Integrated sensor-augmented pump therapy systems [the MiniMed® Paradigm Veo system and the Vibe and G4® PLATINUM CGM (continuous glucose monitoring) system] for managing blood glucose levels in type 1 diabetes: a systematic review and economic evaluation

The MiniMed Paradigm Veo system

The MiniMed Paradigm Veo system has three components:

1. a small glucose sensor, placed under the skin, which measures glucose levels every 5 minutes, 24 hours per day (this sensor must be replaced every 6 days)

2. the MiniLink™ transmitter (Medtronic Inc., Northridge, CA, USA), which sends the information to the Paradigm Veo insulin pump

3. the Paradigm Veo insulin pump.

The system is complete and stand alone and not directly interchangeable with other manufacturers’ pumps or sensors. Many insulin formulations can be used in the insulin pump. In this report, we will focus on only fast-acting insulin formulations, because this type of formulation in the preferred clinical option for use with insulin pumps in the UK. 23

Continuous glucose monitors measure the level of tissue glucose electronically on a continuous basis (every few minutes). They use a subcutaneous, disposable glucose sensor placed just under the skin to measure interstitial glucose levels. The glucose sensor of the Veo system is replaced every 6 days. The sensor is connected to a non-implanted transmitter (MiniLink) which communicates glucose levels wirelessly to the Paradigm Veo pump. The pump displays BG levels with nearly continuous updates, as well as monitoring rising and falling trends. The pump can prompt a person with diabetes, or a carer, to take action to maintain glucose levels. The insulin pump delivers continuous subcutaneous insulin according to a pre-programmed pattern, which can be adapted by the user or a carer in response to real-time glucose trends.

The MiniMed Paradigm Veo system appears to be unique in that it will actively suspend insulin delivery if it predicts a hypoglycaemic episode. This ‘low glucose suspend’ (LGS) function stops insulin delivery for 2 hours if there is no response to a low glucose warning.

Users of this system must perform regular (a minimum of two per day) capillary BG tests (such as a finger prick tests), as CGM measures interstitial fluid glucose levels, not capillary BG levels. Further finger prick tests are required to confirm a CGM value before making any adjustments to diabetes therapy.

The pump can be worn on a belt or in a pouch underneath clothes. Insulin is delivered through a small tube (or ‘infusion set’) placed under the skin. The transmitter is directly connected to the glucose sensor, which is inserted through the skin, usually in the stomach area. The manufacturer’s information for use document states that the infusion set should be replaced every 3 days.

The Vibe and G4 PLATINUM CGM system

The Vibe and G4 PLATINUM CGM system is a CGM-enabled insulin pump, integrated with the G4 PLATINUM sensor. It is similar to the MiniMed Paradigm Veo system in that the glucose sensor is placed under the skin and measures interstitial glucose levels rather than capillary BG levels. Confirmatory capillary BG tests are also required to confirm the value displayed by the continuous glucose monitor before making any adjustments to diabetes therapy. The sensor is approved for up to 7 days of wear.

The insulin pump in the Vibe and G4 PLATINUM CGM system also delivers insulin continuously from a refillable storage reservoir by means of a subcutaneously placed cannula and the pump can be programmed to deliver insulin at a basal rate throughout the day, with the option of triggering higher infusion rates at mealtimes, either as a bolus dose or over time. The pump can be programmed to deliver insulin at different basal rates at different times of the day and night.

The system produces glucose level readings in real time, alerts users of high or low readings, and glucose trend information. It does not have an automated LGS function.

Inclusion and exclusion criteria

Search strategy

Systematic literature searches were conducted to identify studies of SAPT for T1DM (specifically the MiniMed Paradigm Veo system and the Vibe and G4 Platinum system), as well as RCTs and economic evaluations of insulin pump/infusion therapy and MDIs for T1DM. Search strategies were developed using the recommendations of the Centre for Reviews and Dissemination guidance for undertaking reviews in health care,24 and the Cochrane Handbook. 26 The search strategies used relevant search terms, comprising a combination of indexed keywords (e.g. from medical subject headings and the EMBASE thesaurus EMTREE) and free-text terms appearing in the titles and/or abstracts of database records. Search terms were identified through discussion among the review team, by scanning background literature and ‘key articles’ already known to the review team, and by browsing database thesauri. The search strategies were structured using search terms for ‘type 1 diabetes’ in combination with search terms for ‘sensor-augmented pump therapy’. Two further search term facets were included to capture ‘insulin infusion’ and ‘multiple daily injections’. In addition, the search strategy for clinical effectiveness studies included a sensitive methodological search filter designed to identify RCTs. The EMBASE search strategy was translated so that it could be run effectively in each of the databases searched. No date or language limits were applied. The main EMBASE search strategies were independently peer reviewed by a second information specialist using the Canadian Agency for Drugs and Technologies in Health peer review checklist. 27

Details of the full search strategies are presented in Appendix 1.

The following databases and resources were searched for relevant RCTs, systematic reviews and health technology assessments:

• MEDLINE (via OvidSP): 1946–2014/Aug week 4

• MEDLINE In-Process Citations and Daily Update (via OvidSP): up to 4 September 2014

• PubMed (National Library of Medicine): up to 5 September 2014

• EMBASE (via OvidSP): 1974–2014/week 34

• Cochrane Database of Systematic Reviews (Wiley Online Library): issue 9/September 2014

• Cochrane Central Register of Controlled Trials (Wiley Online Library): issue 8/August 2014

• Database of Abstracts of Reviews of Effects (Wiley Online Library): issue 3/July 2014

• Health Technology Assessment (HTA) Database (Wiley Online Library): issue 3/July 2014

• Science Citation Index (Web of Science): 1988–29 August 2014

• Latin American and Caribbean Health Sciences Literature (http://lilacs.bvsalud.org/en/): 1982–5 September 2014

• National Institute for Health Research HTA Programme (www.hta.ac.uk/): up to 5 September 2014

• PROSPERO (www.crd.york.ac.uk/prospero/): up to 5 September 2014

• US Food and Drug Administration (www.fda.gov): up to 5 September 2014

• Medicines and Healthcare products Regulatory Agency (www.mhra.gov.uk/): up to 5 September 2014

Completed and ongoing trials were identified by searches of the following trials registries:

• US National Institutes of Health ClinicalTrials.gov (www.clinicaltrials.gov/): up to 2 September 2014

• Current Controlled Trials (www.controlled-trials.com/): up to 5 September 2014

• World Health Organization International Clinical Trials Registry Platform (www.who.int/ictrp/en/): up to 5 September 2014

Conference proceedings were also searched from the organisations: Diabetes UK, the European Association for the Study of Diabetes and the American Diabetes Association (see Appendix 1).

The bibliographies of identified research and review articles were checked for relevant studies. As a number of databases were searched, there was some degree of duplication. In order to manage this issue, the titles and abstracts of bibliographic records were downloaded and imported into EndNote X7 (Thomson Reuters, CA, USA) reference management software and duplicate records removed. Rigorous records were maintained as part of the searching process. Individual records within the Endnote reference libraries were tagged with searching information, such as searcher, date searched, database host, database searched, search strategy name and iteration, theme and search question. This enabled the information specialist to track the origin of each individual database record and its progress through the screening and review process.

Inclusion screening and data extraction

Two reviewers independently screened the titles and abstracts of all reports identified by searches and any discrepancies were discussed and resolved by consensus. Full-text copies of all studies deemed potentially relevant, after discussion, were obtained and the same two reviewers independently assessed these for inclusion; any disagreements were resolved by consensus. Details of the studies excluded at the full-paper screening stage are presented in Appendix 2.

Data relating to study details, participants, intervention and comparator tests, and outcome measures were extracted by one reviewer, using a piloted, standard data extraction form. A second reviewer checked data extraction and any disagreements were resolved by consensus.

Quality assessment

The methodological quality of included studies was assessed using standard tools. 24 The assessment of the methodological quality of each included study was based on the Cochrane Collaboration quality assessment checklist,26 as detailed in Table 1.

Each study was awarded a ‘yes’, ‘no’ or ‘unclear/unknown’ rating for each individual item in the checklist. Any additional clarifications or comments were also recorded.

Quality assessment was carried out independently by two reviewers. Any disagreements were resolved by consensus. The results of the quality assessment were used for descriptive purposes to provide an evaluation of the overall quality of the included studies and to provide a transparent method of recommendation for the design of any future studies. Based on the findings of the quality assessment, recommendations were made for the conduct of future studies.

Methods of analysis/synthesis

If meta-analysis was considered unsuitable for some or all of the data identified (e.g. because of the heterogeneity and/or small numbers of studies), we employed a narrative synthesis. Typically, this involves the use of text and tables to summarise data. These allow the reader to consider any outcomes in the light of differences in study designs and potential sources of bias for each of the studies being reviewed. Studies were organised according to which therapies were being compared.

The methods used to synthesise the data were dependent on the types of outcome data included, and the clinical effectiveness and statistical similarity of the studies. Possible methods of data synthesis include the types of analysis discussed in the following sections.

Results of literature searches

The literature searches of bibliographic databases identified 9870 references. After initial screening of titles and abstracts, 555 were considered potentially relevant and were ordered for full-paper screening. Of the total of 555 publications considered potentially relevant, 29 could not be obtained within the time scale of this assessment. Most of these 29 unobtainable studies were published before 2000 or were conference abstracts; only four were possibly relevant trials published after 2000, but, based on their abstracts, it was unclear whether or not they fulfilled the inclusion criteria. Figure 1 shows the flow of studies through the review process and Appendix 2 provides details, with reasons for exclusions, of all the publications excluded at the full-paper screening stage.

FIGURE 1.

Flow of studies through the review process.

Based on the searches and inclusion screening described above, 54 publications resulting from 19 studies were included in the review. In addition, 19 publications of 18 ongoing studies were found (see Ongoing studies).

One study32 compared the MiniMed Veo system (with suspend function) with an integrated CSII + CGM system (MiniMed Veo with suspend function turned off) and another33 compared it with CSII + SMBG. Seven other studies compared an integrated CSII + CGM system with CSII + SMBG (three studies)34–36 or with MDI + SMBG (four studies). 37–40 The remaining 10 studies41–50 compared CSII + SMBG with MDI + SMBG. None of the studies included a treatment arm with CSII + CGM or MDI + CGM as a comparator (Table 2). Although several studies included an integrated CSII + CGM system as a treatment arm, it is important to note that none of these studies looked at the Vibe and G4 PLATINUM CGM system; in the included studies, the integrated CSII + CGM system was always a MiniMed Paradigm pump with an integrated CGM system.

Out of the 19 studies, eight were performed in North America32,34,38–40,46,48,49 and eight in Europe. 36,37,41–45,50 The remaining three studies were performed in Australia (two studies33,35) and Israel (one study47). Three out of the eight European studies included patients from the UK. 37,41,45

Twelve studies reported data for adults, five studies reported data for children and five studies reported data for mixed populations (adults and children). Two of these studies reported data for all three groups. One study included pregnant women (Table 3).

