Dr C. W. Howden, Division of Gastroenterology, Northwestern University, Feinberg School of Medicine, 676 N. St Clair Street, Suite 1400, Chicago, IL 60611, USA. E-mail: firstname.lastname@example.org
Background: Proton-pump inhibitors reduce re-bleeding and surgical intervention, but not mortality, after ulcer bleeding.
Aim: To examine the effects of proton-pump inhibitor treatment on transfusion requirements and length of hospital stay in patients with ulcer bleeding.
Methods: For the Cochrane Collaboration meta-analysis of randomized-controlled trials of proton-pump inhibitor therapy for ulcer bleeding, outcomes of transfusion requirements and hospital stay were summarized, respectively, as mean (±s.d.) units transfused and hospital days. We calculated weighted mean difference with 95% confidence interval. We also performed subgroup analyses according to geographical origin of the randomized-controlled trials.
Results: There was significant heterogeneity among randomized-controlled trials for either outcome. Overall, proton-pump inhibitor treatment marginally reduced transfusion requirements (WMD = −0.6 units; 95% CI: −1.1 to 0; P = 0.05) and length of hospitalization (WMD = −1.1 days; 95% CI: −1.5 to −0.7; P < 0.0001). Most of the randomized-controlled trials did not state precise criteria for administering blood transfusion and discharging patients, thereby limiting the strength of conclusions on the pooled effects.
Conclusions: Proton-pump inhibitor treatment for ulcer bleeding produces small, but potentially important, reductions in transfusion requirements and length of hospitalization.
Peptic ulcer is the most frequent cause of acute upper gastrointestinal tract bleeding resulting in hospitalization.1 There is a substantive clinical and economic burden associated with the management of patients with ulcer bleeding. Proton-pump inhibitors (PPIs) are frequently used as part of initial management. The Cochrane Collaboration systematic review and meta-analysis of randomized-controlled trials (RCTs) of PPI treatment for ulcer bleeding found no significant effect on 30-day all-cause mortality, which was the predetermined primary end-point.2, 3 PPI treatment did significantly reduce rates of ulcer re-bleeding and surgical intervention, which were predefined secondary end-points. Some RCTs also provided information on other clinically relevant end-points including blood transfusion requirements and length of hospital stay. Analyses of those end-points were not included in our original publication,3 and are now reported here.
Materials and methods
The search strategy and methods used for the Cochrane Collaboration systematic review and meta-analysis of RCTs comparing PPI treatment with placebo or an histamine-2 receptor antagonist (H2RA) for bleeding ulcer are available on-line from the Cochrane library (http://www.cochrane.org/reviews).2 Details about the design of individual RCTs are available on-line2 and in our initial publication.3 Briefly, we performed computerized searches of MEDLINE, EMBASE and the Cochrane Collaboration's trials register for RCTs that compared a PPI (either orally or i.v.) with placebo or an H2RA for endoscopically confirmed bleeding peptic ulcer. We supplemented that with a hand search of conference proceedings and by requesting any unpublished information from relevant pharmaceutical companies. Two assessors independently reviewed each identified RCT and extracted relevant data onto a purpose-designed form. Any disagreements were resolved by consensus. For inclusion, an RCT had to have confirmed the diagnosis of ulcer bleeding by endoscopy and randomized patients to either the PPI or control treatment. Any other treatment intervention, including endoscopic haemostatic treatment had to have been applied to patients in each arm. RCTs had to have reported at least one of the outcomes of all-cause mortality, re-bleeding and surgical intervention.
We also extracted all available information on blood transfusion requirements and length of hospital stay. Where available, the specific criteria used to determine indications for transfusion and discharge from hospital were recorded. Transfusion requirements were reported as mean (±s.d.) units of blood transfused. Length of hospital stay was recorded as mean (±s.d.) days in hospital. We performed meta-analysis of outcomes by combining trials by inverse variance methods. The software used was the Cochrane Collaboration's revman (version 4.2.2). Statistical heterogeneity was evaluated; P-values of <0.1 were considered statistically significant. We planned to use a fixed effect model unless we found significant heterogeneity, in which case we planned to use a random effects model. Outcomes were summarized as a weighted mean difference (WMD) with its 95% confidence interval (CI). We explored the robustness of the pooled effect estimates for each outcome by sensitivity analysis.
In a post hoc analysis of the Cochrane Collaboration meta-analysis, we reported that RCTs performed in Asia had found quantitatively greater effects of PPI treatment on re-bleeding and surgical intervention than those performed in Europe or elsewhere.4 Furthermore, the only evidence for a reduction in all-cause mortality with PPI treatment for ulcer bleeding came from a pooled subgroup analysis confined to the Asian RCTs.4 We therefore also performed subgroup analyses of RCTs for the end-points of transfusion requirements and length of hospital stay according to geographical origin.
