Surgical Trial In Traumatic intraCerebral Haemorrhage (STITCH): a randomised controlled trial of Early Surgery compared with Initial Conservative Treatment

Inclusion criteria

• Adults aged 14 years or over.

• Evidence of a TICH on CT with a confluent volume of attenuation significantly raised above that of the background white and grey matter that has a total volume greater than 10 ml calculated by (length × width × height)/2 in cm. 22

• Within 24 hours of head injury. This criterion was later increased to 48 hours following discussion at an investigators’ meeting in order to increase the number of patients eligible, allowing more time for patients to reach the neurosurgery department and for the TICH to develop.

• Clinical equipoise. Only patients for whom the responsible neurosurgeon was uncertain about the benefits of either treatment were eligible.

Exclusion criteria

• A significant surface haematoma (EDH or SDH) requiring surgery. (The indications for intervention for these patients were already very well defined.)

• Three or more separate haematomas fulfilling the inclusion criteria.

• If surgery could not be performed within 12 hours of randomisation.

• Severe pre-existing physical or mental disability or severe comorbidity which would lead to a poor outcome even if the patient made a full recovery from the head injury. (Examples would be a high level of dependence before the injury or severe irreversible associated injury such as complete spinal cord injury.)

• Permanent residence outside a study country preventing follow-up.

• Patient and/or relative having a strong preference for one treatment modality.

There was no specified upper age limit. The need for clinical equipoise and explicit exclusion of patients with severe pre-existing physical or mental disability or severe comorbidity which might lead to a poor outcome even if the patient made a good recovery from the head injury excluded the less able older patient while allowing a fit older person to be included. Haematoma rates were known to be more common in the older head-injured patient.

Recruitment was encouraged by quarterly glossy newsletters sent to each centre expressing interest, monthly e-mail newsletters to the site co-ordinators and investigators, regular attendance and presentations at national and international meetings, having stands at national and international meetings, sending an e-mail to the investigators and co-ordinators whenever a patient was recruited and awarding certificates to every surgeon who recruited a patient so that they would be able to demonstrate their involvement in clinical trials.

Baseline

Before randomisation, a one-page baseline form was completed by the responsible neurosurgeon recording demographic (age, gender) and clot characteristics [site, side, length (A), width (B) and height (C) measures to define volume] and status at randomisation (GCS, pupils equal and reacting or not). This information was required in order to randomise the patient.

Two week/discharge

At 2 weeks after randomisation or at discharge or death (whichever occurred first) the discharge/2-week form was completed. This form recorded the date, the event that triggered the form and the patient’s status at that time, whether or not the patient had had surgery and when (and why if randomised to Initial Conservative Treatment or why not if randomised to Early Surgery), the patient’s GCS and localising features for the 5 days following randomisation, the occurrence of any adverse events (including death, pulmonary embolism, deep-vein thrombosis, surgical site infection) following randomisation, past medical history and status prior to the head injury. This form, together with copies of the randomisation CT image and the 5-day post-randomisation CT image, was sent to the STITCH office within 2 weeks. The data manager entered the data into the anonymised password-protected database.

Computed tomography

Copies of two CT images were required: the diagnostic CT image prior to randomisation and a 5-day image. All patients had undergone a diagnostic CT imaging as standard practice. The 5-day image was due between 3 and 7 days after randomisation. Many patients received this as part of standard treatment and the study accepted and used any CT taken for clinical purposes during this period. Only patients who did not receive such a CT during this period required additional imaging.

The preferred scanning technique was CT with volume acquisition 32 × 0.5 mm (or equivalent), 120 kV, 400 mA (or equivalent) and 220 mm field of view. The preferred angle was parallel with the anterior cranial fossa, coverage from base of skull to vertex, reconstruct 5 mm whole head, soft tissue filter.

The preferred method of sending CT images was in DICOM (Digital Imaging and Communications in Medicine)-compatible format, on separate CDs for the two time points, anonymised by use of a patient identifier. They were checked by the data manager initially on receipt at the STITCH(TRAUMA) office to ensure that the haematoma characteristics at randomisation conformed to the required inclusion criteria. Where protocol deviations were suspected, the data manager arranged for the image to be viewed by a trained reader to check whether or not the centre needed to be contacted immediately to prevent repetitions.

The data manager loaded the images into a specialised password-protected image management program. They were allocated a separate randomly created identifier by the data manager, so that it would not be possible for the reader to identify the before-and-after images of the same patient. A separate list identifying patient identifier and scan identifier was kept by the data manager.

The CT images were to be analysed subsequently by trained readers using the image management program and a defined protocol. Their passwords gave access only to scans blinded to treatment group and patient identity.

Follow-up

Postal questionnaires had previously been designed for the STICH4 and STICH II10 studies and were translated into most of the languages required. Where new countries with different languages were recruited, the National Investigator was asked to arrange translation and another principal investigator from the country was asked to check the translation. Postal follow-up was due at 6 months and 12 months. The patient’s general practitioner (in the UK) or consultant (outside the UK) was contacted at 4.5 months and 10 months to check that the patient was alive and to confirm his/her place of residence. At that time the clinician was also requested to complete a major adverse events form to include death, pulmonary embolism, deep-vein thrombosis, stroke or any other serious adverse event. The 6-month outcome questionnaire was mailed to the patient at 5 months for completion by the patient or relative if the patient was unable to complete it him- or herself. If necessary, a reminder was sent at 6 months and telephone follow-up was carried out by ‘blinded’ clerical or nursing staff at 7 months to enhance response rates. Similarly, the 12-month questionnaire was mailed at 11 months with reminders at 12 and 13 months. In countries where the postal system was poor, patients were requested to attend a follow-up clinic at which the questionnaire was distributed and collected. In countries where literacy or language/dialect was a problem, a ‘blinded’ interviewer administered the questionnaire. This methodology had been developed for use and applied to good effect in STICH and STICH II.

