11.2.1. Literature review

Using the same search strategy as for the main systematic reviews but with an additional filter to locate costing information, a search (Appendix 1) was performed of:

• Medline (PubMED)
• Embase
• Health Economic Evaluations Database (HHED) - http://www.ohe-heed.com.
• NHS Economic Evaluations Database (NHS EED) - http://nhscrd.york.ac.uk/nhsdhp.htm.

These strategies were designed to find any economic study related to head injury. Abstracts and database reviews of papers found were reviewed by the health economist and were discarded if they appeared not to contain any economic data or if the focus of the paper was not imaging after trauma. Relevant references in the bibliographies of reviewed papers were also identified and reviewed.

11.2.2. Modelling of cost-effectiveness – intracranial haematoma

A cost analysis was performed for the use of CT scanning on patients who have minor/mild head injury (that is, GCS greater than 12) but some loss of consciousness or amnesia at the time of the impact or thereafter. The reason for selecting this group is that it is assumed that those patients with a more significant loss of consciousness receive CT scanning automatically or are referred to neurosurgery. It is assumed that those who do not experience loss of consciousness or amnesia will not receive CT scanning. These assumptions mirror the methods used to derive the Canadian CT-head rule.

Four alternative strategies were selected for the model (Table 11.1). The first is an approximation of the current (pre-2003) UK system, based on skull X-ray for patients who have experienced loss of consciousness or amnesia. The second and third are the Canadian head rules, which avoid skull X-ray, but allow greater access to CT scanning. Patients with a negative CT scan would be discharged. The fourth strategy is comprehensive scanning and admission of all patients, essentially what happens in the US system.

Table 11.1

Description of different strategies for the target group.

The cost per patient for each strategy was calculated on the basis of the expected usage of skull X-ray, head CT scan and 24 hour observation. It was not possible to quantify differences in health outcomes and other cost outcomes (Table 11.2, outcomes 4–10).

Table 11.2

Health and resource consequences of Canadian CT head rule versus current (pre-2003) UK system.

Usage figures were derived from Nee et al ¹⁷⁶ for the current (pre-2003) UK system and from Stiell et al ²⁵ for the Canadian rules (Table 11.3). For the US model, usage was determined by the model definition.

Table 11.3

Proportion of target group receiving each test.

Stiell et al have not yet put their model into practice; therefore the admission rate figure is provisional. For this model it was assumed that only those with a positive CT scan (ICH or other complication) would be admitted. Another problem was that Stiell et al had already excluded patients without any loss of consciousness or amnesia, whereas the UK paper had not. This problem was tackled by assuming that patients in the UK study who were discharged without a skull X-ray or CT scan were also very low risk (that is, had no loss of consciousness or amnesia).

11.2.3. Modelling of cost-effectiveness – cervical spine injuries

We compared the cost of the two alternative strategies identified as being derived using relatively high quality methods:

• NEXUS study rule ¹²²
• Canadian cervical spine rule ⁵²

These systems evaluate all patients with head trauma, the same cohort as for the intracranial haematoma model.

The expected cost for each strategy was calculated on the basis of the expected usage of cervical spine X-ray, and cervical spine CT scan. It was not possible to quantify differences in health outcomes and other cost outcomes (Table 11.4, outcomes 3–8). Usage figures were derived from the original studies. In the case of the Canadian cervical spine rule, there has not been a validation study hence the figures are from the original derivation study. It was assumed that, for both strategies, 39% of X-rays are inadequate ¹²² and that these are followed up with a CT scan.

Table 11.4

Outcomes from cervical spine scanning.

11.2.4. Unit costs

Average unit costs for X-ray, CT scan and 24 hour observation were taken from the NHS Reference Costs 2005–6¹⁷⁷. A unit cost of 24-hour observation was estimated approximately using the median cost of an excess bed day for a ‘Head injury without significant brain injury: uncomplicated’.

Table 11.5

Unit cost estimates for the UK NHS (updated in 2007).

The NHS reference cost database contains accounting cost data from every NHS hospital trust. Each trust reports an average cost per hospital episode, categorised by type of visit (for example, out-patient, elective in-patient, etc) clinical specialty and Healthcare Resource Group (HRG). Accounting practices do vary between hospitals but the costs should reflect the full cost of the service (including direct, indirect and overhead costs), as described in the NHS Costing Manual.

Sensitivity analyses were conducted to test the sensitivity of the results to the model parameters:

• for the unit costs, the inter-quartile range was used,
• for the probabilities, the confidence intervals were used.

11.3.1. Literature review

Six studies have evaluated the overall impact of different diagnostic testing strategies for patients with minor/mild head injury. The UK studies date back to the early 1980s (pre-CT scanning) and advocate that both skull X-ray and in-patient observation be reduced to save costs.^178–180

Three overseas studies have compared CT scanning with alternative strategies. Ingebrigtsen and Romner ¹⁸¹ found that in-patient observation was not necessary with CT. Therefore CT screening was less costly than skull X-ray screening in Norway because it reduced in-patient stays. Shackford et al ¹⁸² and Stein et al ¹⁸³ had already come to the same conclusion for the USA. However, Stein et al also considered the potential use of X-ray screening without in-patient observation and not surprisingly found this to be the least costly strategy.

Essentially all three studies have concluded that a system of CT scanning high risk patients followed by discharge after a negative CT scan is less costly than skull X-ray and admission for all of these patients. However, this comparison is not strictly relevant to the context of England and Wales because the current system does not admit all patients.

The published evidence from the six studies is not ideal because:

• the resource use and cost for CT scanning is not specific to the UK NHS context or is dated; and
• they have sought to quantify and cost outcomes 1–3 only. For example, the studies did not measure the cost savings and health gain associated with early diagnosis. Stein et al suggested that for those patients who are not diagnosed early there are lost wages and increased costs relating to in-patient stay, rehabilitation, treatment, medication and orthotic devices.

Additional evidence retrieved in 2007 can be found below in 11.3.7.

11.3.2. Cost-effectiveness model – imaging of the head

Using the unit costs and frequencies of testing, the cost per patient of each strategy is shown in Table 11.6. The least cost strategy is the 5-point Canadian CT Head rule. Although the cost of CT scanning is higher than for the current (pre-2003) UK system, the extra cost is more than offset by the reduction in skull X-rays and admissions.

