Meta-analysis: rectal indomethacin for the prevention of post-ERCP pancreatitis

Authors


  • As part of AP&T's peer-review process, a technical check of this meta-analysis was performed by Mr M. Siddiqui.

Summary

Background

Despite initial evidence in the literature, nonsteroidal anti-inflammatory drugs (NSAIDs) have not been widely used to prevent post-endoscopic retrograde cholangiopancreatography (ERCP) pancreatitis (PEP).

Aim

To complete a meta-analysis of high-quality RCTs that included the latest available literature published after past meta-analytical efforts

Methods

A comprehensive electronic literature search was carried out for RCTs comparing peri-procedural rectal indomethacin and placebo in preventing PEP. Methodological quality was assessed by the Cochrane risk of bias tool. Fixed model Mantel–Haenszel meta-analysis, Q test and I2 index were used. Several subgroup and sensitivity analyses were planned.

Results

A total of four of 61 retrieved trials between 2007 and 2012 (n = 1470) were included. No significant publication bias existed. All studies used similar criteria to detect pancreatitis. The pooled proportion estimate of the rate of pancreatitis was 5.1% with indomethacin and 10.3% with placebo. After excluding the high-risk patients, the rates were 3.9% and 7.9% respectively. Fixed model meta-analysis showed that the rate of pancreatitis was significantly lower using indomethacin as compared with placebo [OR = 0.49(0.34–0.71); = 0.0002]. Number needed to treat was 20. There was no significant statistical or clinical heterogeneity. In subgroup analysis, the difference remained unchanged for average-risk population [OR = 0.49(0.28–0.85); = 0.01] or in preventing severe PEP [OR = 0.41(0.21–0.78); = 0.007]. The result of the main outcome remained robust in multiple sensitivity analyses.

Conclusions

Rectal indomethacin used immediately before or after ERCP significantly reduces the risk of PEP to half in both low- and high-risk patients, and with both statistically and clinically significant conclusions. These results suggest that a possible change in routine practice for patients at both low and high risk of developing PEP should be advocated.

Introduction

Acute pancreatitis is the most common major post-endoscopic retrograde cholangiopancreatography (ERCP) complication. Complications range from 1–10% to as high as 25–30% in high-risk patients, including those with patient-related factors such as young age, female gender, pancreas divisum, sphincter of Oddi dysfunction (SOD), prior post-ERCP pancreatitis (PEP), or with procedure-related factors such as difficult cannulation and pancreatic duct injection.[1-3]

Multiple studies have evaluated several pharmacological agents administered prior to, or immediately after, the procedure to prevent PEP. Intravenous gabexate and somatostatin have been investigated with some promise, but have been limited in use because they require continuous infusion and are not commercially available worldwide.[4-6] Nonsteroidal anti-inflammatory drugs (NSAIDs), on the other hand, are inexpensive, readily available and have been proven to be relatively safe for use. NSAIDs are potent inhibitors of phospholipase A2 activity, which is believed to play a critical role in the inflammatory cascade of acute pancreatitis by regulating pro-inflammatory mediators, including arachidonic acid products and platelet-activating factors, making them an attractive potential therapeutic option.[6, 7]

Several NSAIDs have been investigated in this context. Both rectal and oral NSAIDs have shown promising results in retrospective, open-label and randomised clinical trials (RCTs). Two previous meta-analyses have also shown some efficacy attributable to using peri-procedural oral or rectal NSAIDs in preventing PEP. Dai et al. performed a meta-analysis of studies on the use of any NSAIDs in preventing PEP, published up to 2007.[8] The authors suggested that NSAIDs decreased the risk of PEP by half, in the absence of any reported major side effects. However, the data were not sufficient to allow for a stratified analysis by NSAID type or strong enough to end in a change in the standard of care. Moreover, two important RCTs on the use of rectal indomethacin in preventing PEP were published after this meta-analysis was performed.[9, 10] Elmunzer et al. also performed a meta-analysis in 2008, which exhibited similar limitations.[11] A very recently published meta-analysis on NSAID use in preventing PEP did not include a Hungarian RCT by Dobronte et al. and did not provide subgroup analysis, including specifically looking at the use of rectal indomethacin.[12]

As studies on rectal indomethacin have been inconsistent in showing statistically significant results, this practice has not yet been incorporated into contemporary standard of care in major guidelines, presumably due to methodological criticisms, the small sample sizes of published studies and a lack of robust meta-analyses specifically studying rectal indomethacin in PEP prevention. We therefore aimed to complete a meta-analysis of high-quality RCTs that included the latest available literature, including the two large recent RCTs discussed, published after past meta-analytical efforts.

