20.3.1.1. Modelling direct medical costs

The comparison of costs for MDI and mixed insulin regimens aims to highlight differences in costs resulting from different insulin injection regimens. Costs related to the individual insulin dose that were assumed to be common to all treatment alternatives were, therefore, not included in the cost analysis. Costs for consumables required for self-monitoring of blood glucose levels and administration of insulin injections were included, as well as costs associated with the personnel involved in the initial diagnosis and review of treatment.

20.3.1.1.1. Staffing costs

The hourly rates for the healthcare professionals involved in instigating insulin treatment and supporting this process are presented in Table 79. The hourly rates are then multiplied by the total time input to provide an estimate of the total staff cost.

Table 79

Staffing hourly costs.

The costing of staff time was based on the following assumptions.

• Clinical staff and time inputs required at the time of diagnosis are the same for all insulin regimens.
• Time inputs from clinical staff are the same for twice and 3-times daily injection regimens.
• Dietetic and telephone advice for MDI is required more frequently than for twice or 3-times daily injections regimens.
• Dietetic advice for MDI (during home or school visits) requires more time inputs from a dietitian or paediatric diabetes specialist nurse (PDSN) to cover additional carbohydrate counting instruction.
• Time inputs from a multidisciplinary team (PDSN, dietitian, diabetologist and psychologist) during quarterly follow-up consultations are the same for all injection regimens and are, therefore, excluded from the cost analysis.

Table 80 presents an overview of all clinical staff involved at diagnosis and their respective time commitments. Table 81 gives a similar overview for the follow-up after initial diagnosis. The main difference between twice or 3-times daily injections and MDI regimens relates to more time-intensive dietary advice required for carbohydrate counting instruction for MDI both at home and at school. The final additional advice for MDI (delivered by telephone, email or text) is assumed to be delivered every other day for 1 additional month, by which time the individual dose is assumed to be stabilised and patients will not require any further formal advice.

Table 80

Time inputs of clinical staff for multiple daily injections and 2- or 3-times daily injections at diagnosis.

Table 81

Time inputs of clinical staff for MDI and 2/3 times daily injections during early follow-up.

20.3.1.1.2. Consumables

The unit costs of consumables used in insulin administration and self-monitoring of blood glucose (SMBG) are shown in Table 82.

Table 82

Unit costs of consumables.

Costs for consumables are assumed to be different for twice and 3-times daily injections and MDI regimens. Included are costs for consumables required for self-monitoring of blood glucose (SMBG) levels (strips for blood glucose testing and lancets) and administration of insulin injections (prefilled disposable pens and needles). Since meters for blood glucose testing are available free of charge, they are not included in the cost summary (see Table 83).

Table 83

Consumables used in insulin administration and self-monitoring of blood glucose per year.

The total cost of consumables is derived by multiplying the unit costs in Table 82 by the respective quantities in Table 83. A needle is required for each injection and it assumed that MDI consists of 5 daily injections (4 short-acting and 1 long-acting). Similarly, a blood glucose strip and new lancet are required for each self-monitoring blood glucose test and the values in Table 82 are based on 5 such tests per day, as per the recommendations in the guideline (see Section 7.6).

Details of how annual pen usage was estimated are given below. The guideline development group considered that, given the age range covered by the guideline, the total daily insulin dose would vary between 10 and 100 units. For cost purposes a total daily insulin dose of 50 units was assumed as this was thought to be a reasonable approximation of the modal daily dose. It was assumed that 2 units of insulin would be wasted per injection as airshots.

Estimating the annual number of prefilled disposable pens for twice daily (mixed) injections

It is assumed that each injection uses 27 units including 2 units as airshots. Each pen has 300 units in total.

View in own window

Number of injections per penl:	300÷27≈11 injections
Pen life:	11÷2=5.5 days
Pens per year:	365÷5.5≈66 pens

l

With 3 wasted units when the pen is close to empty (11×27=297 units)

Estimating the annual number of prefilled disposable pens for 3-times daily (mixed) injections

It is assumed that each injection uses 19 units including 2 units as airshots. Each pen has 300 units in total.

View in own window

Number of injections per penm:	300÷19≈15 injections
Pen life:	15÷3=5 days
Pens per year:	365÷5=73 pens

m

With 15 wasted units when the pen is close to empty (15×19=285 units)

Estimating the annual number of pre-filled disposable pens for multiple daily injections

It is assumed that a child or young person with type 1 diabetes would be on a daily dose of 25 units of short-acting insulin which would be taken as 4 injections. Therefore it was assumed that each injection would use 8 units including 2 units as airshots. Each pen has 300 units in total.

