Queensland adult ECMO retrieval service: A description of the service and analysis of outcomes

Abstract

Objective

To describe the Queensland Adult ECMO Retrieval Service (QAERS) and assess observed mortality of patients retrieved and treated with ECMO, against benchmarks.

Design

Data was retrospectively collected from clinical and quality assurance databases at the three QAERS hospitals. Demographic data, diagnostic category, and hospital mortality were collected for patients referred to QAERS. Additional data was collected on patients receiving ECMO either before or after transport to a receiving hospital (ECMO patients), enabling calculation of RESP or SAVE scores. In ECMO patients with cardiogenic shock, individual risk of deaths were calculated by SAVE score. Monte Carlo analysis generated a discrete probability distribution function (PDF) of expected number of deaths, with 95 % confidence intervals (CI). The observed number of deaths was compared to this PDF. This was repeated for ECMO patients with respiratory failure, using RESP score.

Setting

ICUs in Queensland and Northern NSW.

Participants

All patients referred to QAERS from May 2017 to December 2023.

Main outcome measures

Predicted and observed mortality of ECMO patients with cardiogenic shock or respiratory failure.

Results

237 patients were referred. 135 were retrieved, with 77 transported on ECMO. 11 commenced ECMO after transfer, giving a total of 88 ECMO patients. 35 ECMO patients had cardiogenic shock and 53 had respiratory failure. 16 cardiogenic shock patients died (95 % CI of PDF 17–28). 7 respiratory failure patients died (95 % CI of PDF 8–19).

Conclusions

Observed mortality of patients retrieved and treated with ECMO was lower than mortality predicted by SAVE and RESP scores.

1. Introduction

1.1. Background

Multiple reports describe Extra-Corporeal Membrane Oxygenation (ECMO) retrieval systems operating outside of Australia.[1], [2], [3] Most use a hub and spoke model with patients retrieved to a single hospital. Case-mix may be adult, paediatric, or both.[3], [4], [5] In Australia, adult ECMO retrieval systems in NSW^6,7 and Victoria^8,9 have been reported. Only some reports include patient outcomes such as survival to ECMO separation or hospital discharge.^5,10,11 Case-mix adjustment is required for valid comparison of outcomes across care providers, but there are no reports assessing performance of an ECMO retrieval service using case-mix adjustment.

Queensland is the second largest Australian state, with a land area of 1.7 million km², larger than New South Wales and Victoria combined.¹² Approximately half the population of 5.6 million resides outside the state capital (Greater Brisbane).^13,14 This geography and population distribution provide unique challenges for ECMO retrieval.

This study assesses the Queensland Adult ECMO Retrieval Service (QAERS) comparing observed mortality of patients retrieved and treated with ECMO, to predicted mortality generated by the SAVE¹⁵ and RESP¹⁶ scores.

1.2. Description of QAERS

The QAERS commenced in 2017 to provide equity of access to ECMO services across Queensland and northern New South Wales (NSW). Three hospitals provide ECMO retrieval, each ‘on-call’ for one week in three in a rotating roster. When a patient is referred, clinical advice is provided by the QAERS intensive care specialist, and if deemed appropriate a retrieval is initiated. When ECMO will be used during transport, an ECMO retrieval team travels to the site, initiates ECMO, then transports the patient back to the receiving hospital. All retrievals on ECMO are by road ambulance or fixed wing aircraft. A detailed description of QAERS is given in Supplemental Appendix S1.

2. Methods

2.1. Data collection

The Metro South Hospital and Health Service Human Research Ethics Committee Ethics granted approval and waiver of consent.

Inclusion criteria:

• 1.Patient referred to QAERS as a potential ECMO candidate.
• 2.Date of referral between 1 May 2017 and 31 December 2023.
• 3.Inter-hospital transport either by road ambulance or aeromedical asset would be required.

