The PEPTIC multi-center RCT compared histamine-2 receptor blockers (H2RBs) versus proton pump inhibitors (PPIs) for stress ulcer prophylaxis among ventilated ICU patients.1 It’s a landmark trial of unprecedented size and statistical power.
PEPTIC was presented at the Critical Care Reviews meeting in Ireland and live-cast around the world (available here). This presentation and the editorials gave a fairly coherent viewpoint of the study, as one which should generally caution us away from using PPIs, due to concerns regarding increased mortality. This viewpoint is summarized by Paul Young in a tweetorial here.
This post will seek to present an alternative perspective on the study, to counterbalance this prevailing interpretation. I won't focus on the fundamentals of the trial, as these have already been nicely covered in a blog by Simon Carley here.
The elephant in the room is whether PPIs kill people. So let’s tackle this issue right away.
part I: do PPIs kill people?
background: what is our pre-test probability that PPIs kill people?
There are numerous reasons to believe that PPIs don’t kill people:
- Affecting all-cause mortality in critically ill patients is extraordinarily difficult. So our pre-test probability that any medication affects mortality should generally be very low.
- PPIs have been blamed for all manner of adverse endpoints (ranging from hip fracture to dementia). However, this seems to reflect association rather than causation, because patients on chronic PPIs tend to have more medical problems.2,3
- PPIs are overall considered to be extremely safe (e.g. they are available over the counter).
- PPI use for stress ulcer prophylaxis involves minimal exposure (e.g. treatment for an average of 3 days).
- Prospective RCTs have not found a mortality difference in patients treated with PPIs, even for extended periods of time. For example, the COMPASS RCT involved eighteen thousand older patients randomized to placebo versus PPI therapy for three years, with no effect on mortality and or other serious adverse events.4
- Prior studies and meta-analyses of PPI use for GI prophylaxis haven’t detected mortality differences.5,6
- There is no known mechanism whereby a few tablets of PPI should kill people. Emerging evidence shows that PPIs don’t cause pneumonia or C. difficile infection (more on this below). PPIs may have some anti-inflammatory effects, but so do many other drugs (such as azithromycin and statins).
The PEPTIC trial was designed using mortality as a primary endpoint, with sufficient power to detect a 2.4% absolute mortality difference. For comparison, the absolute mortality benefit of stenting in acute STEMI is 2%.7 So this trial tested whether different flavors of antacid influence mortality strongly more than stenting a patient with STEMI.
primary mortality endpoint
The primary endpoint was negative – there was no statistically significant difference in mortality between the H2RB group and the PPI group.
There is a trend towards increased mortality in the PPI group, so this data doesn’t exclude the possibility that PPI increases mortality. However, the inability to prove that a drug is safe is an inherent limitation in the statistical basis of medical knowledge. In order to prove zero risk from any medication, an infinitely large study would be required.
mortality differences in the cardiac surgery subgroup
Fifteen subgroup analyses were performed. One of these found a mortality difference among the subgroup of patients status post cardiac surgery. In this group, PPI use was reported as causing a higher risk of death (RR 1.27, with a 95% confidence interval of 1.04-1.57).
This finding was reported as a potential signal that PPIs are dangerous. There are several reasons that this finding is probably spurious.
Perhaps most notably, this finding completely lacks statistical significance when subjected to a Fisher Exact Test (it has a calculated p-value of 0.13 as shown above). What’s going on here? The risk ratios reported in the study are adjusted results, based on complex multivariable models that correct for numerous factors (including randomization time period, ICU clustering unit, etc.). Thus, the raw data isn’t statistically significant – multivariate adjustment is required to obtain a statistically significant result.
There are some additional reasons that this finding may be spurious. This is a subgroup analysis (so arguably the p-value threshold should be reduced below 0.05 to account for multiple comparisons). Finally, there’s no clear reason to expect that PPIs would be especially deadly in this subgroup.
post-hoc mortality analysis by APACHE quartile
A post-hoc analysis of the SUP-ICU trial suggested that PPIs might increase mortality in sicker patients. In attempts to validate this concept, a post-hoc analysis was added to the PEPTIC trial. Patients were divided into quartiles, based on disease severity:
Ultimately this data fails to demonstrate any consistent relationship between adverse effects of PPIs and illness severity. There seems to be a signal of increased mortality in quartile 3, but not quartile 4. C. difficile infections are more problematic in quartile 2, but not quartile 3 or quartile 1. This feels more like statistical noise than signal.
Bayesian interpretation of the mortality endpoint
This post may seem excessively restricted by the traditional rules of frequentist statistics, so here is a very crude Bayesian interpretation of the mortality endpoint to balance things out.
The pre-test probability that PPIs increase mortality is the most challenging aspect to determine (as there is no objective way to quantify this). I would estimate this probability is roughly ~1%, for reasons described earlier. This hypothesis is possible, but is unsupported by prior evidence (both theoretical and empirical).
There is a rough correlation between p-values and Bayes factors, as shown above. A p-value around ~0.05 correlates roughly with a Bayes Factor of ~3.