View in own window

Number of injections per short-acting penn:	300÷8≈37 injections
Short-acting pen life:	37÷4=9 days
Short-acting pens per year:	365÷9≈41 pens

n

With 4 wasted units when the pen is close to empty (37×8=296 units)

It is also assumed that the child or young person would be on a daily dose of 25 units of long-acting insulin which would be taken as a single injection. Therefore it was assumed that each injection would use 27 units including 2 units as airshots. Each pen has 300 units in total.

View in own window

Number of injections per long-acting peno:	300÷27≈11 injections
Long-acting pen life:	11÷1=11 days
Long-acting pens per year:	365÷11≈33 pens

o

With 3 wasted units when the pen is close to empty (11×27=297 units)

20.3.1.1.3. Overall treatment costs

The overall costs of 2- or 3-times daily injections and MDI regimens are summarised in Table 84 and Figure 1.

Table 84

First year treatment costs 2- or 3-times daily injections and multiple daily injection regimens.

Figure 1

Graph showing treatment costs of 2- or 3-times daily injections and multiple daily injections. Source: Costing tool developed for the 2015 update

In the cost effectiveness model described in Section 20.3.2, a treatment cost is required for the first year of treatment and for subsequent years. For the first year treatment costs the model uses the total costs for MDI and 3-times daily mixed insulin as shown in Table 84. For subsequent years the treatment cost is assumed to be given by the consumables costs (insulin and SMBG) shown in Table 84, which is £999 for MDI and £842 for mixed insulin.

In the review protocol (see Appendix E:) mixed insulin was defined as fewer than 4 injections per day. For costing purposes a 3-times daily injection was assumed as this more closely resembles current practice and it reflects the comparator used in the study that was used to estimate the effectiveness of MDI in the cost effectiveness model (Adhikari 2009).

20.3.2.1. Reasons for using the IMS Core Diabetes Model

Type 1 diabetes is an incurable condition with potential implications for health-related quality of life and longevity. Data from trials and observational studies are insufficient to quantify these long-lasting effects. However, the health economic approach adopted in the IMS Core Diabetes Model allows the effects to be modelled by taking data from a wide variety of sources and synthesising them to estimate the lifelong consequences and costs of comparator interventions.

Type 1 diabetes is a condition with many complications and associated comorbidities. Within the timeframe of a clinical guideline it was not possible to model the complex lifetime relationships and therefore the decision was taken across all NICE diabetes guidelines being updated in 2015 to use a bespoke model which was already in use and had been validated. A validation study of the IMS Core Diabetes Model has been recently published (McEwan 2014). Furthermore, the IMS Core Diabetes Model has been used previously in NICE Technology Appraisals (NICE 2008) and therefore its use can be considered to promote a consistent methodological approach to the evaluation of the cost effectiveness of interventions used for type 1 diabetes.

20.3.2.2. IMS Core Diabetes Model approach

The IMS Core Diabetes Model has been described in more detail elsewhere (Palmer 2004). The basic structure of the model is shown in Figure 2.

Figure 2

Schematic of model structure. Source: IMS health (reproduced with permission)

The model is accessed through the Internet and can be used to assess a wide variety of interventions in hypothetical cohorts of patients with either type 1 or type 2 diabetes. The model has a modular format with input parameters across a wide range of patient, clinical and economic characteristics. The input parameters used in this analysis are shown in Section 20.3.2.3.

The IMS Core Diabetes Model consists of 17 interdependent submodels, as shown in Figure 2, as part of a Markov modelling approach. In Markov models patients can be in one of a number of defined but mutually exclusive ‘health states’. Over time, or ‘cycles’, patients may transition to different health states. The developers of the IMS Core Diabetes Model have argued, especially in the context of diabetes, that this may not be realistic as patients can have any number of complications at the same time. To overcome this, the 17 submodels run in parallel allowing patients to develop multiple complications simultaneously across the duration of the simulation.

The model uses Monte Carlo simulation with results for 1000 patients simulated over 1000 iterations. At the start of an iteration, which in this model is at the point of diagnosis, a patient is assigned a number of characteristics such as HbA1c and systolic blood pressure. The patient's health state is them simulated over a maximum period of 90 years^p. At various points in the model the patient encounters ‘chance nodes’ where their progress – death, for example – is determined by a random number generator. Therefore, each patient in each iteration has a ‘random walk’ through the model. In each iteration summary data exist for the hypothetical cohort of 1000 patients and when the simulation has been run this summary exists across 1000 simulations from which summary results are then generated, such as mean QALYs and costs.

