Acute pancreatitis is the most common major complication of endoscopic retrograde cholangiopancreatography (ERCP).1 2 The incidence of post-ERCP pancreatitis (PEP) is generally 1–10%,3^–7 although it may exceed 25% in certain high-risk patient populations.8 While most patients develop mild pancreatitis, a small percentage will follow a severe course, resulting in prolonged hospitalisation, anatomic complications or the need for further intervention.5 6 9

Several mechanical and pharamacological interventions have been evaluated in the prevention of PEP. Prophylactic pancreatic stents have become standard of care for reducing PEP in high-risk cases.10 11 In contrast, trials of pharmacological therapy have generally yielded disappointing results. Intravenous gabexate and somatostatin have shown some promise, although these agents require continuous infusion and are not commercially available worldwide.12 13 To date, no pharmacological prophylaxis for PEP is in widespread clinical use.

Non-steroidal anti-inflammatory drugs (NSAIDs) are potent inhibitors of prostaglandins, phospholipase A2 and neutrophil–endothelial interaction, all believed to play an important role in the pathogenesis of acute pancreatitis.14^–16 It has been demonstrated in animal models that NSAIDs can reduce mortality associated with acute pancreatitis.17^–19 Concordantly, a small human study of indomethacin in acute pancreatitis showed improvement in clinical outcomes.20 NSAIDs are inexpensive, easily administered and have a favourable risk profile when given as a one-time dose, making them an attractive option in the pharmacological prevention of PEP.

Despite these anticipated benefits, prospective clinical trials evaluating NSAIDs in preventing PEP have yielded inconclusive results. The purpose of this meta-analysis is to assess the overall effect of prophylactic rectally administered NSAIDs on PEP.

METHODS

The study was conducted according to standard guidelines.21 A computer-assisted search with the OVID interface to Medline and Embase was conducted to identify potentially relevant papers. A search of these databases from 1966 to 31 December 2007 was performed using the exploded (exp) medical subject heading (MESH) terms exp pancreatitis OR exp post-ERCP pancreatitis AND the exploded terms: non-steroidal anti-inflammatory drugs, diclofenac, indomethacin, aspirin, ibuprofen, naproxen, ketorolac, etodolac, sulindac, COX-2 inhibitors, celecoxib, rofecoxib, valdecoxib. Manual searches of reference lists from potentially relevant papers were performed to identify any additional studies that may have been missed using the computer-assisted strategy. Additionally, all abstracts from the American Gastroenterologic Association, American Society of Gastrointestinal Endoscopy, American College of Gastroenterology and United European Gastroenterology Week 2002–2007 were reviewed for potentially relevant abstracts. The results were limited to human studies.

Three investigators (BJE, JT and SF) independently reviewed titles and abstracts of all citations identified by the literature search. Potentially relevant studies were retrieved and selection criteria applied. The selection criteria were: (1) studies that examine the effect of rectally administered prophylactic NSAIDs on the incidence of PEP; (2) studies that were prospective, randomised and placebo controlled; (3) studies in humans; and (4) data not duplicated in another manuscript. Eligible articles were reviewed and data were abstracted in a duplicate and independent manner by two investigators (BJE and AW). Agreement between investigators was 100%.

Included studies were assessed for methodological quality using criteria set forth by the Evidence-Based Gastroenterology Steering Group.22 These criteria were: (1) concealed random allocation; (2) blinding of patients and caregivers; (3) equal use of co-interventions for treatment and placebo group; (4) complete follow-up of study patients; (5) and use of an intention to treat analysis.

Meta-analysis using the metan command in Stata 9.2 with a fixed effects model and random effects model was performed. Significant heterogeneity was not detected, and therefore fixed effect summary results are reported for the four studies. We evaluated two end points, the overall incidence of pancreatitis and the incidence of moderate to severe pancreatitis, in two separate meta-analyses.

RESULTS

Characteristics of selected studies

Searches of the Medline and EMBASE databases, bibliographies of relevant manuscripts, and conference abstracts yielded 258 potentially relevant articles. Of these, title and abstract review yielded five potentially appropriate articles.23^–27 Four studies met inclusion criteria. A prospective trial of oral diclofenac in predominantly high-risk patients27 was excluded because of the different route of administration of NSAIDs and the statistical heterogeneity it would have introduced to the meta-analysis (see the Discussion section). Characteristics of the four included randomised controlled trials (RCTs) are listed in table 1. Quality assessment revealed that two included studies met 5/5 criteria listed above,23 25 one study met 4/5 criteria24 and one study met 3/5 criteria,26 indicating that all studies were of reasonable methodological quality; none of the studies had any “fatal” methodological flaws.

