Clinical effectiveness and cost-effectiveness of continuous subcutaneous insulin infusion for diabetes: systematic review and economic evaluation

Normal blood glucose control

Glucose is the primary source of fuel for cells in the body. Carbohydrate in food is metabolised to glucose within hours of ingestion, and is absorbed from the blood into cells for use as fuel. Uptake of glucose into cells is regulated by the hormone insulin, which is released from the pancreas in response to rising blood glucose levels. Insulin also regulates the use of glucose by cells, so if there is insufficient insulin, or if cells do not respond properly to insulin (insulin insensitivity or resistance), then glucose is not used efficiently by cells, in terms of either energy or storage.

Type 1 diabetes

In type 1 diabetes mellitus (T1DM), formerly known as ‘insulin-dependent diabetes’, all or nearly all of the β-cells in the pancreas, which produce insulin, have been destroyed, usually by an autoimmune process. The cause is not known. People with T1DM have little or no ability to produce their own insulin, and would die without insulin, and so they have to inject insulin for the rest of their lives. T1DM usually starts in children or young adults, but it can have onset at any age. The incidence (number of new cases per year) has risen considerably over recent decades. Scottish data show that the rate in children has more than trebled over the last 30 years (Figure 1). The rise has been greater in the youngest age group, although absolute numbers are smaller. 1

FIGURE 1.

Scottish Study Group (SSG) incidence by age at diagnosis: 1984–2003.

Similar rises have been reported from the Oxford region by Wilson et al. (2007). 2 The incidence of childhood diabetes (under the age of 15 years) rose from 17 per 100,000 in 1985–90 to 26.5 in 2003–4. The greatest increase was in the under-5s, in which the number who had diabetes by the age of 5 years rose fivefold.

Type 2 diabetes

Type 2 diabetes (T2DM), formerly known as ‘non-insulin-dependent diabetes’ or ‘maturity-onset diabetes’, comes on later in life than T1DM. It used to be seen almost exclusively in people over 45 years old, and was associated with overweight or obesity, but with rising prevalence of obesity it is now increasingly being seen at younger ages and even in children. Some ethnic groups, such as South Asians, have earlier onsets. The York and Humber Public Health Observatory (YHPHO) website (a very useful compendium of data on diabetes) estimates that the total prevalence of T2DM in England is 4.3%, although this includes people with undiagnosed diabetes. 3 [Some people with T2DM, perhaps 20%, have no symptoms and do not know they have it. Hence the current debate on screening, which is covered by another Health Technology Assessment (HTA) report. 4] The YHPHO estimated that the total prevalence of diabetes would rise by 15% between 2001 and 2010, with a 6% increase due to the ageing population and a 9% increase due to increasing obesity. 5

Type 2 diabetes usually starts with insulin resistance, related to overweight, with the pancreas producing more insulin than usual to overcome the resistance. Over time, the pancreas fails to produce enough, insulin production falls, blood glucose rises further, and clinical diabetes ensues. In the UK Prospective Diabetes Study (UKPDS), patients’ insulin production had fallen to about 50% of normal at the time of diagnosis. 6 Treatment starts with lifestyle measures, diet, weight loss and exercise, and, if those fail, oral drugs are added. In most patients T2DM is a progressive disease and, over time, many patients will need insulin. 7 The UKPDS showed that just over one-half of patients initially randomised to sulphonylureas (an oral drug that stimulates pancreatic insulin production) had to switch to insulin by 6 years. 8 In a population-based study in Tayside, Scotland, 6% of patients with T2DM started insulin each year. 9 Most people with T2DM starting insulin nowadays probably start with a once-daily injection of a long-acting analogue, but, over time, some will progress to multiple daily injections (MDI) in order to achieve good control.

Data from other studies have shown that many patients with T2DM are on insulin therapy. The Lothian Audit reported that 32% of people with T2DM are on insulin [J. McKnight, presentation to Royal College of Physicians of Edinburgh (RCPE) conference, 2005, formerly on RCPE website].

Control, glycated haemoglobin and insulin treatment

The term ‘control’ is a recurring one in diabetes. It refers principally to preventing blood glucose from becoming too high, but also applies to preventing it from becoming too low. High blood glucose is known as hyperglycaemia, whereas low blood glucose is called hypoglycaemia.

In the non-diabetic person, blood glucose is kept within a narrow normal range (about 4–5.6 mmol/l) through the action of insulin and other hormones. The pancreas releases a little insulin throughout the 24 hours (known as basal insulin – about 0.5–1 unit per hour in adults), but production of insulin is swiftly and markedly increased with meals, going up 5- to 10-fold in the first 30 minutes. If blood glucose falls too low, counter-regulatory hormones are released and nervous system mechanisms are activated to increase it again. A key aspect is that the brain is dependent on glucose for energy. If blood glucose falls too low, brain function is impaired, as will be described later.

Control of blood glucose is measured in three ways. First, blood glucose can be checked at any time by finger-pricking to produce a drop of blood, and testing it with a testing strip and blood glucose meter. This gives the glucose level at that time, but it may change quite rapidly after meals or either insulin or tablets. Second, longer-term control is measured by glycated haemoglobin (HbA_1c) level, which reflects the average blood glucose level over 2–3 months. HbA_1c has been a major advance in diabetes because by testing every 3 months, it gives an indication of how good control is. However, it provides an average and that can reflect very tight control with little fluctuation in glucose levels, or poorer control with considerable fluctuation. At the risk of considerable simplification, this can be illustrated by the averages of 4 and 8, and 2 and 10 – both equal 6. Third, new devices can now provide frequent automated testing of interstitial tissue glucose, calibrated to reflect plasma glucose, and known as ‘continuous blood glucose monitoring’. These devices are currently used more in research, but are coming into routine clinical practice in some clinics.

In T1DM, control is dependent on injected insulin, and, unfortunately, at present there is no insulin that can exactly mimic production by the normal pancreas. Even the latest rapid-acting insulins can neither achieve as rapid a rise after meals as the pancreas can, nor as rapid a fall. A key point is that natural pancreatic insulin release is regulated by the level of glucose in the blood in a way that injected insulin cannot be. A fall in blood glucose will switch off pancreatic insulin release but cannot affect injected insulin.

There are various forms of insulin, and various combinations, grouped by duration of action.

Short-acting (SA) insulin comes in three types:

• The oldest type is called ‘regular’ or ‘soluble’; we will refer to it as short-acting (SA) soluble because some long-acting analogues are also soluble.

• The next type is SA analogue insulin, with three varieties on the market – aspart, lispro and glulisine. SA soluble starts acting within an hour of injection, peaks at 2–4 hours, and has some effect for up to 8 hours. The SA analogues act a bit more quickly and do not last quite as long. They are therefore regarded as being closer in effect to pancreatic insulin than SA soluble insulin. However, a Cochrane review10 concluded that the advantages of SA analogues over SA soluble were minor – very little difference (0.1%) in HbA_1c or total hypoglycaemic episodes, a greater (50%) but not statistically significant reduction in severe hypoglycaemic (SH) episodes (‘hypos’) in adults, but not adolescents, when used in MDI, but a greater difference (0.2%) in HbA_1c in patients on continuous subcutaneous insulin infusion (CSII). The improvement in HbA_1c with SA analogues rather than SA soluble in CSII, was reported to be 0.26% in a meta-analysis based on the last assessment report for the National Institute for Health and Clinical Excellence (NICE),11 and patient preference was also higher for analogues.

• The third type is inhaled insulin, appraised by NICE in 2006 [Technology Appraisal (TA) 113] but now withdrawn from the market by the manufacturers. 12

Intermediate-acting insulins, such as neutral protamine Hagedorn (NPH) or isophane, start working in 1–2 hours, peak at about 6–10 hours, and have some effect for 16–18 hours. Unfortunately, the peaks may vary unpredictably from injection to injection, and hence from day to day.

Short- and intermediate-acting insulins can be mixed in the same syringe, and can be given as a premixed version twice daily. This is known as ‘conventional’ insulin therapy. The newer long-acting analogue insulins – glargine and detemir – are longer acting than NPH and have a long steady action, being sometimes called ‘peak-less’.

In recent years, since the Diabetes Control and Complications Trial (DCCT)13 showed that good control reduced the adverse effects of T1DM, there has been a move to more intensive insulin treatment. This consists of a combination of basal insulin using NPH (usually twice per day) or a long-acting analogue, with SA insulin at mealtimes, usually called ‘bolus’ insulin – hence the term ‘basal–bolus’ regimens. These can be given in two ways: by MDI or by CSII via insulin pumps.

History of continuous subcutaneous insulin infusion

The first studies of CSII delivered via insulin pumps came from Guy’s Hospital, London, UK in 197814 and Yale, New Haven, CT, USA in 1979. 15 CSII uses a small programmable pump with a fine tube connected to a soft plastic cannula (introduced by needle), which goes into the subcutaneous tissue under the skin, often in the abdomen. The cannula is changed every 2–4 days. The aim of CSII is to try to approximate the insulin delivery profile more closely to the pattern of output behaviour of the normal pancreas, by providing continuously infused, low-volume basal insulin for fasting periods and the delivery of increased rate boluses to cover meals. Only SA (soluble or analogue) insulin is used.

Lenhard and Reeves (2001)16 reviewed the literature in 2001 using MEDLINE only. They noted the rise in popularity of CSII after the introduction of pumps in the late 1970s and early 1980s, followed by a fall because of size, safety and efficacy concerns, followed then by a rise in usage after the publication of the DCCT study. They also noted that the newer pumps were smaller, more reliable and easier to use. They estimated that about 8% of all adults with T1DM in North America were using pumps. They concluded that there was good evidence for benefits in adults (‘comparable or slightly superior to MDI’) and some in pregnancy, but that there was little good-quality evidence in children.

Pickup and Keen,17 who were the originators of CSII, reviewed the history of, and evidence base for, CSII in 2002. They noted the considerable worldwide use of pumps (over 200,000 patients) and the disproportionately low UK use. They concluded that on CSII, blood glucose and HbA_1c levels are similar or slightly lower than when using MDI, that hypoglycaemia is much less frequent, and that ketoacidosis occurs at the same rate. They concluded that the proportion of patients who would be suitable is relatively small. In a complementary paper, Pickup et al. (2002)18 carried out a meta-analysis of randomised controlled trials (RCTs) comparing CSII with MDI. They found that HbA_1c level was about 0.5% better on CSII, but found that few studies reported hypoglycaemic episodes; none appeared to report effect on quality of life. The CSII group needed 14% less insulin.

The previous UK HTA report has been mentioned already and its summary is shown in Appendix 1. The Agence d’Évaluation des Technologies et des Modes d’Intervention en Santé (AETMIS),19 Quebec, Canada, published a report in June 2005, comparing MDI with CSII. It concluded that CSII might be indicated for a limited, selected group of people with T1DM, and cited various selection criteria, including:

• inadequate glycaemic control despite a trial of intensive insulin therapy

• recurrent, unpredictable SH episodes, nocturnal hypoglycaemia or hypoglycaemic unawareness, causing incapacitating anxiety and affecting the quality of life

• morning hyperglycaemic episodes (morning blood glucose level of 8 mmol/l or more)

• and for children, the above plus extreme insulin sensitivity (under 20 units of insulin per day).

At the time the report was written, glargine was not available in Quebec, Canada.

It is always interesting to know what treatments clinicians with diabetes choose for themselves. A survey of the American Association of Diabetes Educators (AADE) and the American Diabetes Association (ADA) asked members if they had diabetes, and, if so, how they were treated. 20 About 6.4% of members had diabetes, of whom 72% had T1DM. The survey found that 96% of those with T1DM used an intensive insulin regimen, and that over one-half (60% of the AADE members with diabetes and 52% of the ADA ones) used an insulin pump.

Search strategy

Sensitive searches of electronic databases were performed in order to retrieve a wide range of different types of evidence and study designs. All bibliographic records retrieved were then manually screened for studies of interest. These included systematic reviews, RCTs, non-randomised trials, observational studies, and studies on economics, costs, quality of life and patient satisfaction.

The following sources were used to identify both published studies and meeting abstracts:

• MEDLINE, 2002–June 2007; EMBASE, 2002–June 2007; Science Citation Index, 2002–June 2007 (limited to meeting abstracts only); Cochrane Library 2007 Issue 1; contact with experts; reference lists; industry submissions; website of ADA for recent meeting abstracts from the 67th Scientific Session June 22–26 2007 Chicago, IL, USA. Searches were limited to English language only.

• Ongoing and recently completed studies were searched for using National Research Register 2007 Issue 2 and Current Controlled Trials June 2007.

