7.3 | 3.62–14.71 | <0.001 |
Removing a 7th variable from the model, younger age (OR=0.986, 95% CI 0.974–0.998, P=0.027), did not significantly affect the predictive value (C-statistic decreased from 0.74 to 0.73) and reduced the complexity of generating the risk score.
Risk score was calculated from cumulative points (Top Panel). Note: the lowest cumulative score is −2 (and not −3) because no patients have both CLD-biliary and CLD-transplant/hepatectomy complications.
Baseline category
CLD, chronic liver disease
Figure 1. Probability of PEP.
for an individual based on cumulative risk score (Table 4), assuming a baseline PEP risk of 3.5% 1.
PEP severity
PEP was mild in 22.7% (n=48), moderate in 66.8% (n=141) and severe in 10.4% (n=22), compared to a systematic review reporting respective frequencies of 45%, 44% and 11%.1 Across PEP severity groups there was no significant difference in median Charlson comorbidity scores or individual comorbidities (data not shown). By univariate analysis (Table 6) PEP severity associated only with the CVD criteria, previous myocardial infarction (MI) (P=0.038) and aspirin use (P=0.046). In the forward stepwise regression multivariate analyses, there were no significant predictors for PEP severity but aspirin use approached statistical significance (OR=2.5, P=0.069).
Table 6.
Univariate Analysis – PEP Severity
| Variable | Mild-PEP (%) | Moderate-PEP (%) | Severe PEP (%) | P-value |
|---|---|---|---|---|
| * Mean age, y (SD) | 46.7 (14.4) | 47.6 (15.9) | 49.0 (14.9) | 0.835 |
| Female | 34/48 (70.8) | 108/141 (76.6) | 16/22 (72.7) | 0.645 |
| Suspected SOD | 17/48 (35.4) | 51/141 (36.2 | 9/22 (40.9) | 0.927 |
| ≥ 2 Pancreatic injections | 0/48 (0) | 9/141 (6.4) | 1/22 (4.5) | 0.202 |
| Pancreatic sphincterotomy | 12/48 (25.0) | 25/141 (17.7) | 4/22 (18.2) | 0.547 |
| Mod./diff. cannulation | 12/48 (25.0) | 46/141 (32.6) | 7/22 (31.8) | 0.623 |
| Pancreatic stent placement | 17/48 (35.4) | 40/141 (28.4) | 7/22 (31.8) | 0.639 |
| Alcohol-use status | 0.213 | |||
| Never drinker | 22/48 (45.8) | 51/141 (36.2) | 9/22 (40.9) | |
| Current drinker | 13/48 (27.0) | 57/141 (40.4) | 11/22 (50.0) | |
| Former drinker | 13/48 (27.0) | 33/141 (22.4) | 2/22 (9.1) | |
| Cigarette-smoking status | 0.377 | |||
| Never smoking | 32/48 (66.7) | 84/141 (59.6) | 15/22 (68.2) | |
| Current smoking | 6/48 (12.5) | 17/141 (12.1) | 0/22 (0) | |
| Former smoking | 10/48 (20.8) | 40/141 (28.4) | 7/22 (31.8) | |
| ** Median Charlson Score (IQR) | 1.0 (0–2.5) | 0 (0–2) | 0 (0–1) | 0.660 |
| Cardiovascular disease (Any below) | 3/48 (6.2) | 14/141 (9.9) | 5/22 (22.7) | 0.115 |
| Cerebral vascular accident | 0/48 (0) | 4/141 (2.8) | 0/22 (0) | 0.727 |
| Transient ischemic attack | 0/48 (0) | 3/141 (2.1) | 0/22 (0) | 0.693 |
| Previous myocardial infarction | 3/48 (6.2) | 8/141 (5.7) | 5/22 (22.7) | 0.038 |
| Angina | 1/48 (2.1) | 1/141 (0.7) | 0/22 (0) | 0.555 |
| Congestive heart failure | 0/48 (0) | 6/141 (4.3) | 0/22 (0) | 0.417 |
| Claudication | 1/48 (2.1) | 1/141 (0.7) | 1/22 (4.5) | 0.153 |
| Cardiovascular stent/angioplasty | 2/48 (4.2) | 2/141 (1.4) | 0/22 (0) | 0.530 |
| Coronary bypass surgery | 0/48 (0) | 5/141 (3.5) | 1/22 (4.5) | 0.331 |
| Hypertension | 11/48 (22.9) | 32/141 (22.7) | 4/22 (18.2) | 0.934 |
| Diabetes | 6/48 (12.5) | 10/141 (7.1) | 1/22 (4.5) | 0.454 |
| * Mean BMI (SD) | 29.8 (7.6) | 27.5 (6.4) | 29.4 (7.3) | 0.325 |
| Aspirin use | 1/48 (2.1) | 17/141 (12.1) | 4/22 (18.2) | 0.046 |
| NSAIDS | 3/48 (6.2) | 5/141 (3.5) | 2/22 (9.1) | 0.246 |
| Chronic liver disease (CLD) | 11/48 (22.9) | 24/141 (17.0) | 2/22 (9.1) | 0.387 |
| Etiology of CLD | 0.900 | |||
| No CLD | 37/48 (77.1) | 117/141 (83.0) | 20/22 (90.9) | Reference |
| Transplant/hepatectomy complications | 5/48 (10.4) | 8/141 (5.7) | 1/22 (4.5) | |
| Biliary (extra-hepatic) | 3/48 (6.3) | 8/141 (5.7) | 0/22 (0) | |
| Non-biliary cancer | 1/48 (2.1) | 2/141 (1.4) | 0/22 (0) | |