In 2003, Murray et al conducted a randomised prospective trial in Scotland comparing 100 mg of rectal diclofenac with placebo in the prevention of PEP. To select for a high-risk study sample, only patients undergoing pancreatography or those with manometrically documented sphincter of Oddi dysfunction were enrolled. Diclofenac or placebo suppositories were administered upon arrival in the recovery area. A total of 110 patients were enrolled in each arm, suggested by the authors to be an adequate sample, although 20 additional patients were enrolled after preliminary results analysis did not demonstrate a statistically significant benefit. The overall incidence of pancreatitis in this trial was 10.9%, slightly higher than the expected overall incidence of PEP, reflecting the study sample. The incidence of pancreatitis in the placebo group was 15.5% (17/110), whereas it was 6.4% in the diclofenac group (7/110). This difference was statistically significant, with a p value of 0.049. Two patients in the control group (1.8%) and none in the diclofenac group developed severe pancreatitis.

In 2007, Sotoudehmanesh et al published a prospective randomised trial from Iran that compared 100 mg of rectally administered indomethacin with placebo in reducing the incidence of PEP. They enrolled a heterogeneous group of 442 patients without selecting for a particular risk group. A sample size was calculated a priori using a power of 90%, α = 0.05, and the event rates observed in the study of Murray et al. An indomethacin or placebo suppository was administered immediately prior to ERCP. The overall incidence of pancreatitis in the study was 4.9%. Fewer than 10% of enrolled patients had sphincter of Oddi dysfunction, perhaps explaining the lower rate of PEP. The incidence of PEP in the placebo group was 6.8% (15/221) whereas the incidence of PEP in the indomethacin group was 3.2% (7/221). This difference was not statistically significant (p = 0.06). A post hoc subgroup analysis revealed a statistically significant protective effect of indomethacin in patients undergoing pancreatic duct injection (p = 0.014). Seven placebo patients developed moderate to severe pancreatitis (2.3%); none of the indomethacin patients developed this outcome.

Also in 2007, Khoshbaten et al published an RCT from Iran comparing 100 mg of rectal diclofenac with placebo in 100 patients undergoing high-risk ERCP. To select high-risk cases, only those undergoing a pancreatogram (± cholangiogram) were enrolled. The study drug or placebo was administered on arrival in the recovery area. The overall incidence of PEP was 15%. The incidence of PEP in the placebo group was 26% (13/50) whereas the incidence of PEP in the diclofenac group was 4% (2/50). This difference was statistically significant, with a p value of <0.01. The study was prematurely terminated after the first 100 patients because of ethical concerns associated with not using prophylactic diclofenac. No patients in this clinical trial developed necrotising pancreatitis or required surgical intervention.

Again in 2007, Montaño Loza et al reported a randomised prospective clinical trial from Mexico comparing indomethacin with placebo in the prevention of PEP. They enrolled patients undergoing ERCP for suspected bile duct obstruction rather than selecting for a high-risk cohort. Rectal indomethacin (100 mg) or placebo was administered prior to ERCP. Seventy-five patients were randomised to each group, a sample size that was calculated a priori to detect a 15% reduction in PEP. The overall incidence of pancreatitis was 10.7%. The incidence of PEP was 16% (12/75) in the placebo group and 5.3% (4/75) in the indomethacin group. This difference was statistically significant, with a p value of 0.034. All pancreatitis cases in both groups were categorised as mild.

Three of the four component studies do not report use of prophylactic pancreatic stents. In the study by Murray et al, 13 pancreatic stents were placed in the diclofenac group and 12 were placed in the placebo group, although it is not specified whether these stents were placed for prophylactic or therapeutic indications.

No adverse events attributable to NSAIDs were reported in any of the trials.

Meta-analysis results

The fixed effects meta-analysis for total pancreatitis had a heterogeneity χ² value of 1.80 (p = 0.615), demonstrating no evidence of statistical heterogeneity within this group of clinical trials. The Mantel–Haenszel pooled relative risk (RR) for PEP after prophylactic administration of NSAIDs was 0.36 (95% CI 0.22 to 0.60) (fig 1). The pooled number of ERCP patients needed to treat (NNT) with NSAIDs to prevent one episode of pancreatitis was 15.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 The four trials of rectal non-steroidal anti-inflammatory drugs (NSAIDs) are presented in a meta-analysis. The p value for heterogeneity was 0.615, and the summary relative risk was 0.36 (95% CI 0.22 to 0.60) for post-endoscopic retrograde cholangiopancreatography pancreatitis in the rectal NSAID arm.

