SUP-ICU is a massive, modern RCT of stress ulcer prophylaxis (SUP) in the ICU.1 With 3298 patients, it is larger than many meta-analyses of SUP.2 3 Its primary mortality endpoint is deeply flawed. However, the study still provides a wealth of information about SUP in the ICU.
Why the primary mortality endpoint was inevitably going to be negative
Big picture: Proving mortality benefit is generally impossible
Fewer than a dozen medications have ever been proven to improve mortality in critical care. There are numerous reasons that it’s nearly impossible to prove mortality benefit, including:
- Mortality rates are falling overall.
- Any signal of cause-specific mortality is easily lost within the noise of all-cause mortality.
- Physicians struggling to keep patients alive will often compensate for the lack of intervention in the control group, by instituting other resuscitative therapies.
Stress ulceration in the ICU doesn’t affect mortality
The SUP-ICU trial based their power calculations upon Krag et al 2015, a large modern descriptive study of gastrointestinal bleeding among ICU patients.4 This study found that stress ulceration wasn’t an independent risk factor for mortality. This implies that GI bleeding doesn’t cause mortality in the ICU.
It makes sense that stress ulceration shouldn’t increase mortality. Stress ulceration is a very treatable entity with modern therapies (e.g. endoscopy, interventional radiology, acid-suppressive medication, blood transfusion). A patient inside the ICU who developed stress ulceration would likely be treated successfully, rather than exsanguinate to death. Numerous other studies have likewise found that stress ulceration doesn’t affect mortality.5 6 7 8 9
If stress ulceration doesn’t cause mortality, that would make it impossible for SUP to affect mortality. Which would make this entire study futile from its inception.
Even if stress ulceration did cause mortality, the difference is too small to prove
Let’s imagine that the following were true:
- 50% of people with stress ulceration bleed to death
- Stress ulcer prophylaxis reduces the rate of stress ulceration by 50% (from 5% to 2.5%)
- The baseline ICU mortality is 25%
In this scenario, the risk of dying from a stress ulceration is:
- without prophylaxis: (50%)(5%) = 2.5%
- with prophylaxis: (50%)(2.5%) = 1.25%
The all-cause mortality would then be:
- without prophylaxis = 25% + 2.5% = 27.5%
- with prophylaxis = 25% + 1.25% = 26.25%
How many patients would be needed to prove this mortality difference? If we run a rough power calculation (with alpha of 0.05 and beta of 0.8) this yields a sample size of 19,700 patients. So even if we imagine that stress ulceration causes a 50% mortality (insanely large), this mortality difference would still be too small to prove without an impossibly large study.
So where did SUP-ICU get its power calculations from?
SUP-ICU was arbitrarily powered to detect a 5% difference in all-cause mortality. This is an unrealistically huge mortality benefit because of the following math. The predicted mortality impact of SUP can be estimated as follows:
(% mortality benefit) = (% baseline frequency of stress ulceration) (% mortality caused by an episode of ulceration) (% fractional reduction in bleeding episodes caused by SUP)
My best guess about a mortality benefit would be as follows:
- Baseline frequency of clinically significant stress ulceration = 4%10
- Attributable mortality from an episode of ulceration = 10% (my guess is that bleeding does increase mortality marginally, but to an extent which is too small to detect in any clinical study)
- Reduction in bleeding episodes from prophylaxis = 50%
Using these assumptions:
% mortality benefit = (4%)(10%)(50%) = 0.2%
Take a minute and play with different numbers, based on your estimation of these values. It’s impossible to get to a 5% absolute mortality difference without making unrealistic assumptions (e.g. very high baseline stress ulceration rate and very high attributable mortality).
To place this in a larger perspective, a 5% impact on all-cause mortality is a huge mortality difference, especially for a mixed population of patients (most of whom would never develop stress ulceration). In comparison, within a population of purely STEMI patients, cardiac catheterization causes a mere 2% improvement in mortality.11 Other authors have previously noted that a 5% mortality benefit from SUP is unrealistic.12
Re-evaluation of the study from the standpoint of other endpoints
The primary mortality endpoint was doomed to be negative. However, this doesn’t mean that we should ignore the entire trial. This remains a massive and well-executed RCT, so the secondary endpoints may be informative. These are shown below:
The legitimacy of using secondary endpoints is controversial. I think this is reasonable when the primary endpoint is severely flawed, as previously discussed here. It’s ideal to focus on a single secondary endpoint which arguably should have been the primary endpoint – which in this case would be the rate of clinically important GI bleeding. Notably, stress ulceration is the primary endpoint of a similar study which is ongoing (REVISE trial). SUP causes a significant reduction in GI bleeding, with a fragility index of seven (which is surprisingly robust). So, it’s reasonable to conclude that pantoprazole decreases the risk of stress ulceration.
