Cost-Effectiveness of Neonatal Extracorporeal Membrane Oxygenation Based on 7-Year Results From the United Kingdom Collaborative ECMO Trial

Clinical Study

The methods for the economic evaluation build on those of the clinical study fully described elsewhere.^8,11 In brief, mature (gestational age at birth ≥35 weeks, birth weight ≥2 kg) newborn infants with severe respiratory failure (oxygenation index ≥40)¹² were enrolled into a pragmatic randomized, controlled trial. At randomization, the infants were already receiving ventilatory support by conventional management in a NICU. The most common diagnoses were persistent pulmonary hypertension because of meconium aspiration, congenital diaphragmatic hernia, isolated persistent fetal circulation, sepsis, and idiopathic respiratory distress syndrome. The infants were recruited from 55 NICUs throughout the United Kingdom between January 1993 and November 1995. Infants allocated to the ECMO group were transferred, by air or road to 1 of 5 specialist regional centers where they were cannulated and commenced ECMO support according to a prespecified protocol. ECMO support was reduced and then discontinued after clinical and radiologic evidence of lung improvement, although moderate levels of conventional ventilation continued to be provided until the infant had stabilized. Infants allocated to the conventional management group continued to receive intensive conventional care in the original hospital, based on trial guidelines for continued ventilation. The primary outcomes of the trial were death and long-term morbidity, which had been assessed previously at age 1 year⁸ and later at 4 years of age.¹³

Standardized neurodevelopmental assessments were subsequently performed in the homes of the surviving children by a developmental psychologist within 3 months of their seventh birthday. Outcome status was assessed across 6 clinical domains (cognitive ability, neuromotor skills, general health, behavior, hearing, and vision) and defined as normal, impaired or mild, moderate, or severely disabled on the basis of the degree of functional loss. A child's overall status was defined by the highest degree of impairment or disability in any of the 6 clinical domains. The individual measures used to assess outcomes and the definitions of impairment and functional disability for each clinical domain are reported in detail elsewhere.¹¹ Ethical approval for the study was obtained from the relevant ethics committees.

Measurement of Resource Use

Data were collected about all of the significant health service resource inputs during the first 7 years of life. The trial data collection forms maintained a record of the mode of transport (road ambulance, fixed wing aircraft, or helicopter), duration of transportation, and distances traveled by the infant initially to the referral center to receive the allocated care and subsequently back to an appropriate local hospital before discharge. The infant's initial stay in hospital was described by 1 of the following 5 mutually exclusive levels of care known to carry different costs¹⁴: (1) days receiving neonatal ECMO, (2) days receiving maximal intensive care (>90% oxygen), (3) days on a ventilator (receiving <90% oxygen), (4) days on supplementary oxygen, and (5) days in standard neonatal care.

Observational work was undertaken to estimate the resource inputs associated with an infant death. These included the resource inputs associated with the postmortem examination and associated procedures and those associated with the transportation of the deceased infant home in an ambulance.

Data on hospital readmissions, outpatient hospital visits, and the use of community and other health care services after the infant's initial stay in hospital were obtained through 3 principal means. First, interviews held over the telephone with the parents of each infant at 4, 8, and 12 months recorded the total hospital and community health service use over each of the previous 4-month periods. The research instruments used as part of the telephone interviews had been piloted to ascertain their acceptability, comprehension, and reliability. Second, the general practitioner of each child was contacted on 2 occasions, first at 4 years and later at 7 years, and asked to provide a detailed profile of each child's attendances and prescriptions of medicine, as well as referrals to hospitals and other community health service providers, over each of the previous 3-year periods. These health care providers were then contacted, first by letter and, where necessary, by telephone, and asked to provide a detailed profile of the care they had provided to the child. Third, 2 face-to-face interviews were held with the parents of each child, the first within 2 weeks of the child's fourth birthday, and the second within 3 months of the child's seventh birthday. The parents were asked a series of structured, closed-ended questions during these face-to-face interviews. At each interview, data were collected about the child's inpatient and outpatient hospital service use over the previous 3-year period, including the name of the hospital provider, its location, the duration of contact, and the ward or clinic attended. Previous research had indicated that the hospital service use of children is accurately recalled by their parents.¹⁵ Therefore, these data were used to validate the information collected directly from the hospital service providers and, where necessary, to supplement missing information. At the 2 face-to-face interviews, the parents were also asked a series of questions about the child's community health service use over the last 6 months, including the professional and agency that provided the service, its location, the frequency of use, and the duration of each service contact, as well as the degree to which the use of each community service had changed when compared with previous periods. When the parental reports of community health service use were extrapolated over the two 3-year periods (1–4 years and 4–7 years) and compared directly with the data collected from the general practitioners and other community health service providers, parental reports tended to slightly underestimate the numbers of contacts. It was, therefore, decided to use the data collected directly from the community health service providers in the final analysis. All of the resource use data were entered directly from the research instruments into a purpose-built data collection program with in-built safeguards against inconsistent entries and then verified by dual coding.

