G.2.3.1. Videofluoroscopic swallow studies (VF)

According to the Committee, VF is readily available in hospitals with an adult speech and language service, but in fewer paediatric departments. One reason is limited access to paediatric speech and language therapists (SLTs) with the necessary competencies to perform VF. Occasionally, paediatric VFs are carried out by the adult SLT department, but with a paediatric SLT in attendance.

Table 1 below presents the national average unit cost of a VF outpatient procedure performed by an imaging service. The Committee noted that there are difficulties in using both VF and FEES in children due to compliance, positioning, availability and level of team expertise, but this is not limited to children with cerebral palsy. The Committee advised that a paediatric centre would undertake at least 20 VF procedures a week, but this would be much greater in adults. For this reason, a procedure with a duration of more than 40 minutes, and a duration of 20 to 40 minutes, could be considered as proxies for children and adults, respectively.

According to the Committee the optimal procedure would involve a radiographer (to obtain the images), radiologist and SLT to interpret the images. Although in some centres, interpretation can be performed solely by the radiologist or SLT. If no radiographer is present then the radiologist would be the operator of the equipment.

The Committee highlighted that there is variable practice across the country surrounding the health care professionals involved during the procedure. This was demonstrated by the Glasgow Royal Infirmary Hospital (cost year 2009) who compared a VF clinic staffed by SLTs and radiographers to consultant radiologists and SLTs. They found a SLT led procedure was more time efficient and gave a more comprehensive assessment of swallowing dangers. This clinic arrangement meant that the cost of a VF was reduced from £345 (using a consultant radiologist and SLT) to £215 (using a radiographer and SLT).

G.2.3.2. Fibreoptic endoscopic evaluation of swallowing (FEES)

According to the Committee, FEES is readily available in hospitals with an adult Ear, Nose and Throat (ENT) service, but used much more rarely than VF in paediatric departments. Most procedures are performed by an ENT specialist and SLT. Similarly to VF, a third health care professional may also be present to operate the equipment, but practice is variable across the country. The Committee noted that children may require a general anaesthetic and overnight stay; hence, Table 2 below presents the cost of a FEES procedure in an outpatient setting and inpatient setting. The Committee also noted that FEES is rarely performed, questioning the reliability of the cost in Table 2 in children and young people with cerebral palsy to assess eating, drinking and swallowing difficulties.

G.3.3.1. Feeding equipment

Most modifications with regards to feeding equipment involve changes to the size or shape of cutlery. However, some children and young people with cerebral palsy may require more high-tech electro/mechanical assistive devices if those modifications prove to be ineffective.

The Committee advised that the Neater Eater is the most commonly used high-tech assistive device in the UK and if such devices are considered a success, they would be used on an ongoing basis – potentially over a person’s lifetime.

Regardless of the specific model, each child or young person with cerebral palsy would receive their own personalised assistive device. These types of devices would generally be used as part of daily life in the same way that we would use cutlery.

Following a successful trial of the device, the child or young person with cerebral palsy and their families or carers would be trained by a company representative, OT and/or SLT over several home visits.

The upfront capital cost of a Neater Eater is relatively expensive at a cost of approximately £2,900 according to the manufacturer. Ideally the device would undergo annual reviews with a SLT at the child or young person’s home where it is most frequently used.

Neater Eaters have a 3 year warranty although the device is expected to last longer than this. The manufacturer offers a refurbishment package for units less than 7 years old that replaces ropes, drive belt, power supply, switches, plates and cutlery and includes a further 12 month warranty (if purchased outside of the original 3 year warranty) and a set up visit from one of their demonstrators. The refurbishment package costs £843 plus value added tax (VAT).

Purchasing an electro/mechanical assistive device is a capital cost, requiring an up-front payment. There are 2 aspects to capital costs:

• opportunity cost – this is the money spent on the device that could have been invested in another venture. This cost is calculated by applying an interest rate on the sum invested in the capital;
• depreciation cost – the device has a certain lifespan and depreciates over time, and will eventually need to be replaced.

The usual practice for economic evaluation is to calculate an ‘annual equivalent cost’. This is calculated by annuitizing the initial capital outlay (including staff/training costs) over the expected life of the device. Calculating the equivalent annual cost means making allowance for the differential timing of costs by discounting.

The formula for calculating the equivalent annual cost is:

Where:

• E = equivalent annual cost
• K = purchase price of the device
• T = training
• S = resale value
• r = discount (interest) rate
• n = device lifespan
• A(n,r) = annuity factor (n years at interest rate r)

Using this formula a cost per person per annum for use of an electro/mechanical assistive device was calculated to allow for comparison.

Table 3 below presents the parameters used to calculate the equivalent annual cost of an electro/mechanical assistive device.

As can be seen from Table 3 electro/mechanical assistive devices have an annual cost per person of approximately £677; driven by the high upfront capital cost.

G.3.3.2. Oral motor devices

One trial (Gisel 2001) included in the clinical evidence review evaluated the Innsbruck Sensiromotor Activator and Regulator (ISMAR). The Committee noted that this is one specific type of oral-motor device and the equivalent in the UK would be a palatal training aid (PTA). Similarly to electro/mechanical assistive devices each child or young person with cerebral palsy would receive their own personalised oral-motor device following a successful assessment, but the upfront cost of a PTA would relatively cheap, costing approximately £50 - the total cost of using oral motor devices would be driven by frequent health care professional contact. However, the Committee stated that their use is not wide spread in the UK and is highly dependent on specialist expertise in assessing, manufacturing, fitting and review.

According to the Committee oral motor devices would be made and fitted by an orthodontist and SLT over several visits to the clinic and reviewed, ideally, every 4 months by an orthodontist and SLT to check the fit and functional impact. This would incur an annual maintenance cost of approximately £99 assuming each consultation lasts 15 minutes (Personal Social Service Research Unit [PSSRU] 2015: community SLT per hour, £44; dental services per hour, £88).

Unlike electro/mechanical devices that are used over a person’s lifetime, some children and young people with cerebral palsy abandon oral-motor devices quite quickly if they are uncomfortable, or not effective, whereas others may use for them for several years. As a result the cost-effectiveness of oral motor devices will depend largely on patient preference.

G.3.3.3. Oral sensorimotor treatment

Beckman was one specific exercise programme identified in the clinical evidence review, but the Committee did not consider it was commonly used in the UK. Instead they advised that eating, drinking and swallowing regimens are often developed individually by dysphagia trained SLTs for children and young people with cerebral palsy to perform at their home or school. This would require at least 1 initial visit with a SLT to teach the child or young person with cerebral palsy and their families or carers on how to perform the techniques. Thereafter the SLT would make follow-up visits every 4 to 6 months to assess the impact and modify the treatment programme as necessary. According to the PSSRU 2015 each 30 minute visit with a Community SLT would cost approximately £22.

The duration and frequency of therapeutic interventions, included in the clinical evidence review varied, from over 30 minutes, 5 days per week (Clawson 2007; Ottenbacher 1981), to 1 hour per week (Sigan 2013). In light of this, the Committee noted that families could potentially struggle in day to day practice to follow the intensive regimens applied in some of the research trials; the most families could be expected to achieve is 30 minutes per day, but 3 days per week would be more realistic. Schools may also be able to perform the exercises, but this would require significant staff training if the care staff did not possess the necessary competencies.

The Committee advised that in UK clinical practice oral sensorimotor exercises are undertaken for less than 1 year, depending on the child or young person’s response. Therefore, if the benefits from oral sensorimotor exercises can be achieved in 1 year and maintained over a person’s lifetime without further treatment, oral sensorimotor exercises could be considered cost-effective compared to lifetime management that requires ongoing resources.

G.4.3.1. Tube feeding

Tube feeding can be used as an adjunct to oral feeding, or if there is clinical concern about the safety of swallowing they can replace oral feeding. Long term interventions to optimise nutritional status include gastrostomy or jejunosotomy tube feeding, whereas nasogastric tube feeding would be used on a shorter term basis. The former are surgical procedures associated with a high cost, whereas the latter can be performed by a nurse as an outpatient procedure. However there are specific clinical implications for long term naso-gastric tube placement that mean they are not the preferred route of enteral feeding beyond short term use.

The costs associated with long-term nutritional supplementation via gastrostomy or nasogastric tube feeding, are outside the scope of NHS Reference Costs and should remain within primary medical services (Department of Health, Reference Costs Guidance 2014-15). For this reason, currency codes related to endoscopic insertions from NHS Reference Costs are presented in Table 4 as a proxy. With regards to nasogastric tube feeding, costs were reported solely for babies under special care (HRG XA03Z); these were considered irrelevant to this review and are not reported.