| Other | 2/48 (4.2) | 6/141 (4.3) | 1/22 (4.5) |
Statistics:
one way ANOVA,
Wilcoxon rank sum test, otherwise Fisher’s exact test.
IQR, interquartile range.
DISCUSSION
By multivariate analysis we identified three independent predictors of protection against PEP and four predictors for PEP but no predictors for PEP severity. The most important new findings are the three predictors of protection, current smoking, CLD-biliary and CLD-transplant/hepatectomy complications. By including these variables with variables independently predictive of PEP (suspected SOD, pancreatic sphincterotomy and MDC) we derived two risk models. The pre-ERCP risk model has four variables (suspected SOD, current smoking, CLD-biliary and CLD-transplant/hepatectomy complications) and the post-ERCP model has all six (pre-ERCP variables plus pancreatic sphincterotomy and MDC). Removing age, a 7th independent variable from the model, did not significantly affect the predictive value (C-statistic=0.73) and significantly reduced the complexity of generating a risk score.
The predictive value for PEP (C-statistic=0.73) of our model is higher compared to the multi-regression logistic model from two (of the four) multicenter studies (C-statistic < 0.70).8,9 Two other studies developed predictive risk models for PEP. Escalating risk for PEP correlated with an increasing risk score derived by Friedland et al3 (from four variables) and with incremental inclusion of four variables identified by Freeman et al2 (women, normal serum bilirubin, difficult cannulation and suspected SOD). In contrast to our current study, however, the investigators did not report the discrimination value of the models (C-statistic, area under the receiver operating characteristic (ROC) curve, or AUC). In addition, these studies had less statistical power due to fewer PEP cases (n=103–131)2,3 and did not include investigational variables examined in the current study.
Similar to Freeman et al,2 our post-ERCP risk model includes suspected SOD and MDC. In contrast, our risk model did not include women, which was a significant variable only in the univariate analysis, but includes pancreatic sphincterotomy, CLD-biliary, CLD-transplant/hepatectomy complications and current smoking. We omitted two variables that added little power to the predictive model, including age (P=0.027) and multiple pancreatic injections (P=0.06). To confirm the generalizability of the risk model to community practice we report that a modified risk model that eliminates liver transplant patients (a tertiary care bias) retains similar predictive power.
Why women did not associate with PEP and younger age added little predictive value to the model is likely because both variables are surrogate markers of suspected SOD. Our data mirrors prior observations of suspected SOD patients,19,20 which were 89% women (predicts SOD with OR=6.8, P<0.001) and significantly younger (43.1 years +/−12.4 vs 52.5 years +/− 16.5, P<0.0001). Hence, these observations would explain why none of the four multicenter prospective studies reported all three variables (women, younger age, suspected SOD) as independent predictors for PEP.2,8–10
In the current, large study (n=211 cases of PEP) we found that current smoking is independently protective against PEP. This finding contrasts with a recent report that active smoking is an independent predictor of PEP, but the study is underpowered (n=36 cases of PEP) and investigators omitted established PEP variables from the analysis (>/= 2 pancreatic injections, moderate/difficult cannulation, suspected SOD, prior history of PEP).21 The importance of a sufficiently powered study is underscored as we previously reported that current cigarette-smoking is protective against PEP in univariate analysis (P=0.001) but not in multivariate analysis (P=0.055),11 in an underpowered study of 123 PEP cases.
