Bleeding and pneumonia in intensive care patients given ranitidine and sucralfate for prevention of stress ulcer: meta-analysis of randomised controlled trialsBMJ 2000; 321 doi: http://dx.doi.org/10.1136/bmj.321.7269.1103 (Published 04 November 2000) Cite this as: BMJ 2000;321:1103
- A Messori, coordinatora,
- S Trippoli, coordinatora,
- M Vaiani, research fellowa,
- M Gorini, staff physicianb,
- A Corrado, headb
- a Laboratorio SIFO di Farmacoeconomia, Centro Informazione Farmaci, Servizio Farmaceutico, Azienda Ospedaliera Careggi, 50134 Florence, Italy,
- b Unita' di Terapia Intensiva Respiratoria, Azienda Ospedaliera Careggi
- Correspondence to: A Messori
- Accepted 1 August 2000
Objectives: To determine the effectiveness of ranitidine and sucralfate in the prevention of stress ulcer in critical patients and to assess if these treatments affect the risk of nosocomial pneumonia.
Design: Published studies retrieved through Medline and other databases. Five meta-analyses evaluated effectiveness in terms of bleeding rates (A: ranitidine v placebo; B: sucralfate v placebo) and infectious complications in terms of incidence of nosocomial pneumonia (C: ranitidine v placebo; D: sucralfate v placebo; E: ranitidine v sucralfate). Trial quality was determined with an empirical ad hoc procedure.
Main outcome measures: Rates of clinically important gastrointestinal bleeding and nosocomial pneumonia (compared between the two study arms and expressed with odds ratios specific for individual studies and meta-analytic summary odds ratios).
Results: Meta-analysis A (five studies) comprised 398 patients; meta-analysis C (three studies) comprised 311 patients; meta-analysis D (two studies) comprised 226 patients: and meta-analysis E (eight studies) comprised 1825 patients. Meta-analysis B was not carried out as the literature search selected only one clinical trial. In meta-analysis A ranitidine was found to have the same effectiveness as placebo (odds ratio of bleeding 0.72, 95% confidence interval 0.30 to 1.70, P=0.46). In placebo controlled studies (meta-analyses C and D) ranitidine and sucralfate had no influence on the incidence of nosocomial pneumonia. In comparison with sucralfate, ranitidine significantly increased the incidence of nosocomial pneumonia (meta-analysis E: 1.35, 1.07 to 1.70, P=0.012). The mean quality score in the four analyses (on a 0 to 10 scale) ranged from 5.6 in meta-analysis E to 6.6 in meta-analysis A.
Conclusions: Ranitidine is ineffective in the prevention of gastrointestinal bleeding in patients in intensive care and might increase the risk of pneumonia. Studies on sucralfate do not provide conclusive results. These findings are based on small numbers of patients, and firm conclusions cannot presently be proposed.
Ranitidine and sucralfate are widely used to prevent stress ulcers in patients admitted to intensive care units.1 A meta-analysis published by Cook et al in 1996 showed that H2 receptor antagonists (such as cimetidine and ranitidine together) are more effective than placebo for this clinical indication.2 With regard to sucralfate, this meta-analysis found a small but significant reduction in overt bleeding but no effect on clinically important events. The meta-analysis did not resolve the question of an increased risk of nosocomial pneumonia related to the use of H2 receptor antagonists.
Several arguments emphasise the need for up to date information on this issue. Firstly, ranitidine has become the main H2 receptor antagonist used for prophylaxis for stress ulcers, and cimetidine has generally been abandoned1; secondly, new findings have been published on effectiveness and complications of ranitidine; and, thirdly, a meta-analytic comparison of ranitidine versus placebo has never been carried out, and as the comparison of sucralfate and placebo made by Cook et al gave no proof of the effectiveness of this drug, ranitidine and sucralfate might both be ineffective. Another problem is that the most recent randomised studies on this topic did not include a group with no prophylaxis and compared supposedly active treatments with one another. 3 4
We conducted a literature search to identify randomised trials, and we carried out a meta-analysis to update the results of Cook's study with regard to effectiveness and infectious complications.
Our Medline search covered the period from 1966 to 20 June 2000 and was based on four key words (stress, pneumonia, ranitidine, sucralfate) and on the extraction of studies published in English (see Appendix). Randomised studies were identified by using the key words “randomized controlled trial” or “random” according to a validated literature search.5
This search was supplemented by examining the Iowa-IDIS system (Iowa Drug Information, Iowa University, United States) from 1966 to December 1999 and Drugdex (CD Rom Drugdex, vol 104, Micromedex, Englewood, Colorado, United States).