Costs

The costs associated with surgical treatment (theatre time, consumables, overheads) were collected from published resources and local cost surveys undertaken by the study health economist. Length of stay, health-care resource use outside hospital, together with productivity costs arising from absence from work, and additional costs for family members through extra caring responsibilities were collected using the additional 3-month postal questionnaire and extended 6-month and 12-month postal questionnaires in the UK. Data were to be combined on quality of life with survival to generate QALYs. This included measurement of health-care costs, quality of life [EQ-5D-3 level (EQ-5D-3L)] work absence [World Health Organization (WHO) Health and Performance Questionnaire – Clinical Trial Version] and carer activities (measured by the discrete choice experiment developed by the Health Economics Research Unit). EQ-5D-3L and survival were collected for all patients by the postal outcome questionnaires in order to generate QALYS for the whole study and for a UK-only analysis.

Serious adverse events

Serious adverse events were recorded on the Major Adverse Events form. Serious adverse events were defined as an adverse experience that resulted in any of the following outcomes:

• death

• life-threatening

• inpatient hospitalisation or prolongation of existing hospitalisation

• persistent or significant disability or incapacity.

All serious adverse events were to be reported to the STITCH Office within 7 days of the local investigator becoming aware of the event and to the local ethics committee or other regulatory bodies as required.

Primary outcome

Analysis was on an intention-to-treat basis. The primary analysis was a simple categorical frequency comparison using the uncorrected chi-squared test for favourable and unfavourable outcomes at 6 months. Patients were categorised as having a favourable outcome if they achieved good recovery or moderate disability on the GOS. Patients were categorised as having an unfavourable outcome if they had severe disability on the GOS, were vegetative or had died. A sensitivity analysis using logistic regression was undertaken to adjust for age, volume of haematoma and GCS.

Secondary outcomes

Secondary outcome analyses included the proportional odds model analysis of GOS, GOSE and Rankin Scale scores at 6 months, Kaplan–Meier survival curve with log-rank test, mortality at 6 months and dichotomised Rankin Scale score (0–2, 3–6) at 6 months (comparing patients able to look after their own affairs with patients who need help). For dichotomised outcomes, absolute differences and 95% confidence intervals (CIs) were reported. Secondary outcomes were also collected at 12 months but were not required by the funding agency.

Subgroup analysis

Minimal subgroup analysis was undertaken and regarded as exploratory. Odds ratios (ORs) and 95% CIs were reported for the following subgroups: age (two bands using randomisation strata: < 50 years and ≥ 50 years – there were very few patients aged over 70 years so the two upper age bands were combined), volume of haematoma (using median split: ≤ 23 ml, > 23 ml), GCS (using standard classification of head injury severe, moderate, minor: 3–8, 9–12, 13–15), time from ictus to randomisation (using median split < 21 hours, ≥ 21 hours) and geographical region (four bands: Europe, India, China, other). Interaction tests (chi-squared test for heterogeneity) were undertaken and relevant p-values reported.

Protocol violations

There were two cases of major protocol violation in that the treatment decision was taken prior to randomisation. One patient randomised to surgery had already had a decision made for no surgery to take place and one case randomised to Initial Conservative Treatment had already had surgery. These two cases will be excluded from all analyses.

Health economics analysis

A cost–utility analysis was conducted from a health service perspective and based on an intention-to-treat principle.

Health-care resource use data were sourced from individual case report forms and participant questionnaires and supplemented using additional questionnaires administered to a sample of trial recruiting centres. Standard unit cost estimates were applied to resource use data for intervention delivery, hospital length of stay and rehospitalisation. Where standard unit costs were not available, supplemental information was collected from the additional questionnaire (e.g. cost of an hour of a surgeon’s time) and WHO choice data for the cost per night in hospital. Intervention and follow-up costs were summed to generate total costs to the health services for each individual within the trial. Costing followed recommended procedures for international studies29,30 and costs were transformed into 2013 international dollars. 31 Standard general linear regression methods were used to estimate the impact of treatment group (conservative or Early Surgery) on costs. Estimates were calculated for all countries as well as country subgroups classified according World Bank classifications32 based on gross national income (GNI) per capita as follows: low income (GNI ≤ Int$1005, e.g. Nepal), lower middle income (GNI Int$1006–3975, e.g. India), upper middle income (GNI Int$3976–12,275, e.g. China), and high income (GNI ≥ Int$12,276, e.g. Western Europe, USA). Owing to small sample sizes, regression analyses were not undertaken.

Reporting statistics

For categorical variables, the number and percentage in each group are reported. Percentages are reported to no decimal places. For continuous variables, the median, quartiles, maximum and minimum are reported. Where p-values are reported, these are given to three decimal places or to one significant figure if four decimal places are required and then < 0.0001 if smaller still. The absence of data is reported. Outcomes are reported as ORs with 95% CIs and to two decimal places and p-values to three decimal places. Absolute benefits with 95% CIs to one decimal place are also reported.

Trial Steering Committee

Independent oversight of the study was provided by a Trial Steering Committee (TSC). The TSC met at 6-monthly intervals during the study. It provided overall supervision of the trial on behalf of the Health Technology Assessment (HTA) programme. It considered progress of the trial (in particular, success in site and patient recruitment), adherence to the protocol, patient safety and consideration of new information. The trial was conducted according to the standards set out in the MRC Guidelines for Good Clinical Practice.

Data Monitoring Committee

In order to monitor accumulating data on patient safety and treatment benefit, a DMC was established. The DMC considered data from interim analyses after 50, 100 and 150 patients had been recruited. It reported to the TSC. Interim analyses were strictly confidential and the committee would recommend stopping the trial early only if one or other treatment showed an advantage at a very high significance level or if the recruitment rates were unexpectedly low. The DMC recommended at each review that the trial should continue.