Table 4 shows the inclusion criteria, regarding the HbA_1c levels and hypoglycaemic events, used in the included studies. Further details of the characteristics of study participants and the interventions, comparators and results are reported in the data extraction tables presented in Appendix 3. It is clear from Table 3 that most studies included patients who had never used a pump before. However, both of the studies looking at the MiniMed Veo system (ASPIRE in-home32 and Ly et al. 33) included patients who had at least 6 months’ experience of using an insulin pump. In addition, baseline HbA_1c levels differ considerably among studies. DeVries et al. 42 included patients with poor control at baseline who were pump-naive. The two studies looking at the MiniMed Veo system included patients with relatively good glycaemic control at baseline; however, that might have been because those patients had been using an insulin pump for at least 6 months. Other studies, such as Bolli et al. ,41 included patients with relatively good glycaemic control at baseline without any previous pump experience. Therefore, there is considerable heterogeneity among the study populations.

Most studies were rated as having a high risk of bias (11 out of 19), four studies were rated as having an unclear risk of bias and another four studies were rated as having a low risk of bias (see Appendix 2). The most problematic factor with regard to the risk-of-bias assessment was the lack of blinding (of participants, physicians and outcome assessors) in the included studies. For participants and physicians, it is almost impossible to perform a trial with true blinding with these types of interventions. Nevertheless, the fact that participants and physicians were not blinded may bias the results, and the outcome assessment for HbA_1c measurement could be performed blinded. Selective outcome reporting was assessed as having a high risk of bias in only three trials. Incomplete data reporting was assessed as having a high risk of bias in 12 trials; this was rated as unclear in four trials. Overall, there was a high risk of bias in the included trials.

In the following chapters, we will discuss the results of the included studies by population (i.e. adults, children and pregnant women) and by follow-up time (i.e. 3 months, 6 months and 9 months or more).

Effectiveness of interventions in adults

We found 12 studies that reported data for adults. 32,34,37–46 As can be seen in Table 5, the age ranges differed considerably; therefore, we asked a panel of four expert committee members whether or not they thought that the results of these studies could be pooled. Three clinical experts agreed that the studies were similar enough to be pooled, as far as the differences in age ranges were concerned, and the fourth clinical expert did not respond.

Veo versus integrated CSII + CGM, CSII + SMBG and MDI + SMBG

For two outcomes [change in HbA_1c levels and diabetic ketoacidosis (DKA)], results of the MiniMed Veo system versus other treatments were available for 3-month follow-up in adults from more than one study,38,39 which could be combined in indirect comparisons. These two outcomes are reported below.

Change in glycated haemoglobin levels at 3-month follow-up

We found four studies32,38,39,42 comparing change in HbA_1c levels at 3-month follow-up in adults, allowing a comparison of the MiniMed Veo system with an integrated CSII + CGM, CSII + SMBG and MDI + SMBG. Figure 2 shows the network linking these interventions and Table 7 shows the results.

FIGURE 2.

Network of studies comparing change in HbA_1c levels and DKA at 3-month follow-up in adults. Note: green boxes represent the interventions; lines represent comparisons between interventions at 3-month follow-up; and transparent boxes represent studies in adults.

TABLE 7 - Results of the indirect comparisons with regard to change in HbA1c at 3-month follow-up

WMD values of < 0 indicate that the results favour the intervention listed in column 1. Differences are significant if the CIs do not include 0 (indicated in bold).

Intervention	Integrated CSII + CGM, WMD (95% CI)	CSII + SMBG, WMD (95% CI)	MDI + SMBG, WMD (95% CI)
Veo	0.04 (–0.07 to 0.15)	0.41 (–0.31 to 1.13)	–0.43 (–0.95 to 0.10)
Integrated CSII + CGM		0.37 (–0.34 to 1.08)	–0.47 (–0.98 to 0.04)
CSII + SMBG			–0.84 (–1.33 to –0.35)

The results of these indirect comparisons show that there are no significant differences between the MiniMed Veo system and any other intervention in change in HbA_1c levels at 3-month follow-up. Similarly, there are no significant differences between the integrated CSII + CGM system and any other intervention in change in HbA_1c levels at 3-month follow-up. The only significant difference found in this analysis was the difference between CSII + SMBG versus MDI + SMBG; in this regard, the results favour CSII + SMBG.

Diabetic ketoacidosis at 3-month follow-up

The same four studies32,38,39,42 provided data for DKA at 3-month follow-up in adults, allowing a comparison of the MiniMed Veo system with an integrated CSII + CGM system, CSII + SMBG and MDI + SMBG. However, the study that compared the MiniMed Veo system with the integrated CSII + CGM system (ASPIRE in-home)32 could not be included in the analysis as no events were reported in either arm. The results of the indirect comparisons for DKA are shown in Table 8.

TABLE 8 - Results of the indirect comparisons with regard to DKA at 3-month follow-up

RR values of < 1 indicate that the results favour the interventions listed in column 1. Differences are significant if the CIs do not include 1.

Intervention	Integrated CSII + CGM, RR (95% CI)	CSII + SMBG, RR (95% CI)	MDI + SMBG, RR (95% CI)
Veo	No events	No events	No events
Integrated CSII + CGM		0.26 (0.01 to 8.53)	0.32 (0.04 to 2.86)
CSII + SMBG			1.25 (0.08 to 19.22)

The results of these indirect comparisons show that there are no significant differences with between the integrated CSII + CGM system and any other intervention with regard to DKA at 3-month follow-up. The comparison between CSII + SMBG and MDI + SMBG also showed no significant difference.

Integrated CSII + CGM versus CSII + SMBG and MDI + SMBG

Results at 3-month follow-up

Proportion of patients with severe hypoglycaemia

The results of these indirect comparisons (Figure 3 and references 38, 39 and 42 therein) suggest that there are no significant differences between the integrated CSII + CGM system and any other intervention with regard to the ‘proportion of patients with severe hypoglycaemia’ at 3-month follow-up. The comparison between CSII + SMBG and MDI + SMBG also showed no significant difference. These findings are summarised in Table 11.

FIGURE 3.

Network of studies comparing ‘severe hypoglycaemia’ at 3-month follow-up in adults.

TABLE 11 - Results of the indirect comparisons with regard to the proportion of patients with severe hypoglycaemia at 3-month follow-up in adults

RR values of < 1 indicate that the results favour the intervention listed in column 1. Differences are significant if the CIs do not include 1 (these are in bold).

Intervention	CSII + SMBG, RR (95% CI)	MDI + SMBG, RR (95% CI)
Integrated CSII + CGM	0.33 (0.03 to 3.87)	0.19 (0.02 to 1.51)
CSII + SMBG		0.63 (0.17 to 2.31)

Results at 6-month follow-up

Change in glycated haemoglobin levels

The results of these indirect comparisons (Figure 4 and references 34, 37 and 41 therein) suggest that there are no significant differences between the integrated CSII + CGM system and CSII + SMBG with regard to change in HbA_1c levels at 6-month follow-up. The comparison between CSII + SMBG and MDI + SMBG also showed no significant difference. The comparison between the integrated CSII + CGM system and MDI + SMBG did show a significant difference, favouring the integrated CSII + CGM system. These findings are summarised in Table 12.

FIGURE 4.

Network of studies comparing change in HbA_1c levels at 6-month follow-up in adults.

TABLE 12 - Results of the indirect comparisons with regard to change in HbA1c levels at 6-month follow-up in adults

WMD values of < 0 indicate that the results favour intervention listed in column 1. Differences are significant if the CIs do not include 0 (these are in bold).

Intervention	CSII + SMBG, WMD (95% CI)	MDI + SMBG, WMD (95% CI)
Integrated CSII + CGM	–0.05 (–0.31 to 0.21)	–1.10 (–1.46 to –0.74)
CSII + SMBG		–0.10 (–0.52 to 0.32)

Proportion of patients achieving glycated haemoglobin levels < 7%

Results of these indirect comparisons (Figure 5 and references 34 and 37 therein) suggest that there are no significant differences between the integrated CSII + CGM system and CSII + SMBG with regard to ‘HbA_1c levels < 7%’ at 6-month follow-up. However, the comparison between the integrated CSII + CGM system and MDI + SMBG did show a significant difference in favour of the integrated CSII + CGM system. Similarly, the comparison between CSII + SMBG and MDI + SMBG showed a significant difference in favour of CSII + SMBG. These findings are summarised in Table 13.

FIGURE 5.

Network of studies comparing ‘HbA_1c levels < 7%’ at 6-month follow-up in adults.

TABLE 13 - Results of the indirect comparisons with regard to HbA1c levels of < 7% at 6-month follow-up in adults

RR values of > 1 indicate that the results favour the intervention listed in column 1. Differences are significant if the CIs do not include 1 (these are in bold).

Intervention	CSII + SMBG, RR (95% CI)	MDI + SMBG, RR (95% CI)
Integrated CSII + CGM	1.45 (0.74 to 2.84)	25.55 (1.58 to 413.59)
CSII + SMBG		17.56 (1.002 to 307.87)

Quality of life

Different tools were used to measure HRQoL (Figure 6). Only those studies using the same questionnaire could be combined in the analysis. Two studies reported results at 6-month follow-up for the Diabetic Treatment Satisfaction Questionnaire (Eurythmics37 and Bolli et al. 41) using a scale from 0 to 36, with higher scores indicating more satisfaction with treatment. These findings are summarised in Table 14. Two studies reported results for the HFS (Eurythmics37 and Thomas et al. 45); however, these could not be analysed together as one reported only the worry subscale of the HFS, whereas the other reported the total score.

FIGURE 6.

Network of studies comparing ‘quality of life’ at 6-month follow-up in adults. DQOL, Diabetes Quality of Life questionnaire; DTSQ, Diabetic Treatment Satisfaction Questionnaire; SF-36, Short Form questionnaire-36 items.

TABLE 14 - Results of the indirect comparisons with regard to quality of life (DTSQ) at 6-month follow-up in adults

DTSQ, Diabetic Treatment Satisfaction Questionnaire.

WMD values of > 0 indicate that the results favour the intervention listed in column 1. Differences are significant if the CIs do not include 0 (these are in bold).

Intervention	CSII + SMBG, WMD (95% CI)	MDI + SMBG, WMD (95% CI)
Integrated CSII + CGM	5.90 (2.22 to 9.58)	8.60 (6.28 to 10.92)
CSII + SMBG		2.70 (–0.16 to 5.56)

The results of these indirect comparisons show that the integrated CSII + CGM system significantly improved the quality-of-life scores at 6-month follow-up when compared with CSII + SMBG or with MDI + SMBG. There was no significant difference between CSII + SMBG and MDI + SMBG.

Effectiveness of interventions in children

We found five studies34,40,47–49 that reported data for children. In addition, there was one study (Ly et al. 33) that included a mixed population of patients between 4 and 50 years old. Approximately 70% of patients were children (< 18 years).