On the basis of the material produced by our search, we carried out five meta-analyses that evaluated data on effectiveness in terms of rates of bleeding (meta-analysis A: ranitidine v placebo; meta-analysis B: sucralfate v placebo) and data of infectious complications in terms of incidence of nosocomial pneumonia (meta-analysis C: ranitidine v placebo; meta-analysis D: sucralfate v placebo; meta-analysis E: ranitidine v sucralfate). We investigated the issue of effectiveness only in studies with a control group that received no prophylaxis, while the issue of infectious complications was also assessed from comparative trials of ranitidine versus sucralfate. Eligible studies were included in meta-analysis A or B if they met the following criteria: patients were admitted to an intensive care unit or were undergoing mechanical ventilation, or both; randomised design; assessment of gastrointestinal bleeding. We excluded studies in which gastrointestinal bleeding did not meet the definition of “clinically important bleeding” according to Cook et al.2 In meta-analyses C, D, and E the inclusion criterion gastrointestinal bleeding was replaced by the assessment of pneumonia. The definitions of pneumonia adopted by the individual investigators were recorded but were not assumed to be an inclusion or exclusion criterion.
Data were extracted by one reviewer (MV) with a structured form and checked for accuracy by a second reviewer (ST). Differences were resolved by consensus.
Assessment of quality of trials
We assessed methodological quality of the trials using a procedure similar to that adopted by Cook et al.2 Five items were evaluated for each trial (patient selection, patient characteristics, randomisation, blinding, definition of bleeding or of pneumonia). Methodological quality was graded for each of the five items on a scale of 0, 1, or 2 (maximum score=10). Three observers independently assessed quality (AM, MV, MG). Differences were resolved by consensus.
Qualitative data synthesis
Descriptive data for each trial included characteristics of participants, details of intervention, and definition of outcomes. Our assessment of clinical heterogeneity was focused in particular on a comparison of the definitions of bleeding and pneumonia across the trials.
Quantitative data synthesis
With regard to the two end points, the odds ratio was used as the principal measure for comparing the treatment effect within each trial. The results specific for trials were combined through standard meta-analytic techniques6–8 to produce the summary odds ratio together with an assessment of the statistical intertrial heterogeneity. The meta-analysis calculations were based both on a fixed effect model 6 7 and on a random effect model. 6 8 The 95% confidence intervals for the odds ratio of individual studies were computed according to Wolf 9 (or by the “rule of three”10 when zeros were present). To avoid the problems of bias and instability associated with estimation of odds ratios, 0.5 was added to each cell of the fourfold tables in keeping with Cook et al.2 Heterogeneity was assessed as previously described.7
Table 1 shows the trial flow for the five meta-analyses. There were 398 patients in meta-analysis A (five studies), 54 in meta-analysis B (one study), 311 in meta-analysis C (three studies), 226 in meta-analysis D (two studies), and 1825 in meta-analysis E (eight studies). Table 2 shows descriptive data for all the trials included in our meta-analyses.
Effectiveness of ranitidine v placebo (meta-analysis A)—Our literature search identified five trials,11–15 three of which had already been included by Cook et al.11–13 The mean (SD) quality score for these trials was 6.6 (0.9). With respect to the end point of clinically important bleeding, this meta-analysis (table 3) failed to show any significant benefit of ranitidine (summary odds ratio 0.72, 95% confidence interval 0.30 to 1.70, P=0.46 for fixed effect model; 0.95, 0.37 to 2.43, P=0.92 for random effect model; χ2 for heterogeneity 6.8, df 4, P=0.15).
Effectiveness of sucralfate v placebo (meta-analysis B)—Our literature search found the same randomised trials that had already been examined by Cook et al (three trials16–18 for the end point of overt bleeding and one trial11 for the end point of clinically important gastrointestinal bleeding). Because our analysis considered the end point of clinically important bleeding, only one trial11 met the inclusion criteria and so no meta-analysis was carried out. The quality score of the trial of Ruiz-Santana et al was 7.0. The results of the study by Cook et al remained unchanged (table 4) with no difference between sucralfate and placebo (1.26, 0.12 to 12.9, P=0.70).
Incidence of pneumonia with ranitidine v placebo (meta-analysis C)—Our third meta-analysis included three randomised studies 12 13 15 that compared the incidence of pneumonia between ranitidine and placebo (two of these trials 12 13 had already been included in the meta-analysis by Cook et al). The mean (SD) quality score for these trials was 6.0 (1.0). The analysis of these three trials (table 5) found no significant difference in the rate of pneumonia with ranitidine and placebo (summary odds ratio 0.98, 0.56 to 1.72, P=0.94 for fixed effect model; 1.10, 0.45 to 2.66, P=0.84 for random effect model; χ2 for heterogeneity 4.38, df 2, P=0.11).