Management committee

This group met weekly to monitor progress and compliance. On a quarterly basis, funnel plots showing the proportion of patients who had died by the number recruited per site and the proportion of crossovers were reported to the group.

National Investigators

In countries with multiple centres, one centre investigator was required to fulfil the role of National Investigator and be responsible for obtaining national ethics approval and other permissions as required, for ensuring that documentation was translated from English as required, for identifying suitable centres within their country, for encouraging recruitment and for acting as a liaison person between the management team and the centres if required.

Centre investigators

Each centre had a centre investigator responsible for the conduct of the trial in that centre.

Patient and public involvement

There are two main charities involved with head injury research in the UK: Headway and UK Acquired Brain Injury Forum. Both organisations were invited to propose representatives to sit on the TSC. These representatives supported the study, providing advice on all aspects of the trial including design of the questionnaires, changes in the protocol and recruitment to the study.

Principal investigators

Professor A David Mendelow, MB BCh PhD FRCS.

Dr Barbara A Gregson, BSc PhD FSS.

Mr Patrick M Mitchell, BA MB BChir BSc FRCS PhD.

Professor Andy Unterberg, MD PhD.

Professor Elaine M McColl, BA MSc PhD.

Dr Iain R Chambers, BSc PhD CEng FIPEM.

Professor Paul McNamee, MA MSc PhD.

Steering committee

Mr J Steers (Independent Chairman).

Dr A Vail (Statistician).

Dr D Birchall (Neuroradiologist – Independent Member).

Mr J Timothy (Neurosurgeon – Independent Member).

Professor L Vale (Health Economist – Independent Member).

Mr A White (Lay Member – Headway).

Mr D O’Meara (Lay Member – UK Acquired Brain Forum).

Professor AD Mendelow.

Dr BA Gregson.

Mr PM Mitchell.

Dr A Unterberg.

Professor EM McColl.

Dr IR Chambers.

Dr P McNamee.

Dr E N Rowan.

Dr D Boyers.

Dr R Francis.

Data Monitoring Committee

Mr P Hutchinson (Chairman and Neurosurgeon).

Professor GD Murray (Independent Statistician).

Dr A Gholkar (Neuroradiologist).

Trial management team

Professor AD Mendelow (Chief Investigator).

Dr BA Gregson (Principal Research Associate and Trial Director).

Dr E Rowan (Senior Research Associate and Trial Manager).

Dr R Francis (Research Associate and Data Manager).

Dr D Boyers (Research Associate and Economist).

Miss H Atkinson (Trial Administrator 2009–2010).

Miss C Howe (Trial Administrator 2010–2013).

Mr P Mitchell (Neurosurgeon).

Radiological committee

Dr A Hassani.

Mr YK Yap.

Dr L Yap.

Dr A Gholkar.

Canada

Toronto, St Michael’s Hospital (0) – RL Macdonald.

Bulgaria

Sofia, University Hospital Pirogov (2) – N Gabrovsky, N Velinov.

The Czech Republic

Brno, University Hospital Brno (1) – M Smrčka, G Hanoun.

Egypt

Alexandria, Alexandria University Hospitals (3) – OS Abdelaziz, I H Zidan.

Germany

Heidelberg, Heidelberg University Hospital (2) – AW Unterberg, C Beynon.

Munich, Bogenhausen Academic Teaching Hospital, Technical University of Munich (1) – CB Lumenta, DB Schul.

Ulm, University of Ulm School of Medicine (0) – M-E Halatsch, A Pala.

Hungary

Szeged, University of Szeged (0) – P Barzo, B Fulop.

India

Bangalore, BGS Global Hospital (1) – SAV Rao, NK Venkataramanaa.

Bangalore, National Institute of Mental Health and Neuro Sciences (NIMHANS) (5) – S Somanna, KVLN Rao, J lal Gangadharan.

Calcutta, Advanced Medical Research Institute (AMRI Hospitals) (0) – RN Bhattacharya.

Chennai, Fortis Malar Hospital (1) – K Sridhar, G Venkatprasanna.

Dehradun, Himalayan Institute of Medical Sciences (13) – KK Bansal, C Gupta, R Kumar.

Lucknow, King George’s Medical University [erstwhile Chhatrapati Shahuji Maharaj (CSM) Medical University] (29) – SK Singh, C Srivastava, BK Ojha, A Chandra.

Ludhiana, Christian Medical College & Hospital (3) – SS Grewal, B Gupta.

Maharashtra, Acharya Vinoba Bhave Rural Hospital (3) – A Agrawal.

Mullana (Ambala), Maharishi Markandeshwar (MM) Institute of Medical Sciences and Research (1) – A Agrawal.

Mysore, Mysore Clinisearch (2) – A Sangli.

New Delhi, All India Institute of Medical Sciences (8) – P Sarat Chandra, BS Sharma.

Visakhapatnam, Care Hospital (8) – PV Ramana, PM Jagannath.

Latvia

Riga, Pauls Stradins Clinical University Hospital (0) – E Valeinis.

Lithuania

Kaunas, Kaunas University of Health Sciences Hospital (2) – A Tamasauskas, R Vilcinis.

Klaipeda, Klaipeda University Hospital (0) – A Gvazdaitis.

Malaysia

Malaysia, Johor Bahru, Hospital Sultanah Aminah Johor Bahru (1) – NAA Rahman, A Ali.

Kota Bharu, Kelantan, Hospital Universiti Sains Malaysia (6) – JM Abdullah, TY Chin.

Nepal

Biratnagar, Neuro Hospital (16) – YB Roka, PR Puri.

Pakistan

Peshawar, Northwest General Hospital & Research Center Peshawar (2) – T Khan, F Filza.