• Details of the search strategies used and a flow chart of studies identified for the clinical effectiveness sections are given in Appendix 3.

Identification of studies

Abstracts returned by the search strategy were examined independently by two researchers, and screened for inclusion and exclusion. Full texts of studies considered to be possible inclusions were obtained. Each was examined by two researchers independently.

Data extraction strategy

For each study, two reviewers extracted data regarding study design and characteristics, details of the intervention and patient characteristics and outcomes into a specially designed form. Differences in data extraction were resolved by discussion, referring back to the original papers. Interobserver differences were few. No formal calculation of interobserver agreement was carried out.

Quality assessment strategy

To assess the quality of the RCTs, the following criteria were used: (1) method and description of randomisation; (2) description of attrition/losses to follow-up; (3) specification of eligibility criteria; (4) power calculation; (5) robustness of outcome measurements; (6) similarity of group participants at baseline; and (7) data analysis. Blinding was not used as a quality criterion in this report, as it is not possible to blind patients to the wearing of an insulin pump.

Overall study quality was rated as follows: A (all quality criteria met); B (one or more of the quality criteria only partially met); or C (one or more criteria not met).

Analysis

We would have considered combining results from the trial by meta-analyses had they been sufficiently similar, but this was not considered appropriate.

Type 1 diabetes

Doyle et al. (2004)110 recruited children and adolescents with T1DM, age range 8–21 years. None had been on glargine or CSII before, and most were on conventional twice-daily insulins. Baseline HbA_1c levels ranged from 6.5% to 11%.

Thomas et al. (2007)111 a pilot study, was a three-arm trial in adults with altered hypoglycaemia awareness and debilitating severe hypoglycaemia. One arm was analogue MDI, another was CSII, and the third (not further mentioned in this report) was of education and relaxation of glycaemic targets. None had been on analogues before, 15 (71%) were using human insulin MDI; and five (29%) were using twice-daily biphasic insulin mixtures. Baseline HbA_1c level was 8.6%.

Maran et al. (2005)112 was a small trial in 10 adults with T1DM who had been on CSII therapy for at least 6 months. Details are sparse but the aim was presumably to find out whether the advent of glargine-based MDI means that patients on CSII could return to MDI. Mean baseline HbA_1c level was 7.7%.

Bolli et al. (2004)113 recruited patients (ages not given) with T1DM naive to CSII and glargine in Italy, the UK and France. Mean baseline HbA_1c level was 7.7%. Details of treatment at recruitment were not given.

Type 2 diabetes

Herman et al. (2005)115 recruited people over 60 years in the USA. Mean baseline HbA_1c level was 8.25%. They were on at least one injection of insulin per day, with or without oral agents.

Wainstein et al. (2005)117 recruited obese people [body mass index (BMI) 30–45 kg/m²] with T2DM age range 30–70 years, who had not been well controlled on two or more injections per day plus metformin. All had HbA_1c level over 8.5%. Insulin dosage before the trial was over 1 unit/kg per day.

Raskin et al. (2003)116 recruited adults over 35 years (mean age 56) with T2DM, on at least one injection of insulin per day, with or without an oral agent. Mean baseline HbA_1c level was 8.1%.

Berthe et al. (2007)114 recruited people aged 40–65 years with a BMI of between 26 and 42 kg/m² and an HbA_1c level of ≥ 6.5% on two determinations to a randomised crossover trial. The mean HbA_1c level was 9.0%.

Raskin et al. (2003)116 was funded by Novo Nordisk Pharmaceutical Industries (who do not manufacture pumps). Thomas et al. (2007)111 was supported by Sanofi-Aventis and Medtronic. Herman et al. (2005)115 was funded by the ADA. Doyle et al. (2004)110 was funded by NIH (National Institutes of Health) and Juvenile Diabetes Research Foundation (JDRF) with additional support from Medtronic. Berthe et al. (2007)114 was supported by Eli Lilly France. No details were given of funding for the Bolli et al. (2004)113, Maran et al. (2005)112 or Wainstein et al. (2005)117 trials. Table 4 gives further details of these trials.

The issue about differences in educational input, usually not reported in these trials, is because the amount of education is potentially a confounding factor – if the CSII group gets more education, any difference observed may be due to that rather than the CSII. The Berthe trial design gives some concern. 114

Internal validity

External validity

Most of the trials were carried out in countries other than the UK.

See Appendix 4 for full details of study quality assessment of the trials.

Mean glycated haemoglobin

Mean blood glucose levels

Mean daily blood glucose level was measured in all four RCTs in T1DM. Bolli et al. (2004)113 reported no significant difference between groups in daily blood glucose levels. Doyle et al. (2004)110 reported that fasting levels were the same on CSII and MDI but that mealtime levels were lower on CSII. Maran et al. (2005)112 reported lower mean glucose levels in CSII (147 mg/dl) than with MDI (189 mg/dl) (p < 0.03).

Thomas et al. (2007)111 reported median glucose levels and the results of CGMS for MDI and CSII groups at 6 months. Daytime median glucoses were similar at 7.6 and 7.8 mmol/l. Night-time median glucose was higher, with CSII at 8.4 mmol/l, but mainly because the MDI group spent, on average, 15% of time with hypoglycaemia (under 2.5 mmol/l), whereas the CSII group had very little hypoglycaemia, with 0.6% of glucose readings under 2.5 mmol/l. Neither group showed a significant change over the 24 weeks.

Glucose variability

Two studies [Bolli et al. (2004)113 and Maran et al. (2005)]112 reported measures of glucose variability. Bolli et al. (2004)113 reported no significant difference between groups in mean amplitude of glycaemic excursions (MAGE) at baseline and 6 months (CSII baseline 8 ± 2.4 to 6 months 6.4 ± 2.2 versus MDI baseline 7.6 ± 1.7 to 6 months 6.4 ± 2.1). Eight-point blood glucose profiles (coefficient of variation) measurements also revealed no significant differences between treatments at baseline and end point.

Maran et al. (2005)112 reported glucose variability as area under the curve (AUC) > 10 mmol/l, time spent at glucose level > 3.6 mmol/l and < 10 mmol/l, and time spent during night-time hours in glucose range > 3.6 mmol/l and < 10 mmol/l. There was significantly less hyperglycaemia during CSII treatment than with MDI (AUC > 10mmol/l CSII end of study 9603 ± 3941 minutes versus MDI 26445 ± 9390 minutes; p < 0.02 between groups). By the end of the study, significantly more time was spent in the glucose range > 3.6 mmol/l and < 10 mmol/l in the CSII group compared with MDI (CSII 1582 ± 212 minutes versus MDI 769 ± 158 minutes, p < 0.02 between groups). Similar results were reported for night-time glucose (CSII 298 ± 63 minutes versus MDI 194 ± 51 minutes; p < 0.02). These results do not quite fit with the lack of difference in HbA_1c, unless the MDI group had much wider variation around the mean.

Berthe et al. (2007)114 used continuous blood glucose monitoring devices, and produced glucose profiles against a target of keeping glucose in the range 3.3–10 mmol/l. The target was achieved for 44% of the time on conventional insulin therapy, for 54% on MDI and for 77% on CSII (p = 0.0095 for CSII versus MDI).

Quality of life

Two studies [Doyle et al. (2004)110 and Thomas et al. (2007)]111 in people with T1DM reported quality of life outcomes. Neither study showed a significant difference in quality of life between CSII and MDI [measured using the Diabetes Quality of Life (DQoL) questionnaire]. In people with T2DM, Herman et al. (2005)115 reported no significant difference in quality of life as measured by Short Form-36 (SF-36) and the Diabetes Quality of Life Clinical Trial Questionnaire (DQoLCTQ). However, Raskin et al. (2003)116 reported a significant improvement in treatment satisfaction in CSII group compared with MDI (p < 0.001).

See Appendix 4, Table 44, for more details of quality of life and patient satisfaction outcomes in the trials.

Hypoglycaemia

All eight RCTs reported the occurrence of hypoglycaemia. 110–116 Only two RCTs in people with T1DM [Doyle et al. (2004)110 and Maran et al. (2005)]112 conducted a statistical analysis of the occurrence of hypoglycaemic episodes. Doyle et al. (2004)110 reported that significantly fewer subjects with T1DM on CSII had severe hypoglycaemia episodes by end of study compared with MDI (CSII two episodes versus MDI five episodes in four patients; p < 0.05 between groups). In contrast, Maran et al. (2005)112 (the smallest trial) reported no significant difference in hypoglycaemic reaction exposure between groups.

Bolli et al. (2004)113 commented that SH episodes were too infrequent to allow meaningful comparison. The frequency of confirmed hypoglycaemic events/patients where blood glucose fell below 4 mmol/l was similar in both groups at 6 months (CSII – 41 events in 28 people versus MDI – 35 events in 29 people). There were only two SH events so no comparison of that was possible.

Thomas et al. (2007),111 with 21 patients followed for 24 weeks, reported that non-significant trends towards reduced incidence of severe and mild symptomatic hypoglycaemia were seen in the MDI and CSII groups in comparison with the third arm, the Education Group, but no difference between MDI and CSII. This may have been due to the very small numbers involved.

None of the four RCTs in people with T2DM reported a significant difference in hypoglycaemic episodes.

See Table 45 in Appendix 4 for more details of adverse events in each of the trials.

Insulin dose

Doyle et al. (2004)110 reported an insignificant difference – 0.6 units/kg per day on CSII and 0.7 units/kg per day on MDI. Thomas et al. (2007)111 reported the daily insulin dose at zero and 24 weeks. There was a non-significant increase in the MDI group and a significant (p = 0.01) decrease in the CSII group. Both groups started on 0.7 units/kg per day. The MDI group ended at 24 weeks on 0.8 units/kg per day and the CSII on 0.4 units/kg per day. They did not test the statistical significance between MDI and CSII. The other two T1DM studies, both abstracts only, gave no results. In T2DM, no differences were found. Wainstein et al. (2005)117 noted a drop in the CSII group in the first period but it did not persist.

Weight

The only T1DM study that reported changes in weight was Thomas et al. (2007)111 where both CSII and MDI showed a non-significant change. All the T2DM studies reported that there were no significant differences in weight changes between CSII and MDI.

Summary

Summary

In terms of HbA_1c levels, three of the new studies show no difference between MDI and CSII;121,123,126 four show differences (0.52%, 0.34%, 0.25% and 0.26%) that are not significant,118,120,124,126 and one shows a larger and statistically significant difference of 0.84%. The lack of statistical significance may sometimes be due to small numbers – the Cohen et al. (2003)118 study reported what would be seen as a clinically useful difference (0.52%), but had only 16 patients.

Caveats

For assessing efficacy, RCTs are the gold standard. Observational studies usually provide poorer quality evidence than RCTs because there is a much greater risk of bias. For example, good results may be obtained because the recruits adhere better to therapy than most patients. Publication bias may be a problem, in that negative observational studies may be less likely to be published than positive ones. For that reason, observational studies are usually excluded from technology assessment reports (TARs) and Cochrane reviews. However, in the last TAR on CSII, we noted that the reduction in hypoglycaemia was greater in some observational studies than in the RCTs. 50 We speculated that this might be because trials were unselective in their recruitment, whereas observational studies might selectively recruit people having particular problems. If so, it is possible that the observational studies will be a better guide to results in routine care. They may also be of longer duration and hence provide useful data on discontinuation rates and side effects. It is also possible that patients will not become fully expert in pump use in short duration trials, in which case long-term follow-up might show better results. Some of the studies may give useful data on the training requirements for people starting CSII.

In this section, we give results from a group of observational studies. Some were comparisons of MDI and CSII, in matched groups or unmatched groups. Because of possible biases, and because we have evidence from RCTs, we have not used the comparative data, but have used only the CSII arms as case series. Nor have we attempted to be comprehensive.

Study characteristics

Details of 48 observational studies are given in Tables 7–9, for the different age groups. Twenty studies included adults (either adults only or mixed ages), 23 studies included children/adolescents and five studies included young children (aged ≤ 7 years).

Studies were conducted in a variety of countries most commonly in the USA (n = 20) and Europe. Three studies were conducted in the UK. 138–140 Other studies were set in Australia, New Zealand, Canada and Israel.

The observational studies incorporated a variety of study designs: surveys, audits, before/after studies with and without control group (matched and unmatched), prospective and retrospective data, primary data, patient records and national registers.

Sample size ranged from 8 to 2702. The majority of studies had a sample size of under 50. In the adult/mixed age groups, the duration of follow-up ranged from 11.5 months to 13 years, the majority being 1–2 years. In the children/adolescent groups, follow-up ranged from 6 months to 5 years, with the majority of studies being 1–2 years. Among younger children, four studies had a follow-up of 1 year and one study of a mean 30 months.