Blood transfusion requirements
Eight RCTs provided information on transfusion requirements.5–12 In four of these, PPI treatment had been compared with placebo;6–8, 10 the remainder compared PPI treatment with an H2RA.5, 9, 11, 12 Together, these eight RCTs comprised 1197 patients with 595 randomized to PPI treatment and 602 to control. There was statistically significant heterogeneity among these RCTs (P < 0.00001). PPI treatment reduced transfusion requirements (WMD = −0.6 units; 95% CI: −1.1 to 0; P = 0.05; Figure 1). This result was not robust to the exclusion of individual trials; by sensitivity analysis, the effect became non-significant when any one of five RCTs was excluded.6, 7, 9–11
We sought to investigate the causes of the heterogeneity among the eight RCTs reporting transfusion requirements. Observation of the forest plot (Figure 1) suggested that the heterogeneity was due to a possible outlier effect of one trial.8 By sensitivity analysis, the remaining trials became non-heterogeneous (P = 0.67) when that trial, but not when any of the others, was excluded. Furthermore, the exclusion of that trial increased the statistical significance of the pooled difference in transfusion requirements in favour of PPI treatment (WMD =−0.3; 95% CI: −0.6 to −0.1; P = 0.01). A funnel plot (Figure 2) was clearly asymmetrical but did not suggest a preponderance of missing small trials with negative results.
Three of the trials had been conducted in Asia7, 8, 10 and five in Europe.5, 6, 9, 11, 12 The three Asian trials comprised 609 patients and there was significant heterogeneity among them for transfusion requirements (P = 0.0005). The pooled WMD was −1.01 units (95% CI: −2.0 to −0.2; P = 0.02). The five European trials comprised 588 patients and there was no significant heterogeneity among them (P = 0.88). The pooled WMD was −0.2 units (95% CI: −0.5 to 0.2; P = 0.31).
In the above analysis all three trials that had been conducted in Asia had used placebo as a comparator while only one6 of the five European trials had done so. Prompted by this observation, we performed a post hoc subgroup analysis of the above trials according to the comparator treatment. For the four European trials that compared PPI with H2RA,5, 9, 11, 12 WMD was −0.1 (95% CI: −0.7 to 0.4). For the one European6 and three Asian7, 8, 10 trials that compared PPI with placebo, WMD was −0.8 (95% CI: −1.6 to 0.05).
Length of hospital stay
Seven trials reported length of hospital stay as mean (±s.d.) days in hospital.5, 7–9, 11–13 Three compared PPI treatment with placebo;7, 8, 13 the remaining four compared it with an H2RA.5, 9, 11, 12 These seven trials included a total of 801 patients with 398 randomized to PPI treatment and 403 to control. There was statistically significant heterogeneity among them (P = 0.07). Figure 3 shows that WMD for each RCT was in favour of PPI treatment, although statistical significance was only reached in three trials;7, 8, 13 the pooled WMD for length of hospital stay was −1.1 days (95% CI: −1.5 to −0.7; P < 0.0001). By sensitivity analysis, this result remained statistically significant when any single trial was excluded. A funnel plot showed no asymmetry, suggesting no publication bias (Figure 4). In investigating possible sources of heterogeneity among these trials, inspection of the forest plot (Figure 3) suggested that this was due to results of two of the trials.7, 13 Although both of these trials demonstrated a statistically significant reduction in length of hospital stay with PPI treatment, the effect was more pronounced in the trial by Javid et al.13 than in that by Kavianni et al.7 with no overlapping of their respective 95% CI for WMD (Figure 3). These trials had similar design characteristics including discharge criteria. Kavianni et al.7 included only patients with active bleeding or non-bleeding visible vessel, while Javid et al.13 also included patients with adherent clot. Javid et al.13 also included relatively fewer patients with gastric ulcer (15 of 166 vs. 37 of 112; P = 0.0002). The different outcomes on hospital stay are not adequately explained by these differences. If either trial was excluded, the remaining trials became non-heterogeneous, while the pooled effect of the analysis remained statistically significant in favour of PPI treatment. With the exclusion of the trial by Javid et al.,13 there was no significant heterogeneity (P =0.29), and the pooled WMD was −0.9 days (95% CI: −1.4 to −0.5). With the exclusion of the trial by Kaviani et al.,7 there was no significant heterogeneity (P =1.00), and the pooled WMD was −1.4 days (95% CI: −1.6 to −1.1).
As the effect of treatment would be expected to be attenuated in patients with in-hospital onset of bleeding, we also performed a sensitivity analysis by excluding the two RCTs that included such patients;6, 12 the pooled result was unaffected (WMD = −1.1 days; 95% CI: −1.6 to −0.6).
Of the seven RCTs included in the above analysis, three were conducted in Asia7, 8, 13 and the remaining four in Europe.5, 9, 11, 12 The three Asian trials comprised 535 patients, and there was significant heterogeneity among them for length of hospital stay (P = 0.003). The pooled WMD was −1.1 days (95% CI: −1.7 to −0.5; P = 0.0002). The four European trials comprised a total of 166 patients with no significant heterogeneity among them (P = 1.00). The pooled WMD was −1.1 days (95% CI: −2.2 to 0.1; P = 0.06).
