After years of waiting, PARAMEDIC 2, the large RCT of epinephrine for out of hospital cardiac arrest, has finally been published. So we can now definitively say that epinephrine is harmful. Wait, maybe it helps? Can a positive study demonstrate harm? Maybe this EBM stuff isn’t so easy after all…
The paper
PARAMEDIC 2: Perkins GD, Ji C, Deakin CD, et al. A Randomized Trial of Epinephrine in Out-of-Hospital Cardiac Arrest. The New England journal of medicine. 2018. PMID: 30021076 [full text]
The Methods
PARAMEDIC 2 is a multicenter, randomized, double-blind, placebo controlled trial
Patients: They included all adult patients with out of hospital cardiac arrest for which advanced life support was provided by trial-trained paramedics.
- Excluded: Patients younger than 16 years old, known or suspected pregnancy, cardiac arrest from anaphylaxis or asthma, or administration of epinephrine before the arrival of the trial trained paramedics.
Intervention: Epinephrine 1 mg IV push every 3-5 minutes.
Comparison: Placebo (saline).
Outcome: The primary outcome was 30 day survival.
- Secondary outcomes included rate of survival until hospital admission, the lengths of stay in the hospital and ICU, the rates of survival at hospital discharge and at 3 months, and the neurologic outcomes at hospital discharge and at 3 months.
The Results
They enrolled 8014 patients. (10,623 were screened, with the majority of exclusions occuring because epinephrine had been given before trial paramedics arrived, or because the patient achieved ROSC before randomization.)
For the primary outcome of 30 day survival, the epinephrine group had a small but statistically significant increase in survival: 3.2% vs 2.4% (OR 1.39; 95% CI 1.06-1.82, p=0.018). The fragility index for this result is 6.
There was no difference in neurologically favourable survival (modified Rankin score 3 or less) at hospital discharge, or at 3 months: 2.1% vs 1.6% (OR 1.31; 95% CI 0.94 – 1.82).
A lot more patients had return of spontaneous circulation in the field with epinephrine (36.3% vs 11.7%). More patients were also admitted to hospital with epinephrine (23.8% vs 8.0%).
My thoughts on PARAMEDIC 2
I have to admit, I was excited for the release of this trial. There has never been convincing evidence that medications help cardiac arrest patients. Observational data has repeatedly shown increased rates of ROSC with epinephrine, but no benefit in terms of neurologically intact survival. (Loomba 2015) The classic OPALS trial demonstrated no improvement in survival with the introduction of advanced life support paramedics. (Stiell 2004) And there was the fantastic prior RCT on this topic which demonstrated no benefit, but unfortunately had enrollment issues and ended up under-powered. (Jacobs 2011) I have always worried that epinephrine might be hurting my patients. I was excited for this trial to be published, so we would finally have a definitive answer. That was my mistake. I should have realized that there is no such thing as a perfect trial. There are always caveats. We don’t get definitive answers.
The big caveat that will drive most opposing views on this trial is the presence of two different but important outcomes: survival, and survival with good neurologic function. That is the crux of the issue, but before we dive in, let’s tackle a few smaller issues.
First, the survival rate was pretty abysmal here. Only 3.2% of the epinephrine group was alive at 30 days. Many communities have much higher survival rates than that (although I think some also have lower rates of survival). In cardiac arrest, we have two groups of patients: dead and only mostly dead. Some patients will not recover no matter what we do. If these patients are over-represented in a trial, they could hide a true benefit in the small subset of patients who actually have a chance of surviving. (EDIT: 615 patients were excluded because of ROSC before the trial paramedics arrived. Therefore, selection bias probably contributes to the low survival seen.)
A related issue is the length of time it took to deliver epinephrine. We know that patients who are likely to achieve ROSC are likely to have it happen early in their resuscitation. Early defibrillation is clearly helpful. If it is going to work, epinephrine probably also needs to be given early. Although response times were pretty good in this trial (6 minutes), the time until the first dose of study drug was very long (about 22 minutes). By 22 minutes, it might simply be too late. (This is not a comment at all on the paramedics here. Working in the prehospital environment is incredibly difficult, and a large number of factors can contribute to delay).
The dose of epinephrine will also likely be discussed. Is 1mg really the best dose? I have always found it weird that I can give a 1 mg IV push of epinephrine to a patient without a pulse, but if I wait just 30 seconds and the patient regains a pulse, that same dose would be considered negligent. Scott Weingart has spoken at length about titrating epinephrine based on arterial line blood pressures, but most of us are still treating empirically. Although people will want to discuss dose, that isn’t what this trial was about.
