Re: An alternative interpretation of the data
Re: An alternative interpretation of the data
We provided pooled mortality rates for the convenience of readers so that they could form some idea of the average risk of mortality in each intervention group. One should not draw the conclusion that there was a true or meaningful difference in mortality between these two groups just because of a small absolute numerical difference. In fact, this difference of 0.6% represents the absolute risk difference (RD) for mortality. One needs to know the 95% CI and the statistical significance of such a difference. These were not provided for RD in the paper as, in consultation with the Cochrane Collaboration, we had pre-determined to use the odds ratio (OR) as the measure of pooled effect. We can, however, now also provide full information on the RD of 0.006 (95% CI = -0.01 to 0.02; P = 0.55). This is clearly non-significant and should be interpreted as insufficient evidence of any effect (either detrimental or beneficial) of PPIs on mortality. This was the principal conclusion stated in our paper.
Had the RD been non-significantly in favour of PPI treatment, we would still have reached the same conclusion.
There is a clear outlier effect from the study of Hasselgren et al in which the mortality rate was higher in the PPI-treated patients than in those given placebo (6.9% vs. 0.6%)1. We assume that Dr Penston would agree that a mortality of 0.6% among elderly patients with ulcer bleeding (which was the group specifically recruited for that trial) is, to say the least, unexpected – especially as many of the patients with high-risk endoscopic stigmata did not receive endoscopic haemostatic therapy. We invite Dr Penston to view the complete on-line version of our manuscript and consider the sensitivity analyses we conducted around this specific issue. When, by sensitivity analysis, we removed this trial, the pooled mortality rates became 5.0% for PPI and 5.2% for control (RD = 0.002; 95%CI = -0.02 to 0.01; OR = 0.95; 95%CI = 0.66 to 1.36). Again this result is far from significant – in any conventional sense of the word.
Furthermore, the non-significant trend in favour of control treatment was clearly not robust to the exclusion of the trial with the most extreme results; in fact the "non-significant trend" can easily be inverted in favour of PPI treatment. Mortality has been surprisingly difficult for us to study because, compared to re-bleeding, it was a relatively uncommon event. We feel justified in stating that “… there may have been too few patients in our pooled analysis of mortality data to enable us to detect a difference” because there were only 71 deaths among 1371 patients on PPI treatment in the pooled mortality analysis, and 65 out of 1403 patients on control treatment. It is hardly surprising that these pooled rates are not robust to sensitivity analysis.
In the complete on-line version of the meta-analysis, we carefully identified those RCTs that were placebo-controlled and those that compared a PPI with an H2-receptor antagonist (H2RA). In one of the pre-specified sub-group analyses, we found essentially the same overall conclusions from the placebo-controlled and the H2RA-controlled trials. We do not, therefore, accept Dr Penston’s imputation that valid conclusions can be drawn only from placebo-controlled trials. However, if he wishes to do so, he will find essentially the same conclusions as from our full analysis.
Dr Penston attempts to support his arguments by “vote counting”: simply adding up the numbers of “positive” and “negative” studies for individual end-points of interest. Such a simplistic approach cannot take account of the relative sample sizes and methodological quality scores of the individual trials and is best avoided lest unsupportable conclusions are prematurely drawn.
Dr Penston also questions our comments on the consistency of our findings. In this, it must be made clear that we did not use consistency to refer to results of individual trials, but rather to the pooled results of the various pre-determined sub-group analyses regarding re-bleeding and surgical intervention rates. (Indeed, had the results of individual RCTs all been consistent with each other, there would have been no need for a systematic review and meta-analysis.) We invite Dr Penston to examine Table 3 in the on-line version of our paper. We should state that such a table was – for the purposes of space – omitted from the print version of the paper but is freely available on-line. In each of these sub-group analyses, the findings were highly consistent – that is that PPI therapy did not influence all-cause mortality but did reduce re-bleeding and usually reduced surgery.
Dr Penston also questions our impartiality. As required by the British Medical Journal and the Cochrane Collaboration (and as we would anyway have insisted upon), we provided a full and detailed summary of our relationships with those pharmaceutical companies marketing PPIs. We believe in the principle of full disclosure and will leave it up to the readership of the BMJ to decide if our judgement has been in any way impaired by these relationships. It is our collective and sincere belief that it has not. Furthermore, this was not a concern voiced by the editorial staff of the BMJ or by its expert external reviewer. Nor was this concern expressed by any of the four international experts who acted as reviewers of our manuscript for the Cochrane Collaboration.
Dr Penston’s closing paragraphs contains errors of perception, errors of fact and differences of opinion and interpretation. First, and as we have extensively reviewed above, we have never attempted to promote the notion that PPIs improve survival from ulcer bleeding. Second, we believe that there is evidence for reduced rates of re-bleeding and surgical intervention resulting from PPI therapy for ulcer bleeding in European trials. However, the magnitude effect is noticeably greater in Asian trials. A post hoc analysis comparing results of trials conducted in Asia and elsewhere is currently in press elsewhere and was presented in 2004 at Digestive Diseases Week in New Orleans 2. Lastly – and something that we can hopefully agree upon as a difference of opinion or emphasis – we do not agree that there is no role for IV PPI therapy for patients with recently bleeding peptic ulcers. For those patients who receive endoscopic haemostatic therapy for ulcer bleeding, there is compelling evidence that such therapy reduces the re-bleeding risk. Although, the individual trials administered IV PPI therapy for 72 hours, it is our practice to switch patients to oral PPI therapy sooner than that if they are clinically stable and can take oral medication. We agree with Dr Penston’s impression that IV PPI therapy may have been over-used in certain circumstances. We cannot make any concluding statements about the value – or lack thereof – of administering IV PPI therapy before diagnostic endoscopy in patients with upper gastrointestinal tract bleeding since our review focused only on patients with endoscopically-confirmed bleeding ulcer.
Grigoris I Leontiadis, consultant gastroenterologist, Department of Gastroenterology, University Hospital of North Durham, Durham DH1 5TW
Virender K Sharma, associate professor, Division of Gastroenterology, Mayo Clinic, Scottsdale, AZ 85259, USA
Colin W Howden, professor, Division of Gastroenterology, Northwestern University Feinberg School of Medicine, Chicago, IL 60611, USA firstname.lastname@example.org
1. Hasselgren G, Lind T, Lundell L, Aadland E, Efskind P, Falk A, et al. Continuous intravenous infusion of omeprazole in elderly patients with peptic ulcer bleeding. Results of a placebo-controlled multicenter study. Scand J Gastroenterol 1997;32: 328-33.
2. GI Leontiadis, L McIntyre, VK Sharma, CW Howden. Influence of geographical location of randomized controlled trials on effectiveness of PPI treatment in ulcer bleeding: A post hoc analysis of a Cochrane Collaboration systematic review. Gastroenterology 2004, 126: A-192.
Competing interests: As stated in our initial publication
Competing interests: No competing interests