The randomised study by Corry 2008 was identified as a relevant source of costing data on tube feeding through ad-hoc searches. This study was included in the Cochrane review on tube feeding for adults with swallowing disturbances. However, it is important to note that Corry 2008 was based on patients with head and neck cancer who required enteral feeding, whose costs may not be generalisable to children and young people with cerebral palsy. They stated that the insertion costs are significantly different as nasogastric tubes are inserted by nursing staff as an outpatient attendance (including the cost of chest X-ray) whereas percutaneous endoscopic gastrostomy tubes are inserted by surgeons in theatre. Table 5 below reports the costs by Corry 2008 alongside inflated sterling prices calculated by the Technical Team.

In addition to the procedure, the Committee advised that some children and young people with cerebral palsy would undergo an intense monitoring schedule during the first few days or weeks with a paediatric nurse specialist. Thereafter the child or young person with cerebral palsy would be monitored on a similar frequency to those receiving high calorie feeds or antimetics, with gastrostomy or jejunosotomy incurring 1 additional visit with their paediatric surgeon each year at a cost of approximately £202 (NHS Reference Costs 2014/15, WF01A, Consultant led Non-Admitted Face to Face Attendance, Follow-up, Paediatric Gastroenterology).

In addition to the monetary cost of tube feeding, the Committee advised that some qualitative reviews show tube feeding can negatively impact a person’s quality of life by affecting social interactions at meal times. Moreover, if the procedure and use of tube feeding is associated with adverse effects, they can incur further treatment costs and decrements in quality of life.

The Committee highlighted that nasogastric tubes frequently fall out and require the cost of a health care professional to reapply to tube if the family/carer were unable to do so. The Committee also added that there are a number of clinical concerns to their long term usage. Equally gastrostomy and jejunostomy tubes, need routine and on occasion emergency replacement which on occasion need professional rather than parent intervention. It was also noted that tube feeding, when used appropriately, positively impacts on clinical wellbeing and health, improving quality of life, justifying the high costs tube feeding can entail in those cases.

G.4.3.2. High calorie feeds

The Committee highlighted 2 commonly used high calorie supplements used to optimise nutrition in children and young people with cerebral palsy: Calogen® and Maxijul®; the cost of these supplements are reported in Table 6.

Ultimately the cost of high calorie feeds will depend on the frequency those feeds are administered. If those feeds were used to substitute rather than complement diet at home, the cost could be substantial. However, the person’s diet would be reviewed and modified prior to consideration of high calorie feeds, hence the health care professional should determine the appropriate frequency of high calorie supplements.

G.4.3.3. Antimetics

Table 7 presents the acquisition cost of antiemetic drugs, over 1 day and 1 month of continued use, according to the cost reported in the October 2016 NHS Electronic Drug Tariff. For this cost description, BNF dosages were the preferred costing method because trial dosages may not reflect UK clinical practice. Moreover, not all interventions have been identified in the clinical evidence review.

For domperidone and metoclopramide the BNF reports a range of doses to prevent nausea and vomiting according to age and weight. To represent the range of conceivable costs Table 7 presents costs for the maximum dose and a midpoint. The full range of preparations is also reported to demonstrate the variability of costs within each drug.

Erythromcin was also considered by the Committee to be used as an antiemetic/pro-motility in low doses (125mg twice daily); however this would be used off-license to prevent nausea and vomiting.

It is evident from Table 7 that oral solutions of domperidone and metoclopramide are substantially more expensive than tablets that cost less than £3 per month. Therefore, when tablets can be tolerated they should be offered instead of oral solutions because they are cheaper and there is no evidence to suggest they are any less effective. If an oral solution is required erythromycin would the cheapest antiemetic at a cost of approximately £10 per month.

G.5.3.1. Pharmacological

G.5.3.2. Non-pharmacological

G.6.3.1. Melatonin

Melatonin is available as a modified-release tablet (Circadin®) and also as unlicensed formulations. Circadin® is licensed for the short-term treatment of primary insomnia in adults over 55 years, but unlicensed immediate-release preparations are available. The BNF reports the following dose for sleep onset insomnia and delayed sleep phase syndrome in children aged 1 month to 18 years:

• initially 2 to 3 mg daily before bedtime;
• increased if necessary after 1 to 2 weeks to 4 to 6 mg daily before bedtime;
• max. 10 mg daily.

Similarly the Committee advised the following dosages for children and young people with cerebral palsy stratified by age:

• 2 to 3mg daily in children under 5;
• 6mg daily in children over 5;
• max. 12mg daily.

Table 13 below presents the acquisition cost of Circadin® across 1 day and 1 month for 3 dosages (3mg, 6mg and 10mg) to reflect a range of conceivable costs. However, other formulations of melatonin are available from ‘special-order’ manufacturers, or specialist importing companies.

G.6.3.2. Sedatives

G.6.3.3. Non-pharmacological

There are many different types of sleep systems available such as postural devices, wedges and supports that vary in price according to the manufacture, attachments and size. Based on this, it would be inappropriate to suggest a “one-price-fits-all” because the equipment would be individualised to the child or young person with cerebral palsy; from a list of manufactures and systems provided by the Committee, the upfront capital cost could range from approximately £100 to £1,000.

The Committee also advised that sleep positioning equipment is usually prescribed with room for growth in mind with a lifespan of approximately 3 to 5 years, but this would vary according to the type of sleep system.

With those factors in mind, there are 2 aspects to capital costs:

• Opportunity cost – this is the money spent on equipment that could have been invested in another venture. This cost is calculated by applying an interest rate on the sum invested in the capital.
• Depreciation cost – the equipment has a certain lifespan and depreciates over time, and will eventually need to be replaced.

The usual practice for economic evaluation is to calculate an ‘annual equivalent cost’. This is calculated by annuitizing the initial capital outlay over the expected life of the equipment. Calculating the equivalent annual cost means making allowance for the differential timing of costs by discounting.

Due to the variations in cost and lifespan described, the equivalent annual cost could range from £26 to £263, for capital costs of £100 to £1,000, respectively, over 4 years.

In addition to the initial capital outlay, the equipment should be reviewed annually, especially with children as the set up would need to be adjusted for growth and changes in their presentation. Ideally the equipment would be reviewed annually by the health care professional who issued the equipment (occupational therapists, physiotherapists or social care occupational therapists), but this may be increased in children and young people with more complex needs, or reduced to when a need is identified by the family or carer. Moreover families or carers should be able to contact services if they identify a need between reviews.

The cost per hour of patient contact with a community physiotherapist or community occupational therapist is £44 according to the PSSRU 2015. The cost per hour includes the costs of overheads, but does not take into account the travel time required by community therapists. Based on 2, 1 hour reviews per year (£88) the total cost per year could range from £114 to £351 (for capital costs of £100 to £1,000, respectively, over 4 years).

An initial assessment by familiar therapists and an equipment representative should be carried out with the child or young person with cerebral palsy and their main carers who will be using the equipment. The equipment should then be left with the family for a few days to see how the child or young person responds. As a result, only children and young people with cerebral palsy who are expected to benefit from sleep systems from this assessment would receive one in clinical practice. However, Lloyd 2014 found no significant difference in sleep initiation and maintenance for sleep systems versus no sleep systems which questions if the benefits of sleep systems justify the cost.

G.7.3.1. Non-pharmacological

The cost per cognitive behavioural therapy (CBT) and psychotherapy attendance is presented in Table 17.

G.7.3.2. Pharmacological: antidepressants & anxiolytics

Pharmacological acquisition costs are presented over the course of 1 day and 1 month of continued use in Table 18 based on the costs reported in the October 2016 NHS Electronic Drug Tariff. For this cost description BNF dosages, unless otherwise stated, were the preferred costing method because trial dosages my not reflect UK clinical practice. Moreover, no pharmacological interventions were identified in the clinical evidence review. Antidepressants and anxiolytics would be administered at home, ideally following an assessment with a specialist psychiatrist (NHS Reference Cost 2015, WF01B, Consultant-led, First Attendance, Non-Admitted Face To Face: Service code 171, Child and Adolescent Psychiatry, £171; Service Code 713, Psychotherapy, £227).

The BNF reports a range of doses; hence, to represent the range of conceivable costs Table 18 presents costs for the maximum dose and an arbitrary midpoint. Appropriate preparations are also reported to demonstrate the variability of costs within each drug.

G.8.3.1. Clinical effectiveness

G.8.3.2. Measurement and valuation of health effects

The quality adjusted life year (QALY) is NICE’s preferred measure of benefit for economic evaluation. This is because it can be seen as a generic measure of health which allows a comparison across interventions which affect different dimensions of health.