Valuation of Resource Use

Unit costs for each resource item were obtained from a variety of sources. All of the unit costs used followed recent guidelines on costing health care services as part of economic evaluation.¹⁶ An average cost per neonatal ECMO day was calculated by sending each specialist regional center a detailed questionnaire; requesting cost data for the main resource categories of drugs, disposables, equipment, staff, and overheads; and then apportioning these to different categories of patient using a “top-down” methodology.⁹ Each specialist regional center was visited to ensure consistency in the apportionment and reporting of cost data. An average cost per day for each level of non-ECMO conventional neonatal care was derived from estimates from a separately funded study¹⁷ that had been weighted by a factor of 10% to take account of the size and case mix of the units participating in the trial.⁹ Road ambulance costs were obtained from the London Ambulance Service and incorporated a fixed fee for the vehicle, a rate for mileage, and an hourly rate for the total time that the ambulance was in use, whereas air transport costs were obtained from the records of the relevant providers. The cost of postmortem examinations was obtained from local pathologists. Hospital readmissions were valued using National Health Service (NHS) Reference Costs, a catalog of costs compiled by the Department of Health in England for acute care interventions that are clinically distinct and have similar implications for resources.¹⁸ The unit costs of community health services were largely derived from national sources,¹⁹ although some were calculated from first principles using established accounting methods.²⁰ The costs of drugs prescribed by general practitioners were obtained from the British National Formulary.²¹ Unit costs were combined with resource volumes to obtain a net cost per child during the trial period. All of the costs are expressed in United Kingdom pounds sterling and valued at 2002–2003 prices (£1 was equivalent to $1.56 in 2003 using purchasing power parities).²²

Measurement and Representation of Cost-Effectiveness

We performed an incremental cost-effectiveness analysis in which we calculated the incremental costs (ΔC) and incremental effectiveness (ΔE) of neonatal ECMO compared with conventional management and expressed these as a ratio. Cost-effectiveness was expressed in 2 forms: first, in terms of incremental cost per additional life year gained, and, second, in terms of incremental cost per additional disability-free life year gained. For the former measure, the date of death was obtained, where necessary, from health service providers. For the latter measure, the period of survival for children whose overall disability status was not classified as mild, moderate, or severely disabled was estimated in terms of disability-free life years, thereby obviating the need for subjective valuations of each state of disability. The probability that neonatal ECMO is cost-effective at 7 years at different values for the willingness of the English NHS to pay (R_c) for an additional life year and for an additional disability-free life year is represented by cost-effectiveness acceptability curves.²³ For the purposes of our analysis, we have indicated these probabilities at an NHS willingness-to-pay threshold of £30000 for each outcome.²⁴

Data Analysis

A number of statistical approaches were tested to impute costs for 18 children with some missing data (7.87% of all resource items were counted as missing, either as a result of an economic questionnaire that was not completed at all or as a result of an economic questionnaire with missing data items). Simple linear regression and simulation-based multiple imputation²⁵ for each disability grouping of children (normal, impaired, mild disability, moderate disability, and severe disability) produced similar results. Therefore, the estimates generated by the simple linear regression were used in this analysis. Costs and health effects accruing beyond the first year were reduced to present values using the 3.5% discount rate currently recommended for health technology appraisal in the United Kingdom.²⁶