People’s Republic of China

Beijing, Tiantan Hospital affiliated to Capital Medical University (30) – J Zhao, L Xu, J Li.

Shanghai, Huashan Hospital, Fudan University (9) – Y Sun, J Hu.

Tianjin, Tianjin Medical University General Hospital (4) – S Yang, R Jiang.

Romani

Romania, Cluj-Napoca, Cluj County Emergency Hospital (1) – I S Florian, M Rus.

Timisoara, Emergency County Hospital Timisoara (7) – H Ples, S M Hanas.

Spain

Santander, University Hospital Marqués de Valdecilla (2) – A Vázquez-Barquero.

Valladolid, Universitario Río Hortega (1) – R Sarabia, I Arrese.

United Kingdom

Cambridge, Addenbrooke’s Hospital, Cambridge University Hospitals NHS Foundation Trust (0) – P J Kirkpatrick, A G Kolias.

Dundee, Ninewells Hospital and Medical School (2) – S Eljamel.

Haywards Heath, Hurstwood Park Neuroscience Centre (0) – G Critchley, J Norris.

Newcastle, Royal Victoria Infirmary (3) – P Bhattathiri, N Ross.

Southampton, Southampton General Hospital (1) – A Belli, D Bulters.

United States of America

Los Angeles, Los Angeles County & University of Southern California Medical Center (0) – JP Gruen.

Philadelphia, Temple University Hospital (0) – M W Weaver, F Sultan.

Portland, Legacy Emanuel Medical Center (0) – J W Chen, S Staat.

Introduction

The objective of the economic evaluation is to assess the cost-effectiveness of a strategy of Early Surgery compared with Initial Conservative Treatment in the management of traumatic ICH. This chapter reports the quality of life outcomes, resource use, costs and cost-effectiveness analyses performed alongside the STITCH(TRAUMA) randomised controlled trial, over a 6-month follow-up period.

It should be noted that the original intention of the economic component of the study was to conduct a cost–utility analysis from the UK NHS perspective, using cost and outcome data collected only on UK patients. We anticipated recruiting 150 UK patients into the trial to achieve this goal. However, UK recruitment achieved only six patients, which was not sufficient to produce a meaningful assessment of costs and outcomes for UK patients. An alternative analysis was therefore conducted, which used data from all participating centres in all countries recruiting into the trial. This approach has the advantage of being more relevant to a wider group of decision-makers. However, such an approach required the collection of additional cost data in non-UK sites, which proved challenging for some centres. Results are presented in terms of all patients recruited as a single analysis, as well as subgroup analyses based on World Bank country income group classifications (low-, lower middle-, upper middle- and high-income countries).

Resource use and costs

The main analysis focuses on results from an international perspective. However, given that the original objective of the cost-effectiveness analysis was to report from a UK perspective, we begin by reporting UK resource use for the six patients recruited into the trial.

Resource use and costs based on a UK analysis

Descriptive results are presented for the resource use and costs associated with UK participants for information only. Costs are assigned to resource use data based on 2013 health resource group (HRG) payment by results data. 1 We apply the non-elective cost associated with intracranial procedures for trauma with a diagnosis of intracranial injury (HRG code AA02). 39 Costs with (£6231 up to 43 days admission + £207 per day thereafter) and without (£4126 up to 18 days + £207 per day thereafter) complications are presented. As our data are presented in days in hospital, we calculate cost per day as the tariff value, divided through by the trim-point time. We then take the average tariff for those with and without complications. This means an overall cost per day applied to resource use in the analysis of £187 per day over the initial episode of care. Further, data from the patient questionnaire are used to estimate the number and length of hospital readmissions at 6 months’ follow-up. As the exact reason for readmission was not reported, we have assumed the same HRG codes would apply to these episodes of care also.

Resource use and costs based on international analysis

All data from the six UK trial participants are used in the costing analysis together with additional data collected from the other trial participating countries. Collection of this supplementary data required the development of a site-specific questionnaire, administered to all participating centres in the trial, to collect data on resource use and unit costs of care. The supplementary questionnaire used for the analysis is included as Appendix 5. Not all sites responded to this questionnaire which generated a substantial amount of missing data. Data were returned for 16 out of 31 sites (52%), representing 115 out of 168 (68%) patients recruited to the study. This required the imputation of cost and resource use data following some plausible assumptions, namely:

1. Where resource use and cost data were missing for some centres recruiting within a country, we have imputed weighted averages based on the number of patients recruited at centres where data are available.

2. Where resource use and cost data were missing for all centres within a country, we have pragmatically imputed weighted average data from all countries in the same income group, with country income subgroups determined according to World Bank classifications.

The impact of these data imputation methods is tested in sensitivity analyses, described in the sensitivity analyses section of this chapter.

The costing analysis is reported on the intention-to-treat principle and undertaken from an international health services perspective. The likely major drivers of costs (i.e. surgery, hospital stay and readmissions) are included. Surgery resource use (including staff time and overheads) and unit costs (e.g. surgeon’s salary and cost of theatre use) were collected using the site-specific questionnaires. Other health-care resource use data were collected, including days in intensive care, high-dependency units and general wards (all sourced from individual case report forms) and hospital readmissions (sourced from the participant 6-month questionnaire). The analysis follows recommendations from Drummond et al. 29 and Manca et al. ,30 reporting resource use and cost data at the country level. The costing of these hospital resource use data is undertaken in two stages. First, we apply country-specific unit costs for nights in hospital (intensive care unit, high-dependency unit, general ward) to resource use data to generate total costs. National average unit costs were not available for the majority of countries in the trial. Costs were therefore sourced directly from finance departments at specific sites and applied to resource use data to generate estimates of costs. Then, country-specific unit costs were transformed into international dollar costs (2013 values), using the Campbell & Cochrane Economics Methods Group purchasing power parity calculator. 31