The QALY reflects the 2 principle objectives of health care:

• increase longevity;
• increase quality of life.

Estimating a QALY involves placing a quality of life weight on a particular health state. This quality weight lies between 0 and 1, where 1 denotes full or ‘perfect health’ and 0 denotes death. For this review question hypothetical quality of life weights and health states based on a 9-point drooling scale have been estimated.

In the model there is 1 user input to consider when estimating the quality of life in children and young people who drool: the disutility associated with increasing scores.

A separate systematic search to identify utility values for children and young people with cerebral palsy was not undertaken. Instead, a search was conducted on the CEA Registry using the term “cerebral palsy” in July 2015. This search identified 4 studies with health states relevant to cerebral palsy (Cahill 2011; Obido 2009; Heintz 2008; Carroll 2006). After title and abstract screening only 1 of those studies identified considered people with cerebral palsy, subsequently the full-text of Heintz 2008 was obtained and assessed for inclusion.

Heintz 2008 estimated QALY weights for individuals diagnosed with cerebral palsy based on the utility values reported in the study by Rosenbaum 2007. Consequently the full-text of Rosenbaum 2007 was retrieved and assessed to inform the utility weights in the model.

Rosenbaum 2007 asked carers to complete the Health Utilities Index Mark 3 (HUI3) – a quality of life survey - on behalf of the person with cerebral palsy they cared for. The resulting score was then transformed into a utility value based on an algorithm using Canadian population values. The utility scores estimated for 192 people with cerebral palsy according to their GMFCS level are presented in Table 21.

The methods used to derive utility values within Rosenbaum 2007 are not in line with the NICE reference case which specifies patients to measure their own quality of life using the EQ-5D valued by a representative sample of the UK population. However in the absence of alternative values this study was considered the best available to inform the model. As can be seen from Table 21 a higher GMFCS level is associated with a lower utility value which reflects a lower quality of life. The Committee noted that utility values of 0.16 and -0.08 were very low for children and young people who require physical assistance for mobility. Following this, the Committee believed utility values would be higher if children and young people with cerebral palsy completed the questionnaire themselves.

In the model the GMFCS level and associated utility value represents the highest achievable utility which would represent a drooling score of 1. In the base-case this was set to GMFCS level II associated with a utility of 0.50 which the Committee believed to represent the quality of life of children and young people with cerebral palsy.

Chang 2012 investigated health-related quality of life using the Paediatric Quality of Life Inventory Version in 47 children with cerebral palsy, with and without drooling according to the TSG score. They found the physical health and psychosocial health summary scores of the children that drooled (16.29 ± 15.97 and 42.92 ± 17.57, respectively) were lower than for the children that did not drool (31.97 ± 22.22 and 57.09 ± 12.21, respectively; P < 0.01) concluding the more severe the drooling was (without considering the type of cerebral palsy), the lower the physical and psychosocial quality of life was in the children with cerebral palsy. However the definition of drooling is unclear and this is not disaggregated according to the TSG score. Ideally further research should be conducted using the EuroQol Group’s EQ-5D instrument (a child-friendly version is available) or the Health Utilities Index (which was developed for children) designed to allow subgroup analysis by severity of cerebral palsy in terms of the GMFCS level and drooling score.

It was assumed that the QALY loss from drooling would increase as the drooling score increases. In the model a disutility per unit increase in score was applied additively to the upper utility value. In the base case the disutility was set to an arbitrary value of 0.025, based on 5% of the utility value for GMFCS level II, but the disutility could be adjusted by the user in the model.

Regardless of the disutility value chosen, a linear relationship with the drooling score is maintained, due to insufficient evidence to suggest otherwise. To reiterate, effectiveness is informed by hypothetical health state utilities, but the structure of the model will allow future research to be incorporated.

The base case health state utilities informed by a disutility of 0.025 (per unit increase in score) and GMFCS level II are presented in Table 22. If we compare no drooling to profuse drooling the utility value is almost halved (0.50 vs. 0.30), somewhat reflecting the physical health summary scores reported by Chang 2012 (31.97 vs. 16.29).

If the drooling score holds a linear relationship between the score and utility value, the GMFCS level will have no effect on the incremental change in QALYs between interventions because this is not intervention dependent; in other words the QALY gain will change proportionally for each intervention. For example, if we compare glycopyrrolate with a mean improvement of 3 to botulinium toxin type A with a mean improvement of 4 the incremental QALYs over 1 year are 0.025 regardless of the GMFCS level, this is illustrated below for GMFCS levels I and III:

• GMFCS level I:

  • 0.075 QALYs gained from glycopyrrolate (post-treatment utility, 0.765 – pre-treatment utility, 0.690)
  • 1.00 QALYs gained from botulinium toxin type A (post-treatment utility, 0.790 - pre-treatment utility, 0.690)
  • incremental difference of 0.025 (1.00 – 0.075) QALYs.
• GMFCS level III:

  • 0.075 QALYs gained from glycopyrrolate (post-treatment utility, 0.315 - pre-treatment utility, 0.240)
  • 1.00 QALYs gained from botulinium toxin type A (post-treatment utility, 0.34 - pre-treatment utility, 0.240)
  • incremental difference of 0.025 (1.00 – 0.075) QALYs.

Changing the disutility value could have an impact on the recommendations because the larger the disutility value the larger the QALY gain - implying a more cost-effective intervention. This will be a proportionate change i.e. doubling the disutility value will double the hypothetical QALY gain, but changing the disutility value will not impact on the ordering of recommendations - only their cost-effectiveness relative to the NICE threshold.

In the model it is implicitly assumed that all interventions have an identical side effect profile. Were this not the case, differences in morbidity would also have to be incorporated into calculating the differential QALY between these interventions. However, due to insufficient side effects profiles reported within the trials this has to be acknowledged as a limitation of the model.

G.8.3.3. Resource and cost use

In accordance with the NICE Guidelines Manual costing was undertaken from the perspective of the NHS and personal social services (PSS). Drug acquisition cost were taken from the October 2016 NHS Electronic Drug Tariff whilst procedures and attendances were taken from NHS Reference Costs 2014/15, unless otherwise stated.

The acquisition cost of glycopyrrolate depends upon a person’s weight. In addition, the duration of treatment for transdermal hyoscine hydrobromide and glycopyrrolate will vary depending on their response. To cost these treatments accurately there are inputs in the model to vary both the weight (kg) of the child or young person with cerebral palsy and time horizon (weeks). In the base case this is set to 47kg and 26 weeks to reflect the trials included in the clinical evidence review. To explore the impact treatment duration has on cost-effectiveness, sensitivity analyses have been conducted where the duration of treatment is increased to lifetime (40 years) and decreased to 8 weeks.

G.8.3.4. Sensitivity analysis

A series of scenario analyses were undertaken in order to test how sensitive the results were to uncertainty in individual parameters. Parameters varied in the scenario analysis were chosen on the basis of uncertainty in their estimation or the potential impact that they had on the results. The values varied, along with their rationale are shown in Table 26.

G.8.5.1. Base case

G.8.5.2. Sensitivity analysis

The fully incremental results from analysis described in Table 26 are presented in Table 31.

The duration of treatment is a key driver of cost-effectiveness. Increasing the duration reduces the cost of surgery relative to the other interventions as this incurs a one-off cost, whereas the other interventions incur an ongoing cost. As a result, surgery dominates transdermal hyoscine hydrobromide, botulinum toxin type A and glycopyrrolate under a lifetime horizon. Conversely, when the time horizon is reduced to 8 weeks, surgery has an ICER substantially above NICE’s threshold and would not be considered cost-effective.

The only scenario where transdermal hyoscine hydrobromide would not considered cost-effective is under a lifetime horizon, all remaining scenarios produce an ICER substantially below NICE’s typical threshold of £20,000 per QALY gained. Therefore transdermal hyoscine hydrobromide should be recommended as a first line intervention.

However, if the Committee believes the clinical effectiveness of botulinum toxin type A is underestimated in the base case, scenario 1 demonstrates that the QALY gains and additional cost from botulinum toxin type A, relative to transdermal hyoscine hydrobromide, would be considered cost-effective with an ICER of £18,856. With regards to implementation, this would lead to a large change in clinical practice as there are currently not enough tertiary centres with the ability to perform botulinum toxin type A injections for this indication. Conversely, if children and young people with cerebral palsy require a general anaesthetic at a cost of £134, botulinum toxin type A would not be considered cost-effective relative to transdermal hyoscine hydrobromide with an ICER above NICE’s advisory upper threshold (£48,394). Combing these 2 scenarios (increasing the effectiveness and cost of botulinum toxin type A) would lead to an ICER of £24,179 which is within NICE’s advisory range for cost-effectiveness (£20,000 to £30,000 per QALY gained).

