Types of studies

We will include all RCTs that compared at least two regimens of the same or different PPI in patients with acute bleeding from peptic ulcer, provided that they meet all the following criteria:

1. concomitant therapy was applied equally to both intervention arms;

2. acute bleeding from peptic ulcer was diagnosed endoscopically;

3. for RCTs that included patients with other causes of upper GI bleeding, the data for patients with peptic ulcer bleeding must be accessible and presented separately;

4. at least one of the following outcomes was reported: mortality, rebleeding, surgical intervention, repeat EHT, length of hospital stay, transfusion requirements or adverse outcomes (defined as total number of patients with any of rebleeding, surgery, endoscopic treatment, and/or death). 

We will include both blinded and unblinded RCTs.

Types of participants

Patients with acute upper gastrointestinal bleeding with an endoscopically‐confirmed diagnosis of bleeding peptic ulcer. The participants can be patients admitted to hospital for the bleeding episode or in‐patients who developed acute bleeding from a peptic ulcer while hospitalized for other reasons.

Types of interventions

The treatment group should have received a PPI (either alone or in combination with other treatment) and the control group a different dose of PPI (alone or in combination with the same other treatment administered to the PPI group). We will, therefore, only include studies in which treatment groups were treated similarly, apart from the dose of PPIs being compared.

Method of delivery of PPI and control treatment will include both i.v. and oral.

We will categorize the studies based on the cumulative dose of PPI received within the first 72 hours of treatment as follows:

• "low‐dose" studies: 120 mg / 72 hours or less;

• "medium‐dose" studies: 121 mg to 599 mg / 72 hours;

• "high‐dose" studies: 600 mg or higher / 72 hours.

Our main analysis will be the comparison of "high‐dose" studies versus studies that had used other dose of PPI ("low‐dose" or "medium‐dose").

We will perform a secondary analysis which will compare "low dose" studies versus studies that had used other doses of PPI ("medium‐dose" or "high‐dose").

We are aware that wherever the cut‐off points are set, they will be arbitrary. Ideally we should be able to evaluate the efficacy of PPIs using PPI dose as a continuous variable, rather than comparing ranges of doses among them. This has led us to choose the cumulative dose of PPI received within the first 72 hours of treatment. It will allow us to assess the potentially modifying effect of the dose of PPI as a continuous variable by meta‐regression. However, our meta‐regression analysis of PPI doses in placebo‐ or H₂RA‐controlled trials can provide only indirect evidence of a dose‐response relationship since there is a paucity of head‐to‐head dose comparisons. We will still have to categorize PPI doses so as to enable meta‐analysis of head‐to‐head comparisons of different doses. We have selected the first 72 hours as the time period for calculating the cumulative dose of PPI because this period is the most critical as it includes the majority of the re‐bleeding episodes (Gralnek 2008). We decided to include two cut‐off points rather than one, so as to reduce the risk of rendering non‐evaluable the RCTs that had compared doses within the same category (i.e. high versus high; or low versus low). The high cut‐off point of 600 mg or higher / 72 hours was chosen so as to isolate the widely used "high‐dose" of 8 mg/hour i.v. infusion following an i.v. bolus of 40 to 80 mg. The low cut‐off point was set at 120 mg/ 72 hours so as to isolate the lowest PPI dose that a clinician is expected to use, i.e. 20 mg twice daily (twice the dose of omeprazole that is prescribed for other indications).

The beneficial effects of PPIs in patients with acute peptic ulcer bleeding are considered to be a class effect. Meta‐regression and subgroup analyses in a previous Cochrane review of RCTs that had compared PPIs with H₂RAs or placebo were consistent with a class effect of PPIs (Leontiadis 2006). Although standard doses of different PPIs have statistically significant differences in the degree of acid inhibition (Calvet 2005), the differences are smaller and inconsistent when the PPIs are compared on a milligram basis, especially when clinical efficacy is assessed (Klok 2003; Yacyshyn 2002). For practical reasons we have considered any potential differences in clinical efficacy between different PPIs to be negligible compared to the wide range of different doses of PPIs (200% to 500%) that will be grouped together in each dose category.