That current smoking protects against PEPseems to contradict clinical observations that smoking is an independent, dose-dependent risk factor for acute pancreatitis22–24 and with experimental data that cigarette smoke increases pancreatic inflammation25,26 and reduces pancreatic blood flow.27 There are at least three hypotheses why current smoking reduces PEP risk. Nicotine (present in cigarettes) may activate the nicotinic anti-inflammatory pathway,28 which reduces pancreatic inflammation and severity of experimental pancreatitis,29 thus mirroring the anti-inflammatory effects of this pathway observed in ulcerative colitis30 and in other conditions.28 Nicotine also relaxes the sphincter of Oddi in experimental models,31 which may reduce sphincter spasm and obstruction as a cause of PEP, and may also have direct, protective effects by reducing secretagogue-evoked cell necrosis in dispersed pancreatic acini.32
The relationship between biliary and/or liver disease and PEP is not clear from previous studies which mostly evaluated the association of bilirubin levels to PEP. Freeman et al2 reported that normal serum bilirubin doubled the risk for PEP. Data from three other large multicenter prospective studies conflicts with this observation,8–10 but could be attributable to using different surrogate markers for normal serum bilirubin. In the current study, there was a significantly lower frequency of PEP in mild-and moderate/severe CLD (see definitions in Methods). By univariate and multivariate analyses, CLD-biliary and CLD-transplant/hepatectomy complications associated with a lower frequency of PEP. This protective relationship may have occurred because both CLD variables predicted lower odds of MDC: CLD-biliary (OR=0.4, 95% CI 0.1–0.9, P=0.037) and CLD-transplant/hepatectomy complications (OR=0.3, 95% CI 0.1–0.7, P=0.004). One possible explanation for easier biliary cannulation is that liver transplantation at the University of Michigan involves dilating the papilla (with a dilating catheter in the operating room) before constructing a biliary duct-to-duct anastomosis over a 5 Fr “stent” crossing the papilla. Why biliary cannulation is easier in CLD-biliary is less clear. Although 52% of this group had primary sclerosing cholangitis (PSC), the frequency of PSC was similar in the control and PEP groups.
The results of our study do not support the hypothesis that CVD associates with PEP or PEP severity. In the univariate analysis of PEP there was a nonsignificant trend towards less frequent CVD in the PEP case sample compared to the control sample (10.4% vs 15.5%; P=0.099). The univariate analysis of PEP severity, however, yielded a significant association of PEP severity with previous MI and aspirin use (the latter we interpret as a marker of CVD and/or previous MI), but no significant variable emerged from the multivariate analysis. There is a possibility of a type II error (false negative association) because only 22 patients had severe PEP.
Although an observational retrospective case control study has inherent disadvantages such as confounding by unknown predictors, missing data and incorrect reporting of the predictors, our study has major strengths including the size and statistical power and the a priori selection of established variables and numerous hypothesis driven variables related to CVD. With the exception of a high rate of missing data for BMI (22.9% [128/559]) due to missing height data, all other variables had an acceptable rate of missing data < 5%.33 The completeness of the data collection was due to the quality of the electronic documentation and the use of the EMERSE search engine to systematically identify data within the electronic medical record.11 The collection of the smoking data was interview-based,11 which is more reliable than self-reporting34 and the accuracy of the smoking data is implied by noting that the smoking-linked disease Chronic Obstructive Pulmonary Disease (COPD) was far more frequent in former smokers (10.6–12.3%) and current smokers (4.3–10.0%) compared to never smokers (0.0–1.6%).
Additional evidence that our data is reliable is that we confirmed that suspected SOD,2,8 pancreatic sphincterotomy,2,10 and MDC2,10 were independent predictors for PEP. Also, we showed that prophylactic pancreatic stent placement4 occurred significantly more frequently in the case sample (30.3% vs 14.9%) by univariate but not multivariate analyses, likely because stenting was a surrogate marker of other predictors for PEP. Similar to our previous study,11 and in contrast to three of the prospective multicenter studies,2,8,9 ≥ 2 pancreatic duct contrast injections was not a significant risk factor for PEP, likely because of non-standardized reporting of injections on procedure reports and flow sheets.