A Bayes factor of 3 increases the likelihood that PPIs increase mortality from a 1% pre-study probability to a 3% post-study probability. So the likelihood that PPIs cause mortality increases, but this hypothesis remains unlikely.
This exercise illustrates my true reservation about the mortality results in PEPTIC, which is the lack of prior evidence supporting mortality risks from PPIs. For example, even if PEPTIC showed that PPIs increase mortality with a “statistically significant” p-value of 0.03, my post-study likelihood would still remain low. Perhaps this reveals that I am excessively swayed by my prior beliefs, but I would argue that it may prevent my practice from whipping around in the breeze based on the whims of the latest study.
OK, enough of this Bayesian interlude, let’s get back to traditional (frequentist) statistics.
conclusion on PPIs and mortality
PEPTIC is the largest and best-powered study in the history of critical care. It very thoroughly evaluates whether PPIs kill people. It adjusts for confounders in an inception-level matrix model. It slices and dices the mortality data in about twenty different ways. Ultimately, the study doesn’t find clear evidence that PPIs cause harm. One could view this as reassuring, rather than incriminating.
There is a trend towards increased mortality in the PPI group. This is frustrating, because it undeniably leaves open the possibility that PPIs kill people. The issue is whether or not we should change practice on the basis of a risk which might exist – but which hasn’t been proven to be real.
part II: the rest of the study
Aside from the confusion about mortality, the remainder of the study is a tour de force. It’s massive, well-designed, and it yields fairly clear results.
primary efficacy endpoint
Clinically significant gastrointestinal bleeding was substantially lower in patients treated with PPIs (0.5% absolute difference, p=0.009, fragility index of 27!). This result is particularly impressive due to cross-over, with many patients in the H2RA group actually receiving PPIs (probably because treating clinicians felt these patients were at very high risk of stress ulceration); such cross-over would be expected to reduce the measured efficacy of PPIs.
Although this was technically a secondary endpoint, I think this would have been more appropriate as a primary endpoint. So, as a rebellious counterpoint to the prevailing interpretation of this study, I might regard this as a positive study in favor of PPIs.
An absolute difference of 0.5% isn’t huge (this correlates with a number needed to treat of 200, in order to prevent one bleed). However, we need to consider that all intubated patients will likely be receiving either a PPI or an H2RB. If all patients will be getting a drug, choosing the most effective therapy makes sense (even if the difference isn’t huge).
This efficacy data indirectly illustrates an important point: stress ulcer prophylaxis still has benefit in 2020. Modern practices, including early resuscitation and nutrition, might reduce the risk of stress ulceration. This has led many to question whether stress ulcer prophylaxis is still worthwhile in a modern ICU context. This evidence suggests that stress ulcer prophylaxis remains useful.
safety endpoints
Perhaps equally important, this data provides substantial evidence of safety. The main concerns regarding the use of PPIs are increased risk of C. difficile colitis or possibly aspiration pneumonia. This data puts these issues to rest nicely:
- C. difficile was actually less likely in patients treated with PPIs. This argues strongly against the concept that PPIs cause C. difficile infection.
- There was no difference in the incidence of ventilator-associated conditions or in the duration of mechanical ventilation. This argues against any increased risk of pneumonia due to PPIs.
comparison of PEPTIC and SUP-ICU
PEPTIC and SUP-ICU are the largest, most recent, and highest quality studies on GI prophylaxis in the ICU. In any meta-analysis of stress ulcer prophylaxis, these studies will tower over all other studies. The studies aren’t exactly the same (PEPTIC compared PPI vs. H2RA, whereas SUP-ICU compared PPI vs. placebo). The main findings from SUP-ICU are:
The performance of PPI is remarkably consistent across both studies :
- PPI therapy is effective at reducing stress ulceration.
- PPI therapy does not increase the risk of C. difficile or pneumonia.
- PPI therapy doesn’t affect mortality (to a statistically significant extent).
If you want a hard-nosed, evidence-based answer to the issue of stress ulcer prophylaxis, this might be it. Mortality trends and post-hoc analyses are certainly provocative, but ultimately we’re not supposed to base our management on them.
- PEPTIC is the largest RCT in the history of critical care. This is undoubtedly a landmark trial, but exactly how to interpret it remains debatable. With great statistical power also comes great statistical noise (due to larger subgroups).
- PPIs are more effective than H2RAs in preventing stress ulceration (with an absolute risk reduction of 0.5%).
- PPIs didn’t cause increased rates of C. difficile or pneumonia. These findings are consistent with results from the SUP-ICU trial.
- There was a non-significant trend towards increased mortality in patients treated with PPIs. However, there is no clear mechanism whereby PPIs should kill people (especially given safety signals regarding C. difficile and pneumonia).
- The pendulum will continue to swing back and forth between H2RAs and PPIs. This study resolves several long-standing safety concerns regarding PPIs (including the risk of C. difficile and the risk of pneumonia). It also validates the ongoing practice of stress ulcer prophylaxis with PPIs (even in a modern ICU setting with early nutrition). However, the study simultaneously raises new concerns regarding the possibility of increased mortality. How to balance definite benefits versus potential risks is controversial.
related
- PEPTIC trial
- Manuscript at JAMA (currently open access)
- St. Emlyn's Blog (Simon Carley) – There are many important point here which I haven't repeated above – so please be sure to read this post as well.
