Vascular access during cardiac arrest is one of those things that we probably don’t think about that much. There is usually quite a lot going on, somebody gets access, the drugs go in and we move on. In most systems the choice is between intravenous (IV) and intraosseous (IO) access, with IO having the obvious advantage that it can usually be obtained pretty quickly when the veins are difficult.
That has certainly influenced my practice over the years. I quite like IO access in resuscitation because it is quick, predictable and avoids repeated attempts at IV access becoming a distraction from everything else that is going on. There is also a logic that earlier vascular access should mean earlier drug administration, and we have data suggesting that earlier adrenaline administration is associated with improved outcomes. However, rapid access to the circulation is not necessarily the same thing as rapid delivery of a drug to the central circulation.
We’ve also had data suggesting that patients receiving drugs by the IO route have worse outcomes (e.g. PARAMEDIC 3). Those studies have always been difficult to interpret because the patients who get IO access are often different from those who get an IV. IO may be used because IV access has already failed, because the patient is technically difficult or because the resuscitation has gone on for longer. As Morley points out in the accompanying editorial, that leaves observational studies particularly vulnerable to confounding and to resuscitation time bias.
Fortunately, we now have randomised evidence from several large trials, most notably IVIO in Denmark and PARAMEDIC3 in the UK, and now an individual patient meta-analysis based on these studies. This new paper brings those data together using an individual participant data meta-analysis, which gives us another opportunity to look at the question.
Abstract
Objective
To perform a systematic review and individual participant data meta-analysis of randomized trials comparing intraosseous-first with intravenous-first vascular access during cardiac arrest.
Methods
The PRISMA-IPD guidelines were followed. We searched Medline, Embase, and the Cochrane Library on March 18, 2026, for individually randomized clinical trials. Pairs of investigators reviewed studies for relevance, extracted data, and assessed risk of bias. Individual participant data meta-analyses were conducted using logistic regression with adjustment for prognostic factors. Bayesian analyses were conducted using prespecified prior distributions. The certainty of evidence was evaluated using GRADE methodology.
Results
Two randomized trials provided individual participant data for 7561 patients with out-of-hospital cardiac arrest. There was no difference in 30-day survival (odds ratio, 1.01; 95% confidence interval, 0.81–1.25; moderate certainty of evidence) and favorable neurological outcome at 30 days or hospital discharge (odds ratio, 1.08; 95% confidence interval, 0.83–1.41; low certainty of evidence) between patients receiving intraosseous compared with intravenous vascular access. For sustained return of spontaneous circulation, the point estimate favored intravenous access (odds ratio, 0.90; 95% confidence interval, 0.81–1.01; low certainty of evidence), although the result was not statistically significant. Bayesian analyses found a high probability that any difference between the two strategies was small.
Conclusions
In this individual participant data meta-analysis of two randomized trials, intraosseous-first compared with intravenous-first vascular access during out-of-hospital cardiac arrest did not result in a statistically significant difference in 30-day survival, sustained return of spontaneous circulation, and favorable neurological outcome.
What did they actually do?
The investigators combined the individual patient data from the randomised trials comparing an IV-first with an IO-first strategy in adult out-of-hospital cardiac arrest. That’s an important distinction. These trials did not allocate patients to IV-only or IO-only resuscitation. They compared the route clinicians should attempt first. If that route failed, they could subsequently cross over to the alternative. So it’s pragmatic, and we like pragmatic approaches at St Emlyn’s as they are ‘real world’.
In practice that happened more commonly following attempted IV access, which is not particularly surprising. In the two trials included in the analysis, around 9–13% crossed from IV to IO compared with around 2–4% going from IO to IV. So the clinical question is therefore not whether IV or IO is the only route we should use rather it’s which one we should try first.
Why an individual patient data meta-analysis?
This is worth a little explanation as it’s one of the reasons this paper adds something to trials we’ve already seen. This isn’t simply a conventional meta-analysis where the authors take the published results from PARAMEDIC3 and IVIO and combine the headline numbers. The investigators obtained the original patient-level data from the trials and analysed them together using common definitions and statistical methods. So in effect we can analyse all the individual patients together, rather than simply combining the results of two separate trials. At first glance that might appear to be the same, but it isn’t.