It has become evident that placement of a prophylactic pancreatic duct stent4 or administering indomethacin suppositories immediately after ERCP5,6 lowers the risk of PEP in “high-risk” groups. Investigators, however, use different variables to define “high-risk”, including “biliary sphincterotomy for SOD, difficult cannulation, precut sphincterotomy, pancreatic sphincterotomy, biliary balloon dilation of intact papilla for stone extraction, endoscopic ampullectomy, pancreatic brush cytology, all ERCPs that were considered to be at high risk,”35 a prior history of PEP, or two or more out of 5 minor criteria.6 Hence to determine which patients would benefit from prophylactic measures requires better definitions of high-risk patients.
In summary, we report three new predictors of protection against PEP (current smoking, CLD-biliary and CLD-transplant/hepatectomy complications) and four independent predictors for PEP (younger age, suspected SOD, pancreatic sphincterotomy, MDC). We derive two risk models using six of the seven independent variables (after omitting age) and show an improved predictive value for PEP (C-statistic = 0.73) compared to prior studies. Our findings are potentially important by identifying overlooked patient predictors of protection against PEP and by generating risk models that conceivably would allow risk assessment pre-ERCP. Prospective study is required to both validate the risk models (to address limitations of the retrospective study design) and to determine whether prophylactic measures lower rates of PEP in high-risk groups defined by our risk models.
Acknowledgments
Financial support: SDS receives research support from the VA HSR&D CDA-2 Career Development Award. MJD receives research support from the National Institutes of Health (K08 DK073298, R21 AA017271), the Michigan Institute for Clinical and Health Research (MICHR) and the University of Michigan Office of Vice President of Research (OVPR).
Abbreviations
- AIDS
acquired immune deficiency syndrome
- BMI
body mass index
- CI
confidence interval
- CLD
chronic liver disease
- CVD
cardiovascular disease
- ERCP
endoscopic retrograde cholangiopancreatography
- IDX
information data exchange
- ICD-9
international classification of diseases 9th edition
- MDC
moderate/difficult cannulation
- NSAIDs
non-steroidal anti-inflammatory drugs
- PEP
post-ERCP pancreatitis
- OR
odds ratio
- SOD
Sphincter of Oddi
Footnotes
Disclosures: MJD received the drug Pioglitazone from Takeda Pharmaceuticals North America for use in an NIH sponsored clinical research trial (2008). MJD received honoraria from Lippincott Williams & Wilkins (Philadelphia, PA, USA) for articles published in Current Opinion in Gastroenterology, Springer (New York, NY, USA) for an article published in Current Gastroenterology Reports and the British Medical Journal (BMJ) Publishing Group Limited for articles published in BMJ Point of Care. MJD also received a consulting fee from MD Evidence (Atlantic City, NJ, USA) for co-authoring a systematic review (2009) entitled, “Systematic review: Pancreatic enzyme treatment for malabsorption associated with chronic pancreatitis”.
References
- 1.Andriulli A, Loperfido S, Napolitano G, et al. Incidence rates of post-ERCP complications: a systematic survey of prospective studies. Am J Gastroenterol. 2007 Aug;102(8):1781–1788. doi: 10.1111/j.1572-0241.2007.01279.x. [DOI] [PubMed] [Google Scholar]
- 2.Freeman ML, DiSario JA, Nelson DB, et al. Risk factors for post-ERCP pancreatitis: a prospective, multicenter study. Gastrointest Endosc. 2001;54(4):425–434. doi: 10.1067/mge.2001.117550. [DOI] [PubMed] [Google Scholar]
