A randomised controlled trial to compare minimally invasive glucose monitoring devices to conventional monitoring in the management of insulin-treated diabetes mellitus (MITRE)

The GlucoWatch^® G2™ Biographer (Animas Corporation, West Chester, PA, USA)

The GlucoWatch (Figure 1) is slightly larger than a watch and can be worn on any part of the body, although the forearm is favoured. The device consists of two parts: (1) a reusable portion, which contains the microprocessor, electronics and output display, and (2) the disposable portion or autosensor that comes into contact with and adheres to skin. The sensor consists of two electrodes and two hydrogel discs that contain glucose oxidase. The sensor extracts fluid electro-osmotically through the skin.

FIGURE 1.

GlucoWatch G2 Biographer (Animas Corporation).

The GlucoWatch requires a 2-hour warm-up period followed by a single capillary glucose estimation for calibration. Following the warm-up period the device may be worn for up to 13 hours, providing the patient with up to 78 estimates over that period. A measurement is made every 10 minutes and the device takes the average of the last two recordings to provide real-time glucose values to the patient every 10 minutes. The usable accurate range is reported to be between 2.2 and 22.0 mmol/l. Over 8500 recordings can be stored in the memory. This information can be downloaded to a personal computer to provide information for the patient and health-care professional on profiles and trends in glucose. Thus, patients can obtain feedback on their glucose profiles without attendance at a clinic.

The device can be programmed to provide audible warnings should the glucose level rise above or fall below preset values; however, the device does not work properly if the skin has excess perspiration or has rapidly changed in temperature, as these will confound the recording. The optimum frequency of use has yet to be determined.

Most studies on the GlucoWatch have focused on the correlation between the GlucoWatch and capillary blood glucose values. Two early studies have shown this to be acceptable with an r-value of 0.85–0.90. 39,40 GlucoWatch readings lag behind blood glucose concentrations by approximately 20 minutes. It is important to note that the GlucoWatch manufacturers stopped selling this device from the end of July 2007.

The Continuous Glucose Monitoring System^® (MiniMed, Northridge, CA, USA)

The MiniMed Continuous Glucose Monitoring System (CGMS) (Figures 2 and 3) is a halter-style device similar in size to a radio pager. The device is worn on the waist and connected via a wire to a subcutaneous sensor. The sensor is a small flexible device containing glucose oxidase that harvests interstitial fluid. This is inserted into the abdominal wall using a rigid introducer and then secured to skin. The patient then wears the device for up to 72 hours. The CGMS does not provide real-time blood glucose readings. Calibration requires the patient to record at least four capillary blood glucose values daily and enter the values into the device. The device samples every 10 seconds and records an average glucose estimation every 5 minutes, i.e. 288 recordings are made in a 24-hour period. The usable accurate range is reported to be between 2.2 and 22.0 mmol/l. A total of 2 weeks of results can be stored in the device.

FIGURE 2.

MiniMed CGMS.

FIGURE 3.

MiniMed CGMS.

The patient is asked to perform frequent capillary glucose testing, if possible, in addition to the recordings required for basic calibration, and to record this information on the device. At the end of the 72-hour period the patient then attends the diabetes unit for downloading of the results to form a glucose profile. This can then be reviewed with a health-care professional and adjustments to treatment made as appropriate. The device may be refitted at intervals to review the change in trends. The optimal frequency of use of the device has not yet been established, although it is suggested that it should be used 4–6 weekly during initial treatment changes and 4–6 monthly during review.

Studies have confirmed that the measurements of interstitial fluid glucose by the device reflect plasma glucose levels across a broad range. 41,42

GlucoDay^® (A. Menarini, Florence, Italy)

The GlucoDay measures glucose through a microdialysis probe, which is inserted into the abdominal wall. A portable unit is wrapped around the wearer’s abdomen. Glucose levels are measured every 3 minutes for 48 hours and only one calibration is required at 48 hours. The device can provide real-time or retrospective readings, depending on how the monitor is set up. Results can be read continuously through an infrared communicating port and downloaded to a personal computer; thus, individual glucose profiles can be observed over a 24-hour continuous monitoring period. A series of alarms are built into the system and can alert the wearer to take appropriate action. 43

Guardian^® Real-Time Continuous Glucose Monitoring System (MiniMed)

The Guardian is the successor to the CGMS and is based on exactly the same technology; however, unlike the CGMS it provides real-time data and hypo- and hyperglycaemic alarms that will sound outside a preset range.

FreeStyle Navigator^® (Abbott Diabetes Care, Alameda, CA, USA)

This uses an enzyme-tipped subcutaneous sensor attached to a transmitter, which sends data to a receiver that can be located up to 10 feet away. The device records glucose readings every minute and presents real-time data. The receiver presents data in various formats including a trend function in which arrows indicate the immediate glucose trend (horizontal: no change; slightly up: increase; strongly up: rapid increase; slightly down: decrease; strongly down: rapid decrease). These data can be downloaded to a computer for analysis. 44

STS™ System (DexCom, San Diego, CA, USA)

The DexCom STS System (3 day and 7 day) uses a disposable sensor placed just below the skin in the abdomen to measure the level of glucose in the fluid found in the body’s tissues (interstitial fluid). Sensor placement causes minimal discomfort and can easily be carried out by the patient. The sensor must be replaced weekly. An alarm can be programmed to sound if a patient’s glucose level reaches preset lows or highs.

It is important to note that none of these devices are intended to provide an alternative to traditional SMBG. FDA labelling states that they should serve as an adjunct to SMBG, supplying additional information on glucose trends that is not available from traditional monitoring.

Summary and critique of available literature evaluating the clinical effectiveness and acceptability of continuous glucose monitoring devices

A descriptive review of the available literature was conducted to assess the evidence of clinical effectiveness, user acceptability and psychological impact of continuous glucose monitoring devices. A summary of this review is presented here.

