Morgenstern, J. PPIs should not be prescribed for upper GI bleeds (pre-endoscopy), First10EM, March 30, 2022. Available at:
The topic of PPIs for upper GI bleeds was one of the first posts on First10EM. There is no new evidence, and the bottom line is the same (just don’t use them), so if you are a long time reader, you can probably skip this post. However, that post was rather messy, and needed some updates. If you are still using PPIs, or know someone who is, this is the evidence summary for you.
A quick overview from Cochrane
The Cochrane review looking at proton pump inhibitor treatment initiated prior to endoscopic diagnosis in upper gastrointestinal bleeding identified 6 RCTs including 2223 patients. (Sreedharan 2010) The key findings from this meta-analysis were:
- No change in mortality (odds ratio (OR)1.12; 95% CI 0.72 to 1.73) 6.1% vs 5.5%
- No change in rebleeding (OR 0.81; 95%CI 0.61 to 1.09) 13.9% vs 16.6%
- No change in need for transfusion (OR 0.95; 95% CI 0.78 to 1.16)
- No change in need for surgery (OR 0.96 95% CI 0.68 to 1.35) 9.9% vs 10.2%
Their conclusion is that “there is no evidence that PPI treatment affects clinically important outcomes, namely mortality, rebleeding or need for surgery.” That is a pretty damning conclusion for such a widespread medical practice. There was one statistically positive finding in this review. PPIs decreased the rate of “stigmata of recent hemorrhage” (37.2% versus 46.5%, OR 0.67; 95% CI 0.54 to 0.84). However, this is not a patient oriented outcome, and the Cochrane review isn’t even convinced it is real. “This result was not robust, that is, it became statistically nonsignificant with the exclusion of one of the trials.”
The numbers in this review make it pretty clear we are not going to see a difference in transfusion or surgery. Rebleeding is a somewhat more nebulous concept, with multiple definitions, and varying impact on patients. There is a suggestion that larger studies might find a benefit in this area. (Which would make sense, as rebleeding is decreased when we give PPIs to patients after the diagnosis of a peptic ulcer is made.) On the other hand, we have to be careful, because the best current summary has the point estimate for mortality being higher in the PPI group, which we will discuss further below.
The key RCTs
Honestly, I think the Cochrane review tells you basically everything you need to know about this topic. There is no clinically important benefit from PPI therapy, and the only purported benefit is highly questionable. However, because the practice is so widespread, it is worth knowing some of the RCTs in detail.
Lau 2007: The positive study that wasn’t
Lau (2007) is the study that really drives much of modern practice, but unfortunately the published manuscript does not accurately represent the original study design. This is a double-blind RCT from a single hospital in Hong Kong that randomized adult patients with overt upper GI bleeds to either omeprazole (80mg IV bolus followed by 8mg/hr IV infusion) or placebo. Patients with hemodynamic instability (HR>110 or BP <90) were resuscitated first, and then only included in the trial if their condition stabilized. If they remained unstable, they went for urgent endoscopy and were not included in the trial. They excluded long term aspirin users, as they were included in a different study. They enrolled 638 patients, with a mean age of 62, approximately 66% were male, only 3% were on anticoagulation, and 60% of whom were ultimately diagnosed with a peptic ulcer at the time of endoscopy. The mean time to endoscopy was 16 hours.
According to clinicaltrials.gov, the original primary outcome of this trial was mortality, and there was no difference (2.5% with omeprazole and 2.2% with placebo, p=0.78). However, the manuscript focuses on a secondary outcome, need for endoscopic intervention, which they claim was their primary outcome. (It was only submitted to clinicaltrials.gov as the primary outcome 2 months after the article was published, so they had clearly looked at their data when they made this change.) 19% of the omeprazole group and 28% of the placebo group (p=0.007) had an endoscopic intervention performed, but this is a subjective and disease oriented outcome, which was also a secondary outcome, so clearly should not drive clinical practice. There were no changes in any of the patient oriented outcomes, including need for urgent endoscopy (0.3% vs 0.2%, p=0.79), need for surgery (2% vs 2%, p=1.00), units of blood transfused (1.5 vs 1.9, p=0.12), or rebleeding (3.5% vs 2.5%, p=0.49).