Databases at QAERS hospitals were examined retrospectively. A basic dataset was collected for all patients referred to QAERS, including demographic and anthropometric information, diagnostic category (cardiogenic shock or respiratory failure), and in-hospital mortality. Additional data was collected on patients that received ECMO, including baseline diagnostic and physiologic information submitted to Extracorporeal Life Support Organisation (ELSO), collected according to the rules in place at the time of data submission. Any missing data was treated as having the value that would minimise mortality. A full description of the data collected is given in Supplemental Appendix S2.

Patients were commonly admitted sequentially to several hospitals. To capture the impact on hospital resources we recorded the duration of the “continuous period of acute care” and the “continuous period of ICU care”.¹⁷ A “continuous period of acute care” is defined as from first presentation to an acute care hospital until discharge to one of the following: home, dedicated rehabilitation ward or hospital, chronic care facility, or death. These end points are consistent with ELSO Registry endpoints, except that the ELSO Registry counts patients transferred to another hospital as survivors, which underestimates mortality.⁹ Patients surviving their continuous period of acute care were counted as survivors. A “continuous period of ICU care” is defined as the time from admission to any ICU to discharge from ICU.

Retrieval time was defined as the duration between the time the team left the QAERS hospital and the time the patient arrived at the destination QAERS hospital.

Two other hospitals provide ECMO in Queensland. Occasionally they commenced a patient on ECMO, then referred them to QAERS for transport to the transplantation hospital for consideration of durable VADs or transplantation. Patients treated with E-CPR, requiring transfer because of ongoing cardiogenic shock, were included in the cardiogenic shock group.

2.2. Statistical methods

For patients treated with ECMO, baseline diagnostic and physiologic information submitted to ELSO was used to calculate the RESP score for each respiratory failure patient¹⁶ and the SAVE score for each cardiogenic shock patient.¹⁵ These scores were used to estimate individual risk of death for the patients in these two groups. Each patient's individual risk of death (iROD) was determined by the average mortality of their risk class. We used Monte Carlo simulation to generate a discrete probability distribution function (pdf) of the expected number of deaths in each group, with 95 % confidence intervals.^17,18 This distribution can be used for statistical inference, which cannot be done with a simple expected versus observed count.

In a single simulation, each patient in the series of patients is assigned a random number between 0 and 1. If this number is higher than their iROD they live, otherwise they die. The deaths in the simulation are added to give the number of deaths for that simulation. Multiple simulations are done, generating a pdf of the number of expected deaths. When the number of simulations is sufficient, the estimate of the pdf stabilises giving a consistent result.

Initially, 1000 simulations were used to generate one pdf. This was repeated 1000 times to produce 1000 pdfs. Each pdf was examined to find which expected number of deaths were outside the 95 % confidence intervals. If this was identical for all 1000 pdfs the final pdf had been found, otherwise the number of simulations was increased by a factor of 10, and the process restarted. Details of the Monte Carlo simulation are in Supplemental Appendix S3.

The primary outcome for each of the two groups (respiratory failure patients treated with ECMO and cardiogenic shock patients treated with ECMO) was whether the observed number of deaths was below (better than expected), within (not significantly different), or above (worse than expected), the 95 % confidence intervals of the expected number of deaths. The two null hypotheses tested were that there was no difference between the observed and expected number of deaths, in the two groups.

3. Results

3.1. Referral and retrieval activity

Patient flow through the referral and retrieval process is shown in Fig. 1. Of the 237 patients referred, 135 were retrieved, with 77 patients cannulated in the referring hospital and transferred on ECMO, and another 11 patients commenced on ECMO following transfer. Fifty-three patients with respiratory failure and 35 patients with cardiogenic shock were treated with ECMO and included in the risk adjusted analysis. There were no deaths during retrieval transport, or at the referring hospital following ECMO cannulation by the QAERS team.

Fig. 1.

Patient flow through the process of referral and retrieval.