Incidence of pneumonia with sucralfate v placebo (meta-analysis D)—Our fourth meta-analysis included two randomised studies 17 18 that compared the incidence of pneumonia between sucralfate and placebo (both trials had already been included in the meta-analysis by Cook et al). The quality score for these trials was 6.0 (1.4). The analysis of these two trials (table 6) found no significant difference in the rate of pneumonia with sucralfate and placebo (summary odds ratio 2.21, 0.86 to 5.65, P=0.10 for fixed effect model; 2.11, 0.79 to 5.64, P=0.14 for random effect model; χ2 for heterogeneity 0.30, df 1, P=0.58).
Incidence of pneumonia with ranitidine v sucralfate (meta-analysis E)—Our fifth meta-analysis included eight randomised studies 3 19–26 that compared the incidence of pneumonia with ranitidine and sucralfate (five of these trials19–23 had already been included in the meta-analysis by Cook et al). The quality score for these trials was 5.6 (2.3). The analysis of these eight trials (table 7) showed a significantly increased risk of pneumonia with ranitidine compared with sucralfate (summary odds ratio 1.35, 1.07 to 1.70, P=0.012 for fixed effect model; 1.51, 1.00 to 2.29, P=0.05 for random effect model; χ2 for heterogeneity 12.9, df 7, P=0.08).
Our overview of the controlled trials of ranitidine or sucralfate compared with placebo provides a picture of poor effectiveness. The single trial available on sucralfate11 does not allow any conclusion to be drawn; the trials on ranitidine11–15 show no difference compared with placebo.
Our results on ranitidine need to be compared with those previously published by Cook et al.2 In their assessment of effectiveness of H2 receptor antagonists Cook et al included five trials that used cimetidine29–33 and three trials with negative results that used ranitidine11–13 (plus one trial on ranitidine published in Spanish34 and one trial with negative results that used a combination of ranitidine and antacids,35 both of which did not meet the criteria for our meta-analysis). Cimetidine is probably effective at statistical levels, as out of the trials that used cimetidine three had positive results, one had significant results in patients at low risk, and one had negative results. A separate meta-analysis that we carried out for this purpose (table 8) showed a significant reduction in bleeding (with P<0.001 and P<0.01 according to the fixed effect and the random effect models, respectively).
Can we accept the hypothesis that cimetidine is effective and ranitidine is ineffective? Is this hypothesis pharmacologically plausible? Our separate analyses for individual drugs did not actually settle the entire question. In fact, the cohorts included in the studies that used these two drugs were small and the incidence of events was low, particularly in the case of ranitidine (10 cases of bleeding with the drug v 15 without prophylaxis; table 3); so the different results might reflect casual variations in the outcome rather than a true difference between the two H2 receptor antagonists. Another point of uncertainty about cimetidine is the methodological quality of studies published many years ago.
Design of new trials
Our results indicate that some points of consensus need to be revised. For example, recent randomised studies on prophylaxis for stress ulcers 3 4 have invariably compared (unproved) active treatments with one another but no longer use a placebo group, and authoritative recommendations suggest the use of ranitidine for prophylaxis for stress ulcers.36
New large scale randomised trials seem to be the only way to resolve this issue. New trials, however, may raise the ethical question of which treatment is appropriate for the control group. One possibility is to give cimetidine to the control group, but this solution raises the question of reusing a drug that has largely been abandoned. Another possibility is to conduct new large scale controlled trials of ranitidine compared with placebo, but the use of placebo can be questionable from an ethical point of view. A third solution could be to design new randomised trials according to a strategy of early treatment of stress ulcer with or without prophylaxis. In this latter case, after randomisation to prophylaxis or placebo the patients could be subjected to intensive gastrointestinal monitoring (for example, by examining nasogastric aspirate at short intervals) and, at the first signs of bleeding, their participation in the trial could be stopped with immediate initiation of an aggressive antisecretory treatment. A drawback of this third solution is that the clinical weight of the end point of early bleeding is less than that of the end point of clinically important bleeding.
Effect on pneumonia
The results of three meta-analyses that evaluated pneumonia were contradictory in some respects (ranitidine v placebo and sucralfate v placebo had the same incidence of pneumonia; for ranitidine v sucralfate there was a significantly higher incidence of pneumonia with ranitidine, P=0.012).The statistical power of these comparisons was better for meta-analysis E (1825 patients) than for meta-analysis C (311 patients); this could in part explain the higher incidence of pneumonia with ranitidine compared with sucralfate but not compared with placebo.
