Treatment of acute pancreatitis with protease inhibitors administered through intravenous infusion: an updated systematic review and meta-analysis

First, we systematically searched PubMed, EMBASE, the Cochrane Library, and Japana Centra Revuo Medicina (the largest database of Japanese articles) for articles of RCTs published from January 1965 to March 2013 (search date: April, 2013) on the effectiveness of protease inhibitors used to treat acute pancreatitis. The electronic database search was conducted using a combination of Medical Subject Heading terms and text words “protease inhibitors” and “acute pancreatitis”. Next, we searched Internet-based clinical trial registries, ClinicalTrials.gov [12] and controlled-trials.com [13], as well as all the three trial registries available in Japan, UMIN [14], JMACCT [15], and JAPIC [16] (search date: April, 2013) for on-going RCTs using the same criteria. References of review articles and previously published meta-analyses were handsearched.

Given that the purpose of the present study was to update the body of evidence regarding the effectiveness of intravenous use of protease inhibitors for acute pancreatitis, the following inclusion criteria were set: 1) randomized placebo-controlled trials of protease inhibitors administered through intravenous infusion; and 2) written in all languages. No restrictions were placed on severity of pancreatitis or type of protease inhibitors. Furthermore, we excluded trials in which 1) both intervention and control groups were administered protease inhibitors; 2) protease inhibitors were administered by intra-arterial or intra-abdominal infusion; 3) subjects included patients with chronic pancreatitis; 4) pancreatitis was noted after endoscopic retrograde cholangiopancreatography (ERCP); 5) patients were administered frozen plasma; 6) the goal was to investigate basic medical science (e.g., pancreatic enzyme research); 7) patients had human immunodeficiency virus; 8) efficacy of antibiotics was being evaluated; 9) patients included post-operative cases; and 10) oral administration of protease inhibitors.

No restrictions were placed on patient age, sex, or cause of acute pancreatitis. Articles in the form of a conference proceeding or full paper were also included. One author (TS) selected articles to be included for analysis, and the other authors verified the process.

The effectiveness of protease inhibitors for acute pancreatitis was evaluated based on the primary outcome of death due to acute pancreatitis, and secondary outcomes including relief from pain, pseudocyst formation, formation of intra-abdominal abscess, surgical intervention, paralytic small bowel obstruction, and other major complications including multiple organ failure.

Study design, participants, mode of intervention, and definition of outcomes were faithfully extracted from the articles included in the final analysis. We also confirmed the status of industrial support for each trial.

We calculated CMR (defined as the number of deaths in the control group divided by the number of patients in the control group) for each RCT.

The quality of each trial retrieved was assessed by the Jadad method [17] on the basis of whether the trial was randomized, the appropriateness of the randomization process (if applicable), whether the trial was double-blinded, the appropriateness of the double-blinding process (if applicable), and withdrawals/dropouts. Each item was assigned a score of 0 or 1, and the total score ranged from 0 to 5. In the present meta-analysis, trials with a Jadad score greater than 3 were defined as high-quality trials. Furthermore, we evaluated risk of bias for each trial and assessed the quality of the body of evidence using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system for grading evidence [18].

One author (TS) conducted the quality assessment and extraction of analyzable data, which were verified by the other authors. Disagreement or uncertainty was resolved among all the authors.

We calculated the risk difference (RD), i.e., risk in the intervention group minus risk in the control group, for the primary outcome of the trials. A negative RD indicated risk reduction due to intervention, and a positive RD, risk increase due to intervention (range, -1 to 1). Whether the treatment or control was favored was denoted by the signs “+” and “-”, respectively. Then, the weighted pooled estimates were calculated for binary data. A fixed-effect model weighted by the Mantel-Haenszel (M-H) method was used for pooling RD [19], followed by a test of homogeneity. Homogeneity among trials was assessed using the I² test [20]. We defined I² value <25% as low, 25 to 50% as moderate, and >50% as high heterogeneity.

If the hypothesis of homogeneity was rejected, a random-effect model using the DerSimonian-Laird (D-L) method was employed [21]. The potential for publication bias was examined by the funnel plot method [22] using the Begg’s [23] or Egger’s test [24]. The number needed to treat (NNT, 1/RD) to prevent one adverse event was also used as a measure of treatment effect. We used the “number needed to treat benefit (NNTB; the number of patients needed to be treated for one additional patient to benefit)” for a positive NNT, and the “number needed to treat harm (NNTH; the number of patients needed to be treated for one additional patient to be harmed)” for a negative NNT. When the upper or lower limit of the 95% confidence interval (CI) was infinity, the NNT scale including infinity was used [25]. All statistical analyses were performed with Stata statistical software [26]. Results were expressed as means and 95% CIs, unless otherwise indicated. P < 0.05 was considered statistically significant.

Our database search yielded 96 articles, and handsearching of bibliographies of retrieved meta-analyses and clinical guidelines yielded additional three and two articles, respectively. There were no on-going trials in the registries. Of the 101 articles, 24 met the inclusion criteria [27‐50]; no multiple publications were found. Reviewers’ selection of relevant articles was completely the same, and there was no unsuitable study for inclusion by authors’ consensus. The 77 excluded articles described ERCP studies (n = 36), studies in which a protease inhibitor was administered to both intervention and control groups (n = 8), arterial infusion studies (n = 6), and other (n = 27) (Figure 1). Next, all authors read the selected 24 articles, reaching a consensus to exclude seven more articles [27, 38‐40, 45, 48, 50] including two in which a protease inhibitor was administered to both intervention and control groups [30, 45], two that were published as comments [38, 50], one in which glucagon was given to the control group [39], one published as an editorial (n = 1) [40], and one reporting an ERCP study [48]. In the end, a total of 17 articles [27‐29, 31‐37, 41‐44, 46, 47, 49] were selected for analysis.