Reasons for starting CSII

Reasons for starting CSII were not reported in 13 studies. All but a few studies included poor metabolic control (including frequent hypoglycaemic episodes and the dawn phenomenon) as a reason for starting CSII.

Among adults and mixed age groups, planning for pregnancy or during pregnancy140–143 and the desire for a more flexible lifestyle140,141,144–146 were commonly cited reasons. Patient preference was included as a reason in four studies. 141,147 Other less commonly cited reasons included: quality of life,141 low insulin requirements, hypoglycaemia unawareness,144 diabetic complications,148 participation in study,142 allergy to insulin,143 gastroparesis140 and lipodystrophy. 140

Among children and adolescents, commonly cited reasons were: request of patient or parents149–156 and the desire for a more flexible lifestyle. 149,157–159 Less common reasons reported included hypoglycaemia unawareness,157 quality of life,157 early onset of diabetic complications,160 too much work with multiple injections150 and problems with injections. 150,158,160

Selection of patients for CSII

Many of the studies gave details of how patients were selected for CSII. Among adult and mixed age studies, patients were selected if already on MDI,138,139,146,161,162 and showed willingness and ability to master intensive management features of CSII. 101,163 In Pickup et al. (2005),138 patients who showed poor compliance or psychological problems were considered unsuitable for pump therapy.

Among studies including only adolescents and children, patients and families needed to demonstrate the desire and ability for intensive management. 149,152,154,157,158,160,164 Two studies reported that patients were already on MDI [Sulli and Shashaj (2003)160 and Berhe et al. (2006)]165 and two studies reported requiring prior documentation of adequate blood glucose testing. 159,166 Other inclusion criteria included good parental supervision, [Garcia-Garcia et al. (2007)167 and Litton et al. (2002)],164 minimum duration of diabetes167,168 and daily insulin requirement of more than 0.75 units/kg. 167 One study reported that patients were excluded if in the honeymoon phase. 165

Education and support for CSII

Seventeen studies did not report details of education and support. A few studies reported identical educational and support programmes for both CSII and MDI groups. [Cersosimo et al. 2002,169 Pickup et al. 138 and Garcia-Garcia et al. )]. 167,168

Several studies described intensive training/education of participants and their families. 101,140,144,146,147,152,154–162,164–166,170–174 Some training programmes included the use of a dummy saline pump. 154,157,158,171 Initiation of pump therapy involved a period of admission to clinic or hospital in a few studies. 152,158,171

Intensive ongoing support was often provided, including initial frequent visits and telephone contact, 152,156,158,161,165,166,172 and 24-hour nurse on call or telephone support. 140,164,166,174

Continuation rates

Continuation rates can be regarded as evidence of patient satisfaction. Fewer than half of the studies (22/48) reported continuation rates. Continuation rates at 1–5 years ranged from 74% to 100%. Continuation rates of 100% were reported in two studies: in adults/mixed age groups144,163 and five studies in adolescents and children. 154,156,159,164,165

A variety of reasons for discontinuing CSII were reported in 12 studies (Tables 10–12). Reasons for discontinuing included: end of pregnancy, lack of tolerance to carry the pump, perception of goals not reached, infection at insulin injection site and hypoglycaemic episodes;141 not able to cope with the technical aspects of using an insulin pump and not convinced of the advantages;145 cost and inconvenience;146 site problems, sweating, costs and other illness;140 patient’s decision (most commonly due to reluctance to wear the system), cutaneous problem and poor compliance;176 dislike of or difficulty with needle insertion, insurance difficulties, trouble keeping the infusion site clean, tape not adhering, and general dislike of the pump;101 wish to return to injections, and worsening control due to omitting bolus insulin;166 extra work involved with changing infusion sets and dislike of something being attached to body;150 psychiatric and dermatological conditions;152 inconvenience in carrying the pump;171 pump limited normal physical activity, recurrent DKA;153 and, DKA due to insulin omission, diabetes burnout, minor problems with infusion site, body image concerns and concerns about weight gain. 155

Glycaemic control as reflected in HbA_1c

Forty-six studies reported comparable before/after data on HbA_1c levels (Table 13). Levels of statistical significance are reported where they were available (some papers did not report whether the change was significant or not).

All of the 18 studies in the adults/mixed age groups showed a significant decrease in HbA_1c levels (ranging from 0.2% to 1.4%) after participants started on pumps.

Few studies reported proportions reaching targets. Targets varied. Pickup et al. (2006)139 reported that 37% of CSII subjects and 13% of MDI ones achieved HbA_1c < 7%; for a target of < 8%, the proportions were 73% and 30%. Radermecker et al. (2005)143 noted that of 95 patients on CSII, only five reached HbA_1c of 7% or less; most (66) were in the range 7.1% to 8.5%. In Reda et al. (2007),146 only 9 out of 105 reached the ADA target (7.0% or less) before CSII, and only 18 afterwards.

There were 23 studies in the children/adolescents age group. Three of the studies showed an increase after using pumps. In Kordonouri et al. (2006)151 (n = 59) and Garcia-Garcia et al. (2007)167 (n = 8) the increases of 0.01% and 0.08%, respectively, were neither clinically or statistically significant. The statistical significance level was not reported in Raile et al. (2002)159 (n = 12), where the increase was 0.6%. The remaining 20 studies showed an overall decrease, ranging from 0.2% to 1.2%. In 13 of these studies, the overall decrease was statistically significant. However, in Mack-Fogg et al. (2005),173 the decrease was only significant in the 10- to 12-year age groups, and not significant in the younger children. Willi et al. (2003)154 also reported the subgroups by age, and found a significance decrease in the 5- to 9-year and 13- to 17-year group, but no change in the 10- to 12-year group.

In four studies the decrease was not significant 150,153,157,183 and three studies171,179,182 did not report the significance level of the decrease. Only one study reported proportions reaching targets. Simmons et al. (2006)179 reported that 75% of 6- to 12-year-olds reached the ADA target of 8% or less (for that age group). Only 15% of adolescents (13–19) reached their age range target of < 7.5%.

There were five studies in young children and all showed decreases, which ranged from 0.2% to 1.6%. 156,164,165,174,184 Four156,164,165,174 showed a statistically significant decrease and one [Alemzadeh et al. (2006),184 n = 14] showed a non-significant decrease. Only one study reported on targets met – Berhe et al. 165 reported that 76% of patients had HbA_1c levels of < 8.5% after CSII compared to 35% before.

In summary, only three out of 46 studies showed an increase in HbA_1c. These studies were all in children/adolescents, and the increases were insignificant in two, and 0.6% in the third (no significance level reported). Most studies showed decreases in HbA_1c varying from 0.2% to 1.6%.

Hypoglycaemic episodes

The main interest was in SH episodes. Data are included from the 26 studies reporting comparable data on the rate of SH episodes before and after CSII was initiated (Table 14). Rates were reported in different units, so rate ratios were calculated to enable comparison between studies.

Ten studies were in adults/mixed age groups. 101,138–140,145–147,162,163,170 Two138,139 did not report any hypoglycaemia before or after pumps use. Of the eight remaining studies, all reported a significant decrease after going on pumps. The rate ratios ranged from 0.07 to 0.4. One of these147 reported no hypoglycaemic episodes in the prepubertal group either before or after pump use.

There were 11 studies in children/adolescents. 149,151,152,155,158,160,166–168,171,173 One study167 had no SH before or after going on pumps. Of the remaining 10 studies, the overall rate ratios varied from 0.12 to 0.80.

In four studies149,155,160,166 the overall decrease was reported as statistically significant. However in Ahern et al. (2002)166 (n = 161) the decreases were not significant when broken down into three age groups (possibly due to smaller sample sizes).

Three studies151,152,171 did not report the significance level of the decrease, but showed substantial reductions (rate ratios of 0.12, 0.30 and 0.35 respectively). In the remaining three studies in children/adolescents the overall decrease was not significant. However in one of the studies173 the reduction was significant in the 10–12 year age group, but not in the 2–4 and 5–9 age groups.

There were five studies in young children,156,164,165,174,184 and the rate ratios ranged from 0 to 0.81. In three studies 156,164,165 this was significant, and the in the other two it was not significant.

In summary, of the 26 studies examined, 15 showed a statistically significant decrease in SH episodes after going on pumps, five showed a non-significant decrease, and three showed a decrease, but the significance level was not reported. The remaining three studies did not report any SH episodes before or after going on pumps.

Diabetic ketoacidosis

As reported in the last assessment report, it is likely that one of the reasons for the low use of CSII in the UK was fear of DKA. People with T1DM on CSII have no insulin store in the body, and if the pump fails, they will rapidly develop metabolic problems.

There were 15 studies that had comparable before/after data on DKA rates. 140,145–147,151,152,156,158,164–167,170,171,173 These are summarised in Table 15.

Five studies were in adults/mixed age groups. 140,145–147,170 One study [Nimri et al. (2006)]147 showed a non-significant increase in prepubertal children and a non-significant decrease in the adolescent and young adult age groups. Two studies [Hunger-Dathe et al. (2003)170 and Rodrigues et al. (2005)]140 showed a statistically significant decrease. The patients in the study by Rodrigues et al. (2005)140 had a higher DKA rate than most other studies, but they included some having particular problems, including with DKA. Two studies [Linkeshova et al. (2002)145 and Reda et al. (2007)]146 did not report the significance level of the reduction.

There were seven studies in children/adolescents. McMahon et al. (2005)152 showed no change. Mack-Fogg et al. (2005)173 and Garcia-Garcia et al. (2007)167 showed a non-significant increase. Ahern et al. (2002)166 showed an increase (of one episode in 161 patients over 1-year) but the significance level was not reported.

Kordonouri et al. (2006)151 showed a significant decrease, and Juliusson et al. (2006)171 and Plotnick et al. (2003)158 showed a non-significant decrease.

There were three studies in young children. 156,164,165 Weinzeimer et al. (2004)156 (n = 65) showed an increase (from 0 to 0.04 per annum) but the significance level was not reported. Berhe et al. (2006)165 (n = 33) and Litton et al. (2002)164 (n = 9) showed no change, but as these studies were small they may have been underpowered to detect a statistically significant difference.

In summary, none of the 15 studies reported a statistically significant increase in DKA rates after patients going on to pumps. Three reported a significant reduction and three studies reported an increase; in one this was not significant and the other two did not report significance levels.

However, a report by Hanas et al. (2005)185 gives worrying data from Sweden, where pump use is common in children. In 1999, 7.5% of children and adolescents used pumps, and this figure rose to 11.2% in 2000. The DKA rate in CSII users was double the overall rate – 3.5 per 100 patient-years versus 1.7, but the true risk ratio will be higher, as the CSII DKA cases will presumably be included in the total for all patients. Hanas et al. note that most DKA occurred soon after CSII initiation, and that, along with the marked rise in CSII use, might perhaps suggest problems with adequate training. Full details will no doubt be published in due course.

Weight change

There were 30 studies in total reporting comparable before/after data on BMI or weight change (Table 16). 138–140,144,145,147,149–156,159,160,162–169,171–173,175,181,182 However, some of the studies that involved adolescents and young children did not take account of changes in the child’s development when considering weight change. 149,159,160,165,172,181,182

Seventeen studies reported a non-significant change in weight. Nimri et al. (2006)147 reported a non-significant change overall, but a significant decrease in the subgroup of young adults (but not in younger age groups).

Linkeshova et al. (2002)145 reported mixed results; i.e. unchanged in 53% of patients, an increase in 22% and decreased in 25%. Juliusson et al. (2006)171 reported a non-significant increase in boys, but a significant increase in girls.

Six studies reported a significant overall increase in BMI. 144,149,172,173,175,182 However, in Mack-Fogg et al. (2005)173 the increase was not significant in the age groups of 2–4 years and 10–12 years (but was highly significant in 5- to 9-year-olds). Also it should be noted that Alemzadeh et al. (2004)149 (mean age 14.7 years, n = 40), Liberatore et al. (2004)172 (mean age 12.9 years, n = 73) and Ugrasbul et al. (2006)182 (age range 4–21 years, n = 131) reported BMI change – not BMI z-scores – so did not take account of the child’s development.

The study by Weinzeimer et al. (2004)156 in young children showed a significant decrease in BMI z-scores. Significance levels were not reported in the four remaining studies. Pickup et al. (2006)139 and Garcia-Garcia et al. (2007)167 showed a decrease, Schiaffini et al. (2005)168 showed an increase and Ahern et al. (2002)166 showed mixed results in different subgroups, i.e. no change in preschool children, an increase in school-aged children, and a decrease in adolescents.