The large trial by Cook et al showed a trend towards an increased incidence of pneumonia with ranitidine v sucralfate (definite pneumonia associated with use of ventilator and pneumonia according to the definition of the Centers for Disease Control and Prevention was more common with ranitidine, at P=0.03 and P=0.13, respectively).3 Discrepancies between large trials and meta-analysis have been pointed out in cases where the meta-analysis was conducted before the large trial.37 In our study, however, the meta-analysis on pneumonia already included the large trial and, more importantly, its results did not differ much from those of the large trial (odds ratio of 1.35 v 1.18, respectively).
These arguments related to statistical power and agreement between large scale trials and meta-analyses do not, however, exclude the other explanation that the increase in rates of pneumonia after ranitidine was simply the result of casual variations in the outcome. In this context, the repeated application of statistical testing to many data sets might have contributed to the generation of some significant results.
Our main conclusion is that there are insufficient data on effectiveness to be able to conclude anything one way or the other. This can be an important argument for further trials.
What is already known on this topic
Ranitidine and sucralfate are widely used to prevent gastrointestinal bleeding in patients in intensive care
Several recommendations suggest this form of prophylaxis, but both the Food and Drug Administration and European Medicines Evaluation Agency have not given their approval
What this study adds
This analysis showed that ranitidine and sucralfate do not prevent gastrointestinal bleeding in patients in intensive care
Ranitidine can increase the risk of nosocomial pneumonia under certain circumstances
These findings are based on small numbers of patients and so firm conclusions cannot presently be proposed
Current recommendations on prophylaxis for stress ulcers should be revised according to these results
We thank Dr Enrico Tendi, head of the pharmaceutical service of the Careggi hospital, for stimulating the discussion on this topic.
Contributors: AM had the original idea for the present study, set up the project, designed the protocol, organised searches, supervised data extraction, supervised cross checking and validation work, assessed methodological quality of the trials, carried out statistical calculations, and discussed analysis and subsequent results. ST had the original idea for the present study, set up the project, designed the protocol, organised searches, checked accuracy of data extraction, supervised cross checking and validation work, carried out statistical calculations, and discussed analysis and subsequent results. MV was involved in the original project, helped to devise protocol, organised searches, extracted data from the studies, assessed methodological quality of the trials, arranged statistical input, collaborated in analyses, and discussed analysis and subsequent results. MG was involved in the original project, helped to devise protocol and organise searches, assessed methodological quality of the trials, collaborated in analyses, and discussed analysis and subsequent results. AC was involved in the original project, helped to devise protocol and organise searches, supervised data extraction, collaborated in analyses, and discussed analysis and subsequent results. The paper was written jointly by AM and ST. AM is guarantor.
Our literature search used the PubMed version of Medline on the internet (www.ncbi.nlm.nih.gov/entrez/). The syntax used for the Medline searches of our five meta-analyses was:
Meta-analysis A: ranitidine AND stress AND eng [LA] AND 1966:2000 [EDAT] AND (Randomized Controlled Trial [PT] OR random* [TW])
Meta-analysis B: sucralfate AND stress AND eng [LA] AND 1966:2000 [EDAT] AND (Randomized Controlled Trial [PT] OR random* [TW])
Meta-analysis C: ranitidine AND pneumonia AND eng [LA] AND 1966:2000 [EDAT] AND Randomized Controlled Trial [PT] OR random* [TW])
Meta-analysis D: sucralfate AND pneumonia AND eng [LA] AND 1966:2000 [EDAT] AND (Randomized Controlled Trial [PT] OR random* [TW])
Meta-analysis E: ranitidine AND sucralfate AND pneumonia AND eng [LA] AND 1966:2000 [EDAT] AND (Randomized Controlled Trial [PT] OR random* [TW])
In this syntax, “eng[LA]” identifies studies published in English language, “1966:2000[EDAT]” selects the year range according to the Entrez date, “Randomized Controlled Trial [PT]” selects this type of trials according to Medical Subjects Headings while “random*”[TW] indicates the word random (in its all variant endings) in the title or abstract; the last two conditions were stated with the “OR” Boolean operator to increase the power of the search.
Competing interests On other occasions, our group has received support from GlaxoWellcome (Italy) in the following terms: AM has received funds for a member of his staff (MV) working on a pharmacoeconomic project on colon cancer; ST has been reimbursed for attending three symposiums; MV received financial support for the colon cancer project and a fee for consulting about remifentanil; AC has received a fee for speaking and a fee for organising an educational seminar.