All of the results are reported as mean values with SDs and as mean differences in costs and effects with 95% confidence intervals (CIs) where applicable. We tested for differences in resource use and costs between the comparator groups using the independent-samples t test procedure and tested for differences in effects between the comparator groups using relative risks. The differences in resource use, costs, and effects were considered significant if 2-tailed P values were ≤0.05. Because the data for costs were skewed, we used nonparametric bootstrap estimation to derive 95% CIs for mean cost differences between the comparator groups.²⁷ Each of these CIs was calculated using 1000 bias-corrected bootstrap replications. Nonparametric bootstrap simulation of the cost-effect pairs was also performed to generate 1000 replications of each of the incremental cost-effectiveness ratios, which were represented graphically on 4 quadrant cost-effectiveness planes.²⁸ Mean net benefits, defined as R_c · ΔE − ΔC,²⁹ were estimated for alternative values of R_c, together with their respective 95% bootstrap CIs. All of the analyses were performed with a microcomputer running Microsoft Excel (Microsoft, Redmond, WA) and SPSS (SPSS, Chicago, IL) software.

We undertook sensitivity analyses to illustrate the impact of the principal aspects of uncertainty on the estimates of cost-effectiveness at 7 years. The values of the following variables were varied as part of the sensitivity analyses: (1) community service use (increased by 10%, 20%, and 30% to reflect underreporting in health economic studies);¹⁵ (2) the per diem costs of each level of hospital inpatient care (10% and 20% decrease and increase to reflect variations in the relative price structures of resource inputs across hospital settings);¹⁶ and (3) the discount rate applied to future costs and health effects (set at 0%, 6%, and 10% to reflect differing views in the health economics literature regarding social time preferences and the social opportunity costs of resources).^16,30,31

Projection of Cost-Effectiveness Throughout Childhood

We undertook a secondary analysis to provide an indication of the longer term cost-effectiveness of neonatal ECMO. Given the uncertainties surrounding the values of key economic parameters over the longer term, we restricted this analysis to the childhood years (the first 18 years of life). It was conservatively assumed that all children who survive to 7 years subsequently survive to 18 years and that the excess annual health service cost attributable to neonatal ECMO that was observed during years 4–7 continues during years 8–18. The incremental cost per life year gained was then recalculated. In addition, it was conservatively assumed that each child's disability status at age 7 does not vary during the remaining childhood years before recalculating the incremental cost per disability-free life year gained.

Resource Use and Unit Costs

Table 2 presents mean health service use by trial group and associated unit costs for each resource item. Infants allocated to the ECMO group spent, on average, a statistically significant greater number of days receiving neonatal ECMO, standard neonatal care, and overall hospital care during the initial hospitalization than infants allocated to the conventional management group (P < .05), reflecting the higher survival rate in the ECMO group. They also made a statistically significant greater number of ambulance journeys, outpatient hospital visits, and health visitor visits (P < .05). In contrast, infants allocated to the conventional management group spent, on average, a statistically significant greater number of days receiving maximal intensive care during the initial hospitalization than infants allocated to the ECMO group (P < .05). There were no statistically significant differences between the trial groups in terms of the number of days on a ventilator or on supplementary oxygen during the initial hospitalization or in terms of the number of days of inpatient hospital readmissions or in the number of visits to general practitioners or other community carers during the first 7 years of life.

Health Service Costs

Mean health service costs during the first 7 years of life were £30270 in the ECMO group and £10229 in the conventional management group, generating a mean cost difference of £20041 (bootstrap mean cost difference: £20057; 95% CI: £13690–26318), which was statistically significant (P < .0001; Table 3). Overall cost differences between the 2 groups can be largely explained by the additional care received by infants in the ECMO group during the neonatal period. Neonatal ECMO increased the cost of transportation by an average of £1851 (bootstrap mean cost difference: £1849; 95% CI: £1251–2497; P < .0001) compared with conventional management. Similarly, neonatal ECMO increased the cost of the initial hospitalization by an average of £16853 (bootstrap mean cost difference: £16826; 95% CI: £11775–22044; P < .0001) and the cost of outpatient hospital care by an average of £474 (bootstrap mean cost difference: £481; 95% CI: £97–875; P = .02) compared with conventional management. The costs associated with the management of infant death were, on average, £412 higher (bootstrap mean cost difference: £406; 95% CI: £209–614; P < .0001) in the conventional management group. Statistical analysis revealed that, at the 5% level, there were no significant differences in the mean cost of inpatient hospital readmissions, community health care, and other health services between the 2 trial groups. A more detailed breakdown of the costs of each trial group is available on request.