In the above analysis all three Asian trials had compared PPI against placebo, while all four European trials had compared PPI against H2RA. Thus, the post hoc analysis of the above trials according to comparator treatment produced numerically identical results: the reduction in hospital stay was statistically significant only in the trials that used placebo as comparator treatment.
Analysis of these tertiary end-points, which are of potential clinical relevance, indicates that PPI treatment for ulcer bleeding reduces transfusion requirements and length of hospital stay. The magnitude of the benefit was greater in Asian than European trials. In the subgroup analyses, these outcomes only achieved statistical significance among the Asian RCTs although there was a clear trend towards significance among the European RCTs for improvement in hospital stay. We have previously reported4 that PPI treatment for ulcer bleeding appears to be more efficacious in Asia than elsewhere regarding mortality, re-bleeding and surgical intervention rates, due possibly to an enhanced pharmacodynamic effect of PPIs among Asian patients. Possible reasons for such an effect might include a lower parietal cell mass,14 higher prevalence of Helicobacter pylori infection15 and higher prevalence of the ‘slow metabolizer’ phenotype through genetic polymorphism for cytochrome P450 2C19,16, 17 as previously discussed elsewhere.4 On the contrary, the observed difference in efficacy regarding transfusion requirements and length of hospital stay could be related to the fact that all Asian trials used in these two analyses had been placebo-controlled, while all European trials (all but one in the analysis of length of hospitalization) had used an H2RA as the comparator treatment. However, this explanation is less likely because it contradicts the results of our main meta-analysis that found no difference in the corresponding subgroup analysis examining primary and secondary end-points. Finally, the observed differences between Asian and European patients for these tertiary end-points may be due to chance given the relatively small number of patients for whom data were obtainable.
The true clinical significance of the observed effects of PPI treatment on transfusion requirements and length of hospital stay are undetermined. Although they are of small magnitude, they may be relevant financially given the frequency of hospitalization for ulcer bleeding. Consideration of these effects should be made during future cost-effectiveness analyses of PPI therapy for ulcer bleeding in health care systems of different countries.
There was significant heterogeneity among the trials for transfusion requirements. It is important to try and identify any possible reasons for heterogeneity in a meta-analysis. The trial by Khuroo et al.8 appeared to be an outlier (Figure 1), and its exclusion in a sensitivity analysis resolved the heterogeneity. This trial had found the largest difference in transfusion requirements in favour of PPI treatment. It differed from the others in that it was the only one to have included patients with spurting bleeding without using prerandomization endoscopic haemostatic treatment. Such patients would be expected to have high transfusion requirements because of ongoing bleeding and more severe episodes of re-bleeding. As a result, and assuming that PPI treatment was effective, this might have enhanced the difference in transfusion requirements between the two treatment arms. The precise criteria for administering blood transfusion were not given for that trial and two others.5, 11 We acknowledge that differences among RCTs for the precise criteria for deciding to administer blood transfusion limit the strength of conclusions on the pooled effects.
We considered whether mortality could have influenced the reporting of transfusion requirements, because those requirements would logically be lower among patients who died shortly after an episode of ulcer bleeding. Had this been the case, PPI treatment should have affected these two outcomes in opposite directions. For example, if there had been increased mortality among the PPI-treated patients, then there should have been a corresponding reduction in transfusion requirements. However, with the exception of one trial,6 this was not the pattern of outcomes observed: PPI treatment otherwise affected mortality and transfusion requirements in the same direction.
Mortality could also have influenced reporting of length of hospitalization; none of the trials distinguished length of hospital stay determined by death as opposed to hospital discharge. If mortality had influenced length of hospital stay, these should have been in opposite directions. For instance, if PPI treatment had decreased mortality, it should have increased the average length of hospital stay. None of the seven trials found a significant effect on mortality although two5, 12 reported non-significant increases in mortality and shorter lengths of stay with PPI treatment. The remaining five trials reported non-significant reductions in mortality and shorter lengths of stay on PPI treatment. Thus, there is insufficient evidence to accept or refute the possibility that mortality may have significantly biased reporting of length of hospital stay. Nevertheless, the strength of the above conclusion is limited by the fact that hospital stay could have been affected by possible variability among RCTs regarding discharge criteria, which were reported only in two RCTs,7, 13 and comorbidity, which could not be reliably quantified in the present meta-analysis.
In conclusion, these planned analyses of predetermined tertiary end-points found small – but potentially important – reductions in transfusion requirements and length of hospital stay with PPI treatment for ulcer bleeding. Effects of PPI treatment on such clinically relevant outcomes should be part of any future cost-effectiveness analyses in different health care delivery models.
Authors thank Ms Iris Gordon, Trial Search Coordinator, Cochrane Collaboration, Upper Gastrointestinal and Pancreatic Diseases Group, University of Leeds, UK for her help with the electronic literature search. Also grateful to Dr Linda McIntyre for assistance with designing the protocol for this review and subsequent data extraction.