Background - Methods - Results - Characetristics of Included Studies - References - Data Tables & Graphs
All except the UK trial had very small numbers of patients. Two of the trials used conventional randomisation with low potential for bias. The other two used less usual designs which have led to difficulties in their interpretation.
All four trials showed a strong benefit of ECMO on mortality (RR 0.44; 95% CI 0.31 to 0.61), especially for babies without congenital diaphragmatic hernia (RR 0.33, 95% CI 0.21 to 0.53). Only the UK trial provided information about death or disability at one and four years, and showed benefit of ECMO at one year (RR 0.56, 95% CI 0.40 to 0.78), and at four years (RR 0.62, 95% CI 0.45 to 0.86). Overall nearly half of the children had died or were severely disabled at four years of age, reflecting the severity of their underlying conditions. Based on economic analysis from the UK trial, the ECMO policy is as cost-effective as other intensive care technologies in common use.
Further studies are needed to refine ECMO techniques; to consider the optimal timing for introducing ECMO; to identify which infants are most likely to benefit; and to address the longer term implications of neonatal ECMO during later childhood and adult life.
The concept arose as an off-shoot of cardiopulmonary by-pass technology. Initially it was used to support adults but early results were poor. Similarly, early attempts to use ECMO in the treatment of newborns were unsuccessful; cannula problems provided the greatest technical difficulty. However in 1975 Bartlett reported the first mature newborn treated successfully with ECMO and other reports soon followed (Bartlett 1976). It subsequently became clear that mature infants with persistent pulmonary hypertension of the newborn (PPHN) were particularly suited to ECMO since the better oxygenation and physiological stability produced by ECMO improved pulmonary blood flow without the risk of further barotrauma.
ECMO is an extremely invasive and technically involved procedure. Traditional ECMO uses two large gauge catheters, one placed in a central vein and the other in a central artery (veno-arterial or V-A). It is essential to achieve adequate flow rates (approximately 100 - 120 mls/kg/min) and as a result cannulae are normally 12 - 14 French gauge. Blood is drained passively via the venous catheter which is inserted into the internal jugular vein and positioned in the right atrium. Blood then passes on to a pump which maintains flow in the circuit. A 'bladder box' and servo system prevent the pump from working if venous drainage becomes inadequate for any reason. Blood then passes to an oxygenator where a sweep gas passes in counter current to the blood. The concentration of oxygen in the sweep gas can be adjusted depending on the needs of the patient. Before re-entering the body warming occurs in a heat exchange column. Blood is returned via the common carotid artery at systemic pressure. This type of ECMO is able to support both pulmonary and cardiac function. More recently veno-venous (V-V) ECMO, which provides just pulmonary support, has become popular. The particular, theoretical, advantage of V-V ECMO is that the cerebral arterial blood supply is not disrupted.
Whilst on ECMO additional gas exchange by the lungs is not essential and therefore ventilation is normally reduced to 'rest' settings. This is typically 5 - 10 cm H2O positive end expiratory pressure and 10 to 20 breaths per minute but the approach does vary from centre to centre. This strategy prevents any further lung damage secondary to barotrauma but arrests the atelectasis which might follow acute withdrawal of respiratory support and enhances clearance of secretions.
The point in an individual baby's course at which ECMO should be considered is debatable. A variety of physiological and clinical parameters have been used. However, over time, oxygenation index (OI) of greater than 40 has probably become the most widely employed, where
OI =(Fi02) * (mean airway pressure cm H20) * 100 / PaO2 mm Hg.
Although the absolute number of babies who reach this level of severity is never likely to be large, the potential benefits of ECMO may be extremely high. The policy is very invasive, however, and because it is so labour intensive, it is likely to be expensive. Hence there is a need for rigorous evaluation of its advantages and disadvantages to guide practice.
Secondary analyses of the primary outcomes (see below) are based on those with and without a primary diagnosis of congenital diaphragmatic hernia, and by severity (oxygenation index between 40-60, and over 60)
Other outcomes include impairment (with or without disability) at one year of age, readmission to hospital in the first year, need for supplemental oxygen at one year of age, tube feeding at one year, weight < the 3rd percentile at one year of age, head circumference < the 3rd percentile at one year of age, head circumference > the 97th percentile at one year of age, visual problems at one year of age, hearing problems at one year of age, on anticonvulsants at one year of age, changes in neuromotor tone at one year of age, asymmetrical neuromotor signs at one year of age, abnormal axial tone at one year of age, abnormal movements at one year of age, motor developmental quotient <50 at one year of age, motor developmental quotient <70 at one year of age, overall developmental quotient <70 at one year of age, professional support for special needs at four years of age.