Table 11.6

Cost per patient for each strategy.

Both Canadian rules could save the NHS money. It would require investment in additional CT scanning facilities but these costs would, be offset by the freeing up of ward space and X-ray capacity.

These results were largely insensitive to the unit costs and probabilities used (Table 11.7). Only when both costs and probabilities were set to favour the current (pre-2003) UK system was the Canadian seven point rule more costly.

Table 11.7

Sensitivity analysis for head CT scanning rules.

This cost analysis was limited because the frequency of testing and admission for each strategy could only be estimated approximately given the currently available data. The Canadian head rule is less costly than the current (pre-2003) UK system because it is assumed that it reduces the number of admissions. In fact Stiell et al ²⁵ have not yet put their model into practice, therefore the admission rate figure is provisional. For this model it was assumed that only those with a positive CT scan (ICH or other complication) would be admitted. If the number of admissions were somewhat higher then this strategy would not be the least cost strategy. Assuming all other parameters in the model remain the same, the five point Canadian head rule is least cost if it reduces in-patient admissions by at least 37%. The seven point Canadian head rule appears to be more expensive even if admissions were entirely eliminated.

Another model parameter which was estimated very approximately was the level of CT use in the current system, because CT scanning use was lower during the Nee et al (1993) study than in the present UK system.

The sensitivity of the results to these particular assumptions is presented in a two-way sensitivity analysis (Table 11.8).

Table 11.8

Additional cost per patient (£) - Canadian seven point rule compared with current (pre-2003) UK system - two-way sensitivity analysis. (Updated 2007).

Another problem was that the study that presented data on the Canadian rules had already excluded patients without loss of consciousness or amnesia, whereas the UK paper had not – this problem was tackled by assuming that patients who were discharged did not receive a skull X-ray. Furthermore the analysis did not include outcomes 4–10 from Table 11.2.

Evidence retrieved in 2007 provides real data on the impact of the Canadian head CT rule on the NHS - see below in 11.3.7.

11.3.3. Health outcomes (4 and 5, see Table 11.2)

A strategy that increases NHS costs would be economically justified if there were associated health gains. Intuitively, we might expect surgical outcomes to improve if intracranial haematomas (ICHs) are detected earlier. There is no direct evidence that a strategy of CT scanning can improve neurosurgical outcomes although there is some evidence that outcomes have been improved in patients with more serious head injuries.¹⁸⁴

UPDATE 2007

However, there is cohort study evidence suggesting reduced mortality associated with prompt surgery^185,186. A paper retrieved during the 2007 update⁷⁶ had estimated the quality-adjusted life-years (QALYs) gained from prompt surgery by comparing the recovery and mortality rates in different case series (see 11.3.7 below).

Any health gains associated with detection could be partially offset by increased cancer risk. There is no direct evidence that exposure to medical X-rays does increase the incidence of cancer, however, there is a general association between radiation and genetic mutation and it is clear that the exposure level is considerably higher with CT scanning than with skull X-ray (see Chapter 10).

11.3.4. Other health service costs (6, see Table 11.2)

The change in health outcomes just mentioned would lead to considerable changes in health service resource use for the particular patients affected. However in both cases the net change in health service costs could go up or down. For example, if an improvement in neurosurgical outcome leads to more patients surviving but those that survive require long term care for chronic brain injury then costs would increase. Alternatively if both mortality and disability were reduced then long term costs are likely to be reduced. However, whichever direction the change is in, the average change in costs per patient scanned is likely to be small given the low likelihood of a change in health outcome.

11.3.5. Patient costs (7&8, see Table 11.2)

The costs (time, lost income, medication purchased, etc) to patients and their families associated with changes in health outcome could be considerable. As with health service costs we could not be certain what the net effect would be for the family. Again when averaged across all patients these cost changes could be quite small because the incidence of these changes in outcomes will be small.

There may be substantial costs associated with the decision to admit but these are likely to differ according to the situation of the family. For example, if a parent is admitted then there might be a need for child-minders but on the other hand the act of regular observation at home is costly in itself and families might find it easier if this burden were undertaken by the hospital.

11.3.6. Litigation costs (9, see Table 11.2)

It has been suggested that litigation might be reduced if more patients were scanned. However, Bramley et al ¹⁸⁷ have estimated that only one in 10,000 patients subsequently turn out to have an intracranial haematoma after being discharged without a CT. Therefore the potential costs saved per patient screened are likely to be small. It should also be born in mind that successful litigation usually arises out of organisations not abiding by guidelines.

11.3.7. Update 2007

We found three new studies that evaluated diagnostic tools: a decision analysis¹⁸⁸ and an RCT⁷⁹ were comparing admission with CT scanning, and a case series¹⁸⁹ was evaluating the use of head MRI as an addition to CT.

A further three new studies evaluated diagnostic decision rules. We found two studies evaluating the implementation of the head CT rule recommended in the original edition of this guideline. A third study compared the Canadian Head CT Rule with various imaging strategies.

A decision analysis¹⁸⁸ compared CT scanning (and discharge after a negative scan) with admission in head injury patients with a GCS of 15 (mild head injury). They found the CT strategy to be cost saving compared with admission. The same team confirmed the results of this study with a randomised controlled trial of 2600 mild head injury patients⁷⁹1. Outcomes were followed up for three months. There were no differences in clinical outcomes (survival and extended Glasgow Outcome scale GOS) but costs were £133 less per patient in the CT arm.

A retrospective case series of 40 patients¹⁸⁹ was used to evaluate the addition of an MRI to CT scanning in patients with traumatic brain injury. The number of lesions diagnosed by CT but not by MRI was 9 out of 40, while the lesions detected by MRI but not by CT were 24 out of 40. The addition of MRI cost more than £1,500 in additional charges per extra lesion diagnosed. However the identification of the additional lesions did not lead to a change in the treatment path and therefore the addition of MRI to CT was neither effective nor cost-effective. However, the cohort was small for estimating the effectiveness with any precision.