Non-parametric bootstrapping methods are used to sample from input parameter distributions where these have been specified with a non-zero standard deviation (SD). Cost effectiveness is presented as a ratio^q and therefore standard methods, such as univariate confidence intervals (CIs), cannot be used to quantify uncertainty around the point estimate. Bootstrapping is a statistical technique for estimating uncertainty of an estimator by repeat sampling from the original sample with replacement. This provides repeated estimates of cost and effect pairs from the MDI and 3-times daily injection interventions. The mean values of these estimates can then be used to calculate incremental cost effectiveness ratios (ICERs) underpinned by the underlying distribution of cost effectiveness. A more detailed description of the approach to probabilistic sensitivity analysis in the IMS Diabetes Core Model is described in more detail elsewhere (IMS CORE Diabetes Model Research Team and IMS Health Economics and Outcomes Research 2014, available from www.core-diabetes.com/).

Within the IMS Core Diabetes Model a number of approaches can be used to estimate QALYs, all of them derived as functions of the diabetes complications experienced in each year of the model along with any acute events that occur. The method presented here is the ‘Core default (minimum approach)’. In this method, if a patient has multiple events within 1 year the lower of the multiple health state utility values is used.

20.3.2.3. Model inputs

The various inputs into the base-case model are described below in Section 20.3.2.3.1 to Section 20.3.2.3.6.

20.3.2.3.1. Model population

The model was run for a population of young people aged 12 years. The incidence of type 1 diabetes follows a bimodal distribution (Felner 2005) with an initial peak in between 4 to 6 years and a bigger peak between 10 to 14 years. Therefore, the guideline development group considered that an age of 12 years at diagnosis was a reasonable age on which to base the model. The group did not consider that sampling from an age distribution would be particularly helpful as age at diagnosis would be unlikely to be an important driver of cost effectiveness given that type 1 diabetes is an incurable, lifelong condition and therefore all young people with type 1 diabetes are likely to derive similar long-term benefits from an intervention that is effective over their remaining lifespan.

The full baseline characteristics of the model population are given in Table 85.

Table 85

Model baseline characteristics.

Within the IMS Core Diabetes Model these input parameters are defined in the ‘Cohort module’. These parameters can be considered as falling into 1 of 2 types:

• continuous – for example HbA1c at baseline
• proportions – for example gender (in this case the characteristic of each patient in the simulation is generated randomly according to the proportion specified; so a patient in the model has a 51.2% chance of being male based on the reported proportion of males in the 10 to 14 age group in England and Wales).

20.3.2.3.2. Costs

The costs, other than costs associated with treatment (MDI or 3-times daily insulin injections in this case), are presented in Table 86. Costs were discounted at 3.5% per year as per the NICE Reference Case. A 20% variation in costs, the model default, was used for sampling in PSA.

Table 86

Model costs.

The costs of treatment are presented in Table 87. Section 20.3.1 described how these costs were derived.

Table 87

Model treatment cost in first year following diagnosis and in subsequent years.

20.3.2.3.3. Health state utilities

Within the IMS Core Diabetes Model are a number of health states, each of which has a health state utility attached. The health state utility is used to derive a QALY for each simulated patient where each health state utility is multiplied by the number of years lived in that state. Table 88 shows the health state utilities associated with model states and events. Event utilities are negative because they indicate that a lower health state utility will be experienced compared with the scenario where no such adverse event had occurred.

Table 88

Model health state utilities.

The CORE default method (minimum approach) was used. In this approach, if a patient has a history of multiple events, the model uses the lowest health state utility from the relevant comorbidities. This assumes that the other conditions will have a negligible further impact on quality of life over and above the most severe condition.

20.3.2.3.4. Ongoing management

Table 89 shows input parameters that relate to various facets of the ongoing management of type 1 diabetes.

Table 89

Management inputs.

20.3.2.3.5. Clinical inputs

These inputs in the IMS Core Diabetes Model come under what is termed the ‘clinical’ module and they are intended to capture the natural history of the disease, to reflect the relationships between risks and events and impact of interventions on risk.

Inputs entered as a proportion indicate the proportion of the simulated cohort who will experience a particular condition. So, for example, the value for ‘Proportion initial CHD event MI female’ would give the proportion of women who experienced a myocardial infarction as their first cardiovascular event. In addition, various risk adjustments can be varied as part of the clinical module.

Various clinical input parameters in the model are shown in Table 90 to Table 102.

Table 90

HbA1c adjustments.

Table 91

Systolic blood pressure adjustments.

Table 92

Myocardial infarction.

Table 93

Myocardial infarction mortality.

Table 94

Stroke.

Table 95

Stroke mortality.

Table 96

Heart failure.

Table 97

Angiotensin converting enzyme inhibitor adjustments for microvascular complications.

Table 98

Angiotensin converting enzyme inhibitor side effects.

Table 99

Probability of adverse events.