Materials and methods

Trial selection criteria

Only RCTs were included. For inclusion, we required that patients in the trial were randomised to rectal indomethacin or placebo prior to, or immediately after, ERCP. No studies were excluded based on the language of publication, quality of study, duration of follow-up or country of origin.

Search strategy

Comprehensive computerised medical literature searches were conducted using OVID MEDLINE (1946 to January 2013), EMBASE (1980 to January 2013), Cochrane library, and ISI Web of knowledge from 1980 to January 2013. Abstracts from major gastroenterology conferences in the past 5 years (including Digestive Disease Week, Canadian Digestive Disease Week, United European Gastroenterology Week, American College of Gastroenterology and the Asia-Pacific Digestive Week) were also searched, as were clinical trial databases (www.clinicaltrials.gov). Other available sources of unpublished data (grey literature) were also searched, where available through cross-referencing. Articles were selected using a highly sensitive search strategy to identify reports of RCTs, with a combination of MeSH headings and text words that included (i) Pancreatitis, (ii) Indomethacin and (iii) ERCP. Recursive searches and cross-referencing were carried out using a ‘similar articles’ function; bibliography of the articles identified after an initial search were also manually reviewed. Nonrandomised trials, studies with insufficient data on clinical response, any study on rescue therapy, paediatric studies and duplicate publications were excluded. Data extraction and quality control were carried out independently by two reviewers (MY and SR). A third reviewer (JMB, RB or ANB) was involved if conflict occurred. Kappa scores were measured to assess the agreement between the two initial reviewers in each step and interpreted as described elsewhere.[13] Methodological quality of the included studies was evaluated by using the Cochrane Collaboration tool for assessing the risk of bias.[14]

Outcome measures

The main analysis addressed the outcome of the rate of post-ERCP pancreatitis in each arm. Subgroup analyses were planned to be carried out on the high-risk vs. low-risk patients and varying doses of rectal indomethacin, if possible.

Statistical analysis

Summary outcomes are described as proportions and 95% CI for categorical and weighted mean difference ± s.d. for continuous data. Cumulative response rates were calculated separately for rectal indomethacin and placebo using the sum of the responders to total numbers of included patients, and were reported as proportions and confidence intervals for each individual comparison. A meta-analysis of intention-to-treat data was performed using the fixed model Mantel–Haenszel method. P-values <0.05 were considered significant. The significance and extent of statistical heterogeneity were calculated using the Q test and I2 index respectively. Random effect modelling was applied if the P-value for the test of heterogeneity was less than 0.10 using the DerSimonian and Laird method.[15] This was chosen rather than the conventional cut-point of P = 0.05, as the test does not have enough power to detect heterogeneity when there are few studies. If not, fixed effect models were constructed. Odds ratios (OR) were calculated for each analysis with the corresponding 95% confidence intervals.

Funnel plots, Begg adjusted rank correlation test and the Egger regression asymmetry test were used to detect the possibility of publication bias.[16, 17] We also planned to perform meta-regression and sensitivity analyses based on the quality and weight of the trials, and by excluding each individual trial in turn as recommended by Cochrane Collaboration open learning material for reviewers.[18] Subgroup analyses were planned for different doses of indomethacin, patients who were at high risk for PEP, those with sphincter of Oddi dysfunction (SOD) or biliary stent insertion and the effect on moderate-to-severe PEP.