- 3.Friedland S, Soetikno RM, Vandervoort J, et al. Bedside scoring system to predict the risk of developing pancreatitis following ERCP. Endoscopy. 2002 Jun;34(6):483–488. doi: 10.1055/s-2002-32004. [DOI] [PubMed] [Google Scholar]
- 4.Choudhary A, Bechtold ML, Arif M, et al. Pancreatic stents for prophylaxis against post-ERCP pancreatitis: a meta-analysis and systematic review. Gastrointestinal endoscopy. 2011 Feb;73(2):275–282. doi: 10.1016/j.gie.2010.10.039. [DOI] [PubMed] [Google Scholar]
- 5.Elmunzer BJ, Waljee AK, Elta GH, et al. A meta-analysis of rectal NSAIDs in the prevention of post-ERCP pancreatitis. Gut. 2008 Sep;57(9):1262–1267. doi: 10.1136/gut.2007.140756. [DOI] [PubMed] [Google Scholar]
- 6.Elmunzer BJ, Scheiman JM, Lehman GA, et al. A randomized trial of rectal indomethacin to prevent post-ERCP pancreatitis. The New England journal of medicine. 2012 Apr 12;366(15):1414–1422. doi: 10.1056/NEJMoa1111103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Feurer ME, Adler DG. Post-ERCP pancreatitis: review of current preventive strategies. Current Opinion in Gastroenterology. 2012 May;28(3):280–286. doi: 10.1097/MOG.0b013e3283528e68. [DOI] [PubMed] [Google Scholar]
- 8.Cheng CL, Sherman S, Watkins JL, et al. Risk factors for post-ERCP pancreatitis: a prospective multicenter study. Am J Gastroenterol. 2006 Jan;101(1):139–147. doi: 10.1111/j.1572-0241.2006.00380.x. [DOI] [PubMed] [Google Scholar]
- 9.Williams EJ, Taylor S, Fairclough P, et al. Risk factors for complication following ERCP; results of a large-scale, prospective multicenter study. Endoscopy. 2007 Sep;39(9):793–801. doi: 10.1055/s-2007-966723. [DOI] [PubMed] [Google Scholar]
- 10.Wang P, Li ZS, Liu F, et al. Risk factors for ERCP-related complications: a prospective multicenter study. Am J Gastroenterol. 2009 Jan;104(1):31–40. doi: 10.1038/ajg.2008.5. [DOI] [PubMed] [Google Scholar]
- 11.Debenedet AT, Raghunathan TE, Wing JJ, et al. Alcohol use and cigarette smoking as risk factors for post-endoscopic retrograde cholangiopancreatography pancreatitis. Clin Gastroenterol Hepatol. 2009 Mar;7(3):353–358e354. doi: 10.1016/j.cgh.2008.11.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Stumpf HH. Microatheromas of very small arteries: unusual lesions involving primarily the pancreas. Hum Pathol. 1983 Dec;14(12):1039–1043. doi: 10.1016/s0046-8177(83)80259-7. [DOI] [PubMed] [Google Scholar]
- 13.Pollak OJ. Human pancreatic atherosclerosis. Ann N Y Acad Sci. 1968 Nov 21;149(2):928–939. doi: 10.1111/j.1749-6632.1968.tb53847.x. [DOI] [PubMed] [Google Scholar]
- 14.Klar E, Messmer K, Warshaw AL, et al. Pancreatic ischaemia in experimental acute pancreatitis: mechanism, significance and therapy. Br J Surg. 1990 Nov;77(11):1205–1210. doi: 10.1002/bjs.1800771104. [DOI] [PubMed] [Google Scholar]
- 15.Cotton PB, Lehman G, Vennes J, et al. Endoscopic sphincterotomy complications and their management: an attempt at consensus. Gastrointest Endosc. 1991 May-Jun;37(3):383–393. doi: 10.1016/s0016-5107(91)70740-2. [DOI] [PubMed] [Google Scholar]
- 16.Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–383. doi: 10.1016/0021-9681(87)90171-8. [DOI] [PubMed] [Google Scholar]
- 17.Anderson KM, Wilson PW, Odell PM, et al. An updated coronary risk profile. A statement for health professionals. Circulation. 1991 Jan;83(1):356–362. doi: 10.1161/01.cir.83.1.356. [DOI] [PubMed] [Google Scholar]
- 18.Pugh RN, Murray-Lyon IM, Dawson JL, et al. Transection of the oesophagus for bleeding oesophageal varices. The British journal of surgery. 1973 Aug;60(8):646–649. doi: 10.1002/bjs.1800600817. [DOI] [PubMed] [Google Scholar]