Table 100

Foot ulcer and amputation.

Table 101

Depression.

Table 102

Other.

In addition there are a number of input parameters governing transition probabilities for renal disease, eye disease, cardiovascular disease (CVD) and depression. There are probability inputs for non-specific mortality and physiological parameter progression tables. Existing values within the IMS Core Diabetes Model were used for all of these inputs with the exception of HbA1c progression which was modified in line with guideline development group opinion to better reflect the progression of this parameter in childhood. The progression specified is shown in Table 103 and in Figure 3.

Table 103

HbA1c progression for type 1 diabetes.

Figure 3

HbA1c progression for multiple daily injections (setting 1) and 3-times daily (mixed) injections (setting 2). Setting 1: multiple daily injections Setting 2: 3-times daily (mixed) injections

20.3.2.3.6. Treatment

The change in HbA1c as a result of treatment was based on a study that was included in the clinical review of the evidence undertaken for this guideline (Adhikari 2009). In this study, the 3-times daily injections group received mixed intermediate-acting insulin (neutral protamine Hagedorn, NPH) and rapid-acting insulin (lispro or aspart) at breakfast, rapid-acting insulin (lispro or aspart) at dinner and intermediate-acting insulin (NPH) at bedtime. Those on multiple daily injections received rapid-acting insulin (lispro or aspart) at mealtimes and a long-acting insulin (glargine) at bedtime. The guideline development group noted that different insulins were used in the different arms of the study, but the group's view was that although glargine may offer marginal benefit with respect to nocturnal hypoglycaemia, there is little evidence of sustained benefit in terms of improved HbA1c. The group also noted that glargine cannot be mixed with fast-acting insulin, meaning that it could not be used in a twice or 3-times daily (mixed) regimen. It is therefore impossible to truly compare like with like. Different insulins are a feature of the different injection regimens and the guideline development group therefore considered the study valid to inform the health economic model.

The change in HbA1c reported in this study between the MDI intervention and 3-times daily injections (mixed) from baseline at 12 months was used as the treatment efficacy. The inputs for treatment efficacy are outlined in Table 104 and are also shown graphically in Figure 3.

Table 104

Change in baseline HbA1c.

Within the IMS Core Diabetes Model these inputs occur within a ‘Treatment Module’. Within that module it was specified that the MDI intervention should have ‘intensive’ insulin therapy and that 3-times daily injections should have ‘conventional’ insulin therapy. Also specified in the treatment module was that ‘Framingham progression’ should be used for systolic blood pressure, total cholesterol, low density lipoprotein, high density lipoprotein and triglycerides. Adverse events were also specified for each treatment as outlined in Table 105. The guideline development group noted that it was not possible to adjust these adverse events by age and instead set them based on their experience of type 1 diabetes in children and young people, but recognised that these rates would not necessarily reflect actual event rates as the children and young people passed into adulthood. However, differences were not outlined between the different treatments and therefore the guideline development group did not consider that this would have an important impact on results.

Table 105

Adverse events.

Finally, risk adjustment for statins and angiotensin converting enzyme (ACE) inhibitors was selected within the ‘Treatment Module’.

20.3.2.4. Results

A PSA suggested that MDI was likely to be cost effective relative to 3-times daily (mixed) injections. Table 106 summarises key discounted results and an incremental comparison of the 2 interventions is given in Table 107. Figure 4 shows the results of all 1000 simulations displayed on the cost effectiveness plane and Figure 5 shows the cost effectiveness acceptability curve.

Table 106

Summary results.

Table 107

Incremental analysis of multiple daily injections relative to 3-times daily (mixed) injections.

Figure 4

Cost effectiveness plane showing incremental costs and quality of life years of multiple daily injections relative to 3-times daily (mixed) injections.

Figure 5

Cost effectiveness acceptability curve.

Figure 5 shows the probability of MDI being cost effective as the willingness to pay for a QALY varies. According to this PSA, MDI has almost a 90% chance of being cheapest and, as Figure 4 shows, nearly all of the simulations produced an incremental QALY gain for MDI. The probability of MDI being cost effective at a willingness to pay of £20,000 per QALY is 98.6%, suggesting a very high likelihood that MDI is cost effective compared with 3-times daily injections at a cost effectiveness threshold often used by NICE.

A range of other outputs from the model are presented in Figure 6 to Figure 13. Setting 1 denotes MDI and setting 2 denotes 3-times daily (mixed) injections.

Figure 6

Breakdown of total costs.

Figure 7

Breakdown of direct costs over time.

Figure 8

Cumulative incidence of eye disease over time.

Figure 9

Cumulative incidence of renal disease over time.

Figure 12

Cumulative incidence of CVD over time.