Glycopyrrolate would not be considered as a cost-effective intervention in any of the scenarios tested. For glycopyrrolate to be considered cost-effective under a threshold of £20,000 per QALY gained, the duration of treatment would need to be less than 10 weeks, given the same QALY gains as botulinum toxin type A. However, given that transdermal hyoscine hydrobromide are believed to be at least equally effective in the short-term, transdermal hyoscine hydrobromide would dominate glycopyrrolate as it is cheaper and at least equally effective.

G.9.3.1. Intermediate to final outcomes

In line with NICE’s preferred method for economic evaluation, a de novo cost-utility model was constructed to determine the most cost-effective interventions for those 3 populations outlined in Section G.9.2.

One outcome to be considered as a possible alternative to quality-adjusted life years (QALYs) was the number of low impact fractures prevented. However, none of the included studies reported on this outcome, and neither did the studies report adverse effects (bone fragility and/or gastric/oesophageal irritation/ulceration). Instead, studies reported on the changes in BMD measured with dual energy X-ray absorptiometry (DEXA) scans at various sites.

BMD is an intermediate outcome that does not incorporate quality or quantity of life. When conducting cost-utility analysis, final endpoints that are related to health related quality of life and hence to utilities and QALYs are required. A link was established between BMD and risk of fracture because the Committee agreed the clinical evidence review on risk factors demonstrated a strong relationship.

A decision tree model was developed in Microsoft Excel® (2013) (Figure 2) from the perspective of the NHS, using 2014/15 costs. For the first 2 populations, the time horizon for the model was a year as this reflected the longest trial follow-up in the clinical evidence review. In addition, those interventions are not generally prescribed on a longer basis due to changes in child or young person’s physique and nutrition, or potential pharmacological toxicities (bisphosphonates). As a result, no discount rate was applied. However, for children and young people with cerebral palsy who use standing frames as part of their postural management programme, the time horizon was increased to 5 years to reflect the lifespan of a standing frame.

Figure 2Decision tree model

* Patient from 1 of 3 defined populations; CYP with CP, children and young people with cerebral palsy

A common algorithm used to evaluate a person’s risk of fracture is the World Health Organisation (WHO) Fracture Risk Assessment (FRAX) tool developed by the University of Sheffield. However, this tool uses BMD measures from the femur neck and is only applicable to people aged 40 years or over. For these reasons the FRAX tool was not considered appropriate for children and young people with cerebral palsy.

To estimate the risk of fracture from BMD, the model utilised an approach used in a study identified in the clinical evidence review (Henderson 2010) who tested the hypotheses that DEXA measures of BMD correlate with bone fractures in children and adolescents with cerebral palsy or muscular dystrophy. This was undertaken by calculating BMD Z-scores for the distal femur and lumbar spine.

Z-scores can be used to compare a measurement to a reference value. The Z-score is the number of standard deviations away from the average value of the reference group using the following formula:

A total of 507 children with cerebral palsy and 112 with muscular dystrophy aged 6 to 18 years from 8 centres in the US, of sufficient severity to significantly impair ambulation, were included in the study by Henderson 2010. Considering that Gross Motor Function Classification System (GMFCS) levels IV and V were identified as risk factors for low BMD in the clinical evidence review, the prevalence of fracture taken from Henderson 2010 would be overestimated if applied to ambulant children i.e. GMFCS levels I to III.

This has to be acknowledged as a limitation for the population with children and young people at increased risk of reduced BMD who will be ambulatory. However, this is unlikely to change the incremental difference between the treatments as the risk would change proportionately for both treatments when they are administered to homogeneous populations.

Henderson 2010 do not specify the type of bone fracture participants experienced. However, they stated that the most common site in healthy children is the forearm, with over 80% of the fractures occurring in the upper limb and fewer than 2% in the femur. They also added that in contrast, over 70% of fractures occur in the lower limb of non-ambulatory children with disabilities such as cerebral palsy or muscular dystrophy, and up to one-half of all fractures are in the femur. The quality of life and implications according to the site of fracture are discussed further in Sections G.9.4 and G.9.5.1.

Table 33 below presents the prevalence of fracture against the anatomical site according to 5 equally sized Z-score groups. It is evident that the probabilities in Table 33 do not follow a linear relationship, however the Committee agreed a linear assumption would be reasonable to inform the model. The prevalence reported was a point estimate. In addition, a participant’s history of prior fracture was obtained at the time of the DEXA scan, but this was categorised into a simple yes, or no, providing no information on the timing or number of fractures participants experienced. Consequently, this has to be acknowledged as a limitation of converting BMD into a probability of fracture for the time horizon in the model. For simplicity, it is also assumed patients can only experience 1 fracture in the model.

The ranges in Table 33 were considered to be too restrictive to demonstrate differences between the interventions within the trials and across the trials included in the clinical evidence review. This was because the majority of interventions were found to lie within the same Z-score percentile; hence the interventions would be associated with the same risk of fracture. For this reason the mid-point of each 20th percentile group was taken to inform a linear trend using Microsoft Excel® (2013). This linear trend estimates the risk of fracture for each intervention, subsequently demonstrating differences in their effectiveness that were unobservable from the Z-score ranges in Table 33. The estimated linear trends used to inform the model are illustrated in Figure 3.

Figure 3Prevalence of fracture illustrated as a linear trend

Figure 3 illustrates the expected negative relationship between the Z-score and the probability of fracture i.e. as the Z-score gets bigger (more negative) the probability of fracture increases. Using this relationship in the model will enable the probability of fracture to be estimated separately for each intervention, given the Z-score is available for the femur or lumbar spine.

G.9.3.2. BMD sites included in the model

As noted in Section G.9.3.1 Henderson 2010 took DEXA measures from the femur or lumbar spine, whereas the clinical evidence report also reported measures from the femur neck.

To decide what assumptions can be made to allow data to be included given heterogeneity between different BMD sites, the Committee was consulted. This led to 2 assumptions that would incorporate the largest evidence base:

1. Distal femur region 1 (F1) used as a proxy for the femur neck
   The Committee noted that in UK clinical practice the most common sites used to measure BMD are the femur neck and lumbar spine. However, not all of the studies included in the clinical review reported BMD at those sites. Consequently an assumption was made that the distal femur region 1 (F1) could be used as a proxy for the femur neck as they were considered to be the closest sites in terms of properties and proximity.
2. When 3 regions of the distal femur are reported, F1 will be used to inform the model
   When studies reported regions of the distal femur, all 3 regions (F1, F2 and F3) were reported, as a result F1 would be the region included in the model as this would facilitate the largest number of comparisons (given the previous assumption).

The correlations reported by Henderson 2002b between BMD sites (Table 34) supported these assumptions. They found that the proximal femur (the closest measurement to the femur neck) was highly correlated with each of the 3 regions of the femur, whereas the lumbar spine was less well correlated with the 3 regions of the femur.

The clinical evidence review also found that within the same study, one site would show a significant difference whereas another may not. For these reasons, both the lumbar spine and F1 are included in the model to capture this uncertainty.

Conflicting results in the clinical evidence review included:

• a clinically beneficial effect of increased time spent on the standing frame compared to no increase in the regular standing duration for vertebral BMD, but no clinically significant difference for proximal tibial BMD;
• a significant difference between the 2 groups for areal BMD of the femur, but no significant difference between home-based cycling program and usual physical activity for areal BMD of the lumbar spine,;
• a clinically beneficial effect of pamidronate disodium compared to placebo for BMD in distal femur region 1, 2 and in the lumbar spine, but no clinically significant difference for BMD in distal femur region 3.

G.9.3.3. Baseline adjustments

Inclusion criteria applied by the studies included in the clinical evidence review varied, as a result this led to heterogeneous populations. For this reason, a single baseline pre-treatment BMD was applied to the population at increased risk of reduced BMD and population with proven osteoporosis as study heterogeneity meant that the baseline BMD in the individual studies differed markedly. Inclusion criteria and pre-treatment BMD reported by the studies included in the model are presented in Table 35 for the 3 populations included in the model.

To obtain homogeneous populations, the active exercise (cycling) study by Chen 2013 was chosen to inform the baseline BMD for the population at increased risk of reduced BMD, and the study by Henderson 2002 that evaluated pamidronate disodium was chosen for the proven osteoporosis population. Both of these studies included measures for F1 and lumbar spine to ensure the pre-treatment BMD from both sites was adjusted. These studies also included participants that reflected the intended population: ambulatory participants and non-ambulatory participants, respectively.