A UK cohort study¹⁶ evaluated the consequences of implementing the NICE guideline. The X-ray and admission-based practice was replaced with the Canadian CT head rule. Cases of head injury were followed up in a regional neurosciences hospital and in a district general hospital for one month, six months before and for one month after the guideline implementation. In the case of the neurosciences hospital the cost per patient was reduced by £34 and it was reduced by £3 per patient at the general hospital. In contrast in a similar cohort study⁸⁸ of 992 patients, costs were found to increase by £77 per patient. Table 1 shows the resource use observed in both studies compared with the predictions in the original edition of this guideline. The evidence from the cohorts suggests that compared with our predictions there was a more modest increase in CT and a more modest decrease in X-ray.

The variation in impact between centres could be due to a number of factors including variation in the baseline position and completeness of adherence to the NICE guideline in the after period of the studies. In the centre that showed an increase in cost, X-rays were very low in number to start with and therefore there was less scope for cost savings; furthermore admissions had inexplicably increased slightly compared with the reductions seen at the other centres. The large amount of variation between centres means that the impact of our recommendations at a national level remains uncertain.

One of the centres in the Hassan study¹⁶ had modified the protocol so that elderly patients with a GCS of 15 seen out of hours could be admitted instead of getting urgent CT. The reasoning involves a combination of factors: a) the cost of out-of-hours radiology was relatively high, b) the elderly represent quite a large group and there are often difficulties in trying to discharge them over night. Hence, the modification is lower cost since out-of-hours radiology is avoided and most would needed admission anyway. We don’t have evidence of effectiveness for this specific patient group but the randomised evidence for the general population showed no difference in outcomes between observation and CT scan⁷⁹. The GDG agreed that this was an acceptable deviation from the head rule and the guideline recommendations were modified accordingly.

A decision analysis⁷⁶ compared the Canadian head CT rule with several strategies including ‘CT all’, ‘admit all’, ‘discharge all’ and ‘X-ray all’ in a US context. Quality-adjusted life-years (QALYs) and costs were estimated for both prompt and delayed surgery by comparing the mortality and recovery rates in different case series. The Canadian rule dominated the other strategies, that is to say it gave the highest number of QALYs and the lowest cost. However, the study did not evaluate the earlier UK guidelines based on skull X-ray and admission. The CT all strategy was just as clinically effective but more costly. The results were sensitive to the probability that prompt surgery leads to a good outcome.

11.4.1. Literature review

There are three cost-effectiveness studies in this area:

• Kaneriya et al ¹⁹⁰ estimated that five view X-ray could save $24 per patient scanned compared with three-view because it reduced the number of subsequent CTs associated with inadequate X-rays by 48%.
• Tan et al ¹⁹¹ estimated the cost-effectiveness of CT scan after inadequate X-ray. They found a cost of $16,900 per potentially (or definitely) unstable fracture and $50,600 per definitely unstable fracture. This is cost-effective given the consequences of paralysis.
• Blackmore et al ¹²¹, using test sensitivities pooled from the published literature, compared CT scanning of the cervical spine with conventional cervical spine X-ray. Using their own risk rating scale, they found CT scanning to be a cost-effective strategy ($16,000 per quality-adjusted life-year gained) for the ‘high’ and ‘moderate’ risk groups (high energy mechanism and age under 50 or moderate energy mechanism and age greater than 50) but not for the low risk group ($84,000 per QALY gained). Unlike the other studies, incorporated into these figures are the costs and morbidity associated with paralysis.
• In addition, two more studies estimated the costs that could be saved by moving from current practice at a particular institution to a particular scanning protocol.^122,192

The above studies are not strictly relevant to the context of England and Wales, not least because the unit costs and the patient groups used in the studies are not from the UK. Furthermore they only attempted to include outcomes 1 and 2 (and in the case of Blackmore et al 4 and 6 as well) and crucially do not address the long term effects of medical radiation, which are likely to be greater in CT scanning of the neck than in CT scanning of the head (see Chapter 10).

The Blackmore analysis suggests for a patient group that is at particularly high risk of paralysis, cervical spine CT could be preferable to X-ray by both improving health outcomes and lowering costs. However, they do not take into account the impact of the large radiation dose received by the thyroid from a cervical spine CT scan. This would be very difficult to model given the lack of empirical evidence on the long term effects of this medical radiation. It was the consensus of the Guideline Development Group that the benefits from CT scanning of the cervical spine do not obviously outweigh the risks.

In light of the review of new clinical and cost-effectiveness evidence, the GDG modified its position to recommend CT scanning in high risk patients. Additional cost-effectiveness evidence retrieved in 2007 can be found below in 11.4.3.

11.4.2. Cost-effectiveness model – imaging of the cervical spine

We conducted our own tentative cost analysis comparing the NEXUS and the Canadian cervical spine rules. We estimated that the Canadian rule could save about £14 per patient (Table 11.10).

Table 11.10

Comparison of the Canadian and NEXUS cervical spine rules (Updated 2007).

The assumption that a CT scan will be performed after all inadequate X-rays may over-estimate the actual cost savings; if we omit them then the cost-savings are £4 per patient scanned. Sensitivity ranges are presented in Table 11.11.

Table 11.11

Sensitivity analysis for cervical spine scanning rules.

The Canadian cervical spine rule could save valuable health service resources but it is yet to be validated and if it was found to be less sensitive it might not be the most cost-effective strategy due to the morbidity and high costs associated with paralysis. This cost analysis was limited because of the use of overseas data and the simplified assumptions regarding dealing with inadequate X-rays. Furthermore the analysis did not include outcomes 3–8 from Table 11.4.

11.4.3. Update 2007

Five new studies were found: a non-randomised controlled trial¹¹⁷, two cohort studies^118,193, a case series¹¹⁹ and a decision model¹²⁰. One study¹⁹³ was evaluating the role of MRI scanning in children, another study ¹¹⁷ was comparing helical CT scanning with X-ray in children, and the rest were comparing CT scanning with X-ray in adults.

A non-RCT ¹¹⁷ compared the costs of helical CT with those of X-ray in a population of 136 children who required cervical spine radiography in addition to cranial CT. The imaging costs including follow-up tests were £100 and £130 respectively for the radiography and CT diagnostic strategies (significance not reported).