In summary, 17 of the 30 studies showed no significant weight change. Six showed a significant increase (but with the caveat that for three of these studies in children/adolescents the change did not measure z-scores on BMI or weight, hence did not take account of the child’s development), and one showed a significant decrease. The remaining studies either did not report significance or showed mixed results.

Insulin dose

There were 21 studies that reported comparable before/after data on insulin dose. 138–140,144,149,151,154,155,157,159–162,164–166,168,169,172,175,181 These are summarised in Table 17.

Of the eight studies in adults,138–140,144,161,162,169,175 five showed a significant decrease138–140,144,162 and one161 showed a decrease, but the significance level was not reported. One study175 showed a significant increase and the other169 showed a non-significant increase.

There were 11 studies in children/adolescents. 149,151,154,155,157,159,160,166,168,172,181 Four155,168,172,181 showed a significant decrease, three showed a decrease but the significance level was not reported149,159,160 and two151,154 showed a non-significant decrease. Ahern et al. (2002)166 showed non-significant change in all subgroups (preschool age group increase and older age groups a decrease). Conrad et al. (2002)157 showed an almost negligible decrease in the prepubertal age group and a significant decrease in the pubertal age group.

There were two studies in young children. 164,165 One164 showed a non-significant increase and the other165 a non-significant decrease.

In summary, of the 21 studies examined, most showed a decrease in insulin dose when patient were on CSII. Five showed an increase in the insulin dose, but this increase was only significant in two studies [Garg et al. (2004),175 in adults; Conrad et al. (2002),157 in the pubertal subgroup]. The reduction in insulin dose will provide some savings to modestly offset the cost of the pump.

Quality of life

Nine studies evaluated aspects of quality of life associated with CSII use from the perspective of health-care professionals, parents or children. 140,141,145,152,153,174,178,180,184

Studies used varying methods to collect data including questionnaires, specified scales, scales developed for the study and interviews. Sample sizes were generally small; only one study evaluated more than 35 patients and this larger study assessed the views of health professionals and not patients/parents. 141

Use of CSII at night-time only

Kanc et al. (1998)72 carried out a small trial to see if good control during the night, with avoidance of hypoglycaemia, would restore hypoglycaemic awareness in patients with T1DM who had lost it. Fourteen patients took part in a crossover study. In one arm, they continued their mealtime SA insulin and bedtime NPH. In the CSII arm, they continued their SA injections but switched to CSII at bedtime. Those who had been experiencing the dawn phenomenon used two or more basal rates during the night.

No differences in HbA_1c were seen between the two arms, but hypoglycaemia was about a third less frequent (p = 0.03) and warning signs were improved. The authors believe that this was due at least partly to avoidance of nocturnal hypoglycaemia, although nocturnal testing was not frequent enough for them to be sure. Total daily insulin requirements were lower with CSII (48 units versus 56 units), with more being taken as SA at mealtimes.

Kaufman et al. (2000)187 carried out a similar study in children aged 7–10 years, with two arms in a crossover trial. In one arm, children continued with three injections of lispro and NPH. In the CSII arm the pump was used to provide basal insulin and the breakfast and dinner lispro. The duration of the trial was short, 4 weeks on each arm. During the CSII period, blood glucose control was better, as reflected in fructosamine levels and five daily measurements, including at 0300. Fear of hypoglycaemia was halved. The authors report that there was less hypoglycaemia but do not provide data. Insulin dosage was reduced from a mean of 0.9 units/kg per day to 0.7 units. Quality of life was reported to be better on CSII but no data are given.

Use of different basal rates

One of the differences between an analogue-based MDI and CSII is that, once glargine or detemir is injected, the basal rate is fixed for the day, and cannot be changed if, for example, unexpected exercise occurs. Nor can the user have different basal rates in, for example, morning and afternoon. However, pump users can programme different basal rates for different times of day, and for different days. We asked INPUT for data on how many different basal rates were used by members, and the results are shown in Table 18.

It would be interesting to look at HbA_1c and hypoglycaemic episode frequency by number of basal rates used, but that is beyond the scope of this review.

The number of basal rates used raises an important issue about expertise in CSII use. Some of the trials are quite short: 16–24 weeks. Nearly all will recruit novice users [Maran et al. (2005)112 being an exception]. How long does it take a pump user to get the full benefit from a pump? Are the trials too short for users to get full benefit? Those randomised to CSII will be testing their blood glucose, and aware of hypoglycaemic episode frequency, but they will get at most one or sometimes two HbA_1c results during the trial. So they will not have time to adjust their regimens, repeat the HbA_1c 3 months later, and adjust again. Nor, perhaps, would they have enough time to try out different basal rates for different combinations of diet and activity.

It would be interesting to have data on HbA_1c and hypoglycaemic episode frequency in pump users at 3-monthly intervals for several years.

Who benefits most from CSII?

Previous meta-analyses reported that CSII gave HbA_1c levels lower, on average by 0.5% than MDI – a clinically significant, but not dramatic, improvement. The trials in these meta-analyses used mainly SA soluble insulin, rather than SA analogues. 18,50 Switching to the latter gives another 0.2% improvement. 11. A later meta-analysis of only studies using analogue insulins as the SA form, which at the time had only three trials, noted that the benefit of CSII relative to MDI was greater in those with high baseline HbA_1c. 188

Pickup et al. (2005)138 explored ability to benefit further. Firstly, they studied the patients to whom NICE guidelines most applied – those who could not achieve good control without disabling hypoglycaemia. (They noted that previous trials often excluded patients who were having problems with hypoglycaemia.) In a before/after study in patients having problems with hypoglycaemia on MDI, they first tried a more intensive period of MDI for 5 months, and then, if control had not been achieved, started CSII. The improvement in HbA_1c was 1.4%.

In an extension of this study with a larger group of patients, Pickup et al. (2006)139 showed that the strongest predictors of improvement in HbA_1c were a high baseline level on MDI and variability of blood glucose. They noted that one of the main reasons for failing to achieve good control was hypoglycaemia, and that hypoglycaemia was associated with large swings in blood glucose, and hypothesised that subjects with wide variability in blood glucose levels would find it most difficult to achieve control on MDI because of high rates of hypoglycaemia.

More recently, Pickup and Sutton (2008)107 carried out a meta-analysis aimed specifically at identifying the impact of CSII in patients with, first, an incidence of severe hypoglycaemia that would make them fit the NICE guideline indication, and, second, an adequate duration of time on CSII (6 months or more). They did not restrict studies to RCTs; most were before/after studies.

The pooled hypoglycaemia rate on MDI was 62 events per 100 patient-years. There was a marked reduction of about 76% on CSII. HbA_1c was reduced by 0.62%. The before/after studies reported a much bigger reduction in HbA_1c (0.72%) than the RCTs (0.22%).

Data from the Insulin Pump Clinical Database

The following information is from unpublished data kindly supplied by R. Feltbower and the Database group, April 2007, but which is being submitted for publication.

A group of centres with considerable pumps experience, and hence with larger numbers of pump users than most clinics, have pooled their data in a project sponsored by Roche Diagnostics (Burgess Hill, UK) but run independently by the Paediatric Epidemiology Group, University of Leeds, UK. The usefulness of this data set is that it reflects, first, results in routine care outwith trials, and, second, it gives results from centres of pumps expertise.

The data currently available to us do not include details of what regimens patients were on before CSII, but we assume MDI. Current UK centres include Harrogate, Bournemouth, Leeds (paediatrics) and Middlesbrough.

The overall reduction in HbA_1c was 0.9%. Table 19 shows mean changes in HbA_1c by age group. These results are in some ways good, but also somewhat disappointing because the means still fell well short of targets.

Short-term studies

Hirsch et al. (2005)189 carried out a crossover study comparing analogue MDI (aspart and glargine) with CSII. It was excluded from our main analysis because of the short-duration – 5 weeks on each arm, and too short to use HbA_1c as a measure of glycaemic control. Fructosamine improved, and nocturnal hypoglycaemia was about 25% less frequent with CSII (p = 0.0024). Insulin dose was slightly lower on CSII.

One problem with short-term studies has been mentioned above. How long does it take users to achieve the maximum benefits of CSII, including the use of multiple basal rates when required? In their study in young children (where parents controlled the pump programming), Wilson et al. (2005)126 noted that pump users started with an average of 2.9 basal rates per day, but by the end of a year were using 4.8 different basal rates.

There are also several very short studies of CSII in insulin-resistant T2DM, used to treat insulin resistance, but we have excluded these.

Quality of life

Barnard et al. (2007)108 recently reviewed studies reporting quality of life aspects on CSII. This group has received support from Roche, one of the pump manufacturers, for pump-related research, and a conference abstract version of the review carries the Roche logo, but the review is high quality and properly critical, and was carried out for a PhD thesis (KD Barnard, University of Southampton, 2007, personal communication), not funded by Roche, and seems free of bias. Most of the 17 studies in their review are included in this review or the 2002 assessment report. 50 Barnard et al. (2007)108 pay particular attention to the quality of life instruments used, and note that some are not validated. They comment on problems with the design of the studies, such as the lack of control groups, and, in particular, the confounding role of structured education, which should be given to all before commencing CSII, but which may not be given to comparator groups if there are any. In before/after studies, it is difficult to say how much of the benefit is due to the education rather than to the CSII. They also note the small numbers in many of the studies.

They conclude that there is currently no consistent quality of life gains from CSII in the current evidence base, but accept that this may be more a problem with lack of evidence than evidence of no benefit: ‘if a minimum standard were assumed to be a randomised controlled trial, which controls for increased education and contact time, uses appropriate sensitive measures, and recruits large numbers of participants to each group, there are no current published studies which meet these criteria.’.

Barnard et al. (2007)108 recommend further research:‘… a large-scale multi-centre patient preference controlled trial is required to focus specifically on quality of life issues surrounding insulin pump therapy… It is important to be clear about what quality of life means, i.e. increased independence, greater freedom, greater flexibility, easier management of diabetes, better control, etc.’.

A key point in quality of life aspects of CSII is that some of the reported benefits are in health-related quality of life, but many are not. Some are in aspects such as social life (not having to worry abut what time meals arrive on social occasions), greater ease of travel, flexibility of lifestyle, and enhanced ability to enjoy physical activities. Once the basal insulin in MDI is in, it is there for the day and night. The pump infusion can be adjusted at any moment.

Having identified a shortage of evidence, Barnard and Skinner carried out two studies. 190 The first was a qualitative study based on interviews with 80 pump users. The study was funded by Roche Diagnostics; the patients were identified by Roche; and the interviews were carried our by Roche staff, trained by the investigators. The potential for bias seems considerable, but we think it provides reliable data, for two reasons. First, the disadvantages of CSII are dealt with as well as the benefits, and the authors report that 60% of the patients reported downsides to CSII. Second, many of the comments match those obtained from pump users in the previous assessment report for NICE, in particular those mentioning flexibility of lifestyle, the feeling of greater control over the disease, less dependence on other family members, and the reduction in hypoglycaemia.

The main disadvantages of CSII include visibility of the pump (although the latest versions may be smaller), problems with breakdowns (21% of respondents), and lack of appropriate advice from health-care professionals. Three of the 80 noted cost was a problem, presumably because they were paying that themselves: some PCTs are very restrictive in funding. But the positives outweighed the negatives, as one would expect in a group of confirmed users. The authors summarise the results thus:‘Participants overwhelmingly reported experiencing benefits and improvements in their quality of life associated with insulin pump use.’.

The second study was provided to us as part of the industry submission but has since been published. 191 It was funded by Roche Diagnostics, and two groups of patients were identified by Roche: the first from their register of pump users, the second from non-CSII users who were on the register of blood glucose monitor users (although some of the latter group were found to be pump users). The authors tried to have four non-pump users for every pump user, but response rates were very different – 85% in the pump user group versus 38% in the non-users, which must be a major source of bias.

This reflects a problem in research involving pump users. They tend to be enthusiastic and highly motivated individuals who are happy to take part in research, which may make it difficult to find a comparison group with the same characteristics.

One strength of this study was that the authors used several instruments to assess quality of life, including the World Health Organization Quality of Life Abbreviated questionnaire (WHOQOL-BREF),192 which has domains for physical health, psychological state, social relationships and ‘environment’, which includes financial matters. They added topic-specific instruments, such as questionnaires assessing aspects of insulin delivery, fear of hypoglycaemia and problem areas in diabetes. Another strength was that the authors anticipated a possible bias arising from the fact that many pump users had to be self-funding, and hence there would be social selection biases operating, and in multiple regression carried out analyses controlling for socioeconomic status.