It is important to note that Henderson 2002 only reported Z-scores, but pre-treatment BMD can be calculated using the following rearrangement of the Z-score:

Henderson 2002 also based the expected BMD for the distal femur on Henderson 2002b which is consistent with the reference data utilised in the model (Section G.9.3.4).

Table 36 below shows the baseline pre-treatment BMD applied to each population.

Further assumptions were required to include the standing frame and vitamin D studies in the model. These studies only reported p values, but it was possible to estimate BMD from the statements that were made within the studies. Because of the estimation techniques used to include these studies, the results inferred must be interpreted with caution.

Caulton 2004 was the only study included in the postural management population, and they reported a 6% mean increase in the vertebral BMD in participants who used standing frames for increased durations. The baseline BMD (0.38) was taken from Jekovec 2000 (calcium and vitamin D), who included participants who had spastic quadriplegia and were bedridden to reflect the non-ambulatory population included in the standing frame study. This was then inflated by 6% based on Caulton 2004 to estimate the post-treatment BMD (0.40).

It is important to note that the comparison analysed in the model was between a standing frame and “no treatment” as this was the comparison of interest to the Committee. For this reason, the additional benefit from an increased standing duration compared to a normal standing duration was assumed to equal the additional benefit of standing frames compared to “no treatment”.

For risedronate plus vitamin D compared with vitamin D, Iwasaki 2008 only reported p values. However, this study was included in the model as a graph presented with a regression line enabled an estimation of pre-treatment BMD (x) and post-treatment BMD (y). The pre-treatment value (x) was arbitrarily chosen from the graph, but the regression line reported by the study means the change in BMD is the same regardless of the baseline value.

Assuming a pre-treatment BMD of 0.32 (x) for vitamin D results in a post-treatment BMD (y) of 0.33 based on the regression line reported by Iwasaki 2008:

Assuming a pre-treatment BMD of 0.45 (x) for risedronate plus vitamin D results in a post-treatment BMD (y) of 0.48:

The regression line was not used to estimate the post-treatment BMD in the base case as this is informed from the relative improvement on the adjusted baseline pre-treatment BMD (Table 66). However, a sensitivity analysis estimated the post-treatment BMD using the regression line.

Iwasaki 2008 undertook a randomised trial, but the 2 arms were not comparable at baseline. For this reason, Iwasaki 2008 compared the 2 arms separately i.e. pre-treatment versus post-treatment.

G.9.3.4. Source of expected BMD to calculate Z-scores

The expected BMD and standard deviation (SD) values to calculate Z-score are only required for the chosen baseline study in each of the 3 populations, not every study included in the model. This is because the expected values reflect a homogenous population defined by the baseline study.

Because a common baseline BMD has been applied to each population, cost-effectiveness is determined from the post-treatment BMD rather than the change in BMD. Moreover, the structure of the decision tree would not enable the change in BMD to be used as an outcome measure.

To calculate the Z-score for each intervention, the participant’s pre-treatment BMD was taken from the mean BMD reported in the baseline study (Chen 2013, cycling/at increased risk of reduced BMD; Henderson 2002, pamidronate disodium /proven osteoporosis; Jekovec 2000 [estimated from Caulton 2004], standing frame) pre-treatment and post-treatment. The DEXA site measured, age and gender of study participants was also recorded to ensure the expected BMD reflected the participants in the trials.

The expected BMD was taken from the National Health and Nutrition Examination Survey (NHANES) 2005-2008. This survey aimed to provide reference values for lumbar spine BMD DEXA measures in a US population in participants aged 8 years and over, disaggregated further by age and nationality.

Henderson 2002b estimated equations to predict distal demur BMD DEXA measures for males and females. This was estimated from 231 Non–African Americans with a mean age of 10.5 years (range 3 years to 18 years 6 months). The equations to calculate the expected BMD and SD for the distal femur region 1 are presented in Table 37. As noted in Section G.9.3.1 this study was used to calculate Z-scores for a correlation study (Henderson 2010) and to calculate Z-scores in a study that included pamidronate disodium as an intervention (Henderson 2002).

In the model the expected BMD and SD was calculated separately for males and females, then weighted according to the proportion of males in the study. Table 38 presents the resulting estimations for expected BMD and SD for the 3 populations.

The difference between the adjusted pre-treatment BMD (common baseline) and unadjusted pre-treatment BMD (study reported) was calculated to find an inflator to apply to the post-treatment BMD. This is described in more detail in Table 64.

Table 39 below presents the adjusted post-treatment BMD scores and resulting Z-scores for each intervention. As stated previously, adjustments to the population at increased risk of reduced BMD were made using active exercise (cycling), and for proven osteoporosis, using pamidronate disodium.

As expected, the population at increased risk of reduced BMD has the highest Z-scores (less negative) which suggests they are closer to the general population that those with proven osteoporosis or those who require a standing frame.

G.9.3.5. Probability of fracture

Table 40 below presents the probability of fracture calculated for each intervention based on the linear relationship estimated from the study by Henderson 2010 (Section G.9.3.1).

A baseline pre-treatment BMD is applied to all studies in a population; hence, the probability of fracture pre-treatment is equal across the interventions in that population. Consequently, the change in the probability of fracture (pre- vs. post-treatment) is inferred from the probability of fracture post-treatment.

For illustrative purposes the change in the risk of fracture (post-treatment vs. pre-treatment) is presented in Table 41. It is evident that the change in the risk of fracture is small except for vitamin D plus calcium and pamidronate disodium which reflects the findings from the clinical evidence review.

In the base case, the probability of fracture over a 5-year time horizon utilises the probabilities reported in Table 43 as it is unclear from Henderson 2010 the point in time prevalence was measured.

It is important to note that vitamin D plus calcium is associated with a 0% probability of fracture post-treatment in children and young people with cerebral palsy at increased risk of reduced BMD which is questionable. This study included participants who had spastic quadriplegia cerebral palsy who were bedridden and required assisted feeding. Those participants provided a low pre-treatment BMD (0.38) whereas the baseline BMD applied to the population at increased risk of reduced BMD was substantially larger (0.58). The model assumed the same relative treatment effect in this population as in the study population (study pre- vs. post- treatment, 0.38 vs. 0.48; adjusted pre- vs. post-treatment, 0.58 vs. 0.73) but this could potentially over-estimate post-treatment BMD in children and young people at increased risk of reduced BMD if they have “less room for improvement”.

G.9.5.1. Cost of fracture

The literature was searched to estimate the cost of a fracture, but the populations identified were based largely on post-menopausal women and considered irrelevant. This included the submission of evidence undertaken by Amgen in NICE TA204 who noted vertebral and femur (or upper limb) fracture ICD diagnosis codes were not included in Healthcare Resource Group (HRG) codes relating to a fracture diagnosis. Instead they used International Classification of Diseases (ICD) HRG codes relating to hip traumas as a proxy.

Overall, NHS Reference Costs 2014/15 do not report the cost of a fracture in the femur or the lower limb, but the Committee thought it was reasonable to use the cost of a hip fracture as a proxy in the absence of alternative reliable costing data to reflect the methods used by Amgen. Procedures that related to complications and comorbidities were also preferred as they were considered applicable to a population with cerebral palsy. The Committee stated that major procedures would include a large share of elderly people, for this reason intermediate procedures were used to inform the cost of fracture in children and young people with cerebral palsy.

The currency codes from NHS Reference Costs 2014/15 used to inform the cost of a fracture to the femur in the base case are presented in Table 44.

To reflect the range of fractures presented in clinical practice a sensitivity analysis (Section G.9.7.1 and Section G.9.9.2) has been conducted using costs for a lower limb fracture considered to be less severe and less costly (NHS Reference Costs 2014/15, HT24D, Intermediate Knee Procedures for Trauma, 18 years and under £2,829).

Use of standing frame as postural management

Each child or young person with cerebral palsy would be provided with their own individualised standing frame. The Committee also noted that some children and young people with cerebral palsy may have 2 standing frames – 1 for use at home and another at their school. However, it was agreed that the cost of 1 standing frame to inform the model would be reasonable.

There are many different types of standing frames available in the UK that vary in price according to the manufacture, model and purpose (upright, prone or standing). Based on this, it would be inappropriate to suggest a “one-price-fits-all” because the equipment would be individualised to the child or young person with cerebral palsy; from a list of manufactures and models provided by the Committee, the upfront capital cost could range from approximately £2,000 to £8,000. Moreover, Jekovec 2000 did not report the specific standing frames used in the study stating that participants used a variety of upright and semi-prone standing frames.