So this is clearly a negative study. The primary outcome was negative, and all the patient important outcomes were negative. It should not have been written up as a positive study. The fact that there is a positive secondary outcome is not surprising, and is more likely to represent a chance finding than a true finding. Either way, it is hypothesis generating only, and shouldn’t drive clinical practice.
There are some other obvious limitations to this study. It is a single center study, and the single center seems to be rather unique. They enrolled 638 patients in 17 months, which is more than 1 upper GI bleed every day. I think that is a lot more than most hospitals. (They excluded another 873 upper GI bleed patients, so they were actually seeing about 3 upper GI bleed patients every day.) This study excluded the sickest patients, so we are simply unable to comment on the impact of omeprazole in critically ill patients based on this data. (Some people use this lack of evidence to argue we should treat the sickest patients, but the absence of evidence doesn’t make a benefit more likely. Harms are just as likely as benefits. Actually, in medicine, harms are usually more likely than benefit from unproven therapies. For PPIs in particular, we have a study with sicker patients showing an increase in mortality with omeprazole.) The exclusion of long term aspirin users is also somewhat awkward, as I think the majority of patients with upper GI bleeds that I see are on aspirin.
Bottom line: Despite the conclusions of the publication, this was clearly a negative study, with no change in the primary outcome, and no change in any clinically important outcomes.
Daneshmend 1992: The other “positive” study
The second study that is sometimes talked about as if it is positive is Daneshmend 1992, but the authors actually appropriately conclude that “omeprazole failed to reduce mortality, rebleeding, or transfusion requirements”. This is a double blind RCT enrolling patients with overt upper GI bleeding at 2 hospitals in England. They excluded patients with such severe bleeding that immediate surgery was required, as well as patients on warfarin. They compared omeprazole (80mg IV initial bolus, then 40mg IV boluses every 8 hours for 3 doses, then 40 mg orally twice a day for 3 days) to matched placebos. Endoscopy was performed within 24 hours of admission, although urgent endoscopy was available if required.
They enrolled 1154 patients. The mean age was 60, about 66% were male, and 25% had known peptic ulcer disease. 92% underwent endoscopy in the first 24 hours, with those who didn’t get endoscopy mostly going for urgent surgery. They don’t state a primary outcome, but instead say that they were looking at “death, rebleeding, and operation rates and transfusion requirements.” There were no statistical differences in any of these outcomes. Mortality was unchanged (6.9% with omeprazole and 5.3% with placebo, which obviously trends in the wrong direction.) Rebleeding was unchanged (18% vs 15%). Transfusions were unchanged (53% vs 52%). Need for surgery was unchanged (11% vs 11%). They do report a statistical difference in signs of bleeding at endoscopy (45% vs 33%), but this is a secondary outcome, and clearly not important if there are no changes in rebleeding, transfusion requirements, surgery, or mortality.
Of course, this isn’t a perfect trial. Once again, the exclusions could be problematic. Anticoagulated patients make up a large proportion of what I see, and we still don’t know what to do with our critically ill patients. This study only took place at 2 hospitals, and I imagine patient populations and general medical management was quite different in 1986 when compared to 2022.
As a comparison to Lau, this study enrolled 1154 patients at 2 hospitals over 3.5 years, which is less than half a patient a day, and seems more in keeping with what I might expect.
Bottom line: Although this study has 4 primary outcomes, there were no differences in any of them. It is a clearly negative study that does not support the use of omeprazole.