- 19.Geenen JE, Hogan WJ, Dodds WJ, et al. The efficacy of endoscopic sphincterotomy after cholecystectomy in patients with sphincter-of-Oddi dysfunction. N Engl J Med. 1989 Jan 12;320(2):82–87. doi: 10.1056/NEJM198901123200203. [DOI] [PubMed] [Google Scholar]
- 20.Kozarek RA. Biliary dyskinesia - are we any closer to defining the entity? Gastrointest Endosc Clin North Am. 1993;3:167–178. [Google Scholar]
- 21.Li N, Tieng A, Novak S, et al. Effects of medications on post-endoscopic retrograde cholangiopancreatography pancreatitis. Pancreatology: official journal of the International Association of Pancreatology. 2010;10(2–3):238–242. doi: 10.1159/000279306. [DOI] [PubMed] [Google Scholar]
- 22.Tolstrup JS, Kristiansen L, Becker U, et al. Smoking and risk of acute and chronic pancreatitis among women and men: a population-based cohort study. Arch Intern Med. 2009 Mar 23;169(6):603–609. doi: 10.1001/archinternmed.2008.601. [DOI] [PubMed] [Google Scholar]
- 23.Lindkvist B, Appelros S, Manjer J, et al. A prospective cohort study of smoking in acute pancreatitis. Pancreatology. 2008 Jan 31;8(1):63–70. doi: 10.1159/000114868. [DOI] [PubMed] [Google Scholar]
- 24.Sadr-Azodi O, Andren-Sandberg A, Orsini N, et al. Cigarette smoking, smoking cessation and acute pancreatitis: a prospective population-based study. Gut. 2012 Feb;61(2):262–267. doi: 10.1136/gutjnl-2011-300566. [DOI] [PubMed] [Google Scholar]
- 25.Wittel UA, Pandey KK, Andrianifahanana M, et al. Chronic pancreatic inflammation induced by environmental tobacco smoke inhalation in rats. Am J Gastroenterol. 2006 Jan;101(1):148–159. doi: 10.1111/j.1572-0241.2006.00405.x. [DOI] [PubMed] [Google Scholar]
- 26.Jianyu H, Guang L, Baosen p. Evidence for cigarette smoke-induced oxidative stress in the rat pancreas. Inhal Toxicol. 2009 Oct;21(12):1007–1012. doi: 10.1080/08958370802665937. [DOI] [PubMed] [Google Scholar]
- 28.Hartwig W, Werner J, Ryschich E, et al. Cigarette smoke enhances ethanol-induced pancreatic injury. Pancreas. 2000;21(3):272–278. doi: 10.1097/00006676-200010000-00009. [DOI] [PubMed] [Google Scholar]
- 28.Tracey KJ. Physiology and immunology of the cholinergic antiinflammatory pathway. J Clin Invest. 2007 Feb;117(2):289–296. doi: 10.1172/JCI30555. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.van Westerloo DJ, Giebelen IA, Florquin S, et al. The vagus nerve and nicotinic receptors modulate experimental pancreatitis severity in mice. Gastroenterology. 2006 Jun;130(6):1822–1830. doi: 10.1053/j.gastro.2006.02.022. [DOI] [PubMed] [Google Scholar]
- 30.McGrath J, McDonald JW, Macdonald JK. Transdermal nicotine for induction of remission in ulcerative colitis. Cochrane Database Syst Rev. 2004;(4):CD004722. doi: 10.1002/14651858.CD004722.pub2. [DOI] [PubMed] [Google Scholar]
- 31.Bagcivan I, Kaya T, Turan M, et al. Investigation of the mechanism of nicotine-induced relaxation on the sheep sphincter of Oddi. Can J Physiol Pharmacol. 2004 Nov;82(11):935–939. doi: 10.1139/y04-073. [DOI] [PubMed] [Google Scholar]
- 32.DiMagno MJ, Shah A, Tesmer T, et al. Cigarette smoke extract (CSE) but not nicotine enhances secretagogue evoked pancreatic acinar necrosis. Pancreas. 2010;39(8):A1318. [Google Scholar]
- 33.Little RJA, Rubin DB. Statistical Analysis With Missing Data. 2. Hoboken, New Jersey: John Wiley & Sons, Inc; 2002. [Google Scholar]
- 34.Patrick DL, Cheadle A, Thompson DC, et al. The validity of self-reported smoking: a review and meta-analysis. Am J Public Health. 1994 Jul;84(7):1086–1093. doi: 10.2105/ajph.84.7.1086. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Freeman ML. Pancreatic stents for prevention of post-endoscopic retrograde cholangiopancreatography pancreatitis. Clin Gastroenterol Hepatol. 2007 Nov;5(11):1354–1365. doi: 10.1016/j.cgh.2007.09.007. [DOI] [PubMed] [Google Scholar]