To account for the highly skewed nature of cost data (i.e. a small proportion of patients incurring very high costs), we use GLM regression models, specifying a gamma family and identity link which best fits the distribution of the cost data. The choice of base-case model for the analysis was made on the basis of the lowest Akaike information criterion score, a method which is recommended as standard best practice. 30 Heteroskedastic robust standard errors were used for all analyses. The model estimates the impact of treatment group (Early Surgery compared with Initial Conservative Treatment) on costs adjusting for patient characteristics (age and sex). GLM models were bootstrapped using non-parametric bootstrapping techniques (n = 1000 repetitions) in Stata Version 13 (StataCorp LP, College Station, TX, USA) to generate data for developing CEACs. 31

Subgroup analyses of costs

Owing to the differences in organisation of care across countries and the value of money differences across jurisdictions, it is likely that the base-case analysis (a single analysis of all trial participants), while statistically more efficient, may have little relevance to informing decision-making at an individual country level. Results are, thus, also reported for groups of countries based on a measure of their development. For this purpose, all countries within the trial were ranked in ascending order of GNI per capita, according to World Bank classifications. The classification groups are low income, GNI per capita equal to Int$1005 or less (including Nepal); lower middle income, GNI per capita between Int$1006 and Int$3975 (including India); upper middle income, GNI per capita between Int$3976 and Int$12,275 (including China); and high income, GNI per capita of Int$12,276 or more (including most Western European countries). The logic is that such countries with similar GNI will deliver broadly comparable levels of care and reporting in this way improves the relevance of the costing for local policy-makers. It also provides an intuitive grouping of countries in the absence of enough data to conduct a country-specific analysis. This approach has been used successfully in a previous study. 40

Quality-adjusted life-years

The EQ-5D-3L generic quality of life instrument41 was administered to all trial participants at 6 months’ follow-up. Responses to the EQ-5D-3L questionnaire were valued using UK general population tariffs32 to generate a utility score for every patient within the trial. We assumed that all patients suitable for randomisation were in an unconscious state and would thus have a baseline utility of –0.402. 42 Given a lack of published tariff data across all trial recruiting countries, we have assumed that the UK tariffs offer a reasonable reflection of quality of life scores across all trial participants. This assumption is associated with limitations and assumes preferences for health states valued on the EQ-5D-3L are similar across countries. However, this approach offered the most pragmatic solution and has the advantage of applying tariffs derived from a standard method (i.e. time trade-off) to all quality of life data.

Quality-of-life data derived from the EQ-5D-3L are combined with mortality data from the trial, using the standard assumption that all participants who have died in the trial will have a utility value of 0. QALYs are then calculated on the basis of these assumptions using an area beneath the curve approach, assuming linear extrapolation of utility between time points [baseline (assumed –0.402) and 6 months’ follow-up]. Where the date of death was available, the QALY calculation has been modified to include this additional information. This introduces some asymmetries into the calculation between those who died and those who survived. The impact of assuming a linear extrapolation between time points for all patients is tested in the sensitivity analysis.

Differences in QALY estimates across groups are analysed using ordinary least squares (OLSs) standard linear regression models. Bootstrapped regressions (1000 repetitions in Stata) are conducted to account for the non-normality of QALY data and regressions are adjusted for patient characteristics, namely age and sex, and to generate data for CEACs. 43 Heteroskedastic robust standard errors are applied to all models. Subgroup analyses are presented for the utility scores and QALYs for each country income group according to that outlined in the Subgroup analyses of costs section. This facilitates the production of incremental cost-effectiveness ratios (ICERs) for each country income group.

Cost–utility analysis

The health economic evaluation is a cost–utility analysis, reporting results as incremental cost per QALY gained for Early Surgery compared with Initial Conservative Treatment. The cost per QALY is presented using the ICER, calculated as the coefficient of treatment effect on costs divided by the coefficient of treatment effect on QALYs from the respective linear regression models. Estimates of the ICER should then be compared with the normal decision-making practice in individual countries or regions.

Costs and QALY differences are presented on the cost-effectiveness plane and CEACs are derived from the net benefit statistic to illustrate the probability of Early Surgery being the most cost-effective option. All statistical analyses were conducted using Stata and Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) was used for calculation of CEACs. Owing to the short period of follow-up of only 6 months, no discounting of costs or QALYs was necessary. No extrapolation to a longer-term time horizon was conducted. Owing to the acute nature of the clinical indication, and the likely recovery time after surgery, it is likely that all patients will either have recovered or died during the trial follow-up period, with no substantial additional costs or QALYs to be accrued over a lifetime horizon.

Resource use and costs

A range of sensitivity analyses were undertaken in order to test the impact of assumptions and uncertainty surrounding resource use and cost data. Specifically:

1. We have tested the impact of variability of resource use and costs at different sites within a country. We conducted an assessment where costs for a given country were estimated on the basis of the highest and lowest cost sites within a country, with those costs applied to all sites in that country. This analysis helps to assess the variability in the organisation of care at different sites within a country and will to some extent address the impact of any differences in the private/public provision of care at different sites.

2. We have tested the impact of variability of resource use and costs at a regional level, where a region is defined by the World Bank country income classifications outlined above. The aim of this sensitivity analysis is to address the uncertainty across countries within an income group and present a plausible range of costs and ICERs for countries (including those not participating in the trial) at a regional level.

These and other assumptions used for the analysis of cost data are presented in Table 14. The table outlines the assumptions made together with justifications for each assumption and any sensitivity analyses conducted to assess the impact on costs and cost-effectiveness.