Meta-analysis was not considered as an appropriate method for summarising the studies identified through the literature searches because of variation in the measurement of the outcomes and in the delivery and content of the interventions.

The MEDLINE (1966–5/2008), EMBASE (1980–5/2008), CINAHL (Cumulative Index to Nursing and Allied Health Literature; 1982–5/2008) and PsycINFO (1972–5/2008) databases were searched using the following search terms: continuous glucose monitoring.

Articles were also identified from the reference lists of studies identified through database searches.

Phase 1

At the baseline clinic visit the research nurse trained and provided participants with the OneTouch Ultra monitor and the GlucoWatch. Participants were asked to use the GlucoWatch at times of their choice but with a minimum attempted use of four times per month and a maximum attempted use of four times per week. In addition, they were told to continue to perform capillary blood glucose monitoring as desired. One calibration finger prick test was required each time they used the GlucoWatch. They were also advised to check capillary glucose if the GlucoWatch sounded a high or low alarm. It was explained to participants that the GlucoWatch must not be relied upon for estimating insulin requirements. During this period, participants were reviewed by the research nurse at 4, 8 and 12 weeks from baseline, at which point the results from both the GlucoWatch and the OneTouch Ultra meter were downloaded, saved and printed. These results were used as the basis for adjustment of treatment regimes when delivering feedback.

Phase 2

Participants were asked to continue using the GlucoWatch as often as they wished but were recommended to use the device at least twice per week. They were reviewed by the research nurse at 6, 12 and 18 months from baseline.

Phase 1

At the baseline clinic visit, the research nurse trained and provided participants with the OneTouch Ultra monitor and the CGMS. The CGMS was fitted by the research nurse and participants were requested to wear it for 72 hours. In addition to wearing the CGMS, participants were asked to continue to perform capillary blood glucose monitoring as desired. They returned to the clinic 72 hours later for the device to be removed or in some cases they removed the device themselves and returned to the clinic as soon as possible but no more than 1 week after device removal. On the return visits, glucose readings from both the CGMS and the OneTouch Ultra were downloaded, saved and printed. These results were reviewed by the research nurse and used to provide feedback for and adjustment to therapy. Participants were also fitted with the device at 6 and 12 weeks from baseline and received nurse feedback sessions 72 hours later.

Phase 2

Participants were fitted with the CGMS and received nurse feedback sessions at 6, 12 and 18 months. During the ‘fitting’ visits, discussion concerning diabetes control was avoided and participants were told that this would be discussed in detail at the visit when the results were downloaded. At each fitting, participants were requested to wear the CGMS for 72 hours. If the device failed within 24 hours of fitting then participants were encouraged to return to the clinic for a refitting. If the device failed more than 24 hours after fitting then the available data were reviewed in the nurse feedback session.

Phase 1

At the baseline clinic visit, the research nurse trained and provided participants with the OneTouch Ultra monitor. Participants were asked to monitor capillary blood glucose at their normal frequency for 3 months and to attend nurse feedback sessions at 4, 8 and 12 weeks. At these feedback sessions the results from the OneTouch Ultra meter were downloaded and used to give feedback.

Phase 2

Participants were asked to continue using the OneTouch Ultra meter at their normal frequency. They were reviewed by the research nurse and provided with feedback on their test results at 6, 12 and 18 months. Throughout phase two, the research nurse was available via the telephone/email to discuss any problems.

Phases 1 and 2

At the baseline clinic visit, the research nurse trained and provided participants with a OneTouch Ultra meter. They were asked to monitor capillary blood glucose at their normal frequency. Participants received standard treatment, which typically consisted of 6-monthly clinic visits and access to diabetes advice when required. At subsequent research visits at 6, 12 and 18 months, no feedback was given by the research nurse.

Inclusion criteria

1. Individuals with insulin-treated diabetes mellitus receiving two or more injections daily [including continuous subcutaneous insulin infusion (CSII) pump users].

2. Age over 18 years.

3. Duration of diabetes over 6 months.

4. Fluent in English, Bengali, Cantonese or Turkish.

5. HbA1c results:

   1. Two HbA1c levels greater than or equal to 7.5%, one in the last 3 months and another within the previous 15 months. Research nurses followed the normal consent procedure for participants fulfilling this criterion.

   2. Individuals with one HbA1c level greater than or equal to 7.5% in the last 3 months and either a second HbA1c level greater than or equal to 7.5% over 15 months previously or no other HbA1c levels greater than or equal to 7.5% were invited to have a screening blood test carried out 3 months later. If that was greater than or equal to 7.5% and the participant consented to the study then this was used as the baseline HbA1c. Research nurses then followed the consent procedure for individuals requiring a screening blood test.

   3. If a participant had an HbA1c level greater than or equal to 7.5% but this had been measured more than 3 months previously then they were invited to have a screening blood test carried out as soon as possible. If that was greater than or equal to 7.5% and the participant consented to the study then this was used as the baseline HbA1c. Research nurses then followed the consent procedure for individuals requiring a screening blood test.

   4. In all cases, the two HbA1c results had to be a minimum of 12 weeks apart. At the outset of the study the inclusion criterion for HbA1c was two consecutive HbA1cs greater than 8.0% (one obtained at screening and one at the last assessment but within 12 months). In light of advances in diabetes management, changes in clinical targets and difficulties experienced in recruitment, the HbA1c inclusion criterion was subsequently changed and the protocol amended (as described above).