All statistical analyses were performed using RevMan. Version 5.0.25, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008 and R version 2.13.0 (R Foundation for Statistical Computing, Vienna, Austria, 2008). The PRISMA statement outline for reporting systematic reviews and meta-analyses was used to report on this work.[19]

Results

Characteristics of included trials

A total of four of 61 identified studies (n = 1470 patients, including 745 in treatment group and 725 in placebo group) were included; all were performed between 2007 and 2012. Three studies were published in English and one in Hungarian. Figure 1 depicts the PRISMA flow diagram. There was excellent inter-reviewer agreement [Kappa = 0.84 (95% CI: 0.56–1.0]. Table 1 shows the methodological and clinical details for each trial. All trials used relatively similar criteria to define post-ERCP pancreatitis, as depicted in Table 1. Figure 2 presents the consensus risk of bias assessments of the included studies.

Table 1. Characteristics of included studies
StudyYearCountryType n Number of PEPsInterventionDefinition of PEPQuality scale
Dobronte et al.2012Hungary-Single centreRCT22822Indo 100 mg PR 10 min qpClinical/Amylase 24 h2
Elmunzer et al.2012US-MulticentreRCT-High-risk pts60279Indo 50 mg PR right after ERCPPain, Amylase> ×3, admission >1 night5
Sotoudehmanesh et al.2007IranRCT49022Indo 100 mg PR right before ERCPPain, Amylase> ×3, admission5
Montaño Loza et al.2007MexicoRCT15016Indo 100 mg PR right before ERCPClinical, amylase level1
Figure 1.

PRISMA flow diagram of included and excluded trials. RCT, randomised controlled trials.

Figure 2.

Consensus risk of bias assessments of the included studies. Green: Low risk, Yellow: Unclear, Red: High risk.

Publication bias

A funnel plot showed that the studies were reasonably well scattered and did not suggest any publication bias; furthermore, the Begg adjusted rank correlation (= 0.45) and Egger regression asymmetry tests (= 0.31) did not demonstrate significant publication bias.

Analysis of main outcome

Rectal indomethacin consistently provided a lower rate of post-ERCP pancreatitis compared with placebo in all included studies, although the difference was not statistically significant in all trials. The pooled proportion estimates of PEP were 5.1% (95% CI: 3.5–16.2) with rectal indomethacin, and 10.3% (95% CI: 8.2–22.9) with placebo. This rate was significantly lower with rectal indomethacin as compared with placebo [OR=0.49 (0.34–0.71); = 0.0002] (Figure 3). The number needed to treat was 20.

Figure 3.

Forest plot of Mantel–Haenszel meta-analysis of the overall rate of post-ERCP pancreatitis with rectal indomethacin. CI, confidence interval; M-H, Mantel–Haenszel.

Statistical and clinical heterogeneity

Statistical heterogeneity was low and not statistically significant in the analyses of the main outcome (χ2 = 1.53, = 0.68, I2 = 0%), subgroup analysis (χ2 = 1.53, = 0.47, I2 = 0%) or any of the sensitivity analyses (= nonsignificant, I2 = 0%). We thus only used fixed effect models.

With regard to clinical heterogeneity, the trial by Elmunzer et al.[10] only included high-risk patients and was the sole trial administering a 50 mg dose of rectal indomethacin, while the other three trials had relatively similar inclusion criteria, all using a rectal dose of 100 mg (Table 1).[9, 20, 21]

Subgroup analyses

We were unable to perform a separate meta-regression analysis, meta-analysis of outcome in patients with SOD or those who underwent prophylactic pancreatic stent placement due to insufficient data.

Low-risk vs. high-risk patients

After excluding the only study on high-risk populations,[10] the rate of PEP remained unchanged and still statistically significant [OR: 0.49 (0.28–0.85); = 0.01], favouring rectal indomethacin. The rate of PEP was 0.49 (0.30–0.81) for high-risk patients in the above-mentioned study.

Rectal indomethacin for preventing moderate-to-severe PEP

All included studies reported the rate of moderate-to-severe pancreatitis. Rectal indomethacin significantly decreased the rate of moderate-to-severe PEP [OR: 0.45 (0.24–0.83); = 0.01].