Figure 13

Cumulative patient survival over time.

20.3.2.5. Discussion

Driving the results of this model is the treatment effect derived from a single US study (Adhikari 2009) showing a significant reduction in HbA1c with MDI compared with 2- to 3-times daily injections in children and young people with newly diagnosed type 1 diabetes at 1 year after baseline. The model assumes that the resulting differential in HbA1c will be lifelong (see Figure 3). If the differential is eroded over time then this model will tend to overstate the cost effectiveness of MDI. Also, the study has a high risk of bias because patients were allocated to the study groups based on physician preferences and therefore confounding variables may also explain some or part of the treatment effect observed.

In this context it should also be noted that the evidence review conducted for this guideline failed to find evidence of benefit of MDI in those with a diagnosis of over 1 year, although it may be that MDI works better in the newly diagnosed when a change of injection strategy is not an issue.

It should be remembered that the different cumulative incidence of particular events is a function of life expectancy as well as diabetic control and so, as in the case of CVD for example, the overall cumulative incidence is higher with MDI because of higher life expectancy (see Figure 10 and Figure 11).

Figure 10

Cumulative incidence of ulcer over time.

Figure 11

Cumulative incidence of cardiovascular disease over time.

20.3.2.6. Conclusion

The results of this analysis strongly suggest that MDI is cost effective relative to a 3-times daily (mixed) regimen. With base-case inputs the model suggested that MDI gave net QALY gains when compared with 3-times daily injections. Furthermore, PSA suggested that there was a high probability that MDI, despite its higher treatment cost, would lead to lower health service costs as a result of reduced incidence of long-term complications.

The model results are based on a population of children and young people starting treatment with newly diagnosed type 1 diabetes. The clinical review did not find evidence of benefit of MDI compared with regimens with fewer than 4 injections per day in a population of children and young people with type 1 diabetes who began using MDI treatment 1 year after diagnosis (see Section 6.1.2.6.8). Therefore, the analysis presented here does not demonstrate the cost effectiveness of MDI in such populations. The rationale for not restricting the recommendation to just those with newly diagnosed type 1 diabetes is presented in Section 6.1.2.6.8.

20.4.1.1. A ‘what-if analysis’

The clinical evidence review did not find any randomised studies that compared different frequencies of finger-prick testing. Therefore, data from observational studies were included and these largely reported outcomes in terms of correlation statistics. However, finding a correlation between increased frequencies of finger-prick testing does not imply causation. It may be that more frequent monitoring leads to better control of blood glucose, as the patient has more data points on which to base action to control blood glucose. On the other hand, it is possible that better motivated patients test more frequently as they have a greater interest in achieving adequate long-term blood glucose control. In that case a positive relationship between the frequency of testing and outcomes may simply reflect that better motivated patients with better blood glucose control test more often and their better blood glucose control reflects their motivation in all aspects of their self-management rather than the extra information derived from the increased finger-prick frequency.

Given this limitation in the data it was decided that a ‘what-if’ analysis would be undertaken. The base-case analysis assumes that any correlation observed between lower HbA1c and increased finger-prick testing is indeed causal and the cost effectiveness of different frequencies is evaluated on that basis. Given the uncertainty as to whether this is a truly or wholly causal effect, additional sensitivity analyses were undertaken to assess how sensitive any conclusion about cost effectiveness is to the magnitude of the observed correlation. The sensitivity analysis took the form of a threshold analysis to determine how much an additional finger-prick test would have to reduce HbA1c for it to be considered cost effective.

20.4.1.2. IMS Core Diabetes Model

The modelling was undertaken using the IMS core diabetes model. This model is described in Section 20.3.2 and its use within the context of this guideline is explained in the analysis comparing the cost effectiveness of MDI with 3 times daily (mixed) injections. The inputs and assumptions used in this model are the same as those used in the MDI versus mixed insulin model unless otherwise stated.

20.4.1.3. Treatment

Studies included in the clinical evidence review for blood glucose monitoring were considered to inform the impact of finger-prick testing frequency for this analysis. A decision was made not to use data from studies containing less than 1000 participants, given the much larger numbers present in other included studies.

Upon reviewing the literature, a decision was made to assess the frequency of monitoring blood glucose levels up to 5 times per day as evidence from a large German database (n=26,723) suggested no further improvement in metabolic control occurs beyond testing 5 times per day (Ziegler 2011).

In the base-case analysis data from a US study were used to estimate the change in blood glucose levels from increased monitoring (Miller 2013). Evidence from participants aged up to 18 years are included (n=11,641) with results presented for children and young people in the following age categories:

• 1 to less than 6 years
• 6 to less than 13 years
• 13 to less than 18 years.