6. Willingness to comply with the consent and trial procedure.

Exclusion criteria

1. Previous inability to use a capillary glucose meter.

2. Previous use of the GlucoWatch or CGMS sensor.

3. Presence of abnormal haemoglobin (presence of elevated levels of HbF or HbS).

4. Pregnancy, or planned pregnancy in the next 18 months.

5. Skin conditions, e.g. eczema, psoriasis or other skin irritation, at the sites of monitor use.

6. Receiving dialysis.

7. Visual or physical impairment limiting ability to use monitors.

8. Planned major surgery (e.g. coronary artery bypass graft, hip replacement) within 3 months of consent.

9. Participation in any other ongoing trial.

Participants who spoke English, Bengali, Cantonese or Turkish were included to ensure that individuals from different ethnic backgrounds were evaluated. When participants were non-English speakers, the nurse feedback sessions were held with the assistance of an appropriately trained translator. All of the questionnaires were translated.

Trial site location

Participants were enrolled from four sites:

• Royal Bournemouth Hospital

• Queen Elizabeth Hospital/Bensham Hospital, Gateshead

• University College London Hospitals (UCLH)

• Whittington Hospital, London.

The sites selected were chosen to improve the diversity of the sample population. The two London sites represent inner-city locations, Gateshead an urban and socioeconomically deprived area and Bournemouth a relatively affluent area with a high proportion of retired people.

Identification of participants

People with diabetes were identified from three sources:

• local diabetes databases

• posters advertising the trial in the waiting rooms of the different sites

• review of clinic notes.

Potential participants were identified primarily by the research nurses at each site assisted by the local investigators. Those identified as potentially eligible were given an information sheet and invited to discuss the trial in more detail with the research nurse or the local investigator. Invitations to participate were issued in person, by telephone contact or through the approved invitation letter.

Consent procedure (see Appendices 3–5)

Eligible participants were provided with a full explanation of each arm of the trial including the potential problems with use of the devices (see Appendix 2). It was clearly explained that participants would have a one in four chance of being in any arm of the trial. Participants were asked to provide verbal agreement that they would not use a non-invasive or minimally invasive blood glucose monitor independently of the trial, regardless of which arm of the trial they were randomised to. Participants were informed that if they were allocated to the attention control or standard treatment arms and if the final results indicated either of the monitors to be beneficial then they would be given priority for use of the GlucoWatch or CGMS on trial completion. Following explanation of the trial, a period of at least 24 hours but no more than 4 weeks had to elapse before written consent was obtained and subsequent randomisation carried out. The exception to this consent procedure was individuals fulfilling inclusion criterion 5(b). In this case participants completed a screening consent form and returned for an HbA1c test 3 months later.

Glycaemic control

Percentage change in HbA1c from baseline to 18 months was the primary outcome in this study. Three blood samples were taken at each assessment by the research nurse. One was analysed locally, one was sent to the Department of Diabetes and Endocrinology at UCLH for analysis and standardisation, and the third sample was retained and stored locally in case of damage or loss to the standardised sample.

Perceived acceptability of the devices

At the outset of the study no suitable measure had been developed that would have been able to adequately assess the acceptability of the minimally invasive blood glucose monitors under evaluation. Hence, a questionnaire was developed for this purpose (see Appendix 7). Details about the process of developing this questionnaire measure are provided in Appendix 8. The number of times that people chose to wear the devices also provided an indicator of acceptability.

Clinical assessments

1. Change in HbA1c (baseline to 3, 6 and 12 months). Percentage change in HbA1c at the end of the intensive phase of the trial (3 months’ follow-up) was measured to assess short-term efficacy. Percentage change in HbA1c was also measured from baseline to 6 and 12 months to assess efficacy in the medium term.

2. Hypoglycaemic episodes (defined as blood sugar ≤ 3.5 mmol/l). Hypoglycaemic episodes, the time and date that they occurred and how they were detected were recorded by the DRN. To collect these data, participants were asked to keep diaries. When possible the incidence of hypoglycaemia was confirmed by the OneTouch Ultra meter and recorded as such in the case report form (CRF). In the case of groups 1 and 2, downloaded data from the GlucoWatch and CGMS were also used. A hypoglycaemic episode was recorded if (a) blood glucose was ≤ 3.5 mmol/l for > 20 minutes (i.e. two or more readings for the GlucoWatch, four or more readings for the CGMS), (b) blood glucose of ≤ 3.5 mmol/l was followed by one or more skipped readings followed by a reading of ≤ 3.5 mmol/l for the GlucoWatch or (c) blood glucose of ≤ 3.5 mmol/l was followed by two or more skipped readings coded PRSP (perspiration) on the GlucoWatch. Awareness of hypoglycaemia was assessed through completion of the Edinburgh Hypoglycaemia Symptoms Scale and the Hypoglycaemia Symptoms Awareness Questionnaire. 107

3. Hyperglycaemic episodes (defined as blood sugar ≥ 10.0 mmol/l). The percentage of finger prick blood glucose values ≥ 10.0 mmol/l were recorded in the CRF by the DRN. These data were drawn from participant diaries and the OneTouch Ultra meters. In the case of groups 1 and 2, downloaded data from the GlucoWatch and CGMS were also used and recorded as the number of glucose readings ≥ 10 mmol/l for > 20 minutes (two or more readings for the GlucoWatch, four or more readings for the CGMS).

4. Skin reactions (GlucoWatch and CGMS groups). During nurse feedback sessions the DRN recorded the extent of any skin irritation for each application of the monitor. These data were drawn from participants’ ratings and recordings in their diaries using the MITRE Skin Scale (see Appendix 6). Data were recorded in the CRF by the DRN at baseline and at each feedback session. Adverse device-related event forms were completed if reactions were rated as severe (i.e. ≥ 6 on the MITRE Skin Scale). If a participant reported a score on the MITRE Skin Scale of ≥  6 between research visits then they were instructed to attend the diabetes clinic for review by the DRN. The DRN reviewed and photographed the site and when appropriate considered the individual for withdrawal from treatment.

5. Side effects. Any side effects reported by participants from either the minimally invasive glucose monitors or the standard monitors were recorded in the CRF by the DRN. Participants were also asked about side effects on the self-reported acceptability questionnaire that was developed for the purpose of this study (see Appendix 7). This questionnaire included asking participants to rate the acceptability of any side effects experienced.