The result suggest that pump users score better on some of the non-health related aspects of quality of life, as well as on having fewer worries about hypoglycaemia. Again, these match the comments received from a far smaller number of users for the last assessment report. 50

New systematic reviews

Weissberg-Benchell et al. 193 published a meta-analysis in 2003 but it included studies only up to 2001, and therefore does not add to the findings of the last assessment report. 50

Retnakaran et al. (2004)188 carried out a meta-analysis of trials, which used rapid-acting analogues in both MDI and CSII, and therefore included only three trials, two of which were included in the last assessment report. 50 The third is De Vries et al. (2002),119 included above. All three trials used NPH as basal insulin.

Siebenhofer et al. (2004)194 included studies comparing SA analogue and SA soluble, for both CSII and injection regimens. They included eight trials in CSII, most of which had been included in the last assessment report50 and in the journal paper by Colquitt et al. (2003). 11 Three trials were not included in the assessment report, two being definite exclusions because of short duration of follow-up. One trial in the assessment report analysis was not included by Siebenhofer.

A high-quality report from AETMIS drew heavily on the last assessment report but added trials published up to 2004. 19,50 These are included in this report.

Indications for CSII in different age groups: children and adolescents

The German working group for insulin pump treatment in paediatric patients195 has identified seven indications for CSII, in a cohort of 1567 children and adolescents:

• dawn phenomenon (27.4%)

• reduction of hypoglycaemia (20%)

• flexibility (22.4%)

• improvement of hyperglycaemia (18.1%)

• motivation (10.4%)

• failure of injection therapy (1.6%), by which they meant CSII was used as a ‘last resort’

• pregnancy (0.1%).

But the proportions varied widely amongst the four age subgroups. In the under-5s, the main reason (42%) was reduction of severe hypoglycaemia, followed by flexibility (22%). In the 5- to 9-year age group, hypoglycaemia was again the top indication (32%), followed closely by the dawn phenomenon (28%). In the 10–14s, dawn phenomenon was the most common reason (32%), followed by flexibility (22%) and hypoglycaemic episode (17%) and hyperglycaemia (18%). In the 15- to 20-year range, there was an even spread – flexibility 26%, dawn phenomenon 22%, hyperglycaemia 21%, motivation 18%.

One important finding of this study was that initial reductions in HbA_1c were not always sustained. Amongst those who started CSII because of hyperglycaemia, HbA_1c fell from an initial mean of 8.8% to 8.5% at 12 months, but then rose back up to 8.8% at 36 months. However, this represents success, as in children HbA_1c usually rises with age. 43 Those who started CSII because of hypoglycaemia had a lower starting HbA_1c of 7.6%, maintained that at 12 months (7.5%), after which it rose to 7.9% at 2 years and 8.1% at 3 years.

Summary of clinical effectiveness

Since the last review, the number of observational studies has increased considerably, and there have been more trials of CSII against NPH-based MDI. We now have some trials in people with T2DM.

Unfortunately, there is a relative scarcity of trials with analogue-based MDI, and some of those are very small.

The one study of CSII versus analogue MDI in adolescents/children shows a reduction of 1% in HbA_1c.

The recent RCTs of CSII versus NPH-based MDI do not add much to the previous review, which found a reduction in HbA_1c of 0.6%, a similar figure to that reported by Pickup et al. in their 2002 meta-analysis. 18

The observational studies have variable results but show larger drops in HbA_1c.

The Insulin Pump Clinical Database also shows a larger reduction, of 0.9% but variable amongst age groups.

Quality of life appears better on CSII, and most patients prefer it.

Most studies show a reduction in hypoglycaemia with CSII, and a reduction in insulin dose required.

Modelling inputs

The cost-effectiveness results presented within the industry submission were reviewed in tandem with additional data supplied through the web-based CORE model implementation. As is clear from the summary of CORE above, from the summary of the population characteristics within the clinical sources used for the CORE model as outlined in Appendix 5, and from communication from the CORE modelling team, it is doubtful whether the CORE model would be applicable to the paediatric or adolescent population of patients with T1DM. The submission was been prudent in this regard, and modelled a cohort of baseline age of 38 years and an average duration of diabetes of 10 years.

The background prevalences for most of the vascular complications arising from diabetes were taken from the DCCT 1994 paper regarding the effect of intensive on the development and progression of long-term complications. 209 Baseline values for aspects such as cholesterol levels and blood pressure were drawn from two references. 68,210 As these references did not provide all of the data necessary for the CORE model, the background prevalences for angina, background diabetic retinopathy, proliferative retinopathy, macular oedema, cataract, foot ulcer and amputation were apparently set to zero. To the degree that background prevalences were underestimates within the modelling, this may have tended to slightly overstate the benefit of the anticipated improvement in HbA_1c arising from adoption of CSII. But given that the baseline age of the cohort simulated it cannot be stated whether this would have necessarily been to the benefit of CSII.

The key clinical effectiveness inputs to the industry submission were drawn from the meta-analysis of Pickup and Sutton (2008),107 which analysed the effect of CSII relative to MDI within a population with T1DM with problems with severe hypoglycaemia episodes and a rate of severe episodes of more than 10 per 100 patient-years. Clinical effectiveness estimates were as shown in Table 20.

The trial based analysis related to the meta-analysis of Pickup and Sutton, which related the average baseline HbA_1c to the change that would be anticipated from the use of CSII. The average baseline HbA_1c within this was 8.11% and the reduction was only –0.62%. The baseline HbA_1c may seem unduly optimistic, which means that patients have less to gain in terms of complications avoided. However, those who stand to gain from CSII include patients with good or even normal HbA_1c, but having problems with severe hypoglycaemia. It is therefore worth including this group.

However, the submission goes on to point out that the UK-based population within the meta-analysis had considerably worse control and an average HbA_1c of 9.4%. This was mapped to given an average UK-relevant improvement from CSII of –1.29%. This might be thought to be inappropriate because the HbA_1c level in all patients might be higher than in those being considered for CSII, who should already be on MDI. However, we know from unpublished data from the Insulin Pump Database (R. Feltbower, University of Leeds, 2007, personal communication) that the mean level before starting CSII was 9.1%, so the figure used in the baseline analysis is not that far away from a representative cohort.

The conservative analysis of –0.95% represents the mid-point between –0.62% and –1.29%. Note that these changes were all applied to a baseline HbA_1c of 9.4%. As a consequence, MDI patients for the trial-based analysis and the conservative UK analysis will have tended to have too high a baseline HbA_1c, which may have tended to bias these analyses to a degree towards CSII.

The data underlying the trial-based analysis, as summarised above (and presented within Table 20 of the economic appendix to the manufacturer’s submission), were not submitted electronically. Based upon the additional data submitted electronically, uncertainty appears to have only entered the model with respect to the baseline HbA_1c level and the effect of CSII upon this.

The baseline level of HbA_1c as applied to the MDI cohort, had a mean of 9.4% but this appears to have been subject to an SD of ± 2.1%. As a consequence, the range of HbA_1c levels within the MDI cohort was somewhat greater than might have appeared to be the case within the submission.

Similarly, it appears that at least for the UK-relevant analysis and the conservative UK analysis, the impact of CSII upon HbA_1c also involved a large range having an SD of ± 2.98%. This uncertainty as to effectiveness does not appear to have been linked to patients’ baseline levels of HbA_1c, as would be implied by the logic applied within the submission to subsets within the Pickup and Sutton107 meta-analysis. This would tend to have increased the effect of CSII in patients with the worst control. Given this, some patients will have been simulated to have worse control under CSII than under MDI, but other patients using CSII will be simulated to have very tight control of HbA_1c indeed.

While the submission is not explicit upon this point, the simulation inputs as uploaded by the CORE team to the CORE website implementation appear to indicate that distributions were placed upon both the baseline HbA_1c and reduction in this associated with CSII. Unfortunately, the current implementation of CORE does not permit a link between baseline HbA_1c and the effect of CSII upon this to be specified in the probabilistic sense: i.e. to specify a positive covariance between these variables. Given this, it may have been more appropriate to have modelled a representative patient than to have placed uncorrelated distributions upon these two variables where a clear covariance structure appears to be implied by the meta-analysis of Pickup and Sutton. 107

The direct costs of CSII and MDI treatment were drawn from industry sources and the British National Formulary (BNF), as outlined in Appendix 6 (and annualised within Table 23 of the economic appendix to the industry submission). It is not immediately clear what dose or patient weight has been assumed, but a point to note is that the industry submission anticipated a 25% reduction in the need for insulin, resulting in a cost saving from CSII of £177 per patient per year, reportedly drawing the dose assumption from the previous HTA and the cost from BNF. But the other costs of CSII more than offset this, with the annualised cost of CSII being £2770 as against £1224: an additional net cost of around £1550 from the use of CSII.

The cost of an SH event was taken to be £413, this being stated as having been drawn from the NICE inhaled insulin TAR,211 which, in turn, cites the NHS reference costs as the source. However, it should be borne in mind that this is likely to relate to a very SH event, the average length of stay within hospital for this reference cost being slightly in excess of 2 days. As outlined within the NICE glargine HTA,75 only a minority of patients are likely to be admitted to hospital following an SH event.

Given the centrality of the effect upon SH events within the submission an average cost of £413 may have been too high, and the £62 of the NICE glargine HTA may have been more appropriate or at a minimum appropriate as a sensitivity analysis. For instance, it appears that given an annual rate of 0.620 SH events under MDI compared with 0.148 under CSII, the annual cost of treating these would be around £280 for MDI compared with around £60 for CSII. This represents an annual saving of £220 from the use of CSII as against MDI. The parallel figures using the lower cost of £62 per SH event would appear to be £42 for MDI, £9 for CSII and a net annual saving of £33 per patient. Given the assumed 75% reduction in SH event from CSII and its centrality to the analysis, the assumed cost of £413 rather than £62 effectively reduces the additional annual cost of treatment with CSII by a little over 10%.

Costs of complications were mainly drawn from the Clarke et al. (2003) UKPDS65 paper,212 while utility values for were mainly drawn from the Clarke et al. (2002) UKPDS62 paper. 213 Note that UKPDS62 relates to patients with T2DM. There is no obvious reason to anticipate that the utility decrements arising from the complications associated with diabetes would be particularly different between patients with T1DM and T2DM. The appropriateness of the baseline utility value within UKPDS62 of 0.814 to patients with T1DM is a matter of conjecture, although an Australian study by Coffey et al. (2002)214 found a somewhat lower baseline value of 0.672 for those with T1DM.

Quality of life values are shown in Table 21, along with their stated sources.

Note that due to their short duration there are limited data on the quality of life impact arising from an SH event. Those studies that exist, for example Davis et al. (2005),219 Lundkvist et al. (2005),220 Tabaei et al. (2004)221 and Wikblad et al. (1996),222 are difficult to interpret due both to confounding variables and indeterminacy in terms of the duration of any quality of life impact from SH events. However, these papers do clearly show a significant effect upon quality of life from patients’ most SH events. The impact may depend on where the episode happened. An event at home may have less impact than one at work, which may lead to time lost, and loss of confidence in both subject and employer.

The submission appears to have assumed that a quality of life detriment of –0.0121 is associated with each SH event. While this appears to have been an arbitrary assumption, the value does not appear to be unreasonable. Perhaps, importantly, and as demonstrated in the discussion and modelling of the subsequent chapter, given the average rate of SH events, the results of the cost-effectiveness modelling are relatively insensitive to the quality of life detriment associated with each SH event. Results are mainly driven by the effect upon glycaemic control.

Industry submission modelling results

Given the assumptions of the modelling and the 50-year time horizon, the aggregate results of the CORE modelling can be summarised as shown in Table 22.

While an analysis based upon the entire Pickup and Sutton (2008)107 meta-analysis results sees a cost-effectiveness of a little over £30,000 per QALY, increasing the average reduction in HbA_1c from CSII to 1.29% effectively doubles the anticipated patient gain from CSII, while also slightly reducing the overall net cost given the reduced rates of complications requiring treatment.

The potential effect of the £413 cost per severe glycaemia event, as opposed to £62, may have had some impact upon the total costs as already noted, possibly being equivalent to a little over a 10% reduction in the net direct treatment costs of CSII.

Note also that within the trial base analysis the cumulative effect of CSII upon macrovascular events over the 50-year time horizon was leading to an absolute reduction in deaths from CHF of 0.7%, of deaths from MI of 0.6% and of deaths from stroke of 0.3%.

A full list of the results of the sensitivity analyses within the industry submission is presented in Appendix 6, the main results of these being summarised below.