One advantage is that we can analyse the data in a consistent way across both trials, including adjusting for the same baseline characteristics. That should give us a more precise estimate of the effect than either trial alone.
It also allows the investigators to look at whether the effect of IV or IO varies according to individual patient characteristics such as age, sex, initial rhythm, witnessed status, BMI or ambulance response time. That’s much more useful than trying to infer subgroup effects by comparing summary data between trials. IPD meta-analysis is often regarded as the gold standard approach to evidence synthesis for those reasons. We would like to see it more often, but it is considerably more work (Ed – I’ve been involved in a few meta analyses, but never an IPD one as they are definitely a step up in terms of workload).
We still need to remember that this analysis is based on the original trials, with all their limitations. IPD analysis can’t correct problems in the original data, and we still need to be cautious about subgroup findings.
Tell me about the patients
The analysis included 7,561 patients from the two large randomised trials, PARAMEDIC3 in the UK and IVIO in Denmark. These were adult out-of-hospital cardiac arrests where clinicians were attempting vascular access as part of advanced life support. This is therefore a population that should look reasonably familiar to those of us working in UK PHEM and ambulance services.
The original trials differed in some important ways, including their EMS systems and the preferred sites for IO insertion. In PARAMEDIC3, more than 73% of IOs were tibial, whereas IVIO also randomised patients to tibial or humeral IO access. Morley’s editorial points out that although IO placement was more successful than IV placement in IVIO (92% vs 80%), displacement and incorrect positioning were more common with humeral than tibial IOs. So IO is not a single entity, perhaps site matters.
What about the outcomes?
The main outcome was survival at 30 days. They also looked at favourable neurological outcome (always St Emlyn’s outcome of choice) and sustained return of spontaneous circulation (ROSC). The authors then used both conventional frequentist and Bayesian approaches to analyse the results.
I quite like the Bayesian part of this paper. We’ve discussed before on St Emlyn’s how unhelpful it can be to reduce a trial to whether the p-value happens to fall either side of 0.05. Here the Bayesian analysis allows us to ask a slightly more useful question: given all the data we now have, how likely is it that one strategy has a clinically important advantage over the other?
What did they find?
The headline figure was that thirty-day survival was almost identical. Survival was 5.9% with an IO-first strategy and 6.1% with an IV-first strategy, with an adjusted odds ratio of 1.01 (95% CI 0.81–1.25). There was also no convincing difference in favourable neurological outcome. So you might argue that you can stop reading now, and do what you like, but as usual there are nuggets of info in the detail.
ROSC was lower in the IO-first group, with an odds ratio of 0.90 (95% CI 0.81–1.01). That just crosses the conventional threshold for statistical significance, so we need to be careful about what we say, but it is consistent with the direction of effect seen in PARAMEDIC3. Morley makes the same point in the editorial. The randomised trials have not shown a survival advantage from routine IO-first access, and there does appear to be a signal towards lower rates of sustained ROSC with IO.
What does the Bayesian analysis tell us?
The Bayesian analysis gives us a little more confidence around what the absence of a survival difference means. With more than 7,500 randomised patients now available, the probability of there being a substantial survival advantage from either strategy is becoming quite small. The Bayesian analysis suggests that any difference in survival is likely to be small. We can’t exclude a small effect, but a large difference between the strategies now looks pretty unlikely.
There could still be a small difference between the strategies, and there may be particular patients where one route is preferable, but another trial of a few hundred patients is unlikely to suddenly demonstrate a large survival effect that has been hiding from us.
What about the subgroup analyses?
As always we are super cautious with these and consider them hypothesis generating rather than defninitive, but there are some potentially interesting findings here, including apparent interactions with sex and ambulance response time.
The response-time interaction disappeared when the investigators excluded a relatively small number of patients with extreme response times. There was no convincing evidence that treatment effect differed according to age, witnessed status, bystander CPR, initial rhythm or BMI.
There is always a temptation following a broadly neutral trial to search through subgroups for the patients in whom the intervention “really works”. Occasionally that identifies something important, but it is also a very good way of finding associations by chance.
Is an IO really the same as an IV?
The IO gives us rapid vascular access, but the reason we want access is to get drugs into the central circulation. During cardiac arrest the circulation is clearly not normal. Drugs administered into the bone marrow still need to reach the central circulation during low-flow CPR, and there are plausible reasons why that might be different from injecting directly into a vein.