Referring hospitals from Sunshine Coast and south of this (SEQ and Northern NSW), are within 250 km of all QAERS hospitals (Fig. 2), a distance considered as the upper limit for road-based retrieval. There were 177 referrals from SEQ and Northern NSW. 103 of these patients were retrieved, all by road ambulance. 54 of these were transported on ECMO, and 9 commenced ECMO after retrieval. Of the 60 referrals from areas north of SEQ, air transport was used for all 32 that were retrieved. 23 of these were transported on ECMO, and 2 commenced ECMO after retrieval.

Fig. 2.

Location of hospitals referring to the QAERS.

Cities with referring hospitals outside of Greater Brisbane shown as filled-in circles. Cities with QAERS ECMO hospitals shown as stars. The straight-line distance from Cairns to Brisbane is 1400 km. Referring hospitals from the Sunshine Coast and south of this, are within 250 km of QAERS ECMO hospitals.

Of the 77 patients retrieved on ECMO, retrieval time was available for 41 road retrievals and 21 aeromedical retrievals. Median road retrieval time was 6h 30m (range 1h 34m–11h 15m), and median aeromedical retrieval time was 17h 0m (range 9h 0m–24h 30m).

155 patients were referred with severe respiratory failure. In this group, 85 patients were retrieved (either on or off ECMO), 17 (20 %) of these retrieved patients died, and 27 (39 %) of the 70 patients that were not retrieved died. 82 patients were referred with cardiogenic shock. In this group, 50 patients were retrieved (either on or off ECMO), 17 (34 %) of these retrieved patients died, and 23 (72 %) of the 32 patients that were not retrieved died.

3.2. ECMO patient characteristics

Baseline parameters prior to commencing ECMO are shown in Table 1 for the cardiogenic shock group and in Table 2 for the respiratory failure group. For both groups, characteristics of ECMO and the patient outcomes are given in Table 3. Complex cannulation configurations, including central VA ECMO, did not preclude retrieval (Supplemental Appendix S4).

Table 1.

Baseline data before commencing ECMO: Cardiogenic shock patients.

Total number of cases		35
Age (years)		54 [18–75]
Sex (male)		22 (63 %)
Weight (kg)		82 [50–150]
Body mass index		26.3 [19.5–39.3]
Acute cardiogenic shock diagnosis group (one or more per patient)
Myocarditis		7 (20 %)
Refractory VT/VF		2 (6 %)
Post-heart or lung transplantation		0 (0 %)
Congenital heart disease		0 (0 %)
Other diagnoses		29 (83 %)
Post non-transplant cardiac surgery		8 (23 %)
STEMI		5 (14 %)
Pulmonary embolism		8 (23 %)
Septic cardiomyopathy		4 (11 %)
Thyrotoxic cardiomyopathy		2 (6 %)
Drug overdose		2 (6 %)
Acute pre-ECMO organ failures
Liver failure		11 (33 %)
CNS dysfunction		4 (12 %)
Acute renal failure		17 (52 %)
Chronic renal failure		0
Duration of intubation prior to ECMO (hours)a
≤ 10		15 (43 %)
11 - 29		13 (37 %)
≥ 30		2 (6 %)
Peak inspiratory pressure ≤ 20		4 (11 %)
Pre-ECMO cardiac arrest		19 (54 %)
Diastolic blood pressure ≥ 40		7 (20 %)
Pulse pressure ≤ 20		9 (26 %)
HCO3 before ECMO ≤ 15		17 (49 %)
Individual risk of death predictions
SAVE risk class	Predicted mortality (%)	Number of patients (%)
I	25 %	1 (3 %)
II	42 %	3 (9 %)
III	58 %	12 (34 %)
IV	70 %	15 (43 %)
V	82 %	4 (11 %)

Data presented as value, or number (% of total cases), or median [range].

VT/VF = Ventricular Tachycardia or Ventricular Fibrillation.

STEMI = ST Elevation Myocardial Infarction.

^a5 patients did not have duration of intubation prior to ECMO recorded; treated as ≤ 10 h when calculating SAVE scores.

Table 2.