The mean HbA1c across these age categories for different frequencies of SMBG were reported as shown in Table 108.

Table 108

Reported association between daily frequency of self-monitoring of blood glucose and mean HbA1c by age.

The mean value of HbA1c across age categories was used and was assumed to apply to the mid-point of the reported SMBG ranges. Linear interpolation was then applied to estimate the mean HbA1c for each daily finger-prick test increment as shown in Table 109 and Figure 14.

Table 109

Estimated change in mean HbA1c from increased daily frequency of self-monitoring of blood glucose.

Figure 14

Graph to show estimated association between HbA1c and daily self-monitoring blood glucose (SMBG) frequency. Source: Estimated from Miller 2013

The reduction in HbA1c is summarised in Table 110.

Table 110

Estimates of reduction in blood glucose level from increased monitoring per day.

20.4.1.4. Costs

The costs of different frequencies of finger-prick testing used in the model are shown in Table 111. These are based on the costs estimated in Section 220.3.1.1.2.3.1.1.3 (Table 82 and Table 82) for the model comparing the cost effectiveness of MDI versus mixed insulin injection regimens.

Table 111

Costs of finger-prick testing per year.

In addition to the treatment costs the model also estimates lifetime complication costs. Costs in the model were taken from an NHS and Personal Social Services (PSS) perspective as per the NICE Reference Case.

20.4.3.1. Comparison across self-monitoring of blood glucose frequencies from 0 to 5 times daily

The results from the base-case analyses are shown in Table 112 and Table 113, and Figure 15 and Figure 16.

Table 112

Total costs and quality adjusted life years with different daily finger-prick testing frequency.

Table 113

Incremental costs and effects of increasing frequency of blood glucose testing.

Figure 15

Graph to show costs associated with different frequencies of finger-prick testing.

Figure 16

Graph to show quality adjusted life years associated with different frequencies of finger-prick testing.

20.4.3.2. Comparison of 4-times daily self-monitoring of blood glucose with 5-times daily self-monitoring of blood glucose

Figure 20 summarises key discounted results and Figure 17 shows the results of all 1000 simulations displayed on the cost effectiveness plane. This suggests that 5-times daily finger-prick testing dominates 4 times daily finger-prick testing, being cheaper and producing longer life expectancy and QALYs.

Figure 20

Cumulative incidence of renal disease.

Figure 17

Cost effectiveness plane (origin 4-times daily testing). QALE quality adjusted life expectancy

A breakdown of costs is displayed in Table 115, showing that the higher monitoring costs of 5-times daily finger-prick testing are more than offset by a reduction in lifetime complication costs, even allowing for the increased life expectancy with 5-times daily testing.

Table 115

Breakdown of total costs.

A range of other outputs from the model are presented in Figure 18 to Figure 24. Setting 1 denotes 4-times daily finger-prick testing and setting 2 denotes 5-times daily finger-prick testing.

Figure 18

Breakdown of total costs.

Figure 19

Cumulative incidence of eye disease.

Figure 21

Cumulative incidence of ulcer.

Figure 22

Cumulative incidence of cardiovascular disease.

Figure 23

Survival curve.

20.4.3.3. What-if analysis

Table 116 shows the incremental costs and QALYs of 5-times daily finger-prick testing relative to 4-times daily finger-prick testing, varying the percentage point reduction in HbA1c as a result of the increased testing. This suggests that 5-times daily finger-prick testing is cost effective compared with 4-times daily finger-prick testing, provided that the additional test leads to a greater than 0.06 percentage point reduction in HbA1c when assessing cost effectiveness at a willingness to pay threshold of £20,000 per QALY.

Table 116

Incremental costs and quality adjusted life years of 5-times daily finger-prick testing relative to 4 times per day finger-prick testing varying the reduction in HbA1c.

20.5.2.1. Population

The model is developed for a population of children and young people with type 1 diabetes.

20.5.2.2. Comparators

Usually it would be important to establish that any form of monitoring represented a cost effective use of resources, but since monitoring of ketones is considered good practice for the prevention of DKA monitoring and forms part of current clinical practice, the 2015 update does not need to include an option for no monitoring. Therefore, this model is restricted to a comparison of urine and blood ketone monitoring.

20.5.2.3. Analysis type

Only 1 of the guideline development group's priority outcomes was reported in the included study (Laffel 2005) and this related to hospital admission. This meant it was not possible to undertake a cost-utility analysis, which is the preferred NICE approach: a cost minimisation approach was undertaken instead based on treatment costs and the costs of hospitalisation. A schematic of the model is presented in Figure 25.

Figure 25

Decision tree for urine ketone monitoring versus blood ketone monitoring.