Outcomes indicating use of resources indicating levels of intensiveness, and therefore cost, of care are: days on ECMO, days on oxygen >90%, days on ventilator, days on supplemental oxygen before first discharge home, and days in hospital before first discharge home. These categories are not mutually exclusive. Further indication of increased or reduced health and other care resource use is given by the outcome 'readmission to hospital in first year'.
Incremental cost per additional survivor and per additional survivor without disability at one year are also reported in local currency values, without summary statistics.
Other outcomes not considered in the review protocol but provided by authors have been given within the Included Studies Table.
The Cochrane Neonatal Group Specialised Register and the Cochrane Controlled Trials Register were searched using keywords ECMO, extra corporeal membrane oxygenation, extracorporeal membrane oxygenation, extra-corporeal membrane oxygenation, and neonat*. MEDLINE was also searched. Searches covered the period 1974 to 2001.
Trial data were extracted by two authors independently
Further information was sought from the authors of the trials, as appropriate.
Analysis is by intention to treat, using Review Manager (RevMan) software. For dichotomous data, summary relative risks are calculated using a fixed effects model providing there is no significant heterogeneity. For continuous data, weighted mean differences are calculated. 95% confidence intervals are used.
Trials that included economic analysis were noted, and associated publication of economic findings referenced. Critical abstracts of economic evaluations of ECMO are available in the NHS economic evaluation database, which is also included in the Cochrane Library. In this review, data about key items of resource use and patient based costs are reported. Where studies meet BMJ criteria for economic evaluation (Drummond et al 1996) and also report measures of incremental cost-effectiveness, this is also reported.
One study was associated with a full economic evaluation, reported separately (Roberts et al 1998)
All the trials except the UK trial had very small numbers of patients. Three of the trials (Boston, Michigan and UK) were stopped early for effectiveness on the advice of the relevant Data Monitoring Committee, in accordance with pre-specified stopping rules in their trial protocols. Nevertheless, as early stopping is often associated with a random high, it is possible that the reported effect sizes may be exaggerations of the true treatment effect.
Two of the trials (Syracuse and UK) used conventional randomisation methods with low potential for selection bias at trial entry. They also used an intention to treat analysis based on patients in the groups to which they were randomised, and with virtually no loss to follow up. Although the treatments could not be masked after randomisation, the outcome measures such as death were unlikely to be subject to observer bias, and the assessor at paediatric follow up in the UK trial was kept unaware of the treatment allocation.
There were more problems about methodological quality in the other two trials which used less usual designs. Both employed a Zelen design (Zelen 1979) in which informed consent to treatment was requested after randomisation, and only to the ECMO arm. This method has high potential for selection biases before and after trial entry if the recruiting clinician, on seeing which treatment has been randomly allocated to a particular patient, then does not ask that patient /parent for consent to the (known) treatment and/or to the follow up; parents may also decide not to consent to a particular treatment, and/or to enter their baby into the trial and/or to give permission for follow up. The potential for bias arises because these decisions are made in knowledge of the allocated treatment and may therefore be differentially affected by that knowledge. This may be even more of a problem if, as in these ECMO trials, a single consent design is used, as one group may not have the opportunity to refuse. The trial reports do not provide sufficient information to be able to assess the extent of these biases (although the Boston trial states that there were no post-randomisation exclusions).
They also used 'response-adaptive' designs. In the Boston trial, this led to a decision to halt randomisation after the fourth death in either trial group. (There was also subsequently a non-randomised phase of this trial, but data from that phase have not been used in this review). In the Michigan trial, the adaptive design used the 'play the winner' strategy in which the first patient was given an equal chance of randomisation to either trial arm, but subsequent allocations were based on the results for the previous allocation, with a higher probability of allocation to the treatment doing better at the time. This has led to a major imbalance in the numbers of infants in each trial arm (only one in the conventional management arm). These unusual designs have led to difficulties in the interpretation of their results.