Quality of life and quality-adjusted life-years

European Quality of Life-5 Dimensions-3 level follow-up responses and death data were fully recorded for all respondents who entered the trial and therefore QALY data were complete, with no missing data needing to be imputed. Owing to a large number of deaths within the trial, we have conducted sensitivity analyses exploring the impact of alternative methods of extrapolation of the EQ-5D-3L utility data for calculation of QALYs in the trial. The base-case analysis imputes a utility score of ‘0’ from the date of death to 6-month follow-up for those who died. Survivors’ QALYs are calculated on the basis of linear interpolation between time points. This reflects the fact that there may, in theory, be a QALY benefit to dying earlier in the trial, as opposed to remaining in a health state valued worse than death for a longer period of time. However, the use of these data in this way creates an asymmetry of information between those who survived and those who died over the trial follow-up, as we have more precise information for those who died. We, therefore, conducted an alternative analysis, in which all patients accrued QALY gains using the same linear interpolation, regardless of whether they survived or died. Although this method addresses the issues of asymmetry in the information available, it does not make use of all information available to us for QALY calculation. Assumptions surrounding QALY calculations, justifications and associated sensitivity analyses conducted where appropriate are outlined in Table 15.

Analysis models of data and the impact of crossovers within the trial

In addition to the sensitivity analyses carried out on the resource use, unit cost and QALY calculations outlined in Tables 14 and 15, we conduct two further sensitivity analyses investigating the impact on results of (1) using an alternative non-parametric bootstrapped OLS regression model to account for the non-normal distribution of both cost and QALY data and (2) including interaction terms in the base-case analysis model to address the impact of crossovers on incremental costs, incremental QALYs and on cost-effectiveness.

Sampling uncertainty

Non-parametric bootstrapping techniques,44 based on 1000 repetitions of the GLM and OLS regressions were undertaken to determine the cost and QALY differences, respectively, between Early Surgery and Initial Conservative Treatment. Data from these bootstrapped regressions for cost and QALY differences were used to develop CEACs and to present scatterplots of cost and QALY differences on the cost-effectiveness plane. 44 CEACs are calculated using a net benefit approach and indicate the probability of an intervention being cost-effective at various threshold values of willingness to pay (WTP) for a QALY gain. They are especially useful when making decisions on cost-effectiveness on the balance of probabilities, and when incremental costs or effects fail to meet the traditional level of statistical significance (i.e. 95% confidence).

Resource use and costs – site-specific costing questionnaire

Full hospital resource use data were available for all patients within the trial (n = 168). These data were supplemented by unit cost information and surgical resource use data collected from the additional site questionnaires. The questionnaires were sent to all 31 recruiting sites in 13 countries worldwide. Data were returned for 16 out of 31 sites (52%), representing 115 out of 168 (68%) patients recruited to the study. Completeness of data from the questionnaires is outlined in Figure 11.

FIGURE 11.

Data completeness (supplementary costing questionnaire) for surgical costs questionnaire, by patients recruited.

Data from the site-specific questionnaires returned were used to make plausible assumptions about resource use and unit costs in other countries where data were missing. These assumptions and sensitivity analyses undertaken have been outlined in Table 14.

The results of the surgery costing exercise show wide variation within countries, for example India (Figure 12) and across countries (Figure 13).

FIGURE 12.

Within country variation of cost of surgery (example of Indian centres). Note that only two sites in India reported complete surgery costing data.

FIGURE 13.

Across country variation in surgery costs (Int$).

Figure 12 shows substantial variation in costs across the two sites which reported data for India. This is because of different public/private mixes of care at these hospitals and illustrates the impact this variability can have on the cost estimates at a country level. The variation between sites in India is the most extreme of all the countries recruiting to the trial; however, it illustrates the need to conduct sensitivity analyses for the imputation of data at all the other centres who did not contribute data to the resource use and costing questionnaire. Figure 13 illustrates the variability in costing across countries.

There is wide variability across the different countries recruiting to the trial. As expected, countries in Europe have much greater treatment costs than those in lower-income countries. However, even within country income groups, there appears to be substantial variability. The most extreme variation in country groups is between the UK and the Czech Republic, which again illustrates the importance of sensitivity analyses to assess the impact of this variation on total cost estimates for the final cost-effectiveness calculations.

Resource use and costs: – UK analysis

The results of the resource use and costs associated with the six patients recruited from the UK into the study are presented in Table 16, using descriptive statistics and are presented for information only.

The remainder of this chapter refers to the costing and cost-effectiveness analysis from an international health-care provider perspective, using data from all participants recruited into the study.

Resource use and costs: international analysis

The results of the base-case costing analysis, showing mean [standard deviation (SD)] resource use and mean (SD) costs in international dollars, based on the intention-to-treat principle are reported in Table 17.

For the whole sample, a comparison of raw mean costs shows that Early Surgery was, on average, Int$476 more costly than Initial Conservative Treatment. Using the general linear modelling estimates, with adjustment for patient characteristics of age and sex, Early Surgery is Int$1774 more costly (95% CI –Int$132 to Int$3679). The results are not significantly different between groups at the traditional 5% level of significance, but are significant at the 10% level. This suggests that there is weak evidence which suggests that Early Surgery is significantly more expensive than Initial Conservative Treatment. This perhaps indicates that patients, who are more likely to survive in the Early Surgery arm, are thus more likely to incur greater health-care costs also, through longer term treatment, and rehospitalisations as a result of their survival.

Costing subgroup analysis

Table 18 presents the cost breakdown of resource use and costs by income subgroups based on World Bank income classifications presented in the Methods section.

Sample sizes were small for country subgroups and results should be interpreted with caution. This is particularly true for low- and high-income subgroups, which recruited 30 participants. Results should therefore be treated as exploratory. Owing to the small sample sizes, there is no evidence of significant differences in costs at the country income subgroup level.