We also performed a meta-analysis of a second end point, the incidence of moderate to severe PEP after administration of prophylactic NSAIDs or placebo. Seven control patients developed moderate to severe pancreatitis (1.5%) whereas none of the NSAID patients was reported to develop this same outcome (0%). The fixed effects meta-analysis for this end point had a heterogeneity χ² value of 0.30 (p = 0.587). The RR of moderate to severe PEP after NSAID administration was 0.10 (95% CI 0.01 to 0.76) (fig 2). This represents a 90% reduction in moderate to severe pancreatitis, with a pooled number of ERCP patients needed to treat to prevent one episode of moderate to severe pancreatitis of 38.8.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 The two trials of rectal non-steroidal anti-inflammatory drugs (NSAIDs) which reported moderate to severe post-endoscopic retrograde cholangiopancreatography pancreatitis (PEP) events are presented in a meta-analysis. The p value for heterogeneity was 0.587, and the summary relative risk was 0.10 (95% CI 0.01 to 0.76) for severe PEP in the rectal NSAID arm. The confidence interval is quite wide due to the fact that only two trials reported severe PEP events, and this should only be considered a very rough estimate of the effect of NSAIDs on severe PEP events.

DISCUSSION

The four prospective RCTs evaluating the protective effect of rectally administered NSAIDs on PEP have yielded statistically inconclusive results. All four of these relatively small studies, however, demonstrate similar trends toward benefit. In determining whether it is appropriate to combine the estimates of effects from clinical trials in a meta-analysis, a statistical test of heterogeneity is performed. If the estimates from the clinical trials are heterogeneous, this test will result in a low p value (<0.1), suggesting that combining the effects from the trials may result in a misleading overall estimate. When comparing the estimates from the clinical trials included in this meta-analysis, a test of heterogeneity had a p value of 0.615, suggesting no evidence of heterogeneity and supporting the calculation of a combined estimate. However, it should be noted that, as only four trials were included, this test is probably underpowered.

Meta-analysis of the four included clinical trials demonstrates that the RR of developing PEP after prophylactic NSAIDs is 0.34 (95% CI 0.22 to 0.60). In other words, patients who received diclofenac or indomethacin in the periprocedure period had a 64% lower risk of pancreatitis. The NNT to prevent one episode of pancreatitis is 15. This meta-analysis compares the outcomes of 456 NSAIDs-treated patients and 456 control patients, for a total of 912 patients. Assuming a two-tailed α = 0.05, a power of 0.9 and PEP incidences of 12.06% and 4.38% in the placebo and NSAIDs groups, respectively (based on the pooled PEP incidences in the four studies included in this meta-analysis), a total of 586 patients would be required to demonstrate the intended decrease in the incidence of PEP. Therefore, in contrast to the individual included trials, this meta-analysis had an adequate pooled sample size to detect a statistically significant benefit, assuming these incidences represent the true rates of PEP in the population of patients undergoing ERCP.

No patient who received prophylactic NSAIDs developed moderate to severe PEP; all cases occurred in those receiving placebo. NSAID-treated patients had a 90% reduction in this outcome, raising the question of whether prophylactic NSAIDs will have their greatest clinical impact in the reduction of risk for severe pancreatitis.

Murray et al and Khoshbaten et al demonstrated a statistically significant benefit in patients at high risk for PEP. Similarly, post hoc analysis of the data of Sotoudehmanesh et al found benefit only in those undergoing pancreatic duct injection, a known risk factor for PEP.5 This combination of results suggests that prophylactic NSAIDs may produce the largest absolute benefit in certain high-risk patients or procedures.

The findings of this meta-analysis are potentially of major economic impact. For example, approximately 750 ERCPs are performed at our institution annually. Assuming a PEP risk of 5%, then 38 cases of PEP occur each year. US Medicare reimbursement for PEP is approximately US$5700, resulting in a total annual cost of US$216 600 for this complication. The average cost of one dose of rectal indomethacin (or diclofenac) is approximately US$2.00 (or US$1.25), resulting in a total annual cost of US$1500 (or US$938). Routine administration of these agents would have reduced the total number of incidences of PEP to approximately 13, decreasing the total annual cost to US$75 600 (US$74 100+US$1500), a saving of approximately US$141 000 or €96 000 annually. This cost reduction is substantial, especially when considered on an international level.

No clinically relevant adverse events attributable to NSAIDs were described in the approximately 450 patients who received them. In particular, post-procedure haemorrhage did not occur despite a 56% sphincterotomy rate (256 of 456 patients). This finding is congruent with previously published data suggesting that NSAIDs in standard doses do not increase the risk of significant bleeding after biliary sphincterotomy.4 28

Of note is that patients with contraindications to NSAIDs, such as renal failure and active peptic ulcer disease, were excluded from these clinical trials. The risk of exacerbating these conditions as well as precipitating cardiovascular events with prophylactic NSAIDs should be considered as part of the overall risk–benefit ratio, particularly in patients prone to these outcomes.