Mortality
Death before discharge home (or to the end of data collection) were
the only outcomes reported for all four trials. For death before discharge
home, each of the four trials showed a strong benefit of ECMO, but as the
three US trials were all very small, the size of effect (typical RR 0.44)
was overwhelmingly determined by the UK trial and the 95% CI was very tight
(0.31 to 0.61), a highly statistically significant benefit (p<0.00001).
This can also be expressed as a difference in rates of -0.32 (95% CI -0.44
to -0.20), implying only three babies need to be treated with ECMO rather
than conventional ventilation to prevent one death. The situation was similar
for deaths to the end of data collection (typical RR 0.50, 95% CI 0.37
to 0.69; p=0.00003), although there were some later deaths in the ECMO
arm (from the trials with follow up).
The majority of patients in these trials did not have congenital diaphragmatic hernia as the primary diagnosis either because this was an exclusion criterion (Boston and Syracuse) or because the numbers with this primary diagnosis were relatively small (1/12 in the Michigan trial and 35/185 in the UK trial). The risk of death by discharge for babies without this diagnosis was reduced even more (typical RR 0.33, 95% CI 0.21 to 0.53; p<0.00001). The results were similar for deaths to the end of data collection (typical RR 0.41, 95% CI 0.27 to 0.63; p=0.00004). Even for the 35 babies in the UK trial with a primary diagnosis of congenital diaphragmatic hernia, the risk of death was reduced (RR 0.72, 95% CI 0.54 to 0.06; p=0.03), but only five infants survived to discharge, and only three children survived to four years of age, all in the ECMO arm (17/17 of the infants in the conventional management arm died before discharge).
Death or disability
Only the UK trial provided information about death or disability at
one and four years. This again showed an overall benefit of ECMO at one
year (RR 0.56, 95% CI 0.40 to 0.78; p=0.006), and at four years (RR 0.62,
95% CI 0.45 to 0.86; p=0.004). The benefit was even more marked in the
subgroup of children who did not have a primary diagnosis of congenital
diaphragmatic hernia at trial entry (RR at one year 0.45, 95% CI 0.28 to
0.72; p=0.009), and at four years (RR 0.49, 95% CI 0.31 to 0.77; p=0.002).
The trend towards benefit for the children with congenital diaphragmatic
hernia at trial entry was much less marked (RR at one year 0.78, 95% CI
0.61 to 1.00; p=0.05), and at four years (RR 0.89, 95% CI 0.75 to 1.05;
p=0.16), with only two children alive and not severely disabled, both in
the ECMO arm.
The Oxygenation Index at trial entry was used as a measure of severity. The effect of a policy of ECMO by four years of age was more marked in the less severe stratum of OI 40-60 (death or severe disability at four years RR 0.52, 95% CI 0.31 to 0.85; p=0.010) than the more severe stratum of OI >60 (death or severe disability at four years RR 0.76, 95% CI 0.52 to 1.12; p=0.16) although the trend is in the same direction.
Disability and impairment
Data from the UK trial at one year showed no clear trend in relation
to the risk of disability or impairment. Assessment of children at one
year is difficult to interpret and hence developmental assessments are
likely to have lacked precision. At 4 years much more detailed information
was available. Five children were lost to follow up (3 in the conventional
management group). Of the 60 randomised to ECMO and assessed at 4 years,
12 appeared normal and 18 had signs of impairment without disability. The
remaining 30 had signs of disability (3 severe). In the conventional arm
35 children were assessed, of whom 4 appeared normal with 9 having signs
of impairment without disability. The other 22 children in this group were
disabled but none were considered severe. The data did not suggest that
an increased risk of particular types of adverse neurodevelopmental outcome
(eg hemiplegia) was associated with either group.
Use of health services
Measures of resource use are analysed as continuous variables. All
four studies reported one or more of the defined resource use outcomes,
but the three American studies provided this information for survivors
only. In the UK trial, data were reported as medians (interquartile ranges
(IQR)). These showed that a policy of ECMO compared to CM led to more days
on ECMO (4 (3-7) vs 0); more days on a ventilator (2 (0.5-4) vs 0 (0-5));
more days on supplemental oxygen (3 (0-12.5) vs 0 (0-5)); fewer days on
oxygen at >90% (0.5 (0.5-1) vs 2 (1-5)); more days in hospital before first
discharge home or death (6 (1-11) vs 0.5 (0-6)); and fewer hospital readmissions
during the first year (0 (0-3) vs 1 (0-7)). Some of the greater resource
use in the ECMO arm is because of the increased survival.