We asked our panel of four expert committee members whether or not they thought that the results of these studies could be pooled, especially whether or not the study by Ly et al. 33 (age range of 4 to 50 years, with 70% of participants < 18 years) could be included as if it was a study in children. One clinical expert agreed that the six studies were similar enough, as far as the differences in age ranges were concerned, to be pooled. A second clinical expert agreed that five of the studies were similar enough, as far as the differences in age ranges were concerned, to be pooled, but given that approximately one-third of participants were aged 18–50, it would be difficult to include the Ly et al. 33 study in the analysis of the interventions in children (if the adult group had been a younger cohort, e.g. 18–25 years, this expert’s conclusion may have been different). The third clinical expert also thought the Ly et al. 33 study could not reasonably be included in analyses for either group (children or adults); this third expert also thought that teenage children behave in a different way from pre-teen children and that, therefore, the 8- to 14-year-old cohort may be significantly different and should perhaps have been excluded from analyses. The fourth clinical expert did not respond.

However, the study by Ly et al. 33 was the only study looking at the MiniMed Veo system in children; therefore, we will present the results from analyses that included this study as if it was a study in children. In addition, the study by Weintrob et al. ,47 with children aged 8 to 14 years old, is the only study with results at 6-month follow-up linking MDI + SMBG to the MiniMed Veo system and the integrated CSII + CGM system; therefore, we included this study in the analyses as well. The results of these analyses should be interpreted with great caution because of the differences in age ranges among the included studies, as shown in Table 15.

Veo versus integrated CSII + CGM and CSII + SMBG

Results at 6-month follow-up: change in glycated haemoglobin levels

The results of the indirect comparison, shown in Figure 7 and Table 17, demonstrate that there were no significant differences between any of the interventions with regard to changes in HbA_1c levels at 6-month follow-up in children.

FIGURE 7.

Network of studies comparing change in HbA_1c levels at 6-month follow-up in children.

TABLE 17 - Results of the indirect comparison of changes in HbA1c levels at 6-month follow-up

WMD values of < 0 indicate that the results favour the interventions listed in column 1. Differences are significant if the CIs do not include 0.

Intervention	Integrated CSII + CGM, WMD (95% CI)	CSII + SMBG, WMD (95% CI)
Veo	0.38 (–0.16 to 0.92)	–0.04 (–0.26 to 0.18)
Integrated CSII + CGM		–0.42 (–0.92 to 0.08)

Effectiveness of interventions in pregnant women

We found one RCT50 that reported data for pregnant women (Table 20). The study included 32 pregnancies in 31 different women. The number of pregnancies was the unit of analysis. The study compared CSII + SMBG with MDI + SMBG; as these are not the relevant interventions described by NICE, the results will not be further discussed in this chapter. Full results are reported in Appendix 3.

Several non-RCTs (controlled clinical trials and observational studies) were identified; however, none of these looked at the MiniMed Veo system or an integrated CSII + CGM system. One ongoing study was identified; this is reported below (see Ongoing studies).

Additional analyses for the economic model

So far, we have adhered to the usual methods of meta-analyses, in accordance with which studies are combined in one analysis only if they compare similar interventions in similar populations at similar follow-up times, using similar outcomes.

We checked with clinical experts/committee members with regard to whether or not they agreed with these intended analyses and there was general agreement on the following points:

• Age Studies in children and adults should be analysed separately and studies in mixed age groups (adults and children), if data are not reported separately by age group, should not be included in analyses for children or adults.

• Follow-up Studies with results at 3-, 6- or 9-month follow-up should be analysed separately. Results from studies reporting outcomes at 2- to 4-month follow-ups can be pooled with results from studies reporting at 3-month follow-up; results from studies reporting at ≥ 9-month follow-up can be pooled in a ≥ 9-month follow-up group.

In cases in which the clinical experts disagreed with our suggested analyses, the clinical experts were always more cautious. For instance, it was suggested that Ly et al. 33 should not be treated as a study in children because one-third of participants were aged 18–50 years; therefore, it would be difficult to include this study with the analysis of children. If the adult age group in this study had been a younger cohort (e.g. 18–25 years) it may have been different. Similarly, teenage children were considered to behave in a different way from pre-teen children; therefore, the study by Weintrob et al. 47 (in which participants were aged 8 to 14 years) may be significantly different from the other studies in children (of up to 18 years) and perhaps should be excluded.

However, because of the lack of data, we have included the studies by Ly et al. 33 and Weintrob et al. 47 in the analyses for children. As a consequence, the results of these analyses are less reliable as a result of clinical heterogeneity between studies.

Despite trying to include as many studies as possible in the analyses for adults, we still have missing results for key comparisons for the economic model. Most importantly, results for comparisons of the MiniMed Veo system and the integrated CSII + CGM system with the comparators CSII + CGM, CSII + SMBG, MDI + CGM and MDI + SMBG are missing for the outcomes change in HbA_1c levels and severe hypoglycaemic event rates. As can be seen in Table 2, none of the included studies looked at CSII + CGM and MDI + CGM. Therefore, a comparison between these comparators cannot be made. However, it is possible to calculate results for these outcomes (change in HbA_1c and severe hypoglycaemic event rates) by comparing the MiniMed Veo system and the integrated CSII + CGM system with CSII + SMBG and with MDI + SMBG in a series of indirect comparisons, if we accept the following assumptions:

• All studies can be pooled, irrespective of length of follow-up (3, 6 or ≥ 9 months).

• Studies in mixed populations (including those of children and adults that do not report separate results by age group) can be pooled in one analysis. This means that we will include O’Connell et al. (30 adults and 32 children),35 RealTrend (81 adults and 51 children)36 and Hirsch et al. (98 adults and 40 children),34 in the analyses for adults. Ly et al. 33 (30 adults and 65 children) will still be excluded from these analyses.

• For event rates, we assumed that if numbers of events were reported, the rate could be derived by assuming that all patients had been observed for the follow-up duration of the trial.

It should be taken into account that the following analyses, including any subsequent analyses, such as the economic model, are based on these assumptions and that the clinical experts advised against using these wide inclusion criteria for pooling studies in one analysis. The results of these analyses are therefore likely to be considerably less reliable because of higher levels of clinical heterogeneity between studies included in these analyses for adults.

Change in glycated haemoglobin levels

The results of the indirect comparison, as shown in Figure 8 and Table 21, demonstrate that there were no significant differences with regard to the change in HbA_1c levels in adults (including mixed populations) between the MiniMed Veo system and the integrated CSII + CGM system. Similarly, there were no significant differences with regard to the change in HbA_1c levels in adults (including mixed populations) between the MiniMed Veo system and the integrated CSII + CGM system on the one hand and CSII + SMBG on the other. There was a significant difference in the change in HbA_1c levels in adults (including mixed populations) between the MiniMed Veo system and the integrated CSII + CGM system if both systems are compared with MDI + SMBG, favouring the MiniMed Veo system and the integrated CSII + CGM system over MDI + SMBG.

FIGURE 8.

Network of studies32,34–46 comparing change in HbA_1c levels at all follow-up times in adults and mixed populations. Green boxes represent the interventions; lines represent comparisons between interventions at different follow-up times (blue line, 3 months; green line, 6 months; black line, ≥ 9 months); and transparent boxes represent studies (blue, mixed population; black, adult population).

TABLE 21 - Results of the indirect comparison with regard to change in HbA1c levels at all follow-up times in adults and mixed populations

WMD values of < 0 indicate that the results favour the interventions listed in column 1. Statistically significant differences are those where the 95% CIs do not include 0 (shown in bold).

This result was from a random-effects analysis as I² was 62.5%.

This result was from a random-effects analysis as I² was 80.2%.

Intervention	Integrated CSII + CGM, WMD (95% CI)	CSII + SMBG, WMD (95% CI)	MDI + SMBG, WMD (95% CI)
Veo	0.04 (–0.07 to 0.15)	–0.07 (–0.31 to 0.17)	–0.66 (–1.05 to –0.27)
Integrated CSII + CGM		–0.11 (–0.32 to 0.10)	–0.70 (–1.05 to –0.30) a
CSII + SMBG			–0.46 (–1.18 to 0.27)b

Overall, integrated systems (the MiniMed Veo system and the integrated CSII + CGM system) are superior to SMBG (with CSII or MDIs) in terms of HbA_1c levels. However, as reported above, the reliability of the results of these analyses is reduced because of a relatively high level of heterogeneity between the studies included in the analyses. This is particularly true for the comparison between the MiniMed Veo system and CSII + SMBG, which is based not only on an indirect comparison (using data from the ASPIRE in-home trial,32 O’Connell et al. ,35 Hirsch et al. 34 and RealTrend36), but also on data from 3-month follow-up (ASPIRE in-home32 and O’Connell et al. 35) combined with data from 6-month follow-up (Hirsch et al. 34 and RealTrend36), and on data from adults (ASPIRE in-home32 and Hirsch et al. 34) and mixed populations (O’Connell et al. 35 and RealTrend36).

Severe hypoglycaemic event rate

The results of the indirect comparison, as shown in Figure 9 and Table 22, show that there were no significant differences in the severe hypoglycaemic event rate in adults (including mixed populations) between the MiniMed Veo system and any of the other treatments. Similarly, there were no significant differences in the change in severe hypoglycaemic event rate between the integrated CSII + CGM system and MDI + SMBG. There was a significant difference in the severe hypoglycaemic event rate between the integrated CSII + CGM system and CSII + SMBG, in favour of CSII + SMBG. However, as reported above, the reliability of the results of these analyses is reduced because of a relatively high level of heterogeneity between the studies included in the analyses. With regard to the significant difference in particular, it is important to point out that this result relies upon the data from three trials with different follow-up times (3 months for O’Connell et al. 35 and 6 months for Hirsch et al. 34 and RealTrend36), and that data from all three trials are from mixed populations, including adults and children.

FIGURE 9.

Network of studies32,34–46 comparing severe hypoglycaemic event rate at all follow-up times in adults and mixed populations. Green boxes represent the interventions; lines represent comparisons between interventions at different follow-up times (blue line, 3 months; green line, 6 months; black line, ≥ 9 months); transparent boxes represent studies (blue, mixed population; black, adult population; green, adults and all hypoglycaemic events).

TABLE 22 - Results of the indirect comparison for severe hypoglycaemic event rate at all follow-up times in adults and mixed populations

Rate ratio values of < 1 indicate that the results favour the intervention listed in column 1. Statistically significant differences are those where the 95% CIs do not include 1 (shown in bold).

Intervention	Integrated CSII + CGM, rate ratio (95% CI)	CSII + SMBG, rate ratio (95% CI)	MDI + SMBG, rate ratio (95% CI)
Veo	0.12 (0.01 to 2.14)	0.39 (0.02 to 8.40)	0.10 (0.01 to 1.93)
Integrated CSII + CGM		3.23 (1.10 to 9.49)	0.86 (0.51 to 1.46)
CSII + SMBG			0.67 (0.38 to 1.20)

Overall, the main conclusion regarding the evidence for hypoglycaemic event rate, and change in HbA_1c levels, in adults is that the evidence is limited and when all available evidence is combined, the results become highly unreliable.