For costing purposes, an upfront capital cost of £4,500 was chosen to represent the cost of the average standing frame in the model. The Committee stated that a person with cerebral palsy would have 2 to 3 standing frames over their childhood and early adult years to match their growth. For this reason, it was assumed that a standing frame can be used for approximately 5 years before it needs to be replaced.

To reflect the range of standing frames available a sensitivity analysis varying the upfront capital cost by ±50% was undertaken (Section 73G.9.7.1 and Section G.9.9.2).

Table 45 below presents the parameters used to calculate the equivalent annual cost.

In addition to the initial capital outlay, the frame should be reviewed annually, especially with children as the set up would need to be adjusted for growth and changes in the person’s presentation. This would usually be performed by a physiotherapist or occupational therapist. Adding this annual review to the equivalent annual cost results in a total annual cost of £999 per person.

Use of vibration therapy as passive exercise

Purchasing vibration equipment is a capital cost, requiring an up-front payment. Ruck 2010 included in the review used a Vibraflex Home Edition II, but a cost for this particular vibration plate was unable to be sourced. Subsequrntly, a cost of approximately £500 (excluding VAT) was used to inform the model based on the Committee’s experience with vibration equipment. It was also assumed that the equipment can be used for approximately 5 years before it needs to be replaced. Table 46 below presents the parameters used to calculate the equivalent annual cost.

If each child or young person with cerebral palsy was provided with their own personal vibration equipment the cost per person per year would equal £107. However, in clinical practice the equipment would be purchased by a school, child development centre or cerebral palsy centre to be used by all of their eligible members with cerebral palsy. In this case a cost per person per annum can be calculated based on the typical use of the equipment per annum. The cost per person will vary depending on the usage of the equipment i.e. the more the equipment is used the lower the cost per person. The formula for calculating the cost per visit is:

As can be seen from Table 47, the cost of the vibration device is negligible compared to the cost of the visit with a physiotherapist. According to the Committee, vibration therapy would need to be used 5 times per week for the benefits of this intervention to be seen with regards to BMD; however, they acknowledged that this could be quite a burden to the child and young person with cerebral palsy and their family or carers.

Based on those assumptions, the cost per year per person based on 5 sessions per week would equal approximately £4,682. To reflect the range of vibration plates available, a sensitivity analysis varying the upfront capital cost by ±50% was undertaken (Section G.9.7.1 and Section G.9.9.2).

As an aside, the Committee noted that vibration therapy can aid exercise programmes to provide benefits beyond BMD that may entail less intensive regimens. Moreover, vibration therapy may be used alongside other interventions to prevent reductions in BMD; however, combined interventions were not identified in the clinical evidence review.

Active exercise programmes

The programmes included in the trials were weight-bearing exercise and home-based virtual cycling. Other forms of active exercise, such as rebound therapy (trampolining), sports or other fitness equipment such as treadmills, were not identified in the clinical evidence review. The cost of these programmes including the cost of equipment would be similar to cycling, but there is no evidence to show equivalent efficacy; hence the model focussed on cycling as the active exercise programme.

To explore the cost-effectiveness of active exercise without the need to purchase equipment, a sensitivity analysis was undertaken with zero equipment cost (Section G.9.7.1 and Section G.9.9.2).

The Committee agreed that weight-bearing exercise and home-based virtual cycling could be performed without supervision from a health care professional at the child or young person’s home or school. The activities would require at least 1 initial visit from a physiotherapist to teach the child or young person with cerebral palsy and their family or carers how to perform the techniques. Thereafter, the physiotherapist would make follow-up visits ideally every 3 months to assess the impact of the programme and modify as necessary. According to the PSSRU each 30 minute visit from a community occupational therapist or physiotherapist would cost approximately £18.

The trials included in the clinical evidence review performed exercise up to 5 days per week. Daily regimens would be ideal for the benefits of the programmes to be realised, but the Committee noted that families would not be able to follow such intensive regimens. The most families would be expected to achieve is 2 to 3 sessions per week as they may have additional programmes to follow. Overall, this could be quite a burden to the child and young person with cerebral palsy and their family or carers, unless it is something that they choose to do, or enjoy.

On the other hand, active exercise programmes could be incorporated into normal school sports or play, which could increase adherence. Table 48 below presents the cost of an active exercise programme that involves cycling. An exercise bike is assumed to cost £200, but over the lifespan of the equipment (5 years) the equivalent annual cost would be £43. Assuming each child or young person with cerebral palsy has their own bike and has input from their physiotherapist 4 times a year, the total cost per annum is approximately £133.

Excluding the cost of a bike, weight bearing activities are assumed to cost £90 per annum based on an equivalent monitoring schedule with a physiotherapist.

G.9.6.1. Supplementation

G.9.6.2. Bisphosphonates

G.9.6.3. “No treatment”

The Committee noted that “no treatment” would not incur zero cost in clinical practice as children and young people with cerebral palsy would undergo bone, blood, liver, renal, and vitamin D investigations at least every 12 months. Moreover, these investigations would not be common practice across all the comparators under consideration. The cost of the individual tests would be relatively inexpensive (NHS Reference Costs 2014/15, Pathology services, DAPS05, Haematology, £3). Including staff time, the Committee advised a cost of £30 would be reasonable to inform the model.

G.9.6.4. Summary of treatment costs

Table 57 below presents the total annual costs included in the model for each intervention.

G.9.7.1. Deterministic sensitivity analysis

A series of scenario analyses were undertaken in order to test how sensitive the results are to uncertainty in individual parameters. Parameters varied in the scenario analysis were chosen on the basis of uncertainty in their estimation or the potential impact that they had on the results. The values varied, along with their rationale are shown in Table 58.

G.9.7.2. Probabilistic sensitivity analysis (PSA)

PSA was conducted in the model to take account of the simultaneous effect of uncertainty relating to model parameter values. Key parameters were varied by sampling from probability distributions.

The model was run for 1,000 simulations to generate estimates of total costs and QALYs for each treatment arm by varying probabilities, costs and utilities simultaneously. Those simulations are presented on cost-effectiveness planes with a willingness-to-pay (WTP) threshold of £20,000 per QALY and over several thresholds using cost-effectiveness acceptability curves (CEACs). The model structure and model settings were kept constant. Drug costs are known with certainty hence only the total treatment cost has been varied which includes monitoring, and/or the purchase of equipment.

A beta probability distributions was employed for probabilities and utilities, whilst a gamma probability distribution was employed for costs. It is important to note that utilities are assumed to be bound by 0 and 1; hence values worse than death are not possible, consequently the probabilistic values are positively skewed towards 0 given the low GMFCS levels under consideration. A starting point for unknown data was to assume the value of the standard deviation was 20% of the expected input parameter mean – such conservative methods are often used by manufacturers in Technology Appraisal submissions (such as NICE TA354 and NICE TA327) when the distribution of the data is not available. PSA parameters are provided in Table 67.

If the underlying probability distribution is unknown then PSA may be less informative in quantifying the uncertainty arising from model inputs. However, undertaking PSA was considered to be useful to the Committee to illustrate the possible simulations that could result from the assumptions made in the model.

G.9.9.1. Base case

Table 59 below presents the total costs and total QALYs associated with each intervention in the model. It is important to note that the total costs incorporate the cost of treatment to prevent reduced BMD (Table 57) plus the expected cost of fracture treatment (Table 44).

For example, if 1 child or young person with cerebral palsy enters the population at increased risk of reduced BMD and receives vitamin D as an intervention the cost of treatment is £53. They have a 10.8% probability of receiving fracture treatment at a cost of £7,104 which results in an expected fracture cost of £765 (10.8% x £7,104). As a result, the total expected cost for that person over 1 year is £828 (£63 + £765).

The same principle of expected values are used to calculate the total QALYs. Our child or young person with cerebral palsy in the first line population receiving vitamin D will experience a fracture with a probability of 10.8% associated with a utility of 0.3 or not experience a fracture a probability of 89.2% associated with a utility of 0.5. Therefore, over 1 year that person will expect to accrue a total of 0.4785 QALYs (0.108 × 0.3 + 0.892 × 0.5).

It is important to note that the difference in total QALYs across the treatments is small and can lead to substantial changes in the resulting ICER.