That could potentially explain the signal towards lower ROSC with IO despite no detectable difference in longer-term survival but it could also be completely irrelevant. The trials weren’t designed to answer the pharmacokinetic question, so we should be careful not to invent a mechanism simply because it fits the results.
Also worth mentioning that there are those differences between where you put the IO and also the IV. I’m personally a fan of the external jugular in cardiac arrest for IV access which I think gets drugs to the heart faster, and in the first part of my career I always put subclavians central lines in at arrests (I did a lot of those, which has turned an advantage in my later career as we now use them a lot in major trauma patients). Anecdotally it certainly felt that a more central location for IV also had a faster and more profound effect.
What about speed?
The major practical argument for IO has always been speed. That only matters if it actually gets the drugs in earlier. In IVIO, median time from emergency call to drug administration was about 15 minutes and was essentially the same between the two strategies. In PARAMEDIC3 it was around 24 minutes and again similar between groups.
If an IO doesn’t result in earlier drug administration, then one of its main theoretical advantages has disappeared. This is also where local practice matters. A team that can reliably get an IV in within seconds is in a different position from one where repeated attempts are common. Similarly, a service that uses IO frequently and trains with it regularly may perform differently from one where it is an unusual procedure.
So should we stop using IO?
No, and I don’t think that’s what these data tell us. IO access appears to be remarkably safe. Morley cites a Danish study of 5,387 patients receiving IO access in which long-term complications were exceptionally uncommon, with no reported osteonecrosis, osteomyelitis or compartment syndrome. And we have a lovely BestBet on this too (COI – – co-written by Emma Carley).
It is also very useful when IV access is difficult or impossible. The more interesting question is whether IO should be the routine first choice simply because it can be technically easier and quicker to insert and these trials don’t provide much support for that approach. The choice probably depends upon local IV and IO success rates, training, device availability, displacement rates and, importantly, the effect of each technique on time to drug administration. Clinicians should anticipate the need for IO when IV access is likely to be difficult rather than waiting through repeated unsuccessful IV attempts.
That makes sense to me.
What does this paper add?
With more than 7,500 randomised patients now available, there is little evidence that routinely choosing IO first improves survival in adult OHCA. There may be a signal towards better ROSC with IV, but that doesn’t translate into a demonstrated longer-term survival advantage.
Previous observational studies suggested that patients receiving IO drugs did worse. The randomised trials suggest that much of that apparent difference may have been related to the patients and circumstances in which IO was used, rather than the route itself.
The bottom line
If IV access looks straightforward and can be achieved quickly do that first. If the veins are poor, an IV attempt fails, or vascular access is beginning to consume time and cognitive bandwidth during the resuscitation, then IO remains an excellent alternative.
vb
S
References
- Holmberg MJ, Couper K, Andersen LW, Lall R, Granfeldt A, Ji C, Perkins GD, Vallentin MF. Intravenous or intraosseous vascular access in cardiac arrest: an individual participant data meta-analysis. Resuscitation. 2026 Sep 11;228:111304. doi: 10.1016/j.resuscitation.2026.111304. Epub ahead of print. PMID: 42727699.
- Morley PT. IO or not IO, is that the question? Resuscitation. 2025;207:110521. doi:10.1016/j.resuscitation.2025.110521.
- Couper K, Ji C, Deakin CD, et al. A randomized trial of drug route in out-of-hospital cardiac arrest. N Engl J Med. 2024. 1-s2.0-S0300957225000334-main.pdfPDF
- Vallentin MF, Granfeldt A, Klitgaard TL, et al. Intraosseous or intravenous vascular access for out-of-hospital cardiac arrest. N Engl J Med. 2024.
- Nolan JP, Deakin CD, Ji C, et al. Intraosseous versus intravenous administration of adrenaline in patients with out-of-hospital cardiac arrest: a secondary analysis of the PARAMEDIC2 placebo-controlled trial. Intensive Care Med. 2020;46:954–962.
- Carley ELJ, Yates GP. In children requiring emergency vascular access, are intraosseous lines associated with significant long-term complications? Emerg Med J. 2026 Aug 11:emermed-2026-216214. doi: 10.1136/emermed-2026-216214. Epub ahead of print. PMID: 42580875.