Quality-adjusted life-years

For the purposes of this analysis, we assume that all patients in the trial started with an EQ-5D-3L health state of unconscious, corresponding to a baseline utility value of –0.402, applied to all patients in the trial. Figure 14 details the results of the responses to the individual EQ-5D-3L domains at 6 months’ follow-up. Data are presented on the basis of the percentage of respondents to the questionnaire who reported any problems on any of the EQ-5D-3L domains, broken down by randomised group.

FIGURE 14.

Responses to the EQ-5D-3L.

The results show that a greater proportion of respondents in the Early Surgery group experienced at least some problems in each of the five domains of the EQ-5D-3L, when compared with respondents in the Initial Conservative Treatment group. The most notable differences were in the Usual Activities, Pain/Discomfort and Anxiety/Depression domains, in which at least 20% more respondents in the Early Surgery group had at least some problems. In contrast, the proportion of patients who had died over the course of the 6-month follow-up period was twice as high in the Initial Conservative Treatment group as in the Early Surgery group (33% vs. 15% respectively). These data suggest that as more people survive in the Early Surgery group, they continue to have some problems in their quality of life at 6 months’ follow-up.

Table 19 presents the descriptive statistics for the QALY analysis, based on the assumption of all patients commencing with a baseline utility score of –0.402, equivalent to being unconscious. Owing to the non-normal nature of the QALY data, means (SD) are presented together with median (intraquartile range) for information. Differences between groups are estimated using the bootstrapped regressions described in the Methods section.

Based on the regression analysis, incorporating date of death into the QALY calculation, and adjusting for patients’ characteristics of age and sex, we find that on average, patients randomised to the Early Surgery group had an average gain of 0.019 QALYs over a 6-month period, 95% bootstrapped CI (–0.004 to 0.043), when compared with those randomised to the Initial Conservative Treatment. This is equivalent to an incremental QALY gain of 3.5 days over a 6-month period. The broad QALY gains are driven primarily by the increased chance of survival in the Early Surgery group.

Subgroup analysis of quality-adjusted life-years data

Again applying UK-specific population weights to the quality of life scores from the EQ-5D-3L questionnaire, Table 20 presents the results by country income subgroup as classified by the World Bank country income subgroups described in the Methods section.

For all income subgroups, utility scores were, on average, higher in the Early Surgery group than in the Initial Conservative Treatment group at 6 months’ follow-up. There were negligible differences in raw mean QALYs for the low- and high-income countries, although sample sizes were very small and the regression outputs should be interpreted with caution. However, lower middle- and upper middle-income countries tended to show average QALY gains for those in the Early Surgery group. The results indicate that, although not reaching statistically significant QALY gains, patients in these groups are likely to experience QALY gains from Early Surgery. The results are promising for Early Surgery and indicate broad generalisability across the largest recruiting countries. Owing to the very small sample sizes recruited, it is not possible to draw conclusions on QALY outcomes for the low- and high-income countries.

Cost-effectiveness

The results of the base-case cost-effectiveness analysis are presented in Table 21.

For the base-case analysis, the incremental costs for the whole group together were Int$1774, with average QALY gains of 0.019. Therefore, on this basis, the base-case ICER is Int$1774/0.019 = Int$93,368 per QALY gained for Early Surgery when compared with Initial Conservative Treatment.

However, the ICER in itself should be interpreted with caution, given the range of countries which contributed data to the costing process, and also relating to alternative thresholds of cost-effectiveness which may be used by decision-makers in different jurisdictions. This complicates interpretation of the ICER. It may thus be more informative to examine the incremental costs and QALYs for individual subgroups based on their country income classification. These data are presented in Table 22. Owing to uncertainty in incremental costs and effects, we have not presented ICERs in this table. Decision-makers are instead referred to the CEAC presented in Figure 17, which illustrates the uncertainty in cost-effectiveness at the subgroup level.

As expected, there are substantial differences in costs across the subgroups. It appears that incremental costs of Early Surgery versus Initial Conservative Treatment could be decreasing for more developed countries. However, the results could equally be skewed by outliers in the data, which would have a large impact, given the small numbers. Despite substantial uncertainty in the presented results at a subgroup level, the data, on balance, suggest that favourable results could be achieved for Early Surgery in all subgroups with the exception of low-income countries.

Sensitivity analyses

A range of sensitivity analyses were undertaken focusing on the assumptions used to impute data from the site-specific questionnaire, QALY calculation methods and methods of analysis of the data. The impact of these analyses on cost-effectiveness results is outlined in Table 23. Sensitivity analyses on cost and QALY calculations are based on the assumptions outlined in the methods section and cross-referenced in the table below. Further analyses explore the impact of the model of analysis of cost data and the impact of crossovers within the study. Two ICERs are produced for each sensitivity analysis, the first based on a comparison of raw mean data across groups and the second based on the incremental costs and QALYs calculated using regression analyses described.

Owing to the small sample size in two of the income subgroups, sensitivity analyses are performed only on the whole sample of patients recruited into the trial.

The results of the sensitivity analyses show some important impacts on cost-effectiveness calculations depending on the assumptions used in the analysis. For the costing assumptions, as expected, imputing higher cost estimates increased the ICERs, and lower cost estimates reduced the ICER. Based on the range of cost imputations tested in the sensitivity analyses, the ICER ranged from Int$15,227 to Int$141,724. This serves to illustrate the impact of within-country variation in reported unit costs and resource use from the centre-specific questionnaires, and the impact of assumptions around missing data on cost estimates used for the cost-effectiveness analysis. This variation is probably driven by the differing public/private mix of care in specific countries, and especially in relation to data provided for India. The results were less sensitive to cross-country variation, within World Bank country income groups. This adds some confidence to the reliability of cross-country comparisons within income subgroups and suggests results may to some extent be generalisable to other countries within income groups.