Only pair-wise comparisons were estimated in the model because there is insufficient clinical data to present a fully incremental comparison for one BMD site; consequently the site of BMD is a user input in the model when more than one site is available for each treatment. Furthermore, interventions are not mutually exclusive in the population at increased risk of reduced BMD, and there is no evidence on their combined effect. Consequently, there are a substantial number of comparisons that can be made for the populations at increased risk of reduced BMD and with proven osteoporosis. For pragmatic reasons the Committee was asked to prioritise their comparisons of interest in those populations:

1. Population at increased risk of reduced BMD:

   • vitamin D plus calcium vs. “no treatment”
   • aerobic exercise (cycling) vs. “no treatment”
2. Proven osteoporosis population:

   • risedronate (plus vitamin D) vs. pamidronate disodium

For those comparisons, cost-effectiveness planes (CE planes) are presented in addition to the ICER. Sensitivity analysis was also conducted on those comparisons (Section G.9.7 and Section G.9.9.2).

Site of fracture

For any given Z-score, the lumbar spine reported a higher probability of fracture than the distal femur (F1). When the Z-score gets bigger (more negative) the difference in probability of fracture estimated between the 2 sites increases. If this difference is large enough, the resulting ICER can change substantially when a lumbar spine site is compared to a F1 site.

When the probability of fracture increases, the expected QALYs are reduced and the expected cost of fracture treatment is increased. Consequently, a treatment that was dominant (less expensive and more effective) or had a positive ICER (more expensive and more effective) under a F1 measure could become a less expensive and less effective, or dominated (more expensive and less effective) under a lumbar spine measure if the lumbar spine is not associated with a lower (less negative) Z-score.

Because interventions with an F1 measure may be considered favourable compared to those with only a lumbar spine measure. Therefore, comparisons using equivalent sites are preferable to comparisons using different sites, when possible.

Population at increased risk of reduced BMD

The interventions under consideration in this population are not mutually exclusive; however, combined interventions were not identified in the clinical evidence review. Only comparisons against “no treatment” are discussed and presented here (Table 60). All remaining comparisons estimated by the model are reported in in Table 61.

Table 60Population at increased risk of reduced BMD, base case results (ICER)

View in own window

Intervention ▼vs. ▶		“No treatment”
		F1	Lumbar
Active exercise (cycling)	F1	£95,041	£6,261
	Lumbar	Dominated a	Dominated a
Weight-bearing exercise	F1	Dominant b	Dominant b
	Lumbar	NC	NC
Vitamin D	F1	Dominant c	Dominant c
	Lumbar	NC	NC
Vitamin D plus calcium	F1	NC	NC
	Lumbar	Dominant d	Dominant d
Vibration therapy	F1	£2,044,947	£1,154,538
	Lumbar	£5,288,632	£1,791,107

ICER, incremental cost-effectiveness ratio; NC, not calculable

(a)

Cycling more expensive and less effective than “no treatment”

(b)

Weights less expensive and more effective than “no treatment”

(c)

Vitamin D less expensive and more effective than “no treatment”

(d)

Vitamin D plus calcium less expensive and more effective than “no treatment”

Intervention: exercise (cycling and weight-beating)

When the effectiveness of cycling is based on the lumbar spine site it is dominated (more expensive and less effective) by “no treatment”. However, when the effectiveness of cycling is based on the F1 site, cycling becomes more effective than “no treatment”. As a result, there are different conclusions according to the site of BMD used for “no treatment”. When F1 sites are compared, cycling would not be considered cost-effective under NICE’s advisory cost-effective threshold. Conversely, cycling would be considered cost-effective when compared to the lumbar site for “no treatment”. As stated previously, the results based on different BMD sites should be interpreted with caution. Figure 4 below presents the cost-effectiveness planes for cycling vs. “no treatment” for all BMD site comparisons.

Weight-bearing exercise would be considered cost-effective to limit reductions in BMD as it dominates (less expensive and more effective) “no treatment”.

Both weight-bearing exercise and cycling are active exercise programmes with negligible difference in their treatment costs. However, their difference in clinical effectiveness is evident as the risk of fracture under cycling is higher than weight-bearing exercise (lumbar, 12.4%; F1, 10.9% vs. F1, 10.1%), but with the caveat that fracture risks are point estimates with some underlying uncertainty as to their true value. From these results, exercise that aims to improve muscle tone from weight-bearing activities would be cost-effective relative to aerobic exercise that improves cardiovascular ability.

Figure 4CE plane, cycling vs. “no treatment”

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Intervention: vitamin D

Vitamin D would be considered cost-effective to limit reductions in BMD as it is dominates (less expensive and more effective) “no treatment”.

Intervention: vitamin D plus calcium

Vitamin D plus calcium dominates (less expensive and more effective) “no treatment”. However, it is important to note that this intervention is assumed to have a zero risk of fracture post-treatment in the model (Section G.9.3.5). Figure 5 below presents the cost-effectiveness planes for cycling vs. “no treatment” for all BMD site comparisons

Figure 5CE plane, vitamin D plus calcium vs. “no treatment”

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Intervention: vibration therapy

Vibration therapy would not be considered cost-effective to limit reductions in BMD as it is has an ICER substantially greater than NICE’s advisory cost-effective threshold. This is driven by the high treatment cost associated with vibration therapy, which cannot be offset by larger gains in clinical effectiveness.

Summary

Overall weight-bearing active exercise, vitamin D, or vitamin D plus calcium would be considered cost-effective interventions to limit reductions in BMD in a population of children and young people with cerebral palsy at increased risk of reduced BMD compared to “no treatment” based on the results presented in Table 60. On the other hand, cycling and vibration therapy may not be considered cost-effective compared to “no treatment”.

Children and young people with cerebral palsy with proven osteoporosis

Intervention: vitamin D

Vitamin D would be considered cost-effective, but whether this is a decision of disinvestment (less effective and less expensive) or dominance depends on the comparators site of BMD. When compared to F1 measures i.e. risedronate plus vitamin D or pamidronate disodium, vitamin D is less expensive and less effective. In other words, pamidronate disodium and risedronate plus vitamin D have ICERs above NICE’s advisory cost-effective threshold and neither would not be considered cost-effective relative to vitamin D.

Conversely, when compared to lumbar spine measures i.e. calcium plus vitamin D or pamidronate disodium, vitamin D is less expensive and more effective (dominant). As stated previously, for any given Z-score the lumbar spine reports a higher probability of fracture than the distal femur (F1); hence comparing F1 measures against lumbar spine measures could favour the former intervention. Overall, regardless of the site of BMD vitamin D always produces the lowest total cost.

Intervention: risedronate plus vitamin D

Risedronate plus vitamin D would not be considered cost-effective under NICE’s cost-effective advisory threshold when compared to vitamin D or vitamin D plus calcium.

Risedronate plus vitamin D would be considered cost-effective relative to pamidronate disodium. When F1 measures are compared, pamidronate disodium is more expensive and more effective than risedronate plus vitamin D, but the ICER is substantially above NICE’s advisory cost-effective threshold, hence pamidronate disodium would not be considered cost-effective. When risedronate plus vitamin D is compared to pamidronate disodium using a lumbar spine site, risedronate plus vitamin D is less expensive and more effective than pamidronate disodium, hence pamidronate disodium is dominated by risedronate plus vitamin D. Figure 6 below presents the cost-effectiveness plane for pamidronate disodium compared to risedronate plus vitamin D.

Figure 6CE plane, pamidronate vs. risedronate plus vitamin D

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Intervention: calcium plus vitamin D

Calcium plus vitamin D is dominated (more expensive and less effective) by vitamin D. However, calcium plus vitamin D is a lot cheaper and slightly less effective than pamidronate disodium and risedronate plus vitamin D leading to a decision of disinvestment (less expensive and less effective) with substantial cost savings per QALY loss. In other words, pamidronate disodium and risedronate plus vitamin D have ICERs above NICE’s advisory cost-effective threshold and neither would not be considered cost-effective relative to calcium plus vitamin D.

Intervention: pamidronate disodium

Pamidronate disodium would not be considered cost-effective relative to any of the interventions included in the model as it is either dominated or substantially above NICE’s advisory cost-effective threshold.