The present meta-analysis of the retrieved competent studies included 15 RCTs from the handsearch [27, 28, 31‐37, 41‐43, 46, 47, 49], one [29] from a previous meta-analysis [51], and one [44] from guidelines [52], with the total sample size of 1,697 patients. Of the 15 articles manually searched, 10 [33, 34, 37, 41‐44, 46, 47, 49] were used in our previous meta-analysis [10].

All articles evaluated death due to acute pancreatitis, in addition to other outcomes such as pain relief (n = 2) [31, 49], pseudocyst formation (n = 5) [29, 37, 41, 43, 46], intra-abdominal abscess formation (n = 4) [37, 41, 43, 46], surgical intervention (n = 3) [44, 47, 49], paralytic small bowel obstruction (n = 3) [41, 47, 49], and other complications (n = 5) [36, 37, 43, 46, 47]. Of the 17 articles, 11 [29‐31, 33‐36, 41, 42, 44, 46, 47] were conducted in a multicenter setting, and six [27, 28, 32, 37, 42, 49] in a single-center setting. Sixteen articles were published as a full paper [27‐29, 31‐37, 41‐43, 46, 47], and one as an abstract only [44]. Fourteen articles were written in English [27‐29, 31, 33‐35, 37, 41, 43, 44, 46, 47, 49], two in German [32, 42], and one in French [36]. Eleven trials used aprotinin [27‐29, 31‐37, 41], and six trials used gabexate mesilate [42‐44, 46, 47, 49], all against the placebo control. Four articles [36, 41, 47, 49] described tangible outcomes in the method section, whereas 13 provided no description [27‐29, 31‐35, 37, 42‐44, 46]. The sample size was pre-calculated only in one article [47]. Acute pancreatitis was defined as clinical manifestation with elevated serum pancreatic enzyme levels in 11 articles [27‐29, 34, 35, 37, 41, 43, 46, 47, 49], or as clinical manifestation with elevated urine pancreatic enzyme levels in two [32, 33]; no description regarding the definition was found in the remaining four articles [31, 36, 42, 44]. Nine articles described sources of funding or medicine support from industries [27, 29, 31, 34, 35, 37, 41‐44]. The severity of acute pancreatitis was reported using the Ranson’s score in two articles [47, 49], or Acute Pathophysiology and Chronic Health Evaluation (APACHE-II) score in one article [49]. In addition to these, we evaluated the severity of acute pancreatitis according to CMR in the present analysis, and found that 10 articles had CMR >0.10 [27, 28, 34‐36, 41, 43, 44, 47, 49]. No articles mentioned the presence or absence of adverse events. Seven articles observed death due to acute pancreatitis as the primary outcome [27, 28, 32‐35, 47]. Six articles reported no secondary outcome [27, 28, 32‐35], while one article observed surgical intervention and bowel obstruction as the secondary outcomes [47]. Ten articles reported undistinguishable outcomes [29, 31, 36, 37, 41‐45, 49]. We performed quality of evidence assessment and found death, pseudocyst, intra-abdominal abscess, and any major complications to be low quality, and abdominal pain, surgical intervention, and bowel obstruction to be very low quality. Table 1 shows heterogeneity of each outcome with I² values.

The overall mean Jadad score was 2.1 (range 0–5). The mean Jadad score increased to 2.6 if three RCTs with a 0 score [28, 32, 43] were excluded. Six trials [31, 34, 35, 37, 46, 47] were considered high quality (Jadad score ≥3), with the mean Jadad score of 3.7. With respect to risk of bias of each trial, two described random allocation [46, 47], 13 described allocation concealment [27, 29, 31, 33‐37, 41, 42, 44, 46, 47], 13 described blinding [27, 29, 31, 33‐37, 41, 42, 44, 46, 47], three described outcome data addressed [44, 46, 47], and four described selective outcome reporting [36, 41, 47, 49]; no other bias was described. One RCT described intention to treat (ITT) analysis [47]. The quality of each outcome in terms of evidence was graded low to very low.

All trials reported death due to acute pancreatitis (Table 1 and Figures 2 and 3), 11 with aprotinin [27‐29, 31‐37, 41], and six with gabexate mesilate [42‐44, 46, 47, 49]. Overall results of the 17 trials showed no significant risk reduction (low heterogeneity) in mortality with the use of protease inhibitors. Furthermore, subgroup analyses revealed no significant results (Table 3).

The funnel plot was not symmetric; however, neither one of the statistical tests revealed significant publication bias (p = 0.653 and p = 0.736, respectively).

Using the main outcome, i.e., overall mortality from acute pancreatitis, we performed a sensitivity analysis according to the Jadad score. The effect size and level of statistical significance did not decrease with increasing trial quality. We then performed another sensitivity analysis according to CMR. In all trials and trials with CMR <0.20, no significant effectiveness of protease inhibitors was shown. However, a significant effectiveness was found when trials were limited to those with CMR >0.20.