Costs and cost effectiveness
Only one study (UK 1996) included costs of health care over the year, and this was reported separately (Roberts TE et al 1998). The median cost/case for patients receiving extracorporeal membrane oxygenation was £15276 (IQR £11242-£24786) (mean £20,826 ) versus £3702 (IQR 2314-£9649) (mean £7,002) for patients receiving conventional treatment (1994-95 UK sterling prices). When compared to the gain in survival, the additional cost per additional survivor at one year was £51,222, and the additional cost per additional survivor without severe disability was £75,327. Sensitivity analysis for uncertainty about transport costs, staffing levels in neonatal and ECMO units, and odds of survival, found that the range of cost per additional survivor could be between £34,346 and £110,593. The purchasing power parity between UK£ to US$ in 1996 was £0.644GB=$1US (OECD 2001).
The diagnosis of severe but potentially reversible respiratory failure is not straightforward. Over the time that ECMO has been available a variety of indices have been used in this role. All are intended to identify babies with a high probability of death from continued conventional therapy. The results of this review would indicate that they achieve this aim. The various measures used to identify suitable infants have not been compared but this seems unnecessary given the randomised nature of the subsequent studies.
The invasive nature of ECMO has been the cause of much concern. The potential for acute problems related to the ECMO circuit and the inevitable disruption to the cerebral circulation led many to make the broad assumption that there was an inherent risk attached to the use of ECMO which would inevitably result in increased morbidity. These concerns have not been born out. Since the risks are undeniable it would appear that the damaging effect of prolonged exposure to aggressive conventional therapy are even greater. It is important to note that only a minority of all recruited infants could be considered normal survivors at four years. Although ECMO has been considered as a single entity in this comparison there was significant use of the veno venous technique in the UK study whilst this was not the case in earlier trials.
The majority of patients in these trials did not have congenital diaphragmatic hernia as the primary diagnosis either because this was an exclusion criterion (Boston and Syracuse) or because the numbers with this primary diagnosis were small (1/12 in the Michigan trial and 35/185 in the UK trial). Although the balance of benefit was still in favour of the ECMO policy (17/17 of the infants in the conventional management arm died before discharge), by the age of 4 years, 16/18 of those in the ECMO arm had also died or were severely disabled.
There was no evidence that the severity of illness as judged by an OI of 40-60 or over 60 affected the benefit of the ECMO policy.
Although there is a clear benefit for the ECMO policy, overall nearly half of the children had died or were severely disabled at four years of age, reflecting the severity of their underlying conditions. Nevertheless, based on the economic analysis from the UK trial (Roberts et al, 1998), the ECMO policy is not only clinically effective but also as cost-effective as other intensive care technologies in common use.
The situation for babies with diaphragmatic hernia is less clear since, despite their common underlying anomaly, they do not represent a homogeneous group. It would appear that ECMO offers short term benefits but the overall effect of employing ECMO in this group is not clear. In the absence of a definitive study the use of ECMO can only be recommended on clinical grounds i.e. where it can be used to stabilise a baby thought to be potentially viable but failing more conventional support.
Cost effectiveness is sensitive to the organisation of health care for ECMO and intensive neonatal care. Lower cot occupancy and higher staff to cot ratios increase costs, as do long travel times and distances.
The identification of suitable infants also merits further consideration. At present infants are referred for ECMO when other therapies have failed and the baby is continuing to deteriorate. Outcomes might be improved by introducing ECMO earlier, ie as soon as all other therapies have failed.
The longer term effects of neonatal ECMO (eg during later childhood, adolescence and adult life) remain unclear. Studies to address these issues are clearly important if infants are going to continue to be offered this form of life support. A seven year follow up is in progress for the UK trial.
The correct approach to the management of infants with diaphragmatic hernia is not known. Large randomised studies, with long term follow up, are needed in order to establish both the best approach to acute management and the extent to which "normal survival" is achievable with our present treatment options. There is some uncertainty about what constitutes "present treatment options" and establishing the test arms would clearly be the first step in developing such a study.