- Tweetorial by Paul Young here (it's been converted into a single webpage).
- Background
references
- 1.Young PJ, Bagshaw SM, et al. Effect of Stress Ulcer Prophylaxis With Proton Pump Inhibitors vs Histamine-2 Receptor Blockers on In-Hospital Mortality Among ICU Patients Receiving Invasive Mechanical Ventilation. JAMA. January 2020. doi:10.1001/jama.2019.22190
- 2.Vaezi M, Yang Y, Howden C. Complications of Proton Pump Inhibitor Therapy. Gastroenterology. 2017;153(1):35-48. doi:10.1053/j.gastro.2017.04.047
- 3.Corley D. Safety and Complications of Long-Term Proton Pump Inhibitor Therapy: Getting Closer to the Truth. Gastroenterology. 2019;157(3):604-607. doi:10.1053/j.gastro.2019.07.039
- 4.Moayyedi P, Eikelboom J, Bosch J, et al. Safety of Proton Pump Inhibitors Based on a Large, Multi-Year, Randomized Trial of Patients Receiving Rivaroxaban or Aspirin. Gastroenterology. 2019;157(3):682-691.e2. doi:10.1053/j.gastro.2019.05.056
- 5.Wang Y, Ye Z, Ge L, et al. Efficacy and safety of gastrointestinal bleeding prophylaxis in critically ill patients: systematic review and network meta-analysis. BMJ. 2020;368:l6744. doi:10.1136/bmj.l6744
- 6.Krag M, Marker S, Perner A, et al. Pantoprazole in Patients at Risk for Gastrointestinal Bleeding in the ICU. N Engl J Med. 2018;379(23):2199-2208. doi:10.1056/NEJMoa1714919
- 7.Keeley E, Boura J, Grines C. Primary angioplasty versus intravenous thrombolytic therapy for acute myocardial infarction: a quantitative review of 23 randomised trials. Lancet. 2003;361(9351):13-20. doi:10.1016/S0140-6736(03)12113-7
- PulmCrit Wee: Rational selection of infusion rate based on loading dose - June 25, 2024
- PulmCrit: PPIs are safe and effective for GI prophylaxis… the end. - June 18, 2024
- PulmCrit: Bilevel Sequence Intubation (BSI) – The new standard - June 17, 2024
One possibility I didn’t see addressed by the subgroup/non-primary analyses that could be a mechanism/account for the mortality differences would be the use of omeprazole interacting with clopidogrel. The supplement shows many sites used omeprazole, and for some it was the main PPI used. Esomeprazole may have a similar effect and was also used. Didn’t see how many pts were taking clopidogrel but probably more of the post op cardio pts (where the mortality did was sig, supporting this idea). I know the clinical relevance of omeprazole impacting CV outcomes in pts on clopidogrel isn’t clear/controversial but the increase in… Read more »
We do not have these data I am afraid.
I had the same thought, at my old shop any patient who was on an antiplatelet would get a H2B as opposed to a PPI.
Nice post Josh. A couple of point to clarify: 1. Analyses that do not account for clustering effects are very misleading/incorrect. The statistically correct analyses are the ones presented. 2. As outlined in the supplement, our sample sample size calculation actually dramatically underestimated our true power. We had 80% power to detect a 1.8% absolute difference in mortality and 90% power to detect a 2.2% difference. 3. The increased risk of death was seen in the patients with high illness acuity who had a substantially longer exposure to study treatment than the average as shown in table 3 of the… Read more »
Sorry for lots of typos in my first comment. Critically, it should say 3% not 2%!
Paul, thank you for your thoughtful comments. I agree with your points and always learn a lot from your arguments. Overall, I think the real challenge here is one of interpretation. Mortality is obviously an incredibly important outcome, so even a trend towards mortality is worrisome. However, if we were to analyze treatments with no mortality effect, many would display a trend towards increased mortality due to the play of chance. If we take the viewpoint that any trend towards increased mortality represents an unacceptable risk, this could lead us to throw out many safe and useful therapies. Indeed, just… Read more »
So, here’s the thing. For comparisons of standard therapies that are cost-neutral I think we serve our patients best when we both act on uncertain evidence and continue to demand better data. I think that given the SUP-ICU findings, the recent BMJ meta-analysis, and the PEPTIC findings, we have seen a consistent and reproducible signal suggesting a small increase in mortality risk with PPIs. Based on the very large number of patients studied, I consider that the probability of a true effect on mortality is higher than you do, and I am particularly concerned about the potential for harm in… Read more »
In a nutshell, this study shows (as several others have done previously) that PPIs significantly reduce the incidence of upper gastrointestinal bleed among critically ill patients. There is a relatively weak signal that they may increase the 90-d mortality. However, as far as I can see, the purported mechanisms of harm from PPIs – C. difficile infection, and ventilator-associated conditions were no different between treatments. There is no plausible answer to what may have caused excessive mortality with PPIs. I am no PPI fan; however, in my humble opinion, this trial does not establish that the use of PPIs as… Read more »