Table 61Proven osteoporosis base case results (ICER) for all comparisons

View in own window

Intervention		Vitamin D		Risedronate + vitamin D		Calcium + vitamin D		Pamidronate disodium
		F1	Lumbar	F1	Lumbar	F1	Lumbar	F1	Lumbar
Vitamin D	F1	-		£1,799,501 (SW)a	NC	NC	Dominant b	£1,371,205 (SW) c	Dominant d
	Lumbar			NC	NC	NC	NC	NC	NC
Risedronate plus vitamin D	F1	£1,799,501 (NE)	NC	-		NC	£81,047 (NE)	£1,344,044 (SW) e	Dominant f
	Lumbar	NC	NC			NC	NC	NC	NC
Calcium plus vitamin D	F1	NC	NC	NC	NC	-		NC	NC
	Lumbar	Dominated g	NC	£81,047 (SW) h	NC			£930,466 (SW) i	£4,782,071 (SW) i
Pamidronate disodium	F1	£1,371,205 (NE)	NC	£1,344,044 (NE)	NC	NC	£930,466 (NE)	-
	Lumbar	NC	NC	Dominated j	NC	NC	£4,782,071 (NE)

South-west ICERs represent the costs saved per QALY loss

NC, not calculable; NE, north-east quadrant on the cost-effectiveness plane (more expensive and more effective than the comparator); SW, south-west quadrant on the cost-effectiveness plane (less expensive and less effective than the comparator)

(a)

vitamin D less expensive and less effective than risedronate plus vitamin D

(b)

vitamin D less expensive and more effective than calcium plus vitamin D

(c)

vitamin D less expensive and less effective than pamidronate

(d)

vitamin D less expensive and more effective than pamidronate

(e)

risedronate plus vitamin D less expensive and less effective than pamidronate

(f)

risedronate plus vitamin D less expensive and more effective than pamidronate

(g)

calcium plus vitamin D more expensive and less effective than vitamin D

(h)

calcium plus vitamin D less expensive and less effective than risedronate plus vitamin D

(i)

calcium plus vitamin D less expensive and less effective than pamidronate

(j)

pamidronate more expensive and less effective than risedronate plus vitamin D

Standing frame

Caulton 2004 took BMD measures from the vertebrae and proximal tibia, however for reasons described in Section G.9.3.3 only the vertebral site could be included in the model from the data they reported.

They found a 6% mean increase in vertebral vTBMD in the intervention group; whereas the proximal tibial vTBMD in the intervention group showed a change of −0.85 mg/cm3 compared to the control group (95% CI −16.83 to 15.13; p = 0.92). Overall, these results suggested that a longer period of standing leads to a significant increase in vertebral but not proximal tibial vTBMD in non-ambulant cerebral palsy children, demonstrating that different BMD sites can lead to different conclusions.

The standing frame would not be considered cost-effective to prevent reduced BMD based on the vertebral site as the ICER is substantially higher than NICE’s advisory cost-effective threshold when compared to “no treatment” (Table 62). This combined with their findings that there were no significant difference pre- and post- intervention on the proximal tibial vTBMD reiterates that the standing frame should not be recommended to prevent reduced BMD. However, it is important to note that the normal standing duration has been used as a proxy for “no treatment” which may underestimate the incremental effectiveness of the standing frame. Figure 7 below presents the cost-effectiveness plane.

Table 62Standing frame results

View in own window

Treatment	Total cost	Total QALYs	Inc. cost	Inc. QALYs	ICER
“No treatment”	£2,475	0.7183	-	-	-
Standing frame	£7,165	0.7190	£4,691	0.0007	£6,896,073

ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life years

Costs and benefits discounted at 3.5%/year over a 5-year time horizon

Figure 7CE plane, standing frame vs. “no treatment”

QALYs, quality-adjusted life years; WTP, willingness-to-pay

G.9.9.2. Deterministic sensitivity analysis

Comparisons of interest in the population at increased risk of reduced BMD are limited to vitamin D plus calcium, cycling and “no treatment” and in the proven osteoporosis population risedronate plus vitamin D vs. pamidronate disodium, unless otherwise stated. Table 63 presents the resulting ICERs for these comparisons.

G.9.9.3. Probabilistic sensitivity analysis (PSA)

Comparisons of interest in the population at increased risk of reduced BMD are limited to vitamin D plus calcium vs. “no treatment”, and cycling vs. “no treatment”, and in the proven osteoporosis population risedronate plus vitamin D vs. pamidronate disodium. For the reasons described in Section G.9.9.1, only comparisons of equivalent BMD sites (i.e. F1 vs. F1 or lumbar spine vs. lumbar spine) will be presented here.

Population at increased risk of reduced BMD

Intervention: Vitamin D plus calcium

It is evident from Figure 8 and Figure 9 that all simulations predict vitamin D plus calcium to be cost-effective relative to “no treatment” under a threshold of £20,000 per QALY.

Moreover, assuming a WTP threshold of £20,000 or £30,000 per QALY there is a 100% probability that calcium plus vitamin D is optimal.

Vitamin D plus calcium has a deterministic input in PSA as the probability of fracture cannot fall below 0%. For this reason almost all simulations find calcium plus vitamin D to be more effective than “no treatment”.

Figure 8CEAC, vitamin D plus calcium (lumbar spine) vs. “no treatment” (lumbar spine)

WTP, willingness-to-pay

Figure 9CE plane, 1,000 simulations, vitamin D plus calcium (lumbar spine) vs. “no treatment” (lumbar spine)

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Intervention: active exercise (cycling)

With regards to cycling, the simulations in Figure 11 are distributed across all 4 quadrants of the cost-effectiveness plane, and almost half of simulations (46%) at a WTP threshold of £20,000 per QALY predict cycling to be cost-effective relative to “no treatment” when F1 sites are compared. As a result this questions if the deterministic ICER of £95,401 is overestimated. However, assuming a WTP threshold of £20,000 or £30,000 the probability that cycling is optimal is never greater than “no treatment” (Figure 10). Overall, the underlying probability distributions included in PSA are arbitrary due to insufficient data to estimate confidence intervals, also questioning their reliability.

Figure 10CEAC, cycling (F1) vs. “no treatment” (F1)

CEAC, cost-effectiveness acceptability curve; WTP, willingness-to-pay

Figure 11CE plane, 1,000 simulations, cycling (F1) vs, “no treatment” (F1)

QALYs, quality-adjusted life years; WTP, willingness-to-pay

When lumbar sites are compared in Figure 12 and Figure 13, the majority of simulations lie in the north-west quadrant, reflecting the base case where cycling is dominated by “no treatment”. Specifically, cycling is considered cost-effective relative to “no treatment” in 28% of simulations.

Simulations that are more expensive are generally less effective, this is because poor QALY gains are associated with the cost of fracture treatment; hence, the inverse relationship between QALYs and costs is to be expected in Figure 11 and Figure 13.

The simulations from both comparisons reflect the base-case uncertainty surrounding the chosen BMD sites, but the PSA has demonstrated that the cost-effectiveness of cycling may be more nuanced than implied by the deterministic results.

Figure 12CEAC, cycling (lumbar) vs. “no treatment” (lumbar)

CEAC, cost-effectiveness acceptability curve; WTP, willingness-to-pay

Figure 13CE plane, 1,000 simulations, cycling (lumbar) vs. “no treatment” (lumbar)

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Proven osteoporosis

All simulations in Figure 16 predict pamidronate disodium to be more expensive than risedronate plus vitamin D. In terms of drug treatment costs, risedronate plus vitamin D is a lot cheaper than pamidronate disodium (£601 vs. £7,103 /year) so it reasonable for none of the simulations to lie in the south quadrants. As a result, the probability of fracture or the cost of treating fractures would have to be a lot larger for risedronate plus vitamin D to have a greater total cost than pamidronate disodium.

In addition, pamidronate disodium is almost always shown to be more effective than risedronate plus vitamin D in Figure 16, with the majority of simulations lying in the (north-) east quadrant. This is reasonable because pamidronate disodium is always associated with a lower probability of fracture (where a fracture is associated with a disutility) than risedronate plus vitamin D (Figure 14). Overall, pamidronate disodium would not be considered cost-effective in any of the simulations under a £20,000 WTP threshold as the simulated ICER’s are much greater than this.

At all feasible WTP thresholds the probability that pamidronate disodium is optimal is zero (Figure 15).

Figure 141,000 simulated fracture probabilities for pamidronate vs. risedronate plus vitamin D

Figure 15CEAC, pamidronate (F1) vs. risedronate plus vitamin D (F1)

CEAC, cost-effectiveness acceptability curve; WTP, willingness-to-pay

Figure 16CE plane, 1,000 simulations pamidronate (F1) vs. risedronate plus vitamin D (F1)

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Postural management

In Figure 17, the standing frame is cost-effective relative to “no treatment” in 0% of simulations. Similarly, Figure 18 illustrates that at a WTP threshold of £20,000, or £30,000 the probability that the standing frame is cost-effective is zero.

Figure 17CE plane, 1,000 simulations, standing frame vs. “no treatment”

QALYs, quality-adjusted life years; WTP, willingness-to-pay

Figure 18CEAC, standing frame vs. “no treatment”

CEAC, cost-effectiveness acceptability curve; WTP, willingness-to-pay