20.5.2.4. Hospitalisation events

Table 117 presents DKA-related hospital admissions for a 6-month period using data from 1 US study (Laffel 2005). The guideline development group was of the opinion that these hospital admission rates were higher than would be experienced in England and Wales. From their own experience and the National Paediatric Diabetes Audit Report 2011-12, Part 2, Hospital Admissions and Complications, they thought that an admission rate of 5% to 10% per year (or equivalently 5 to 10 per 100 patient years) would be more typical in patients for whom this guideline is intended. The guideline development group considered that there might be lower rates in England and Wales compared with the USA because of easier access to advice. Most paediatric units in England and Wales offer 24-hour advice and so when children and young people are ill they can often be cared for at home. Geography may also play a part as distance from home to diabetes units is shorter in England and Wales than in the USA. There is also an option of GP involvement in the care of children and young people at an early stage of illness. Additionally, the costs of medical care in the USA may act as a deterrent to advice being sought at an early stage. The effects of lower admission rates are explored in a sensitivity analysis (Section 20.5.3.3.1).

Table 117

Hospital admission for 6 months.

Table 118 shows the model parameters derived from the data, including those used for PSA.

Table 118

Model hospitalisation parameters.

20.5.2.5. Costs

The costs are based on an NHS perspective and are for a cost year of 2012/13.

Cost inputs relate to the cost of ketone testing and different forms of hospital admission. For costing purposes the guideline development group suggested that a hospitalisation admission should be considered as a stay in a high dependency unit of 2 to 3 days and that an emergency room admission should be considered as a 24-hour stay on a paediatric ward. The model allows for the costs of a follow-up consultation to be included, although this does not form part of the base-case analysis. The model also allows the user to choose alternative types of paediatric intensive care, but the base-case is set to high dependency paediatric intensive care. A cost for a paediatric ward stay was not found in NHS Reference Costs and therefore the NHS Reference Cost for ‘Diabetic mellitus with ketoacidosis or coma’ was used as a proxy. The unit costs are summarised in Table 119.

Table 119

Unit costs.

Table 120 presents further assumptions with respect to resource use.

Table 120

Ketone monitoring pack use and hospital length of stay.

20.5.2.6. Sensitivity analyses

To assess the robustness of the base-case results and to take account of data uncertainty, a number of sensitivity analyses were undertaken. These involved a mixture of 1-way sensitivity analyses (where a single model parameter is changed), 2-way sensitivity analyses and PSA using Monte Carlo simulation methods. PSA was undertaken for base-case data inputs and for each 1-way sensitivity analysis. Each PSA calculates the probability that blood ketone monitoring will be cost effective compared with urine ketone monitoring.

A ‘what-if’ analysis was used to assess the impact of assigning a QALY loss to an emergency room admission and hospitalisation.

20.5.3.1. Base-case analysis

The base-case deterministic results are summarised in Table 121 and depicted graphically in Figure 26. The results suggest that blood ketone monitoring is £519 cheaper per patient than urine ketone monitoring.

Table 121

Costs of blood and urine ketone monitoring.

Figure 26

Graph comparing costs of blood and urine ketone monitoring.

A PSA of 1000 simulations suggested that blood ketone monitoring was cheaper than urine ketone monitoring in 97.8% of simulations (see Figure 27). The diagonal line in Figure 27 indicates the threshold of equal cost for both monitoring strategies.

Figure 27

Probabilistic sensitivity analysis results based on base-case inputs.

20.5.3.2. One-way sensitivity analyses

20.5.3.2.1. Including outpatient follow-up

The results of 1-way sensitivity analysis including outpatient follow-up are shown in Table 122 and, for the PSA, in Figure 28 The probability that blood ketone monitoring was cheaper than urine ketone monitoring was 96.9% in this PSA.

Table 122

Costs of blood and urine ketone monitoring when including the costs of an outpatient follow-up appointment.

Figure 28

Probabilistic sensitivity analysis based on base case inputs and including outpatient follow-up.

20.5.3.2.2. Including paediatric intensive care follow-up

The results of 1-way sensitivity analysis including paediatric intensive care follow-up are shown in Table 123 and, for the PSA, in Figure 29. The probability that blood ketone monitoring was cheaper than urine ketone monitoring was 97.5% in this PSA.

Table 123

Costs of blood and urine ketone monitoring when including the costs of a paediatric intensive care follow-up appointment.

Figure 29

Probabilistic sensitivity analysis based on base-case inputs and including paediatric intensive care follow-up.

