In this study, we retrospectively applied five alternative ECPR selection criteria to an ECPR dataset from three metropolitan ECPR centres in Australia. We found that applying different selection criteria had a large impact on the population included and survival rates. We showed that more restrictive criteria could potentially result in (1) a reduction of the total number of ECPR cases performed (2) a higher survival rate in those cases that are performed but (3) could potentially also increase the number of missed survivors that did not fulfil all criteria but may have still survived. The reverse is true for more liberal criteria. We also found that a greater proportion of the IHCA ECPR population would have been excluded from ECPR compared to OHCA if strictly adhering to the current selection criteria.

The survival rates in this study, when applying the five various selection criteria, demonstrated similar trends to the original trials they were based upon. The major limitation being that we were not able to review patient level data for comparison instead we compared by selection criteria only. The dataset is consistent but it represents ‘real-life’ data and is not prospective. The PRAGUE trial selection criteria included patients most liberally in our cohort. We found applying these criteria resulted in the greatest number of survivors, but also more non-survivors, with overall the lowest survival rate. For OHCA the survival rates in our analysis were similar to the original study. It showed 39/124 (31.5%) neurologically intact survival (CPC1 or 2) at 180 days, as compared to 16/73 (21.9%) survival to discharge for all patients fulfilling the Prague criteria in this cohort. The most restrictive criteria were the CHEER criteria, with age < 65 limit, arrest time limit to < 45 min and being limited to VF alone. This resulted in a survival rate of 7/16 (43.8%), which was similarly high compared to the original study 3/9 (33.3%).

The INCEPTION criteria included restrictive variables (witnessed by bystander, bystander CPR and an initial shockable rhythm). However, despite early transport and an intention to exclude patients in the field with an anticipated start to cannulation > 60 min, the actual delay to start of cannulation hindered a restrictive time criterion and a large number of patients were cannulated > 60 min post arrest. 14/70 (20%) survived in the intention-to-treat ECPR arm, however only 46 of these patients received ECPR with a survival of 5/46 (10.9%) compared to 15/55 (27.3%) applying the INCEPTION criteria to this dataset. The median time to ECMO support was very similar at 74 min (vs. 77 min in our cohort) suggesting that system factors such as first responders, pre-hospital care and experience in simultaneous advanced life support and ECPR cannulation play a vital role for outcomes.

Randomised data demonstrating survival benefit for ECPR in IHCA is lacking. However, survival rates for in-hospital populations are described in the range of 20–40% [15, 16], substantially higher than the comparable ECPR survival form registry data for OHCA 8.4–16.4% [7,8,9].

Overall, survival following IHCA ECPR was two times that for OHCA in our cohort. By strictly applying the selection criteria, even higher survival post ECPR for IHCA was achieved. Restrictive criteria (like the CHEER criteria) applied to our data resulted in 10/15 (66.7%) survival versus 9/15 (60%) in the original study at the cost of excluding a group of patients from ECMO that may still have benefited from ECPR support.

Despite the fact that in-hospital cardiac arrest patients were on average 4 years older and had more comorbidities and a lower rate of shockable rhythms, the survival outcomes were substantially better than for OHCA (see Table 2). We speculate that the shorter time to establish ECPR, a well described prognostic variable, coupled with immediate trained CPR in the hospital setting, likely accounts for a substantial portion of this difference. Survival predictors for ECPR in the in-hospital setting have been recently described in the (RESCUE-IHCA) score [17]. Aside from in-hospital illness categories (medical cardiac vs surgical cardiac vs surgical non-cardiac) and renal insufficiency all other factors have been previously described: age, duration of cardiac arrest, time of day and presenting rhythm. The only moderate performance of the predictive score raises the question whether entirely different criteria are needed for IHCA, and should be an area of future research.

This study has several strengths. We included a fairly large, multiregional cohort of ECPR patients, which improves the generalisability of the findings. We developed clear definitions a priori to reduce bias in data collection. Three experienced researchers used multiple primary sources to cross check the validity of the data. However, there are several limitations. The decision to initiate ECMO is complex, and additional unmeasured factors may have influenced clinicians. The prehospital and ECPR practise itself was not mandated in this observational study—however practice is similar across both states of Australia. Finally, many patients were initiated on ECPR despite being outside local protocols. This can be seen in that > 20% of patients did not fulfil even the liberal criteria. This unpredictable application of the criteria may have influenced the final cohort and the overall findings of the study. Future work will focus on the impact of different criteria on ECPR usage and outcomes.

For ECPR criteria to be effective in including most potential survivors, the data need to be both simple and readily available (< 5 min) to inform clinicians whether or not to commence ECPR for both OHCA and IHCA. However, such criteria also inherently reduce specificity, potentially resulting in the initiation of cardiac arrest patients who are unlikely to benefit. Future work on refining criteria for OHCA and developing specific criteria for IHCA is a high priority for future ECPR research.