A retrospective cohort study ¹¹⁸ based on an adult population of 573 trauma patients undergoing spinal imaging (the proportion with head injury was not reported) compared the costs of helical CT with X-ray. Unlike the non-RCT, this study found the cost of CT was no greater than X-ray (£36 vs £35) due to the staff time involved with CT being substantially less.

In a case series study ¹¹⁹, 407 adult patients in a trauma centre underwent both X-ray and helical CT (again the proportion with head injury was not reported). The reference standard was represented by two radiologists independently reviewing both the HCT and plain X-ray results together with hospital case notes. The sensitivity yielded by X-ray was 45% while the sensitivity yielded by the helical CT intervention was 98%. The helical CT strategy was more costly than a strategy of helical CT after inadequate X-ray. From their figures, we calculate that this strategy costs an extra £7,300 per fracture detected. Using the model by Blackmore and colleagues¹²¹, as follows, we can see that this is highly cost-effective. The model estimated that 5% of fractures would lead to paralysis and that paralysis is associated with 16 QALYs lost. Hence £7,300 per fracture detected would translate to only £9,125 per QALY gained and that is without taking in to account the considerable cost savings from averting paralysis.

The decision analysis of helical CT vs X-ray of the cervical spine in patients undergoing cranial CT for head injury by Grogan et al¹²⁰ was based on an earlier model by Blackmore and colleagues¹²¹ looking at conventional CT vs X-ray. It considered only patients at medium and high risk:

• Focal neuro-deficit or severe head injury or high energy impact, or
• Moderate energy impact and age more than 50

Helical CT cost an additional £37,000 per paralysis averted in this group. This would imply that the helical CT strategy is cost saving when the very high cost of treating paralysis is taken into account.

A retrospective cohort study with a historical control published in 2002 ¹⁹³ evaluated a protocol of MRI scanning patients whose cervical spine had not been cleared within 72 hours. The control strategy was not clearly defined. This study was conducted in a specific population of patients consisting of 102 children (age 0 to 17) who were intubated at the time of hospital admission and who remained in the intensive care unit for at least 3 days. Among the 51 patients in the control group, 19 underwent MRI, whereas it was required for 31 patients in the post-protocol group.

The MRI group had reduced hospital charges (£18,000 vs £24,000; significance not reported) attributable to reduced stay in hospital and in intensive care. However, sample variation and a general trend over time towards reduced stay might explain this difference.

11.5.1. Conclusions from the 2007 update

A randomised controlled trial has confirmed that to discharge patients with mild head injury (GCS15) after a negative CT scan, as recommended in this guideline, is both safe and cost saving.

The impact of the Canadian CT rule as advocated in the original edition of this guideline has varied considerably but reassuringly in some centres it has reduced costs. A published model that took into account long term treatment costs and health consequences indicated that the Canadian head CT rule is more cost-effective than a number of alternative strategies based on CT, X-ray or admission. However, none of the evidence has taken into account the impact of the increased radiation exposure.

Updating the costs to 2005–6 prices makes the Canadian CT head rule even more cost-effective, since the cost of imaging has fallen.

A modification of the rule so that elderly patients with a GCS of 15 seen out of hours could be admitted instead of getting urgent CT is a safe strategy and could be cost saving for services where out of hours radiography costs are prohibitively high.

The new studies add to existing evidence, in suggesting that CT scanning of the cervical spine is cost-effective in higher risk groups who are already undergoing head CT. However, none of these studies have taken into account the costs and health consequences associated with the increased radiation exposure – it is possible that CT is no longer cost-effective when these are taken into account. It is difficult to model the impact of radiation exposure on cost-effectiveness since there are a large number of uncertainties: a) the amount of radiation received at different parts of the body, b) the relationship between exposure and cancer, c) the types of cancer caused, d) the pattern of resource use in the diagnosis and treatment of the cancer, and e) the timing of cancer, treatment and death. Another limitation with regard to cervical spine imaging is that all the studies were conducted in the USA; the observed healthcare costs and savings might not be transferable to a UK NHS setting. As the cost of CT scanning, as with most medical care, is lower in the UK, if it is cost-effective in the USA then it is likely to be cost-effective for the NHS. However, the cost savings from paralysis care averted are also likely to be lower.

11.6.1. Literature review

We did not find any cost-effectiveness evidence for this question but we did find two simulation models, which we will refer to as the London and Staffordshire models. We have reviewed these models in some detail, as follows.

11.6.2. London model

The report¹⁹⁶ summarises the findings of a review conducted by the London Severe Injury Working Group focusing on the Trauma services provided in London, including care, treatment and transfer of severely injured patients. Severe injury was defined as the need for Intensive Care.

The analysis of the current service highlights some key issues:

• high secondary referral rate (two thirds of the severely injured patients group),
• evidence of problems associated with such transfers (adverse clinical events during transfer, delay to definitive intervention, low level of staff and standard of care), and
• difficulties for hospitals in transferring patients for specialist care, especially for neurosurgery (stabilisation of patient first, co-ordination between the first hospital and the specialist hospital and consequent long delays).

Methods

A modelling of the flow of trauma patients was carried out to determine the best trauma service configuration for adult trauma patients with severe injury in the London area. The model was designed to estimate the time from injury to:

• Critical Intervention (urgent life saving interventions such as intubation); these interventions are crucial for all trauma patients
• Definitive Intervention (specialist interventions such as neurosurgery); these interventions vary according to the site of the trauma

The specific aims of the modelling exercise were to evaluate the effect on time to intervention of:

1. different bypass strategies
2. improving the current system by reducing time taken in pre-hospital and in-hospital trauma management.
3. a doctor in the pre-hospital phase provided by the London Helicopter Emergency Medical Service (HEMS).

The model simulated results based on about 10,000 actual severe injuries from the London region. Of these 33% had isolated head injury and a further 18% had non-isolated head injury.

The model estimates time to intervention using flow charts. Figure 1 shows the flowchart for an isolated head injury patient with the average times based on current practice. Similar flowcharts were devised for the different types of trauma. The timings were based on ambulance service records and expert opinion.

Figure 1

London Model flowchart for isolated head injury patients (figures in parentheses are average time in minutes).