Baseline data before commencing ECMO: Respiratory failure patients.

Total number of cases		53
Age (years)		39 [17–67]
Sex (male)		31 (58 %)
Weight (kg)		90 [48–200]
Body mass index		30.2 [17–78.1]
Acute respiratory diagnosis group (only one per patient)
Viral pneumonia		24 (45 %)
Bacterial pneumonia		18 (34 %)
Asthma		2 (4 %)
Trauma and burn		2 (4 %)
Aspiration pneumonitis		3 (6 %)
Other acute respiratory diagnosis		4 (8 %)
Non-respiratory and chronic respiratory diagnosis		0 (0 %)
Immunocompromised		2 (4 %)
Mechanical ventilation prior to ECMO
< 48 h		36 (68 %)
48hrs - 7 days		10 (19 %)
> 7 days		7 (13 %)
CNS dysfunction		2 (4 %)
Acute associated non-pulmonary infection		3 (6 %)
Neuromuscular blockade agents prior to ECMO		40 (77 %)
NO use before ECMO		13 (25 %)
HCO3 infusion before ECMO		7 (13 %)
Cardiac arrest before ECMO		2 (4 %)
PaCO2 => 75		9 (17 %)
Peak inspiratory pressure => 42		3 (6 %)
Individual risk of death predictions
RESP risk class	Predicted mortality (%)	Number of patients (%)
I	8 %	18 (34 %)
II	24 %	20 (38 %)
III	43 %	12 (23 %)
IV	67 %	2 (4 %)
V	82 %	1 (2 %)

Data presented as value, or number (% of total cases), or median [range].

NO = nitric oxide, HCO3 = Sodium Bicarbonate, CNS = Central Nervous System.

Table 3.

Characteristics and outcomes of ECMO support.

Indication for ECMO	Cardiogenic shock	Respiratory failure
Total number of cases	35	53
Place ECMO first initiated
Emergency department	3 (9 %)	0 (0 %)
ICU	25 (71 %)	53 (100 %)
Operating theatre	6 (17 %)	0 (0 %)
Cardiac catheterization laboratory	1 (3 %)	0 (0 %)
ECMO initiated prior to transfer	33 (94 %)	44 (83 %)
Duration of ECMO support (days)	6 [2–26]	11.5 [4–63]
Continuous period of ICU care (days)	15 [2–91]	26 [4–102]
Duration of invasive mechanical ventilation (days)	12 [2–45]	22 [5–87]
Continuous period of acute care (days)	27 [2–178]	38 [9–129]
Total hospital admission including rehabilitation (days)	43 [2–371]	38 [9–423]
Renal replacement therapy while in ICU	28 (80 %)	30 (57 %)
RBCs transfused (units)	8 [0–97]	6 [0–84]
Outcome of ECMO
Died on ECMO or palliative withdrawal of ECMO	8 (23 %)	4 (8 %)
Survived ECMO	27 (77 %)	49 (92 %)
Discharge destination at end of continuous period of acute care
Died	16 (46 %)	7 (13 %)
Dedicated rehabilitation ward or hospital	5 (14%)	12 (23 %)
Chronic care facility	0 (0 %)	0 (0 %)
Home	14 (40%)	34 (64 %)
Eventual discharge destination
Died	16 (46 %)	7 (13 %)
Chronic care facility	0 (0 %)	0 (0 %)
Home	19 (54 %)	46 (87 %)

Data presented as value, or number (% of total cases), or median [range].

RBCs = red blood cells, ICU = Intensive Care Unit.

2 retrievals were for repatriation to ECMO centres in other Australian states: 1 cardiogenic shock patient and 1 respiratory failure patient.

3.3. Risk adjusted outcomes

In the 53 patients with respiratory failure treated with ECMO, 7 deaths were observed during the continuous period of acute care, below the 95 % confidence intervals[8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19] for deaths predicted by the RESP score class (p < 0.05, Supplemental Fig. S1). All survivors in this group were eventually discharged home.