Study | Methods | Participants | Interventions | Outcomes | Notes | Allocation concealment |
Boston 1989 | Adaptive design with single consent Zelen randomisation. No post- randomisation exclusions. Randomisation in balanced blocks of size 4 and planned to cease after 4th death in either group. Phase II was non -randomised enrolment in group with <4 deaths until 4th death in that group or or number of survivors significantly larger than number of survivors in arm discontinued first | 19 infants with severe persistent pulmonary hypertension and respiratory failure. Birthweight >= 2.5 kg, gestational age >= 38 weeks, normal cranial ultrasound, severe hypoxemia, 80% predicted mortality based on PaO2/PAO2 <=0.15 on 2 occasions > 30 mins apart between 12 and 72 hours after birth. Exclusions: congenital diaphragmatic hernia, heart diesase. | Extra Corporeal Membrane Oxygenation (venoarterial) usually involving
transport to a multidisciplinary intensive care unit (within Boston).
Conventional treatment remained on optimal ventilatory support in initial neonatal intensive care unit. |
Death, duration of ventilation and of supplemental oxygen, intracranial haemorrhage, complications of ECMO | Methodological quality
Masking of intervention (not possible) Completeness of follow up (yes, until discharge) Masking of outcome assessment (mortality outcome so masking not appropriate) |
B |
Michigan 1985 | Adaptive design with single consent Zelen randomisation.
'Play the winner' - 1st patient equal chance of randomisation to either arm, but subsequent assigments based on results for previous patients - higher probability for treatment doing better. |
12 infants with newborn respiratory failure; > 2kg birthweight; any
of following:
1 acute deterioration PaO2<40 mmHg of pH<7.15 for 2 hours 2. Unresponsive- ness (2 of 3 indication for 3 hours - PaO2<55, pH<7.4 or hypotension 3. barotrauma 4. congenital diaphragmatic hernia 5 80%+ mortality index at 24 hours Exclusions: intracranial haemorrhage grade II or more; > 7 days; incompatible with normal quality life |
Single centre for both treatments. Extra Corporeal Membrane Oxygenation
(venoarterial if signs of haemodynamic instability, otherwise veno-venous).
Conventional treatment remained on optimal ventilatory support. |
Death, duration of ventilation and of hospital stay, intracranial haemorrhage, complications of ECMO some follow up | Methodological quality
Masking of intervention (not possible) Completeness of follow up (yes, until discharge) Masking of outcome assessment (mortality outcome so masking not appropriate) |
B |
Syracuse 1992 | Assigned randomly (no other details). | 28 infants with respiratory failure - oxygenation index >40 for 4 hours; >35 weeks; >= 2 kg; 10 days; Exclusions: intraventricular haemorrhage, structural heart disease, congenital diaphragmatic hernia; severe congenital anomaly | Single centre for conventional treatment.
Transport for Extra Corporeal Membrane Oxygenation (venoarterial) in one of 3 centres. Conventional treatment remained on optimal ventilatory support. |
Death, duration of ventilation, of supplemental oxygen, and of hospital stay, intracranial haemorrhage, complications of ECMO. Follow up to 2 years - neurological abnormality, Bayley scores. | Methodological quality
Masking of intervention (not possible) Completeness of follow up (yes, until discharge, and good at follow up) Masking of outcome assessment (masking not appropriate for mortality outcome; not clear if paediatric assessor masked) |
A |
UK 1996 | Central telephone randomisation with minimisation on primary diagnosis, severity, and referral hospital and ECMO centre | 185 infants with severe respiratory failure (oxygenation index >40);
> 2kg birthweight; >35 weeks gestation;
<10 days high pressure ventilation; < 28 days old; no contraindiction for ECMO (ventricular haemorrhage, irreversible cardiopulmonary disease, asystole, necrotising enterocolitis); no major congenital anomaly |
Transport for Extra Corporeal Membrane Oxygenation (venoarterial) in
one of 5 centres.