For each type of injury, a group of clinical experts decided on a target time for intervention. For head injury, it was considered that it was crucial to carry out neurosurgery within 4 hours of the injury, based on some evidence¹⁸⁶. For each service configuration scenario, the primary outcomes were:

• the median times to critical and definitive interventions.
• the proportion of patients receiving critical and definitive interventions within the relevant time target.

Model Results

11.13 shows the median time to critical/definitive intervention by type of injury and by bypass strategy used. On the left side of the table the results are based on current timings. On the right hand side the results are based on improved timings. In the case of the isolated head injury patient the median time to neurosurgery is 4.8 hours currently but would fall to 3.4 hours when bypassing patients who are less than 20 minutes from a specialist centre. Table 11.14 shows the proportion of patients that receive interventions within the target time. In the case of the isolated head injury patient the number receiving neurosurgery within 4 hours would increase from 23% with no bypass to 74% with bypassing patients who are less than 20 minutes from a specialist centre. However, on the negative side with this bypass strategy only 84% (compared with 91%) would receive critical intervention within 60 minutes. The group that is made worse off by bypass is those patients with isolated orthopaedic injury: only 25% would receive their definitive intervention within their 2 hour target (compared with 30% without bypass).

Table 11.13

London Model: Median time (hours) to critical/definitive interventions, by bypass strategy.

Table 11.14

London Model: Proportion of patients receiving critical/definitive interventions within target time, by bypass strategy.

For the injuries that can be treated in every hospital the most rapid movement to Definitive Intervention was achieved by the models without bypass, and with improvement in hospital times.

For injuries requiring specialist management the best models for providing early Definitive Intervention included 20 minutes bypass, improvement in hospital times and use of the London HEMS.

11.6.3. Staffordshire model

The link between time and health outcomes missed by the London model was captured to some extent in the Staffordshire model⁶⁸.

It evaluated the impact of 10 different transport strategies on survival of patients with serious or worse HI (AIS more than 2). In the model, survival was determined by a number of variables including: a) head AIS score, b) non-head AIS score, c) time to surgery, d) grade of staff during transfer, e) incidence of hypoxia and hypotension, g) distance from hospitals. Some of these variables are patient-specific (a,b,g), some are service-specific (d) and some are determined by the transport strategy (c,e). The data used in the model came from a variety of sources including a large trauma database, the published literature and expert opinion. Monte Carlo simulation (that is repeatedly generating new results by simultaneously drawing at random from the distribution of each model parameter) was used to simulate 10,000 head injury patients and their outcomes under each strategy.

Table 11.15 shows the results for each strategy. All direct transport strategies had higher expected survival than a strategy of sending all patients to the nearest emergency department but strategies 2–6 were the most effective. Among these strategies, strategy 4 (direct transport of patients with critical head injury, AIS=5) required the least number of patients being diverted to specialist centres. The results were not sensitive to the parameters that were determined by expert opinion.

Table 11.15

Stevenson’s Transport model - results.

An important limitation that was acknowledged by the authors was that AIS score is determined after treatment and therefore assessment of patients at the scene of the injury is less accurate. The implication is that the survival gain observed in this model is probably larger than can be achieved in reality, although the pattern should be the same. There are different costs associated with each strategy and therefore a cost-effectiveness analysis is needed to assess which of the 10 strategies is the most cost-effective.

In conclusion, the simulation study shows that survival of severe head injury patients could be substantially improved by transporting patients directly from the injury scene to a hospital with a specialist neurosciences centre. Cost-effectiveness of these strategies was determined as described in 11.6.4.

11.6.4. Cost-effectiveness model – Direct transport

We conducted a cost-effectiveness analysis of transporting patients with serious head injury directly from the injury scene to a specialist neurosciences hospital (NSH). This was compared to initially transporting such patients to the nearest emergency department and then later transferring them to the NSH after stabilising the patient.

The following general principles were adhered to:

• The GDG was consulted during the construction and interpretation of the models.
• The sources of data are published studies and expert opinion.
• Model assumptions were reported fully and transparently.
• The results were subject to sensitivity analysis and limitations were discussed.
• We followed the methods of the NICE reference case. Therefore costs were calculated from a health services perspective. Health gain was measured in terms of quality-adjusted life-years (QALYs) gained.

11.6.4.1. General method

The model is represented by a decision tree (Fig. 2): once the ambulance crews arrive at the accident scene, the patient can be transported either to the nearest District General Hospital (DGH) or to a Neurosciences Hospital (NSH). Severe head injury patients initially admitted to the DGH will be subsequently referred to the NSH. Patients that survive will require rehabilitation and frequently some kind of long term care. The number of survivors is different in the different strategies.

Fig. 2

Transport model decision tree.

To assess the cost-effectiveness of direct transport we need to assess not just changes to ambulance and emergency department costs associated with each strategy but also any changes in rehabilitation and long term care costs arising from the different strategies. These have to be balanced against the health gain.

We could not find evidence of effectiveness that perfectly suits this question. We therefore constructed two similar models based on different empirical studies:

Model A: We based this model on the only study in the clinical literature review that reported both mortality and health status (Glasgow Outcome Scale, GOS) in head injury patients– Poon et al 1991¹³⁵. This study compared a cohort of patients that had been directly transported to NSH to another cohort that were transferred from DGH. This study allows us to estimate both the QALYs gained and the cost savings attributable to improved care status in patients being directly transported. However, there was concern that this study was biased, since case-mix was not properly controlled for. For this reason we developed a more conservative model.

Model B, a conservative model, calculates only the health gain attributable to those patients who survive with direct transport but would not survive with a secondary transfer strategy. The number of these extra survivors is estimated using the results of a decision model that was explicitly answering our question – Stevenson et al 2001⁶⁸ (see 11.6.3). Model B does not take into account health gain for patients who survive under both strategies but have an improved health status with the direct transport strategy.

Each model has advantages and limitations (Table 11.16).

Table 11.16

Summary of the models.

For each strategy in both models, the expected healthcare costs and the expected QALYs were calculated by estimating the costs and QALYs for each GOS state and then multiplying them by the proportion of patients that would be in that state as determined by the strategy taken. Health state defined by the GOS state was assumed to be fixed over the lifetime.