Selection and development of psychological measures

Primary end point: glycaemic control at 18 months’ follow-up

Percentage change in HbA1c from baseline was measured, looking for a clinically important difference of 12.5%. This proportional change took into account the baseline HbA1c of the participants and equates to an absolute drop of 1% in participants entering the study with an HbA1c of 8%, or an absolute drop of 1.1% in participants entering the study with an HbA1c of 9%. Based on clinical data, a 5% mean change from baseline was expected in the standard care control group, with a standard deviation of 15.5%. Allowing for a 10% attrition rate, 100 participants per arm were calculated to provide 90% power to detect a 12.5% reduction in HbA1c from baseline at the 5% significance level. A total of 400 participants were therefore required from the four participating centres.

In the original protocol the sample size target was 600 based upon the proportion of patients allocated achieving an absolute reduction of 1% in HbA1c from baseline to 18 months’ follow-up. However, in light of advances in diabetes management and changes in clinical targets combined with difficulties in recruitment, the protocol was amended so that the power calculations were based upon identifying a 12.5% reduction from baseline instead.

Primary end point: glycaemic control

All analyses were conducted on an intention to treat basis, comparing each of the device groups with the standard care and attention control groups. The trial was not powered to make a direct comparison between the GlucoWatch and CGMS groups. The primary outcome, the percentage change in HbA1c from baseline, was calculated looking for a mean relative reduction in HbA1c of 12.5% from baseline to 18 months, for example a baseline HbA1c of 10% decreasing to 8.75% at 18 months’ follow-up.

When available, HbA1c results from routine clinic/GP appointments were recorded if participants missed study visits or if participants who had withdrawn from all other aspects of the study gave their consent. All of the HbA1c results analysed in the study were those obtained from the local laboratories at each of the four centres. All four laboratories used standardised (DCCT-aligned) methods for measurement of HbA1c. Each individual’s HbA1c results across the duration of the study period were analysed at the same local laboratory. HbA1c results were included in the analysis if they fell within a prespecified period of time around the research visit:

• 3 months: up to 4 weeks before and 4 weeks after the date of the scheduled visit

• 6 months: up to 8 weeks before and 8 weeks after the date of the scheduled visit

• 12 and 18 months: up to 12 weeks before and 12 weeks after the date of the scheduled visit.

Primary end point: acceptability

Comparison of the GlucoWatch and CGMS groups to assess whether the two non-invasive devices differed in levels of acceptability was by parametric/non-parametric tests depending on the distribution of the data.

Secondary end point: glycaemic control

The same analyses of percentage change in HbA1c from baseline to 18 months were carried out as on the data for 3, 6 and 12 months described above. Analyses of variance (ANOVA) were performed to examine the relative reduction in HbA1c from baseline at each of the follow-up periods (follow-up HbA1c –  baseline HbA1c/baseline HbA1c ×  100).

Secondary end point: psychosocial outcomes

Groups were compared on the psychosocial data (quality of life, diabetes self-care activities, treatment satisfaction, etc.) using repeated measures, ANOVA, when the data were normally distributed and using non-parametric analyses when distributions were skewed.

Cost-effectiveness analysis methods

Economic evaluation of health-care treatments combines measures of outcome with measures of opportunity cost, to answer the question of whether reallocating resources to a programme would result in a more efficient allocation of resources. The most commonly followed practice in economic evaluation is cost-effectiveness analysis in which the aim is to maximise outcomes given the constrained resources in the NHS. In cost-effectiveness analysis both the costs and consequences of an intervention are considered simultaneously against other relevant comparators (e.g. best alternative care). The comparative nature of these evaluations is key as it is not possible to establish cost-effectiveness without formal comparison with other ways of using these resources. 117

Quality-adjusted life-years

Quality-adjusted life-years (QALYs) are a generic (non-disease-specific) measure of health outcome that simultaneously captures morbidity [health-related quality of life (HRQoL) gains] and mortality (survival duration gains) and combines the two into a single measure. QALYs are generated by the summation across all health states of the length of time in a particular health state multiplied by a weight representing the HRQoL (utility value) attached to that health state. The utility values are based on a scale in which 1 represents full health and 0 represents death. The utility values of the participants in the MITRE trial were measured using the EQ-5D (European Quality of Life – 5 dimensions) questionnaire. 118 The EQ-5D is a standardised instrument for the measurement of health outcome and will be discussed further later in this report.

Decision rules

Let A and B represent two alternative treatments. If intervention A is less costly and more effective, it is said to ‘dominate’ B. Similarly, if A is more costly and less effective, it would be dominated by B. Under either of these conditions it is easy to conclude that the dominant option is the more cost-effective. In practice it is rare that the cost and outcomes lend themselves to the dominance rule, and it is usually the case that an intervention is more effective but also more costly. The critical issue here is whether the additional (incremental) cost is worth paying for the incremental benefits. The decision rules developed to address this issue focus on the incremental cost-effectiveness ratio (ICER), which is defined as:

At this point the decision about whether an intervention is considered cost-effective hinges on the cost-effectiveness threshold, which is considered to represent a reasonable ‘willingness to pay’ for an additional QALY. The threshold considered to be appropriate by the National Institute for Health and Clinical Excellence (NICE) is between £20,000 and £30,000. 80 If the ICER of the intervention is lower than this threshold then the intervention can be viewed as a cost-effective use of NHS resources.