Time horizon

As usual with diabetes, improved control now reduces complications years into the future, and so discounting has a large effect. In a relatively newly-diagnosed patient, the costs of CSII will be incurred now and every year thereafter, but the savings from, for example avoiding or postponing dialysis for end-stage renal failure, may not occur for 20–30 years. The industry submission includes various sensitivity analyses that alter the discount rates, which for the previous NICE discount rates of 1.5% for health effects and 6.0% for costs reduced the incremental cost-effectiveness ratio (ICER) for the trial-based analysis from £34,330 per QALY to £18,997 per QALY (see Table 33). The results of this sensitivity analysis were not reported for the other scenarios deemed to be more relevant to the UK setting within the industry submission.

As noted within the cost-effectiveness literature review, there may be some concerns around the possibility of CORE modelling tending to overestimate macrovascular events and in turn mortality within the population of those with T1DM. In parallel with the sensitivity analyses for discount rates, the results of sensitivity analyses on the time horizon appear only to be reported for the trials-based analyses: time horizons of 15 years, 10 years and 5 years, increasing the ICER from the base-case value of £34,330, to £42,039, £47,921 and £63,795 per QALY, respectively.

Hypoglycaemia

From the submission, it was not clear quite what the industry modelling includes in terms of the impact of reduction of hypoglycaemic episodes on quality of life, although the electronic modelling inputs uploaded to CORE website indicate a quality of life loss from each SH event of 0.0121, and also a quality of life loss from each non-SH event of 0.0052. As mentioned above, the cost of SH episodes is included at £413. The CORE model has a section for hypoglycaemic episodes, but Tables 27 and 28 of the industry submission do not mention hypoglycaemia; Table 29 includes the cost of hypoglycaemic episodes. As already noted, it appears that the industry submission has conservatively assumed that SH events have no death rate associated with them.

Fear of hypoglycaemia

The submission does not appear to include allowance for benefits such as reduction in fear of hypoglycaemias, noted in the NICE appraisal of glargine. 51 In that TA (TA 53) of long-acting insulin analogues (at that time only glargine), the NICE Appraisal Committee accepted that both hypoglycaemic episodes and the fear of such episodes recurring caused significant disutility. The relevant paragraph states:

The Committee accepted that episodes of hypoglycaemia are potentially detrimental to an individual’s quality of life. This is partly the result of an individual’s objective fear of symptomatic hypoglycaemic attacks as indicated in the economic models reviewed in the Assessment Report. In addition, as reported by the experts who attended the appraisal meeting, individuals’ quality of life is affected by increased awareness and uncertainty of their daily blood glucose status and their recognition of the need to achieve a balance between the risk of hypoglycaemia and the benefits of longer-term glycaemic control. The Committee understood that improvement in this area of concern regarding the balance between hypoglycaemia and hyperglycaemia could have a significant effect on an individual’s quality of life.

However, the guidance did not specify the amount of utility lost because of fear of hypoglycaemic episodes, and nor did the HTA report,75 because it was based on the industry submission from Aventis, which was classed as confidential. But clearly the utility gain from reducing the fear of hypoglycaemic episodes was enough to change a very large cost per QALY to an affordable one.

Other benefits not included

The submission does not factor into the ICER calculations, aspects of quality of life, reduction in depression, less cognitive impairment in children, and non-health-related quality of life gains, such as flexibility of lifestyle. Some of these omissions are understandable due to unavailability of data. In particular, it would be controversial to try to assess the impact of cognitive impairment on quality of life. If a child loses 10 points in IQ score, that does not mean that quality of life is reduced. In other cases, such as flexibility of lifestyle, the measures used most often in quality of life studies in diabetes may not capture the effect. 223

RCT studies: T1DM

Weintrob et al. (2003)225 performed a randomised crossover trial of CSII versus MDI among 23 Israeli children with T1DM, aged 9–14 years, with a crossover period of 3.5 months after a 2-week run-in period. Quality of life aspects were measured with the DTSQ and for more general quality of life through the DQoL-Y. All children completed the two study arms. There were no significant differences in glycaemic control between the two arms. However, there was a significant difference in DTSQ scores, which averaged 21.4 at baseline, 21.9 at the end of the MDI arm and 30.6 at the end of the CSII arm. No statistically significant differences were recorded within the DQoL-Y subscales, with the end of MDI arms and the end of CSII arm displaying similar central estimates for all DQoL-Y subscales. At the end of the trial, patients were asked which regime they preferred: 70% preferred CSII on grounds of greater mealtime flexibility, avoidance of the pain of injections and better glycaemic control or profiling. Of those preferring MDI, concern as to glycaemic control, overeating and weight gain were cited, coupled with the desire to keep diabetes a secret and shame at wearing the pump were also cited, as was the required frequency of self-monitoring.

Hoogma et al. (2006)226 reported the results of a five nation randomised controlled crossover trial of CSII against MDI among 272 patients with T1DM of whom 223 completed the trial. Patient selection and patient characteristics were not presented in the paper. While not explicitly stated, it appears that patients were randomised to starting an intensified regime of either CSII or MDI, with a run-in period of 2 months. Trial duration thereafter was 6 months. HbA_1c results were similar after run-in across both arms, with the mean difference at end of trial being in favour of CSII. Non-inferiority of CSII in terms of HbA_1c was demonstrated, with subsequent analysis indicating statistically significant superiority. Similarly, the rate of respondent-defined SH events at 0.2 per years for CSII compared with 0.5 for MDI was also statistically significantly different.

Against this background, patients completed the DQoL and Short Form-12 (SF-12) questionnaires at baseline and at end of treatment. The overall DQoL score at end of treatment was significantly higher for CSII, the individual subscales of satisfaction, impact and diabetes-related worry also being statistically significantly better, although the social/vocational worry subscale did not reach significance. Within the SF-12 questionnaire, no significant differences in physical health were recorded, but the composite mental health subscale for CSII was statistically superior to that of MDI. The authors concluded by noting that once patients had experienced CSII they were more likely to recommend it than the MDI regimen.

Hans DeVries et al. (2002)119 reported the results of a crossover trial of 79 Dutch patients with T1DM with an average age of 37 years. Unfortunately, due to dropouts at crossover, the trial was analysed as a parallel trial using only the first half of the crossover phase. The authors noted that 11% of patients at randomisation or at crossover refused to start CSII. The trial found statistically significant differences in the general health and mental health subscales of SF-36, with CSII recording improvements of 5.9 and 5.2 on these subscales, respectively, as against falls of 1.2 and 0.6 for MDI. Scores for the other subscales were not reported. Overall treatment satisfaction was assessed using the DTSQ. No statistically significant difference was recorded, with CSII scoring an increase of 1.3 and MDI scoring an increase of 0.2.

In an RCT of CSII versus MDI in US children of under 5 years of age with T1DM, Dimeglio et al. (2004)120 reported that CSII was generally well tolerated with 19 out of the 20 families opting to continue with CSII after 6 months, rather than switch to MDI. The preferences of those in the MDI arm with regards switching of therapy at the 6-month point were not reported.

Fox et al. (2005)121 performed a 6-month RCT of CSII versus MDI among 26 US children with T1DM, average age of a little under 4 years. The children were randomly assigned to continue receiving MDI, or to switch to CSII. Given the age of participants, aspects of parental rather than patient quality of life were measured, coupled with parental perceptions of patient quality of life. While both mothers and fathers reported more psychological distress in the MDI group than in the CSII group, these differences were not statistically significant when baseline differences were controlled for. Mothers in the MDI arm reported significantly greater stress at baseline, but this difference was no longer significant at 6 months’ follow-up. However, fathers reported significantly more positive quality of life changes in the CSII group over the 6 months. The authors noted that at the end of follow-up all MDI patients started CSII therapy, while all CSII patients continued on CSII therapy.

RCT studies: T2DM

Raskin et al. (2003)116 undertook a randomised parallel trial over 24 months of CSII versus MDI among 132 US CSII-naive patients with T2DM, aged > 35 years. Patient satisfaction was measured through administering components of a poorly documented PHASE V Technologies Outcome Informations System Questionnaire. Overall satisfaction with treatment, as measured over the 10 subscales administered, was scored as an increase from 59 to 79 in the CSII group compared with an increased from 64 to 70 among the MDI group, which was described as being statistically significant. It was unclear whether the baseline for the CSII group was prior to the initiation of CSII therapy or shortly thereafter. CSII was also statistically significantly superior to MDI in all subscales other than pain, where no statistically significant difference was noted.

Herman et al. (2005)115 in a 12-month RCT of CSII versus MDI among 107 older patients with T2DM, average age 66 years, evaluated patient satisfaction and quality of life through both the DQoLCTQ and the SF-36 questionnaire. Over the period of the trial both arms reported similar increases in satisfaction with their treatment, DQoL scores for CSII increasing from 52 to 81 and for MDI increasing from 50 to 78. Likewise, changes in the composite SF-36 physical and mental health subscales were similar between both arms: the physical score for CSII rising from 40.5 to 41.1, compared with a rise from 40.6 to 41.0 for MDI, while the mental health score fell slightly for CSII from 51.0 to 50.0, compared with a slightly greater fall for MDI from 53.0 to 50.5. None of these changes was statistically significant.

Case–control study: T1DM

Hoogma et al. (2004)227 presented the results of a cross-sectional study among 128 Dutch patients with T1DM with an average age of 42 years, 49 of whom received CSII and 79 of whom received MDI. The design of the study intentionally recruited twice as many patients using MDI as patients using CSII. The selection criteria for patients using CSII patients participating in the study was not stated, although patients using MDI were reportedly randomly selected from the same outpatient clinics. Quality of life aspects were evaluated through three questionnaires: DQoL, the DTSQ and the WHO Wellbeing Questionnaire.

Self-measurement of blood glucose was once daily in 81% of CSII patients as against 63% in patients using MDI. The majority of the remaining patients using CSII, 10%, measured blood glucose one or two times a week compared with 19% for the MDI arm. Full daily blood glucose profiling showed a similar profile, with more frequent profiling being performed within the CSII arm. No differences were uncovered regarding the outcomes of DQoL, and the treatment satisfaction again showed no differences between the groups, even with regards to hypo- and hyperglycaemia. With regards to the general wellbeing questionnaire, again no statistically significant differences were noted, except for the ‘energy’ subscale, for which the MDI arm showed somewhat better results with a score of 8.7 as against 7.5 for CSII. As with much of the patient preference and quality of life literature, these results are difficult to interpret as patient characteristics and the reasons for receiving CSII were not well documented.

Before/after studies: T1DM

Garmo et al. (2004)228 report a study of 27 Swedish adults with T1DM, average age 41 years, changing from MDI at baseline to CSII with follow-up at 6 months. It appears that the change to CSII was medically driven rather than part of a research programme per se, which may make results more applicable to the sort of patients who would start CSII in routine care. Satisfaction with treatment was measured using the DTSQ, with possible values ranging from 0 to 36. Prior to the change to CSII the median DTSQ score was 20, with a range of 4–32, while at 6-month follow-up after changing to CSII the median DTSQ score was 32, with a range of 17–36. However, these results are subject to a number of criticisms, the most significant of which is the before/after non-randomised nature of the study coupled with patients being in some sense self selecting due to their previous MDI therapy being presumably unacceptable for either clinical or personal reasons. Respondents were presumably failing on MDI, this also being reflected in a self-reported fall in the frequency of unacceptable hypoglycaemic events.

Bruttomesso et al. (2002)230 retrospectively review quality of life using the DQoL questionnaire among 138 Italian patients with T1DM, who were receiving CSII, and had been on average for 7.4 years. Ninety-eight of the patients surveyed completed at least one aspect of the DQoL questionnaire, all 98 completing the satisfaction subscale, average score 72.5, and the impact of diabetes subscale, average score 71.3. 95 completed the diabetes worry subscale, average score 80.2, but only 51 completed the social worry subscale, average score 67.8. The overall average DQoL score across patients and subscales was 73.0.

Sanfield et al. (2002)229 reported the results of a before/after study of 104 US patients with T1DM prior to the initiation of CSII therapy. Patient characteristics were not reported. Prior to initiating CSII, patients underwent up to three outpatient sessions to assess suitability, a large degree of which appears to have been education around self-selection for suitability for CSII. One of the criteria within this was financial, in terms of patients having adequate financial resources to cover the initial and ongoing costs of CSII. Of the 104 patients, 35% did not proceed through all three outpatient sessions to receive CSII. The reasons for this are not outlined, but given the financial criterion the relevance of this to the UK setting is questionable.