The base case models assume that only patients with serious head injury would be transported. A concern is the ability of ambulance crews to determine the severity of the head injury at the scene. There might be a risk of overestimating the number of severely injured patients and therefore of sending too many patients to the NSH, which would mean that cost-effectiveness is reduced and would be risky for patients with multiple trauma. For this purpose, we conducted a sensitivity analysis on the number of false positives (patients erroneously deemed having a serious head injury) that would be transported to the specialist centre without requiring neurosurgical care.

11.6.4.2. Methods: Effectiveness

In Model A, the mortality rate together with the outcomes were derived from a study by Poon at al ¹³⁵ in which a group of patients having an extradural haematoma was directly transported to the NSH while another group was only secondarily transferred there (Table 11.17). The mortality and the outcomes were assessed six months after the injury.

Table 11.17

GOS score and death rate after neurosurgical care in a NSH (Model A).

The survival gain in Model B was derived from the results of a simulation model by Stevenson et al⁶⁸, where the target patient population were adults with a serious head injury (AIS of 3 or more) – see 11.6.3.

The model evaluated 10 different strategies of transporting patients directly to the NSH, which selected patients by different criteria (relating to level of AIS score, presence of multiple injuries, possibility of pre-hospital intubation, out of hours). Directly transporting all serious head injury patients to the NSH led to an estimated increase in survival of 4.5% for injury scenes near to the NSH and 3.4% for more distant injury scenes.

Stevenson et al estimated only mortality and not health status. We assumed that health status in the additional survivors would be similar to the general population of patients with serious head injury treated in a NSH. We used 6-month GOS data from the surviving patients in a UK study, Patel 2002¹⁹⁷ (Table 11.18). The study population had all had a severe head injury (GCS 8 or less) and had been treated in a Neurosciences Critical Care Unit.

Table 11.18

GOS score after neurosurgical care in a NSH (Model B).

We estimated the health loss associated with false positives. In fact, for these patients the longer the journey from the accident scene to the hospital, the higher is the risk of death from hypotension. In the case of a distant NSH (53 minutes, as reported in Stevenson’s model), the mortality increases by 0.05%, while it increases by 0.03% if the NSH is near (20 minutes). These figures derived from the calculation of the probability of death based on clinical estimates (see 11.6.4.7).

11.6.4.3. Methods: Estimating QALYs

For each health state we estimated QALYs (Quality-Adjusted Life Years) by multiplying the discounted life expectancy by the utility score associated with each state. The expected QALYs for each strategy are then estimated by summing up the QALYs for each state weighted by the proportion of patients in that state.

In order to calculate the QALYs we combined data on life expectancy with data on quality of life.

Life expectancy

The life expectancy of patients in a vegetative state (VS) was assumed to be 10 years ^198,199. In the case of a 60 year old patient in a VS, the life expectancy would be shorter and was assumed to be the same as for a patient in the severe disability state (see below).

To calculate the life expectancy for health states other than VS, we applied the standardised mortality rate (SMR), reported for 2,320 traumatic brain injured patients in Shavelle 2001 ²⁰⁰, to the general population of England and Wales, using the Life Tables. According to Shavelle, the SMR decreases during the first 4 years post-injury but remains constant afterwards. In Shavelle 2001 the SMR was distinguished according to three levels of ambulation: a) none, b) some, c) stairs, which we matched approximately to the levels of disability of the GOS (a=SD, b=MD and c=GR).

Life expectancy was discounted at a rate of 3.5% per year, as required by NICE.

For our base case analysis we estimated life expectancy for men aged 40 (the average age of a patient in the Stevenson study). For our sensitivity analysis, we also calculated life-years for patients aged 20 and 60.

Quality of life

The utility scores in Table 11.19 are a measure of the quality of life associated with each of the health states on a scale from 0 (death) to 1 (perfect health). For the good recovery (GR) outcome, we used the EQ-5D score of 0.83 reported for the United Kingdom population ²⁰¹. The other utility scores were taken from a decision analysis, Aoki 1998 ²⁰². The mean utilities for each GOS score were elicited from a sample of 140 subjects with a clinical background using the standard-gamble method. The GOS states in this study were expressed as the degree of disability due to brain damage caused by subarachnoid haemorrhage.

Table 11.19

Health Utilities by Glasgow Outcome Scale (GOS) state.

The Poon et al study (Model A) did not distinguish between patients that were severely disabled (SD) and those that were moderately disabled (MD). For these patients we used the simple average of the two SMRs and the simple average of the two utilities.

Another study was found, Tsauo 1999²⁰³, which reported the utility scores associated with each GOS score obtained from health professionals in the UK using the standard gamble method. We did not use this study in our base case model for the following reasons:

-

scores were presented for a number of time points and there seemed to be inconsistency between the estimates

-

the figures were skewed towards high values (i.e. the utility associated with a moderate disability was higher than the average EQ5D utility score for the general population in the UK²⁰¹)

-

the value for the vegetative state was missing

-

the number of the health professionals interviewed for the elicitation of the utility scores was not reported.

Therefore, we used this study only for the purpose of sensitivity analysis.

In the sensitivity analysis on the assessment at the scene, we assumed that the false positives, if they survive the longer transport, would have had the same expected QALYs as the good recovery (GR) patient.

Calculating QALYs gained

For Model A, the QALYs gained are calculated as follows:

QALYs gained= Q₁-Q₀

Q_i = (Pⁱ_GR x LE_GR x U_GR) + (Pⁱ_D x LE_D x U_D)

where

Q_i =the expected QALYs per patient (i=1: with bypass, i=0: without bypass)

Pⁱ_GR, Pⁱ_D, = proportion of patients in each of the GOS states at 6 months by strategy (where D is both mild disability and severe disability combined).

LE_GR, LE_D, = the discounted life expectancy of patients by GOS states at 6 months

U_GR, U_D, = the utility score for each GOS state.