The decision rules of cost-effectiveness analysis can be extended to deal with multiple treatment comparisons. Further discussions of such extensions and the related net benefit framework can be found in Drummond et al. 117

Overview of economic analysis

The aim of the economic analysis is to compare the costs and consequences (in terms of utilities) of the four trial arms of the MITRE trial. These include the GlucoWatch, CGMS, attention control and standard care control arms. The costs to be considered are those faced by the NHS in terms of health service resource use. The unit costs/prices used are for 2005–6. The consequences to be considered are those to the treated patients, which will be measured in terms of utilities using the EQ-5D questionnaire. Costs and consequences should be measured or extrapolated over the time that they could be expected to differ between treatment arms.

Data collection methods and frequency

Unit costs

Unit costs at 2005–6 prices were used to value the resource use measured in the trial when available. These were average costs. It was assumed that the cost of a nurse’s time for a patient interview or feedback session is the same as that for an appointment with a nurse. As only the type of medication and not the brand was specified in the CRFs, it was assumed that the participants were prescribed the most commonly prescribed brand (as given by Prescription cost analysis: England 2004119). Although there may have been some change in the most commonly prescribed drug brands, the earliest date of randomisation in the trial was 20 May 2003 and so the use of Prescription cost analysis: England 2004 provides a reasonably accurate indication of the medications that patients were taking.

Resource costs

Multiplication of resource use by the unit costs gives the resource costs. The costs presented represent the costs over the 3 months preceding each nurse feedback session (or patient interview for the standard care arm). As the data collected for medication were not based on a 3-month period but instead on what patients were taking either at the time of the visit or in the previous week, it has been assumed that the patients were on the same medications recorded for the 3 months preceding each nurse feedback session (or patient interview for those in the standard care arm).

As a consequence of problems with missing data (which are discussed further in the section on statistical methods), the resource costs are presented at a higher level of aggregation than the resource use data (e.g. the total diabetic clinic cost per period is presented, not by type of visit). The resource cost components that are presented are shown in Table 2.

The device costs for the GlucoWatch arm were based on the total number of sensor boxes (a box contains 16 sensors) that would be required given the number of sensors used, for example an individual using 17 sensors would be charged for two boxes of sensors. As an individual might use one box of sensors over several trial periods, the analysis has taken the total number of sensors used during the trial, converted it to the number of boxes needed and then split the costs of the boxes equally over the trial period.

As the interventions investigated in the MITRE trial were primarily community based, it is unsurprising that the number of hospitalisations observed in the trial is very low and mostly uncorrelated with the disease condition. For these reasons, the health economists in the study felt that the inclusion of hospitalisation costs in the main analysis might confound the economic results. Therefore, it was decided to perform analyses with and without hospitalisation costs to assess whether their inclusion was likely to have affected the main conclusions.

European Quality of Life – 5 dimensions

The EQ-5D questionnaire was completed by trial participants to provide preference data for the estimation of utilities. 118,120 The EQ-5D is a standardised instrument for the measurement of health outcome. It is applicable to a wide range of health conditions and treatments and provides a simple descriptive profile and a single index value for a patient’s health status. Its descriptive system consists of five dimensions (mobility, self-care, usual activity, pain/discomfort and anxiety/depression) with each dimension having three levels (no problem, some problem or extreme problem). The five dimensions with three levels each yield 243 possible health states. These health states have been valued on the 0 (equivalent to dead) to 1 (equivalent to full health) utility scale, using a community sample of people from the UK who valued the health states using the time trade-off technique. 121

Within the trial the EQ-5D questionnaire was completed at baseline and at 3, 6, 12 and 18 months (with the exception of the standard care arm whose participants did not complete the questionnaire at 3 months). At baseline the questionnaire was completed by participants during the baseline assessment period. They were expected to complete it in privacy although they could ask for instruction if they were unclear how to complete an assessment. At 3, 6, 12 and 18 months the participants were requested to complete the questionnaire before the assessment session.

Within-trial analysis

The within-trial analysis involved quantification of the mean resource use and costs at 18 months, as well as estimation of the mean EQ-5D scores at baseline and at different follow-up points. Estimates are reported together with an appropriate measure of sampling uncertainty (e.g. standard deviation) at different follow-up times in the four arms of the trial. QALYs were not calculated because, in chronic health conditions, costs and health benefits manifest themselves over the entire lifetime of the patient. Any intervention that aims to affect future costs and/or health benefits in this patient population needs to be evaluated within the relevant time horizon. It follows that estimation of within-trial QALYs is often inappropriate if not misleading, as benefits extend beyond the trial follow-up period. In this case a long-term extrapolation of the results of the trial is needed. In light of this, the within-trial analyses are reported as summary statistics for resource use, costs and EQ-5D scores in each arm of the trial at baseline and at different follow-up points.

To estimate the cost-effectiveness of alternative minimally invasive continuous glucose monitoring devices against conventional monitoring, a beyond follow-up analysis is required to estimate long-term resource use, costs and QALYs over an appropriate time horizon. However, this would only be necessary if the difference in clinical outcome between the trial arms was found to be potentially clinically and economically significant, as in this case a difference in costs and benefits in the long run would be expected.

Statistical methods

Long-term impact on HbA1c (18 months)

The primary analysis of the study was intention to treat to determine the long-term impact on HbA1c of wearing minimally invasive continuous glucose monitors. These findings are displayed in Table 12 and Figure 7.

FIGURE 7.

Mean percentage change in HbA1c from baseline to 3, 6, 12 and 18 months; follow-up with confidence intervals.

These data indicate that all arms showed a reduction in HbA1c by 18 months’ follow-up. There was, however, no statistically significant advantage to the continuous glucose monitoring devices at 18 months.

Secondary clinical outcomes

Secondary end points: short- and medium-term impact on HbA1c

Medium-term impact on HbA1c (6 and 12 months)

The medium-term impact on HbA1c was also examined by assessing the change in HbA1c from baseline to 6 and 12 months’ follow-up. These findings are displayed in Tables 14 and 15, respectively, and in Figure 8.

TABLE 14 - Relative percentage change in HbA1c from baseline to 6 months’ follow-up

CI, confidence interval.