Ongoing studies

We found 18 ongoing studies51–68 – 17 RCTs51–55,57–68 and one observational study56 looking at the use of a threshold suspend feature at home with a sensor-augmented insulin pump (SAP) (MiniMed 530G). Most ongoing studies are in children (12 out of 18 studies51,53,54,56,60–62,64–68), five are in a general population (adults or adults and children)52,55,57,59,63 and one study is in pregnant women. 58 Seven studies include the MiniMed Veo system51,52,54–56,59,64 and four studies include the integrated CSII + CGM system. 55,63,64,66 Details of ongoing studies are reported in Table 23.

Studies in adults

Twelve studies were included in the analyses for adults. 32,34,37–46 Only one of these studies (Hirsch et al. 34) reported the change in HbA_1c levels separately for adults. None of these studies looked at CSII or MDI + CGM. Table 5 shows an overview of these 12 studies, their comparisons and their baseline data. Further details are reported in Appendix 3.

Studies in children

Six studies were included in the analyses for children. 33,34,40,47–49 One of these studies (Hirsch et al. 34) reported only the change in HbA_1c levels separately for children. None of these studies looked at CSII or MDI + CGM. Table 15 shows an overview of these six studies, their comparisons and their baseline data. Further details are reported in Appendix 3.

Studies in pregnant women

We found one RCT that reported data for pregnant women. 57 The study included 32 pregnancies in 31 different pregnant women. The number of pregnancies was the unit of analysis. The study compared CSII + SMBG with MDI + SMBG; therefore, the results are not relevant for comparisons with the MiniMed Veo system or the integrated CSII + CGM system.

Search methods

Literature searches were undertaken to identify published economic evaluations of the MiniMed Paradigm Veo system and the Vibe and G4 PLATINUM CGM system. The search strategy for economic evaluations included a filter designed to identify cost and economic studies in databases that are not health economics specific.

The following databases and resources were searched for relevant economic evaluations and cost studies:

• NHS Economic Evaluation Database (Wiley Online Library): issue 3/July 2014

• Health Economic Evaluations Database (Wiley Online Library): up to 5 September 2014

• MEDLINE (via OvidSP): 1946–2014/August week 4

• MEDLINE In-Process Citations and Daily Update (via OvidSP): up to 5 September 2014

• PubMed (via National Library of Medicine): up to 5 September 2014

• EMBASE (via OvidSP): 1974–2014/week 34

• EconLit (EBSCOhost): 1969–1 August 2014

• Cost-effectiveness Analysis Registry (www.cearegistry.org): up to 5 September 2014

• Research Papers in Economics (http://repec.org/): up to 5 September 2014.

In addition, economic searches specifically for the MiniMed Paradigm Veo system, and Vibe and G4 PLATINUM CGM system were conducted using the same resources listed above.

The full search strategies are presented in Appendix 1.

Relevant studies were then identified in two stages. Titles and abstracts returned by the search strategy were examined independently by two researchers (Maiwenn Al and Isaac Corro Ramos) and screened for possible inclusion. Disagreements were resolved by discussion. Full texts of the identified studies were obtained. Two researchers (Maiwenn Al and Isaac Corro Ramos) examined these independently for inclusion or exclusion, and disagreements were resolved by discussion.

Inclusion criteria

The initial search identified a total of eight abstracts, six of which were of conference abstracts and were thus not included. Both of the full-text papers were identified as relevant to our review. These studies were by Kamble et al. 69 and Ly et al. 70 The study by Kamble et al. 69 evaluated integrated CSII + CGM versus MDI + SMBG in the USA, whereas the study by Ly et al. 70 evaluated the MiniMed Paradigm Veo system versus CSII + SMBG in Australia. The first evaluation69 showed that the integrated CSII + CGM system was not cost-effective compared with MDI + SMBG, despite taking all health effects into account through the IMS Centre for Outcomes Research and Effectiveness diabetes model (IMS CDM) version 8.5 (IMS Health, Danbury, CT, USA). On the other hand, the second study70 showed that the MiniMed Veo system was cost-effective compared with CSII + SMBG, if only the impact on the reduction of severe hypoglycaemic events was taken into account.

The characteristics of these studies are summarised in Table 24.

Of the six (excluded) conference abstracts, one was an abstract that was later published as a full-text paper71 and was already included as one of the two selected full-text papers. 69 While we will not formally discuss the conference abstracts,72–76 their characteristics, as far as they can be found in these abstracts, are presented in Table 25.

Quality assessment

A quality appraisal was carried out on the two studies,69,70 using the Drummond checklist. 77 A summary of the results are provided in Table 26.

Results

Model structure

The IMS CDM is an internet-based, interactive simulation model that predicts the long-term health outcomes and costs associated with the management of T1DM and T2DM. It is suitable for running cohort (bootstrap) and individual patient-level simulations. It was first developed by the Centre for Outcomes Research and Effectiveness and details of the first version were published by Palmer et al. in 2004. 79 It is widely used in diabetes cost-effectiveness research, both by health technology companies as well as those who pay for such technologies, and it has also been used in previous NICE technology assessments and clinical guidelines. 14,81–85 The model has been extensively validated. Since 1999, it has been examined at Mount Hood conferences, during which health economic models on diabetes are compared with each other in terms of their structure, performance and validity. 86–88 Two major validation papers on the IMS CDM have been published to date. 89,90 The latest one,90 from 2014, is the basis for the technical model description provided in this report. This description is consistent with the latest version of the model (version 8.5). Given the degree of validation of the model, and in order to be in line with the T1DM NICE guideline,81 it was deemed important not to use an alternative model or develop a de novo cost-effectiveness model for this evaluation.

The structure of the IMS CDM (from McEwan et al. 90) is shown in Figure 10. The IMS CDM comprises 17 interdependent submodels, which represent the most common diabetes-related complications: angina pectoris, myocardial infarction (MI), congestive heart failure (CHF), stroke, peripheral vascular disease (PVD), diabetic retinopathy, cataracts, hypoglycaemia, DKA, nephropathy, neuropathy, foot ulcer/amputation, macular oedema, lactic acidosis (T2DM only), (peripheral) oedema (T2DM only) and depression. A submodel for non-specific mortality is also included. Each of these submodels is a Markov model that includes different health states depicting the severity/stage of the complication. Transition probabilities in between the states of a complication submodel can be dependent on time, demographics, health state, physiological factors and diabetes type.

FIGURE 10.

IMS CDM model structure. Reprinted from Value in Health, 17/6, Phil McEwan, Volker Foos, James L. Palmer, Mark Lamotte, Adam Lloyd, David Grant, Validation of the IMS CORE Diabetes Model, Pages 714–724, Copyright (2014), with permission from Elsevier. 90 ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; CHF, congestive heart failure; CVD, cardiovascular disease; Inc., including; MI, myocardial infarction; PVD, peripheral vascular disease.

In addition, the non-parametric bootstrapping approach provides additional information on the uncertainty surrounding the long-term outcomes provided by the model. In this approach, a cohort population (with a size that can be defined by the model user) is created. Each patient in this population is unique in the sense of its baseline characteristics (demographics, existing baseline complications, baseline physiological risk factors and other risk factors, e.g. the number of cigarettes smoked per day). Within the bootstrapping simulation approach, two types of analysis are possible: deterministic and probabilistic. In the deterministic simulation, the continuous input parameters (baseline age, diabetes duration, HbA_1c levels, etc.) of each patient in the cohort that is created (e.g. 1000 patients) will be identical, but binary variables will differ (gender, presence of a diabetes-related complication, e.g. MI, etc.). In each iteration, one of the patients in this cohort is sampled with replacement and entered into the simulation (i.e. the complication submodels) until the patient dies. Applied treatment effects, utilities, costs and coefficients of cardiovascular disease (CVD) events will then be identical in each iteration. However, results will differ per iteration because of the differences in the binary input parameters in the baseline cohort and the way a patient progresses through the model (random walk). In the probabilistic simulation, all variables that are subjected to random sampling (i.e. cohort baseline parameters, treatment effects, coefficients of the CVD risk equations, health-state utilities/adverse event disutilities and costs) are randomly assigned at the beginning of the first iteration according to pre-defined probability distributions. Then all the patients in the cohort (e.g. 1000) are processed through the model while the parameters assigned at the start of the iteration are held constant. Those patients will only differ as a result of binary variables and random walk. When the model progresses to the next iteration, parameters are resampled again and the next 1000 patients are progressed though the model while parameters are held constant again. This process is repeated for all the bootstrap iterations.

However, it should be noted that because of computational time requirements, not all parameters in the model are subjected to random sampling. For instance, among the baseline risk factors, cigarette and alcohol consumption per day are not subjected to sampling. The same is true for minor and severe hypoglycaemia/ketoacidosis rates and coefficients from non-CVD-related risk adjustment equations.

Transition probabilities within each submodel (i.e. the annual probability of a change in health state) are dependent on baseline demographic and current physiological patient characteristics [HbA_1c levels, body mass index (BMI), etc.], and the existence of other complications and concomitant treatments (e.g. angiotensin-converting enzyme inhibitor, statin or laser). Transition probabilities are further calculated based on established regression or risk adjustment functions from the literature. 91–93 State transitions of a cohort occur simultaneously in each submodel. Therefore, it is possible that a patient will develop multiple complications in 1 year. In the IMS CDM model, diabetes-specific mortality is assumed to be caused by the following complications: MI, stroke, CHF, nephropathy, foot ulcer/amputation, hypoglycaemia, DKA and lactic acidosis. However, non-specific mortality is based on UK life tables. 94 Additional details on the submodels of the IMS CDM are given in Appendix 5.

An important limitation of the model is that it is not suitable for modelling long-term outcomes for children or adolescent populations, because the background risk adjustment/risk factor progression equations (such as those based on the Framingham studies)93,95–97 are all based on adult populations. Hence, we had to limit all our analyses to the adult population.

Baseline population characteristics

If possible, we estimated cohort baseline parameters based on the studies identified in our systematic review to properly reflect our base-case population (i.e. T1DM patients eligible for an insulin pump). In this case, only the study by Bergenstal et al. 32 provided reliable information for some patient characteristics. For the characteristics not reported in Bergenstal et al. ,32 we used those from the general T1DM population, as in the latest diabetes NICE guideline. 81 The cohort baseline characteristics used in our base-case analysis and their sources can be seen in Table 28. For the probabilistic sensitivity analysis (PSA) the input parameters age, duration of diabetes and baseline risk factors, for HbA_1c levels, systolic blood pressure (SBP), BMI, total cholesterol and low-density lipoproteins, are sampled from a normal distribution; the means and SDs are given in Table 28. Baseline triglyceride and high-density lipoprotein levels are sampled from a gamma distribution with the following parameters: alpha = mean²/SD² and beta = mean/SD².

Costs

The direct costs included in the model are for:

• management (for primary prevention of complications)

• diabetes-related complications

• the treatment of diabetes (this also includes the cost of the pump and/or glucose monitor)

• other hospital costs.

Indirect costs parameters were set to zero in the model as these were not included in our analyses, given the perspective of the NHS. Treatment costs were not included in the PSA because this was not possible using the current version of the IMS CDM, as the model developers argue that the uncertainty around the pharmacy/treatment administration costs is very small.