Types of outcome measures

Electronic searches

We will identify trials by searching the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE and EMBASE. There will be no language restrictions. A search strategy will be constructed by using a combination of subject headings and text words relating to the use of PPIs for the treatment of bleeding peptic ulcers. The standard Cochrane search strategy filter for identifying RCTs will be applied to all searches.

We will search registers of controlled trials (both active and archived registers by metaRegister of Controlled Trials: www.controlled‐trials.com/mrct/) for ongoing or unpublished studies.

Searching other resources

We will evaluate reference lists from trials and review articles retrieved by electronic searching to identify further relevant trials.

We will also search abstracts from the conference proceedings of: American Digestive Disease Week, United European Gastroenterology Week, American College of Gastroenterology, Canadian Digestive Diseases Week, World Congress of Gastroenterology and British Society of Gastroenterology.

We will contact experts in the field and pharmaceutical companies which market PPIs in the US or Europe for any additional published or unpublished data.

Selection of studies

Two reviewers (J Martin and JC Claro) will independently check titles and abstracts of the references identified from the search and remove obviously irrelevant reports. They will obtain the full text of all potentially relevant studies and assess them for fulfillment of the pre‐defined inclusion criteria. A third reviewer (GI Leontiadis) will act as arbiter in case of disagreement. If duplicate publications of the same patient groups are identified, we will include only the most recent version. If the published information is not sufficient to clarify study eligibility, we will contact the authors for further information.

Data extraction and management

Two reviewers (L Letelier and I Neumann) will independently extract data in pre‐designed electronic data extraction and validity assessment forms. A third reviewer (CW Howden) will act as an arbiter in case of disagreement.

The following data will be extracted where possible:

• method of randomization;

• criteria for patient inclusion and exclusion;

• details of intervention including dose, route of administration, duration;

• details of any co‐interventions including initial EHT;

• details of post‐intervention treatment;

• patient characteristics including mean age, age range, sex, ethnicity;

• number of patients assigned to each treatment group;

• number of patients with co‐morbidity per treatment group;

• number of patients with duodenal and gastric ulcers per treatment group;

• number of patients with each class of stigmata of recent hemorrhage (spurting, oozing, non‐bleeding visible vessel, and adherent clot) at index endoscopy per treatment group;

• number of patients per treatment group with the following outcomes: mortality, surgical interventions, further EHT, blood transfusions; length of hospital stay, adverse outcomes (defined as total number of patients with any of rebleeding, surgery, endoscopic treatment, and/or death);

• timing of outcome assessment;

• definition of rebleeding;

• indications for repeat endoscopy, further EHT, surgery, blood transfusions, discharge;

• causes of death per treatment group;

• outcomes reported by class of stigmata of hemorrhage at initial endoscopy;

• number of patients with adverse reactions per treatment group and method used to seek for adverse reactions;

• baseline comparability of treatment groups;

• blinding of outcome assessor, patients and carers;

• dropouts, with reasons.

Assessment of risk of bias in included studies

We will assess risk of bias in included studies using the "Risk of bias table" which is the tool recommended by The Cochrane Collaboration (Higgins 2008). For each study we will provide a description and a judgement for each one of the following domains: sequence generation, allocation sequence concealment, blinding, incomplete outcome data, selective outcome reporting and other potential sources of bias. The "Risk of bias table" will be prepared by two reviewers independently, with a third reviewer acting as arbiter. We will contact the original authors for further clarification wherever necessary.

Measures of treatment effect

We will report pooled outcomes as risk ratio (RR) with 95% CI for dichotomous outcomes, and as mean difference (MD) with 95% CI for continuous outcomes.