ANOVA: F = 1.97 (3, 337), p = 0.12.

	n	Baseline HbA1c (%), mean (SD)	6-month HbA1c (%), mean (SD)	Relative percentage change in HbA1c
				Mean difference within group (SD)	Mean difference compared with standard care control	95% CI compared with standard care control
GlucoWatch	81	9.0 (1.2)	8.7 (1.3)	–2.5 (12.9)	3.4	–0.4 to 7.3
CGMS	88	9.1 (1.2)	8.4 (1.4)	–6.7 (10.6)	–0.8	–4.2 to 2.6
Attention control	86	8.9 (1.1)	8.3 (1.1)	–6.0 (13.5)	–0.1	–3.9 to 3.8
Standard care control	86	9.5 (1.4)	8.8 (1.4)	–5.9 (12.1)
Total	341	9.1 (1.2)	8.6 (1.3)	–5.3 (12.4)

TABLE 15 - Relative percentage change in HbA1c from baseline to 12 months’ follow-up

CI, confidence interval.

ANOVA: F = 3.39, (3, 335), p = 0.02.

	n	Baseline HbA1c (%), mean (SD)	12-month HbA1c (%), mean (SD)	Relative percentage change in HbA1c
				Mean difference within group (SD)	Mean difference compared with standard care control	95% CI compared with standard care control
GlucoWatch	84	9.1 (1.4)	9.0 (1.6)	–0.9 (14.5)	5.7	1.4–10.0
CGMS	86	8.9 (1.0)	8.4 (1.1)	–5.1 (12.4)	1.5	–2.4 to 5.5
Attention Control	89	8.9 (1.1)	8.3 (1.2)	–6.6 (13.4)	0.0	–4.0 to 4.1
Standard care control	80	9.4 (1.3)	8.7 (1.4)	–6.6 (13.4)
Total	339	9.1 (1.2)	8.6 (1.3)	–4.8 (13.6)

FIGURE 8.

Percentage of participants maintaining a clinically significant reduction in HbA1c (12.5%) at 6, 12 and 18 months by trial arm.

As at the 18-month follow-up time point, at both 6 and 12 months’ follow-up all of the groups showed a reduction in HbA1c from baseline. Although the GlucoWatch group seemed to do less well at the 6-month follow-up point than the other groups, there was no significant group effect at this assessment point. There was a statistically significant difference in HbA1c levels between the groups at 12 months (p = 0.02), indicating slightly less improvement in the GlucoWatch group than in the other groups. The mean relative reduction in HbA1c ranged from 0.8% (GlucoWatch) to 5% (CGMS) and 7% (attention and standard care control groups). This translates to absolute mean differences in HbA1c from baseline of 0.1% in the GlucoWatch group, 0.5% in the CGMS group and 0.6% in the other groups. This reflected the relatively poorer performance of the GlucoWatch group at the other time points but, given the number of comparisons performed, may have occurred by chance.

Proportion of individuals achieving a clinically meaningful reduction in HbA1c

As part of the secondary analysis the proportion of participants achieving a 12.5% reduction in HbA1c from baseline at each follow-up period was examined and tested using the chi-squared test. This analysis is presented in Table 16. There were no significant differences between the groups in the proportion of people achieving a 12.5% reduction in HbA1c. Overall, almost 25% of the total group achieved this reduction at each follow-up period.

Percentage maintaining a clinically meaningful reduction in HbA1c

Further descriptive analysis, was undertaken to examine how many participants maintained a reduction of 12.5% in their HbAlc levels. Maintenance was defined as demonstrating a 12.5% reduction in HbA1c at two consecutive visits. Only patients with consecutive visits were included in this comparison, which is shown in Figure 8.

Although the proportion attaining and maintaining benefit in the GlucoWatch arm was consistently lower than the proportions in the other three arms, the difference did not achieve statistical significance.

Hypoglycaemic episodes

The following data were derived from the Lifescan meters used by all participants in all arms of the trial. At each research visit the number of hypoglycaemic readings reported over the past 28 days was recorded. A hypoglycaemic reading was defined as a glucose reading of ≤ 3.5 mmol/l or self-reported hypoglycaemia even if a glucose reading (taken at the time) was > 3.5 mmol/l . Lifescan data were downloaded at each research visit and information on hypoglycaemic episodes was gathered from the downloaded data. In the few instances in which participants did not remember to bring their meters in but data were available from their diaries, the data were used instead. The data on hypoglycaemic episodes are displayed in Table 17. The baseline data has not been reported here as this was based on retrospective reporting of hypoglycaemic episodes in the past 28 days (with or without diaries/meters). These data showed some differences in the percentage of participants reporting hypoglycaemic episodes, as well as in the proportion of total glucose readings that were classified as hypoglycaemic. No consistent pattern emerged between groups or over time.

Severe hypoglycaemia, when the person with diabetes is unable to recognise the symptoms of low blood glucose and requires assistance from another person to administer treatment, was also assessed. Throughout the trial, the number of respondents reporting severe episodes of hypoglycaemia was very low – five or less people in any one trial arm (Table 18). These episodes of severe hypoglycaemia accounted for less than 0.02% of the total hypoglycaemic episodes throughout the course of the trial.

At each assessment period, participants were asked whether they knew when hypoglycaemia was commencing. They rated this on a visual analogue scale from 1 (never) to 5 (always). Table 19 documents the number and percentage of respondents within each trial arm who scored 1–5 on that scale. Table 20 shows the median scores by trial arm at each assessment point. From these tables it can be seen that the trial population included few people suffering from problems with hypoglycaemia awareness. At each assessment point the majority of people scored 4 or 5 on this scale.

Hyperglycaemic episodes

Data on hyperglycaemic episodes were collected as described for hypoglycaemic episodes. A hyperglycaemic reading was defined as ≥  10.0 mmol/l . These data are presented in Table 21.