20.5.3.2.3. Reducing the rate of emergency admission with urine ketone monitoring to 13% per year

In this analysis the rate of emergency admission with urine ketone monitoring is reduced to 13%, approximately half that of blood ketone monitoring while maintaining all other inputs at their default values. The results of this analysis are shown in Table 124 and, for the PSA, in Figure 30. The probability that blood ketone monitoring was cheaper than urine ketone monitoring was 85.4% in this PSA.

Table 124

Costs of blood and urine ketone monitoring when reducing the rate of emergency admission with urine ketone monitoring.

Figure 30

PSA based on base-case inputs but reducing the rate of emergency admission with urine ketone monitoring.

20.5.3.2.4. Eight blood ketone monitoring packs per year

This sensitivity analysis more than doubles blood ketone monitoring costs while maintaining other inputs at their default values. The results are given in Table 125 and, for the PSA, in Figure 31. The probability that blood ketone monitoring was cheaper than urine ketone monitoring was 94.5% in this PSA.

Table 125

Costs of blood and urine ketone monitoring when doubling the number of blood ketone monitoring packs.

Figure 31

PSA based on base-case inputs but doubling the number of blood ketone monitoring packs per year.

20.5.3.3. Two-way sensitivity analysis

Although 1-way sensitivity analysis is useful in demonstrating the impact of varying 1 parameter in the model, changing 2 variables simultaneously and examining their relationship may aid interpretation when investigating uncertainty around the estimated results. It can be used to show a threshold where blood or urine ketone monitoring becomes cheaper and a view can be made as to whether the threshold value could plausibly be crossed given existing uncertainty.

The model allows 2 parameters to be varied simultaneously. Using a Visual Basic® macro 10,000 combinations of these parameters were compared. A lower and upper limit is set for each parameter and 100 equally spaced parameter values between these limits are evaluated (100×100).

Results are shown graphically as a threshold analysis for each combination of values within a given range. Any 2 of the following parameters can be selected for 2-way sensitivity analysis:

• rate for emergency room admission – blood
• rate for emergency room admission – urine
• rate for hospitalisation – blood
• rate for hospitalisation – urine
• number of test packs – blood
• number of test packs – urine
• cost per pack of test strips – blood
• cost per pack of test strips – urine
• high dependency unit cost
• length of stay.

A number of 2-way sensitivity analyses are described below.

20.5.3.3.1. Varying the emergency room admission and hospitalisation rate for blood ketone monitoring

The results for this 2-way sensitivity analysis are shown in Figure 32. It shows the combinations of emergency room admission rates and hospitalisation rates with blood ketone monitoring compared with blood ketone monitoring holding all other inputs constant at their base-case values. The base-case inputs for these 2 variables falls well within the green shaded cost effectiveness region.

Figure 32

Graph of 2-way sensitivity varying emergency room admission and hospitalisation rate for blood ketone monitoring.

In addition, further probabilistic sensitivity analysis demonstrated that the probabilities of ketone monitoring being cost effective relative to urine ketone monitoring were 72.2% and 68.4% if both emergency room and hospitalisation rates were assumed to be 25% and 20% of the levels reported in the trials, respectively.

20.5.3.3.2. Varying the emergency room admission rate for blood and urine ketone monitoring

Figure 33 shows the impact of varying the emergency room admission rate for blood and urine ketone monitoring. Again, the default emergency admission rates lie well within the green shaded area for cost effectiveness.

Figure 33

Graph of 2-way sensitivity varying emergency room admission rates for blood and urine ketone monitoring.

20.5.3.3.3. Varying the cost of a blood ketone pack and the hospitalisation rate for blood ketone monitoring

In this sensitivity analysis the cost of a pack of blood monitoring ketone strips is varied along with the blood ketone monitoring hospitalisation rate and the results are displayed in Figure 34. It should be noted that the upper limit of the cost of a pack of blood monitoring ketone strips has been set to a level well in excess of the default value (over which there is no uncertainty). However, the analysis does show the trade-off between these variables necessary to retain cost effectiveness.

Figure 34

Graph of 2-way sensitivity varying the cost of a blood ketone pack and the hospitalisation rate for blood ketone monitoring emergency.

20.5.3.4. ‘What-if’ quality adjusted life year analysis

A QALY loss of 0.003 was assigned to a hospitalisation and an emergency room admission, while keeping all other inputs at their base-case values. The value of this QALY is not evidence based but is designed to reflect the fact that DKA is a short-lived medical emergency. The value of the ‘what-if’ QALY loss used in this sensitivity analysis might be interpreted as upper bound estimate of the QALY loss associated with a DKA hospital admission. It is based on a health state utility equivalent to death but experienced only for a period of 24 hours.

The results of this analysis are shown in Table 126.

Table 126

Incremental cost effectiveness of blood ketone monitoring assuming a quality adjusted life year loss is attributable to an admission.