Whether the route of delivery of prophylactic NSAIDs affects clinical efficacy is an area of uncertainty. A recent study by Cheon et al demonstrated no difference between oral diclofenac and placebo in preventing PEP in 207 predominantly high-risk patients.26 Our meta-analysis remains positive (albeit less powerful) even if this study is included, with an RR of pancreatitis of 0.51 (95% CI 0.35 of 0.74) (fig 3). The addition of this study to the meta-analysis, however, decreases the heterogeneity p value from 0.615 to 0.117, indicating that this study should not be combined with the others in a single estimate. A l’Abbe’s plot reveals that the Cheon study is indeed an outlier that would compromise the statistical integrity of the meta-analysis if included (fig 4).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 The five trials of prophylactic non-steroidal anti-inflammatory drugs (NSAIDs) are presented in a meta-analysis. The p value for heterogeneity was 0.117, and the summary relative risk was 0.51 (95% CI 0.35 to 0.74) for post-endoscopic retrograde cholangiopancreatography pancreatitis in the NSAID arm. However, the low p value for heterogeneity indicates that the trials have statistically different outcomes and should not be combined in a meta-analysis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4 The five trials of prophylactic non-steroidal anti-inflammatory drugs (NSAIDs) are presented in a L’Abbe plot, which compares the effect size in the active and placebo arms. The oral NSAID study (Cheon) lies near the no effect line, while a linear regression on the remaining four studies forms a distinct line, which supports the finding that the outcome of the Cheon et al study comes from a distribution different from the rectal NSAID studies. PEP, post-endoscopic retrograde cholangiopancreatography pancreatitis.

From a clinical standpoint, all four studies evaluating rectal NSAIDs in preventing PEP are positive or demonstrate a trend toward positivity, whereas the single published study evaluating oral NSAIDs is negative. One potential explanation for this difference in outcomes is that peak plasma concentrations of NSAIDs are reached in 30 min when administered rectally compared with 2 h when administered orally.29 Perhaps the more rapid onset of action of rectal NSAIDs makes this route better suited to inhibit the initial pathophysiological events that eventually result in pancreatitis after ERCP. To elucidate this and other potential mechanistic differences further, additional studies addressing the route of administration of prophylactic NSAIDs are required.

This meta-analysis answers an important clinical question with studies that exhibit strong statistical homogeneity despite methodological differences in design. Murray et al and Khoshbaten et al enrolled only patients at high risk for PEP, whereas the other two studies enrolled all-comers, regardless of patient- or procedure-related risk. Despite this difference, there is homogeneity of the odds ratios for the protective effect of NSAIDs in these four trials. This finding is somewhat surprising, as one might expect a larger relative benefit in trials selecting for high-risk patients. One possible explanation for this homogeneity in odds ratios could be that these four trials, despite differing inclusion criteria, simply enrolled subjects at similar risk for PEP. However, the rate of PEP in the placebo group of the all-comers trials was 8.3% (23/277) whereas the rate of PEP in the placebo group of high-risk trials was 18.8% (30/160), showing that this was unlikely to be the case. It appears more plausible that the homogeneity in odds ratios indicates that the relative benefit of NSAIDs is actually similar in low- and high-risk cases. This suggests that the efficacy of prophylactic NSAIDs applies to both low-risk and high-risk patients with a similar RR reduction, although the largest absolute benefit may be seen in high-risk cases.

An additional difference between the trials is that Murray et al and Khoshbaten et al administered prophylactic diclofenac whereas the other two studies used indomethacin. This difference is unlikely to be clinically relevant because NSAIDs have been shown to be equivalent to one another in many different inflammatory states, despite varying pharmacokinetics and cyclooxygenase selectivity. Moreover, interruption of the inflammatory cascade in acute pancreatitis appears to be a class effect, evidenced by diclofenac and indomethacin sharing similar phospholipase A2 inhibitory properties in vitro.14 Therefore, although these differences represent heterogeneities in study design, they did not result in clinically significant differences in outcomes.

In summary, the results of this meta-analysis support the use of prophylactic diclofenac or indomethacin in the prevention of PEP. Widespread prophylactic administration of these agents may result in substantial reduction in the incidence of PEP, translating into major economic benefit. Many other medications have been studied in the prophylaxis of PEP. Despite initial positive studies, subsequent investigation has not proven efficacy. This has led to a general scepticism regarding prophylactic medications for PEP. Given this climate of scepticism, further prospective multicentre studies are needed to confirm the results of this meta-analysis. In addition, further studies are necessary to refine our estimate of the substantial protection from severe pancreatitis conferred by NSAIDs and to establish whether these medications act synergistically with other prophylactic interventions, such as pancreatic stenting.