In the 35 patients with cardiogenic shock treated with ECMO, 16 deaths were observed during the continuous period of acute care, below the 95 % confidence intervals[17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28] for deaths predicted by the SAVE score class (Supplemental Fig. S2). All survivors in this group were eventually discharged home. When we adjusted the risk of death prediction, based on data from Amin et al.¹⁹ which suggested the original SAVE score classes overestimated mortality, our results were not significantly different from predictions (Supplemental Fig. S3). This cardiogenic shock group includes five patients transferred on ECMO following successful ECPR commenced by referral hospitals, of which three survived.

4. Discussion

4.1. Overcoming the tyranny of distance to provide equity of access

Retrievals from Cairns to Brisbane are approximately 1400 km. While much longer international ECMO retrievals have been reported,^1,20 these long-distance retrievals by QAERS are within one state, not infrequent, and are often logistically complex. They take at least 3 senior staff away from the QAERS hospital for over 24 h. Equitable distribution of workload was one reason why the 3-centre model was selected.

Approximately three quarters of the Queensland population lives within Southeast Queensland (SEQ).¹³ Hospitals in SEQ and Northern NSW can be serviced by a road based ECMO retrieval system, but road transport outside of this area is impractical due to distance.

The proportion of patients referred to QAERS from a hospital outside of SEQ and Northern NSW was 25 %, closely mirroring the population distribution. Proportions were similar for patients retrieved to QAERS hospitals, for patients retrieved on ECMO, and for patients that were retrieved and received ECMO. This suggests reasonable equity of access to ECMO services.

4.2. Case-mix adjustment

True utility and cost-effectiveness of an ECMO retrieval service is difficult to ascertain. There may not be equipoise for a RCT testing conventional transfer vs. retrieval on ECMO, notwithstanding the challenges of conducting such a trial. In this setting, reporting risk adjusted survival and benchmarking against successful systems is critical to maintaining a sustainable high-quality service.

Previous publications describing ECMO retrieval services^5,6,[9], [10], [11] do not include case-mix adjustment. Mortality in ECMO patients is highly dependent on diagnosis and physiologic status prior to ECMO initiation.^15,16,21 Case-mix adjustment is required to allow valid comparison of outcomes across care providers. This is the first publication assessing performance of an ECMO retrieval service using a case-mix adjustment.

QAERS performance was satisfactory compared to outcomes of the ELSO registry, as assessed by individual ROD analysis based on SAVE and RESP scores. In QAERS patients treated with ECMO, observed mortality for cardiogenic shock and severe respiratory failure was lower than predicted mortality.

The SAVE and RESP scores do not provide perfect risk of death prediction.^22,23 Two small North American studies suggested SAVE scores overestimate mortality.^19,24 When mortality predicted by SAVE scores was recalibrated using data from Amin et al.,¹⁹ there was no difference between observed and predicted mortality in our cardiogenic shock patients. This is consistent with outcome improving, though the recalibration dataset is only 120 patients from a single centre, limiting interpretation.

It is less clear that VV ECMO mortality is improving. Analysis of the ELSO database showed no difference in survival between 2012 and 2016 (62.3 %) and 2017–2019 (61.7 %), but survival was lower in 2020–2022 (51.8 %)²⁵ (table e4 in reference). In the first year of the pandemic, mortality for COVID treated with ECMO was similar to non-COVID related ARDS.²⁶ Subsequent studies demonstrated mortality worsened over the course of the pandemic.²⁷ RESP score has poor predictive ability in COVID-19 ARDS,^28,29 but as only one patient treated with ECMO in our cohort had COVID-19, this is unlikely to affect our results.

Data for RESP and SAVE scores was collected using data definitions in place at the time of data collection, but ELSO data definitions have changed over time, potentially reducing accuracy of SAVE and RESP mortality predictions.