I could discuss a number of issues with the modified Rankin score. A cutoff of 2 instead of 3 might be a better representation of the kind of outcomes we would all want. There are a number of ways this data could be dissected or presented, but I have previously discussed how subjective or inaccurate this score can be. But the thing that bothers me the most might seem like a really small point: the use a logarithmic scale when displaying these results, which completely distorts the outcomes. This chart will be widely spread, but it is very misleading.
But let’s get back to the real issue with this trial: what outcome should we care about? The primary outcome of this trial was 30 day mortality, and epinephrine was better. Sounds pretty simple. We should use epinephrine. Unfortunately, increasing quantity of life does not necessarily increase quality of life, and our patients actually care more about quality than quantity. Therefore, the fact that epinephrine failed to improve survival with good neurologic outcome seems to make this a negative trial.
I would argue that any intervention that increases survival, but doesn’t increase survival with good neurologic outcomes, is clearly harmful. That combination means that patients “saved” are left with poor functional outcomes and a quality of life that might not be appreciated.
And there are further harms. Epinephrine increased ROSC. It also increased admissions to hospital. At face value, those sound like good outcomes, but when you consider the fact that these patients were not going to walk out of the hospital, I think this is clearly harmful. Leaving aside the added costs, I don’t know any person who wants to spend the final days of their life in an ICU. We are willing to tolerate ICU care if there is hope. If there is a chance at survival. But epinephrine seems to fill our ICUs, and subject our patients to aggressive care, without any real chance that the patients will go home.
Optimists will often argue with me on this point. “Let’s just get these patients into the ICU”, they say, “and then we can focus on other interventions like TTM or ECMO to improve their neurologic outcomes”. It is true that we will never be able to improve outcomes if we give up on our patients too early. I don’t want to be overly pessimistic. But for now, none of those downstream interventions have shown any benefit. I think it is a mistake to allow our optimism for the future to translate into harm for our current patients.
My comments thus far have all assumed that epinephrine does not improve survival with good neurologic outcomes. If that is the case, I would argue very strongly that epinephrine is harmful and should be abandoned. Unfortunately, if we take a close look at the data, things get a little more complicated.
In the epinephrine group, 2.2% of patients survived to hospital discharge with a favourable neurologic outcome. In the placebo group, it was 1.9%. Although this was not a statistically significant difference, the point estimate is still higher in the epinephrine group. At 3 months, the results are similar (2.1% vs 1.6%). What stands out to me are the odds ratios. The odds ratio for 30 day survival was 1.39. The odds ratio for a good neurologic outcome at 3 months was 1.31. Those numbers are very similar. The absolute numbers are tiny, but that might be primarily driven by the tiny number of survivors over all. If those same odds ratios held true in a population with a 10% baseline survival rate, the absolutely differences might actually be clinically important.
Ultimately, the key question here is one of values, not of statistics or methodology. It seems pretty clear that epinephrine will result in more return of spontaneous circulation, and more admissions to hospital. It probably increases overall survival, but that survival might come with a significantly decreased quality of life.
If we take the numbers here at face value, and forget statistics for a minute, for every 1000 out of hospital cardiac arrests, the use of epinephrine will result in 246 extra cases of ROSC, 158 extra admissions to hospital, and 8 extra survivors at 30 days. Of those 8 extra survivors, 3 would have a good neurologic outcome and 5 would have a bad outcome.
Those are the numbers, but I am not sure what they mean. Are the 3 good neurologic outcomes worth the 5 bad outcomes? Are they worth the 158 extra ICU admissions? Does the benefit outweigh the harm?
I give a lot of credit to these authors. They provide us with one of the best discussion sections I have ever read. They talked to patients and the public before starting this trial to determine what outcomes were important. They note that a number needed to treat of 112 is below what is generally considered to be the minimal clinically important outcome. They tell us that patients care more about neurologic outcomes than survival alone. They note that “patients may be less willing to accept burdensome treatments if the chances of recovery are small or the risk of survival with an impaired neurologic outcome is high.”
These are the important questions. In the coming days and months, physicians are going to spend a lot of time debating the implications of this trial. No amount of debate is going to provide us with clarity. These are not questions that should be left in the hands of clinicians. These are questions that we need to address as a society. What counts as a good outcome? How small a difference are we willing to chase? And what costs are we willing to pay? Where does organ donation fit into this conversation? We don’t need another RCT – we need research that involves the public and endeavours to answer these very difficult questions.