As with the data on hypoglycaemic episodes, there did not appear to be any consistent differences between the study groups in either the number of people reporting hyperglycaemic episodes or the proportion of total readings that were hyperglycaemic.

Long-term impact on HbA1c – per protocol analysis

There was no group effect in terms of relative change in HbA1c from baseline to 18 months in the per protocol analysis (Table 23).

Short- and medium-term impact on HbA1c – per protocol analysis

Tables 24–26 display the results for the short- and medium-term impact of the devices on HbA1c for the per protocol analysis (3, 6 and 12 months’ follow-up respectively).

The per protocol analysis examining the impact of the monitors when worn a minimum number of occasions on HbA1c in the short and medium term mirrored the results of the intention to treat analysis. As in the intention to treat analysis, at the 3-, 6- and 12-month follow-up, all of the groups showed some reduction in HbA1c from baseline. The GlucoWatch group appeared to show the least improvement in comparison with the other groups, although this was not statistically significant.

Proportion of individuals achieving a clinically meaningful reduction in HbA1c – per protocol analysis

The per protocol analysis of the proportion of individuals achieving a clinically meaningful reduction in HbA1c is displayed in Table 27. No significant difference was found between the groups.

Monitor use

Table 31 displays the number of times that the devices were worn for each participant at each time point, as well as the cumulative number (%) of people who had stopped using the devices and their reasons for stopping. The overall percentages continuing to use the devices are shown in Figure 10.

FIGURE 10.

Percentage of participants continuing to use the devices over the course of the study (data relate to those people who attended research visits).

In phase 1 of the trial GlucoWatch patients were asked to use the watch at times of their choice but with a minimum attempted use of four times per month (12 times in phase 1, the first 3-month period) and a maximum attempted use of four times per week (52 times in phase 1). It was planned that the CGMS group would have the device fitted at baseline and at 6 and 12 weeks, that is, three times in phase 1 of the trial.

The per protocol analysis for phase 1 (0–3 months) was defined as:

• CGMS: worn at least once

• GlucoWatch: worn at least three times.

During this period the majority of the CGMS group wore the device at least once (n = 98, 96%). In the GlucoWatch group, 68 (68%) individuals wore the device at least three times during phase 1; median use of the devices was five times by 4 weeks’ follow-up, three times between 4 and 8 weeks’ follow-up and once between the 8- and 12-week visits. Six participants allocated to the GlucoWatch and four CGMS participants chose not to wear the devices at all throughout the course of the study.

During phase 2 (3–18 months) of the study, use of both devices declined, although the decline was considerably greater in the GlucoWatch group. Skin reactions were the most common reason given for stopping use in this group.

As can be seen in Table 31 and Figure 10, a greater number of CGMS participants than GlucoWatch participants used the device throughout the course of the study. In both arms of the trial it was unusual to see participants interrupt using the device and then start using it again (n = 9 GlucoWatch versus n = 3 CGMS).

Skin reactions over the course of the study (0–18 months)

Skin reactions are known to be commonly associated with use of the GlucoWatch. In Table 32 the number of people reporting skin reactions, the duration of skin problems, the number of people who removed the monitor because of skin problems, the typical MITRE Skin Scale score for those reactions and the number of people reporting severe skin reactions are presented. The majority of the GlucoWatch group reported skin reactions. The median duration of skin problems ranged from 3 to 60 days in this group. A higher MITRE Skin Scale score indicates a greater severity of skin reaction. The median MITRE Skin Scale scores reported in the GlucoWatch group tended to be higher than those in the CGMS group. No-one in the CGMS group reported a severe skin reaction (MITRE Skin Scale score ≥  6). The proportion of people reporting severe skin reactions in the GlucoWatch group ranged from 14% to 48%. The number of people removing the device because of skin reactions provides an indicator of how willing participants were to tolerate skin reactions. In the GlucoWatch group the proportion of people removing the device at each time point because of skin problems ranged from 9% to 23%.

Secondary data analysis: factors associated with use of the devices

As part of an exploratory secondary data analysis, four factors assessed at baseline and agreed a priori (age, sex, type of diabetes and fear of hypoglycaemia) were examined to see if there were differences in the level of use of the two devices during the intensive phase of the study (0–3 months) according to these factors. For these analyses, the distributions for the frequency of use data were examined to determine appropriate categories of use. The following categories were applied: CGMS worn one or two times versus three times; GlucoWatch worn one or two times, three to six times, 7–15 times or ≥ 16 times. Chi-squared tests, ANOVAs and t-tests were used to perform these analyses. There were no significant differences in frequency of use according to any of these factors. In the GlucoWatch group, those in the highest use group (worn 16 times or more during phase 1) reported the highest scores on the fear of hypoglycaemia scale, indicating greater fear, but this was not statistically significant (data not presented here).

Side effects, interference with lifestyle and impact of wearing the monitors during phase 1 (0–3 months)

This section presents the results of the analysis of the acceptability data collected at 3 months’ follow-up, that is, after completion of phase 1, the intensive part of the study. A total of 10 participants (six GlucoWatch, four CGMS) did not wear the devices at any point during the study, and so it was not appropriate for them to complete the acceptability questionnaire as the questions are based on use of the devices. In addition, 43 participants (23 GlucoWatch, 20 CGMS) did not complete the acceptability questionnaire at 3 months’ follow-up. For these reasons the valid numbers in the following analyses do not correspond to those for the analyses on HbA1c and monitor use, as detailed previously.

Side effects

Participants were asked whether they had experienced any of nine specific side effects. If they answered ‘yes’ to any of these questions, they were then asked to rate how acceptable these side effects were to them. Tables 33 and 34 show the numbers of people reporting each of the nine side effects and their acceptability ratings for both devices respectively.