Different Dosage of rectal indomethacin (100 mg vs. 50 mg):

Elmunzer et al. was the only study using a dose of 50 mg and therefore a meta-analysis analysing this dosage was not possible.[10] A subgroup meta-analysis of three studies using rectal indomethacin 100 mg showed that this dose was significantly superior to placebo in preventing PEP [OR: 0.49 (0.28–0.85); p=0.01].

Sensitivity analyses

Excluding each trial in turn and the effect of the size of trial

The effect of therapeutic procedure on symptom resolution remained statistically significant after removing each included trial (OR: 0.45–0.52, < 0.01).

Effect of the methodological quality of studies

The comparison remained unchanged after excluding two studies with higher risk of bias [OR: 0.48 (0.31–0.75); = 0.001].[9, 21]

Effect of the largest included trial

After excluding the largest included trial,[10] the rate of PEP remained unchanged and still statistically significant [OR: 0.49 (0.28–0.85); = 0.01].

Discussions

In this meta-analysis, we showed that rectal indomethacin is superior to placebo in preventing PEP in both average- and high-risk patients undergoing ERCP. It decreases the rate of PEP by half, with a number needed to treat of around 20. Sensitivity analyses showed that the results remained robust and unchanged when excluding any of the included studies in turn, or when stratifying according to PEP risk, or the dose of indomethacin, thus confirming the robustness of the findings.

To our knowledge, this study is thus far the most contemporary meta-analysis of available randomised controlled trials on administering rectal indomethacin for the prevention of PEP. In half of our included trials, rectal indomethacin was not shown to be statistically superior to placebo in preventing PEP, although all studies demonstrated an arithmetical trend favouring the former medication. This is consistent with the absence of statistically significant heterogeneity shown in the analysis. Moreover, the designs of the included RCTs were very similar thereby resulting in the observed low clinical heterogeneity, not measurable by statistical means, which might further enhance the validity of the findings.

Thus far, stent placement in the pancreatic duct has been the only proven effective intervention for PEP prophylaxis with an OR of 0.44 (0.24–0.81);[22-24] however, more than 20% of endoscopists do not perform prophylactic pancreatic stenting in any situations.[23] In fact, pancreatic stent placement has been shown not to be cost-effective in patients at average risk for PEP and requires endoscopic expertise for insertion.[25] On the other hand, a recent post hoc analysis by Elmunzer et al. showed that the rate of PEP following combined rectal indomethacin and pancreatic stenting was greater than for rectal indomethacin alone (9.7% vs. 6.3% respectively).[26] Although one might argue the possibility of a selection bias by using prophylactic pancreatic stent placement in patients at higher risk for PEP, caution must be exercised and controlled trials are needed to compare both or combination of prophylactic approaches. Moreover, their cost-effectiveness analysis demonstrated that rectal indomethacin alone is more cost-effective than placebo, prophylactic pancreatic duct stent placement or a combination of rectal indomethacin and stent placement.[26] Furthermore, a very recent network meta-analysis, based on both direct and indirect evidence, showed that rectal NSAIDs were superior to prophylactic pancreatic stent placement [OR: 0.48 (0.26–0.87)] and a combination of rectal NSAIDs and prophylactic pancreatic stent placement was not superior to rectal NSAIDs [OR:1.46 (0.79–2.69)].[27] In addition, rectal indomethacin is relatively safe, inexpensive and easy to administer. Rectal administration of indomethacin results in faster absorption than administration by the oral route. The peak plasma concentration of oral NSAIDs is about 2 h vs. only 30 min when administered rectally.[28] Although one could argue that the oral administration of NSAIDs 2 h before ERCP might have the same effect, a head-to-head comparison has not been carried out and rectal administration might be preferred by physicians who would rather order strict fasting 4–6 h before the procedure.