Bottom line
I think people will use PARAMEDIC 2 to support their existing belief, whatever that is. This data demonstrates that epinephrine has, at best, a very minimal impact on patient oriented outcomes. It is not clear that the benefit outweighs the harm.
UPDATE: Since this post on PARAMEDIC 2, I have completed a more thorough review of the literature on epinephrine for out of hospital cardiac arrest, which can be found here.
Other FOAMed
JC: Does Epinephrine work in Cardiac Arrest on St. Emlyn’s
EM Nerd: The Case of the Costly Compound
REBEL Cast Ep56 – PARAMEDIC 2: Time to Abandon Epinephrine in OHCA?
Great conversation with one of the PARAMEDIC 2 authors on The Resus Room
References
Jacobs IG, Finn JC, Jelinek GA, Oxer HF, Thompson PL. Effect of adrenaline on survival in out-of-hospital cardiac arrest: A randomised double-blind placebo-controlled trial. Resuscitation. 2011; 82(9):1138-43. [pubmed]
Loomba RS, Nijhawan K, Aggarwal S, Arora RR. Increased return of spontaneous circulation at the expense of neurologic outcomes: Is prehospital epinephrine for out-of-hospital cardiac arrest really worth it? Journal of critical care. 2015; 30(6):1376-81. [pubmed]
Stiell IG, Wells GA, Field B, et al. Advanced cardiac life support in out-of-hospital cardiac arrest. The New England journal of medicine. 2004; 351(7):647-56. [pubmed]
Morgenstern, J. Paramedic 2: Epinephrine harms/helps in out of hospital cardiac arrest, First10EM,
July 20, 2018. Available at:
https://doi.org/10.51684/FIRS.6140
22 thoughts on “Paramedic 2: Epinephrine harms/helps in out of hospital cardiac arrest”
It isn’t selection bias to not enroll patients into a study who had ROSC prior to the administration of the drug or the placebo. It was part of the exclusion criteria. Anyone who achieved ROSC prior wasn’t included in their survival data; the 3.1% is not indicative of overall OHCA survival during the study period.
“Although response times were pretty good in this trial (6 minutes), the time until the first dose of study drug was very long (about 22 minutes)” This is largely because in the UK, drugs aren’t given until after the 3rd shock (for shockable rhythms), so you will have a built in delay. I’m not aware of any evidence to support the idea that epi given earlier produces better results though; maybe that’s something that needs to be studied but it isn’t a known.
“The primary outcome of this trial was 30 day mortality, and epinephrine was better.” Barley, and in it was practically an insignificant number, meaning it reached statistical significance, but really not practical significance. Better, yes, but…
“In the epinephrine group, 2.2% of patients survived to hospital discharge with a favourable neurologic outcome. In the placebo group, it was 1.9%. Although this was not a statistically significant difference, the point estimate is still higher in the epinephrine group. At 3 months, the results are similar (2.1% vs 1.6%). What stands out to me are the odds ratios. The odds ratio for 30 day survival was 1.39. The odds ratio for a good neurologic outcome at 3 months was 1.31. Those numbers are very similar. The absolute numbers are tiny, but that might be primarily driven by the tiny number of survivors over all.” But those were the survival numbers, so the data IS meaningful. I’m not sure attempting to parse this out in this manner is really helpful. Making assumptions outside of the dataset isn’t normally a great idea, no?
I do agree 100% with your bottom line.
Thanks for the comment Scott. Lot’s of good points to address
1) I do think that the exclusion of ROSC is selection bias. Selection biases are often embedded into the inclusion/exclusion criteria, because a selection bias is anything that makes the study population different from the patients you might be trying to apply the research to. The context of that comment is trying to explain the low survival in this cohort, which is well explained by the exclusion of patients with early ROSC. That being said, it might not matter, because you would never give epi to those patients with early ROSC.
2) There is observational data that suggest early epinephrine is better than late, although only in terms of mortality and nor good neurologic outcomes. The timing needs to be considered, because giving medications to a dead patient clearly is not going to work. (If the epi had been given at 1 hour here, we would all agree that is too late. What we don’t know is what is early enough to give the patient a chance.)
3) I agree that a 8 in 1000 increase in survival seems clinically insignificant. Lots of people disagree. (See twitter). That is why I think this discussion needs to be taken out of the hands of clinicians and into the hands of all our potential patients.
4) I am not sure I understand your last question. We always extrapolate study results to other populations. That is the basis of science. In general, relative risks seem to stay pretty steady across populations, but with changing baseline risk, the absolute numbers change. (This is why there is so much debate about things like statins in low risk populations – in relative number they still look great, but the risk is so low, there is not absolute benefit.) Therefore, when extrapolating these numbers, you might expect to see a bigger absolute difference in a trial with a higher survival rate. However, you do have to look at the confidence intervals, which here include harm as well as benefit.