All other direct costs can be included in the PSA. Although cost parameters are typically sampled from different distributions independently in other economic evaluations, in the IMS CDM all direct costs are multiplied by the same positive factor which is sampled from a log-normal distribution with a mean of 1 and a user-defined coefficient of variation. In line with the latest diabetes NICE guideline,81 for our analyses we assumed a 20% deviation from the mean as it is assumed that this would represent a reasonable range of variation. Detailed descriptions of all four direct cost categories are given in the following sections.

Utilities

Health benefits were expressed in terms of life-years and QALYs gained. If more than one complication occurs at a time, a multiplicative approach is applied. 125 For the PSA, utility and disutility values are sampled from a beta distribution. Means and SDs are inputs for the IMS CDM; these are parameterized into parameters a and b of the beta distribution as follows: a = ((mean²) × (1 – mean)/(SD²)); and b = (mean × (1 – mean)/(SD²)) – ((mean²) × (1 – mean)/(SD²)). The utilities used in the model are summarised in Table 39.

Treatment effects

We used the reduction in HbA_1c baseline levels and the number of severe hypoglycaemic events as the outcomes to characterise treatment effectiveness. We considered using the number of minor hypoglycaemic and DKA events as well but not enough reliable data were found to make comparisons.

For HbA_1c levels, a baseline value had to be established onto which the treatment effect could be applied [i.e. the value at the start of treatment (time zero)]. The mean baseline value was 7.26% (standard error 0.71%), based on the relevant population, as shown in Table 28. Treatment effects were then estimated as the mean reduction from the baseline value, determined from our systematic review. An indirect meta-analysis was conducted to estimate the WMD between the MiniMed Paradigm Veo system and integrated CSII + CGM (used to inform the Vibe and G4 PLATINUM CGM system), CSII + CGM, CSII + SMBG, MDI + CGM and MDI + SMBG. Because of a lack of published clinical data, MDI + CGM had to be excluded from the analysis (see Figure 8) and treatment effects of integrated CSII + CGM and non-integrated CSII + CGM were assumed to be identical (see Figure 8 and Table 21). After calculating the change in HbA_1c levels from baseline in Bergenstal et al. 32 as –0.02, the change in HbA_1c levels for other treatments could be found. These values are listed in Table 40.

Since there is uncertainty and there are limitations in the indirect meta-analysis (because of heterogeneity and differences in baseline HbA_1c levels), to explore the impact of different HbA_1c change levels, we analysed a hypothetical situation in which the baseline HbA_1c levels do not change after the initiation of treatment in a separate scenario. It should be noted that, in the IMS CDM, the change in HbA_1c level is assumed to occur within the first 12 months. After this, an annual progression rate is applied. For the base case we followed the approach in NICE Guideline NG17,81 in which an annual progression of 0.045% (derived from the DCCT)92 was used.

For severe hypoglycaemic events, it is not necessary to set a baseline value since the IMS CDM assumes that this is a treatment-specific parameter. Treatment effects were estimated as the rate ratio of event rates per 100 patient-years obtained from our systematic review (see Figure 9 and Table 22). This was then applied to a reference value for integrated CSII + CGM, which was derived from a weighted average (by sample size) of the event rates observed in the CSII + CGM arms of the trials. These values are shown in Table 41.

For the PSA, treatment effects on HbA_1c levels at baseline are sampled from a beta distribution (mean and SD are converted into beta distribution-specific parameters, as explained in Utilities). The event rates of severe hypoglycaemic events are fixed in the IMS CDM and therefore they are not included in the PSA. In order to explore the uncertainty of the effects of severe hypoglycaemic episodes on long-term outcomes, several scenarios with different treatment-specific rates were analysed (see Treatment effects part II: severe hypoglycaemic event rates).

Disease management parameters

These parameters will determine the proportion of patients that will receive disease management regimens, such as preventative treatments or screening programmes. These parameters and their sources are shown in Table 42. With the exception of the proportion on the UK-specific foot ulcer prevention programme, for which we followed the approach in NICE Guideline NG17,81 the majority of the inputs are the default values from the IMS CDM and were also used in the latest diabetes NICE guideline. 81

Disease natural history parameters

These are the parameters that will determine the natural course of the disease. These parameters are either transition probabilities, that is the probability of each of the events (e.g. diabetic retinopathy or MI) or the (relative) risk of an event, given a particular risk factor; risk factors are based on physiological measures, such as HbA_1c levels, BMI, SBP or characteristics like the presence of microalbuminuria. We considered the same values as in NICE Guideline NG17,81 most of which were the same as the IMS CDM default values. For that reason, and because the number of parameters is so large that it may distract the reader’s attention, we have decided to show these parameters in Appendix 6.

It should be noted that one of these parameters is the probability of death from a severe hypoglycaemic event. In line with NICE Guideline NG17,81 this was assumed to be zero for the base case. However, as deaths due to severe hypoglycaemic events have been reported,138,139 we expect that this parameter may have an impact on our results, as one of the key features of the MiniMed Paradigm Veo is the LGS function, which was shown to reduce the number of severe hypoglycaemic events, and thus the number of deaths caused by severe hypoglycaemia. Therefore, other options for this mortality rate were explored in additional scenarios.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis was used to explore the impact of statistical uncertainties regarding the model’s input parameters. PSA is an in-built feature of the IMS CDM, activated if the second order with sampling option is selected.

Probabilistic sensitivity analysis results were presented in the cost-effectiveness plane for all the treatments compared. Cost-effectiveness acceptability curves (CEACs) were used to describe the probability of a treatment being considered cost-effective given a threshold incremental cost-effectiveness ratio (ICER). The probability distributions used in the PSA are described throughout the Model input parameters section.

Scenario analyses

Scenario analyses were performed to explore the impact on costs and QALYs of using different assumptions on the baseline population characteristics, on the number of blood tests (finger prick tests) conducted per day, on the amount of insulin used, on the inclusion of HbA_1c progression after year 1, on treatment effects (both in terms of HbA_1c level change and in terms of the number of severe hypoglycaemic episodes per treatment), on the inclusion of a non-zero probability of death as a result of hypoglycaemia, on time horizon, on QALY estimation methods, on utility benefits associated with less fear of hypoglycaemia, and on the cost of the stand-alone insulin pump and CGM devices.

Base-case results

The base-case results from the full incremental analysis reported as cost per QALY gained (ICER) per technology for adult T1DM patients are summarised in Table 47.

First, it should be noted that since the same treatment effects were assumed for stand-alone and integrated CSII + CGM, the latter is dominated by the former (i.e. effectiveness is the same for the integrated as for the stand-alone technology, but the integrated technology is more expensive, as shown in Table 38). As expected, MDI + SMBG is the cheapest treatment but also the one that provides the lowest number of QALYs. The ICER of CSII + SMBG compared with MDI + SMBG is £52,381. MiniMed Paradigm Veo is extendedly dominated by stand-alone CSII + CGM. Essentially, this means that, in a full incremental analysis, where all the interventions and comparators are considered, CSII + CGM is better value for money than MiniMed Veo. This is because, from our systematic review, the decrease in HbA_1c levels with respect to baseline was highest for stand-alone CSII + CGM; this decrease in HbA_1c levels leads to a decrease in the number of complications that occur over a lifetime to such an extent that it compensates for the higher number of hypoglycaemic events. In any case, the ICER of stand-alone CSII + CGM compared with CSII + SMBG is £660,376. Thus, given the common threshold ICER of £30,000, it is clear that stand-alone CSII + CGM is not cost-effective.

Alternatively, we present the base-case ICERs for the two interventions against every comparator in Table 48. Integrated CSII + CGM (Vibe) is dominated by stand-alone CSII + CGM. It should be noted that when the MiniMed Veo system is compared with stand-alone CSII + CGM, the ICER obtained is high (£422,849) but that this results from both negative incremental QALYs and incremental costs (i.e. the ICER is in the south-west quadrant of the cost-effectiveness plane). In this case, the cost savings outweigh the loss in QALYs and therefore the MiniMed Veo system is more cost-effective than stand-alone CSII + CGM. This might not be immediately apparent when looking at the full incremental results in Table 47 because, in this table, the MiniMed Veo system is in position of extended dominance. The lowest ICERs are obtained when the interventions are compared with MDI + SMBG, but these are above £100,000 in the north-east quadrant of the cost-effectiveness plane. When the interventions are compared with CSII + SMBG, the highest ICERs are obtained (around £700,000 in the north-east quadrant of the cost-effectiveness plane). Thus, given the common threshold ICER of £30,000, the interventions are not cost-effective.

In the deterministic simulation, the cost-effectiveness results are very similar except that, in this simulation, MiniMed Veo is not extendedly dominated by stand-alone CSII + CGM. These results are shown in Table 24. Although overall cost and QALY estimates are higher than in the probabilistic simulation, the ICERs and the main conclusions from the simulation presented in Table 49 are similar to the conclusions drawn from the simulation presented in Table 47.

The base-case ICERs for the two interventions compared with every comparator in the deterministic simulation are shown in Table 50. These results are similar to those presented in Table 48 and so are the conclusions drawn.

When we looked at the breakdown of the total costs, we observed that treatment costs always represent the largest proportion of the total costs, independently of the treatment chosen. In Figure 11, the treatment costs constitute 79% of the total direct costs for the MiniMed Paradigm Veo system, and integrated and stand-alone CSII + CGM. For CSII + SMBG, treatment costs represent 66% of the total costs and for MDI + SMBG this is 41%. For each treatment, the foot ulcer/amputation/neuropathy cost category is the second largest, and eye diseases and renal diseases are the third and fourth largest cost categories, respectively. MDI + SMBG has higher complication incidences (CVD, ulcer, eye disease, etc.), whereas for the other four treatments these complication incidences are similar. Lifetime hypoglycaemic events were least reported for the MiniMed Paradigm Veo system (0.622 severe hypoglycaemic events per patient), and were most reported for MDI + SMBG (5.412 severe hypoglycaemic events per patient).

FIGURE 11.

Breakdown of costs per treatment: (a) breakdown of MiniMed Veo system costs; (b) breakdown of integrated CSII + CGM costs; (c) breakdown of stand-alone CSII + CGM costs; (d) breakdown of CSII + SMBG costs; and (e) breakdown of MDI + SMBG costs.

Results of the probabilistic sensitivity analyses

Statistical uncertainties in the model were investigated in the PSA. Since we compared five treatments simultaneously, the scatterplot of the PSA outcomes in the cost-effectiveness plane was not very informative (Figure 12). Nevertheless, we can observe a clear positive correlation between costs and QALYs and that the treatments including CGM are similarly scattered, showing that they are more expensive but also provide more QALYs.

FIGURE 12.

Cost-effectiveness plane with PSA outcomes for all treatments in T1DM patients.