Patients proceeding to CSII completed SF-36 and three additional trial-specific quality of life questionnaires. The great majority, 97%, of patients initiating CSII and remaining in the area remained on CSII after over 2.5 years. Few details are provided as to the quality of life measures, but the paper notes that statistically significant improvements occurred over time in the SF-36 parameters of general health, ability to perform activities, energy and physical pain. Within the trial-specific quality of life questionnaires, patients are reported as stating that eating, working, sleeping, bathing and sexual activity were the most important aspects of life. CSII was found to interfere with bathing and sexual activity, although many patients removed pumps during these activities.

McMahon et al. (2005)152 prospectively followed 100 Australian children and adolescents with T1DM who were starting CSII therapy. Those selected for CSII therapy had demonstrated some or all of severe hypoglycaemia, poor glycaemic control and erratic lifestyle with regards sport, food or routine, although commencement was typically at parental request. Quality of life was measured prior to starting CSII and after 6 months of CSII among the first 51 patients being switched to CSII through a modified DQoL questionnaire and the SED scale, with respondents being of more than 10 years of age. Within the DQoL results, the impact of diabetes score fell, on average, from 55.4 to 50.2, a difference significant at the 5% level. Worries about diabetes and satisfaction with life scores showed no significant change over the 6 months. The SED scale improved from 159 to 174, which was described as being statistically significant. But in common with the Fox et al. (2005) study,121 these results require care in their interpretation given the before/after nature of the study, coupled with patients in some sense having been failing on their previous therapy.

Juliusson et al. (2006)171 performed another before/after study of the adoption of CSII upon patient quality of life, this being among 31 children with T1DM, average age of 14 years, and who were poorly regulated on MDI with an average HbA_1c of 10.4%. 171 Quality of life was measured with the DQoL questionnaire and generic CHQ-CF87 questionnaire prior to starting CSII and twice during 15 months of follow-up. Average differences in subscales of the CHQ-CF87 uniformly showed improvements, but these only reached statistical significance for the ‘family activities’ subscale. Similarly, while an improvement was recorded across the dimensions of the DQoL, none of these reached statistical significance. The authors concluded that respondents might have had moderate improvements in diabetes-specific quality of life scores, and were unlikely to have had appreciable deteriorations in them.

Rodrigues et al. (2005)140 reported the results of a before/after study of 40 UK patients with T1DM, average age 33 years, the study aiming to identify if current guidelines would correctly identify patients who would benefit from CSII. Twenty-five of the 40 patients were initiated on CSII for reasons other than SH events, while 15 patients reportedly had contraindications to CSII. The median follow-up was 20 months. Quality of life was measured although the DQoL, with the SF-36 and the Hypoglycemia Fear Survey also being administered. Only 33 questionnaires were returned, including four responses from patients who had discontinued with CSII. These four responses from those who had discontinued with CSII were excluded from the analysis. No significant differences were observed within any of the DQoL subscales. The only significant difference within the SF-36 reported was not between baseline and follow-up, but between those with and without contraindications to CSII. Those with contraindications had a significantly lower score on the mental health subscale of 47.5 than those without, who scored 69.9. Having excluded those discontinuing CSII from the analysis, all respondents, as expected, preferred CSII to their previous treatment.

Mednick et al. (2004)178 surveyed 22 US children with T1DM, average age of 14 years, and their parents, after having transferred to CSII and remaining on CSII. Data were collected between three and 22 months after CSII was initiated. Telephone interviews were conducted using the IPTSQ, which had been specifically designed for the survey by the authors, composed of 10 items ranked on a Likert scale, coupled with three open-ended questions as to life changes from pump use, the most challenging aspects of the pump and advice to prospective users to maximise their benefits from pump use. Satisfaction ratings were derived for parents and children. None was dissatisfied, and the average response on the 5-point Likert scale was consistently above ‘3’ and typically above ‘4’.

Shehadeh et al. (2004)174 briefly reported the results of a before after study of CSII among 15 Israeli children with T1DM, aged 1–6 years. Treatment satisfaction and quality of life were measured through DTSQ and DQoL, respectively, for parents at baseline and at 4 months. Both the overall DTSQ scores and the DQoL scores showed a statistically significant improvement at 4 months compared with baseline.

With all of the above before/after studies, many patients will have commenced CSII due to failure on their previous regime. As a consequence, these results may be of limited relevance to the consideration of whether those moving on to intensive insulin therapy would be best starting on CSII or starting on MDI. However, the results appear to be more useful than if the results had been representative of the total T1DM population for the situation where CSII is used in patients who are in some way failing on current therapy, as with the current NICE guidance. There is the additional difficulty that participation in the study may have affected results over the time period of the study.

Diabetic clinic survey: T1DM

Bruttomesso et al. (2006)141 reported the results of a survey of Italian diabetic centres, with 145 centres out of 179 centres responding. These covered a total of 2702 patients using CSII, of whom the average age was 39 and among whom 97% were patients with T1DM who had previously received MDI. The main reason for starting CSII was given as poor metabolic control, although quality of life, flexibility, reducing the number of hypoglycaemic events and correcting the dawn phenomenon were also cited. Reasons for not starting CSII were the inability to cope with the pump technology, lack of compliance, psychiatric problems and unwillingness to check blood glucose frequently. Notably 571 patients abandoned CSII, although quite what the relevant baseline or denominator figure for this was not clear from the paper. In total, 187 of these stopped at the end of pregnancy. Intolerance of the pump was also cited as a reason for discontinuing, coupled with disappointment as to the goals attained, infection at the infusion site and hypoglycaemia.

Meeting abstracts

Barnard and Skinner (2006)231 briefly reported the results of a qualitative telephone survey of 80 insulin pump users. The patient characteristics and method of respondent selection were not specified. Respondents were asked about the benefits of insulin pump use, and quality of life effects, any downsides to insulin pump use and whether they wished to raise any other issues. A number of positive themes emerged from the survey, with 56% reporting greater control, 41% reporting greater flexibility, 35% reporting increased freedom, 9% reporting greater convenience and 6% reporting greater independence. However, 59% also reported downsides, with 31% reporting difficulties with concealment, 21% reporting technical issues when things go wrong, 6% reporting site reactions and pain and 4% reporting cost. The authors noted that these negative factors may explain why a number of patients only remain on pump therapy for a short period of time. Given this, there may have been a degree of sample selection bias, with any results relating more to the quality of life of those finding insulin pumps of benefit in the longer term.

Reid and Lawson (2002)232 reported the results of a Canadian cross-sectional survey of 74 children with T1DM, 28 of whom had been using CSII for more than 4 months and among whom the average duration of CSII usage was 16 months. The average age of those using CSII was 14 years. Patients were matched for sex, age and duration of diabetes, with a linear regression model comparing values from the DQoL-Y questionnaire. Treatment satisfaction was significantly higher in the CSII group with a score of 33.8 as against 27.5 in the MDI group. However, despite metabolic control also being significantly better in the CSII group, with 7.7% compared with 8.9% for the MDI group, no significant differences were observed in the DQoL-Y dimensions of satisfaction, impact or worry. The authors concluded that DQoL-Y may not be the most appropriate measure of quality of life in this group.

Galatzer et al. (2002)233 compared the treatment satisfaction among 208 CSII and MDI people with T1DM of average age of 20 years, although ages ranged from 10–50 years through the use of the DQoL questionnaire. Patient selection and other characteristics were not reported. The overall treatment satisfaction score was significantly different at 30.7 among CSII patients compared with 22.7 among patients using MDI. Overall, 86% of CSII patients were reported as recommending CSII to other patients, compared with only 19% of patients using MDI.

Schweitzer et al. (2006)234 reported the results of a postal questionnaire sent to 36,450 patients using CSII, from whom a 38% response rate was achieved. The abstract restricted itself to the results of the 729 responses received from patients with T2DM, these having an average age of 56 years, an average duration of diabetes of 17 years and an average duration of CSII of 3.4 years. Most, 76%, reported selecting CSII due to poor glycaemic control on other therapies, with 44% also citing uncontrolled blood glucose fluctuations. A minority, 34% mentioned a high insulin requirement when using injections as a reason. Many (44%) patients initiated therapy as inpatients (which would be unusual in the UK), with 46% starting at a specialist diabetologist’s office, and 5% as a hospital outpatient. Only 3% initiated therapy via a GP. The authors concluded that patients were highly satisfied with CSII therapy, although the means of assessing this and associated results were not presented.

Ulahannan study

The industry submission was also accompanied by a 2007 study by Ulahannan et al. ,245 which reported the costs associated with CSII, and made the case that there could be short-term savings from adopting CSII. It will probably be widely quoted in support of CSII use, when pump clinics negotiate with PCTs over funding.

The study was carried out in Gloucester, UK. The aim of Ulahannan et al. (2007)245 was to collect data on NHS resource use by 34 patients starting CSII between June 2000 and June 2005. The resource use included:

• diabetic clinical visits

• appointments at other outpatient clinics

• hospital admissions, whether related to diabetes or not

• primary care contacts, at GP surgery, home visits, out of hours calls, and telephone contacts.

The aim was to collect such data for up to 5 years before and after, although for most patients it would be for much less. The exception was the subset of 17 patients for whom primary care data were collected, where the aim was to collect data for 2 years before and after. They also collected data on HbA_1c levels. The average length of follow-up for hospital data was 31 months prior to CSII and 34 afterwards. The before/after HbA_1c level showed an average drop of 1.2%.

The impact of CSII upon hospital resource use was reported as:

• Consultant outpatient visits fell from 2.4 to 1.3 per year; these were costed at £88 each.

• Nurse appointments were little changed, from 5.3 per year to 4.8 per year.

• Hospital admissions were reported to have fallen, but few details are given. The median number of admissions per month in both periods was nil, so most patients were not admitted before or after. Means were not given, and would have been more useful. Those who were admitted before were assumed to have been so because of severe hypoglycaemia, and costed accordingly, but this could not be confirmed, because the authors received only a crude breakdown of reason for admissions into diabetes or other cause. The average length of stay for diabetes admissions in the hospital was 10.7 days. The length of stay for a hypoglycaemic episode would be much shorter. The costs were given as £757 for a hypoglycaemic episode admission, and £1932.50 for other diabetes admissions.

• Primary care contacts were reported to have fallen by about half, from 11 to six appointments per year.

There are several problems with this study, due to there being various data deficits (i.e. some data were not available to the authors), and some inconsistencies in the paper:

• The reduction in outpatient appointments would not release any real savings, in that the consultant would not be paid less but would do other things – an opportunity cost gain but not a financial saving. The figure used to estimate cost savings was 1.44 fewer diabetes consultant appointments per year, which does not tally with the figure of 1.1 provided from hospital records in the previous table.

• Second, the lack of data on hospital admission costs makes any cost calculation unsafe. Table 3 of the paper suggests that 0.132 admissions per patient per year would be saved, but this is not compatible with the range of admissions per year given in the previous table, of 0–1.2.

• Third, any reduction in admissions would not release real savings unless beds were closed and staff made redundant, which would not happen given the very small number of admissions involved. Again, although this may yield an opportunity cost gain there are unlikely to be any associated financial savings.

• Similarly, if there is a reduction in primary care contacts, no funds would be released. The practices would be very slightly less busy.

Although the study provides some evidence for asserting that CSII may provide some gains in terms of opportunity costs and the freeing of resources to undertake other activities, the assertion that CSII will be accompanied by short-term savings in hospital and primary care cannot be supported by this study. A more detailed study is required.

Cost meeting abstracts

Bolli et al. (2004)113 undertook a 6-month multicentre RCT of CSII versus MDI among 57 patients with T1DM. Both arms of the trial showed similar improvements in glycaemic control, blood pressure and glycaemic events, with any differences in outcomes being non-significant. While treatment costs were not reported, the authors did report that CSII treatment costs were 400% those of MDI. The authors concluded that a glargine-based MDI regime was more cost-effective than CSII in an unselected T1DM population.

Clinical effectiveness

The key clinical elements within the modelling relate to the differences between the HbA_1c level and the rate of severe hypoglycaemia episodes between MDI and CSII. The base case uses the CORE model to assess the cost-effectiveness of CSII in a population of average age 40 years with T1DM within whom it seems well suited. While the CORE model is not suited to modelling adolescent and paediatric use, the effect of a younger cohort will be explored through an adoption of an average age of 30 years.

The base-case effect on the baseline HbA_1c level assumes a baseline level of 8.8% HbA_1c on MDI, reduced by 0.9% to 7.9% by CSII (based on the results from the Insulin Pumps Clinical Database for the 20–39 age range, which seems the most appropriate group for our purposes). Three sensitivity analyses are undertaken with regards to the effect of CSII upon glycaemic control.