For Model B, the QALYs gained are calculated as follows:

QALYs gained=Q_i-Q₀ = ES_i x ( ( P_GR x LE_GR x U_GR) + ( P_MD x LE_MD x U_MD) + ( P_SD X LE_SD x U_SD) + ( P_vs x LE_VS x U_VS) )

where

Q_i =the expected QALYs per patient associated with bypass strategy i,

Q₀ = the expected QALYs per patient associated with no bypass,

ES_i = extra survivors=the proportion of patients surviving under strategy i that would not have survived under the no bypass strategy

P_GR, P_MD, P_SD, P_VS, = the proportion of extra survivors in each of the GOS states at 6 months

LE_GR, LE_MD, LE_SD, LE_VS, = the discounted life expectancy of patients by GOS states at 6 months

U_GR, U_MD, U_SD, U_VS, = the utility score for each GOS state.

11.6.4.5. Methods: Rehabilitation and care costs

We derived the cost of rehabilitation from two UK studies: one, Wood 1999¹⁴⁷, applicable to the severely disable patients and the other one, Nyein 1999²⁰⁵, applicable to the moderately disabled patients (Table 11.20). The length of rehabilitation for the severely disabled group was 14 months, while it was 75 days for the moderately disabled group. We assumed patients who had a good recovery to undergo the same intensity of rehabilitation as the moderately disabled group, given the fact that the good outcome was assessed six months post-injury. Patients in a vegetative state were assumed not to receive any specific rehabilitative therapy. If any rehabilitation service was provided to them, its cost was assumed to be incorporated in to the cost of long term care.

Table 11.20

Cost of rehabilitation and long term care.

The same two UK studies were used to calculate the annual care costs (Tab.11.20); in the case of severely disabled patients, the long term care was the community care support required after rehabilitation and it was based on the cost of a support worker. Similarly, the long term annual cost for the moderate disability group was calculated from the weekly cost of care three months after discharge from the rehabilitation. Patients having a good recovery were assumed not to incur any long term costs. Patients in a vegetative state were assumed to have the same annual care costs as those who are in the severe disability state.

Care costs were discounted at a rate of 3.5% per year, as required by NICE.

Thus the model takes into account the increased costs of rehabilitation and care due to people surviving under direct transport, who would not survive under the current system. It could be that costs of neurosurgery and intensive care are also increased if patients are now making it to the NSH who would have died in transit. Since we do not have data on the timing of deaths, we have not included such costs in the base case. However, for a sensitivity analysis we added on the cost of 3 days of level 3 neurosurgical intensive care for each additional survivor. The costs of care in an ICU were calculated from the NHS Reference Costs 2005–2006¹⁷⁷ at £1,338 per day.

Calculating incremental cost

For Model A the incremental cost is calculated as follows:

Incremental cost = Cost_NSU - Cost_DGH

Cost_NSU = (p^NSU_GR x (RH_GR + LE_GR x ACC_GR)) + p^NSU_D x (RH_D + LE_D x ACC_D))

Cost_DGH = (p^DGH_GR x (RH_GR + (LE_GR x ACC_GR))) + (p^DGH_D x (RH_D + (LE_D x ACC_D))) + TC

where

Cost_NSU = the expected cost per patient associated with direct transport to the NSU

Cost_DGH = the expected cost per patient associated with a secondary referral to the NSU from a DGH

P^NSU_GR, P^NSU_D = the proportion of survivors in good recovery or mild/severe disability at 6 months with direct transport to the NSU

P^DGH_GR, P^DGH_D = the proportion of survivors in good recovery or mild/severe disability at 6 months with a secondary referral

RH_GR, RH_D = the cost of rehabilitation by GOS state at 6 months (where D is both mild disability and severe disability combined)

LE_GR, LE_D = the discounted life expectancy of patients by GOS state at 6 months

ACC_GR, ACC_D = annual care cost by GOS state at 6 months

TC = cost of transport in secondary referral

For Model B the incremental cost is calculated as follows:

Incremental cost = Cost_i - Cost₀ = ES_i x ((P_GR x (RH_GR + (LE_GR x ACC_GR)) + (P_MD x (RH_MD + (LE_MD x ACC_MD))) + (P_SD x (RH_SD + (LE_SD x ACC_SD))) + P_VS x (RH_VS + (LE_VS x ACC_VS)))) - (TC x P_DT)

where

Cost_{i =} the expected cost per patient associated with bypass strategy i

Cost_{0 =} the expected cost per patient associated with secondary referral

ES_i = the proportion of patients surviving under strategy i that would not have survived under the no bypass strategy

P_GR, P_MD, P_SD, P_VS, = the proportion of extra survivors in each of the GOS states at 6 months

RH_GD, RH_MD, RH_SD, RH_VS = the cost of rehabilitation by GOS states at 6 months

LE_GR, LE_MD, LE_SD, LE_VS, = the discounted life expectancy of patients by GOS states at 6 months

ACC_GR, ACC_MD, ACC_SD, ACC_VS = annual care cost by GOS states at 6 months

TC = cost of transport in secondary referral

P_DT = proportion of patients directly transported to the NSU

11.6.4.6. Probabilistic sensitivity analysis

A probabilistic sensitivity analysis was performed to assess the robustness of the model results to plausible variations in the model parameters. This analysis was applied exclusively to the strategy of transporting all patients to the NSU (strategy 2) compared no bypass in the conservative model B.

Probability distributions were assigned to each model parameter, where there was some measure of parameter variability (11.21). We then re-estimated the main results 5000 times, each time each of the model parameters were set simultaneously selecting from the respective parameter distribution at random.

Table 11.21

Parameters used in the probabilistic sensitivity analysis.

11.6.4.7. Results of the cost-effectiveness analysis

According to Model A there are large QALY gains and large cost savings associated with direct transport to the NSH – direct transport is dominant (Table 11.22). With Model B – the conservative model - the QALYs gained are smaller and costs are not decreased overall (Table 11.23 and Table 11.24). However, even with this conservative model, direct transport is cost-effective (below £20,000 per QALY gained).

Table 11.22

Results - Model A.

Table 11.23

Results - Model B – Far from NSU.

Table 11.24

Results - Model B - Near NSU.

We chose the group of patients who were 40 years old at the time of injury to represent the results (Table 11.22, Table 11.23 and Table 11.24). In the tables we report the results for the groups of patients of 20 and 60 of age as well. In these cases, direct transport was the dominant strategy in Model A and the incremental cost-effectiveness ratio was still below the threshold of £ 20,000 per QALY in Model B.