Cost-Effectiveness at 7 Years

The incremental cost-effectiveness of neonatal ECMO compared with conventional management with respect to the principal outcome measures is shown in Table 4. Although neonatal ECMO was effective at increasing the number of survivors and the number of survivors formally classified as free of disability, it also led to an increase in health care costs. The incremental cost per life year gained was estimated at £13385 (95% CI: £7967–27672). The incremental cost per disability-free life year gained was estimated at £23566 (95% CI: £9751–107632). Variability around the baseline estimates of cost-effectiveness for the principal outcomes is shown in the cost-effectiveness planes displayed in Figs 1 and 2. In addition, cost-effectiveness acceptability curves for the principal outcome measures are shown in Fig 3. These indicate the probabilities that neonatal ECMO is cost-effective at 7 years according to NHS willingness-to-pay thresholds for an additional life year and for an additional disability-free life year. At the notional willingness-to-pay threshold of £30000 for an additional life year, the probability that neonatal ECMO is cost-effective at 7 years was estimated at 0.98. Similarly, at the notional willingness-to-pay threshold of £30000 for an additional disability-free life year, the probability that neonatal ECMO is cost-effective at 7 years was estimated at 0.69.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Cost-effectiveness plane, expressed in terms of incremental costs and incremental life years gained.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

Cost-effectiveness plane, expressed in terms of incremental costs and incremental disability-free life years gained.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3

Cost-effectiveness acceptability curves, probability that neonatal ECMO is cost-effective after 7 years.

Sensitivity Analyses

The estimates of the costs and consequences of neonatal ECMO reported in this article were derived from a rigorously conducted randomized, controlled trial, with the uncertainty surrounding the ceiling cost-effectiveness ratio that decision-makers would consider acceptable dealt with through the use of cost-effectiveness acceptability curves. However, we also performed sensitivity analyses to determine the impact that uncertainty surrounding individual parameter values might have on the incremental cost-effectiveness for the principal outcomes (Table 4). The sensitivity analyses did not find the incremental cost-effectiveness ratios sensitive to the number of community service contacts reported by community health service providers or to the discount rate applied to future costs and health effects. However, assuming that per diem costs for each level of hospital inpatient care were 20% less (and greater) than those generated by our accounting methods had the effect of reducing (and increasing) the incremental cost per life year gained by £2303 and the incremental cost per disability-free life year gained by £4056.

Mean Net Benefits

Mean net benefits were estimated for alternative willingness-to-pay thresholds for an additional life year and for an additional disability-free life year (Table 5). Assuming that Rc equals £10000 for an additional life year generates a mean net benefit to the health services attributable to neonatal ECMO of −£5299 (ie, there is a net loss to the health services in monetary terms). Similarly, assuming that Rc equals £10000 for an additional disability-free life year generates a mean net benefit to the health services attributable to neonatal ECMO of −£11387. Increasing the value of Rc had the effect of increasing the mean net benefit to the health services of neonatal ECMO. For example, assuming that Rc equals £30000 for an additional life year results in mean net benefit of £24362. Similarly, assuming that Rc equals £30000 for an additional disability-free life year generates a mean net benefit of £5737.

Projection of Longer Term Cost-Effectiveness

When the time horizon of the economic evaluation was extended to the childhood years, the cost-effectiveness of neonatal ECMO improved substantially. On the basis of the conservative assumptions specified above, the incremental cost per life year gained falls to £7344 (95% CI: £4239–13496), and the incremental cost per disability-free life year gained falls to £11802 (95% CI: £5234–56199) over an 18-year time horizon. Assuming that Rc equals £30000 for an additional life year generates a mean net benefit to the health services attributable to neonatal ECMO of £77733 over this time horizon. Similarly, assuming that Rc equals £30000 for an additional disability-free life year generates a mean net benefit to the health services attributable to neonatal ECMO of £33772 over this time horizon.