The CEACs for each treatment are shown in Figure 13. These CEACs confirm that only the treatments including SMBG are considered cost-effective. At ceiling ratio values of < £52,381, MDI + SMBG was the treatment with the highest probability of being cost-effective. When that threshold is exceeded, then CSII + SMBG was the treatment with the highest probability of being cost-effective. It should be noted that, for all three treatments including CGM, the cost-effectiveness probability was zero for all the ceiling ratios considered in the analysis. This was expected as the difference in costs between CGM treatments and SMBG treatments was too large to outweigh the additional QALYs gained by using CGM.

FIGURE 13.

Cost-effectiveness acceptability curves for all treatments in T1DM patients.

Multiple daily insulin injection-unsuitable subgroup

As mentioned in Chapter 1 (see Comparators), insulin pumps are recommended for people with T1DM for whom MDIs are not suitable. Therefore, we questioned the extent to which insulin pumps (especially modern pumps such as the integrated systems) and MDIs are used in similar populations. This seemed a reasonable question in light of the lack of studies found by our systematic review that compared these two treatments. If MDI + SMBG is not considered in the analysis, the ICERs from the full incremental analysis were the same as those reported in Table 47, but excluding the first row. It appears that CSII + SMBG is the strategy most likely to be cost-effective, independent of the ceiling ratio value (up to £100,000 per QALY), as shown in Figure 14.

FIGURE 14.

Cost-effectiveness acceptability curves for all non-MDI treatments in T1DM patients.

Continuous glucose monitoring-indicated/self-monitoring of blood glucose-unsuitable subgroup

In the analysis for the CGM-indicated/SMBG-unsuitable subgroup, we excluded SMBG-based treatment options from the analysis on the presumption that the most relevant population comprises those who find it difficult to perform SMBG often or adequately enough. In this situation, integrated CSII + CGM (Vibe) is dominated by stand-alone CSII + CGM, as shown in Table 47 and the only relevant comparison is the MiniMed Veo system with stand-alone CSII + CGM. The ICER is £422,849 (in the south-west quadrant of the cost-effectiveness plane), as shown in Table 48. The corresponding CEACs are shown in Figure 15. These CEACs indicate that the MiniMed Veo system is the CGM treatment most likely to be cost-effective for all the ceiling ratios considered in the analysis. However, as the ceiling ratio increases, the CEACs for the MiniMed Paradigm Veo system and stand-alone CSII + CGM seem to converge. As expected, the CEAC for integrated CSII + CGM was always zero for all the ceiling ratios considered in the analysis, since this was dominated by the stand-alone combination of CSII and CGM.

FIGURE 15.

Cost-effectiveness acceptability curves for CGM treatments in T1DM patients.

Results of scenario analyses

In the scenarios presented below, only the ICERs from the full incremental analysis are discussed. The ICERs for the two interventions against every comparator are shown in Appendix 7.

Treatment effects part I: change in glycated haemoglobin levels in the first year

In this scenario analysis, we assumed that the baseline HbA_1c value is stabilised for 1 year and that it does not change in any of the treatments (i.e. 0% change in HbA_1c levels in the first year). The model results for this scenario are shown in Table 54.

TABLE 54 - Cost-effectiveness results when no treatment effect (in terms of change in HbA1c levels) is assumed in the first year (for all technologies)

Intervention	QALYs	Cost (£)	Incremental QALY	Incremental cost (£)	ICER (£)
CSII + CGM	12.0006	146,632	Dominated by MDI + SMBG
Integrated CSII + CGM (Vibe)	12.0006	147,304	Dominated by MDI + SMBG
MDI + SMBG	12.0016	56,928	–	–	–
CSII + SMBG	12.0160	90,178	0.0144	33,250	2,309,028
MiniMed Veo system	12.0260	138,538	0.0099	48,360	4,871,356

The QALY expectations for all treatments are very similar. The minor differences in QALYs can be explained by the differences in severe hypoglycaemic episode rates. It should be noted that although the rate of severe hypoglycaemic events for MDI + SMBG was estimated to be higher than the rate for integrated CSII + CGM (see Treatment effects), MDI + SMBG resulted in a slightly higher gain in QALYs which could be due to randomness. CSII + CGM systems were dominated by MDI + SMBG. Furthermore, CSII + SMBG and the MiniMed Veo system are on the efficient frontier but with extremely high ICER values. As can be seen in the resulting CEACs in Figure 16, MDI + SMBG was the most cost-effective treatment for all the values of the ceiling ratio considered in the analysis.

FIGURE 16.

Cost-effectiveness acceptability curves for all treatments when there is no HbA_1c treatment effect.

Non-zero probability of death caused by severe hypoglycaemia

In this scenario, we assumed a mortality due to severe hypoglycaemia of 4.9%, as derived from Ben-Ami et al. 142 The model results are shown in Table 56.

TABLE 56 - Cost-effectiveness results for the mortality due to severe hypoglycaemia scenario (all technologies)

Intervention	QALYs	Cost (£)	Incremental QALY	Incremental cost (£)	ICER (£)
MDI + SMBG	11.1041	58,510	–	–	–
CSII + CGM	11.7701	142,215	Dominated by CSII + SMBG
Integrated CSII + CGM (Vibe)	11.7701	142,872	Dominated by CSII + SMBG
CSII + SMBG	11.8781	89,475	0.774	30,965	40,006
MiniMed Veo system	12.0071	137,801	0.129	8326	374,531

In this scenario, both integrated and stand-alone CSII + CGM were dominated by CSII + SMBG. The ICER of CSII + SMBG compared with MDI + SMBG was £40,006, and the ICER of MiniMed Veo compared with CSII + SMBG was £374,531. Thus, these treatments are not cost-effective given the common threshold ICER of £30,000. Both cost-effectiveness plane scatterplots and CEACs are similar to those for the base-case scenario and therefore they are not shown here. If only the CGM treatments were considered, the probability of the MiniMed Paradigm Veo system being cost-effective was equal to 1 for almost all the values of the ceiling ratio considered in the analysis; this is shown in Figure 17.

FIGURE 17.

Cost-effectiveness acceptability curves for only CGM treatments for the non-zero mortality due to severe hypoglycaemia scenario.

Different time horizon

In this scenario, we assumed a 4-year time horizon, which corresponds to the average lifetime of an insulin pump. These results are shown in Table 58.

TABLE 58 - The 4-year time horizon scenario (all technologies)

Intervention	QALYs	Cost (£)	Incremental QALY	Incremental cost (£)	ICER (£)
MDI + SMBG	2.7718	6706	–	–	–
CSII + CGM	2.7882	24,803	Dominated by CSII + SMBG
Integrated CSII + CGM (Vibe)	2.7886	24,939	Dominated by CSII + SMBG
CSII + SMBG	2.7906	13,365	0.0188	6659	354,202
MiniMed Paradigm Veo	2.7928	23,144	0.0022	9778	4,461,063

We observed that both stand-alone and integrated CSII + CGM are dominated by CSII + SMBG. Although the MiniMed Paradigm Veo system is the treatment with the highest number of QALYs gained, its high cost when compared with CSII + SMBG does not outweigh this gain in QALYs, and results in an ICER of £4,461,063. Therefore, for this scenario also, it is very unlikely that MiniMed Paradigm Veo will be deemed cost-effective, as illustrated by the corresponding CEACs in Figure 18.

FIGURE 18.

Cost-effectiveness acceptability curves for all treatments for the 4-year time horizon scenario.

If only the CGM treatments are considered, the MiniMed Paradigm Veo system is clearly the treatment with the highest probability of being cost-effective, as shown in Figure 19.

FIGURE 19.

Cost-effectiveness acceptability curves for CGM treatments only: 4-year time horizon scenario.

Fear of hypoglycaemia unawareness

Table 59 shows the results obtained when the utility increment (0.0329) from Kamble et al. 69 was used to represent the reduced fear of hypoglycaemia. We applied this utility increment throughout the remaining lifetimes of patients using integrated devices (the MiniMed Paradigm Veo system and integrated CSII + CGM). This benefit was not applied to non-integrated devices (stand-alone CSII + CGM, CSII + SMBG and MDI + SMBG), as these non-integrated devices do not give a warning or activate/stop the release of insulin automatically in response to low BG levels.

TABLE 59 - Cost-effectiveness results for the fear of hypoglycaemia scenario (all technologies)

Intervention	QALYs	Cost (£)	Incremental QALY	Incremental cost (£)	ICER (£)
MDI + SMBG	11.4146	61,050	–	–	–
CSII + SMBG	11.9756	90,436	0.5610	259,386	52,381
CSII + CGM	12.0604	146,476	Extendedly dominated by MiniMed Veo system
MiniMed Veo system	12.6224	138,357	0.6468	47,920	74,088
Integrated CSII + CGM (Vibe)	12.6429	147,150	0.0205	8792	428,595

For this scenario, the main difference with respect to the base-case scenario is that stand-alone CSII and stand-alone CGM devices is extendedly dominated by the MiniMed Paradigm Veo system, which has an ICER compared with CSII + SMBG of £74,088. Moreover, in this scenario, integrated CSII + CGM is not dominated by the corresponding stand-alone combination, as the utility increment for the integrated system led to a larger number of QALYs accumulated than the non-integrated options. Nevertheless, the ICER of integrated CSII + CGM compared with the MiniMed Paradigm Veo system is still very large (£428,595).

The scatterplot of the PSA outcomes in the CE plane is very similar to the one in the base-case scenario and therefore we decided not to show it here. The CEACs for each treatment are shown in Figure 20. These CEACs demonstrate that, compared with the base-case scenario, the probability of being cost-effective for CSII + SMBG starts decreasing at approximately £60,000. As the ceiling ratio increases, the probability of being cost-effective for the MiniMed Paradigm Veo system and integrated CSII + CGM systems also increases. At ceiling ratio values larger than (approximately) £75,000, the MiniMed Paradigm Veo system is the treatment with the highest probability of being cost-effective, followed by integrated CSII + CGM systems, at ceiling ratio values of more than (approximately) £90,000. It should be noted that for stand-alone CSII + CGM, the cost-effectiveness probability was zero for all of the ceiling ratios considered in the analysis.

FIGURE 20.

Cost-effectiveness acceptability curves for reduced fear of hypoglycaemia scenario.

If only the CGM treatments were considered, we observed similar CEACs (Figure 21) to those observed for the base case (see Figure 14), but in this scenario the role of integrated and stand-alone CSII + CGM was interchanged in the CEAC.

FIGURE 21.

Cost-effectiveness acceptability curves for only CGM treatments for the fear of hypoglycaemia scenario.

Cost of stand-alone insulin pumps and continuous glucose monitoring devices

In this scenario, we assumed that the yearly cost of stand-alone CSII + CGM could be estimated from the average costs of the different stand-alone devices, as shown in Tables 32 and 33, but without the weighting for market share from White et al. 121,122 Therefore, in this scenario, the estimated yearly cost of stand-alone CSII + CGM was £5460. The results from this scenario are shown in Table 60.

TABLE 60 - Cost-effectiveness results for cost of stand-alone CSII + CGM without market share scenario (all technologies)

Intervention	QALYs	Cost (£)	Incremental QALY	Incremental cost (£)	ICER (£)
MDI + SMBG	11.4146	61,050	–	–	–
CSII + SMBG	11.9756	92,272	0.5610	31,222	55,654
MiniMed Veo system	12.0412	138,357	Extendedly dominated by integrated CSII + CGM
Integrated CSII + CGM (Vibe)	12.0604	147,150	0.0849	54,878	646,692
CSII + CGM	12.0604	150,063	Dominated by integrated CSII + CGM

The main difference in these results, with respect to the base-case scenario, was that, as expected, stand-alone CSII + CGM was more expensive than integrated CSII + CGM (Vibe). Since both technologies are assumed to have the same efficacy, integrated CSII + CGM (Vibe) dominated stand-alone CSII + CGM. The CEACs for each treatment are shown in Figure 22. These are very similar to those for the base-case scenario. The higher cost of stand-alone CSII + CGM had almost no impact on the cost-effectiveness probability since MDI + SMBG and CSII + SMBG are the only strategies that are considered cost-effective.

FIGURE 22.

Cost-effectiveness acceptability curves for cost of stand-alone CSII and stand-alone CGM devices CGM without market share scenario.

If only the CGM treatments were considered, we observed similar CEACs (Figure 23) as those observed for the base-case scenario (see Figure 15) but, as expected, in this scenario the role of integrated CSII + CGM (Vibe) and stand-alone CSII + CGM was interchanged in the CEAC.

FIGURE 23.

Cost-effectiveness acceptability curves for CGM treatments only for the cost of stand-alone CSII and stand-alone CGM devices without market share scenario.

Parameters subject to extreme uncertainty in the clinical effectiveness evidence for child and adolescent patients

These are all parameters for treatment effects on both HbA_1c levels and hypoglycaemic event rates for all six treatment options (i.e. essentially 12 different parameters).

For the MiniMed Veo system, our systematic review identified only one study in children: Ly et al. 33 This study included patients between 4 and 50 years old, 70% of whom were children (4–18 years old). However, data were not reported separately by age group; therefore, we could use only the data for the total population and assume that it would apply to children.

However, our clinical experts advised us not to use this study as a study in children for two main reasons: (1) children behave differently to adults and, therefore, results for children are not the same as those for adults; and (2) pre-teen children behave differently from teenagers and, therefore, the 4- to 12-year-old age group would be different from a 12- to 18-year-old age group and the influence of parents on younger children would have to be taken into account. Indeed, this further subdivision of children essentially implies a doubling of the number of parameters for which there is no evidence of any treatment effect.

The only reason that we presented the data from this study in Chapter 3 (see Assessment of clinical effectiveness) is that, without it, there would have been no evidence at all with regard to the effectiveness of the MiniMed Veo system in children. Therefore, for the MiniMed Veo system (and the assessment of severe hypoglycaemic events), we have data from only one study and this does not properly apply to children.

In addition, we found two trials presenting evidence for the integrated CSII + CGM system versus CSII + SMBG34 and versus MDI + SMBG,40 and three trials comparing CSII + SMBG with MDI + SMBG. 47–49

However, these studies differed with regard to the age groups included (12–17 years,34 7–18 years,40 8–14 years,47 8–18 years48 and 8–21 years49), whether or not patients had pump experience, baseline HbA_1c levels (8–11.5%), follow-up times (3, 6 and 12 months), hypoglycaemic status at baseline (in one study, patients with hypoglycaemia unawareness were excluded;40 in another study, only patients with impaired awareness of hypoglycaemia were included;33 and other studies had no exclusions34 or no information47–49), and country (Israel,47 the USA,34,48,49 and the USA and Canada40). None of the studies was performed in the UK. Therefore, there is considerable heterogeneity between the studies, which makes any pooling of results invalid.

Uncertainties around the parameters for disease progression and treatment within the IMS CDM for child and adolescent patients

Several additional modelling uncertainties with regard to using the IMS CDM for children and adolescents have been identified. Indeed, the CDM structure is limited in that it lacks crucial parameters to inform the long-term effects of hypoglycaemia. These uncertainties have been summarised in Table 61, along with those regarding the treatment effects on HbA_1c levels and hypoglycaemic event rate.

Health economic analyses of type 1 diabetes for children and adolescent patients in other National Institute for Health and Care Excellence guidelines/assessment reports

Clinical effectiveness

Nineteen trials were included, 12 reported data for adults,32,34,37–46 six reported data for children33,34,40,47–49 and one trial reported data for pregnant women. 50 Four trials were in mixed populations (adults and children); two of these reported data separately for adults and children and are included in both the 12 trials for adults and the six trials for children. 34,40 Two trials did not report data separately for adults and children (O’Connell et al. 35 and RealTrend36). Therefore, the results from these trials were not used in the main analyses.

Twelve studies were included in the analyses for adults. 32,34,37–46 The main conclusion from these trials is that the MiniMed Paradigm Veo system reduces hypoglycaemic events in adults in comparison with the integrated CSII + CGM system, without any differences in other outcomes, including change in HbA_1c levels. Nocturnal hypoglycemic events occurred 31.8% less frequently in the MiniMed Veo group than in the integrated CSII + CGM group [1.5 events (SD 1.0 event) vs. 2.2 events (SD 1.3 events) per patient per week; p < 0.001]. Similarly, the MiniMed Veo group had significantly lower weekly rates of combined daytime and night-time events than the integrated CSII + CGM group (p < 0.001). Indirect evidence seems to suggest that that there are no significant differences between the MiniMed Paradigm Veo system and CSII + SMBG or MDI + SMBG with regard to the change in HbA_1c levels at 3-month follow-up. However, if all studies are combined (combining different follow-up times and including mixed populations), the MiniMed Paradigm Veo system is significantly better than MDI + SMBG in terms of HbA_1c levels.

For the integrated CSII + CGM system (MiniMed Paradigm REAL-Time 722 System) versus other treatments, the results suggest a significant effect in favour of the integrated CSII + CGM system over MDI + SMBG for HbA_1c levels, but not if compared with CSII + SMBG, and a significant effect in favour of the integrated CSII + CGM system over MDI + SMBG and CSII + SMBG with regard to quality of life.

With regard to comparisons of CSII and MDIs, only one of the six trials41–46 showed a significant difference between CSII + SMBG and MDI + SMBG in terms of change in HbA_1c levels. DeVries et al. 42 found a significant difference in favour of CSII + CGM: at 16 weeks, mean HbA_1c levels were 0.84% lower (mean = –0.84%, 95% CI –1.31% to –0.36%) in the CSII + SMBG group than the MDI + SMBG group. No differences were found in any trial with regard to the number of severe hypoglycaemic events.

Six studies were included in the analyses for children. 33,34,40,47–49 None of the studies in children made a direct comparison between the MiniMed Paradigm Veo system and the integrated CSII + CGM system. An indirect comparison was possible, using data from Ly et al. 33 and Hirsch et al. 34 at 6-month follow-up, but only for HbA_1c levels, which showed no significant difference between groups.

One study33 compared the MiniMed Paradigm Veo system with CSII + SMBG. The only significant difference between treatment groups was the rate of moderate and severe hypoglycaemic events, which favoured the MiniMed Paradigm Veo system.

One study34 compared the integrated CSII + CGM system with CSII + SMBG; this trial found no significant difference in HbA_1c levels between groups. One study40 compared the integrated CSII + CGM system with MDI + SMBG; this trial found a significant difference in HbA_1c change scores in favour of the integrated CSII + CGM system, but no significant difference in the number of children achieving HbA_1c levels of ≤ 7%. The hyperglycaemic AUC was significantly lower in the integrated CSII + CGM group, but the hypoglycaemic AUC showed no significant difference. Other outcomes showed no significant differences between groups.

For pregnant women, we found only one trial50 comparing CSII + SMBG with MDI + SMBG, which is not relevant to the decision problem.

Cost-effectiveness

We assessed the cost-effectiveness of the MiniMed Paradigm Veo system and the Vibe and G4 PLATINUM CGM system compared with stand-alone CSII + CGM, CSII + SMBG, MDI + CGM and MDI + SMBG for the management of T1DM in adults.

In addition to the literature limitations regarding the population subgroups of interest (i.e. children and pregnant women) mentioned above, the model employed to conduct the cost-effectiveness analyses, the IMS CDM, is not suitable for modelling long-term outcomes for child/adolescent or pregnant woman populations, because all of the background risk adjustment/risk factor progression equations are based on adult populations.

The comparator MDI + CGM was not included in the cost-effectiveness analyses as no relevant evidence for this comparator was found in the systematic review. Moreover, in the absence of data comparing the clinical effectiveness of integrated CSII + CGM systems with stand-alone CSII + CGM systems, we assumed, in our analyses, that both technologies would be equally effective, which seems to be plausible. The immediate consequence of this assumption was that stand-alone CSII + CGM systems dominated the integrated CSII + CGM systems since the stand-alone system was cheaper, according to our estimated cost, while being equally effective.

Overall, the cost-effectiveness results suggest that the technologies using SMBG (either with CSII or MDIs) are more likely to be cost-effective, since the higher quality of life provided by the technologies that use CGM does not outweigh their higher costs. This is in line with the findings in the currently updated T1DM guideline,81 in which CGM was compared with several SMBG setups and was found not to be cost-effective. In particular, the base-case results show that MDI + SMBG is the cheapest treatment, but also the one that provides the lowest number of QALYs. The ICER of CSII + SMBG compared with MDI + SMBG is £52,381. The MiniMed Paradigm Veo system is extendedly dominated by stand-alone CSII + CGM. This is mainly because, according to our systematic review, the decrease in HbA_1c levels with respect to baseline was highest for integrated CSII + CGM, and this decrease in HbA_1c leads to a decrease in the number of complications that occur over a lifetime to such an extent that it compensates for the higher number of severe hypoglycaemic events. In any case, the ICER of stand-alone CSII + CGM compared with CSII + SMBG was £660,376. Thus, given the common threshold ICER of £30,000, it is clear that stand-alone CSII + CGM would not be cost-effective.

We also considered two additional base-case analyses. Since insulin pumps are recommended for people with T1DM for whom MDIs are not suitable, we excluded MDI-containing technologies from the analysis. In this scenario, the CSII + SMBG appeared to be the strategy most likely to be cost-effective, with a cost-effectiveness probability equal to almost 1 for all of the ceiling ratios considered in the analysis. Following this, we also excluded SMBG treatments from the analysis in order to capture the effect of the LGS function of the MiniMed Paradigm Veo system, which is expected to have an influence on reducing the number of severe hypoglycaemic events, and thus on the number of QALYs gained. In this situation, the only relevant comparison was the MiniMed Veo system versus stand-alone CSII + CGM, since the Vibe and G4 PLATINUM CGM system was dominated by the stand-alone combination of CSII and CGM. The corresponding results showed that when the MiniMed Veo system was compared with stand-alone CSII + CGM, the ICER obtained was high (£422,849). However, this results from both negative incremental QALYs and incremental costs (i.e. the ICER is in the south-west quadrant of the cost-effectiveness plane). In this case, the higher the ICER, the better (i.e. any cost saving could be used on other patients in order to generate QALYs that could ‘outweigh’ the loss in QALYs). Therefore, at a ceiling ratio of £30,000 per QALY, the MiniMed Veo system would be more cost-effective than stand-alone CSII + CGM. This is demonstrated by the corresponding CEACs, since the MiniMed Paradigm Veo system is the CGM treatment most likely to be cost-effective for all of the ceiling ratios considered in the analysis. However, as the ceiling ratio increases, the CEACs for the MiniMed Paradigm Veo system and stand-alone CSII + CGM seem to converge. As expected, the PSA showed that, for the Vibe and G4 PLATINUM CGM system, the probability of this system being cost-effective is always zero for all of the ceiling ratios considered in the analysis.

The results of these different scenario analyses did not differ much from the base-case results. The scenario that was most favourable with regard to the MiniMed Paradigm Veo system was the one that considered an additional utility decrement associated with the fear of hypoglycaemia. In this scenario, the ICER of the MiniMed Paradigm Veo system compared with CSII + SMBG was equal to £74,088 (the lowest found in all analyses). However, given the common threshold ICER of £30,000, the MiniMed Paradigm Veo system would not be considered cost-effective. For the Vibe and G4 PLATINUM CGM system, when it was not (extendedly) dominated by other strategies, the lowest ICER obtained was £428,595 when compared with the MiniMed Paradigm Veo system. This was also the case for the scenario in which a utility increment associated with reducing the fear of hypoglycaemia was considered.

Clinical effectiveness

Overall, the evidence seems to suggest that the MiniMed Paradigm Veo system reduces hypoglycaemic events in comparison with other treatments, without any differences in other outcomes, including change in HbA_1c levels. In addition, we found significant results in favour of the integrated CSII + CGM system in comparison with MDI + SMBG with regard to HbA_1c levels and quality of life. However, the evidence base was poor. The quality of included studies was generally low and often there was only one study that compared treatments in a specific population and at a specific follow-up time. In particular, the evidence for the two interventions of interest was limited, with only one study comparing the MiniMed Paradigm Veo system with an integrated CSII + CGM system,32 and only one study, in a mixed population, comparing the MiniMed Paradigm Veo system with CSII + SMBG. 33 In addition, although several studies included the integrated CSII + CGM system as a treatment arm, it is important to note that none of these studies looked at the Vibe and G4 PLATINUM CGM system; in the included studies, the integrated CSII + CGM system was always a MiniMed Paradigm pump with integrated CGM system (MiniMed Paradigm REAL-Time 722 System). This also means that all of the studies that assessed the effectiveness of the integrated CSII + CGM system were performed in the USA. Overall, only 337,41,45 out of the 19 included studies included patients from the UK, and only one of these was completely performed in the UK (Thomas et al. ). 45 Interactions between patients and health-care providers may show considerable differences in different countries, which will affect patients’ behaviour and therefore the effectiveness of insulin pumps and monitors. Therefore, the results from the included studies may not be representative of the UK situation.

Unfortunately, many studies had to be excluded because they compared CSII with MDIs, without specifying the type of monitoring, or CGM with SMBG, without specifying the type of insulin delivery. Two studies149,150 with 2 × 2 factorial design, including CSII + CGM, CSII + SMBG, MDI + CGM and MDI + SMBG, had to be excluded because the results were reported for only CSII versus MDIs and CGM versus SMBG. One of these studies was in children (Mauras et al. 149) and one was in adults [Little et al. (HypoCOMPaSS trial. 150)] These studies were excluded because they could not be classified as one of the relevant comparators defined by NICE and they could not be compared with the MiniMed Paradigm Veo system or an integrated CSII + CGM system.

In addition, we had problems differentiating stand-alone and integrated CSII + CGM interventions because the interventions were often poorly described, making it difficult to be sure which type of intervention was used. Sometimes researchers indicated no differences between these two types of treatments and provided patients in the same treatment arm with stand-alone and integrated CSII + CGM systems (see Beck151).

Four of the included studies were in mixed populations (Ly et al. 33 used 65 children and 30 adults with an age range of 4–50 years; O’Connell et al. 35 used 32 children and 30 adults with an age range of 13–40 years; RealTrend36 used 51 children and 81 adults with an age range of 2–65 years; and Hirsch et al. 34 used 40 children and 98 adults with an age range of 12–72 years). The advice from clinical experts was not to combine results from adults and children and vice versa. Therefore, these studies were, in the first instance, excluded from our analyses. Only if results were reported separately for adults and children were results included in the analyses or if there would have been no data without using a mixed adult/child population study, as in the case of Ly et al. ,33 which was used as a study in children to make a comparison between the MiniMed Paradigm Veo system and other treatments.

As reported in Chapter 1 of this report (see Comparators), there is a problem with the comparability of populations in studies evaluating insulin pumps and MDIs. NICE recommended CSII as a potential treatment for children ≥ 12 years and adults, who have disabling hypoglycaemia (including anxiety about hypoglycaemia) when trying to attain HbA_1c < 7.5%, or HbA_1c is constantly > 8.5%, while undergoing MDIT. 14 In other words, insulin pumps are recommended for people with T1DM for whom MDIs are not suitable. Therefore, it was anticipated that it would be problematic finding studies comparing insulin pumps (especially modern pumps such as the integrated systems) with MDIs in similar populations.

Most studies comparing CSII with MDIs show no difference with regard to HbA_1c levels. One trial found a significant difference in the change in HbA_1c levels at follow-up (DeVries et al. 42). In this trial, patients with persistent poor control, defined as a mean of all HbA_1c levels of ≥ 8.5% in the 6 months before the trial, were included. Partly based on this trial, NICE14 concluded that CSII would most likely be cost-effective in patients with poorly controlled diabetes. Our current systematic review shows that nothing has changed in the evidence base with regard to CSII versus MDIs. The trial by DeVries et al. 42 is still the only trial showing significant differences in HbA_1c levels at follow-up between CSII + SMBG and MDI + SMBG. This highlights the problem with identifying the correct population for comparisons between the interventions relevant to this appraisal. For the comparison of the MiniMed Paradigm Veo system with the integrated CSII + CGM system, we have included a general population of T1DM patients. However, if we compare these interventions with CSII + SMBG or MDI + SMBG in general populations, we will obscure the differences that exist between CSII and MDIs in diabetes patients with poor control at baseline.

For the comparison of CSII with MDIs, it is important to differentiate between populations with good HbA_1c control at baseline and populations with poor control. However, if we compare the MiniMed Paradigm Veo system with the integrated CSII + CGM system and with CSII + SMBG, all patients will be using a pump and, in most studies comparing different types of pumps, patients will have been using a pump for > 6 months. In such studies, baseline HbA_1c levels will be relatively low because of long-term pump use. Therefore, it is difficult to assess how valid comparisons are between those patients and patients involved in trials comparing pump use with MDIs.

Given these problems resulting from the heterogeneity among RCT populations, we did not consider including any further observational studies, as these problems would be even more apparent if results from observational studies were compared.

For pregnant women, we found no studies looking at the MiniMed Paradigm Veo system or the integrated CSII + CGM system.

Cost-effectiveness

An important strength of the current cost-effectiveness evaluation is that we used a well-validated diabetes model (IMS CDM) that has been used for many assessments, including submissions for NICE. 14,81–85 In particular, this model was used to assess the cost-effectiveness of CSII versus MDIs for T1DM patients in a 2010 HTA report80 and in the current update of the NICE Guideline on T1DM (NG17). 81 Since 1999, the model has been used at Mount Hood conferences, during which health economic models on diabetes are compared with each other in terms of their structure, performance and validity. 86–88 Two major validation papers on the IMS CDM have been published to date. 89,90 The latest one,90 from 2014, is the basis for the technical model description provided in this report. This description is consistent with the latest version of the model (version 8.5). Given the degree of validation of the model, and in order to be in line with the currently updated T1DM guideline81 from which we sourced many input parameters, it was deemed important not to use an alternative model or develop a de novo cost-effectiveness model for this evaluation. The most recent unit cost data were obtained for the analyses, including detailed data on equipment costs obtained from the relevant companies.

Although many of our input parameters are the same as those described in NICE Guideline NG17,81 we have also considered interventions that were not assessed in the guideline. Furthermore, we have considered a large variety of scenarios and performed PSAs for all of them.

A major limitation of the model is that the IMS CDM is not appropriate for analysing health economic outcomes for paediatric/adolescent populations. This was reported in the 2010 assessment of CSII versus MDIs for T1DM patients80 and confirmed by the model developers, who also mentioned that the model is not appropriate for pregnant women either. Therefore, these two subgroup populations were not included in the cost-effectiveness analyses.

Another limitation of the IMS CDM is that not all input parameters can be included in a PSA because of the technical constraints of the model. It is likely that the most important parameter not included in the PSA was the rate of severe hypoglycaemic events, as this is considered to be one of the key drivers of the model results, especially with regard to the MiniMed Paradigm Veo system. As a consequence, the uncertainty regarding the ICERs is currently somewhat underestimated. However, the ICERs themselves are not influenced by this limitation.

Another major limitation is the lack of comparability of treatments and clinical trials to estimate the treatment effect for stand-alone CSII + CGM. In the current analysis, we had to assume equal effectiveness of integrated and stand-alone CSII + CGM, thus assuring that stand-alone CSII + CGM would always dominate integrated CSII + CGM. Moreover, it was difficult to determine the extent to which the effect of the LGS function of the MiniMed Paradigm Veo system was captured in the model results. Furthermore, we found no reliable data on minor hypoglycaemic events and DKA events. The impact of these parameters on the cost-effectiveness results is difficult to predict, but we expect them to have less of an impact than the other treatment effect parameters (e.g. reduction in HbA_1c levels and rate of severe hypoglycaemic events).

Finally, information was limited for the estimation of the cost of the stand-alone insulin pump. Although we do not expect a large difference in our estimated costs, it may have a major implication for the comparison of stand-alone CSII + CGM versus the integrated Vibe and G4 PLATINUM CGM system, as both are equally effective. Thus, depending on the price, one of these two options will dominate the other.

Clinical effectiveness

The main uncertainties with regard to clinical effectiveness are the general lack of data (especially for children and pregnant women) and the poor quality of the available data. In addition, there were problems with differentiating interventions (in particular integrated and stand-alone CSII + CGM systems) and with interpreting results from mixed populations (adults and children).

Because of inherent differences in patient characteristics at baseline, it was difficult to compare MDI + SMBG with any of the other interventions in this assessment.

Cost-effectiveness

The uncertainties described for clinical effectiveness also apply to the assessment of cost-effectiveness. In addition, it is uncertain how realistic it is to assume a continuous increase in HbA_1c levels over the first year of treatment. It seems likely that, in clinical practice, efforts would be made to keep HbA_1c levels as low as possible, so periods of increase may be followed by decreases. It is unclear at this moment what the most realistic scenario will be in the long term.

EMBASE (via OvidSP)

Date range searched: 1974–2014/week 34.

Date searched: 5 September 2014.

MEDLINE (via OvidSP)

Date range searched: 1946–2014/August week 4.

Date searched: 5 September 2014.