• The first uses the results from the meta-analysis by Pickup and Sutton, (2008)107 of a reduction of 0.6%.

• The second uses the results from the study by Pickup et al. (2005)138 that show a reduction of 1.4% from a baseline of 9.0%, but no reduction in the rate of SH events

• The third assumes that some pump users who have a high rate of SH events will gain no reduction in HbA_1c levels because they start with good control hence a baseline of 7.5%, but achieve a reduction in their rate of SH events.

A general population of those with T1DM can be characterised as having a rate of SH events of 18.7 per 100 patient-years as within the NICE appraisal of the cost-effectiveness of glargine. The costing within this report has assumed the use of glargine or other long-acting analogue, which the Assessment Group for the glargine appraisal estimated could reduce the severe hypoglycaemia rate to as little as 8.8 per 100 patient-years. However, the Final Appraisal Document noted that this might be an overestimate and as a consequence a baseline rate for the general population with T1DM of 18.7 per 100 patient-years will be assumed. The effect of CSII upon this rate will be explored as a 50% reduction and a 75% reduction in SH events, with no effect upon SH events also being explored as a sensitivity analysis.

It seems likely that the main focus for NICE for the application of CSII will be on patients with more severe problems with hypoglycaemia. In common with the industry submission, this will be explored through the assumption of a rate of 62 per 100 patient-years, with reductions of 50% and 75% within this group. Simulations will also explore the possible effects of alternative cost scenarios as outlined within the costs section above. As noted within the literature review, the CORE model may have a tendency to overstate overall mortality within T1DM over longer time horizons, which may be due to a possible overestimation of macrovascular complications and their associated mortality. As a consequence, this will also be explored through 50-, 30- and 10-year time horizons coupled with a reporting of the evolution of the estimated macrovascular events and survival over this period. While the 10-year time horizon is too short for an accurate evaluation of the cost-effectiveness of CSII relative to MDI, it permits the evolution of cost-effectiveness to be explored and highlight timing of the anticipated main gains.

The group of patients who start with good control as reflected in HbA_1c, but achieved at the cost of more problems with severe hypoglycaemia, will get little or no reduction in HbA_1c levels. The effect of CSII upon HbA_1c levels will be set to nothing for this group. However, the baseline rate of SH events will be increased to 134 per 100 patient-years as in Boland et al. (1999),99 with the effect of CSII being explored through reductions of 50% and 75%, the 50% reduction corresponding closely to the reduction to 76 per 100 patient-years from CSII as reported by Boland et al.

Treatment costs

More detail on costs is given in Appendix 7.

Results

Conclusions

The possible benefits of CSII include:

• improved glycaemic control

• reduced frequency of hypoglycaemia

• a reduction in the chronic fear of severe hypoglycaemia

• a reduction in insulin dose giving modest savings

• improved non-health related quality of life arising from greater flexibility of lifestyle.

If severe hypoglycaemia has a cost of only £65 per event, whether CSII has no effect on frequency of events, halves it or reduces it by 75% has little impact upon the estimated cost-effectiveness of CSII. This is because the main driver of cost-effectiveness in the CORE model is HbA_1c level. Severe hypoglycaemia has little effect because episodes are of short duration.

This is underlined by simulation among those with high rates of severe glycaemia events but good baseline HbA_1c levels, in whom CSII has no effect upon glycaemic control, but causes reductions of 50% and 75% in the rate of severe glycaemic events. Despite the larger absolute falls in severe glycaemia events within this group, the estimates of the cost-effectiveness of CSII relative to MDI rise to six figure sums per QALY. This does not take chronic fear of hypoglycaemia into account.

For the base-case estimate of a 0.9% reduction upon a baseline 8.8% HbA_1c level, the central estimate of the cost-effectiveness of CSII relative to MDI is estimated as £36,587 per QALY. These figures are based upon an estimated reduction in SH events from 62 per 100 patient-years to 31 per 100 patient-years. If CSII reduces this rate to only 15.5 per 100 patient-years the cost-effectiveness of CSII for a 0.9% reduction to a baseline 8.8% HbA_1c level is only slightly reduced to an estimate of £33,361 per QALY.

A lesser effect upon glycaemic control of only a 0.6% reduction upon a baseline 8.8% HbA_1c level while retaining a 50% reduction in SH events leads to a cost-effectiveness estimate for CSII of £53,788. A greater effect upon glycaemic control of a 1.4% reduction upon a baseline 9.0% HbA_1c level with no reduction in SH events leads to a cost-effectiveness estimate for CSII of £24,720 per QALY.

There is uncertainty as to the net treatment cost of CSII, it being plausible for this to range from a central estimate of £1700 up to a high of around £1900, and down to a low of around £1500. Despite this range, the estimate of the cost-effectiveness of CSII relative to MDI remains slightly above the usual cost-effectiveness thresholds.

However, if the net price of CSII is lower than that assumed for the base case, the impact of a 75% reduction in SH events as opposed to a 50% reduction is just sufficient to tip the cost-effectiveness of CSII from being slightly above £30,000 per QALY to being slightly below £30,000 per QALY.

One of the main uncertainties relates to the additional quality of life benefit from a reduction in the fear of severe hypoglycaemia. While this effect may apply only to a subgroup of patients, the quality of life gain from a reduced fear of severe hypoglycaemia would have to be only relatively minor to make CSII cost-effective.

An annual gain of only 0.01 in quality of life from the reduced fear of severe hypoglycaemia would be sufficient to reduce the ICER for CSII to around the £30,000 per QALY, while an annual gain of around 0.03 in quality of life would reduce it to around £20,000 per QALY. But note that this is within the base-case population: those patients with T1DM who are poorly controlled on MDI and for whom CSII results in a 0.9% improvement in glycaemic control and a halving in the rate of SH events.

Background

Many centres across the UK are experiencing a demand for insulin pumps from patients, as the awareness of the success of this form of therapy has entered the public domain. This has become particularly so for children and young people. 253

This section examines the accounts of parents with young children with T1DM who have changed from using injections to using pump therapy. We believe provision of CSII to young children may be one of the current growth areas and perhaps one of the most marked changes since the last assessment report. 50

Understanding patient-centred care for children and young people with T1DM, as outlined in the National Service Frameworks for both Children (DoH 2004)254 and diabetes (Scottish Executive 2002),255 requires an understanding of children’s individual autonomy as well as the executive role of parents, and the important contribution they make to the successful management of diabetes. In this section, parents’ accounts of switching to pumps from MDI are divided into themed headings. Time did not allow a larger study. This report aims to offer some background to choices of therapy made by parents with children with T1DM, and the problems they faced.

Methods and subjects

Details of methods and rationale are given in Appendix 11. The main questions posed are given below:

• Why did parents decide to use CSII, how did they obtain information, how was CSII managed, to what extent has CSII affected diabetes outcome, and what lessons have been learned?

• To what extent have different clinical practices affected parental use of CSII in young children, and what factors explain the variation in opinion about the use of CSII in young children, i.e. to gain a better understanding as to why the previous NICE guidance74 has been implemented to varying extents in different parts of the country?

• What are the benefits and challenges of using CSII in young children?

The ages at onset of diabetes, at start of CSII, and at time of interview are shown in Table 38, along with funding arrangements and HbA_1c changes.

Results

Discussion

This qualitative study examined the beliefs and attitudes of parents of young children who have been successfully started on pump therapy in the UK in the last 5 years. While accepting that this is a biased account from parents who have had considerable difficulties in starting their children on CSII, they nonetheless reflect some important aspects of the benefits and challenges of pump therapy.

Interestingly, none of the children concerned had treated themselves for any length of time with ‘classical’ intensive insulin injection therapy – MDI. This was because of several reasons: very young age, erratic eating and exercise patterns, and a difficulty in interpreting multiple blood glucose results. However, perhaps the major factor was the reluctance of schools to take on children requiring MDI, forcing the parents to use ‘conventional’ twice daily injections of insulin. Schools frequently insisted on extra staff to supervise the intensive therapy. The parents, therefore, moved rapidly in the course of their young child’s diabetes on to pump therapy, with few difficulties. This was reflected by the schools, who also seemed more at ease with pump therapy, accepting that the parents and the children themselves were the experts and no additional supervision was necessary.

Overall for this age group the benefits of the pump outweighed significantly the challenges and difficulties. Glycaemic excursions were dramatically reduced, with improvement in overall glycaemic control, less hypoglycaemia and no episodes of DKA. The children felt better. This was with fewer injections than with MDI. The wearing of the pump did not produce significant difficulties and these young children appeared to cope well with: the practical issues; pump technology; wearing of the pump; and managing diabetes with the pump. The ‘quality of life’ for both parents and children appeared to be markedly improved.

These were all committed parents who had to seek information about the availability and the practical issues of pump therapy. A significant amount of their information came from outwith the UK. Their local diabetes teams for the majority were not supportive of pump therapy and the parents appeared to become evangelists in order to seek out pump therapy, often travelling some distance to receive sympathetic and expert advice. The majority felt this process had delayed the placing of their child on a pump and most believe that children of all ages should be considered for pump therapy from diagnosis.

The parents expressed a strong view that considerable commitment was required to master pump therapy and accepted that not all parents would be prepared and/or able to give this commitment. However, a significant factor in allowing parents to consider pump therapy would be the valuation that their clinical team places on this form of therapy.

The parents were aware of the cost issues for the NHS. However, all had made a case to their funding committees in their local PCTs within the NHS. (Eight out of the ten had their pumps paid for totally by the NHS; two parents were paying towards the cost of the consumables.) Undoubtedly, all felt that the benefits of pump therapy made it cost-effective.

Acknowledgements

We are grateful for INPUT for recruiting the families and thank the parents who freely gave us their time and intimate thoughts.

Caveats

The patients’ perspectives section is based largely on written statements from pump users, and several caveats are required. First, most comments have come from members of INPUT, who have responded to a request for comments. They are likely to be a more motivated group than average and some are clearly highly organised individuals. This does not affect the validity of their comments, but may have implications for generalisability. Second, they are successful pump users and tend to be enthusiasts for the technology. That is less important because those who do not succeed will not incur the ongoing costs of pumps. Third, most have had to pay for the pumps and consumables themselves, which creates another selection bias. Fourth, because pumps are little used in the UK – it appears that most of those who have gone on to CSII have done so because they have had a lot of trouble with control of blood glucose or frequent hypoglycaemic episodes, that is, a severity bias. They may have more to gain than the average person with insulin-treated diabetes. Again, this does not affect the validity of the findings, but will be relevant to discussions about the proportion of people with diabetes who should be considered for CSII. 11

The respondents for the last assessment report mentioned the expected gains in HbA_1c and hypoglycaemic episodes, but also emphasised flexibility of lifestyle and working patterns, having more energy, feeling in control, less visibility of diabetes (being able to take bolus insulin without anyone noticing), and better moods in children (an almost universal comment in submissions by parents, but not mentioned in the trials):50

Her HbA1c levels dropped from 10.6 to 8.2, and her mood and personality changed – we got our little girl back again.

(Parent 8)

One comment from several mothers was that they found it very difficult to work full-time with a young diabetic child:

I have found it hard to go back to work as I seem to be on call for him all the time. For example, I will drop him off at school at 9.30 and by 11 am they can phone me because he has gone low, and I have to go back to the school

(Parent 3)

There were many problems with schools, but the comments were consistent in saying that school life was easier on CSII and, because staff cannot take responsibility for injections, children find it easier to look after themselves, have fewer hypoglycaemic episodes, do not need to eat at special times, can miss meals if necessary, and do not need to carry insulin syringes and vials.

Other types of CSII

Intraperitoneal infusion would be more physiological than subcutaneous, as insulin normally goes into the liver via the portal vein. This could be done in two ways:

• From an external pump, as used for CSII but with the catheter into the peritoneal cavity. This would create a risk of infection, at entry site, along the tunnel through the subcutaneous tissues, and peritonitis.

• From an implanted pump. Such pumps have been in use for over 20 years in countries such as France, the USA and the Netherlands. 268,269 The EVADIAC (Evaluation dans le Diabéte du Traitement par Implants Artifs) group, Toulouse, France, which maintains a register of patients with implanted pumps, suggests three main indications:269

  1. – poor glycaemic control despite intensive CSII with good patient education and close follow-up

  2. – good control achieved but with unacceptable hypoglycaemia

  3. – improved quality of life.

Other forms of childhood diabetes

We are aware that CSII is used in children with other much less common forms of diabetes, such as cystic fibrosis-related diabetes (not common in children with cystic fibrosis, but very common in the over-15s) and associated with treatment for acute leukaemia. No research has been published for such groups. Numbers may be too small for research to be worthwhile.