4.3. Model of care

ECMO retrieval systems often use a ‘hub and spoke’ model with referrals to a single centre.[30], [31], [32], [33], [34] This study demonstrates satisfactory results can be achieved when workload is distributed between several centres. This maintains ECMO case volume at all the centres, distributes workload avoiding excessive demand on any single centre's resources, enables ECMO for trauma patients, and facilitates ECMO for patients that cannot be weaned from cardiopulmonary bypass at the non-transplant hospitals. While the non-transplant hospitals had relatively low numbers of ECMO patients, they are large tertiary hospitals with a high volume of patients with acute severe respiratory failure and cardiogenic shock. They have demonstrated satisfactory ECMO outcomes.¹⁷

There are close links between the hospital providing durable ventricular support and heart-lung transplantation services and the other QAERS hospitals, with strong personal relationships built up over years. All QAERS hospitals participate in statewide ECMO quality and data review meetings, using the same incident monitoring system. There is central coordination of the ECMO retrieval system by the medical clinical lead and clinical nurse consultant, simulation using new equipment and transport assets before they are put into service, and largely standardised equipment for transports. The model of care is not “scoop and run”,³⁵ but “stabilise then controlled transport”.

4.4. Other study limitations

Five patients in the cardiogenic shock group were missing information about ‘duration of intubation prior to ECMO’ and were treated as if this value was ≤10 h, minimising predicted mortality. This may make QAERS performance appear worse than it was. Five patients in the cardiogenic shock group initially received ECPR, then were transferred by QAERS due to ongoing cardiogenic shock. While ECPR outcomes are worse than for cardiogenic shock (38), these patients had survived initial resuscitation and stabilised enough to be retrieved. Comparing them to the ELSO ECPR cohort would be innately biased.

This study compared survival of ECMO patients retrieved by QAERS, to predictions of the SAVE and RESP scores, which were derived from the ELSO database and contain some patients that were retrieved and others that were not. Available data suggest no difference in mortality between patients retrieved on ECMO and those in which ECMO is initiated in a retrieval centre,³⁶ but this remains a potential source of bias. No analysis was undertaken comparing mortality between retrieved and non-retrieved patients, as such comparisons would be highly confounded.

This study examined mortality at discharge from acute care and at hospital discharge. Long-term mortality, functional status, and quality of life measures are more robust patient centred outcomes. However, we followed patients until their final discharge destination, in contrast to most other ECMO studies^1,6,10,11 that treat patients as survivors if they leave the acute care facility alive. All survivors were eventually discharged home.

In common with statistical methods such as binomial exact test and standardised mortality ratio, the Monte Carlo method used in this study assumes model predictions have perfect discrimination and calibration. If they do not, there is risk of overdispersion. In the external validation of the SAVE and RESP studies, the area under the receiver operating characteristic curves were 0.92 and 0.90 respectively. In the external validation of the SAVE study the Hosmer–Lemeshow C-statistic (HLC) was 13.97 (P = 0.08). While no HLC is available for the external validation of the RESP study, HLC for the internal validation was 1.281. This excellent discrimination and good calibration suggests overdispersion is unlikely to have a major effect.

5. Conclusion

In patients retrieved by the QAERS and treated with ECMO either before or after transport, observed mortality during acute hospital care was lower than mortality predicted by RESP and SAVE scores. When the SAVE score was recalibrated with a limited contemporary dataset, there was no difference between observed and predicted mortality in cardiogenic shock patients.

CRediT authorship contribution statement

Nihal Kumpta - Conceptualization, methodology, investigation, data curation, and writing the original draft. Chris Joyce - Conceptualization, methodology, formal analysis, supervision, and writing (review and editing). Germaine Kenny and Jason Meyer - Conceptualization, methodology, investigation and writing (review and editing). James Winearls, James McCullough, Kiran Shekar, Jayshree Lavana, Anand Krishnan, Kyle White, David Cook – Conceptualization, methodology, and writing (review and editing).

Conflict of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Appendix A. Supplementary data

The following are the Supplementary data to this article: