The post-cardiac arrest learning gap
The technology is already here. The culture isn't.
I recently joined Eoin Walker on the Pre-Hospital Care Podcast to talk about what happens to a cardiac arrest case after the resuscitation ends. This is the short version of that conversation.
The starting position is one I’d defend fairly hard. The outcome for the patient in front of you is mostly decided in the moment. Bystander CPR, time to first defibrillation and who delivered it, compression quality, the post-ROSC bundle you put together before handover. That’s the case. What gets decided after the case is the outcome for the next hundred patients, and the next thousand. That’s the whole argument for taking post-event review seriously.
The guidelines aren’t the problem
At a guideline level this is done. ILCOR, ERC, AHA and ANZCOR all recommend or suggest some form of structured debriefing, and the most recent ILCOR EIT review found it associated with either no effect or improvements in ROSC, survival to discharge and neurological outcome.
So why is adoption still so variable? The barriers aren’t scientific. They’re operational and cultural:
- Time. Someone wants you to go green. Calls are stacking. There’s no protected time in most clinicians’ days to sit in the ambulance bay and debrief, and a cold debrief weeks later often means paying people overtime to come in on a day off.
- Tooling. The ePCR sits in one place, the monitor data in another, and some services are still running paper case records. Lining those up with the crew’s recollection is genuinely hard if a service hasn’t invested in bringing them together.
- Psychological safety. Plenty of services have historically equated “review” with “investigation”. We’ve moved a long way, but that memory is long.
- Facilitator skill. The word doing the work in all of those guidelines is structured. A good debrief is a learned skill, and as Andy Bell, Nick Overington and Megan Currie put it on Clinical Practice Radio recently, a hot debrief done badly can do harm. You need a framework, and most services haven’t resourced anyone to build one.
If I’m going to throw the fox into the hen house, it’s a leadership problem, top down. Most clinicians on road want to improve. The job of leadership is making sure they have the space to do it.
“Good data” is wider than the crew
When we talk about post-arrest data we tend to mean what the crew did on scene, but the 2024 Utstein update pushes us to look at the whole chain of survival, and a debrief should follow it. If nothing happened in the first ten minutes before we arrived, of course the outcome wasn’t great. Working forward from the call:
- Dispatch recognition. Some numbers put it around 75%, so roughly one in four arrests not recognised, and in some services recognised late in the call, which is its own lost opportunity. The biggest driver is agonal breathing going unidentified.
- The language of that call. Marine Riou and colleagues at the pre-hospital research unit at Curtin published “She’s sort of breathing” in 2018, looking at how callers describe gasping, snoring, and that hesitation before they answer. We’ve spent decades arguing about adrenaline and hand placement. Bringing linguistics researchers into dispatch science is one of the more interesting things happening in resuscitation.
- Telephone CPR and AED activation. Was dispatcher-assisted CPR delivered? Was a metronome used? Did someone get sent for an AED, and does the system even know where its AEDs are? There are still services in North America not doing telephone CPR at all.
- Community response. Bystander CPR roughly doubles 30-day survival in some registry data. Was an AED actually used? Was a community first responder activated? The UK’s AED network is arguably world-leading here and The Circuit do a cracking job of it, while some Australian states still don’t have a registry of community AEDs at all. Western Australia has something like 13,000 on its network.
- CPR quality. Rate, depth, full recoil, leaning. Sensing compressions through the pads alone gets you rate and not much else, so the rest needs a dedicated feedback sensor, and even where services have them some clinicians stay hesitant to use them. Chest compression fraction sits here too: some guidelines will accept 60%, I’d argue we aim as close to 100 as we can get and certainly above 80.
- The rest of the EMS response. Peri-shock pauses, because time off the chest costs coronary perfusion pressure and shock success. Rhythm changes, and whether the shock was timely and correct. End-tidal CO2. Drug timestamps, which in plenty of services still live on a notepad or somebody’s glove. Pad placement and transthoracic impedance, which is about the only way to get at placement retrospectively at all.
- After handover. ROSC, sustained ROSC, what it looked like in ED, cath lab activation, survival to discharge, CPC or mRS. For a lot of services this is the hardest data to get back, and without it you can’t know whether any of it is working.
Why crew recall isn’t enough
I don’t mean any offence by this, but crew recall is consistently unreliable. That’s not because we’re a bunch of liars or psychopaths. It’s because these are high acuity, low occurrence events. Outside critical care and high acuity response roles, most crews work maybe one or two viable arrests a year. Then put that person under cognitive load, with adrenaline up, in a noisy room, with family screaming and crying and the dog running around, and their perception of time goes. Paramedics overestimate compression depth, get rate wrong in both directions, and therefore overestimate compression fraction. One simulation study found only around 10% of compressions satisfactory when crews self-reported high confidence, and that’s a simulation, not even a real scene.
Real-time feedback closes some of that gap during the arrest. The data afterwards closes the rest.
Data without context is its own hazard
This is the caution I’d want people to take away. It is very easy to sit in a well-lit office with your AirPods in and your favourite music on, look at a rhythm strip and some CPR feedback data, see the gaps, and conclude the crew didn’t do a very good job.
But that gap might be there because the patient’s brother, who’d just taken a frantic phone call, barrelled into the room asking what was happening, and the crew’s attention went to their own safety for thirty seconds. That’s not the crew’s fault. It also won’t appear in the case sheet, because when I’m writing a case sheet I’m documenting the care I gave the patient, not the family dynamics in the hallway.
Which is why an automated report landing in someone’s inbox at the start of their next shift, unreviewed and unqualified, can feel like an attack even when nobody meant it that way. Adding the context costs someone’s time, and I don’t think that time is optional.
Oceans of data, a puddle’s worth used
We’re collecting oceans worth of data and using a puddle’s worth of it. The control room audits for AMPDS compliance, but does that flow anywhere else? Does hospital outcome data come back to us? Everyone has their own silo, their own privacy constraint, their own system that won’t interface with the next one, so someone ends up pulling a spreadsheet. Integration is the challenge now, not collection.
Close behind it: who is actually responsible for getting the learning back to the crew? A dedicated resuscitation improvement team, or a clinical leader with five thousand other things on their list already?
Pushing data forward, not just backwards
The same argument runs in the other direction, towards the hospital. Most EMS monitors can transmit something in real time now, whether that’s vital signs, the ECG, or full audio and video alongside live vitals. (I quoted a smaller figure on the episode, which I think I’d picked up from an in-hospital study. The paper below is the better source.) A 2024 systematic review and meta-analysis in CJC Open pooled 17 observational studies covering 4,306 patients and found pre-hospital digital ECG transmission associated with door-to-device times around 33 minutes shorter, first medical contact-to-device around 25 minutes shorter, and mortality of 8.9% against 13.7%. It’s observational data with considerable heterogeneity on the time outcomes, so hold the causal claim loosely. But the direction is consistent across almost every study in it, the mortality finding had no meaningful heterogeneity at all, and the intervention is close to free.
The technology isn’t the blocker. The blocker is that the ambulance service and the receiving hospital often sit in different trusts, different parts of a health department, or in North America different private providers entirely, and everyone now has a cybersecurity position on what’s allowed on their network.
There’s a secondary use here I think is underrated. Live vital signs plus a senior clinician on the other end is genuinely useful remote decision support for a junior workforce. We lost a lot of senior clinicians around COVID, and with a paramedic career lifespan sitting around six or seven years, there are crews out there without anyone senior to lean on.
Technology or culture?
Eoin asked whether quality improvement here is a technology problem or a culture problem. It’s a bit of a false dichotomy, but if forced to pick, culture wins. The technology is largely already here. Monitors record event data, most services have an ePCR, interoperability standards exist, and machine learning is perfectly capable of drafting a report for a senior clinician to review and contextualise before it goes out. Implementation science and a genuine no-blame culture are the parts we’re still working on. Technology takes the friction out of the learning loop. Culture decides whether the loop actually runs.
The full episode is on the Pre-Hospital Care Podcast, part of the Medic’s Academy Network: Spotify or Apple Podcasts.
That’s the high-level version. A few things here deserve their own post: what a genuinely useful debrief report actually contains, the dispatch end of the chain, and the case for a structured resuscitation data set that’s comparable between services and eventually between countries. I’ll link them back here as they go up.