Overall, this data indicates that SUP prevented stress ulceration without any signal of harm. That is exactly what SUP is supposed to do. It’s not going to save your life or wax your car, it’s going to prevent stress ulceration. Prophylaxis wasn’t 100% effective, but prophylaxis for most conditions isn’t perfect (e.g. DVT prophylaxis is only ~50% effective).
Preferred agent for SUP?
Proton pump inhibitors more effectively suppress gastric pH than H2-receptor blockers, making them more effective at stress ulcer prevention.7 13 However, proton pump inhibitors have previously been avoided due to fears that neutralization of gastric pH could increase the risk of clostridium difficile and pneumonia.14 15
Secondary endpoints in SUP-ICU allay these fears. There were actually fewer cases of clostridium difficile among patients treated with pantoprazole. The rates of pneumonia were identical across both groups. Given how well the study was powered, this makes it unlikely that pantoprazole causes a significant increase in either complication.
Removal of concerns about pneumonia or clostridium difficile would favor moving pantoprazole to a front-line agent for SUP, based on its higher efficacy. There is also some suggestion that pantoprazole may cause less delirium than H2-blockers, supporting pantoprazole as a front-line agent.16 17
Number needed to treat?
A 1.7% difference in GI bleeding equates to a number needed to treat (NNT) of 59. That’s not impressive. However, it must be borne in mind that the patient population wasn’t very high-risk for GI bleeding (e.g. only 78% were intubated). Inclusion in the study required only a single risk factor for stress ulceration, from among the following: shock, anticoagulation, renal-replacement therapy, intubation, any history of liver disease, any history of coagulopathy, or ongoing coagulopathy.
More work needs to be done to define the appropriate population to receive SUP. Application of SUP to a higher-risk group would improve the benefit of therapy (with a commensurate decrease in the NNT). For example, it might be reasonable to restrict SUP to intubated patients with either coagulopathy or hypotension.18
DVT prophylaxis is widely accepted, whereas SUP remains controversial. This leads to some interesting comparisons.
The most important point here is that DVT prophylaxis doesn’t improve mortality.19 20 21 This illustrates that an intervention doesn’t need to decrease mortality in order to be extremely useful.
The relative effectiveness of DVT prophylaxis is roughly a ~50% reduction in events. Once again, this runs parallel to the results obtained from SUP-ICU.
The main difference between DVT prophylaxis and SUP is the baseline event rate. In the absence of any prophylaxis, the risk of DVT is much higher than the risk of stress ulceration (perhaps roughly ~30% vs ~4%). This means that DVT prophylaxis offers patients a much larger magnitude of benefit, with a lower number needed to treat (NNT).
This comparison overall suggests that SUP likely does have value. However, given the lower incidence of stress ulceration, it might be sensible to restrict SUP to patients at the highest risk of stress ulceration. Further work is needed to sort out exactly who these patients are in the context of modern critical care practices (e.g. early enteral nutrition).22
- Stress ulceration doesn’t significantly affect mortality. Therefore, there is no way that preventing stress ulceration could possibly improve mortality. The fact that SUP-ICU detected no mortality benefit was inevitable and doesn’t provide any new information about SUP. DVT prophylaxis doesn’t affect mortality either.
- Pantoprazole did its job in this trial: it reduced the incidence of stress ulceration (by 50%) without any detectable toxicity (no increase in clostridium difficile or pneumonia rates). Lack of toxicity may favor use of pantoprazole as a front-line agent for SUP.
- Pantoprazole reduced the incidence of clinically important GI bleeding by 1.7%, yielding a number needed to treat (NNT) of 59. This suggests that SUP should be restricted to a more highly selected population of patients, who are at greater risk of stress ulceration.
- Further work is needed to define which patients may benefit from SUP. There is probably a role for it, but we’re using it too often.
- Chasing mortality endpoints is a fool's errand
- Secondary endpoints: Can we separate the wheat from the chaff?