The sensitivity analysis with one of the greatest impacts on the ICER related to the method of QALY calculation. The base-case analysis makes use of all available information, imputing a utility score of ‘0’ for those who have died, from the date of death to the end of follow-up. We chose to use the date of death for QALY estimation because, given the initial negative utility, patients could in theory attain a QALY advantage from dying earlier in the follow-up period. However, the use of date of death will create an asymmetry of information on the time point at which utility is accrued in the trial between those who have died and those who survived, as we have no comparable information for survivors, other than at their 6-month follow up. Therefore, in order to illustrate the impact of this asymmetry, we have conducted an analysis assuming a linear interpolation between time points when calculating QALYs. This effectively ignores the date of death in the calculations but addresses the asymmetry of information. Take, for example, a patient who died at 10 days after surgery. In the base-case analysis, they would have [(–0.402 + 0)/2] × (10/365)] = –0.005507 QALY, over the first 10 days + 0 QALYs between death and 6 months. For the sensitivity analysis, the equivalent calculation would be –0.1005 QALYs [(–0.402 + 0)/2] × (182.5/365)]. These results show substantial differences in QALY calculation depending on whether or not date of death is accounted for.

In addition to the base-case GLM model, we conducted exploratory analysis on the impact of model of analysis on trial outcomes. For example, running a bootstrapped OLS regression model to account for the skewed distribution of the cost data shows Early Surgery being Int$314 more costly [95% bootstrapped CI (–Int$5139 to Int$5766)]. The resultant ICER falls to Int$16,526 per QALY gained. However, this analysis should only be interpreted as exploratory and closer examination of the data confirms that the GLM model with gamma distribution is the most appropriate fit to the data. However, the analysis serves to illustrate the potential uncertainty and impact of analysis model on the costing results.

Running an alternative model as a sensitivity analysis, which includes interaction terms to address crossovers in the GLM model, suggests on average higher costs and lower quality of life for groups which have crossed over in both directions, therefore indicating that crossovers have an important impact on results. Increases in costs and deterioration in QALYs were significant at the 10% and 5% levels respectively for crossovers from initial conservative management to surgery. These results would be expected, given that crossovers to surgery were likely a result of emergency circumstances. In the model which adjusts for crossovers, being randomised to Early Surgery, and receiving the randomised allocation suggests incremental costs of Int$1021 (95% CI –Int$367 to Int$2368) compared with being randomised to and receiving conservative management.

Sampling uncertainty

In order to address sampling uncertainty in our estimates, we present the bootstrapped iterations of incremental costs and incremental outcomes on a scatterplot of the cost-effectiveness plane. These are also presented in the form of a CEAC, outlining the probability of Early Surgery intervention being a cost-effective use of scarce health-care resource use. Figures 15 and 16 illustrate the probability of cost-effectiveness for the base-case analysis.

FIGURE 15.

Scatterplot of the cost-effectiveness plane: Early Surgery vs. Initial Conservative Treatment (all randomised patients – all countries). ES, Early Surgery; ICT, Initial Conservative Treatment.

FIGURE 16.

Cost-effectiveness acceptability curve: base-case analysis. ES, Early Surgery; ICT, Initial Conservative Treatment.

The graphical illustrations presented above indicate the probability of cost-effectiveness of Early Surgery compared with Initial Conservative Treatment and illustrate the sampling uncertainty surrounding the cost-effectiveness calculations. The scatterplot shows that there is a high probability of Early Surgery delivering improved QALYs; however, there is much uncertainty surrounding the incremental cost estimates. Figure 16 shows the probability of Early Surgery being cost-effective at certain threshold values of WTP for a QALY gain. The probability of cost-effectiveness increases as WTP increases, indicating approximately 50% probability of cost-effectiveness at a threshold value of WTP for a QALY gain of Int$50,000, increasing to approximately 65% when the threshold increases to Int$100,000. However, this graph should be interpreted with caution. Conclusions on cost-effectiveness would depend on a number of issues, including (1) how reflective these costs, which are based on all recruiting countries, are of individual country circumstances and (2) what amount of money decision-makers are willing to pay in international dollars for a QALY gain in their country. In the light of these two issues, Figure 17 presents the CEACs calculated for each of the individual income subgroups recruiting into the trial and may be more informative at a local decision-making level.

FIGURE 17.

Cost-effectiveness acceptability curves for base-case and country income subgroups. ES, Early Surgery; ICT, Initial Conservative Treatment.

The data broken down by subgroup and presented above indicate wide variation in the probability of cost-effectiveness depending on the country income subgroup considered. The probability of cost-effectiveness appears to be lowest for the low-income country group. However, data for both low- and high-income groups are based on very small numbers recruited and are as such unreliable to draw firm conclusions. Higher-income (e.g. UK, Germany), upper middle-income (e.g. China) and lower middle-income (e.g. India) countries have a probability of cost-effectiveness between 70% and 80% at a Int$50,000 threshold value of WTP for a QALY gain.

Interpretation of these country income subgroup CEACs is likely to depend on the wealth of individual countries and countries in income subgroups here should draw conclusions based on the data presented for their income subgroup but also in conjunction with normal threshold values of WTP for a QALY in their jurisdiction. The WHO Choosing Interventions that are Cost-Effective (CHOICE) project has issued guidance to assist with this process, suggesting that an ICER less than three times gross domestic product (GDP) per capita might be considered cost-effective and an ICER that is less than GDP per capita would be considered highly cost-effective.

Conclusions

The cost-effectiveness analysis indicates that Early Surgery may be associated with additional QALYs and increases in health-care expenditures. However, differences in costs and QALYs do not reach statistical significance. The results of our analyses, especially in relation to costs, should be interpreted with caution, in light of the assumptions outlined in this chapter. Further research is required to determine more conclusively whether Early Surgery is more cost-effective than Initial Conservative Treatment.