Our results are consistent with those from previous studies. However, previous meta-analyses mainly aimed at evaluating the role of NSAIDs as a class of pharmacological agents and, to our knowledge, none reported individual results for rectal indomethacin. As a result, it would have been difficult to advocate a specific treatment as standard of care based on these antecedent analyses. The most recent published meta-analysis included 10 RCTs comparing any oral or rectal NSAIDs with placebo, showing similar results, however, the authors did not include a large trial (n = 228) by Dobronte et al.[9] /Five included studies used rectal indomethacin or diclofenac, but the meta-analysis did not plan on performing subgroup analysis for any specific medications or route of administration, and therefore the authors only proposed using NSAIDs, in general, to prevent PEP. They obtained an odds ratio of 0.57 (0.38–0.86) for the main analysis, and 0.46 (0.25–0.75) for the subgroup of moderate-to-severe PEP, which is relatively similar to the point estimate in the current effort.[12]

Two other meta-analyses were published in 2008 and 2009. Elmunzer et al. aimed at including all RCTs on rectally administered NSAIDs.[11] They showed a relative risk of 0.36 (0.22–0.60) for preventing PEP. Similar results were reported for moderate-to-severe PEP. However, they could not report any results for subgroup analysis of individual NSAIDs, mainly due to the small number of included trials, as their publication predated three more recent RCTs that were subsequently published. Dai et al. also published a meta-analysis on the subject.[8] Unfortunately, the investigators did not exclude a duplicate publication by Montaño Loza et al., while still exhibiting the aforementioned weaknesses of other former meta-analyses.[21]

In our study, the number needed to treat was 20. One might argue not adopting the intervention on this basis. On the other hand, one could consider adopting such a strategy, given the low cost of the intervention, its negligible side effects, and the fact that one of every two episodes of PEP could be prevented with the intervention could potentially advocate its use. Alternately, high-risk groups could be targeted.

A concern is that the rate of PEP in RCTs included in our meta-analysis was higher than reported in previous observational studies.[27] One explanation might be that we only used ITT data and therefore the withdrawals were counted as failures. The data from observational studies could therefore underestimate the rate of PEP. For example, some patients, in particular those with mild symptoms, might not seek medical attention, while patients included in RCTs are carefully monitored for the development of PEP using both abdominal pain and amylase level. The methodology of randomised controlled trial and using placebo control should adjust for these contributing factors. The weaknesses of our study relate to the methodology of meta-analyses. The possibility of missed trials cannot be completely ruled out, especially when only four studies were identified. We tried to minimise this possibility by including several types of publications, search methods and all languages as well as performing several statistical analyses to rule out publication bias. Our study did not include any patient-level data analysis, and therefore is unable to characterise individual predictors of outcomes, nor were there sufficient trials to perform a meta-regression. If additional trials or patient-level data were available, this would have helped in identifying patients who are likely to benefit from one type of therapy more than the other. The main limitation of the current analysis is the small number of trials that it includes, which made it impossible to perform all planned subgroup analysis. Despite this, the main results of the study are significant as discussed above, and remain robust after many sensitivity analyses across outcomes with both statistically and clinically significant conclusions and the resulting implications as to a possible change in routine practice for both patients at low and high risk of developing PEP.

In conclusion, this meta-analysis suggests that rectal indomethacin is probably effective in preventing half of clinical PEP cases. Although the possibility of publication bias could not be completely ruled out, the results appear significant and should be considered as part of the current evidence in preparing clinical practice guidelines, given the proven cost-effectiveness of rectal indomethacin as compared with other available modalities and the significant clinical burden of disease. We suggest future research to focus on an individual patient meta-analysis when more data become available to better identify the optimal patient population that will benefit most or not from this approach, as well as a randomised controlled trial comparing rectal indomethacin and prophylactic pancreatic stent placement in preventing PEP.

Authorship

Guarantor of the article: Dr Alan Barkun.

Author contributions: MY, SR, KAW, JM-B, MM, RB, PS and ANB: conception and design, or analysis and interpretation of data. MY, SR, KAW, RB and ANB: drafting the article or revising it critically for important intellectual content and final approval of the version to be published.

Acknowledgements

Declaration of personal interests: The authors would like to thank Dr Constantine Soulellis for his helpful comments on the manuscript. ANB is a consultant for Takeda Canada and Olympus Canada, received research funding from Boston Scientific Inc and Cook, and has served as a speaker for AstraZeneca Inc and Takeda Inc.

Declaration of funding interests: None.

Ancillary