TABLE 33 - Number of CGMS participants reporting side effects and their acceptability ratings in phase 1 (0–3 months)

	Not reporting side effect, n (%)	Reporting side effect, n (%)	Acceptability of side effect
			Not at all acceptable, n (%)	Slightly acceptable, n (%)	Moderately acceptable, n (%)	Very acceptable, n (%)	Completely acceptable, n (%)
Itching	50 (66)	26 (34)	1 (1)	9 (12)	7 (9)	3 (4)	6 (8)
Tingling	73 (96)	3 (4)	0	3 (4)	0	0	0
Soreness	58 (76)	18 (24)	3 (4)	6 (8)	7 (9)	0	2 (3)
Dry skin	71 (93)	5 (7)	1 (1)	0	4 (5)	0	0
Red marks	50 (66)	26 (34)	2 (3)	4 (5)	8 (11)	4 (5)	8 (11)
Discomfort	50 (66)	26 (34)	2 (3)	10 (13)	9 (12)	2 (3)	3 (4)
Bruising	68 (90)	8 (11)	3 (4)	2 (3)	1 (1)	1 (1)	1 (1)
Pain	65 (86)	11 (14)	2 (3)	5 (7)	2 (3)	2 (3)	0
Blisters	71 (93)	5 (7)	1 (1)	2 (3)	2 (3)	0	0

TABLE 34 - Number of GlucoWatch participants reporting side effects and their acceptability ratings in phase 1 (0–3 months)

	Not reporting side effect, n (%)	Reporting side effect, n (%)	Acceptability of side effect
			Not at all acceptable, n (%)	Slightly acceptable, n (%)	Moderately acceptable, n (%)	Very acceptable, n (%)	Completely acceptable, n (%)
Itching	5 (7)	66 (93)	22 (31)	27 (38)	15 (21)	0	2 (3)
Tingling	24 (34)	47 (66)	9 (13)	25 (35)	8 (11)	1 (1)	4 (6)
Soreness	19 (27)	52 (73)	25 (35)	19 (27)	7 (10)	0	1 (1)
Dry skin	29 (42)	40 (58)	11 (16)	18 (26)	7 (10)	2 (3)	2 (3)
Red marks	5 (7)	66 (93)	28 (39)	22 (31)	12 (17)	0	4 (6)
Discomfort	18 (25)	53 (75)	20 (28)	22 (31)	10 (14)	0	1 (1)
Bruising	54 (76)	17 (24)	6 (8)	3 (4)	5 (7)	1 (1)	2 (3)
Pain	35 (49)	36 (51)	13 (18)	12 (17)	10 (14)	0	1 (1)
Blisters	25 (35)	46 (65)	26 (37)	12 (17)	6 (8)	1 (1)	1 (1)

In total, 63% (48/76) of the CGMS group and 97% (69/71) of the GlucoWatch group reported at least one side effect. In the CGMS group 17% (8/48) of those who reported one or more side effect rated at least one of these as ‘not at all acceptable’. The figure for the GlucoWatch group was 63% (44/71).

The relationships between the percentage of each device group experiencing side effects and the number of times that each device was worn are shown in Figures 11 and 12 respectively; there would appear to be no relationship for the GlucoWatch but a suggestion of an increase in the percentage of the device group experiencing side effects with an increase in the number of times that the device was worn in the case of the CGMS.

FIGURE 11.

Percentage of GlucoWatch group experiencing side effects by number of times worn.

FIGURE 12.

Percentage of CGMS group experiencing side effects by number of times worn.

Impact of wearing the monitor

Participants were asked to complete a 33-item questionnaire that related more generally to the impact of wearing the monitor. For each of the 33 statements they were asked to indicate the extent to which they agreed or disagreed on a five-point Likert scale. The positively worded items were reverse scored so that a higher score meant a more negative impact from wearing the device.

Principal components analysis

A principal components analysis (PCA) was conducted to examine the factor structure of this questionnaire. PCA is a factor analytic technique that assists in detecting structure in the relationships between variables (questionnaire items) and thereby allows a reduction of the number of variables. A PCA of this section of the questionnaire indicated a three-factor solution that included 24 items and accounted for 42% of the variance. The components were:

• Factor 1: Ease of use, practicality (10 items, scored 0–50). Example items:

  1. – I could not always enter information into the machine as instructed to

  2. – I found the use of the monitor required careful planning.

• Factor 2: Value of the device, improvement in glycaemic control (eight items, scored 0–40). Example items:

  1. – Wearing the monitor has helped me reduce the number of hypoglycaemic episodes I experience

  2. – I would be interested in using the machine in the future.

• Factor 3: Appearance, self-consciousness, disclosure (six items, scored 0–30). Example items:

  1. – I felt more self-conscious of my appearance when I was wearing the monitor

  2. – I was happy to explain what the monitor was to anyone who asked.

There were nine redundant items in total. The authors would recommend omitting these nine items in future work with this questionnaire. The complete list of questionnaire items and their factor loadings is provided in Appendix 9.

The two devices were compared on each of these three factors using t-tests (Table 42). The mean scores for each device on each subscale of the questionnaire are also displayed in Figure 13.

TABLE 42 - Comparison of mean scores on the subscales of the impact questionnaire

Subscales	Valid, n	Mean score (SD)	t-value	df	p-value
Ease of use
GlucoWatch	69	33.4 (8.2)	–9.3	145	0.000
CGMS	78	21.8 (6.9)
Value of device
GlucoWatch	69	23.1 (8.3)	–2.6	130	0.010
CGMS	78	19.9 (6.7)
Appearance, self-consciousness, disclosure
GlucoWatch	69	13.0 (4.8)	–9.6	145	0.339
CGMS	78	12.2 (4.8)

FIGURE 13.

Mean scores on the subscales of the impact questionnaire (higher scores indicate a more negative impact from wearing the device). ns, not significant.