Conventional care in centre accustomed to providing optimal ventilatory support |
Death, duration of ventilation, of supplemental oxygen, and of hospital stay, intracranial haemorrhage, complications of ECMO. Follow up to 1 and 4 years completed - respiratory, growth, vision, hearing, neuromotor/neurological abnormality, Griffith scores; disability and impairment; health service use and cost effectiveness. 7 years on-going | Methodological quality
Masking of intervention (not possible) Completeness of follow up (yes, until discharge, and good at follow up) Masking of outcome assessment (masking not appropriate for mortality outcome, and paediatric assessment was masked) |
A |
O'Rourke PP, Crone RK, Vacanti JP, Ware JH, Lillehei CW, Parad RB, Epstein MF. Extracoporeal membrane oxygenation and conventional medical therapy in neonates with persistent pulmonary hypertension of the newborn: A prospective randomized study. Pediatrics 1989;84:957-63.
Michigan 1985 {published data only}
Bartlett RH, Roloff DW, Cornell RG, Andrews AF, Dillon PW, Zwischenberger JB. Extracorporeal circulation in neonatal respiratory failure: A prospective randomized study. Pediatrics 1985;76:479-87.
Syracuse 1992 {published and unpublished data}
* Bifano EM, Hakanson DO, Hingre RV, Gross SJ. Prospective randomized controlled trial of conventional treatment or transport for ECMO in infants with persistent pulmonary hypertension (PPHN). Pediatr Res 1992;31:196A.
Gross SJ, Bifano EM, D'Eugenio D, Hakanson DO, Hingre RV. Prospective randomized controlled trial of conventional treatment or transport for ECMO in infants with severe persistent pulmonary hypertension (PPHN): two year follow up. Pediatr Res 1994;36:17A.
* UK Collaborative ECMO Trial Group. UK collaborative randomised trial of neonatal extracorporeal membrane oxygenation. Lancet 1996;348:75-82.
UK Collaborative ECMO Trial Group. The Collaborative UK ECMO Trial: Follow-up to 1 year of age. Pediatrics (URL: http://www.pediatrics.org/cgi/contents/full/101/4/e1) 1998;101(4).
Roberts T and the Extracorporeal Membrane Oxygenation Economics Working Group. Economic evaluation and randomised controlled trial of extracorporeal membrane oxygenation: UK Collaborative Trial. BMJ 1998;317:911-16.
Bennett C, Johnson A, Field D, Elbourne D for UK Collaborative ECMO trial group. UK collaborative randomised trial of neonatal extracorporeal membrane oxygenation: follow up to age 4 years. Lancet 2001;357:1094-6.
* indicates the primary reference for the study
Bartlett RH, Gazzaniga AB, Jefferies MR et al. Extracorporeal membrane oxygenation (ECMO) cardiopulmonary support in infancy. Trans Am Soc Artif Intern Organs 1976;22:80-93.
Drummond MF, Jefferson TO. Guidelines for authors and peer reviewers of economic submissions to the BMJ. The BMJ Economic Evaluation Working Party. BMJ 1996;313:275-83.
Organisation for Economic Cooperation and Development. Main economic indicators. http://www1.oecd.org October 2001.
Zelen M. A new design for randomized clinical trials. N Eng J Med 1979;300:1242-1245.
02 Infants without congenital diaphragmatic hernia as principal diagnosis
02.01 Death
before discharge home
02.02 Death
in the first year of life
02.03 Death
at any time to the end of data collection
02.04 Severe
disability in survivors at one year of age
02.05 Disability
(severe and not severe) in survivors at one year of age
02.06 Death
or severe disability at one year of age
02.07 Death
or severe disability at 4 years of age
03 Infants with congenital diaphragmatic hernia as principal diagnosis
03.01 Death
before discharge home
03.02 Death
in the first year of life
03.03 Death
at any time to the end of data collection
03.04 Death
or severe disability at one year of age
03.05 Death
or severe disability at 4 years of age
04 Infants with oxygenation index 40-60 at trial entry
04.01 Severe
disability in survivors at one year of age
04.02 Disability
(severe and not severe) in survivors at one year of age
04.03 Death
or severe disability at 4 years of age
05 Infants with oxygenation index > 60 at trial entry
05.01 Severe
disability in survivors at one year of age
05.02 Disability
(severe and not severe) in survivors at one year of age
05.03 Death
or severe disability at 4 years of age
Miranda Mugford
Professor of Health Economics
School of Medicine, Health Policy and Practice
University of East Anglia
Norwish
UK
NR4 7TJ
Telephone 1: +44 1603 593583
Facsimile: +44 1603 593604
E-mail: M.Mugford@uea.ac.uk