Hawkey 2001: Another negative study with a positive spin
This is a randomized, double blind, double dummy trial that looked at both lansoprazole (60mg orally, the 30 mg PO QID) and TXA, as compared to placebo in patients with suspected upper GI bleeds at 2 hospitals in England. Patients were excluded if they needed immediate surgery or had any coagulopathy (including therapeutic anticoagulation). The trial wasn’t registered. They talk about a lot of different outcomes, and it isn’t clear which was supposed to be their primary outcome. There were 4 endoscopic outcomes and 4 clinical outcomes. They use an intention to treat analysis for the clinical outcomes, but for the endoscopic outcomes they exclude everyone who didn’t have an endoscopy, or who had an endoscopy but didn’t have signs of bleeding, or were ‘not evaluable’. This is a big problem, as they retrospectively exclude 186 patients out of only 414 randomized, and all of these patients received the study drugs.
The population is what you would expect: mean age about 60, and about 60% male. Median time to endoscopy was 19 hours after hospital admission. The final study group only includes 228 patients out of the 414 enrolled. There were no differences in any of the clinical outcomes: blood transfusions, rebleeding, surgery, or death. There were no differences in 3 of the 4 endoscopic findings: stigmata of hemorrhage, active bleeding, or fresh blood in the stomach. There was a difference in total blood in stomach based on a qualitative scoring system, but they only present the statistics after fitting their data to a model, and only in the small subset of patients with signs of blood at endoscopy. Furthermore, the amount of blood in the stomach can’t possibly be clinically significant if there is no difference in any of the clinical outcomes.
Bottom line: This is pretty clearly a negative study, with no clinical benefit and only 1 of 4 endoscopic findings being statistically positive positive
RCTs I cannot access
Unfortunately, there are 3 other RCTs included in the Cochrane review that I cannot find anywhere:
- Hulagu S, Demirturk L, GulS, Yazgan Y, Altin M, Danaci M. The effect of omeprazole or ranitidine intravenous on upper gastrointestinal bleeding. Endoskopi Journal 1995;2:35-43.
- Naumovski-Mihalic S, Katicic M, Colic-Cvlje V, Bozek T, Prskalo M, SabaricB, et al. Intravenous proton pump inhibitor in ulcer bleeding in patients admitted to an intensive care unit. Gastroenterology 2005;128 suppl 4:W1578
- Wallner G, Ciechanski A, Wesolowski M, Sory A, Misiuna P. Treatment of acute upper gastrointestinal bleeding with intravenous omeprazole or ranitidine. European Journal of Clinical Research 1996;8:235-43.
If anyone has access to these papers, send them my way, and I will add summaries to this post.
If you are tempted to believe the secondary outcomes, you would want to see consistency across the studies. That consistency doesn’t exist. For example, although Daneshmend 1992 found a statistical difference in “stigmata of recent hemorrhage”, the Cochrane review notes that there is not a statistical difference when all the other studies are pooled. (Sreedharan 2010) Each study seems to find a positive secondary outcome, but it is different in each case, suggesting these blips are more likely to be noise than signal. More importantly, these disease oriented surrogate outcomes cannot possibly be important if they don’t have any effective on clinical outcomes, which they don’t in these studies.
I often hear people argue that these studies don’t apply to their patients, because patients were scoped immediately. I am not sure where that meme originated, but in all of these studies endoscopy was scheduled for the next day, which is a similar practice to all the hospitals I have worked in. The other concern is that critically ill patients were excluded. Although that is a definite shortcoming of this data, every single study is negative, so there is absolutely no reason to believe that PPIs would be any more effective in the critically ill than they are in the studied populations. Furthermore, as is discussed below, although PPIs probably have a role after the diagnosis of a bleeding peptic ulcer is made, they still do not change mortality, so there is no reason to think they will save our critically ill patients.
Some other data
The most common argument for using PPIs prior to endoscopy is that they have to be given to most patients after endoscopy anyway. However, even after endoscopy, the benefit of PPIs is relatively limited.
The Cochrane review on the topic identified 24 RCTs encompassing 4373 patients. (Leontiadis 2006) PPIs have no effect on mortality (3.9% vs 3.8%, OR 1.01, 95% CI 0.74 to 1.40). Obviously these are small studies, and the confidence intervals are still large, but it is important to know that PPIs are not life saving medications, so there is really no rush to get them started.
There does seem to be benefit from using PPIs in patients diagnosed with peptic ulcer bleeding, but the overall benefit is small and one might question the true benefit to the patient. There is a decrease in rebleeding (11% vs 17%, OR 0.49, 95%CI 0.37-0.65, with significant heterogeneity). However, with no change in mortality, the importance of these rebleeds might be questioned. There is also a decrease in surgery (6% vs 9%, OR 0.61, 95%CI 0.48-0.78). I would definitely like to avoid surgery, but a NNT of more than 30 certainly doesn’t suggest any need to rush to PPIs in the ED. (And PPIs started in the ED have no effect on surgery or rebleeding.)
The data on PPIs post-endoscopy does get more interesting, as there is some indication of a variance in effect between Asian and Western countries. In the 8 trials that took place in Asia, there was a statistically significant decrease in mortality with PPI therapy (1.5% vs 4.4%, OR 0.35, 95% CI 0.16-0.74). However, considering this complete lack of benefit in the total population, this apparent benefit in a subgroup must be balanced out by harm in another subgroup. Although not statistically significant, mortality does look higher in the non-Asian countries (4.8% vs 3.6%, OR 1.36, 95%CI 0.94-1.96). Furthermore, they run a sensitivity analysis in which this increase in mortality is statistically significant in non-Asian countries. Although subgroups are notoriously unreliable, for me in Canada, this suggests that PPIs either have exactly no effect on mortality, or increase mortality. Neither is a very good reason to prescribe these drugs.
As mentioned above, in the largest trial of PPIs in the pre-endoscopy phase, mortality was higher in the PPI group, although not statistically significant. (Daneshmend 1992) That is somewhat concerning, in light of the Cochrane analysis suggesting possible increased mortality in non-Asian countries when PPIs are given post-endoscopy. The point estimate for mortality was also higher with PPIs in the pre-endoscopy trials, although the confidence intervals are huge – odds ratio 1.12; 95% CI 0.72 to 1.73. (Sreedharan 2010)
Adding to this harm signal, one of the highest quality post-endoscopy trials had to be stopped early because of a signal of increased mortality in the PPI group. (Hasselgren 1997) This was a multi-centre, double-blind, placebo controlled RCT from 29 hospitals in Sweden and Norway. Patients over the age of 60 with an upper GI bleed and a peptic ulcer in the stomach were randomized to omeprazole (80mg IV bolus then 8mg/hr infusion) or placebo. They stopped the study after enrolling 333 patients. There was no difference in rebleeding or endoscopic treatment. There was a decrease in surgery with omeprazole, but there was a large increase in 21 day mortality (6.9% versus 0.6%, p=0.12) that resulted in the trial being stopped for safety reasons. However, despite this increase in mortality, they use a weird 5 point scale as their primary outcome, which essentially counts mortality as equivalent with blood transfusions and endoscopic treatment, and published their results as if they are positive. This abstract is probably in the running for the most biased ever written. They stopped the trial because of a 6.3% absolute increase in mortality, and they conclude “three days’ infusion of omeprazole improved overall outcome and reduced need for intervention in PUB patients.” That seems absolutely ridiculous to me.
I think the complete lack of a dose response relationship with PPIs is also subtle evidence of the overall lack of benefit. Intermittent intravenous therapy is no different than continuous. (Sachar 2014) High dose PPI (80mg bolus followed by 8 mg/hr) is no different than lower doses. (Neumann2013; Villalon 2016; Wu 2010) Oral PPIs are no worse, and perhaps better than intravenous. (Rodriguez 2017) If it doesn’t matter how the drug is given, or how much is given, it might not matter if the drug is given at all.
I don’t find any of the secondary outcomes suggesting a benefit in stigmata of hemorrhage or need for endoscopic intervention convincing. They were secondary outcomes, with different trials finding different benefits. More importantly, these are surrogate outcomes, and there is absolutely no indication that they are translating into clinical benefits for patients. (In fact, the best evidence seems to suggest that we might be killing patients). However, if you are someone who finds these endoscopic surrogates interesting, it is important to know that they are subjective and incredibly unreliable. If you show expert endoscopists the exact same video-taped endoscopy, they don’t agree at all on the classification of the lesions. For the very obvious high-grade lesions (1a and 1b lesions), there is poor to moderate agreement (kappa of 0.76 and 0.61 respectively). For all other lesions, agreement was horrible, with kappas between 0.44 and 0.49. (Mondardini 1998)
Although we tend to think of PPIs as incredibly safe medications, there is some evidence that we might be wrong. In one meta-analysis of 28,559 patients comparing PPIs to histamine-2 receptor antagonists in stress ulcer prophylaxis in critical care, they found a small but statistically significant increase in mortality with PPIs (RR 1.05; 95% confidence interval 1.00-1.10; with an estimated risk difference for mortality of 9 additional deaths per 1000 patients exposed to PPI). (Lee 2021)
Despite the fact that many admitting doctors seem to consider PPIs standard of care, the official guidelines are quite lukewarm. The American College of Gastroenterology only says that PPIs “may be considered to decrease the proportion of patients who have higher risk stigmata of hemorrhage at endoscopy and who receive endoscopic therapy. However, PPIs do not improve clinical outcomes such as further bleeding, surgery, or death (Conditional recommendation).” (Laine 2012) In my eyes, that reads very close to ‘should not be used routinely’, and actually seems to conflict with current practice.
I think the NICE guidelines (UK National Institute for Health and Care Excellence) are even more appropriate: “Do not offer acid-suppression drugs (proton pump inhibitors or H2-receptor antagonists) before endoscopy to patients with suspected non-variceal upper gastrointestinal bleeding.” (Dworzynski 2012)
The clinical summary
Although there are still large gaps in this evidence base, I think the clinical conclusion is very clear: PPIs should not be given. Every RCT conducted to date has been negative. There is no indication of clinical benefit, and we should be concerned about the possibility of an increase in mortality.
Given the (marginal) benefit in patients with known bleeding ulcers, if a patient had a recent endoscopy with known peptic ulcer disease, I will give a single IV bolus of PPI. Given that the largest studies excluded the most critically ill patients, it seems reasonable to run further RCTs focusing specifically on that population. However, without further studies, I think it is very clear that we should not be prescribing PPIs, given their potential for harm and absolute lack of proven benefit.
Daneshmend TK, Hawkey CJ, Langman MJ, Logan RF, Long RG, Walt RP. Omeprazole versus placebo for acute upper gastrointestinal bleeding: randomised double blind controlled trial. BMJ. 1992 Jan 18;304(6820):143-7. doi: 10.1136/bmj.304.6820.143. PMID: 1737157
Dworzynski K, Pollit V, Kelsey A, Higgins B, Palmer K; Guideline Development Group. Management of acute upper gastrointestinal bleeding: summary of NICE guidance. BMJ. 2012 Jun 13;344:e3412. doi: 10.1136/bmj.e3412. PMID: 22695897
Hawkey GM, Cole AT, McIntyre AS, Long RG, Hawkey CJ. Drug treatments in upper gastrointestinal bleeding: value of endoscopic findings as surrogate end points. Gut. 2001 Sep;49(3):372-9. doi: 10.1136/gut.49.3.372. PMID: 11511559
Hasselgren G, Lind T, Lundell L, Aadland E, Efskind P, Falk A, Hyltander A, Söderlund C, Eriksson S, Fernström P. Continuous intravenous infusion of omeprazole in elderly patients with peptic ulcer bleeding. Results of a placebo-controlled multicenter study. Scand J Gastroenterol. 1997 Apr;32(4):328-33. doi: 10.3109/00365529709007680. PMID: 9140154
Lau JY, Leung WK, Wu JC, Chan FK, Wong VW, Chiu PW, Lee VW, Lee KK, Cheung FK, Siu P, Ng EK, Sung JJ. Omeprazole before endoscopy in patients with gastrointestinal bleeding. N Engl J Med. 2007 Apr 19;356(16):1631-40. doi: 10.1056/NEJMoa065703. PMID: 17442905
Laine L, Jensen DM. Management of patients with ulcer bleeding. Am J Gastroenterol. 2012 Mar;107(3):345-60; quiz 361. doi: 10.1038/ajg.2011.480. Epub 2012 Feb 7. PMID: 22310222
Lee TC, Goodwin Wilson M, Lawandi A, McDonald EG. Proton Pump Inhibitors Versus Histamine-2 Receptor Antagonists Likely Increase Mortality in Critical Care: An Updated Meta-Analysis. Am J Med. 2021 Mar;134(3):e184-e188. doi: 10.1016/j.amjmed.2020.08.004. Epub 2020 Sep 12. PMID: 32931766
Leontiadis GI, Sharma VK, Howden CW. Proton pump inhibitor treatment for acute peptic ulcer bleeding. Cochrane Database Syst Rev. 2006 Jan 25;(1):CD002094. doi: 10.1002/14651858.CD002094.pub3. Update in: Cochrane Database Syst Rev. 2010;(5):CD002094. PMID: 16437441
Mondardini A, Barletti C, Rocca G, Garripoli A, Sambataro A, Perotto C, Repici A, Ferrari A. Non-variceal upper gastrointestinal bleeding and Forrest’s classification: diagnostic agreement between endoscopists from the same area. Endoscopy. 1998 Aug;30(6):508-12. doi: 10.1055/s-2007-1001335. PMID: 9746157
Neumann I, Letelier LM, Rada G, Claro JC, Martin J, Howden CW, Yuan Y, Leontiadis GI. Comparison of different regimens of proton pump inhibitors for acute peptic ulcer bleeding. Cochrane Database Syst Rev. 2013 Jun 12;(6):CD007999. doi: 10.1002/14651858.CD007999.pub2. PMID: 23760821
Rodriguez EA, Donath E, Waljee AK, Sussman DA. Value of Oral Proton Pump Inhibitors in Acute, Nonvariceal Upper Gastrointestinal Bleeding: A Network Meta-Analysis. J Clin Gastroenterol. 2017 Sep;51(8):707-719. doi: 10.1097/MCG.0000000000000625. PMID: 27505402
Sachar H, Vaidya K, Laine L. Intermittent vs continuous proton pump inhibitor therapy for high-risk bleeding ulcers: a systematic review and meta-analysis. JAMA Intern Med. 2014 Nov;174(11):1755-62. doi: 10.1001/jamainternmed.2014.4056. PMID: 25201154
Sreedharan A, Martin J, Leontiadis GI, Dorward S, Howden CW, Forman D, Moayyedi P. Proton pump inhibitor treatment initiated prior to endoscopic diagnosis in upper gastrointestinal bleeding. Cochrane Database Syst Rev. 2010 Jul 7;2010(7):CD005415. doi: 10.1002/14651858.CD005415.pub3. Update in: Cochrane Database Syst Rev. 2022 Jan 7;1:CD005415. PMID: 20614440
Villalón A, Olmos R, Rada G. Are higher doses of proton pump inhibitors better in acute peptic bleeding? Medwave. 2016 Jun 24;16 Suppl 2:e6476. English, Spanish. doi: 10.5867/medwave.2016.6476. PMID: 27390875
Wu LC, Cao YF, Huang JH, Liao C, Gao F. High-dose vs low-dose proton pump inhibitors for upper gastrointestinal bleeding: a meta-analysis. World J Gastroenterol. 2010 May 28;16(20):2558-65. doi: 10.3748/wjg.v16.i20.2558. PMID: 20503458