Unit of analysis issues

Since only standard‐design RCTs will be included, each patient will have received only one intervention. In studies where more than two interventions have been compared (e.g. PPI regimen 1 versus PPI regimen 2 versus placebo), extra care will be taken so as avoid including the same patients in the meta‐analysis twice.

Dealing with missing data

We will contact the original investigators to request missing data. We will analyze only the available data and address the potential impact of missing data on the findings of the review in the Discussion section.

Assessment of heterogeneity

We will assess heterogeneity statistically with the Mantel‐Haenszel Chi² test and the I² statistic. Statistically significant heterogeneity will be defined as at least one positive test (either P < 0.10 using the Mantel‐Haenszel Chi² test, or > 50% using the I² statistic).

Assessment of reporting biases

We will investigate the presence of publication bias and other reporting biases visually with the use of funnel plots. In the analyses for dichotomous outcomes (mortality, rebleeding and surgical intervention rates) we will also assess publication bias statistically with the use of Egger's test (Egger 1997) performed with Stata software (Stata 2003). We will base evidence of asymmetry on P < 0.10, and present intercepts with 90% CIs.

Data synthesis

Data synthesis will be performed with the Mantel‐Haenszel random‐effects method. In case of statistically significant heterogeneity (P < 0.10 for the Chi² test, or > 50% for the I² statistic), a meta‐analysis will still be performed, although the readers will be strongly advised to interpret the results with caution. Furthermore, heterogeneity will be thoroughly investigated with predetermined subgroup analyses and meta‐regression (as explained in the next section).

Subgroup analysis and investigation of heterogeneity

The influence of the following factors on the outcomes will be assessed by subgroup analyses:

• risk of bias of primary study ("high" versus "low" or "unclear"; "low" defined as no bias for all key domains, "unclear" defined as unclear risk of bias for one or more key domains, "high" defined as presence of bias for one or more key domains; key domains: adequate sequence generation, allocation concealment, blinding, incomplete outcome data addressed, free of selective reporting, free of other bias);

• site of ulcer (gastric versus duodenal);

• initial endoscopic treatment (yes versus none);

• route of administration of PPI (oral versus intravenous);

• mode of intravenous administration (continuous infusions perfusion versus bolus administration);

• specific PPI used (omeprazole versus all others; pantoprazole versus all others);

• Indo‐Asian population versus non‐Indo‐Asian population.

Where sufficient data are available, we will assess the influence of the above factors on the size of treatment effect (log odds ratio for mortality, rebleeding and surgical intervention) and on the heterogeneity of the analyses using meta‐regression analysis (random‐effects model, within‐study variance estimated with the restricted maximum‐likelihood method). Furthermore, the cumulative dose of PPI received within the first 72 hours of treatment will also be assessed by meta‐regression as a potential effect modifier. Meta‐regression analysis will be performed with Stata software (Stata 2003) or Comprehensive Meta‐Analysis software (CMA).

Sensitivity analysis

We plan to undertake the following sensitivity analyses:

1. the analysis of "high‐dose" studies versus other dose studies, the analysis of "low‐dose" studies versus other dose studies, and the subgroup analyses according to geographical location will be restricted to patients with active bleeding or a non‐bleeding visible vessel at index endoscopy who received appropriate EHT at index endoscopy;

2. "medium‐dose" studies will be excluded from the analysis of "high‐dose" studies versus other dose studies (resulting in a comparison of "high‐dose" versus "low‐dose" studies);

3. "low‐dose" studies will be excluded from the analysis of "high‐dose" studies versus other dose studies (resulting in a comparison of "high‐dose" versus "medium‐dose" studies);

4. "high‐dose" studies will be excluded from the analysis of "low‐dose" versus other dose studies (resulting in a comparison of "low‐dose" versus "medium‐dose" studies);

5. "low‐dose" oral studies or "low‐dose" i.v. studies will be excluded from the analysis of "low‐dose" versus other dose studies.