Thank you sir.
I think we just disagree about the selection bias portion; selection bias means that because of the selection criteria, proper randomization doesn’t occur (or proper recruitment). They weren’t looking at total survival, just those who got enrolled in the epi/placebo arms. They weren’t looking at total / overall OCHA survival, so I just to see this as a bias. If they were looking for total survival and survival to discharge and excluded that population, than I would see the bias.
I see where you were coming from in #2, thank you for the clarification. #3 is going to be a long-term debate I think, I was just looking at it from a strictly data point-of-view. I think we agree on the last point as well. I didn’t understand the point you were making (obviously) originally, but your explanation is clear.
Again, thank you for your responses.
I had this problem when writing my glossary of biases. https://first10em.com/bias/
There is no agreed on definition for any of the biases. There are multiple subtypes of selection bias, and you might only be describing one of them. Finding bias in a trial has nothing to do with criticizing the researchers or trial design. It is more about identifying methodological issues that will impact the way you interpret or apply study results. In that context, I think this is a bias. The way that patients were selected significantly impacts my application of the results; it biases the results so that they cannot be applied to all comers. I know the authors did this on purpose. I think this was the right way to design the trial. But it still needs to be discussed in order to properly interpret and apply the results.
But now we are probably just getting overly pedantic.
I think you’re suffering from secondary outcome syndrome. As the bar graph clearly shows, the outcome that matters most to patients is survival with good neurological outcome. This excellent trial gave adrenaline ample opportunity to demonstrate that it could favourably influence this outcome and it failed. Now I know that overall survival to 30 days (which was improved by adrenaline) was the nominal primary outcome but it clearly shouldn’t have been. If someone wants to push the proposition that adrenaline should still be used because of this outcome (or any other one) then they need to acknowledge that they are not addressing what patients really value.
I agree, except that the trial wasn’t powered for survival with good neurological outcome. It would have had to have been much larger, and it if was it may have shown a statistically significant outcome. Rather than focus on that, the real debate is whether than difference would be clinically significant. There are still people out there who would be excited about a 0.3% increase in neurologically intact survival. The key is to focus on the harms and have a societal discussion about what harms and costs we are willing to accept for significantly diminishing returns.
I agree with Scott. Patients were not randomised to either arm until the third shock (if in shockable rythm). So, survival rate is looking at those sicker patients who did not respond to CPR and Defib, both of which have a much lower NNT than adrenaline and a stronger evidence base linked to survival.
Interesting analysis. But, more important, at light of this RCT, what will you do from now on?
Ahhh.. you noticed I left that part out? I honestly don’t know. I was hesitant to use epinephrine before this trial came out, and this makes me even more hesitant. However, most of ACLS is so protocolized, that I actually don’t make this decision – it just happens as part of a well functioning team resuscitation. I am interesting in hearing what other smart people are doing. Scott Weingart’s titration based on diastolic pressure makes some sense. Another option might be 1 or 2 early bolus doses, based on the assumption that earlier ROSC is good, but no extra doses, as higher doses could be presumed to be causing the harm seen here.
The way you adressed this issue to society opinion is interesting, but the only way to do it correctly with current data would be – NOT the majority opinion, but induvidual opinion (something like “do not resuscitate documentations”), which makes it more ethical and complicated in the end, even as far, as i am writing it, it gets more and more complicated than i thought from originally 🙂
6 minute response time and 22 minutes to give first dose? What were they doing for 16 minutes.
Even if it was only 2 people (which I doubt…they likely had an engine), they should have the first epi on-board in under 10 minutes (but more realistically under 5). Do they not have IO access?
Do not underestimate how much this affects the study outcome. This study is about whether epinephrine helps very-dead people survive. In many areas, they discontinue resuscitation efforts after 20 minutes. They were giving epi right when reaching that point.
Is there something I’m misreading here?
In the UK there is no “engine”. An ambulance response is a Paramedic and an EMT on an ambulance, not a fire engine. So the answer to your question is there are indeed only two people but trying to do the work of 4.
Its not that you are misreading something rather you are assuming something.
The default should be no code. If the patient doesn’t have a Do Resuscitate bracelet on we should do nothing. I am not joking. We have this backwards. I would much rather be dead than living strapped to a wheelchair unable to follow soap opera plots. It is interesting that the only time you can deliberately touch another person without their permission is when you code them.