After running the Model B 5,000 times, the probability that directly transporting all the patients to the NSU is cost-effective (i.e. probability that the cost-effectiveness ratio is below £20,000 per QALY gained) is 73% when the NSU near the incident scene (within 20 minutes). In the cases of a patient aged 20 or 60, the probability falls to 66%.

For Model B, we performed a sensitivity analysis on the length of stay in the ICU: assuming that the most costly level 3 of care applies to all the outcome grades, the analysis shows that the direct transport would still be cost-effective as long as the increased length of stay does not exceed 3 days per additional survivor. Furthermore, even if the LOS were longer than this, these costs could be counteracted by additional complications in those patients who are secondarily transported to the NSH and had delayed surgery.

Using model B, we conducted a threshold sensitivity analysis to take into account the negative effects of overestimating the number of patients to be taken to the NSH. We define the positive predictive value as the proportion of patients transported directly to the NSH who are correctly diagnosed with a severe head injury. It is the number of true positives divided by the sum of both the true positives and false positives. In the case that the NSH is far from the accident scene (53 minutes), the strategy of taking all the patients directly to the NSH is cost-effective as long as the positive predictive value is more than 28%. If the NSH is near the accident scene (20 minutes), the direct transport to the NSH is marginally cost-effective strategy even if the positive predictive value is as low as 10%.

Using model B we performed a sensitivity analysis by using an alternative set of utility scores. The result was that direct transport strategy proved to be even more cost-effective than in the original model (Table 11.25).

Table 11.25

Results of the sensitivity analysis on the utility – Model B.

11.6.4.8. Discussion

We found that direct transport is potentially cost saving if the health status of patients are substantially improved as was indicated by the Poon study. Even in our conservative model we find that direct transport is cost-effective. But our analysis is limited for a number of reasons.

First, some of our assumptions regarding cost and survival were based on proxies or were extrapolated in to the long term.

Our conservative model, Model B, was based on the mortality results of a previous simulation model. Some of the parameters in the simulation model were based on expert judgement (those listed in Table 11.26). The main clinical outcomes from which the probability of death derives were estimated by experts. In particular, experts were asked to estimate the number of patients that would have survived assuming they received the appropriate care (critical intervention or neurosurgery) at time zero. The actual time elapsed since the accident and its related probability of death was taken from the database. Having these two points on the probability of death graph, a straight line was drawn. The authors found that the results were not sensitive to the slope of the line. However, the curve representing the real relationship between time to intervention and probability of death could have a different shape.

Table 11.26

Parameters for which the value was estimated by clinicians.

For simplicity, neither model considers the change in health status during the patient’s lifetime - they assume that the GOS score (assessed six months after the head injury) remains constant. If instead patients continue to improve after 6 months then our conservative model is underestimating the health gain and cost-effectiveness associated with direct transport. Likewise, our assumption that mortality is increased compared with the general population for survivors over their entire lifetime is a conservative one.

We have probably underestimated the cost savings attributable to direct transport because we included only hospital personnel (one anaesthetist and a nurse), omitting for the costs of drugs, equipment and ambulance. However, we have also omitted additional acute costs associated with direct transport in the treatment of complications such as hypoxia and hypotension, which are less likely if the patient has been stabilised earlier. This would require additional treatments such as volume replacement, blood transfusion, and in some extreme cases they would require surgery or ventilatory support for weeks.

A strategy of direct transport from the injury scene to an NSH will inevitably mean that the unit sees more patients than previously, even though many patients currently being taken to the nearest emergency department are subsequently transferred to the NSH. From the viewpoint of the NSH there will be a substantial cost impact in particular in terms of ITU beds.

In the long-term, this should not represent an increase in cost to the NHS since patients and their treatment costs are merely being shifted from one hospital to another. Furthermore we have no reason to believe that ITU costs are higher at the NSH; indeed according to the 2006 Reference Costs¹⁷⁷, the cost of a bed in a neurosurgical ITU is lower than the cost of a bed in a general ITU. Hence we did not include ITU costs in our base case analysis.

In the short-term, the resource impact is less clear and will depend on local circumstances. A DGH might not achieve the full cost savings from seeing fewer patients as typically it would be losing only ¼ of an ITU bed. However, staff costs and consumables would be redeployed almost immediately. The bed could also be re-deployed if there is currently under-capacity. If so more patients would be treated in ITU as a result of the increased capacity at DGHs but this would not necessarily produce a reduction in costs to the Trust. However, this increase in ITU capacity could lead to cost savings from reduced transfers.

To implement a direct transport strategy, NSH units will need to invest in extra ITU beds. This will be offset by cost savings at DGHs. However the cost savings will not necessarily offset the cost fully in the short-term. The implementation costs associated with shifting patients will have to be taken in to account in any cost impact analysis conducted for the purposes of implementation.

A US study²⁰⁶ reports a successful rate of GCS assessment (410/412 patients) by ambulance crews at the incident site, after an 8-hour training course. Hence, training for ambulance staff in the assessment of head injury patients would be necessary to safeguard the effectiveness and cost-effectiveness of the direct transport strategy.

Since we do not have survival outcomes for the other simulation model based in London (see 11.6.2) we could not use it to estimate cost-effectiveness. However, there is no reason to believe that it would effect our conclusions for near hospitals: if the specialist hospital is ≤20 minutes from the injury scene then direct transport is likely to be cost-effective. For distances greater than 20 minutes, the authors of the London model have erred on the side of caution by not recommending bypass. It seems logical that the further away is the specialist hospital the more risky is direct transport. Given the uncertainty of the evidence in this area, if we are to recommend direct transport at all then it probably is better to use some kind of cut-off but it is unclear how the authors of the London model made this decision since analyses based on transport times longer than 20 minutes are not present in the report.

The London model assumed that not just neurosciences but also other specialist services were available at the specialist centres. If specialist centres contain the whole range of services then the issue of whether ambulance crews can diagnose isolated head injury becomes less of an issue (this problem had been raised by several stakeholders), as long as specialist hospitals have adequate provision of beds, etc. Perhaps we should be recommending that bypass strategies are developed at a regional level to take into account local service configurations.